1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4 #include <linux/bpf.h>
5 #include <linux/bpf-cgroup.h>
6 #include <linux/bpf_trace.h>
7 #include <linux/bpf_lirc.h>
8 #include <linux/bpf_verifier.h>
9 #include <linux/bsearch.h>
10 #include <linux/btf.h>
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 #include <linux/vmalloc.h>
15 #include <linux/mmzone.h>
16 #include <linux/anon_inodes.h>
17 #include <linux/fdtable.h>
18 #include <linux/file.h>
19 #include <linux/fs.h>
20 #include <linux/license.h>
21 #include <linux/filter.h>
22 #include <linux/kernel.h>
23 #include <linux/idr.h>
24 #include <linux/cred.h>
25 #include <linux/timekeeping.h>
26 #include <linux/ctype.h>
27 #include <linux/nospec.h>
28 #include <linux/audit.h>
29 #include <uapi/linux/btf.h>
30 #include <linux/pgtable.h>
31 #include <linux/bpf_lsm.h>
32 #include <linux/poll.h>
33 #include <linux/sort.h>
34 #include <linux/bpf-netns.h>
35 #include <linux/rcupdate_trace.h>
36 #include <linux/memcontrol.h>
37 #include <linux/trace_events.h>
38 #include <net/netfilter/nf_bpf_link.h>
39
40 #include <net/tcx.h>
41
42 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
43 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
44 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
45 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
46 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
47 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
48 IS_FD_HASH(map))
49
50 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
51
52 DEFINE_PER_CPU(int, bpf_prog_active);
53 static DEFINE_IDR(prog_idr);
54 static DEFINE_SPINLOCK(prog_idr_lock);
55 static DEFINE_IDR(map_idr);
56 static DEFINE_SPINLOCK(map_idr_lock);
57 static DEFINE_IDR(link_idr);
58 static DEFINE_SPINLOCK(link_idr_lock);
59
60 int sysctl_unprivileged_bpf_disabled __read_mostly =
61 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
62
63 static const struct bpf_map_ops * const bpf_map_types[] = {
64 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
65 #define BPF_MAP_TYPE(_id, _ops) \
66 [_id] = &_ops,
67 #define BPF_LINK_TYPE(_id, _name)
68 #include <linux/bpf_types.h>
69 #undef BPF_PROG_TYPE
70 #undef BPF_MAP_TYPE
71 #undef BPF_LINK_TYPE
72 };
73
74 /*
75 * If we're handed a bigger struct than we know of, ensure all the unknown bits
76 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
77 * we don't know about yet.
78 *
79 * There is a ToCToU between this function call and the following
80 * copy_from_user() call. However, this is not a concern since this function is
81 * meant to be a future-proofing of bits.
82 */
bpf_check_uarg_tail_zero(bpfptr_t uaddr,size_t expected_size,size_t actual_size)83 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
84 size_t expected_size,
85 size_t actual_size)
86 {
87 int res;
88
89 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
90 return -E2BIG;
91
92 if (actual_size <= expected_size)
93 return 0;
94
95 if (uaddr.is_kernel)
96 res = memchr_inv(uaddr.kernel + expected_size, 0,
97 actual_size - expected_size) == NULL;
98 else
99 res = check_zeroed_user(uaddr.user + expected_size,
100 actual_size - expected_size);
101 if (res < 0)
102 return res;
103 return res ? 0 : -E2BIG;
104 }
105
106 const struct bpf_map_ops bpf_map_offload_ops = {
107 .map_meta_equal = bpf_map_meta_equal,
108 .map_alloc = bpf_map_offload_map_alloc,
109 .map_free = bpf_map_offload_map_free,
110 .map_check_btf = map_check_no_btf,
111 .map_mem_usage = bpf_map_offload_map_mem_usage,
112 };
113
bpf_map_write_active_inc(struct bpf_map * map)114 static void bpf_map_write_active_inc(struct bpf_map *map)
115 {
116 atomic64_inc(&map->writecnt);
117 }
118
bpf_map_write_active_dec(struct bpf_map * map)119 static void bpf_map_write_active_dec(struct bpf_map *map)
120 {
121 atomic64_dec(&map->writecnt);
122 }
123
bpf_map_write_active(const struct bpf_map * map)124 bool bpf_map_write_active(const struct bpf_map *map)
125 {
126 return atomic64_read(&map->writecnt) != 0;
127 }
128
bpf_map_value_size(const struct bpf_map * map)129 static u32 bpf_map_value_size(const struct bpf_map *map)
130 {
131 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
132 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
133 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
134 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
135 return round_up(map->value_size, 8) * num_possible_cpus();
136 else if (IS_FD_MAP(map))
137 return sizeof(u32);
138 else
139 return map->value_size;
140 }
141
maybe_wait_bpf_programs(struct bpf_map * map)142 static void maybe_wait_bpf_programs(struct bpf_map *map)
143 {
144 /* Wait for any running BPF programs to complete so that
145 * userspace, when we return to it, knows that all programs
146 * that could be running use the new map value.
147 */
148 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
149 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
150 synchronize_rcu();
151 }
152
bpf_map_update_value(struct bpf_map * map,struct file * map_file,void * key,void * value,__u64 flags)153 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
154 void *key, void *value, __u64 flags)
155 {
156 int err;
157
158 /* Need to create a kthread, thus must support schedule */
159 if (bpf_map_is_offloaded(map)) {
160 return bpf_map_offload_update_elem(map, key, value, flags);
161 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
162 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
163 return map->ops->map_update_elem(map, key, value, flags);
164 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
165 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
166 return sock_map_update_elem_sys(map, key, value, flags);
167 } else if (IS_FD_PROG_ARRAY(map)) {
168 return bpf_fd_array_map_update_elem(map, map_file, key, value,
169 flags);
170 }
171
172 bpf_disable_instrumentation();
173 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
174 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
175 err = bpf_percpu_hash_update(map, key, value, flags);
176 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
177 err = bpf_percpu_array_update(map, key, value, flags);
178 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
179 err = bpf_percpu_cgroup_storage_update(map, key, value,
180 flags);
181 } else if (IS_FD_ARRAY(map)) {
182 rcu_read_lock();
183 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
184 flags);
185 rcu_read_unlock();
186 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
187 rcu_read_lock();
188 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
189 flags);
190 rcu_read_unlock();
191 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
192 /* rcu_read_lock() is not needed */
193 err = bpf_fd_reuseport_array_update_elem(map, key, value,
194 flags);
195 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
196 map->map_type == BPF_MAP_TYPE_STACK ||
197 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
198 err = map->ops->map_push_elem(map, value, flags);
199 } else {
200 rcu_read_lock();
201 err = map->ops->map_update_elem(map, key, value, flags);
202 rcu_read_unlock();
203 }
204 bpf_enable_instrumentation();
205 maybe_wait_bpf_programs(map);
206
207 return err;
208 }
209
bpf_map_copy_value(struct bpf_map * map,void * key,void * value,__u64 flags)210 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
211 __u64 flags)
212 {
213 void *ptr;
214 int err;
215
216 if (bpf_map_is_offloaded(map))
217 return bpf_map_offload_lookup_elem(map, key, value);
218
219 bpf_disable_instrumentation();
220 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
221 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
222 err = bpf_percpu_hash_copy(map, key, value);
223 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
224 err = bpf_percpu_array_copy(map, key, value);
225 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
226 err = bpf_percpu_cgroup_storage_copy(map, key, value);
227 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
228 err = bpf_stackmap_copy(map, key, value);
229 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
230 err = bpf_fd_array_map_lookup_elem(map, key, value);
231 } else if (IS_FD_HASH(map)) {
232 err = bpf_fd_htab_map_lookup_elem(map, key, value);
233 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
234 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
235 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
236 map->map_type == BPF_MAP_TYPE_STACK ||
237 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
238 err = map->ops->map_peek_elem(map, value);
239 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
240 /* struct_ops map requires directly updating "value" */
241 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
242 } else {
243 rcu_read_lock();
244 if (map->ops->map_lookup_elem_sys_only)
245 ptr = map->ops->map_lookup_elem_sys_only(map, key);
246 else
247 ptr = map->ops->map_lookup_elem(map, key);
248 if (IS_ERR(ptr)) {
249 err = PTR_ERR(ptr);
250 } else if (!ptr) {
251 err = -ENOENT;
252 } else {
253 err = 0;
254 if (flags & BPF_F_LOCK)
255 /* lock 'ptr' and copy everything but lock */
256 copy_map_value_locked(map, value, ptr, true);
257 else
258 copy_map_value(map, value, ptr);
259 /* mask lock and timer, since value wasn't zero inited */
260 check_and_init_map_value(map, value);
261 }
262 rcu_read_unlock();
263 }
264
265 bpf_enable_instrumentation();
266 maybe_wait_bpf_programs(map);
267
268 return err;
269 }
270
271 /* Please, do not use this function outside from the map creation path
272 * (e.g. in map update path) without taking care of setting the active
273 * memory cgroup (see at bpf_map_kmalloc_node() for example).
274 */
__bpf_map_area_alloc(u64 size,int numa_node,bool mmapable)275 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
276 {
277 /* We really just want to fail instead of triggering OOM killer
278 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
279 * which is used for lower order allocation requests.
280 *
281 * It has been observed that higher order allocation requests done by
282 * vmalloc with __GFP_NORETRY being set might fail due to not trying
283 * to reclaim memory from the page cache, thus we set
284 * __GFP_RETRY_MAYFAIL to avoid such situations.
285 */
286
287 gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
288 unsigned int flags = 0;
289 unsigned long align = 1;
290 void *area;
291
292 if (size >= SIZE_MAX)
293 return NULL;
294
295 /* kmalloc()'ed memory can't be mmap()'ed */
296 if (mmapable) {
297 BUG_ON(!PAGE_ALIGNED(size));
298 align = SHMLBA;
299 flags = VM_USERMAP;
300 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
301 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
302 numa_node);
303 if (area != NULL)
304 return area;
305 }
306
307 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
308 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
309 flags, numa_node, __builtin_return_address(0));
310 }
311
bpf_map_area_alloc(u64 size,int numa_node)312 void *bpf_map_area_alloc(u64 size, int numa_node)
313 {
314 return __bpf_map_area_alloc(size, numa_node, false);
315 }
316
bpf_map_area_mmapable_alloc(u64 size,int numa_node)317 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
318 {
319 return __bpf_map_area_alloc(size, numa_node, true);
320 }
321
bpf_map_area_free(void * area)322 void bpf_map_area_free(void *area)
323 {
324 kvfree(area);
325 }
326
bpf_map_flags_retain_permanent(u32 flags)327 static u32 bpf_map_flags_retain_permanent(u32 flags)
328 {
329 /* Some map creation flags are not tied to the map object but
330 * rather to the map fd instead, so they have no meaning upon
331 * map object inspection since multiple file descriptors with
332 * different (access) properties can exist here. Thus, given
333 * this has zero meaning for the map itself, lets clear these
334 * from here.
335 */
336 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
337 }
338
bpf_map_init_from_attr(struct bpf_map * map,union bpf_attr * attr)339 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
340 {
341 map->map_type = attr->map_type;
342 map->key_size = attr->key_size;
343 map->value_size = attr->value_size;
344 map->max_entries = attr->max_entries;
345 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
346 map->numa_node = bpf_map_attr_numa_node(attr);
347 map->map_extra = attr->map_extra;
348 }
349
bpf_map_alloc_id(struct bpf_map * map)350 static int bpf_map_alloc_id(struct bpf_map *map)
351 {
352 int id;
353
354 idr_preload(GFP_KERNEL);
355 spin_lock_bh(&map_idr_lock);
356 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
357 if (id > 0)
358 map->id = id;
359 spin_unlock_bh(&map_idr_lock);
360 idr_preload_end();
361
362 if (WARN_ON_ONCE(!id))
363 return -ENOSPC;
364
365 return id > 0 ? 0 : id;
366 }
367
bpf_map_free_id(struct bpf_map * map)368 void bpf_map_free_id(struct bpf_map *map)
369 {
370 unsigned long flags;
371
372 /* Offloaded maps are removed from the IDR store when their device
373 * disappears - even if someone holds an fd to them they are unusable,
374 * the memory is gone, all ops will fail; they are simply waiting for
375 * refcnt to drop to be freed.
376 */
377 if (!map->id)
378 return;
379
380 spin_lock_irqsave(&map_idr_lock, flags);
381
382 idr_remove(&map_idr, map->id);
383 map->id = 0;
384
385 spin_unlock_irqrestore(&map_idr_lock, flags);
386 }
387
388 #ifdef CONFIG_MEMCG_KMEM
bpf_map_save_memcg(struct bpf_map * map)389 static void bpf_map_save_memcg(struct bpf_map *map)
390 {
391 /* Currently if a map is created by a process belonging to the root
392 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
393 * So we have to check map->objcg for being NULL each time it's
394 * being used.
395 */
396 if (memcg_bpf_enabled())
397 map->objcg = get_obj_cgroup_from_current();
398 }
399
bpf_map_release_memcg(struct bpf_map * map)400 static void bpf_map_release_memcg(struct bpf_map *map)
401 {
402 if (map->objcg)
403 obj_cgroup_put(map->objcg);
404 }
405
bpf_map_get_memcg(const struct bpf_map * map)406 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
407 {
408 if (map->objcg)
409 return get_mem_cgroup_from_objcg(map->objcg);
410
411 return root_mem_cgroup;
412 }
413
bpf_map_kmalloc_node(const struct bpf_map * map,size_t size,gfp_t flags,int node)414 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
415 int node)
416 {
417 struct mem_cgroup *memcg, *old_memcg;
418 void *ptr;
419
420 memcg = bpf_map_get_memcg(map);
421 old_memcg = set_active_memcg(memcg);
422 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
423 set_active_memcg(old_memcg);
424 mem_cgroup_put(memcg);
425
426 return ptr;
427 }
428
bpf_map_kzalloc(const struct bpf_map * map,size_t size,gfp_t flags)429 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
430 {
431 struct mem_cgroup *memcg, *old_memcg;
432 void *ptr;
433
434 memcg = bpf_map_get_memcg(map);
435 old_memcg = set_active_memcg(memcg);
436 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
437 set_active_memcg(old_memcg);
438 mem_cgroup_put(memcg);
439
440 return ptr;
441 }
442
bpf_map_kvcalloc(struct bpf_map * map,size_t n,size_t size,gfp_t flags)443 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
444 gfp_t flags)
445 {
446 struct mem_cgroup *memcg, *old_memcg;
447 void *ptr;
448
449 memcg = bpf_map_get_memcg(map);
450 old_memcg = set_active_memcg(memcg);
451 ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
452 set_active_memcg(old_memcg);
453 mem_cgroup_put(memcg);
454
455 return ptr;
456 }
457
bpf_map_alloc_percpu(const struct bpf_map * map,size_t size,size_t align,gfp_t flags)458 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
459 size_t align, gfp_t flags)
460 {
461 struct mem_cgroup *memcg, *old_memcg;
462 void __percpu *ptr;
463
464 memcg = bpf_map_get_memcg(map);
465 old_memcg = set_active_memcg(memcg);
466 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
467 set_active_memcg(old_memcg);
468 mem_cgroup_put(memcg);
469
470 return ptr;
471 }
472
473 #else
bpf_map_save_memcg(struct bpf_map * map)474 static void bpf_map_save_memcg(struct bpf_map *map)
475 {
476 }
477
bpf_map_release_memcg(struct bpf_map * map)478 static void bpf_map_release_memcg(struct bpf_map *map)
479 {
480 }
481 #endif
482
btf_field_cmp(const void * a,const void * b)483 static int btf_field_cmp(const void *a, const void *b)
484 {
485 const struct btf_field *f1 = a, *f2 = b;
486
487 if (f1->offset < f2->offset)
488 return -1;
489 else if (f1->offset > f2->offset)
490 return 1;
491 return 0;
492 }
493
btf_record_find(const struct btf_record * rec,u32 offset,u32 field_mask)494 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
495 u32 field_mask)
496 {
497 struct btf_field *field;
498
499 if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
500 return NULL;
501 field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
502 if (!field || !(field->type & field_mask))
503 return NULL;
504 return field;
505 }
506
btf_record_free(struct btf_record * rec)507 void btf_record_free(struct btf_record *rec)
508 {
509 int i;
510
511 if (IS_ERR_OR_NULL(rec))
512 return;
513 for (i = 0; i < rec->cnt; i++) {
514 switch (rec->fields[i].type) {
515 case BPF_KPTR_UNREF:
516 case BPF_KPTR_REF:
517 if (rec->fields[i].kptr.module)
518 module_put(rec->fields[i].kptr.module);
519 btf_put(rec->fields[i].kptr.btf);
520 break;
521 case BPF_LIST_HEAD:
522 case BPF_LIST_NODE:
523 case BPF_RB_ROOT:
524 case BPF_RB_NODE:
525 case BPF_SPIN_LOCK:
526 case BPF_TIMER:
527 case BPF_REFCOUNT:
528 /* Nothing to release */
529 break;
530 default:
531 WARN_ON_ONCE(1);
532 continue;
533 }
534 }
535 kfree(rec);
536 }
537
bpf_map_free_record(struct bpf_map * map)538 void bpf_map_free_record(struct bpf_map *map)
539 {
540 btf_record_free(map->record);
541 map->record = NULL;
542 }
543
btf_record_dup(const struct btf_record * rec)544 struct btf_record *btf_record_dup(const struct btf_record *rec)
545 {
546 const struct btf_field *fields;
547 struct btf_record *new_rec;
548 int ret, size, i;
549
550 if (IS_ERR_OR_NULL(rec))
551 return NULL;
552 size = offsetof(struct btf_record, fields[rec->cnt]);
553 new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
554 if (!new_rec)
555 return ERR_PTR(-ENOMEM);
556 /* Do a deep copy of the btf_record */
557 fields = rec->fields;
558 new_rec->cnt = 0;
559 for (i = 0; i < rec->cnt; i++) {
560 switch (fields[i].type) {
561 case BPF_KPTR_UNREF:
562 case BPF_KPTR_REF:
563 btf_get(fields[i].kptr.btf);
564 if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
565 ret = -ENXIO;
566 goto free;
567 }
568 break;
569 case BPF_LIST_HEAD:
570 case BPF_LIST_NODE:
571 case BPF_RB_ROOT:
572 case BPF_RB_NODE:
573 case BPF_SPIN_LOCK:
574 case BPF_TIMER:
575 case BPF_REFCOUNT:
576 /* Nothing to acquire */
577 break;
578 default:
579 ret = -EFAULT;
580 WARN_ON_ONCE(1);
581 goto free;
582 }
583 new_rec->cnt++;
584 }
585 return new_rec;
586 free:
587 btf_record_free(new_rec);
588 return ERR_PTR(ret);
589 }
590
btf_record_equal(const struct btf_record * rec_a,const struct btf_record * rec_b)591 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
592 {
593 bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
594 int size;
595
596 if (!a_has_fields && !b_has_fields)
597 return true;
598 if (a_has_fields != b_has_fields)
599 return false;
600 if (rec_a->cnt != rec_b->cnt)
601 return false;
602 size = offsetof(struct btf_record, fields[rec_a->cnt]);
603 /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
604 * members are zeroed out. So memcmp is safe to do without worrying
605 * about padding/unused fields.
606 *
607 * While spin_lock, timer, and kptr have no relation to map BTF,
608 * list_head metadata is specific to map BTF, the btf and value_rec
609 * members in particular. btf is the map BTF, while value_rec points to
610 * btf_record in that map BTF.
611 *
612 * So while by default, we don't rely on the map BTF (which the records
613 * were parsed from) matching for both records, which is not backwards
614 * compatible, in case list_head is part of it, we implicitly rely on
615 * that by way of depending on memcmp succeeding for it.
616 */
617 return !memcmp(rec_a, rec_b, size);
618 }
619
bpf_obj_free_timer(const struct btf_record * rec,void * obj)620 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
621 {
622 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
623 return;
624 bpf_timer_cancel_and_free(obj + rec->timer_off);
625 }
626
627 extern void __bpf_obj_drop_impl(void *p, const struct btf_record *rec);
628
bpf_obj_free_fields(const struct btf_record * rec,void * obj)629 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
630 {
631 const struct btf_field *fields;
632 int i;
633
634 if (IS_ERR_OR_NULL(rec))
635 return;
636 fields = rec->fields;
637 for (i = 0; i < rec->cnt; i++) {
638 struct btf_struct_meta *pointee_struct_meta;
639 const struct btf_field *field = &fields[i];
640 void *field_ptr = obj + field->offset;
641 void *xchgd_field;
642
643 switch (fields[i].type) {
644 case BPF_SPIN_LOCK:
645 break;
646 case BPF_TIMER:
647 bpf_timer_cancel_and_free(field_ptr);
648 break;
649 case BPF_KPTR_UNREF:
650 WRITE_ONCE(*(u64 *)field_ptr, 0);
651 break;
652 case BPF_KPTR_REF:
653 xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
654 if (!xchgd_field)
655 break;
656
657 if (!btf_is_kernel(field->kptr.btf)) {
658 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
659 field->kptr.btf_id);
660 migrate_disable();
661 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
662 pointee_struct_meta->record :
663 NULL);
664 migrate_enable();
665 } else {
666 field->kptr.dtor(xchgd_field);
667 }
668 break;
669 case BPF_LIST_HEAD:
670 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
671 continue;
672 bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
673 break;
674 case BPF_RB_ROOT:
675 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
676 continue;
677 bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
678 break;
679 case BPF_LIST_NODE:
680 case BPF_RB_NODE:
681 case BPF_REFCOUNT:
682 break;
683 default:
684 WARN_ON_ONCE(1);
685 continue;
686 }
687 }
688 }
689
690 /* called from workqueue */
bpf_map_free_deferred(struct work_struct * work)691 static void bpf_map_free_deferred(struct work_struct *work)
692 {
693 struct bpf_map *map = container_of(work, struct bpf_map, work);
694 struct btf_record *rec = map->record;
695 struct btf *btf = map->btf;
696
697 security_bpf_map_free(map);
698 bpf_map_release_memcg(map);
699 /* implementation dependent freeing */
700 map->ops->map_free(map);
701 /* Delay freeing of btf_record for maps, as map_free
702 * callback usually needs access to them. It is better to do it here
703 * than require each callback to do the free itself manually.
704 *
705 * Note that the btf_record stashed in map->inner_map_meta->record was
706 * already freed using the map_free callback for map in map case which
707 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
708 * template bpf_map struct used during verification.
709 */
710 btf_record_free(rec);
711 /* Delay freeing of btf for maps, as map_free callback may need
712 * struct_meta info which will be freed with btf_put().
713 */
714 btf_put(btf);
715 }
716
bpf_map_put_uref(struct bpf_map * map)717 static void bpf_map_put_uref(struct bpf_map *map)
718 {
719 if (atomic64_dec_and_test(&map->usercnt)) {
720 if (map->ops->map_release_uref)
721 map->ops->map_release_uref(map);
722 }
723 }
724
bpf_map_free_in_work(struct bpf_map * map)725 static void bpf_map_free_in_work(struct bpf_map *map)
726 {
727 INIT_WORK(&map->work, bpf_map_free_deferred);
728 /* Avoid spawning kworkers, since they all might contend
729 * for the same mutex like slab_mutex.
