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