1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
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>
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>
39 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
40 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
41 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
42 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
43 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
44 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
47 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
49 DEFINE_PER_CPU(int, bpf_prog_active);
50 static DEFINE_IDR(prog_idr);
51 static DEFINE_SPINLOCK(prog_idr_lock);
52 static DEFINE_IDR(map_idr);
53 static DEFINE_SPINLOCK(map_idr_lock);
54 static DEFINE_IDR(link_idr);
55 static DEFINE_SPINLOCK(link_idr_lock);
57 int sysctl_unprivileged_bpf_disabled __read_mostly =
58 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
60 static const struct bpf_map_ops * const bpf_map_types[] = {
61 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
62 #define BPF_MAP_TYPE(_id, _ops) \
64 #define BPF_LINK_TYPE(_id, _name)
65 #include <linux/bpf_types.h>
72 * If we're handed a bigger struct than we know of, ensure all the unknown bits
73 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
74 * we don't know about yet.
76 * There is a ToCToU between this function call and the following
77 * copy_from_user() call. However, this is not a concern since this function is
78 * meant to be a future-proofing of bits.
80 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
86 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
89 if (actual_size <= expected_size)
93 res = memchr_inv(uaddr.kernel + expected_size, 0,
94 actual_size - expected_size) == NULL;
96 res = check_zeroed_user(uaddr.user + expected_size,
97 actual_size - expected_size);
100 return res ? 0 : -E2BIG;
103 const struct bpf_map_ops bpf_map_offload_ops = {
104 .map_meta_equal = bpf_map_meta_equal,
105 .map_alloc = bpf_map_offload_map_alloc,
106 .map_free = bpf_map_offload_map_free,
107 .map_check_btf = map_check_no_btf,
110 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
112 const struct bpf_map_ops *ops;
113 u32 type = attr->map_type;
117 if (type >= ARRAY_SIZE(bpf_map_types))
118 return ERR_PTR(-EINVAL);
119 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
120 ops = bpf_map_types[type];
122 return ERR_PTR(-EINVAL);
124 if (ops->map_alloc_check) {
125 err = ops->map_alloc_check(attr);
129 if (attr->map_ifindex)
130 ops = &bpf_map_offload_ops;
131 map = ops->map_alloc(attr);
135 map->map_type = type;
139 static void bpf_map_write_active_inc(struct bpf_map *map)
141 atomic64_inc(&map->writecnt);
144 static void bpf_map_write_active_dec(struct bpf_map *map)
146 atomic64_dec(&map->writecnt);
149 bool bpf_map_write_active(const struct bpf_map *map)
151 return atomic64_read(&map->writecnt) != 0;
154 static u32 bpf_map_value_size(const struct bpf_map *map)
156 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
157 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
158 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
159 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
160 return round_up(map->value_size, 8) * num_possible_cpus();
161 else if (IS_FD_MAP(map))
164 return map->value_size;
167 static void maybe_wait_bpf_programs(struct bpf_map *map)
169 /* Wait for any running BPF programs to complete so that
170 * userspace, when we return to it, knows that all programs
171 * that could be running use the new map value.
173 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
174 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
178 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
179 void *key, void *value, __u64 flags)
183 /* Need to create a kthread, thus must support schedule */
184 if (bpf_map_is_offloaded(map)) {
185 return bpf_map_offload_update_elem(map, key, value, flags);
186 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
187 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
188 return map->ops->map_update_elem(map, key, value, flags);
189 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
190 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
191 return sock_map_update_elem_sys(map, key, value, flags);
192 } else if (IS_FD_PROG_ARRAY(map)) {
193 return bpf_fd_array_map_update_elem(map, map_file, key, value,
197 bpf_disable_instrumentation();
198 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
199 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
200 err = bpf_percpu_hash_update(map, key, value, flags);
201 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
202 err = bpf_percpu_array_update(map, key, value, flags);
203 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
204 err = bpf_percpu_cgroup_storage_update(map, key, value,
206 } else if (IS_FD_ARRAY(map)) {
208 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
211 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
213 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
216 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
217 /* rcu_read_lock() is not needed */
218 err = bpf_fd_reuseport_array_update_elem(map, key, value,
220 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
221 map->map_type == BPF_MAP_TYPE_STACK ||
222 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
223 err = map->ops->map_push_elem(map, value, flags);
226 err = map->ops->map_update_elem(map, key, value, flags);