730 */
731 queue_work(system_unbound_wq, &map->work);
732 }
733
bpf_map_free_rcu_gp(struct rcu_head * rcu)734 static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
735 {
736 bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
737 }
738
bpf_map_free_mult_rcu_gp(struct rcu_head * rcu)739 static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
740 {
741 if (rcu_trace_implies_rcu_gp())
742 bpf_map_free_rcu_gp(rcu);
743 else
744 call_rcu(rcu, bpf_map_free_rcu_gp);
745 }
746
747 /* decrement map refcnt and schedule it for freeing via workqueue
748 * (underlying map implementation ops->map_free() might sleep)
749 */
bpf_map_put(struct bpf_map * map)750 void bpf_map_put(struct bpf_map *map)
751 {
752 if (atomic64_dec_and_test(&map->refcnt)) {
753 /* bpf_map_free_id() must be called first */
754 bpf_map_free_id(map);
755
756 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
757 if (READ_ONCE(map->free_after_mult_rcu_gp))
758 call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
759 else if (READ_ONCE(map->free_after_rcu_gp))
760 call_rcu(&map->rcu, bpf_map_free_rcu_gp);
761 else
762 bpf_map_free_in_work(map);
763 }
764 }
765 EXPORT_SYMBOL_GPL(bpf_map_put);
766
bpf_map_put_with_uref(struct bpf_map * map)767 void bpf_map_put_with_uref(struct bpf_map *map)
768 {
769 bpf_map_put_uref(map);
770 bpf_map_put(map);
771 }
772
bpf_map_release(struct inode * inode,struct file * filp)773 static int bpf_map_release(struct inode *inode, struct file *filp)
774 {
775 struct bpf_map *map = filp->private_data;
776
777 if (map->ops->map_release)
778 map->ops->map_release(map, filp);
779
780 bpf_map_put_with_uref(map);
781 return 0;
782 }
783
map_get_sys_perms(struct bpf_map * map,struct fd f)784 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
785 {
786 fmode_t mode = f.file->f_mode;
787
788 /* Our file permissions may have been overridden by global
789 * map permissions facing syscall side.
790 */
791 if (READ_ONCE(map->frozen))
792 mode &= ~FMODE_CAN_WRITE;
793 return mode;
794 }
795
796 #ifdef CONFIG_PROC_FS
797 /* Show the memory usage of a bpf map */
bpf_map_memory_usage(const struct bpf_map * map)798 static u64 bpf_map_memory_usage(const struct bpf_map *map)
799 {
800 return map->ops->map_mem_usage(map);
801 }
802
bpf_map_show_fdinfo(struct seq_file * m,struct file * filp)803 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
804 {
805 struct bpf_map *map = filp->private_data;
806 u32 type = 0, jited = 0;
807
808 if (map_type_contains_progs(map)) {
809 spin_lock(&map->owner.lock);
810 type = map->owner.type;
811 jited = map->owner.jited;
812 spin_unlock(&map->owner.lock);
813 }
814
815 seq_printf(m,
816 "map_type:\t%u\n"
817 "key_size:\t%u\n"
818 "value_size:\t%u\n"
819 "max_entries:\t%u\n"
820 "map_flags:\t%#x\n"
821 "map_extra:\t%#llx\n"
822 "memlock:\t%llu\n"
823 "map_id:\t%u\n"
824 "frozen:\t%u\n",
825 map->map_type,
826 map->key_size,
827 map->value_size,
828 map->max_entries,
829 map->map_flags,
830 (unsigned long long)map->map_extra,
831 bpf_map_memory_usage(map),
832 map->id,
833 READ_ONCE(map->frozen));
834 if (type) {
835 seq_printf(m, "owner_prog_type:\t%u\n", type);
836 seq_printf(m, "owner_jited:\t%u\n", jited);
837 }
838 }
839 #endif
840
bpf_dummy_read(struct file * filp,char __user * buf,size_t siz,loff_t * ppos)841 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
842 loff_t *ppos)
843 {
844 /* We need this handler such that alloc_file() enables
845 * f_mode with FMODE_CAN_READ.
846 */
847 return -EINVAL;
848 }
849
bpf_dummy_write(struct file * filp,const char __user * buf,size_t siz,loff_t * ppos)850 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
851 size_t siz, loff_t *ppos)
852 {
853 /* We need this handler such that alloc_file() enables
854 * f_mode with FMODE_CAN_WRITE.
855 */
856 return -EINVAL;
857 }
858
859 /* called for any extra memory-mapped regions (except initial) */
bpf_map_mmap_open(struct vm_area_struct * vma)860 static void bpf_map_mmap_open(struct vm_area_struct *vma)
861 {
862 struct bpf_map *map = vma->vm_file->private_data;
863
864 if (vma->vm_flags & VM_MAYWRITE)
865 bpf_map_write_active_inc(map);
866 }
867
868 /* called for all unmapped memory region (including initial) */
bpf_map_mmap_close(struct vm_area_struct * vma)869 static void bpf_map_mmap_close(struct vm_area_struct *vma)
870 {
871 struct bpf_map *map = vma->vm_file->private_data;
872
873 if (vma->vm_flags & VM_MAYWRITE)
874 bpf_map_write_active_dec(map);
875 }
876
877 static const struct vm_operations_struct bpf_map_default_vmops = {
878 .open = bpf_map_mmap_open,
879 .close = bpf_map_mmap_close,
880 };
881
bpf_map_mmap(struct file * filp,struct vm_area_struct * vma)882 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
883 {
884 struct bpf_map *map = filp->private_data;
885 int err;
886
887 if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
888 return -ENOTSUPP;
889
890 if (!(vma->vm_flags & VM_SHARED))
891 return -EINVAL;
892
893 mutex_lock(&map->freeze_mutex);
894
895 if (vma->vm_flags & VM_WRITE) {
896 if (map->frozen) {
897 err = -EPERM;
898 goto out;
899 }
900 /* map is meant to be read-only, so do not allow mapping as
901 * writable, because it's possible to leak a writable page
902 * reference and allows user-space to still modify it after
903 * freezing, while verifier will assume contents do not change
904 */
905 if (map->map_flags & BPF_F_RDONLY_PROG) {
906 err = -EACCES;
907 goto out;
908 }
909 }
910
911 /* set default open/close callbacks */
912 vma->vm_ops = &bpf_map_default_vmops;
913 vma->vm_private_data = map;
914 vm_flags_clear(vma, VM_MAYEXEC);
915 if (!(vma->vm_flags & VM_WRITE))
916 /* disallow re-mapping with PROT_WRITE */
917 vm_flags_clear(vma, VM_MAYWRITE);
918
919 err = map->ops->map_mmap(map, vma);
920 if (err)
921 goto out;
922
923 if (vma->vm_flags & VM_MAYWRITE)
924 bpf_map_write_active_inc(map);
925 out:
926 mutex_unlock(&map->freeze_mutex);
927 return err;
928 }
929
bpf_map_poll(struct file * filp,struct poll_table_struct * pts)930 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
931 {
932 struct bpf_map *map = filp->private_data;
933
934 if (map->ops->map_poll)
935 return map->ops->map_poll(map, filp, pts);
936
937 return EPOLLERR;
938 }
939
940 const struct file_operations bpf_map_fops = {
941 #ifdef CONFIG_PROC_FS
942 .show_fdinfo = bpf_map_show_fdinfo,
943 #endif
944 .release = bpf_map_release,
945 .read = bpf_dummy_read,
946 .write = bpf_dummy_write,
947 .mmap = bpf_map_mmap,
948 .poll = bpf_map_poll,
949 };
950
bpf_map_new_fd(struct bpf_map * map,int flags)951 int bpf_map_new_fd(struct bpf_map *map, int flags)
952 {
953 int ret;
954
955 ret = security_bpf_map(map, OPEN_FMODE(flags));
956 if (ret < 0)
957 return ret;
958
959 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
960 flags | O_CLOEXEC);
961 }
962
bpf_get_file_flag(int flags)963 int bpf_get_file_flag(int flags)
964 {
965 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
966 return -EINVAL;
967 if (flags & BPF_F_RDONLY)
968 return O_RDONLY;
969 if (flags & BPF_F_WRONLY)
970 return O_WRONLY;
971 return O_RDWR;
972 }
973
974 /* helper macro to check that unused fields 'union bpf_attr' are zero */
975 #define CHECK_ATTR(CMD) \
976 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
977 sizeof(attr->CMD##_LAST_FIELD), 0, \
978 sizeof(*attr) - \
979 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
980 sizeof(attr->CMD##_LAST_FIELD)) != NULL
981
982 /* dst and src must have at least "size" number of bytes.
983 * Return strlen on success and < 0 on error.
984 */
bpf_obj_name_cpy(char * dst,const char * src,unsigned int size)985 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
986 {
987 const char *end = src + size;
988 const char *orig_src = src;
989
990 memset(dst, 0, size);
991 /* Copy all isalnum(), '_' and '.' chars. */
992 while (src < end && *src) {
993 if (!isalnum(*src) &&
994 *src != '_' && *src != '.')
995 return -EINVAL;
996 *dst++ = *src++;
997 }
998
999 /* No '\0' found in "size" number of bytes */
1000 if (src == end)
1001 return -EINVAL;
1002
1003 return src - orig_src;
1004 }
1005
map_check_no_btf(const struct bpf_map * map,const struct btf * btf,const struct btf_type * key_type,const struct btf_type * value_type)1006 int map_check_no_btf(const struct bpf_map *map,
1007 const struct btf *btf,
1008 const struct btf_type *key_type,
1009 const struct btf_type *value_type)
1010 {
1011 return -ENOTSUPP;
1012 }
1013
map_check_btf(struct bpf_map * map,const struct btf * btf,u32 btf_key_id,u32 btf_value_id)1014 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
1015 u32 btf_key_id, u32 btf_value_id)
1016 {
1017 const struct btf_type *key_type, *value_type;
1018 u32 key_size, value_size;
1019 int ret = 0;
1020
1021 /* Some maps allow key to be unspecified. */
1022 if (btf_key_id) {
1023 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1024 if (!key_type || key_size != map->key_size)
1025 return -EINVAL;
1026 } else {
1027 key_type = btf_type_by_id(btf, 0);
1028 if (!map->ops->map_check_btf)
1029 return -EINVAL;
1030 }
1031
1032 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1033 if (!value_type || value_size != map->value_size)
1034 return -EINVAL;
1035
1036 map->record = btf_parse_fields(btf, value_type,
1037 BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1038 BPF_RB_ROOT | BPF_REFCOUNT,
1039 map->value_size);
1040 if (!IS_ERR_OR_NULL(map->record)) {
1041 int i;
1042
1043 if (!bpf_capable()) {
1044 ret = -EPERM;
1045 goto free_map_tab;
1046 }
1047 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1048 ret = -EACCES;
1049 goto free_map_tab;
1050 }
1051 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1052 switch (map->record->field_mask & (1 << i)) {
1053 case 0:
1054 continue;
1055 case BPF_SPIN_LOCK:
1056 if (map->map_type != BPF_MAP_TYPE_HASH &&
1057 map->map_type != BPF_MAP_TYPE_ARRAY &&
1058 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1059 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1060 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1061 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1062 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1063 ret = -EOPNOTSUPP;
1064 goto free_map_tab;
1065 }
1066 break;
1067 case BPF_TIMER:
1068 if (map->map_type != BPF_MAP_TYPE_HASH &&
1069 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1070 map->map_type != BPF_MAP_TYPE_ARRAY) {
1071 ret = -EOPNOTSUPP;
1072 goto free_map_tab;
1073 }
1074 break;
1075 case BPF_KPTR_UNREF:
1076 case BPF_KPTR_REF:
1077 case BPF_REFCOUNT:
1078 if (map->map_type != BPF_MAP_TYPE_HASH &&
1079 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1080 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1081 map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1082 map->map_type != BPF_MAP_TYPE_ARRAY &&
1083 map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1084 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1085 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1086 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1087 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1088 ret = -EOPNOTSUPP;
1089 goto free_map_tab;
1090 }
1091 break;
1092 case BPF_LIST_HEAD:
1093 case BPF_RB_ROOT:
1094 if (map->map_type != BPF_MAP_TYPE_HASH &&
1095 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1096 map->map_type != BPF_MAP_TYPE_ARRAY) {
1097 ret = -EOPNOTSUPP;
1098 goto free_map_tab;
1099 }
1100 break;
1101 default:
1102 /* Fail if map_type checks are missing for a field type */
1103 ret = -EOPNOTSUPP;
1104 goto free_map_tab;
1105 }
1106 }
1107 }
1108
1109 ret = btf_check_and_fixup_fields(btf, map->record);
1110 if (ret < 0)
1111 goto free_map_tab;
1112
1113 if (map->ops->map_check_btf) {
1114 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1115 if (ret < 0)
1116 goto free_map_tab;
1117 }
1118
1119 return ret;
1120 free_map_tab:
1121 bpf_map_free_record(map);
1122 return ret;
1123 }
1124
1125 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1126 /* called via syscall */
map_create(union bpf_attr * attr)1127 static int map_create(union bpf_attr *attr)
1128 {
1129 const struct bpf_map_ops *ops;
1130 int numa_node = bpf_map_attr_numa_node(attr);
1131 u32 map_type = attr->map_type;
1132 struct bpf_map *map;
1133 int f_flags;
1134 int err;
1135
1136 err = CHECK_ATTR(BPF_MAP_CREATE);
1137 if (err)
1138 return -EINVAL;
1139
1140 if (attr->btf_vmlinux_value_type_id) {
1141 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1142 attr->btf_key_type_id || attr->btf_value_type_id)
1143 return -EINVAL;
1144 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1145 return -EINVAL;
1146 }
1147
1148 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1149 attr->map_extra != 0)
1150 return -EINVAL;
1151
1152 f_flags = bpf_get_file_flag(attr->map_flags);
1153 if (f_flags < 0)
1154 return f_flags;
1155
1156 if (numa_node != NUMA_NO_NODE &&
1157 ((unsigned int)numa_node >= nr_node_ids ||
1158 !node_online(numa_node)))
1159 return -EINVAL;
1160
1161 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1162 map_type = attr->map_type;
1163 if (map_type >= ARRAY_SIZE(bpf_map_types))
1164 return -EINVAL;
1165 map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1166 ops = bpf_map_types[map_type];
1167 if (!ops)
1168 return -EINVAL;
1169
1170 if (ops->map_alloc_check) {
1171 err = ops->map_alloc_check(attr);
1172 if (err)
1173 return err;
1174 }
1175 if (attr->map_ifindex)
1176 ops = &bpf_map_offload_ops;
1177 if (!ops->map_mem_usage)
1178 return -EINVAL;
1179
1180 /* Intent here is for unprivileged_bpf_disabled to block BPF map
1181 * creation for unprivileged users; other actions depend
1182 * on fd availability and access to bpffs, so are dependent on
1183 * object creation success. Even with unprivileged BPF disabled,
1184 * capability checks are still carried out.
1185 */
1186 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
1187 return -EPERM;
1188
1189 /* check privileged map type permissions */
1190 switch (map_type) {
1191 case BPF_MAP_TYPE_ARRAY:
1192 case BPF_MAP_TYPE_PERCPU_ARRAY:
1193 case BPF_MAP_TYPE_PROG_ARRAY:
1194 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1195 case BPF_MAP_TYPE_CGROUP_ARRAY:
1196 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1197 case BPF_MAP_TYPE_HASH:
1198 case BPF_MAP_TYPE_PERCPU_HASH:
1199 case BPF_MAP_TYPE_HASH_OF_MAPS:
1200 case BPF_MAP_TYPE_RINGBUF:
1201 case BPF_MAP_TYPE_USER_RINGBUF:
1202 case BPF_MAP_TYPE_CGROUP_STORAGE:
1203 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1204 /* unprivileged */
1205 break;
1206 case BPF_MAP_TYPE_SK_STORAGE:
1207 case BPF_MAP_TYPE_INODE_STORAGE:
1208 case BPF_MAP_TYPE_TASK_STORAGE:
1209 case BPF_MAP_TYPE_CGRP_STORAGE:
1210 case BPF_MAP_TYPE_BLOOM_FILTER:
1211 case BPF_MAP_TYPE_LPM_TRIE:
1212 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1213 case BPF_MAP_TYPE_STACK_TRACE:
1214 case BPF_MAP_TYPE_QUEUE:
1215 case BPF_MAP_TYPE_STACK:
1216 case BPF_MAP_TYPE_LRU_HASH:
1217 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1218 case BPF_MAP_TYPE_STRUCT_OPS:
1219 case BPF_MAP_TYPE_CPUMAP:
1220 if (!bpf_capable())
1221 return -EPERM;
1222 break;
1223 case BPF_MAP_TYPE_SOCKMAP:
1224 case BPF_MAP_TYPE_SOCKHASH:
1225 case BPF_MAP_TYPE_DEVMAP:
1226 case BPF_MAP_TYPE_DEVMAP_HASH:
1227 case BPF_MAP_TYPE_XSKMAP:
1228 if (!capable(CAP_NET_ADMIN))
1229 return -EPERM;
1230 break;
1231 default:
1232 WARN(1, "unsupported map type %d", map_type);
1233 return -EPERM;
1234 }
1235
1236 map = ops->map_alloc(attr);
1237 if (IS_ERR(map))
1238 return PTR_ERR(map);
1239 map->ops = ops;
1240 map->map_type = map_type;
1241
1242 err = bpf_obj_name_cpy(map->name, attr->map_name,
1243 sizeof(attr->map_name));
1244 if (err < 0)
1245 goto free_map;
1246
1247 atomic64_set(&map->refcnt, 1);
1248 atomic64_set(&map->usercnt, 1);
1249 mutex_init(&map->freeze_mutex);
1250 spin_lock_init(&map->owner.lock);
1251
1252 if (attr->btf_key_type_id || attr->btf_value_type_id ||
1253 /* Even the map's value is a kernel's struct,
1254 * the bpf_prog.o must have BTF to begin with
1255 * to figure out the corresponding kernel's
1256 * counter part. Thus, attr->btf_fd has
1257 * to be valid also.
1258 */
1259 attr->btf_vmlinux_value_type_id) {
1260 struct btf *btf;
1261
1262 btf = btf_get_by_fd(attr->btf_fd);
1263 if (IS_ERR(btf)) {
1264 err = PTR_ERR(btf);
1265 goto free_map;
1266 }
1267 if (btf_is_kernel(btf)) {
1268 btf_put(btf);
1269 err = -EACCES;
1270 goto free_map;
1271 }
1272 map->btf = btf;
1273
1274 if (attr->btf_value_type_id) {
1275 err = map_check_btf(map, btf, attr->btf_key_type_id,
1276 attr->btf_value_type_id);
1277 if (err)
1278 goto free_map;
1279 }
1280
1281 map->btf_key_type_id = attr->btf_key_type_id;
1282 map->btf_value_type_id = attr->btf_value_type_id;
1283 map->btf_vmlinux_value_type_id =
1284 attr->btf_vmlinux_value_type_id;
1285 }
1286
1287 err = security_bpf_map_alloc(map);
1288 if (err)
1289 goto free_map;
1290
1291 err = bpf_map_alloc_id(map);
1292 if (err)
1293 goto free_map_sec;
1294
1295 bpf_map_save_memcg(map);
1296
1297 err = bpf_map_new_fd(map, f_flags);
1298 if (err < 0) {
1299 /* failed to allocate fd.
1300 * bpf_map_put_with_uref() is needed because the above
1301 * bpf_map_alloc_id() has published the map
1302 * to the userspace and the userspace may
1303 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1304 */
1305 bpf_map_put_with_uref(map);
1306 return err;
1307 }
1308
1309 return err;
1310
1311 free_map_sec:
1312 security_bpf_map_free(map);
1313 free_map:
1314 btf_put(map->btf);
1315 map->ops->map_free(map);
1316 return err;
1317 }
1318
1319 /* if error is returned, fd is released.
1320 * On success caller should complete fd access with matching fdput()
1321 */
__bpf_map_get(struct fd f)1322 struct bpf_map *__bpf_map_get(struct fd f)
1323 {
1324 if (!f.file)
1325 return ERR_PTR(-EBADF);
1326 if (f.file->f_op != &bpf_map_fops) {
1327 fdput(f);
1328 return ERR_PTR(-EINVAL);
1329 }
1330
1331 return f.file->private_data;
1332 }
1333
bpf_map_inc(struct bpf_map * map)1334 void bpf_map_inc(struct bpf_map *map)
1335 {
1336 atomic64_inc(&map->refcnt);
1337 }
1338 EXPORT_SYMBOL_GPL(bpf_map_inc);
1339
bpf_map_inc_with_uref(struct bpf_map * map)1340 void bpf_map_inc_with_uref(struct bpf_map *map)
1341 {
1342 atomic64_inc(&map->refcnt);
1343 atomic64_inc(&map->usercnt);
1344 }
1345 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1346
bpf_map_get(u32 ufd)1347 struct bpf_map *bpf_map_get(u32 ufd)
1348 {
1349 struct fd f = fdget(ufd);
1350 struct bpf_map *map;
1351
1352 map = __bpf_map_get(f);
1353 if (IS_ERR(map))
1354 return map;
1355
1356 bpf_map_inc(map);
1357 fdput(f);
1358
1359 return map;
1360 }
1361 EXPORT_SYMBOL(bpf_map_get);
1362
bpf_map_get_with_uref(u32 ufd)1363 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1364 {
1365 struct fd f = fdget(ufd);
1366 struct bpf_map *map;
1367
1368 map = __bpf_map_get(f);
1369 if (IS_ERR(map))
1370 return map;
1371
1372 bpf_map_inc_with_uref(map);
1373 fdput(f);
1374
1375 return map;
1376 }
1377
1378 /* map_idr_lock should have been held or the map should have been
1379 * protected by rcu read lock.
1380 */
__bpf_map_inc_not_zero(struct bpf_map * map,bool uref)1381 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1382 {
1383 int refold;
1384
1385 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1386 if (!refold)
1387 return ERR_PTR(-ENOENT);
1388 if (uref)
1389 atomic64_inc(&map->usercnt);
1390
1391 return map;
1392 }
1393
bpf_map_inc_not_zero(struct bpf_map * map)1394 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1395 {
1396 spin_lock_bh(&map_idr_lock);
1397 map = __bpf_map_inc_not_zero(map, false);
1398 spin_unlock_bh(&map_idr_lock);
1399
1400 return map;
1401 }
1402 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1403
bpf_stackmap_copy(struct bpf_map * map,void * key,void * value)1404 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1405 {
1406 return -ENOTSUPP;
1407 }
1408
__bpf_copy_key(void __user * ukey,u64 key_size)1409 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1410 {
1411 if (key_size)
1412 return vmemdup_user(ukey, key_size);
1413
1414 if (ukey)
1415 return ERR_PTR(-EINVAL);
1416
1417 return NULL;
1418 }
1419
___bpf_copy_key(bpfptr_t ukey,u64 key_size)1420 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1421 {
1422 if (key_size)
1423 return kvmemdup_bpfptr(ukey, key_size);
1424
1425 if (!bpfptr_is_null(ukey))
1426 return ERR_PTR(-EINVAL);
1427
1428 return NULL;
1429 }
1430
1431 /* last field in 'union bpf_attr' used by this command */
1432 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1433
map_lookup_elem(union bpf_attr * attr)1434 static int map_lookup_elem(union bpf_attr *attr)
1435 {
1436 void __user *ukey = u64_to_user_ptr(attr->key);
1437 void __user *uvalue = u64_to_user_ptr(attr->value);
1438 int ufd = attr->map_fd;
1439 struct bpf_map *map;
1440 void *key, *value;
1441 u32 value_size;
1442 struct fd f;
1443 int err;
1444
1445 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1446 return -EINVAL;
1447
1448 if (attr->flags & ~BPF_F_LOCK)
1449 return -EINVAL;
1450
1451 f = fdget(ufd);
1452 map = __bpf_map_get(f);
1453 if (IS_ERR(map))
1454 return PTR_ERR(map);
1455 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1456 err = -EPERM;
1457 goto err_put;
1458 }
1459
1460 if ((attr->flags & BPF_F_LOCK) &&
1461 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1462 err = -EINVAL;
1463 goto err_put;
1464 }
1465
1466 key = __bpf_copy_key(ukey, map->key_size);
1467 if (IS_ERR(key)) {
1468 err = PTR_ERR(key);
1469 goto err_put;
1470 }
1471
1472 value_size = bpf_map_value_size(map);
1473
1474 err = -ENOMEM;
1475 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1476 if (!value)
1477 goto free_key;
1478
1479 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1480 if (copy_from_user(value, uvalue, value_size))
1481 err = -EFAULT;
1482 else
1483 err = bpf_map_copy_value(map, key, value, attr->flags);
1484 goto free_value;
1485 }
1486
1487 err = bpf_map_copy_value(map, key, value, attr->flags);
1488 if (err)
1489 goto free_value;
1490
1491 err = -EFAULT;
1492 if (copy_to_user(uvalue, value, value_size) != 0)
1493 goto free_value;
1494
1495 err = 0;
1496
1497 free_value:
1498 kvfree(value);
1499 free_key:
1500 kvfree(key);
1501 err_put:
1502 fdput(f);
1503 return err;
1504 }
1505
1506
1507 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1508
map_update_elem(union bpf_attr * attr,bpfptr_t uattr)1509 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1510 {
1511 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1512 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1513 int ufd = attr->map_fd;
1514 struct bpf_map *map;
1515 void *key, *value;
1516 u32 value_size;
1517 struct fd f;
1518 int err;
1519
1520 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1521 return -EINVAL;
1522
1523 f = fdget(ufd);
1524 map = __bpf_map_get(f);
1525 if (IS_ERR(map))
1526 return PTR_ERR(map);
1527 bpf_map_write_active_inc(map);
1528 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1529 err = -EPERM;
1530 goto err_put;
1531 }
1532
1533 if ((attr->flags & BPF_F_LOCK) &&
1534 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1535 err = -EINVAL;
1536 goto err_put;
1537 }
1538
1539 key = ___bpf_copy_key(ukey, map->key_size);
1540 if (IS_ERR(key)) {
1541 err = PTR_ERR(key);
1542 goto err_put;
1543 }
1544
1545 value_size = bpf_map_value_size(map);
1546 value = kvmemdup_bpfptr(uvalue, value_size);
1547 if (IS_ERR(value)) {
1548 err = PTR_ERR(value);
1549 goto free_key;
1550 }
1551
1552 err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1553
1554 kvfree(value);
1555 free_key:
1556 kvfree(key);
1557 err_put:
1558 bpf_map_write_active_dec(map);
1559 fdput(f);
1560 return err;
1561 }
1562
1563 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1564
map_delete_elem(union bpf_attr * attr,bpfptr_t uattr)1565 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1566 {
1567 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1568 int ufd = attr->map_fd;
1569 struct bpf_map *map;
1570 struct fd f;
1571 void *key;
1572 int err;
1573
1574 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1575 return -EINVAL;
1576
1577 f = fdget(ufd);
1578 map = __bpf_map_get(f);
1579 if (IS_ERR(map))
1580 return PTR_ERR(map);
1581 bpf_map_write_active_inc(map);
1582 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1583 err = -EPERM;
1584 goto err_put;
1585 }
1586
1587 key = ___bpf_copy_key(ukey, map->key_size);
1588 if (IS_ERR(key)) {
1589 err = PTR_ERR(key);
1590 goto err_put;
1591 }
1592
1593 if (bpf_map_is_offloaded(map)) {
1594 err = bpf_map_offload_delete_elem(map, key);
1595 goto out;
1596 } else if (IS_FD_PROG_ARRAY(map) ||
1597 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1598 /* These maps require sleepable context */
1599 err = map->ops->map_delete_elem(map, key);
1600 goto out;
1601 }
1602
1603 bpf_disable_instrumentation();
1604 rcu_read_lock();
1605 err = map->ops->map_delete_elem(map, key);
1606 rcu_read_unlock();
1607 bpf_enable_instrumentation();
1608 maybe_wait_bpf_programs(map);
1609 out:
1610 kvfree(key);
1611 err_put:
1612 bpf_map_write_active_dec(map);
1613 fdput(f);
1614 return err;
1615 }
1616
1617 /* last field in 'union bpf_attr' used by this command */
1618 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1619
map_get_next_key(union bpf_attr * attr)1620 static int map_get_next_key(union bpf_attr *attr)
1621 {
1622 void __user *ukey = u64_to_user_ptr(attr->key);
1623 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1624 int ufd = attr->map_fd;
1625 struct bpf_map *map;
1626 void *key, *next_key;
1627 struct fd f;
1628 int err;
1629
1630 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1631 return -EINVAL;
1632
1633 f = fdget(ufd);
1634 map = __bpf_map_get(f);
1635 if (IS_ERR(map))
1636 return PTR_ERR(map);
1637 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1638 err = -EPERM;
1639 goto err_put;
1640 }
1641
1642 if (ukey) {
1643 key = __bpf_copy_key(ukey, map->key_size);
1644 if (IS_ERR(key)) {
1645 err = PTR_ERR(key);
1646 goto err_put;
1647 }
1648 } else {
1649 key = NULL;
1650 }
1651
1652 err = -ENOMEM;
1653 next_key = kvmalloc(map->key_size, GFP_USER);
1654 if (!next_key)
1655 goto free_key;
1656
1657 if (bpf_map_is_offloaded(map)) {
1658 err = bpf_map_offload_get_next_key(map, key, next_key);
1659 goto out;
1660 }
1661
1662 rcu_read_lock();
1663 err = map->ops->map_get_next_key(map, key, next_key);
1664 rcu_read_unlock();
1665 out:
1666 if (err)
1667 goto free_next_key;
1668
1669 err = -EFAULT;
1670 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1671 goto free_next_key;
1672
1673 err = 0;
1674
1675 free_next_key:
1676 kvfree(next_key);
1677 free_key:
1678 kvfree(key);
1679 err_put:
1680 fdput(f);
1681 return err;
1682 }
1683
generic_map_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1684 int generic_map_delete_batch(struct bpf_map *map,
1685 const union bpf_attr *attr,
1686 union bpf_attr __user *uattr)
1687 {
1688 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1689 u32 cp, max_count;
1690 int err = 0;
1691 void *key;
1692
1693 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1694 return -EINVAL;
1695
1696 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1697 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1698 return -EINVAL;
1699 }
1700
1701 max_count = attr->batch.count;
1702 if (!max_count)
1703 return 0;
1704
1705 if (put_user(0, &uattr->batch.count))
1706 return -EFAULT;
1707
1708 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1709 if (!key)
1710 return -ENOMEM;
1711
1712 for (cp = 0; cp < max_count; cp++) {
1713 err = -EFAULT;
1714 if (copy_from_user(key, keys + cp * map->key_size,
1715 map->key_size))
1716 break;
1717
1718 if (bpf_map_is_offloaded(map)) {
1719 err = bpf_map_offload_delete_elem(map, key);
1720 break;
1721 }
1722
1723 bpf_disable_instrumentation();
1724 rcu_read_lock();
1725 err = map->ops->map_delete_elem(map, key);
1726 rcu_read_unlock();
1727 bpf_enable_instrumentation();
1728 if (err)
1729 break;
1730 cond_resched();
1731 }
1732 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1733 err = -EFAULT;
1734
1735 kvfree(key);
1736
1737 maybe_wait_bpf_programs(map);
1738 return err;
1739 }
1740
generic_map_update_batch(struct bpf_map * map,struct file * map_file,const union bpf_attr * attr,union bpf_attr __user * uattr)1741 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1742 const union bpf_attr *attr,
1743 union bpf_attr __user *uattr)
1744 {
1745 void __user *values = u64_to_user_ptr(attr->batch.values);
1746 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1747 u32 value_size, cp, max_count;
1748 void *key, *value;
1749 int err = 0;
1750
1751 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1752 return -EINVAL;
1753
1754 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1755 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1756 return -EINVAL;
1757 }
1758
1759 value_size = bpf_map_value_size(map);
1760
1761 max_count = attr->batch.count;
1762 if (!max_count)
1763 return 0;
1764
1765 if (put_user(0, &uattr->batch.count))
1766 return -EFAULT;
1767
1768 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1769 if (!key)
1770 return -ENOMEM;
1771
1772 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1773 if (!value) {
1774 kvfree(key);
1775 return -ENOMEM;
1776 }
1777
1778 for (cp = 0; cp < max_count; cp++) {
1779 err = -EFAULT;
1780 if (copy_from_user(key, keys + cp * map->key_size,
1781 map->key_size) ||
1782 copy_from_user(value, values + cp * value_size, value_size))
1783 break;
1784
1785 err = bpf_map_update_value(map, map_file, key, value,
1786 attr->batch.elem_flags);
1787
1788 if (err)
1789 break;
1790 cond_resched();
1791 }
1792
1793 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1794 err = -EFAULT;
1795
1796 kvfree(value);
1797 kvfree(key);
1798 return err;
1799 }
1800
1801 #define MAP_LOOKUP_RETRIES 3
1802
generic_map_lookup_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1803 int generic_map_lookup_batch(struct bpf_map *map,
1804 const union bpf_attr *attr,
1805 union bpf_attr __user *uattr)
1806 {
1807 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1808 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1809 void __user *values = u64_to_user_ptr(attr->batch.values);
1810 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1811 void *buf, *buf_prevkey, *prev_key, *key, *value;
1812 int err, retry = MAP_LOOKUP_RETRIES;
1813 u32 value_size, cp, max_count;
1814
1815 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1816 return -EINVAL;
1817
1818 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1819 !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1820 return -EINVAL;
1821
1822 value_size = bpf_map_value_size(map);
1823
1824 max_count = attr->batch.count;
1825 if (!max_count)
1826 return 0;
1827
1828 if (put_user(0, &uattr->batch.count))
1829 return -EFAULT;
1830
1831 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1832 if (!buf_prevkey)
1833 return -ENOMEM;
1834
1835 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1836 if (!buf) {
1837 kvfree(buf_prevkey);
1838 return -ENOMEM;
1839 }
1840
1841 err = -EFAULT;
1842 prev_key = NULL;
1843 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1844 goto free_buf;
1845 key = buf;
1846 value = key + map->key_size;
1847 if (ubatch)
1848 prev_key = buf_prevkey;
1849
1850 for (cp = 0; cp < max_count;) {
1851 rcu_read_lock();
1852 err = map->ops->map_get_next_key(map, prev_key, key);
1853 rcu_read_unlock();
1854 if (err)
1855 break;
1856 err = bpf_map_copy_value(map, key, value,
1857 attr->batch.elem_flags);
1858
1859 if (err == -ENOENT) {
1860 if (retry) {
1861 retry--;
1862 continue;
1863 }
1864 err = -EINTR;
1865 break;
1866 }
1867
1868 if (err)
1869 goto free_buf;
1870
1871 if (copy_to_user(keys + cp * map->key_size, key,
1872 map->key_size)) {
1873 err = -EFAULT;
1874 goto free_buf;
1875 }
1876 if (copy_to_user(values + cp * value_size, value, value_size)) {
1877 err = -EFAULT;
1878 goto free_buf;
1879 }
1880
1881 if (!prev_key)
1882 prev_key = buf_prevkey;
1883
1884 swap(prev_key, key);
1885 retry = MAP_LOOKUP_RETRIES;
1886 cp++;
1887 cond_resched();
1888 }
1889
1890 if (err == -EFAULT)
1891 goto free_buf;
1892
1893 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1894 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1895 err = -EFAULT;
1896
1897 free_buf:
1898 kvfree(buf_prevkey);
1899 kvfree(buf);
1900 return err;
1901 }
1902
1903 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1904
map_lookup_and_delete_elem(union bpf_attr * attr)1905 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1906 {
1907 void __user *ukey = u64_to_user_ptr(attr->key);
1908 void __user *uvalue = u64_to_user_ptr(attr->value);
1909 int ufd = attr->map_fd;
1910 struct bpf_map *map;
1911 void *key, *value;
1912 u32 value_size;
1913 struct fd f;
1914 int err;
1915
1916 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1917 return -EINVAL;
1918
1919 if (attr->flags & ~BPF_F_LOCK)
1920 return -EINVAL;
1921
1922 f = fdget(ufd);
1923 map = __bpf_map_get(f);
1924 if (IS_ERR(map))
1925 return PTR_ERR(map);
1926 bpf_map_write_active_inc(map);
1927 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1928 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1929 err = -EPERM;
1930 goto err_put;
1931 }
1932
1933 if (attr->flags &&
1934 (map->map_type == BPF_MAP_TYPE_QUEUE ||
1935 map->map_type == BPF_MAP_TYPE_STACK)) {
1936 err = -EINVAL;
1937 goto err_put;
1938 }
1939
1940 if ((attr->flags & BPF_F_LOCK) &&
1941 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1942 err = -EINVAL;
1943 goto err_put;
1944 }
1945
1946 key = __bpf_copy_key(ukey, map->key_size);
1947 if (IS_ERR(key)) {
1948 err = PTR_ERR(key);
1949 goto err_put;
1950 }
1951
1952 value_size = bpf_map_value_size(map);
1953
1954 err = -ENOMEM;
1955 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1956 if (!value)
1957 goto free_key;
1958
1959 err = -ENOTSUPP;
1960 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1961 map->map_type == BPF_MAP_TYPE_STACK) {
1962 err = map->ops->map_pop_elem(map, value);
1963 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1964 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1965 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1966 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1967 if (!bpf_map_is_offloaded(map)) {
1968 bpf_disable_instrumentation();
1969 rcu_read_lock();
1970 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1971 rcu_read_unlock();
1972 bpf_enable_instrumentation();
1973 }
1974 }
1975
1976 if (err)
1977 goto free_value;
1978
1979 if (copy_to_user(uvalue, value, value_size) != 0) {
1980 err = -EFAULT;
1981 goto free_value;
1982 }
1983
1984 err = 0;
1985
1986 free_value:
1987 kvfree(value);
1988 free_key:
1989 kvfree(key);
1990 err_put:
1991 bpf_map_write_active_dec(map);
1992 fdput(f);
1993 return err;
1994 }
1995
1996 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1997
map_freeze(const union bpf_attr * attr)1998 static int map_freeze(const union bpf_attr *attr)
1999 {
2000 int err = 0, ufd = attr->map_fd;
2001 struct bpf_map *map;
2002 struct fd f;
2003
2004 if (CHECK_ATTR(BPF_MAP_FREEZE))
2005 return -EINVAL;
2006
2007 f = fdget(ufd);
2008 map = __bpf_map_get(f);
2009 if (IS_ERR(map))
2010 return PTR_ERR(map);
2011
2012 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
2013 fdput(f);
2014 return -ENOTSUPP;
2015 }
2016
2017 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2018 fdput(f);
2019 return -EPERM;
2020 }
2021
2022 mutex_lock(&map->freeze_mutex);
2023 if (bpf_map_write_active(map)) {
2024 err = -EBUSY;
2025 goto err_put;
2026 }
2027 if (READ_ONCE(map->frozen)) {
2028 err = -EBUSY;
2029 goto err_put;
2030 }
2031
2032 WRITE_ONCE(map->frozen, true);
2033 err_put:
2034 mutex_unlock(&map->freeze_mutex);
2035 fdput(f);
2036 return err;
2037 }
2038
2039 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2040 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2041 [_id] = & _name ## _prog_ops,
2042 #define BPF_MAP_TYPE(_id, _ops)
2043 #define BPF_LINK_TYPE(_id, _name)
2044 #include <linux/bpf_types.h>
2045 #undef BPF_PROG_TYPE
2046 #undef BPF_MAP_TYPE
2047 #undef BPF_LINK_TYPE
2048 };
2049
find_prog_type(enum bpf_prog_type type,struct bpf_prog * prog)2050 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2051 {
2052 const struct bpf_prog_ops *ops;
2053
2054 if (type >= ARRAY_SIZE(bpf_prog_types))
2055 return -EINVAL;
2056 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2057 ops = bpf_prog_types[type];
2058 if (!ops)
2059 return -EINVAL;
2060
2061 if (!bpf_prog_is_offloaded(prog->aux))
2062 prog->aux->ops = ops;
2063 else
2064 prog->aux->ops = &bpf_offload_prog_ops;
2065 prog->type = type;
2066 return 0;
2067 }
2068
2069 enum bpf_audit {
2070 BPF_AUDIT_LOAD,
2071 BPF_AUDIT_UNLOAD,
2072 BPF_AUDIT_MAX,
2073 };
2074
2075 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2076 [BPF_AUDIT_LOAD] = "LOAD",
2077 [BPF_AUDIT_UNLOAD] = "UNLOAD",
2078 };
2079
bpf_audit_prog(const struct bpf_prog * prog,unsigned int op)2080 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2081 {
2082 struct audit_context *ctx = NULL;
2083 struct audit_buffer *ab;
2084
2085 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2086 return;
2087 if (audit_enabled == AUDIT_OFF)
2088 return;
2089 if (!in_irq() && !irqs_disabled())
2090 ctx = audit_context();
2091 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2092 if (unlikely(!ab))
2093 return;
2094 audit_log_format(ab, "prog-id=%u op=%s",
2095 prog->aux->id, bpf_audit_str[op]);
2096 audit_log_end(ab);
2097 }
2098
bpf_prog_alloc_id(struct bpf_prog * prog)2099 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2100 {
2101 int id;
2102
2103 idr_preload(GFP_KERNEL);
2104 spin_lock_bh(&prog_idr_lock);
2105 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2106 if (id > 0)
2107 prog->aux->id = id;
2108 spin_unlock_bh(&prog_idr_lock);
2109 idr_preload_end();
2110
2111 /* id is in [1, INT_MAX) */
2112 if (WARN_ON_ONCE(!id))
2113 return -ENOSPC;
2114
2115 return id > 0 ? 0 : id;
2116 }
2117
bpf_prog_free_id(struct bpf_prog * prog)2118 void bpf_prog_free_id(struct bpf_prog *prog)
2119 {
2120 unsigned long flags;
2121
2122 /* cBPF to eBPF migrations are currently not in the idr store.
2123 * Offloaded programs are removed from the store when their device
2124 * disappears - even if someone grabs an fd to them they are unusable,
2125 * simply waiting for refcnt to drop to be freed.
2126 */
2127 if (!prog->aux->id)
2128 return;
2129
2130 spin_lock_irqsave(&prog_idr_lock, flags);
2131 idr_remove(&prog_idr, prog->aux->id);
2132 prog->aux->id = 0;
2133 spin_unlock_irqrestore(&prog_idr_lock, flags);
2134 }
2135
__bpf_prog_put_rcu(struct rcu_head * rcu)2136 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2137 {
2138 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2139
2140 kvfree(aux->func_info);
2141 kfree(aux->func_info_aux);
2142 free_uid(aux->user);
2143 security_bpf_prog_free(aux);
2144 bpf_prog_free(aux->prog);
2145 }
2146
__bpf_prog_put_noref(struct bpf_prog * prog,bool deferred)2147 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2148 {
2149 bpf_prog_kallsyms_del_all(prog);
2150 btf_put(prog->aux->btf);
2151 module_put(prog->aux->mod);
2152 kvfree(prog->aux->jited_linfo);
2153 kvfree(prog->aux->linfo);
2154 kfree(prog->aux->kfunc_tab);
2155 if (prog->aux->attach_btf)
2156 btf_put(prog->aux->attach_btf);
2157
2158 if (deferred) {
2159 if (prog->aux->sleepable)
2160 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2161 else
2162 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2163 } else {
2164 __bpf_prog_put_rcu(&prog->aux->rcu);
2165 }
2166 }
2167
bpf_prog_put_deferred(struct work_struct * work)2168 static void bpf_prog_put_deferred(struct work_struct *work)
2169 {
2170 struct bpf_prog_aux *aux;
2171 struct bpf_prog *prog;
2172
2173 aux = container_of(work, struct bpf_prog_aux, work);
2174 prog = aux->prog;
2175 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2176 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2177 bpf_prog_free_id(prog);
2178 __bpf_prog_put_noref(prog, true);
2179 }
2180
__bpf_prog_put(struct bpf_prog * prog)2181 static void __bpf_prog_put(struct bpf_prog *prog)
2182 {
2183 struct bpf_prog_aux *aux = prog->aux;
2184
2185 if (atomic64_dec_and_test(&aux->refcnt)) {
2186 if (in_irq() || irqs_disabled()) {
2187 INIT_WORK(&aux->work, bpf_prog_put_deferred);
2188 schedule_work(&aux->work);
2189 } else {
2190 bpf_prog_put_deferred(&aux->work);
2191 }
2192 }
2193 }
2194
bpf_prog_put(struct bpf_prog * prog)2195 void bpf_prog_put(struct bpf_prog *prog)
2196 {
2197 __bpf_prog_put(prog);
2198 }
2199 EXPORT_SYMBOL_GPL(bpf_prog_put);
2200
bpf_prog_release(struct inode * inode,struct file * filp)2201 static int bpf_prog_release(struct inode *inode, struct file *filp)
2202 {
2203 struct bpf_prog *prog = filp->private_data;
2204
2205 bpf_prog_put(prog);
2206 return 0;
2207 }
2208
2209 struct bpf_prog_kstats {
2210 u64 nsecs;
2211 u64 cnt;
2212 u64 misses;
2213 };
2214
bpf_prog_inc_misses_counter(struct bpf_prog * prog)2215 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2216 {
2217 struct bpf_prog_stats *stats;
2218 unsigned int flags;
2219
2220 stats = this_cpu_ptr(prog->stats);
2221 flags = u64_stats_update_begin_irqsave(&stats->syncp);
2222 u64_stats_inc(&stats->misses);
2223 u64_stats_update_end_irqrestore(&stats->syncp, flags);
2224 }
2225
bpf_prog_get_stats(const struct bpf_prog * prog,struct bpf_prog_kstats * stats)2226 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2227 struct bpf_prog_kstats *stats)
2228 {
2229 u64 nsecs = 0, cnt = 0, misses = 0;
2230 int cpu;
2231
2232 for_each_possible_cpu(cpu) {
2233 const struct bpf_prog_stats *st;
2234 unsigned int start;
2235 u64 tnsecs, tcnt, tmisses;
2236
2237 st = per_cpu_ptr(prog->stats, cpu);
2238 do {
2239 start = u64_stats_fetch_begin(&st->syncp);
2240 tnsecs = u64_stats_read(&st->nsecs);
2241 tcnt = u64_stats_read(&st->cnt);
2242 tmisses = u64_stats_read(&st->misses);
2243 } while (u64_stats_fetch_retry(&st->syncp, start));
2244 nsecs += tnsecs;
2245 cnt += tcnt;
2246 misses += tmisses;
2247 }
2248 stats->nsecs = nsecs;
2249 stats->cnt = cnt;
2250 stats->misses = misses;
2251 }
2252
2253 #ifdef CONFIG_PROC_FS
bpf_prog_show_fdinfo(struct seq_file * m,struct file * filp)2254 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2255 {
2256 const struct bpf_prog *prog = filp->private_data;
2257 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2258 struct bpf_prog_kstats stats;
2259
2260 bpf_prog_get_stats(prog, &stats);
2261 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2262 seq_printf(m,
2263 "prog_type:\t%u\n"
2264 "prog_jited:\t%u\n"
2265 "prog_tag:\t%s\n"
2266 "memlock:\t%llu\n"
2267 "prog_id:\t%u\n"
2268 "run_time_ns:\t%llu\n"
2269 "run_cnt:\t%llu\n"
2270 "recursion_misses:\t%llu\n"
2271 "verified_insns:\t%u\n",
2272 prog->type,
2273 prog->jited,
2274 prog_tag,
2275 prog->pages * 1ULL << PAGE_SHIFT,
2276 prog->aux->id,
2277 stats.nsecs,
2278 stats.cnt,
2279 stats.misses,
2280 prog->aux->verified_insns);
2281 }
2282 #endif
2283
2284 const struct file_operations bpf_prog_fops = {
2285 #ifdef CONFIG_PROC_FS
2286 .show_fdinfo = bpf_prog_show_fdinfo,
2287 #endif
2288 .release = bpf_prog_release,
2289 .read = bpf_dummy_read,
2290 .write = bpf_dummy_write,
2291 };
2292
bpf_prog_new_fd(struct bpf_prog * prog)2293 int bpf_prog_new_fd(struct bpf_prog *prog)
2294 {
2295 int ret;
2296
2297 ret = security_bpf_prog(prog);
2298 if (ret < 0)
2299 return ret;
2300
2301 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2302 O_RDWR | O_CLOEXEC);
2303 }
2304
____bpf_prog_get(struct fd f)2305 static struct bpf_prog *____bpf_prog_get(struct fd f)
2306 {
2307 if (!f.file)
2308 return ERR_PTR(-EBADF);
2309 if (f.file->f_op != &bpf_prog_fops) {
2310 fdput(f);
2311 return ERR_PTR(-EINVAL);
2312 }
2313
2314 return f.file->private_data;
2315 }
2316
bpf_prog_add(struct bpf_prog * prog,int i)2317 void bpf_prog_add(struct bpf_prog *prog, int i)
2318 {
2319 atomic64_add(i, &prog->aux->refcnt);
2320 }
2321 EXPORT_SYMBOL_GPL(bpf_prog_add);
2322
bpf_prog_sub(struct bpf_prog * prog,int i)2323 void bpf_prog_sub(struct bpf_prog *prog, int i)
2324 {
2325 /* Only to be used for undoing previous bpf_prog_add() in some
2326 * error path. We still know that another entity in our call
2327 * path holds a reference to the program, thus atomic_sub() can
2328 * be safely used in such cases!
2329 */
2330 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2331 }
2332 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2333
bpf_prog_inc(struct bpf_prog * prog)2334 void bpf_prog_inc(struct bpf_prog *prog)
2335 {
2336 atomic64_inc(&prog->aux->refcnt);
2337 }
2338 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2339
2340 /* prog_idr_lock should have been held */
bpf_prog_inc_not_zero(struct bpf_prog * prog)2341 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2342 {
2343 int refold;
2344
2345 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2346
2347 if (!refold)
2348 return ERR_PTR(-ENOENT);
2349
2350 return prog;
2351 }
2352 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2353
bpf_prog_get_ok(struct bpf_prog * prog,enum bpf_prog_type * attach_type,bool attach_drv)2354 bool bpf_prog_get_ok(struct bpf_prog *prog,
2355 enum bpf_prog_type *attach_type, bool attach_drv)
2356 {
2357 /* not an attachment, just a refcount inc, always allow */
2358 if (!attach_type)
2359 return true;
2360
2361 if (prog->type != *attach_type)
2362 return false;
2363 if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2364 return false;
2365
2366 return true;
2367 }
2368
__bpf_prog_get(u32 ufd,enum bpf_prog_type * attach_type,bool attach_drv)2369 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2370 bool attach_drv)
2371 {
2372 struct fd f = fdget(ufd);
2373 struct bpf_prog *prog;
2374
2375 prog = ____bpf_prog_get(f);
2376 if (IS_ERR(prog))
2377 return prog;
2378 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2379 prog = ERR_PTR(-EINVAL);
2380 goto out;
2381 }
2382
2383 bpf_prog_inc(prog);
2384 out:
2385 fdput(f);
2386 return prog;
2387 }
2388
bpf_prog_get(u32 ufd)2389 struct bpf_prog *bpf_prog_get(u32 ufd)
2390 {
2391 return __bpf_prog_get(ufd, NULL, false);
2392 }
2393
bpf_prog_get_type_dev(u32 ufd,enum bpf_prog_type type,bool attach_drv)2394 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2395 bool attach_drv)
2396 {
2397 return __bpf_prog_get(ufd, &type, attach_drv);
2398 }
2399 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2400
2401 /* Initially all BPF programs could be loaded w/o specifying
2402 * expected_attach_type. Later for some of them specifying expected_attach_type
2403 * at load time became required so that program could be validated properly.
2404 * Programs of types that are allowed to be loaded both w/ and w/o (for
2405 * backward compatibility) expected_attach_type, should have the default attach
2406 * type assigned to expected_attach_type for the latter case, so that it can be
2407 * validated later at attach time.
2408 *
2409 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2410 * prog type requires it but has some attach types that have to be backward
2411 * compatible.
2412 */
bpf_prog_load_fixup_attach_type(union bpf_attr * attr)2413 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2414 {
2415 switch (attr->prog_type) {
2416 case BPF_PROG_TYPE_CGROUP_SOCK:
2417 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2418 * exist so checking for non-zero is the way to go here.
2419 */
2420 if (!attr->expected_attach_type)
2421 attr->expected_attach_type =
2422 BPF_CGROUP_INET_SOCK_CREATE;
2423 break;
2424 case BPF_PROG_TYPE_SK_REUSEPORT:
2425 if (!attr->expected_attach_type)
2426 attr->expected_attach_type =
2427 BPF_SK_REUSEPORT_SELECT;
2428 break;
2429 }
2430 }
2431
2432 static int
bpf_prog_load_check_attach(enum bpf_prog_type prog_type,enum bpf_attach_type expected_attach_type,struct btf * attach_btf,u32 btf_id,struct bpf_prog * dst_prog)2433 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2434 enum bpf_attach_type expected_attach_type,
2435 struct btf *attach_btf, u32 btf_id,
2436 struct bpf_prog *dst_prog)
2437 {
2438 if (btf_id) {
2439 if (btf_id > BTF_MAX_TYPE)
2440 return -EINVAL;
2441
2442 if (!attach_btf && !dst_prog)
2443 return -EINVAL;
2444
2445 switch (prog_type) {
2446 case BPF_PROG_TYPE_TRACING:
2447 case BPF_PROG_TYPE_LSM:
2448 case BPF_PROG_TYPE_STRUCT_OPS:
2449 case BPF_PROG_TYPE_EXT:
2450 break;
2451 default:
2452 return -EINVAL;
2453 }
2454 }
2455
2456 if (attach_btf && (!btf_id || dst_prog))
2457 return -EINVAL;
2458
2459 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2460 prog_type != BPF_PROG_TYPE_EXT)
2461 return -EINVAL;
2462
2463 switch (prog_type) {
2464 case BPF_PROG_TYPE_CGROUP_SOCK:
2465 switch (expected_attach_type) {
2466 case BPF_CGROUP_INET_SOCK_CREATE:
2467 case BPF_CGROUP_INET_SOCK_RELEASE:
2468 case BPF_CGROUP_INET4_POST_BIND:
2469 case BPF_CGROUP_INET6_POST_BIND:
2470 return 0;
2471 default:
2472 return -EINVAL;
2473 }
2474 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2475 switch (expected_attach_type) {
2476 case BPF_CGROUP_INET4_BIND:
2477 case BPF_CGROUP_INET6_BIND:
2478 case BPF_CGROUP_INET4_CONNECT:
2479 case BPF_CGROUP_INET6_CONNECT:
2480 case BPF_CGROUP_INET4_GETPEERNAME:
2481 case BPF_CGROUP_INET6_GETPEERNAME:
2482 case BPF_CGROUP_INET4_GETSOCKNAME:
2483 case BPF_CGROUP_INET6_GETSOCKNAME:
2484 case BPF_CGROUP_UDP4_SENDMSG:
2485 case BPF_CGROUP_UDP6_SENDMSG:
2486 case BPF_CGROUP_UDP4_RECVMSG:
2487 case BPF_CGROUP_UDP6_RECVMSG:
2488 return 0;
2489 default:
2490 return -EINVAL;
2491 }
2492 case BPF_PROG_TYPE_CGROUP_SKB:
2493 switch (expected_attach_type) {
2494 case BPF_CGROUP_INET_INGRESS:
2495 case BPF_CGROUP_INET_EGRESS:
2496 return 0;
2497 default:
2498 return -EINVAL;
2499 }
2500 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2501 switch (expected_attach_type) {
2502 case BPF_CGROUP_SETSOCKOPT:
2503 case BPF_CGROUP_GETSOCKOPT:
2504 return 0;
2505 default:
2506 return -EINVAL;
2507 }
2508 case BPF_PROG_TYPE_SK_LOOKUP:
2509 if (expected_attach_type == BPF_SK_LOOKUP)
2510 return 0;
2511 return -EINVAL;
2512 case BPF_PROG_TYPE_SK_REUSEPORT:
2513 switch (expected_attach_type) {
2514 case BPF_SK_REUSEPORT_SELECT:
2515 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2516 return 0;
2517 default:
2518 return -EINVAL;
2519 }
2520 case BPF_PROG_TYPE_NETFILTER:
2521 if (expected_attach_type == BPF_NETFILTER)
2522 return 0;
2523 return -EINVAL;
2524 case BPF_PROG_TYPE_SYSCALL:
2525 case BPF_PROG_TYPE_EXT:
2526 if (expected_attach_type)
2527 return -EINVAL;
2528 fallthrough;
2529 default:
2530 return 0;
2531 }
2532 }
2533
is_net_admin_prog_type(enum bpf_prog_type prog_type)2534 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2535 {
2536 switch (prog_type) {
2537 case BPF_PROG_TYPE_SCHED_CLS:
2538 case BPF_PROG_TYPE_SCHED_ACT:
2539 case BPF_PROG_TYPE_XDP:
2540 case BPF_PROG_TYPE_LWT_IN:
2541 case BPF_PROG_TYPE_LWT_OUT:
2542 case BPF_PROG_TYPE_LWT_XMIT:
2543 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2544 case BPF_PROG_TYPE_SK_SKB:
2545 case BPF_PROG_TYPE_SK_MSG:
2546 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2547 case BPF_PROG_TYPE_CGROUP_DEVICE:
2548 case BPF_PROG_TYPE_CGROUP_SOCK:
2549 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2550 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2551 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2552 case BPF_PROG_TYPE_SOCK_OPS:
2553 case BPF_PROG_TYPE_EXT: /* extends any prog */
2554 case BPF_PROG_TYPE_NETFILTER:
2555 return true;
2556 case BPF_PROG_TYPE_CGROUP_SKB:
2557 /* always unpriv */
2558 case BPF_PROG_TYPE_SK_REUSEPORT:
2559 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2560 default:
2561 return false;
2562 }
2563 }
2564
is_perfmon_prog_type(enum bpf_prog_type prog_type)2565 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2566 {
2567 switch (prog_type) {
2568 case BPF_PROG_TYPE_KPROBE:
2569 case BPF_PROG_TYPE_TRACEPOINT:
2570 case BPF_PROG_TYPE_PERF_EVENT:
2571 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2572 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2573 case BPF_PROG_TYPE_TRACING:
2574 case BPF_PROG_TYPE_LSM:
2575 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2576 case BPF_PROG_TYPE_EXT: /* extends any prog */
2577 return true;
2578 default:
2579 return false;
2580 }
2581 }
2582
2583 /* last field in 'union bpf_attr' used by this command */
2584 #define BPF_PROG_LOAD_LAST_FIELD log_true_size
2585
bpf_prog_load(union bpf_attr * attr,bpfptr_t uattr,u32 uattr_size)2586 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2587 {
2588 enum bpf_prog_type type = attr->prog_type;
2589 struct bpf_prog *prog, *dst_prog = NULL;
2590 struct btf *attach_btf = NULL;
2591 int err;
2592 char license[128];
2593
2594 if (CHECK_ATTR(BPF_PROG_LOAD))
2595 return -EINVAL;
2596
2597 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2598 BPF_F_ANY_ALIGNMENT |
2599 BPF_F_TEST_STATE_FREQ |
2600 BPF_F_SLEEPABLE |
2601 BPF_F_TEST_RND_HI32 |
2602 BPF_F_XDP_HAS_FRAGS |
2603 BPF_F_XDP_DEV_BOUND_ONLY))
2604 return -EINVAL;
2605
2606 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2607 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2608 !bpf_capable())
2609 return -EPERM;
2610
2611 /* Intent here is for unprivileged_bpf_disabled to block BPF program
2612 * creation for unprivileged users; other actions depend
2613 * on fd availability and access to bpffs, so are dependent on
2614 * object creation success. Even with unprivileged BPF disabled,
2615 * capability checks are still carried out for these
2616 * and other operations.
2617 */
2618 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
2619 return -EPERM;
2620
2621 if (attr->insn_cnt == 0 ||
2622 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2623 return -E2BIG;
2624 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2625 type != BPF_PROG_TYPE_CGROUP_SKB &&
2626 !bpf_capable())
2627 return -EPERM;
2628
2629 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2630 return -EPERM;
2631 if (is_perfmon_prog_type(type) && !perfmon_capable())
2632 return -EPERM;
2633
2634 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2635 * or btf, we need to check which one it is
2636 */
2637 if (attr->attach_prog_fd) {
2638 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2639 if (IS_ERR(dst_prog)) {
2640 dst_prog = NULL;
2641 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2642 if (IS_ERR(attach_btf))
2643 return -EINVAL;
2644 if (!btf_is_kernel(attach_btf)) {
2645 /* attaching through specifying bpf_prog's BTF
2646 * objects directly might be supported eventually
2647 */
2648 btf_put(attach_btf);
2649 return -ENOTSUPP;
2650 }
2651 }
2652 } else if (attr->attach_btf_id) {
2653 /* fall back to vmlinux BTF, if BTF type ID is specified */
2654 attach_btf = bpf_get_btf_vmlinux();
2655 if (IS_ERR(attach_btf))
2656 return PTR_ERR(attach_btf);
2657 if (!attach_btf)
2658 return -EINVAL;
2659 btf_get(attach_btf);
2660 }
2661
2662 bpf_prog_load_fixup_attach_type(attr);
2663 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2664 attach_btf, attr->attach_btf_id,
2665 dst_prog)) {
2666 if (dst_prog)
2667 bpf_prog_put(dst_prog);
2668 if (attach_btf)
2669 btf_put(attach_btf);
2670 return -EINVAL;
2671 }
2672
2673 /* plain bpf_prog allocation */
2674 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2675 if (!prog) {
2676 if (dst_prog)
2677 bpf_prog_put(dst_prog);
2678 if (attach_btf)
2679 btf_put(attach_btf);
2680 return -ENOMEM;
2681 }
2682
2683 prog->expected_attach_type = attr->expected_attach_type;
2684 prog->aux->attach_btf = attach_btf;
2685 prog->aux->attach_btf_id = attr->attach_btf_id;
2686 prog->aux->dst_prog = dst_prog;
2687 prog->aux->dev_bound = !!attr->prog_ifindex;
2688 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2689 prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2690
2691 err = security_bpf_prog_alloc(prog->aux);
2692 if (err)
2693 goto free_prog;
2694
2695 prog->aux->user = get_current_user();
2696 prog->len = attr->insn_cnt;
2697
2698 err = -EFAULT;
2699 if (copy_from_bpfptr(prog->insns,
2700 make_bpfptr(attr->insns, uattr.is_kernel),
2701 bpf_prog_insn_size(prog)) != 0)
2702 goto free_prog_sec;
2703 /* copy eBPF program license from user space */
2704 if (strncpy_from_bpfptr(license,
2705 make_bpfptr(attr->license, uattr.is_kernel),
2706 sizeof(license) - 1) < 0)
2707 goto free_prog_sec;
2708 license[sizeof(license) - 1] = 0;
2709
2710 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2711 prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2712
2713 prog->orig_prog = NULL;
2714 prog->jited = 0;
2715
2716 atomic64_set(&prog->aux->refcnt, 1);
2717
2718 if (bpf_prog_is_dev_bound(prog->aux)) {
2719 err = bpf_prog_dev_bound_init(prog, attr);
2720 if (err)
2721 goto free_prog_sec;
2722 }
2723
2724 if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2725 bpf_prog_is_dev_bound(dst_prog->aux)) {
2726 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2727 if (err)
2728 goto free_prog_sec;
2729 }
2730
2731 /* find program type: socket_filter vs tracing_filter */
2732 err = find_prog_type(type, prog);
2733 if (err < 0)
2734 goto free_prog_sec;
2735
2736 prog->aux->load_time = ktime_get_boottime_ns();
2737 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2738 sizeof(attr->prog_name));
2739 if (err < 0)
2740 goto free_prog_sec;
2741
2742 /* run eBPF verifier */
2743 err = bpf_check(&prog, attr, uattr, uattr_size);
2744 if (err < 0)
2745 goto free_used_maps;
2746
2747 prog = bpf_prog_select_runtime(prog, &err);
2748 if (err < 0)
2749 goto free_used_maps;
2750
2751 err = bpf_prog_alloc_id(prog);
2752 if (err)
2753 goto free_used_maps;
2754
2755 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2756 * effectively publicly exposed. However, retrieving via
2757 * bpf_prog_get_fd_by_id() will take another reference,
2758 * therefore it cannot be gone underneath us.
2759 *
2760 * Only for the time /after/ successful bpf_prog_new_fd()
2761 * and before returning to userspace, we might just hold
2762 * one reference and any parallel close on that fd could
2763 * rip everything out. Hence, below notifications must
2764 * happen before bpf_prog_new_fd().
2765 *
2766 * Also, any failure handling from this point onwards must
2767 * be using bpf_prog_put() given the program is exposed.
2768 */
2769 bpf_prog_kallsyms_add(prog);
2770 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2771 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2772
2773 err = bpf_prog_new_fd(prog);
2774 if (err < 0)
2775 bpf_prog_put(prog);
2776 return err;
2777
2778 free_used_maps:
2779 /* In case we have subprogs, we need to wait for a grace
2780 * period before we can tear down JIT memory since symbols
2781 * are already exposed under kallsyms.
2782 */
2783 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2784 return err;
2785 free_prog_sec:
2786 free_uid(prog->aux->user);
2787 security_bpf_prog_free(prog->aux);
2788 free_prog:
2789 if (prog->aux->attach_btf)
2790 btf_put(prog->aux->attach_btf);
2791 bpf_prog_free(prog);
2792 return err;
2793 }
2794
2795 #define BPF_OBJ_LAST_FIELD path_fd
2796
bpf_obj_pin(const union bpf_attr * attr)2797 static int bpf_obj_pin(const union bpf_attr *attr)
2798 {
2799 int path_fd;
2800
2801 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2802 return -EINVAL;
2803
2804 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2805 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2806 return -EINVAL;
2807
2808 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2809 return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2810 u64_to_user_ptr(attr->pathname));
2811 }
2812
bpf_obj_get(const union bpf_attr * attr)2813 static int bpf_obj_get(const union bpf_attr *attr)
2814 {
2815 int path_fd;
2816
2817 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2818 attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2819 return -EINVAL;
2820
2821 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2822 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2823 return -EINVAL;
2824
2825 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2826 return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2827 attr->file_flags);
2828 }
2829
bpf_link_init(struct bpf_link * link,enum bpf_link_type type,const struct bpf_link_ops * ops,struct bpf_prog * prog)2830 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2831 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2832 {
2833 WARN_ON(ops->dealloc && ops->dealloc_deferred);
2834 atomic64_set(&link->refcnt, 1);
2835 link->type = type;
2836 link->id = 0;
2837 link->ops = ops;
2838 link->prog = prog;
2839 }
2840
bpf_link_free_id(int id)2841 static void bpf_link_free_id(int id)
2842 {
2843 if (!id)
2844 return;
2845
2846 spin_lock_bh(&link_idr_lock);
2847 idr_remove(&link_idr, id);
2848 spin_unlock_bh(&link_idr_lock);
2849 }
2850
2851 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2852 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2853 * anon_inode's release() call. This helper marks bpf_link as
2854 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2855 * is not decremented, it's the responsibility of a calling code that failed
2856 * to complete bpf_link initialization.
2857 * This helper eventually calls link's dealloc callback, but does not call
2858 * link's release callback.
2859 */
bpf_link_cleanup(struct bpf_link_primer * primer)2860 void bpf_link_cleanup(struct bpf_link_primer *primer)
2861 {
2862 primer->link->prog = NULL;
2863 bpf_link_free_id(primer->id);
2864 fput(primer->file);
2865 put_unused_fd(primer->fd);
2866 }
2867
bpf_link_inc(struct bpf_link * link)2868 void bpf_link_inc(struct bpf_link *link)
2869 {
2870 atomic64_inc(&link->refcnt);
2871 }
2872
bpf_link_defer_dealloc_rcu_gp(struct rcu_head * rcu)2873 static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
2874 {
2875 struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
2876
2877 /* free bpf_link and its containing memory */
2878 link->ops->dealloc_deferred(link);
2879 }
2880
bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head * rcu)2881 static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
2882 {
2883 if (rcu_trace_implies_rcu_gp())
2884 bpf_link_defer_dealloc_rcu_gp(rcu);
2885 else
2886 call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
2887 }
2888
2889 /* bpf_link_free is guaranteed to be called from process context */
bpf_link_free(struct bpf_link * link)2890 static void bpf_link_free(struct bpf_link *link)
2891 {
2892 const struct bpf_link_ops *ops = link->ops;
2893 bool sleepable = false;
2894
2895 bpf_link_free_id(link->id);
2896 if (link->prog) {
2897 sleepable = link->prog->aux->sleepable;
2898 /* detach BPF program, clean up used resources */
2899 ops->release(link);
2900 bpf_prog_put(link->prog);
2901 }
2902 if (ops->dealloc_deferred) {
2903 /* schedule BPF link deallocation; if underlying BPF program
2904 * is sleepable, we need to first wait for RCU tasks trace
2905 * sync, then go through "classic" RCU grace period
2906 */
2907 if (sleepable)
2908 call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
2909 else
2910 call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
2911 } else if (ops->dealloc)
2912 ops->dealloc(link);
2913 }
2914
bpf_link_put_deferred(struct work_struct * work)2915 static void bpf_link_put_deferred(struct work_struct *work)
2916 {
2917 struct bpf_link *link = container_of(work, struct bpf_link, work);
2918
2919 bpf_link_free(link);
2920 }
2921
2922 /* bpf_link_put might be called from atomic context. It needs to be called
2923 * from sleepable context in order to acquire sleeping locks during the process.
2924 */
bpf_link_put(struct bpf_link * link)2925 void bpf_link_put(struct bpf_link *link)
2926 {
2927 if (!atomic64_dec_and_test(&link->refcnt))
2928 return;
2929
2930 INIT_WORK(&link->work, bpf_link_put_deferred);
2931 schedule_work(&link->work);
2932 }
2933 EXPORT_SYMBOL(bpf_link_put);
2934
bpf_link_put_direct(struct bpf_link * link)2935 static void bpf_link_put_direct(struct bpf_link *link)
2936 {
2937 if (!atomic64_dec_and_test(&link->refcnt))
2938 return;
2939 bpf_link_free(link);
2940 }
2941
bpf_link_release(struct inode * inode,struct file * filp)2942 static int bpf_link_release(struct inode *inode, struct file *filp)
2943 {
2944 struct bpf_link *link = filp->private_data;
2945
2946 bpf_link_put_direct(link);
2947 return 0;
2948 }
2949
2950 #ifdef CONFIG_PROC_FS
2951 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2952 #define BPF_MAP_TYPE(_id, _ops)
2953 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2954 static const char *bpf_link_type_strs[] = {
2955 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2956 #include <linux/bpf_types.h>
2957 };
2958 #undef BPF_PROG_TYPE
2959 #undef BPF_MAP_TYPE
2960 #undef BPF_LINK_TYPE
2961
bpf_link_show_fdinfo(struct seq_file * m,struct file * filp)2962 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2963 {
2964 const struct bpf_link *link = filp->private_data;
2965 const struct bpf_prog *prog = link->prog;
2966 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2967
2968 seq_printf(m,
2969 "link_type:\t%s\n"
2970 "link_id:\t%u\n",
2971 bpf_link_type_strs[link->type],
2972 link->id);
2973 if (prog) {
2974 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2975 seq_printf(m,
2976 "prog_tag:\t%s\n"
2977 "prog_id:\t%u\n",
2978 prog_tag,
2979 prog->aux->id);
2980 }
2981 if (link->ops->show_fdinfo)
2982 link->ops->show_fdinfo(link, m);
2983 }
2984 #endif
2985
2986 static const struct file_operations bpf_link_fops = {
2987 #ifdef CONFIG_PROC_FS
2988 .show_fdinfo = bpf_link_show_fdinfo,
2989 #endif
2990 .release = bpf_link_release,
2991 .read = bpf_dummy_read,
2992 .write = bpf_dummy_write,
2993 };
2994
bpf_link_alloc_id(struct bpf_link * link)2995 static int bpf_link_alloc_id(struct bpf_link *link)
2996 {
2997 int id;
2998
2999 idr_preload(GFP_KERNEL);
3000 spin_lock_bh(&link_idr_lock);
3001 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3002 spin_unlock_bh(&link_idr_lock);
3003 idr_preload_end();
3004
3005 return id;
3006 }
3007
3008 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3009 * reserving unused FD and allocating ID from link_idr. This is to be paired
3010 * with bpf_link_settle() to install FD and ID and expose bpf_link to
3011 * user-space, if bpf_link is successfully attached. If not, bpf_link and
3012 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3013 * transient state is passed around in struct bpf_link_primer.
3014 * This is preferred way to create and initialize bpf_link, especially when
3015 * there are complicated and expensive operations in between creating bpf_link
3016 * itself and attaching it to BPF hook. By using bpf_link_prime() and
3017 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3018 * expensive (and potentially failing) roll back operations in a rare case
3019 * that file, FD, or ID can't be allocated.
3020 */
bpf_link_prime(struct bpf_link * link,struct bpf_link_primer * primer)3021 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3022 {
3023 struct file *file;
3024 int fd, id;
3025
3026 fd = get_unused_fd_flags(O_CLOEXEC);
3027 if (fd < 0)
3028 return fd;
3029
3030
3031 id = bpf_link_alloc_id(link);
3032 if (id < 0) {
3033 put_unused_fd(fd);
3034 return id;
3035 }
3036
3037 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
3038 if (IS_ERR(file)) {
3039 bpf_link_free_id(id);
3040 put_unused_fd(fd);
3041 return PTR_ERR(file);
3042 }
3043
3044 primer->link = link;
3045 primer->file = file;
3046 primer->fd = fd;
3047 primer->id = id;
3048 return 0;
3049 }
3050
bpf_link_settle(struct bpf_link_primer * primer)3051 int bpf_link_settle(struct bpf_link_primer *primer)
3052 {
3053 /* make bpf_link fetchable by ID */
3054 spin_lock_bh(&link_idr_lock);
3055 primer->link->id = primer->id;
3056 spin_unlock_bh(&link_idr_lock);
3057 /* make bpf_link fetchable by FD */
3058 fd_install(primer->fd, primer->file);
3059 /* pass through installed FD */
3060 return primer->fd;
3061 }
3062
bpf_link_new_fd(struct bpf_link * link)3063 int bpf_link_new_fd(struct bpf_link *link)
3064 {
3065 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
3066 }
3067
bpf_link_get_from_fd(u32 ufd)3068 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3069 {
3070 struct fd f = fdget(ufd);
3071 struct bpf_link *link;
3072
3073 if (!f.file)
3074 return ERR_PTR(-EBADF);
3075 if (f.file->f_op != &bpf_link_fops) {
3076 fdput(f);
3077 return ERR_PTR(-EINVAL);
3078 }
3079
3080 link = f.file->private_data;
3081 bpf_link_inc(link);
3082 fdput(f);
3083
3084 return link;
3085 }
3086 EXPORT_SYMBOL(bpf_link_get_from_fd);
3087
bpf_tracing_link_release(struct bpf_link * link)3088 static void bpf_tracing_link_release(struct bpf_link *link)
3089 {
3090 struct bpf_tracing_link *tr_link =
3091 container_of(link, struct bpf_tracing_link, link.link);
3092
3093 WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3094 tr_link->trampoline));
3095
3096 bpf_trampoline_put(tr_link->trampoline);
3097
3098 /* tgt_prog is NULL if target is a kernel function */
3099 if (tr_link->tgt_prog)
3100 bpf_prog_put(tr_link->tgt_prog);
3101 }
3102
bpf_tracing_link_dealloc(struct bpf_link * link)3103 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3104 {
3105 struct bpf_tracing_link *tr_link =
3106 container_of(link, struct bpf_tracing_link, link.link);
3107
3108 kfree(tr_link);
3109 }
3110
bpf_tracing_link_show_fdinfo(const struct bpf_link * link,struct seq_file * seq)3111 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3112 struct seq_file *seq)
3113 {
3114 struct bpf_tracing_link *tr_link =
3115 container_of(link, struct bpf_tracing_link, link.link);
3116 u32 target_btf_id, target_obj_id;
3117
3118 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3119 &target_obj_id, &target_btf_id);
3120 seq_printf(seq,
3121 "attach_type:\t%d\n"
3122 "target_obj_id:\t%u\n"
3123 "target_btf_id:\t%u\n",
3124 tr_link->attach_type,
3125 target_obj_id,
3126 target_btf_id);
3127 }
3128
bpf_tracing_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)3129 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3130 struct bpf_link_info *info)
3131 {
3132 struct bpf_tracing_link *tr_link =
3133 container_of(link, struct bpf_tracing_link, link.link);
3134
3135 info->tracing.attach_type = tr_link->attach_type;
3136 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3137 &info->tracing.target_obj_id,
3138 &info->tracing.target_btf_id);
3139
3140 return 0;
3141 }
3142
3143 static const struct bpf_link_ops bpf_tracing_link_lops = {
3144 .release = bpf_tracing_link_release,
3145 .dealloc = bpf_tracing_link_dealloc,
3146 .show_fdinfo = bpf_tracing_link_show_fdinfo,
3147 .fill_link_info = bpf_tracing_link_fill_link_info,
3148 };
3149
bpf_tracing_prog_attach(struct bpf_prog * prog,int tgt_prog_fd,u32 btf_id,u64 bpf_cookie)3150 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3151 int tgt_prog_fd,
3152 u32 btf_id,
3153 u64 bpf_cookie)
3154 {
3155 struct bpf_link_primer link_primer;
3156 struct bpf_prog *tgt_prog = NULL;
3157 struct bpf_trampoline *tr = NULL;
3158 struct bpf_tracing_link *link;
3159 u64 key = 0;
3160 int err;
3161
3162 switch (prog->type) {
3163 case BPF_PROG_TYPE_TRACING:
3164 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3165 prog->expected_attach_type != BPF_TRACE_FEXIT &&
3166 prog->expected_attach_type != BPF_MODIFY_RETURN) {
3167 err = -EINVAL;
3168 goto out_put_prog;
3169 }
3170 break;
3171 case BPF_PROG_TYPE_EXT:
3172 if (prog->expected_attach_type != 0) {
3173 err = -EINVAL;
3174 goto out_put_prog;
3175 }
3176 break;
3177 case BPF_PROG_TYPE_LSM:
3178 if (prog->expected_attach_type != BPF_LSM_MAC) {
3179 err = -EINVAL;
3180 goto out_put_prog;
3181 }
3182 break;
3183 default:
3184 err = -EINVAL;
3185 goto out_put_prog;
3186 }
3187
3188 if (!!tgt_prog_fd != !!btf_id) {
3189 err = -EINVAL;
3190 goto out_put_prog;
3191 }
3192
3193 if (tgt_prog_fd) {
3194 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3195 if (prog->type != BPF_PROG_TYPE_EXT) {
3196 err = -EINVAL;
3197 goto out_put_prog;
3198 }
3199
3200 tgt_prog = bpf_prog_get(tgt_prog_fd);
3201 if (IS_ERR(tgt_prog)) {
3202 err = PTR_ERR(tgt_prog);
3203 tgt_prog = NULL;
3204 goto out_put_prog;
3205 }
3206
3207 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3208 }
3209
3210 link = kzalloc(sizeof(*link), GFP_USER);
3211 if (!link) {
3212 err = -ENOMEM;
3213 goto out_put_prog;
3214 }
3215 bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3216 &bpf_tracing_link_lops, prog);
3217 link->attach_type = prog->expected_attach_type;
3218 link->link.cookie = bpf_cookie;
3219
3220 mutex_lock(&prog->aux->dst_mutex);
3221
3222 /* There are a few possible cases here:
3223 *
3224 * - if prog->aux->dst_trampoline is set, the program was just loaded
3225 * and not yet attached to anything, so we can use the values stored
3226 * in prog->aux
3227 *
3228 * - if prog->aux->dst_trampoline is NULL, the program has already been
3229 * attached to a target and its initial target was cleared (below)
3230 *
3231 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3232 * target_btf_id using the link_create API.
3233 *
3234 * - if tgt_prog == NULL when this function was called using the old
3235 * raw_tracepoint_open API, and we need a target from prog->aux
3236 *
3237 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3238 * was detached and is going for re-attachment.
3239 *
3240 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3241 * are NULL, then program was already attached and user did not provide
3242 * tgt_prog_fd so we have no way to find out or create trampoline
3243 */
3244 if (!prog->aux->dst_trampoline && !tgt_prog) {
3245 /*
3246 * Allow re-attach for TRACING and LSM programs. If it's
3247 * currently linked, bpf_trampoline_link_prog will fail.
3248 * EXT programs need to specify tgt_prog_fd, so they
3249 * re-attach in separate code path.
3250 */
3251 if (prog->type != BPF_PROG_TYPE_TRACING &&
3252 prog->type != BPF_PROG_TYPE_LSM) {
3253 err = -EINVAL;
3254 goto out_unlock;
3255 }
3256 /* We can allow re-attach only if we have valid attach_btf. */
3257 if (!prog->aux->attach_btf) {
3258 err = -EINVAL;
3259 goto out_unlock;
3260 }
3261 btf_id = prog->aux->attach_btf_id;
3262 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3263 }
3264
3265 if (!prog->aux->dst_trampoline ||
3266 (key && key != prog->aux->dst_trampoline->key)) {
3267 /* If there is no saved target, or the specified target is
3268 * different from the destination specified at load time, we
3269 * need a new trampoline and a check for compatibility
3270 */
3271 struct bpf_attach_target_info tgt_info = {};
3272
3273 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3274 &tgt_info);
3275 if (err)
3276 goto out_unlock;
3277
3278 if (tgt_info.tgt_mod) {
3279 module_put(prog->aux->mod);
3280 prog->aux->mod = tgt_info.tgt_mod;
3281 }
3282
3283 tr = bpf_trampoline_get(key, &tgt_info);
3284 if (!tr) {
3285 err = -ENOMEM;
3286 goto out_unlock;
3287 }
3288 } else {
3289 /* The caller didn't specify a target, or the target was the
3290 * same as the destination supplied during program load. This
3291 * means we can reuse the trampoline and reference from program
3292 * load time, and there is no need to allocate a new one. This
3293 * can only happen once for any program, as the saved values in
3294 * prog->aux are cleared below.
3295 */
3296 tr = prog->aux->dst_trampoline;
3297 tgt_prog = prog->aux->dst_prog;
3298 }
3299
3300 err = bpf_link_prime(&link->link.link, &link_primer);
3301 if (err)
3302 goto out_unlock;
3303
3304 err = bpf_trampoline_link_prog(&link->link, tr);
3305 if (err) {
3306 bpf_link_cleanup(&link_primer);
3307 link = NULL;
3308 goto out_unlock;
3309 }
3310
3311 link->tgt_prog = tgt_prog;
3312 link->trampoline = tr;
3313
3314 /* Always clear the trampoline and target prog from prog->aux to make
3315 * sure the original attach destination is not kept alive after a
3316 * program is (re-)attached to another target.
3317 */
3318 if (prog->aux->dst_prog &&
3319 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3320 /* got extra prog ref from syscall, or attaching to different prog */
3321 bpf_prog_put(prog->aux->dst_prog);
3322 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3323 /* we allocated a new trampoline, so free the old one */
3324 bpf_trampoline_put(prog->aux->dst_trampoline);
3325
3326 prog->aux->dst_prog = NULL;
3327 prog->aux->dst_trampoline = NULL;
3328 mutex_unlock(&prog->aux->dst_mutex);
3329
3330 return bpf_link_settle(&link_primer);
3331 out_unlock:
3332 if (tr && tr != prog->aux->dst_trampoline)
3333 bpf_trampoline_put(tr);
3334 mutex_unlock(&prog->aux->dst_mutex);
3335 kfree(link);
3336 out_put_prog:
3337 if (tgt_prog_fd && tgt_prog)
3338 bpf_prog_put(tgt_prog);
3339 return err;
3340 }
3341
3342 struct bpf_raw_tp_link {
3343 struct bpf_link link;
3344 struct bpf_raw_event_map *btp;
3345 };
3346
bpf_raw_tp_link_release(struct bpf_link * link)3347 static void bpf_raw_tp_link_release(struct bpf_link *link)
3348 {
3349 struct bpf_raw_tp_link *raw_tp =
3350 container_of(link, struct bpf_raw_tp_link, link);
3351
3352 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3353 bpf_put_raw_tracepoint(raw_tp->btp);
3354 }
3355
bpf_raw_tp_link_dealloc(struct bpf_link * link)3356 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3357 {
3358 struct bpf_raw_tp_link *raw_tp =
3359 container_of(link, struct bpf_raw_tp_link, link);
3360
3361 kfree(raw_tp);
3362 }
3363
bpf_raw_tp_link_show_fdinfo(const struct bpf_link * link,struct seq_file * seq)3364 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3365 struct seq_file *seq)
3366 {
3367 struct bpf_raw_tp_link *raw_tp_link =
3368 container_of(link, struct bpf_raw_tp_link, link);
3369
3370 seq_printf(seq,
3371 "tp_name:\t%s\n",
3372 raw_tp_link->btp->tp->name);
3373 }
3374
bpf_copy_to_user(char __user * ubuf,const char * buf,u32 ulen,u32 len)3375 static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3376 u32 len)
3377 {
3378 if (ulen >= len + 1) {
3379 if (copy_to_user(ubuf, buf, len + 1))
3380 return -EFAULT;
3381 } else {
3382 char zero = '\0';
3383
3384 if (copy_to_user(ubuf, buf, ulen - 1))
3385 return -EFAULT;
3386 if (put_user(zero, ubuf + ulen - 1))
3387 return -EFAULT;
3388 return -ENOSPC;
3389 }
3390
3391 return 0;
3392 }
3393
bpf_raw_tp_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)3394 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3395 struct bpf_link_info *info)
3396 {
3397 struct bpf_raw_tp_link *raw_tp_link =
3398 container_of(link, struct bpf_raw_tp_link, link);
3399 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3400 const char *tp_name = raw_tp_link->btp->tp->name;
3401 u32 ulen = info->raw_tracepoint.tp_name_len;
3402 size_t tp_len = strlen(tp_name);
3403
3404 if (!ulen ^ !ubuf)
3405 return -EINVAL;
3406
3407 info->raw_tracepoint.tp_name_len = tp_len + 1;
3408
3409 if (!ubuf)
3410 return 0;
3411
3412 return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3413 }
3414
3415 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3416 .release = bpf_raw_tp_link_release,
3417 .dealloc_deferred = bpf_raw_tp_link_dealloc,
3418 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3419 .fill_link_info = bpf_raw_tp_link_fill_link_info,
3420 };
3421
3422 #ifdef CONFIG_PERF_EVENTS
3423 struct bpf_perf_link {
3424 struct bpf_link link;
3425 struct file *perf_file;
3426 };
3427
bpf_perf_link_release(struct bpf_link * link)3428 static void bpf_perf_link_release(struct bpf_link *link)
3429 {
3430 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3431 struct perf_event *event = perf_link->perf_file->private_data;
3432
3433 perf_event_free_bpf_prog(event);
3434 fput(perf_link->perf_file);
3435 }
3436
bpf_perf_link_dealloc(struct bpf_link * link)3437 static void bpf_perf_link_dealloc(struct bpf_link *link)
3438 {
3439 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3440
3441 kfree(perf_link);
3442 }
3443
bpf_perf_link_fill_common(const struct perf_event * event,char __user * uname,u32 ulen,u64 * probe_offset,u64 * probe_addr,u32 * fd_type)3444 static int bpf_perf_link_fill_common(const struct perf_event *event,
3445 char __user *uname, u32 ulen,
3446 u64 *probe_offset, u64 *probe_addr,
3447 u32 *fd_type)
3448 {
3449 const char *buf;
3450 u32 prog_id;
3451 size_t len;
3452 int err;
3453
3454 if (!ulen ^ !uname)
3455 return -EINVAL;
3456
3457 err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3458 probe_offset, probe_addr);
3459 if (err)
3460 return err;
3461 if (!uname)
3462 return 0;
3463 if (buf) {
3464 len = strlen(buf);
3465 err = bpf_copy_to_user(uname, buf, ulen, len);
3466 if (err)
3467 return err;
3468 } else {
3469 char zero = '\0';
3470
3471 if (put_user(zero, uname))
3472 return -EFAULT;
3473 }
3474 return 0;
3475 }
3476
3477 #ifdef CONFIG_KPROBE_EVENTS
bpf_perf_link_fill_kprobe(const struct perf_event * event,struct bpf_link_info * info)3478 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3479 struct bpf_link_info *info)
3480 {
3481 char __user *uname;
3482 u64 addr, offset;
3483 u32 ulen, type;
3484 int err;
3485
3486 uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3487 ulen = info->perf_event.kprobe.name_len;
3488 err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3489 &type);
3490 if (err)
3491 return err;
3492 if (type == BPF_FD_TYPE_KRETPROBE)
3493 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3494 else
3495 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3496
3497 info->perf_event.kprobe.offset = offset;
3498 if (!kallsyms_show_value(current_cred()))
3499 addr = 0;
3500 info->perf_event.kprobe.addr = addr;
3501 return 0;
3502 }
3503 #endif
3504
3505 #ifdef CONFIG_UPROBE_EVENTS
bpf_perf_link_fill_uprobe(const struct perf_event * event,struct bpf_link_info * info)3506 static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3507 struct bpf_link_info *info)
3508 {
3509 char __user *uname;
3510 u64 addr, offset;
3511 u32 ulen, type;
3512 int err;
3513
3514 uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3515 ulen = info->perf_event.uprobe.name_len;
3516 err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3517 &type);
3518 if (err)
3519 return err;
3520
3521 if (type == BPF_FD_TYPE_URETPROBE)
3522 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3523 else
3524 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3525 info->perf_event.uprobe.offset = offset;
3526 return 0;
3527 }
3528 #endif
3529
bpf_perf_link_fill_probe(const struct perf_event * event,struct bpf_link_info * info)3530 static int bpf_perf_link_fill_probe(const struct perf_event *event,
3531 struct bpf_link_info *info)
3532 {
3533 #ifdef CONFIG_KPROBE_EVENTS
3534 if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3535 return bpf_perf_link_fill_kprobe(event, info);
3536 #endif
3537 #ifdef CONFIG_UPROBE_EVENTS
3538 if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3539 return bpf_perf_link_fill_uprobe(event, info);
3540 #endif
3541 return -EOPNOTSUPP;
3542 }
3543
bpf_perf_link_fill_tracepoint(const struct perf_event * event,struct bpf_link_info * info)3544 static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3545 struct bpf_link_info *info)
3546 {
3547 char __user *uname;
3548 u32 ulen;
3549
3550 uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3551 ulen = info->perf_event.tracepoint.name_len;
3552 info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3553 return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL);
3554 }
3555
bpf_perf_link_fill_perf_event(const struct perf_event * event,struct bpf_link_info * info)3556 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3557 struct bpf_link_info *info)
3558 {
3559 info->perf_event.event.type = event->attr.type;
3560 info->perf_event.event.config = event->attr.config;
3561 info->perf_event.type = BPF_PERF_EVENT_EVENT;
3562 return 0;
3563 }
3564
bpf_perf_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)3565 static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3566 struct bpf_link_info *info)
3567 {
3568 struct bpf_perf_link *perf_link;
3569 const struct perf_event *event;
3570
3571 perf_link = container_of(link, struct bpf_perf_link, link);
3572 event = perf_get_event(perf_link->perf_file);
3573 if (IS_ERR(event))
3574 return PTR_ERR(event);
3575
3576 switch (event->prog->type) {
3577 case BPF_PROG_TYPE_PERF_EVENT:
3578 return bpf_perf_link_fill_perf_event(event, info);
3579 case BPF_PROG_TYPE_TRACEPOINT:
3580 return bpf_perf_link_fill_tracepoint(event, info);
3581 case BPF_PROG_TYPE_KPROBE:
3582 return bpf_perf_link_fill_probe(event, info);
3583 default:
3584 return -EOPNOTSUPP;
3585 }
3586 }
3587
3588 static const struct bpf_link_ops bpf_perf_link_lops = {
3589 .release = bpf_perf_link_release,
3590 .dealloc = bpf_perf_link_dealloc,
3591 .fill_link_info = bpf_perf_link_fill_link_info,
3592 };
3593
bpf_perf_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)3594 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3595 {
3596 struct bpf_link_primer link_primer;
3597 struct bpf_perf_link *link;
3598 struct perf_event *event;
3599 struct file *perf_file;
3600 int err;
3601
3602 if (attr->link_create.flags)
3603 return -EINVAL;
3604
3605 perf_file = perf_event_get(attr->link_create.target_fd);
3606 if (IS_ERR(perf_file))
3607 return PTR_ERR(perf_file);
3608
3609 link = kzalloc(sizeof(*link), GFP_USER);
3610 if (!link) {
3611 err = -ENOMEM;
3612 goto out_put_file;
3613 }
3614 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3615 link->perf_file = perf_file;
3616
3617 err = bpf_link_prime(&link->link, &link_primer);
3618 if (err) {
3619 kfree(link);
3620 goto out_put_file;
3621 }
3622
3623 event = perf_file->private_data;
3624 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3625 if (err) {
3626 bpf_link_cleanup(&link_primer);
3627 goto out_put_file;
3628 }
3629 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3630 bpf_prog_inc(prog);
3631
3632 return bpf_link_settle(&link_primer);
3633
3634 out_put_file:
3635 fput(perf_file);
3636 return err;
3637 }
3638 #else
bpf_perf_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)3639 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3640 {
3641 return -EOPNOTSUPP;
3642 }
3643 #endif /* CONFIG_PERF_EVENTS */
3644
bpf_raw_tp_link_attach(struct bpf_prog * prog,const char __user * user_tp_name)3645 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3646 const char __user *user_tp_name)
3647 {
3648 struct bpf_link_primer link_primer;
3649 struct bpf_raw_tp_link *link;
3650 struct bpf_raw_event_map *btp;
3651 const char *tp_name;
3652 char buf[128];
3653 int err;
3654
3655 switch (prog->type) {
3656 case BPF_PROG_TYPE_TRACING:
3657 case BPF_PROG_TYPE_EXT:
3658 case BPF_PROG_TYPE_LSM:
3659 if (user_tp_name)
3660 /* The attach point for this category of programs
3661 * should be specified via btf_id during program load.
3662 */
3663 return -EINVAL;
3664 if (prog->type == BPF_PROG_TYPE_TRACING &&
3665 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3666 tp_name = prog->aux->attach_func_name;
3667 break;
3668 }
3669 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3670 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3671 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3672 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3673 return -EFAULT;
3674 buf[sizeof(buf) - 1] = 0;
3675 tp_name = buf;
3676 break;
3677 default:
3678 return -EINVAL;
3679 }
3680
3681 btp = bpf_get_raw_tracepoint(tp_name);
3682 if (!btp)
3683 return -ENOENT;
3684
3685 link = kzalloc(sizeof(*link), GFP_USER);
3686 if (!link) {
3687 err = -ENOMEM;
3688 goto out_put_btp;
3689 }
3690 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3691 &bpf_raw_tp_link_lops, prog);
3692 link->btp = btp;
3693
3694 err = bpf_link_prime(&link->link, &link_primer);
3695 if (err) {
3696 kfree(link);
3697 goto out_put_btp;
3698 }
3699
3700 err = bpf_probe_register(link->btp, prog);
3701 if (err) {
3702 bpf_link_cleanup(&link_primer);
3703 goto out_put_btp;
3704 }
3705
3706 return bpf_link_settle(&link_primer);
3707
3708 out_put_btp:
3709 bpf_put_raw_tracepoint(btp);
3710 return err;
3711 }
3712
3713 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3714
bpf_raw_tracepoint_open(const union bpf_attr * attr)3715 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3716 {
3717 struct bpf_prog *prog;
3718 int fd;
3719
3720 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3721 return -EINVAL;
3722
3723 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3724 if (IS_ERR(prog))
3725 return PTR_ERR(prog);
3726
3727 fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3728 if (fd < 0)
3729 bpf_prog_put(prog);
3730 return fd;
3731 }
3732
3733 static enum bpf_prog_type
attach_type_to_prog_type(enum bpf_attach_type attach_type)3734 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3735 {
3736 switch (attach_type) {
3737 case BPF_CGROUP_INET_INGRESS:
3738 case BPF_CGROUP_INET_EGRESS:
3739 return BPF_PROG_TYPE_CGROUP_SKB;
3740 case BPF_CGROUP_INET_SOCK_CREATE:
3741 case BPF_CGROUP_INET_SOCK_RELEASE:
3742 case BPF_CGROUP_INET4_POST_BIND:
3743 case BPF_CGROUP_INET6_POST_BIND:
3744 return BPF_PROG_TYPE_CGROUP_SOCK;
3745 case BPF_CGROUP_INET4_BIND:
3746 case BPF_CGROUP_INET6_BIND:
3747 case BPF_CGROUP_INET4_CONNECT:
3748 case BPF_CGROUP_INET6_CONNECT:
3749 case BPF_CGROUP_INET4_GETPEERNAME:
3750 case BPF_CGROUP_INET6_GETPEERNAME:
3751 case BPF_CGROUP_INET4_GETSOCKNAME:
3752 case BPF_CGROUP_INET6_GETSOCKNAME:
3753 case BPF_CGROUP_UDP4_SENDMSG:
3754 case BPF_CGROUP_UDP6_SENDMSG:
3755 case BPF_CGROUP_UDP4_RECVMSG:
3756 case BPF_CGROUP_UDP6_RECVMSG:
3757 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3758 case BPF_CGROUP_SOCK_OPS:
3759 return BPF_PROG_TYPE_SOCK_OPS;
3760 case BPF_CGROUP_DEVICE:
3761 return BPF_PROG_TYPE_CGROUP_DEVICE;
3762 case BPF_SK_MSG_VERDICT:
3763 return BPF_PROG_TYPE_SK_MSG;
3764 case BPF_SK_SKB_STREAM_PARSER:
3765 case BPF_SK_SKB_STREAM_VERDICT:
3766 case BPF_SK_SKB_VERDICT:
3767 return BPF_PROG_TYPE_SK_SKB;
3768 case BPF_LIRC_MODE2:
3769 return BPF_PROG_TYPE_LIRC_MODE2;
3770 case BPF_FLOW_DISSECTOR:
3771 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3772 case BPF_CGROUP_SYSCTL:
3773 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3774 case BPF_CGROUP_GETSOCKOPT:
3775 case BPF_CGROUP_SETSOCKOPT:
3776 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3777 case BPF_TRACE_ITER:
3778 case BPF_TRACE_RAW_TP:
3779 case BPF_TRACE_FENTRY:
3780 case BPF_TRACE_FEXIT:
3781 case BPF_MODIFY_RETURN:
3782 return BPF_PROG_TYPE_TRACING;
3783 case BPF_LSM_MAC:
3784 return BPF_PROG_TYPE_LSM;
3785 case BPF_SK_LOOKUP:
3786 return BPF_PROG_TYPE_SK_LOOKUP;
3787 case BPF_XDP:
3788 return BPF_PROG_TYPE_XDP;
3789 case BPF_LSM_CGROUP:
3790 return BPF_PROG_TYPE_LSM;
3791 case BPF_TCX_INGRESS:
3792 case BPF_TCX_EGRESS:
3793 return BPF_PROG_TYPE_SCHED_CLS;
3794 default:
3795 return BPF_PROG_TYPE_UNSPEC;
3796 }
3797 }
3798
bpf_prog_attach_check_attach_type(const struct bpf_prog * prog,enum bpf_attach_type attach_type)3799 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3800 enum bpf_attach_type attach_type)
3801 {
3802 enum bpf_prog_type ptype;
3803
3804 switch (prog->type) {
3805 case BPF_PROG_TYPE_CGROUP_SOCK:
3806 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3807 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3808 case BPF_PROG_TYPE_SK_LOOKUP:
3809 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3810 case BPF_PROG_TYPE_CGROUP_SKB:
3811 if (!capable(CAP_NET_ADMIN))
3812 /* cg-skb progs can be loaded by unpriv user.
3813 * check permissions at attach time.
3814 */
3815 return -EPERM;
3816
3817 ptype = attach_type_to_prog_type(attach_type);
3818 if (prog->type != ptype)
3819 return -EINVAL;
3820
3821 return prog->enforce_expected_attach_type &&
3822 prog->expected_attach_type != attach_type ?
3823 -EINVAL : 0;
3824 case BPF_PROG_TYPE_EXT:
3825 return 0;
3826 case BPF_PROG_TYPE_NETFILTER:
3827 if (attach_type != BPF_NETFILTER)
3828 return -EINVAL;
3829 return 0;
3830 case BPF_PROG_TYPE_PERF_EVENT:
3831 case BPF_PROG_TYPE_TRACEPOINT:
3832 if (attach_type != BPF_PERF_EVENT)
3833 return -EINVAL;
3834 return 0;
3835 case BPF_PROG_TYPE_KPROBE:
3836 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
3837 attach_type != BPF_TRACE_KPROBE_MULTI)
3838 return -EINVAL;
3839 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
3840 attach_type != BPF_TRACE_UPROBE_MULTI)
3841 return -EINVAL;
3842 if (attach_type != BPF_PERF_EVENT &&
3843 attach_type != BPF_TRACE_KPROBE_MULTI &&
3844 attach_type != BPF_TRACE_UPROBE_MULTI)
3845 return -EINVAL;
3846 return 0;
3847 case BPF_PROG_TYPE_SCHED_CLS:
3848 if (attach_type != BPF_TCX_INGRESS &&
3849 attach_type != BPF_TCX_EGRESS)
3850 return -EINVAL;
3851 return 0;
3852 default:
3853 ptype = attach_type_to_prog_type(attach_type);
3854 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
3855 return -EINVAL;
3856 return 0;
3857 }
3858 }
3859
3860 #define BPF_PROG_ATTACH_LAST_FIELD expected_revision
3861
3862 #define BPF_F_ATTACH_MASK_BASE \
3863 (BPF_F_ALLOW_OVERRIDE | \
3864 BPF_F_ALLOW_MULTI | \
3865 BPF_F_REPLACE)
3866
3867 #define BPF_F_ATTACH_MASK_MPROG \
3868 (BPF_F_REPLACE | \
3869 BPF_F_BEFORE | \
3870 BPF_F_AFTER | \
3871 BPF_F_ID | \
3872 BPF_F_LINK)
3873
bpf_prog_attach(const union bpf_attr * attr)3874 static int bpf_prog_attach(const union bpf_attr *attr)
3875 {
3876 enum bpf_prog_type ptype;
3877 struct bpf_prog *prog;
3878 int ret;
3879
3880 if (CHECK_ATTR(BPF_PROG_ATTACH))
3881 return -EINVAL;
3882
3883 ptype = attach_type_to_prog_type(attr->attach_type);
3884 if (ptype == BPF_PROG_TYPE_UNSPEC)
3885 return -EINVAL;
3886 if (bpf_mprog_supported(ptype)) {
3887 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3888 return -EINVAL;
3889 } else {
3890 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
3891 return -EINVAL;
3892 if (attr->relative_fd ||
3893 attr->expected_revision)
3894 return -EINVAL;
3895 }
3896
3897 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3898 if (IS_ERR(prog))
3899 return PTR_ERR(prog);
3900
3901 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3902 bpf_prog_put(prog);
3903 return -EINVAL;
3904 }
3905
3906 switch (ptype) {
3907 case BPF_PROG_TYPE_SK_SKB:
3908 case BPF_PROG_TYPE_SK_MSG:
3909 ret = sock_map_get_from_fd(attr, prog);
3910 break;
3911 case BPF_PROG_TYPE_LIRC_MODE2:
3912 ret = lirc_prog_attach(attr, prog);
3913 break;
3914 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3915 ret = netns_bpf_prog_attach(attr, prog);
3916 break;
3917 case BPF_PROG_TYPE_CGROUP_DEVICE:
3918 case BPF_PROG_TYPE_CGROUP_SKB:
3919 case BPF_PROG_TYPE_CGROUP_SOCK:
3920 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3921 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3922 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3923 case BPF_PROG_TYPE_SOCK_OPS:
3924 case BPF_PROG_TYPE_LSM:
3925 if (ptype == BPF_PROG_TYPE_LSM &&
3926 prog->expected_attach_type != BPF_LSM_CGROUP)
3927 ret = -EINVAL;
3928 else
3929 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3930 break;
3931 case BPF_PROG_TYPE_SCHED_CLS:
3932 ret = tcx_prog_attach(attr, prog);
3933 break;
3934 default:
3935 ret = -EINVAL;
3936 }
3937
3938 if (ret)
3939 bpf_prog_put(prog);
3940 return ret;
3941 }
3942
3943 #define BPF_PROG_DETACH_LAST_FIELD expected_revision
3944
bpf_prog_detach(const union bpf_attr * attr)3945 static int bpf_prog_detach(const union bpf_attr *attr)
3946 {
3947 struct bpf_prog *prog = NULL;
3948 enum bpf_prog_type ptype;
3949 int ret;
3950
3951 if (CHECK_ATTR(BPF_PROG_DETACH))
3952 return -EINVAL;
3953
3954 ptype = attach_type_to_prog_type(attr->attach_type);
3955 if (bpf_mprog_supported(ptype)) {
3956 if (ptype == BPF_PROG_TYPE_UNSPEC)
3957 return -EINVAL;
3958 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3959 return -EINVAL;
3960 if (attr->attach_bpf_fd) {
3961 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3962 if (IS_ERR(prog))
3963 return PTR_ERR(prog);
3964 }
3965 } else if (attr->attach_flags ||
3966 attr->relative_fd ||
3967 attr->expected_revision) {
3968 return -EINVAL;
3969 }
3970
3971 switch (ptype) {
3972 case BPF_PROG_TYPE_SK_MSG:
3973 case BPF_PROG_TYPE_SK_SKB:
3974 ret = sock_map_prog_detach(attr, ptype);
3975 break;
3976 case BPF_PROG_TYPE_LIRC_MODE2:
3977 ret = lirc_prog_detach(attr);
3978 break;
3979 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3980 ret = netns_bpf_prog_detach(attr, ptype);
3981 break;
3982 case BPF_PROG_TYPE_CGROUP_DEVICE:
3983 case BPF_PROG_TYPE_CGROUP_SKB:
3984 case BPF_PROG_TYPE_CGROUP_SOCK:
3985 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3986 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3987 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3988 case BPF_PROG_TYPE_SOCK_OPS:
3989 case BPF_PROG_TYPE_LSM:
3990 ret = cgroup_bpf_prog_detach(attr, ptype);
3991 break;
3992 case BPF_PROG_TYPE_SCHED_CLS:
3993 ret = tcx_prog_detach(attr, prog);
3994 break;
3995 default:
3996 ret = -EINVAL;
3997 }
3998
3999 if (prog)
4000 bpf_prog_put(prog);
4001 return ret;
4002 }
4003
4004 #define BPF_PROG_QUERY_LAST_FIELD query.revision
4005
bpf_prog_query(const union bpf_attr * attr,union bpf_attr __user * uattr)4006 static int bpf_prog_query(const union bpf_attr *attr,
4007 union bpf_attr __user *uattr)
4008 {
4009 if (!capable(CAP_NET_ADMIN))
4010 return -EPERM;
4011 if (CHECK_ATTR(BPF_PROG_QUERY))
4012 return -EINVAL;
4013 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4014 return -EINVAL;
4015
4016 switch (attr->query.attach_type) {
4017 case BPF_CGROUP_INET_INGRESS:
4018 case BPF_CGROUP_INET_EGRESS:
4019 case BPF_CGROUP_INET_SOCK_CREATE:
4020 case BPF_CGROUP_INET_SOCK_RELEASE:
4021 case BPF_CGROUP_INET4_BIND:
4022 case BPF_CGROUP_INET6_BIND:
4023 case BPF_CGROUP_INET4_POST_BIND:
4024 case BPF_CGROUP_INET6_POST_BIND:
4025 case BPF_CGROUP_INET4_CONNECT:
4026 case BPF_CGROUP_INET6_CONNECT:
4027 case BPF_CGROUP_INET4_GETPEERNAME:
4028 case BPF_CGROUP_INET6_GETPEERNAME:
4029 case BPF_CGROUP_INET4_GETSOCKNAME:
4030 case BPF_CGROUP_INET6_GETSOCKNAME:
4031 case BPF_CGROUP_UDP4_SENDMSG:
4032 case BPF_CGROUP_UDP6_SENDMSG:
4033 case BPF_CGROUP_UDP4_RECVMSG:
4034 case BPF_CGROUP_UDP6_RECVMSG:
4035 case BPF_CGROUP_SOCK_OPS:
4036 case BPF_CGROUP_DEVICE:
4037 case BPF_CGROUP_SYSCTL:
4038 case BPF_CGROUP_GETSOCKOPT:
4039 case BPF_CGROUP_SETSOCKOPT:
4040 case BPF_LSM_CGROUP:
4041 return cgroup_bpf_prog_query(attr, uattr);
4042 case BPF_LIRC_MODE2:
4043 return lirc_prog_query(attr, uattr);
4044 case BPF_FLOW_DISSECTOR:
4045 case BPF_SK_LOOKUP:
4046 return netns_bpf_prog_query(attr, uattr);
4047 case BPF_SK_SKB_STREAM_PARSER:
4048 case BPF_SK_SKB_STREAM_VERDICT:
4049 case BPF_SK_MSG_VERDICT:
4050 case BPF_SK_SKB_VERDICT:
4051 return sock_map_bpf_prog_query(attr, uattr);
4052 case BPF_TCX_INGRESS:
4053 case BPF_TCX_EGRESS:
4054 return tcx_prog_query(attr, uattr);
4055 default:
4056 return -EINVAL;
4057 }
4058 }
4059
4060 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4061
bpf_prog_test_run(const union bpf_attr * attr,union bpf_attr __user * uattr)4062 static int bpf_prog_test_run(const union bpf_attr *attr,
4063 union bpf_attr __user *uattr)
4064 {
4065 struct bpf_prog *prog;
4066 int ret = -ENOTSUPP;
4067
4068 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4069 return -EINVAL;
4070
4071 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4072 (!attr->test.ctx_size_in && attr->test.ctx_in))
4073 return -EINVAL;
4074
4075 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4076 (!attr->test.ctx_size_out && attr->test.ctx_out))
4077 return -EINVAL;
4078
4079 prog = bpf_prog_get(attr->test.prog_fd);
4080 if (IS_ERR(prog))
4081 return PTR_ERR(prog);
4082
4083 if (prog->aux->ops->test_run)
4084 ret = prog->aux->ops->test_run(prog, attr, uattr);
4085
4086 bpf_prog_put(prog);
4087 return ret;
4088 }
4089
4090 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4091
bpf_obj_get_next_id(const union bpf_attr * attr,union bpf_attr __user * uattr,struct idr * idr,spinlock_t * lock)4092 static int bpf_obj_get_next_id(const union bpf_attr *attr,
4093 union bpf_attr __user *uattr,
4094 struct idr *idr,
4095 spinlock_t *lock)
4096 {
4097 u32 next_id = attr->start_id;
4098 int err = 0;
4099
4100 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4101 return -EINVAL;
4102
4103 if (!capable(CAP_SYS_ADMIN))
4104 return -EPERM;
4105
4106 next_id++;
4107 spin_lock_bh(lock);
4108 if (!idr_get_next(idr, &next_id))
4109 err = -ENOENT;
4110 spin_unlock_bh(lock);
4111
4112 if (!err)
4113 err = put_user(next_id, &uattr->next_id);
4114
4115 return err;
4116 }
4117
bpf_map_get_curr_or_next(u32 * id)4118 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4119 {
4120 struct bpf_map *map;
4121
4122 spin_lock_bh(&map_idr_lock);
4123 again:
4124 map = idr_get_next(&map_idr, id);
4125 if (map) {
4126 map = __bpf_map_inc_not_zero(map, false);
4127 if (IS_ERR(map)) {
4128 (*id)++;
4129 goto again;
4130 }
4131 }
4132 spin_unlock_bh(&map_idr_lock);
4133
4134 return map;
4135 }
4136
bpf_prog_get_curr_or_next(u32 * id)4137 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4138 {
4139 struct bpf_prog *prog;
4140
4141 spin_lock_bh(&prog_idr_lock);
4142 again:
4143 prog = idr_get_next(&prog_idr, id);
4144 if (prog) {
4145 prog = bpf_prog_inc_not_zero(prog);
4146 if (IS_ERR(prog)) {
4147 (*id)++;
4148 goto again;
4149 }
4150 }
4151 spin_unlock_bh(&prog_idr_lock);
4152
4153 return prog;
4154 }
4155
4156 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4157
bpf_prog_by_id(u32 id)4158 struct bpf_prog *bpf_prog_by_id(u32 id)
4159 {
4160 struct bpf_prog *prog;
4161
4162 if (!id)
4163 return ERR_PTR(-ENOENT);
4164
4165 spin_lock_bh(&prog_idr_lock);
4166 prog = idr_find(&prog_idr, id);
4167 if (prog)
4168 prog = bpf_prog_inc_not_zero(prog);
4169 else
4170 prog = ERR_PTR(-ENOENT);
4171 spin_unlock_bh(&prog_idr_lock);
4172 return prog;
4173 }
4174
bpf_prog_get_fd_by_id(const union bpf_attr * attr)4175 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4176 {
4177 struct bpf_prog *prog;
4178 u32 id = attr->prog_id;
4179 int fd;
4180
4181 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4182 return -EINVAL;
4183
4184 if (!capable(CAP_SYS_ADMIN))
4185 return -EPERM;
4186
4187 prog = bpf_prog_by_id(id);
4188 if (IS_ERR(prog))
4189 return PTR_ERR(prog);
4190
4191 fd = bpf_prog_new_fd(prog);
4192 if (fd < 0)
4193 bpf_prog_put(prog);
4194
4195 return fd;
4196 }
4197
4198 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4199
bpf_map_get_fd_by_id(const union bpf_attr * attr)4200 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4201 {
4202 struct bpf_map *map;
4203 u32 id = attr->map_id;
4204 int f_flags;
4205 int fd;
4206
4207 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4208 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4209 return -EINVAL;
4210
4211 if (!capable(CAP_SYS_ADMIN))
4212 return -EPERM;
4213
4214 f_flags = bpf_get_file_flag(attr->open_flags);
4215 if (f_flags < 0)
4216 return f_flags;
4217
4218 spin_lock_bh(&map_idr_lock);
4219 map = idr_find(&map_idr, id);
4220 if (map)
4221 map = __bpf_map_inc_not_zero(map, true);
4222 else
4223 map = ERR_PTR(-ENOENT);
4224 spin_unlock_bh(&map_idr_lock);
4225
4226 if (IS_ERR(map))
4227 return PTR_ERR(map);
4228
4229 fd = bpf_map_new_fd(map, f_flags);
4230 if (fd < 0)
4231 bpf_map_put_with_uref(map);
4232
4233 return fd;
4234 }
4235
bpf_map_from_imm(const struct bpf_prog * prog,unsigned long addr,u32 * off,u32 * type)4236 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4237 unsigned long addr, u32 *off,
4238 u32 *type)
4239 {
4240 const struct bpf_map *map;
4241 int i;
4242
4243 mutex_lock(&prog->aux->used_maps_mutex);
4244 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4245 map = prog->aux->used_maps[i];
4246 if (map == (void *)addr) {
4247 *type = BPF_PSEUDO_MAP_FD;
4248 goto out;
4249 }
4250 if (!map->ops->map_direct_value_meta)
4251 continue;
4252 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4253 *type = BPF_PSEUDO_MAP_VALUE;
4254 goto out;
4255 }
4256 }
4257 map = NULL;
4258
4259 out:
4260 mutex_unlock(&prog->aux->used_maps_mutex);
4261 return map;
4262 }
4263
bpf_insn_prepare_dump(const struct bpf_prog * prog,const struct cred * f_cred)4264 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4265 const struct cred *f_cred)
4266 {
4267 const struct bpf_map *map;
4268 struct bpf_insn *insns;
4269 u32 off, type;
4270 u64 imm;
4271 u8 code;
4272 int i;
4273
4274 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4275 GFP_USER);
4276 if (!insns)
4277 return insns;
4278
4279 for (i = 0; i < prog->len; i++) {
4280 code = insns[i].code;
4281
4282 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4283 insns[i].code = BPF_JMP | BPF_CALL;
4284 insns[i].imm = BPF_FUNC_tail_call;
4285 /* fall-through */
4286 }
4287 if (code == (BPF_JMP | BPF_CALL) ||
4288 code == (BPF_JMP | BPF_CALL_ARGS)) {
4289 if (code == (BPF_JMP | BPF_CALL_ARGS))
4290 insns[i].code = BPF_JMP | BPF_CALL;
4291 if (!bpf_dump_raw_ok(f_cred))
4292 insns[i].imm = 0;
4293 continue;
4294 }
4295 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4296 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4297 continue;
4298 }
4299
4300 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4301 continue;
4302
4303 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4304 map = bpf_map_from_imm(prog, imm, &off, &type);
4305 if (map) {
4306 insns[i].src_reg = type;
4307 insns[i].imm = map->id;
4308 insns[i + 1].imm = off;
4309 continue;
4310 }
4311 }
4312
4313 return insns;
4314 }
4315
set_info_rec_size(struct bpf_prog_info * info)4316 static int set_info_rec_size(struct bpf_prog_info *info)
4317 {
4318 /*
4319 * Ensure info.*_rec_size is the same as kernel expected size
4320 *
4321 * or
4322 *
4323 * Only allow zero *_rec_size if both _rec_size and _cnt are
4324 * zero. In this case, the kernel will set the expected
4325 * _rec_size back to the info.
4326 */
4327
4328 if ((info->nr_func_info || info->func_info_rec_size) &&
4329 info->func_info_rec_size != sizeof(struct bpf_func_info))
4330 return -EINVAL;
4331
4332 if ((info->nr_line_info || info->line_info_rec_size) &&
4333 info->line_info_rec_size != sizeof(struct bpf_line_info))
4334 return -EINVAL;
4335
4336 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4337 info->jited_line_info_rec_size != sizeof(__u64))
4338 return -EINVAL;
4339
4340 info->func_info_rec_size = sizeof(struct bpf_func_info);
4341 info->line_info_rec_size = sizeof(struct bpf_line_info);
4342 info->jited_line_info_rec_size = sizeof(__u64);
4343
4344 return 0;
4345 }
4346
bpf_prog_get_info_by_fd(struct file * file,struct bpf_prog * prog,const union bpf_attr * attr,union bpf_attr __user * uattr)4347 static int bpf_prog_get_info_by_fd(struct file *file,
4348 struct bpf_prog *prog,
4349 const union bpf_attr *attr,
4350 union bpf_attr __user *uattr)
4351 {
4352 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4353 struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4354 struct bpf_prog_info info;
4355 u32 info_len = attr->info.info_len;
4356 struct bpf_prog_kstats stats;
4357 char __user *uinsns;
4358 u32 ulen;
4359 int err;
4360
4361 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4362 if (err)
4363 return err;
4364 info_len = min_t(u32, sizeof(info), info_len);
4365
4366 memset(&info, 0, sizeof(info));
4367 if (copy_from_user(&info, uinfo, info_len))
4368 return -EFAULT;
4369
4370 info.type = prog->type;
4371 info.id = prog->aux->id;
4372 info.load_time = prog->aux->load_time;
4373 info.created_by_uid = from_kuid_munged(current_user_ns(),
4374 prog->aux->user->uid);
4375 info.gpl_compatible = prog->gpl_compatible;
4376
4377 memcpy(info.tag, prog->tag, sizeof(prog->tag));
4378 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4379
4380 mutex_lock(&prog->aux->used_maps_mutex);
4381 ulen = info.nr_map_ids;
4382 info.nr_map_ids = prog->aux->used_map_cnt;
4383 ulen = min_t(u32, info.nr_map_ids, ulen);
4384 if (ulen) {
4385 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4386 u32 i;
4387
4388 for (i = 0; i < ulen; i++)
4389 if (put_user(prog->aux->used_maps[i]->id,
4390 &user_map_ids[i])) {
4391 mutex_unlock(&prog->aux->used_maps_mutex);
4392 return -EFAULT;
4393 }
4394 }
4395 mutex_unlock(&prog->aux->used_maps_mutex);
4396
4397 err = set_info_rec_size(&info);
4398 if (err)
4399 return err;
4400
4401 bpf_prog_get_stats(prog, &stats);
4402 info.run_time_ns = stats.nsecs;
4403 info.run_cnt = stats.cnt;
4404 info.recursion_misses = stats.misses;
4405
4406 info.verified_insns = prog->aux->verified_insns;
4407
4408 if (!bpf_capable()) {
4409 info.jited_prog_len = 0;
4410 info.xlated_prog_len = 0;
4411 info.nr_jited_ksyms = 0;
4412 info.nr_jited_func_lens = 0;
4413 info.nr_func_info = 0;
4414 info.nr_line_info = 0;
4415 info.nr_jited_line_info = 0;
4416 goto done;
4417 }
4418
4419 ulen = info.xlated_prog_len;
4420 info.xlated_prog_len = bpf_prog_insn_size(prog);
4421 if (info.xlated_prog_len && ulen) {
4422 struct bpf_insn *insns_sanitized;
4423 bool fault;
4424
4425 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4426 info.xlated_prog_insns = 0;
4427 goto done;
4428 }
4429 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4430 if (!insns_sanitized)
4431 return -ENOMEM;
4432 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4433 ulen = min_t(u32, info.xlated_prog_len, ulen);
4434 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4435 kfree(insns_sanitized);
4436 if (fault)
4437 return -EFAULT;
4438 }
4439
4440 if (bpf_prog_is_offloaded(prog->aux)) {
4441 err = bpf_prog_offload_info_fill(&info, prog);
4442 if (err)
4443 return err;
4444 goto done;
4445 }
4446
4447 /* NOTE: the following code is supposed to be skipped for offload.
4448 * bpf_prog_offload_info_fill() is the place to fill similar fields
4449 * for offload.
4450 */
4451 ulen = info.jited_prog_len;
4452 if (prog->aux->func_cnt) {
4453 u32 i;
4454
4455 info.jited_prog_len = 0;
4456 for (i = 0; i < prog->aux->func_cnt; i++)
4457 info.jited_prog_len += prog->aux->func[i]->jited_len;
4458 } else {
4459 info.jited_prog_len = prog->jited_len;
4460 }
4461
4462 if (info.jited_prog_len && ulen) {
4463 if (bpf_dump_raw_ok(file->f_cred)) {
4464 uinsns = u64_to_user_ptr(info.jited_prog_insns);
4465 ulen = min_t(u32, info.jited_prog_len, ulen);
4466
4467 /* for multi-function programs, copy the JITed
4468 * instructions for all the functions
4469 */
4470 if (prog->aux->func_cnt) {
4471 u32 len, free, i;
4472 u8 *img;
4473
4474 free = ulen;
4475 for (i = 0; i < prog->aux->func_cnt; i++) {
4476 len = prog->aux->func[i]->jited_len;
4477 len = min_t(u32, len, free);
4478 img = (u8 *) prog->aux->func[i]->bpf_func;
4479 if (copy_to_user(uinsns, img, len))
4480 return -EFAULT;
4481 uinsns += len;
4482 free -= len;
4483 if (!free)
4484 break;
4485 }
4486 } else {
4487 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4488 return -EFAULT;
4489 }
4490 } else {
4491 info.jited_prog_insns = 0;
4492 }
4493 }
4494
4495 ulen = info.nr_jited_ksyms;
4496 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4497 if (ulen) {
4498 if (bpf_dump_raw_ok(file->f_cred)) {
4499 unsigned long ksym_addr;
4500 u64 __user *user_ksyms;
4501 u32 i;
4502
4503 /* copy the address of the kernel symbol
4504 * corresponding to each function
4505 */
4506 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4507 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4508 if (prog->aux->func_cnt) {
4509 for (i = 0; i < ulen; i++) {
4510 ksym_addr = (unsigned long)
4511 prog->aux->func[i]->bpf_func;
4512 if (put_user((u64) ksym_addr,
4513 &user_ksyms[i]))
4514 return -EFAULT;
4515 }
4516 } else {
4517 ksym_addr = (unsigned long) prog->bpf_func;
4518 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4519 return -EFAULT;
4520 }
4521 } else {
4522 info.jited_ksyms = 0;
4523 }
4524 }
4525
4526 ulen = info.nr_jited_func_lens;
4527 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4528 if (ulen) {
4529 if (bpf_dump_raw_ok(file->f_cred)) {
4530 u32 __user *user_lens;
4531 u32 func_len, i;
4532
4533 /* copy the JITed image lengths for each function */
4534 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4535 user_lens = u64_to_user_ptr(info.jited_func_lens);
4536 if (prog->aux->func_cnt) {
4537 for (i = 0; i < ulen; i++) {
4538 func_len =
4539 prog->aux->func[i]->jited_len;
4540 if (put_user(func_len, &user_lens[i]))
4541 return -EFAULT;
4542 }
4543 } else {
4544 func_len = prog->jited_len;
4545 if (put_user(func_len, &user_lens[0]))
4546 return -EFAULT;
4547 }
4548 } else {
4549 info.jited_func_lens = 0;
4550 }
4551 }
4552
4553 if (prog->aux->btf)
4554 info.btf_id = btf_obj_id(prog->aux->btf);
4555 info.attach_btf_id = prog->aux->attach_btf_id;
4556 if (attach_btf)
4557 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4558
4559 ulen = info.nr_func_info;
4560 info.nr_func_info = prog->aux->func_info_cnt;
4561 if (info.nr_func_info && ulen) {
4562 char __user *user_finfo;
4563
4564 user_finfo = u64_to_user_ptr(info.func_info);
4565 ulen = min_t(u32, info.nr_func_info, ulen);
4566 if (copy_to_user(user_finfo, prog->aux->func_info,
4567 info.func_info_rec_size * ulen))
4568 return -EFAULT;
4569 }
4570
4571 ulen = info.nr_line_info;
4572 info.nr_line_info = prog->aux->nr_linfo;
4573 if (info.nr_line_info && ulen) {
4574 __u8 __user *user_linfo;
4575
4576 user_linfo = u64_to_user_ptr(info.line_info);
4577 ulen = min_t(u32, info.nr_line_info, ulen);
4578 if (copy_to_user(user_linfo, prog->aux->linfo,
4579 info.line_info_rec_size * ulen))
4580 return -EFAULT;
4581 }
4582
4583 ulen = info.nr_jited_line_info;
4584 if (prog->aux->jited_linfo)
4585 info.nr_jited_line_info = prog->aux->nr_linfo;
4586 else
4587 info.nr_jited_line_info = 0;
4588 if (info.nr_jited_line_info && ulen) {
4589 if (bpf_dump_raw_ok(file->f_cred)) {
4590 unsigned long line_addr;
4591 __u64 __user *user_linfo;
4592 u32 i;
4593
4594 user_linfo = u64_to_user_ptr(info.jited_line_info);
4595 ulen = min_t(u32, info.nr_jited_line_info, ulen);
4596 for (i = 0; i < ulen; i++) {
4597 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4598 if (put_user((__u64)line_addr, &user_linfo[i]))
4599 return -EFAULT;
4600 }
4601 } else {
4602 info.jited_line_info = 0;
4603 }
4604 }
4605
4606 ulen = info.nr_prog_tags;
4607 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4608 if (ulen) {
4609 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4610 u32 i;
4611
4612 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4613 ulen = min_t(u32, info.nr_prog_tags, ulen);
4614 if (prog->aux->func_cnt) {
4615 for (i = 0; i < ulen; i++) {
4616 if (copy_to_user(user_prog_tags[i],
4617 prog->aux->func[i]->tag,
4618 BPF_TAG_SIZE))
4619 return -EFAULT;
4620 }
4621 } else {
4622 if (copy_to_user(user_prog_tags[0],
4623 prog->tag, BPF_TAG_SIZE))
4624 return -EFAULT;
4625 }
4626 }
4627
4628 done:
4629 if (copy_to_user(uinfo, &info, info_len) ||
4630 put_user(info_len, &uattr->info.info_len))
4631 return -EFAULT;
4632
4633 return 0;
4634 }
4635
bpf_map_get_info_by_fd(struct file * file,struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)4636 static int bpf_map_get_info_by_fd(struct file *file,
4637 struct bpf_map *map,
4638 const union bpf_attr *attr,
4639 union bpf_attr __user *uattr)
4640 {
4641 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4642 struct bpf_map_info info;
4643 u32 info_len = attr->info.info_len;
4644 int err;
4645
4646 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4647 if (err)
4648 return err;
4649 info_len = min_t(u32, sizeof(info), info_len);
4650
4651 memset(&info, 0, sizeof(info));
4652 info.type = map->map_type;
4653 info.id = map->id;
4654 info.key_size = map->key_size;
4655 info.value_size = map->value_size;
4656 info.max_entries = map->max_entries;
4657 info.map_flags = map->map_flags;
4658 info.map_extra = map->map_extra;
4659 memcpy(info.name, map->name, sizeof(map->name));
4660
4661 if (map->btf) {
4662 info.btf_id = btf_obj_id(map->btf);
4663 info.btf_key_type_id = map->btf_key_type_id;
4664 info.btf_value_type_id = map->btf_value_type_id;
4665 }
4666 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4667
4668 if (bpf_map_is_offloaded(map)) {
4669 err = bpf_map_offload_info_fill(&info, map);
4670 if (err)
4671 return err;
4672 }
4673
4674 if (copy_to_user(uinfo, &info, info_len) ||
4675 put_user(info_len, &uattr->info.info_len))
4676 return -EFAULT;
4677
4678 return 0;
4679 }
4680
bpf_btf_get_info_by_fd(struct file * file,struct btf * btf,const union bpf_attr * attr,union bpf_attr __user * uattr)4681 static int bpf_btf_get_info_by_fd(struct file *file,
4682 struct btf *btf,
4683 const union bpf_attr *attr,
4684 union bpf_attr __user *uattr)
4685 {
4686 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4687 u32 info_len = attr->info.info_len;
4688 int err;
4689
4690 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4691 if (err)
4692 return err;
4693
4694 return btf_get_info_by_fd(btf, attr, uattr);
4695 }
4696
bpf_link_get_info_by_fd(struct file * file,struct bpf_link * link,const union bpf_attr * attr,union bpf_attr __user * uattr)4697 static int bpf_link_get_info_by_fd(struct file *file,
4698 struct bpf_link *link,
4699 const union bpf_attr *attr,
4700 union bpf_attr __user *uattr)
4701 {
4702 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4703 struct bpf_link_info info;
4704 u32 info_len = attr->info.info_len;
4705 int err;
4706
4707 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4708 if (err)
4709 return err;
4710 info_len = min_t(u32, sizeof(info), info_len);
4711
4712 memset(&info, 0, sizeof(info));
4713 if (copy_from_user(&info, uinfo, info_len))
4714 return -EFAULT;
4715
4716 info.type = link->type;
4717 info.id = link->id;
4718 if (link->prog)
4719 info.prog_id = link->prog->aux->id;
4720
4721 if (link->ops->fill_link_info) {
4722 err = link->ops->fill_link_info(link, &info);
4723 if (err)
4724 return err;
4725 }
4726
4727 if (copy_to_user(uinfo, &info, info_len) ||
4728 put_user(info_len, &uattr->info.info_len))
4729 return -EFAULT;
4730
4731 return 0;
4732 }
4733
4734
4735 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4736
bpf_obj_get_info_by_fd(const union bpf_attr * attr,union bpf_attr __user * uattr)4737 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4738 union bpf_attr __user *uattr)
4739 {
4740 int ufd = attr->info.bpf_fd;
4741 struct fd f;
4742 int err;
4743
4744 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4745 return -EINVAL;
4746
4747 f = fdget(ufd);
4748 if (!f.file)
4749 return -EBADFD;
4750
4751 if (f.file->f_op == &bpf_prog_fops)
4752 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4753 uattr);
4754 else if (f.file->f_op == &bpf_map_fops)
4755 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4756 uattr);
4757 else if (f.file->f_op == &btf_fops)
4758 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4759 else if (f.file->f_op == &bpf_link_fops)
4760 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4761 attr, uattr);
4762 else
4763 err = -EINVAL;
4764
4765 fdput(f);
4766 return err;
4767 }
4768
4769 #define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4770
bpf_btf_load(const union bpf_attr * attr,bpfptr_t uattr,__u32 uattr_size)4771 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4772 {
4773 if (CHECK_ATTR(BPF_BTF_LOAD))
4774 return -EINVAL;
4775
4776 if (!bpf_capable())
4777 return -EPERM;
4778
4779 return btf_new_fd(attr, uattr, uattr_size);
4780 }
4781
4782 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4783
bpf_btf_get_fd_by_id(const union bpf_attr * attr)4784 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4785 {
4786 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4787 return -EINVAL;
4788
4789 if (!capable(CAP_SYS_ADMIN))
4790 return -EPERM;
4791
4792 return btf_get_fd_by_id(attr->btf_id);
4793 }
4794
bpf_task_fd_query_copy(const union bpf_attr * attr,union bpf_attr __user * uattr,u32 prog_id,u32 fd_type,const char * buf,u64 probe_offset,u64 probe_addr)4795 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4796 union bpf_attr __user *uattr,
4797 u32 prog_id, u32 fd_type,
4798 const char *buf, u64 probe_offset,
4799 u64 probe_addr)
4800 {
4801 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4802 u32 len = buf ? strlen(buf) : 0, input_len;
4803 int err = 0;
4804
4805 if (put_user(len, &uattr->task_fd_query.buf_len))
4806 return -EFAULT;
4807 input_len = attr->task_fd_query.buf_len;
4808 if (input_len && ubuf) {
4809 if (!len) {
4810 /* nothing to copy, just make ubuf NULL terminated */
4811 char zero = '\0';
4812
4813 if (put_user(zero, ubuf))
4814 return -EFAULT;
4815 } else if (input_len >= len + 1) {
4816 /* ubuf can hold the string with NULL terminator */
4817 if (copy_to_user(ubuf, buf, len + 1))
4818 return -EFAULT;
4819 } else {
4820 /* ubuf cannot hold the string with NULL terminator,
4821 * do a partial copy with NULL terminator.
4822 */
4823 char zero = '\0';
4824
4825 err = -ENOSPC;
4826 if (copy_to_user(ubuf, buf, input_len - 1))
4827 return -EFAULT;
4828 if (put_user(zero, ubuf + input_len - 1))
4829 return -EFAULT;
4830 }
4831 }
4832
4833 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4834 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4835 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4836 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4837 return -EFAULT;
4838
4839 return err;
4840 }
4841
4842 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4843
bpf_task_fd_query(const union bpf_attr * attr,union bpf_attr __user * uattr)4844 static int bpf_task_fd_query(const union bpf_attr *attr,
4845 union bpf_attr __user *uattr)
4846 {
4847 pid_t pid = attr->task_fd_query.pid;
4848 u32 fd = attr->task_fd_query.fd;
4849 const struct perf_event *event;
4850 struct task_struct *task;
4851 struct file *file;
4852 int err;
4853
4854 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4855 return -EINVAL;
4856
4857 if (!capable(CAP_SYS_ADMIN))
4858 return -EPERM;
4859
4860 if (attr->task_fd_query.flags != 0)
4861 return -EINVAL;
4862
4863 rcu_read_lock();
4864 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4865 rcu_read_unlock();
4866 if (!task)
4867 return -ENOENT;
4868
4869 err = 0;
4870 file = fget_task(task, fd);
4871 put_task_struct(task);
4872 if (!file)
4873 return -EBADF;
4874
4875 if (file->f_op == &bpf_link_fops) {
4876 struct bpf_link *link = file->private_data;
4877
4878 if (link->ops == &bpf_raw_tp_link_lops) {
4879 struct bpf_raw_tp_link *raw_tp =
4880 container_of(link, struct bpf_raw_tp_link, link);
4881 struct bpf_raw_event_map *btp = raw_tp->btp;
4882
4883 err = bpf_task_fd_query_copy(attr, uattr,
4884 raw_tp->link.prog->aux->id,
4885 BPF_FD_TYPE_RAW_TRACEPOINT,
4886 btp->tp->name, 0, 0);
4887 goto put_file;
4888 }
4889 goto out_not_supp;
4890 }
4891
4892 event = perf_get_event(file);
4893 if (!IS_ERR(event)) {
4894 u64 probe_offset, probe_addr;
4895 u32 prog_id, fd_type;
4896 const char *buf;
4897
4898 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4899 &buf, &probe_offset,
4900 &probe_addr);
4901 if (!err)
4902 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4903 fd_type, buf,
4904 probe_offset,
4905 probe_addr);
4906 goto put_file;
4907 }
4908
4909 out_not_supp:
4910 err = -ENOTSUPP;
4911 put_file:
4912 fput(file);
4913 return err;
4914 }
4915
4916 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4917
4918 #define BPF_DO_BATCH(fn, ...) \
4919 do { \
4920 if (!fn) { \
4921 err = -ENOTSUPP; \
4922 goto err_put; \
4923 } \
4924 err = fn(__VA_ARGS__); \
4925 } while (0)
4926
bpf_map_do_batch(const union bpf_attr * attr,union bpf_attr __user * uattr,int cmd)4927 static int bpf_map_do_batch(const union bpf_attr *attr,
4928 union bpf_attr __user *uattr,
4929 int cmd)
4930 {
4931 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
4932 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4933 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4934 struct bpf_map *map;
4935 int err, ufd;
4936 struct fd f;
4937
4938 if (CHECK_ATTR(BPF_MAP_BATCH))
4939 return -EINVAL;
4940
4941 ufd = attr->batch.map_fd;
4942 f = fdget(ufd);
4943 map = __bpf_map_get(f);
4944 if (IS_ERR(map))
4945 return PTR_ERR(map);
4946 if (has_write)
4947 bpf_map_write_active_inc(map);
4948 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4949 err = -EPERM;
4950 goto err_put;
4951 }
4952 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4953 err = -EPERM;
4954 goto err_put;
4955 }
4956
4957 if (cmd == BPF_MAP_LOOKUP_BATCH)
4958 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4959 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4960 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4961 else if (cmd == BPF_MAP_UPDATE_BATCH)
4962 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4963 else
4964 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4965 err_put:
4966 if (has_write)
4967 bpf_map_write_active_dec(map);
4968 fdput(f);
4969 return err;
4970 }
4971
4972 #define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
link_create(union bpf_attr * attr,bpfptr_t uattr)4973 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4974 {
4975 struct bpf_prog *prog;
4976 int ret;
4977
4978 if (CHECK_ATTR(BPF_LINK_CREATE))
4979 return -EINVAL;
4980
4981 if (attr->link_create.attach_type == BPF_STRUCT_OPS)
4982 return bpf_struct_ops_link_create(attr);
4983
4984 prog = bpf_prog_get(attr->link_create.prog_fd);
4985 if (IS_ERR(prog))
4986 return PTR_ERR(prog);
4987
4988 ret = bpf_prog_attach_check_attach_type(prog,
4989 attr->link_create.attach_type);
4990 if (ret)
4991 goto out;
4992
4993 switch (prog->type) {
4994 case BPF_PROG_TYPE_CGROUP_SKB:
4995 case BPF_PROG_TYPE_CGROUP_SOCK:
4996 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4997 case BPF_PROG_TYPE_SOCK_OPS:
4998 case BPF_PROG_TYPE_CGROUP_DEVICE:
4999 case BPF_PROG_TYPE_CGROUP_SYSCTL:
5000 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5001 ret = cgroup_bpf_link_attach(attr, prog);
5002 break;
5003 case BPF_PROG_TYPE_EXT:
5004 ret = bpf_tracing_prog_attach(prog,
5005 attr->link_create.target_fd,
5006 attr->link_create.target_btf_id,
5007 attr->link_create.tracing.cookie);
5008 break;
5009 case BPF_PROG_TYPE_LSM:
5010 case BPF_PROG_TYPE_TRACING:
5011 if (attr->link_create.attach_type != prog->expected_attach_type) {
5012 ret = -EINVAL;
5013 goto out;
5014 }
5015 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5016 ret = bpf_raw_tp_link_attach(prog, NULL);
5017 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5018 ret = bpf_iter_link_attach(attr, uattr, prog);
5019 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5020 ret = cgroup_bpf_link_attach(attr, prog);
5021 else
5022 ret = bpf_tracing_prog_attach(prog,
5023 attr->link_create.target_fd,
5024 attr->link_create.target_btf_id,
5025 attr->link_create.tracing.cookie);
5026 break;
5027 case BPF_PROG_TYPE_FLOW_DISSECTOR:
5028 case BPF_PROG_TYPE_SK_LOOKUP:
5029 ret = netns_bpf_link_create(attr, prog);
5030 break;
5031 #ifdef CONFIG_NET
5032 case BPF_PROG_TYPE_XDP:
5033 ret = bpf_xdp_link_attach(attr, prog);
5034 break;
5035 case BPF_PROG_TYPE_SCHED_CLS:
5036 ret = tcx_link_attach(attr, prog);
5037 break;
5038 case BPF_PROG_TYPE_NETFILTER:
5039 ret = bpf_nf_link_attach(attr, prog);
5040 break;
5041 #endif
5042 case BPF_PROG_TYPE_PERF_EVENT:
5043 case BPF_PROG_TYPE_TRACEPOINT:
5044 ret = bpf_perf_link_attach(attr, prog);
5045 break;
5046 case BPF_PROG_TYPE_KPROBE:
5047 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5048 ret = bpf_perf_link_attach(attr, prog);
5049 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI)
5050 ret = bpf_kprobe_multi_link_attach(attr, prog);
5051 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5052 ret = bpf_uprobe_multi_link_attach(attr, prog);
5053 break;
5054 default:
5055 ret = -EINVAL;
5056 }
5057
5058 out:
5059 if (ret < 0)
5060 bpf_prog_put(prog);
5061 return ret;
5062 }
5063
link_update_map(struct bpf_link * link,union bpf_attr * attr)5064 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5065 {
5066 struct bpf_map *new_map, *old_map = NULL;
5067 int ret;
5068
5069 new_map = bpf_map_get(attr->link_update.new_map_fd);
5070 if (IS_ERR(new_map))
5071 return PTR_ERR(new_map);
5072
5073 if (attr->link_update.flags & BPF_F_REPLACE) {
5074 old_map = bpf_map_get(attr->link_update.old_map_fd);
5075 if (IS_ERR(old_map)) {
5076 ret = PTR_ERR(old_map);
5077 goto out_put;
5078 }
5079 } else if (attr->link_update.old_map_fd) {
5080 ret = -EINVAL;
5081 goto out_put;
5082 }
5083
5084 ret = link->ops->update_map(link, new_map, old_map);
5085
5086 if (old_map)
5087 bpf_map_put(old_map);
5088 out_put:
5089 bpf_map_put(new_map);
5090 return ret;
5091 }
5092
5093 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5094
link_update(union bpf_attr * attr)5095 static int link_update(union bpf_attr *attr)
5096 {
5097 struct bpf_prog *old_prog = NULL, *new_prog;
5098 struct bpf_link *link;
5099 u32 flags;
5100 int ret;
5101
5102 if (CHECK_ATTR(BPF_LINK_UPDATE))
5103 return -EINVAL;
5104
5105 flags = attr->link_update.flags;
5106 if (flags & ~BPF_F_REPLACE)
5107 return -EINVAL;
5108
5109 link = bpf_link_get_from_fd(attr->link_update.link_fd);
5110 if (IS_ERR(link))
5111 return PTR_ERR(link);
5112
5113 if (link->ops->update_map) {
5114 ret = link_update_map(link, attr);
5115 goto out_put_link;
5116 }
5117
5118 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5119 if (IS_ERR(new_prog)) {
5120 ret = PTR_ERR(new_prog);
5121 goto out_put_link;
5122 }
5123
5124 if (flags & BPF_F_REPLACE) {
5125 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5126 if (IS_ERR(old_prog)) {
5127 ret = PTR_ERR(old_prog);
5128 old_prog = NULL;
5129 goto out_put_progs;
5130 }
5131 } else if (attr->link_update.old_prog_fd) {
5132 ret = -EINVAL;
5133 goto out_put_progs;
5134 }
5135
5136 if (link->ops->update_prog)
5137 ret = link->ops->update_prog(link, new_prog, old_prog);
5138 else
5139 ret = -EINVAL;
5140
5141 out_put_progs:
5142 if (old_prog)
5143 bpf_prog_put(old_prog);
5144 if (ret)
5145 bpf_prog_put(new_prog);
5146 out_put_link:
5147 bpf_link_put_direct(link);
5148 return ret;
5149 }
5150
5151 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5152
link_detach(union bpf_attr * attr)5153 static int link_detach(union bpf_attr *attr)
5154 {
5155 struct bpf_link *link;
5156 int ret;
5157
5158 if (CHECK_ATTR(BPF_LINK_DETACH))
5159 return -EINVAL;
5160
5161 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5162 if (IS_ERR(link))
5163 return PTR_ERR(link);
5164
5165 if (link->ops->detach)
5166 ret = link->ops->detach(link);
5167 else
5168 ret = -EOPNOTSUPP;
5169
5170 bpf_link_put_direct(link);
5171 return ret;
5172 }
5173
bpf_link_inc_not_zero(struct bpf_link * link)5174 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5175 {
5176 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5177 }
5178
bpf_link_by_id(u32 id)5179 struct bpf_link *bpf_link_by_id(u32 id)
5180 {
5181 struct bpf_link *link;
5182
5183 if (!id)
5184 return ERR_PTR(-ENOENT);
5185
5186 spin_lock_bh(&link_idr_lock);
5187 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5188 link = idr_find(&link_idr, id);
5189 if (link) {
5190 if (link->id)
5191 link = bpf_link_inc_not_zero(link);
5192 else
5193 link = ERR_PTR(-EAGAIN);
5194 } else {
5195 link = ERR_PTR(-ENOENT);
5196 }
5197 spin_unlock_bh(&link_idr_lock);
5198 return link;
5199 }
5200
bpf_link_get_curr_or_next(u32 * id)5201 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5202 {
5203 struct bpf_link *link;
5204
5205 spin_lock_bh(&link_idr_lock);
5206 again:
5207 link = idr_get_next(&link_idr, id);
5208 if (link) {
5209 link = bpf_link_inc_not_zero(link);
5210 if (IS_ERR(link)) {
5211 (*id)++;
5212 goto again;
5213 }
5214 }
5215 spin_unlock_bh(&link_idr_lock);
5216
5217 return link;
5218 }
5219
5220 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5221
bpf_link_get_fd_by_id(const union bpf_attr * attr)5222 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5223 {
5224 struct bpf_link *link;
5225 u32 id = attr->link_id;
5226 int fd;
5227
5228 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5229 return -EINVAL;
5230
5231 if (!capable(CAP_SYS_ADMIN))
5232 return -EPERM;
5233
5234 link = bpf_link_by_id(id);
5235 if (IS_ERR(link))
5236 return PTR_ERR(link);
5237
5238 fd = bpf_link_new_fd(link);
5239 if (fd < 0)
5240 bpf_link_put_direct(link);
5241
5242 return fd;
5243 }
5244
5245 DEFINE_MUTEX(bpf_stats_enabled_mutex);
5246
bpf_stats_release(struct inode * inode,struct file * file)5247 static int bpf_stats_release(struct inode *inode, struct file *file)
5248 {
5249 mutex_lock(&bpf_stats_enabled_mutex);
5250 static_key_slow_dec(&bpf_stats_enabled_key.key);
5251 mutex_unlock(&bpf_stats_enabled_mutex);
5252 return 0;
5253 }
5254
5255 static const struct file_operations bpf_stats_fops = {
5256 .release = bpf_stats_release,
5257 };
5258
bpf_enable_runtime_stats(void)5259 static int bpf_enable_runtime_stats(void)
5260 {
5261 int fd;
5262
5263 mutex_lock(&bpf_stats_enabled_mutex);
5264
5265 /* Set a very high limit to avoid overflow */
5266 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5267 mutex_unlock(&bpf_stats_enabled_mutex);
5268 return -EBUSY;
5269 }
5270
5271 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5272 if (fd >= 0)
5273 static_key_slow_inc(&bpf_stats_enabled_key.key);
5274
5275 mutex_unlock(&bpf_stats_enabled_mutex);
5276 return fd;
5277 }
5278
5279 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5280
bpf_enable_stats(union bpf_attr * attr)5281 static int bpf_enable_stats(union bpf_attr *attr)
5282 {
5283
5284 if (CHECK_ATTR(BPF_ENABLE_STATS))
5285 return -EINVAL;
5286
5287 if (!capable(CAP_SYS_ADMIN))
5288 return -EPERM;
5289
5290 switch (attr->enable_stats.type) {
5291 case BPF_STATS_RUN_TIME:
5292 return bpf_enable_runtime_stats();
5293 default:
5294 break;
5295 }
5296 return -EINVAL;
5297 }
5298
5299 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5300
bpf_iter_create(union bpf_attr * attr)5301 static int bpf_iter_create(union bpf_attr *attr)
5302 {
5303 struct bpf_link *link;
5304 int err;
5305
5306 if (CHECK_ATTR(BPF_ITER_CREATE))
5307 return -EINVAL;
5308
5309 if (attr->iter_create.flags)
5310 return -EINVAL;
5311
5312 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5313 if (IS_ERR(link))
5314 return PTR_ERR(link);
5315
5316 err = bpf_iter_new_fd(link);
5317 bpf_link_put_direct(link);
5318
5319 return err;
5320 }
5321
5322 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5323
bpf_prog_bind_map(union bpf_attr * attr)5324 static int bpf_prog_bind_map(union bpf_attr *attr)
5325 {
5326 struct bpf_prog *prog;
5327 struct bpf_map *map;
5328 struct bpf_map **used_maps_old, **used_maps_new;
5329 int i, ret = 0;
5330
5331 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5332 return -EINVAL;
5333
5334 if (attr->prog_bind_map.flags)
5335 return -EINVAL;
5336
5337 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5338 if (IS_ERR(prog))
5339 return PTR_ERR(prog);
5340
5341 map = bpf_map_get(attr->prog_bind_map.map_fd);
5342 if (IS_ERR(map)) {
5343 ret = PTR_ERR(map);
5344 goto out_prog_put;
5345 }
5346
5347 mutex_lock(&prog->aux->used_maps_mutex);
5348
5349 used_maps_old = prog->aux->used_maps;
5350
5351 for (i = 0; i < prog->aux->used_map_cnt; i++)
5352 if (used_maps_old[i] == map) {
5353 bpf_map_put(map);
5354 goto out_unlock;
5355 }
5356
5357 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5358 sizeof(used_maps_new[0]),
5359 GFP_KERNEL);
5360 if (!used_maps_new) {
5361 ret = -ENOMEM;
5362 goto out_unlock;
5363 }
5364
5365 /* The bpf program will not access the bpf map, but for the sake of
5366 * simplicity, increase sleepable_refcnt for sleepable program as well.
5367 */
5368 if (prog->aux->sleepable)
5369 atomic64_inc(&map->sleepable_refcnt);
5370 memcpy(used_maps_new, used_maps_old,
5371 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5372 used_maps_new[prog->aux->used_map_cnt] = map;
5373
5374 prog->aux->used_map_cnt++;
5375 prog->aux->used_maps = used_maps_new;
5376
5377 kfree(used_maps_old);
5378
5379 out_unlock:
5380 mutex_unlock(&prog->aux->used_maps_mutex);
5381
5382 if (ret)
5383 bpf_map_put(map);
5384 out_prog_put:
5385 bpf_prog_put(prog);
5386 return ret;
5387 }
5388
__sys_bpf(int cmd,bpfptr_t uattr,unsigned int size)5389 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5390 {
5391 union bpf_attr attr;
5392 int err;
5393
5394 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5395 if (err)
5396 return err;
5397 size = min_t(u32, size, sizeof(attr));
5398
5399 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5400 memset(&attr, 0, sizeof(attr));
5401 if (copy_from_bpfptr(&attr, uattr, size) != 0)
5402 return -EFAULT;
5403
5404 err = security_bpf(cmd, &attr, size);
5405 if (err < 0)
5406 return err;
5407
5408 switch (cmd) {
5409 case BPF_MAP_CREATE:
5410 err = map_create(&attr);
5411 break;
5412 case BPF_MAP_LOOKUP_ELEM:
5413 err = map_lookup_elem(&attr);
5414 break;
5415 case BPF_MAP_UPDATE_ELEM:
5416 err = map_update_elem(&attr, uattr);
5417 break;
5418 case BPF_MAP_DELETE_ELEM:
5419 err = map_delete_elem(&attr, uattr);
5420 break;
5421 case BPF_MAP_GET_NEXT_KEY:
5422 err = map_get_next_key(&attr);
5423 break;
5424 case BPF_MAP_FREEZE:
5425 err = map_freeze(&attr);
5426 break;
5427 case BPF_PROG_LOAD:
5428 err = bpf_prog_load(&attr, uattr, size);
5429 break;
5430 case BPF_OBJ_PIN:
5431 err = bpf_obj_pin(&attr);
5432 break;
5433 case BPF_OBJ_GET:
5434 err = bpf_obj_get(&attr);
5435 break;
5436 case BPF_PROG_ATTACH:
5437 err = bpf_prog_attach(&attr);
5438 break;
5439 case BPF_PROG_DETACH:
5440 err = bpf_prog_detach(&attr);
5441 break;
5442 case BPF_PROG_QUERY:
5443 err = bpf_prog_query(&attr, uattr.user);
5444 break;
5445 case BPF_PROG_TEST_RUN:
5446 err = bpf_prog_test_run(&attr, uattr.user);
5447 break;
5448 case BPF_PROG_GET_NEXT_ID:
5449 err = bpf_obj_get_next_id(&attr, uattr.user,
5450 &prog_idr, &prog_idr_lock);
5451 break;
5452 case BPF_MAP_GET_NEXT_ID:
5453 err = bpf_obj_get_next_id(&attr, uattr.user,
5454 &map_idr, &map_idr_lock);
5455 break;
5456 case BPF_BTF_GET_NEXT_ID:
5457 err = bpf_obj_get_next_id(&attr, uattr.user,
5458 &btf_idr, &btf_idr_lock);
5459 break;
5460 case BPF_PROG_GET_FD_BY_ID:
5461 err = bpf_prog_get_fd_by_id(&attr);
5462 break;
5463 case BPF_MAP_GET_FD_BY_ID:
5464 err = bpf_map_get_fd_by_id(&attr);
5465 break;
5466 case BPF_OBJ_GET_INFO_BY_FD:
5467 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5468 break;
5469 case BPF_RAW_TRACEPOINT_OPEN:
5470 err = bpf_raw_tracepoint_open(&attr);
5471 break;
5472 case BPF_BTF_LOAD:
5473 err = bpf_btf_load(&attr, uattr, size);
5474 break;
5475 case BPF_BTF_GET_FD_BY_ID:
5476 err = bpf_btf_get_fd_by_id(&attr);
5477 break;
5478 case BPF_TASK_FD_QUERY:
5479 err = bpf_task_fd_query(&attr, uattr.user);
5480 break;
5481 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5482 err = map_lookup_and_delete_elem(&attr);
5483 break;
5484 case BPF_MAP_LOOKUP_BATCH:
5485 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5486 break;
5487 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5488 err = bpf_map_do_batch(&attr, uattr.user,
5489 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5490 break;
5491 case BPF_MAP_UPDATE_BATCH:
5492 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5493 break;
5494 case BPF_MAP_DELETE_BATCH:
5495 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5496 break;
5497 case BPF_LINK_CREATE:
5498 err = link_create(&attr, uattr);
5499 break;
5500 case BPF_LINK_UPDATE:
5501 err = link_update(&attr);
5502 break;
5503 case BPF_LINK_GET_FD_BY_ID:
5504 err = bpf_link_get_fd_by_id(&attr);
5505 break;
5506 case BPF_LINK_GET_NEXT_ID:
5507 err = bpf_obj_get_next_id(&attr, uattr.user,
5508 &link_idr, &link_idr_lock);
5509 break;
5510 case BPF_ENABLE_STATS:
5511 err = bpf_enable_stats(&attr);
5512 break;
5513 case BPF_ITER_CREATE:
5514 err = bpf_iter_create(&attr);
5515 break;
5516 case BPF_LINK_DETACH:
5517 err = link_detach(&attr);
5518 break;
5519 case BPF_PROG_BIND_MAP:
5520 err = bpf_prog_bind_map(&attr);
5521 break;
5522 default:
5523 err = -EINVAL;
5524 break;
5525 }
5526
5527 return err;
5528 }
5529
SYSCALL_DEFINE3(bpf,int,cmd,union bpf_attr __user *,uattr,unsigned int,size)5530 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5531 {
5532 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5533 }
5534
syscall_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)5535 static bool syscall_prog_is_valid_access(int off, int size,
5536 enum bpf_access_type type,
5537 const struct bpf_prog *prog,
5538 struct bpf_insn_access_aux *info)
5539 {
5540 if (off < 0 || off >= U16_MAX)
5541 return false;
5542 if (off % size != 0)
5543 return false;
5544 return true;
5545 }
5546
BPF_CALL_3(bpf_sys_bpf,int,cmd,union bpf_attr *,attr,u32,attr_size)5547 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5548 {
5549 switch (cmd) {
5550 case BPF_MAP_CREATE:
5551 case BPF_MAP_DELETE_ELEM:
5552 case BPF_MAP_UPDATE_ELEM:
5553 case BPF_MAP_FREEZE:
5554 case BPF_MAP_GET_FD_BY_ID:
5555 case BPF_PROG_LOAD:
5556 case BPF_BTF_LOAD:
5557 case BPF_LINK_CREATE:
5558 case BPF_RAW_TRACEPOINT_OPEN:
5559 break;
5560 default:
5561 return -EINVAL;
5562 }
5563 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5564 }
5565
5566
5567 /* To shut up -Wmissing-prototypes.
5568 * This function is used by the kernel light skeleton
5569 * to load bpf programs when modules are loaded or during kernel boot.
5570 * See tools/lib/bpf/skel_internal.h
5571 */
5572 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5573
kern_sys_bpf(int cmd,union bpf_attr * attr,unsigned int size)5574 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5575 {
5576 struct bpf_prog * __maybe_unused prog;
5577 struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5578
5579 switch (cmd) {
5580 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5581 case BPF_PROG_TEST_RUN:
5582 if (attr->test.data_in || attr->test.data_out ||
5583 attr->test.ctx_out || attr->test.duration ||
5584 attr->test.repeat || attr->test.flags)
5585 return -EINVAL;
5586
5587 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5588 if (IS_ERR(prog))
5589 return PTR_ERR(prog);
5590
5591 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5592 attr->test.ctx_size_in > U16_MAX) {
5593 bpf_prog_put(prog);
5594 return -EINVAL;
5595 }
5596
5597 run_ctx.bpf_cookie = 0;
5598 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5599 /* recursion detected */
5600 __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5601 bpf_prog_put(prog);
5602 return -EBUSY;
5603 }
5604 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5605 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5606 &run_ctx);
5607 bpf_prog_put(prog);
5608 return 0;
5609 #endif
5610 default:
5611 return ____bpf_sys_bpf(cmd, attr, size);
5612 }
5613 }
5614 EXPORT_SYMBOL(kern_sys_bpf);
5615
5616 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5617 .func = bpf_sys_bpf,
5618 .gpl_only = false,
5619 .ret_type = RET_INTEGER,
5620 .arg1_type = ARG_ANYTHING,
5621 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5622 .arg3_type = ARG_CONST_SIZE,
5623 };
5624
5625 const struct bpf_func_proto * __weak
tracing_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)5626 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5627 {
5628 return bpf_base_func_proto(func_id);
5629 }
5630
BPF_CALL_1(bpf_sys_close,u32,fd)5631 BPF_CALL_1(bpf_sys_close, u32, fd)
5632 {
5633 /* When bpf program calls this helper there should not be
5634 * an fdget() without matching completed fdput().
5635 * This helper is allowed in the following callchain only:
5636 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5637 */
5638 return close_fd(fd);
5639 }
5640
5641 static const struct bpf_func_proto bpf_sys_close_proto = {
5642 .func = bpf_sys_close,
5643 .gpl_only = false,
5644 .ret_type = RET_INTEGER,
5645 .arg1_type = ARG_ANYTHING,
5646 };
5647
BPF_CALL_4(bpf_kallsyms_lookup_name,const char *,name,int,name_sz,int,flags,u64 *,res)5648 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5649 {
5650 if (flags)
5651 return -EINVAL;
5652
5653 if (name_sz <= 1 || name[name_sz - 1])
5654 return -EINVAL;
5655
5656 if (!bpf_dump_raw_ok(current_cred()))
5657 return -EPERM;
5658
5659 *res = kallsyms_lookup_name(name);
5660 return *res ? 0 : -ENOENT;
5661 }
5662
5663 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5664 .func = bpf_kallsyms_lookup_name,
5665 .gpl_only = false,
5666 .ret_type = RET_INTEGER,
5667 .arg1_type = ARG_PTR_TO_MEM,
5668 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
5669 .arg3_type = ARG_ANYTHING,
5670 .arg4_type = ARG_PTR_TO_LONG,
5671 };
5672
5673 static const struct bpf_func_proto *
syscall_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)5674 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5675 {
5676 switch (func_id) {
5677 case BPF_FUNC_sys_bpf:
5678 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5679 case BPF_FUNC_btf_find_by_name_kind:
5680 return &bpf_btf_find_by_name_kind_proto;
5681 case BPF_FUNC_sys_close:
5682 return &bpf_sys_close_proto;
5683 case BPF_FUNC_kallsyms_lookup_name:
5684 return &bpf_kallsyms_lookup_name_proto;
5685 default:
5686 return tracing_prog_func_proto(func_id, prog);
5687 }
5688 }
5689
5690 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5691 .get_func_proto = syscall_prog_func_proto,
5692 .is_valid_access = syscall_prog_is_valid_access,
5693 };
5694
5695 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5696 .test_run = bpf_prog_test_run_syscall,
5697 };
5698
5699 #ifdef CONFIG_SYSCTL
bpf_stats_handler(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)5700 static int bpf_stats_handler(struct ctl_table *table, int write,
5701 void *buffer, size_t *lenp, loff_t *ppos)
5702 {
5703 struct static_key *key = (struct static_key *)table->data;
5704 static int saved_val;
5705 int val, ret;
5706 struct ctl_table tmp = {
5707 .data = &val,
5708 .maxlen = sizeof(val),
5709 .mode = table->mode,
5710 .extra1 = SYSCTL_ZERO,
5711 .extra2 = SYSCTL_ONE,
5712 };
5713
5714 if (write && !capable(CAP_SYS_ADMIN))
5715 return -EPERM;
5716
5717 mutex_lock(&bpf_stats_enabled_mutex);
5718 val = saved_val;
5719 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5720 if (write && !ret && val != saved_val) {
5721 if (val)
5722 static_key_slow_inc(key);
5723 else
5724 static_key_slow_dec(key);
5725 saved_val = val;
5726 }
5727 mutex_unlock(&bpf_stats_enabled_mutex);
5728 return ret;
5729 }
5730
unpriv_ebpf_notify(int new_state)5731 void __weak unpriv_ebpf_notify(int new_state)
5732 {
5733 }
5734
bpf_unpriv_handler(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)5735 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5736 void *buffer, size_t *lenp, loff_t *ppos)
5737 {
5738 int ret, unpriv_enable = *(int *)table->data;
5739 bool locked_state = unpriv_enable == 1;
5740 struct ctl_table tmp = *table;
5741
5742 if (write && !capable(CAP_SYS_ADMIN))
5743 return -EPERM;
5744
5745 tmp.data = &unpriv_enable;
5746 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5747 if (write && !ret) {
5748 if (locked_state && unpriv_enable != 1)
5749 return -EPERM;
5750 *(int *)table->data = unpriv_enable;
5751 }
5752
5753 if (write)
5754 unpriv_ebpf_notify(unpriv_enable);
5755
5756 return ret;
5757 }
5758
5759 static struct ctl_table bpf_syscall_table[] = {
5760 {
5761 .procname = "unprivileged_bpf_disabled",
5762 .data = &sysctl_unprivileged_bpf_disabled,
5763 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
5764 .mode = 0644,
5765 .proc_handler = bpf_unpriv_handler,
5766 .extra1 = SYSCTL_ZERO,
5767 .extra2 = SYSCTL_TWO,
5768 },
5769 {
5770 .procname = "bpf_stats_enabled",
5771 .data = &bpf_stats_enabled_key.key,
5772 .mode = 0644,
5773 .proc_handler = bpf_stats_handler,
5774 },
5775 { }
5776 };
5777
bpf_syscall_sysctl_init(void)5778 static int __init bpf_syscall_sysctl_init(void)
5779 {
5780 register_sysctl_init("kernel", bpf_syscall_table);
5781 return 0;
5782 }
5783 late_initcall(bpf_syscall_sysctl_init);
5784 #endif /* CONFIG_SYSCTL */
5785