229 bpf_enable_instrumentation();
230 maybe_wait_bpf_programs(map);
235 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
241 if (bpf_map_is_offloaded(map))
242 return bpf_map_offload_lookup_elem(map, key, value);
244 bpf_disable_instrumentation();
245 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
246 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
247 err = bpf_percpu_hash_copy(map, key, value);
248 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
249 err = bpf_percpu_array_copy(map, key, value);
250 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
251 err = bpf_percpu_cgroup_storage_copy(map, key, value);
252 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
253 err = bpf_stackmap_copy(map, key, value);
254 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
255 err = bpf_fd_array_map_lookup_elem(map, key, value);
256 } else if (IS_FD_HASH(map)) {
257 err = bpf_fd_htab_map_lookup_elem(map, key, value);
258 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
259 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
260 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
261 map->map_type == BPF_MAP_TYPE_STACK ||
262 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
263 err = map->ops->map_peek_elem(map, value);
264 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
265 /* struct_ops map requires directly updating "value" */
266 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
269 if (map->ops->map_lookup_elem_sys_only)
270 ptr = map->ops->map_lookup_elem_sys_only(map, key);
272 ptr = map->ops->map_lookup_elem(map, key);
279 if (flags & BPF_F_LOCK)
280 /* lock 'ptr' and copy everything but lock */
281 copy_map_value_locked(map, value, ptr, true);
283 copy_map_value(map, value, ptr);
284 /* mask lock and timer, since value wasn't zero inited */
285 check_and_init_map_value(map, value);
290 bpf_enable_instrumentation();
291 maybe_wait_bpf_programs(map);
296 /* Please, do not use this function outside from the map creation path
297 * (e.g. in map update path) without taking care of setting the active
298 * memory cgroup (see at bpf_map_kmalloc_node() for example).
300 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
302 /* We really just want to fail instead of triggering OOM killer
303 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
304 * which is used for lower order allocation requests.
306 * It has been observed that higher order allocation requests done by
307 * vmalloc with __GFP_NORETRY being set might fail due to not trying
308 * to reclaim memory from the page cache, thus we set
309 * __GFP_RETRY_MAYFAIL to avoid such situations.
312 gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
313 unsigned int flags = 0;
314 unsigned long align = 1;
317 if (size >= SIZE_MAX)
320 /* kmalloc()'ed memory can't be mmap()'ed */
322 BUG_ON(!PAGE_ALIGNED(size));
325 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
326 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
332 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
333 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
334 flags, numa_node, __builtin_return_address(0));
337 void *bpf_map_area_alloc(u64 size, int numa_node)
339 return __bpf_map_area_alloc(size, numa_node, false);
342 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
344 return __bpf_map_area_alloc(size, numa_node, true);
347 void bpf_map_area_free(void *area)
352 static u32 bpf_map_flags_retain_permanent(u32 flags)
354 /* Some map creation flags are not tied to the map object but
355 * rather to the map fd instead, so they have no meaning upon
356 * map object inspection since multiple file descriptors with
357 * different (access) properties can exist here. Thus, given
358 * this has zero meaning for the map itself, lets clear these
361 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
364 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
366 map->map_type = attr->map_type;
367 map->key_size = attr->key_size;
368 map->value_size = attr->value_size;
369 map->max_entries = attr->max_entries;
370 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
371 map->numa_node = bpf_map_attr_numa_node(attr);
372 map->map_extra = attr->map_extra;
375 static int bpf_map_alloc_id(struct bpf_map *map)
379 idr_preload(GFP_KERNEL);
380 spin_lock_bh(&map_idr_lock);
381 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
384 spin_unlock_bh(&map_idr_lock);
387 if (WARN_ON_ONCE(!id))
390 return id > 0 ? 0 : id;
393 void bpf_map_free_id(struct bpf_map *map)
397 /* Offloaded maps are removed from the IDR store when their device
398 * disappears - even if someone holds an fd to them they are unusable,
399 * the memory is gone, all ops will fail; they are simply waiting for
400 * refcnt to drop to be freed.
405 spin_lock_irqsave(&map_idr_lock, flags);
407 idr_remove(&map_idr, map->id);
410 spin_unlock_irqrestore(&map_idr_lock, flags);
413 #ifdef CONFIG_MEMCG_KMEM
414 static void bpf_map_save_memcg(struct bpf_map *map)
416 /* Currently if a map is created by a process belonging to the root
417 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
418 * So we have to check map->objcg for being NULL each time it's
421 if (memcg_bpf_enabled())
422 map->objcg = get_obj_cgroup_from_current();
425 static void bpf_map_release_memcg(struct bpf_map *map)
428 obj_cgroup_put(map->objcg);
431 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
434 return get_mem_cgroup_from_objcg(map->objcg);
436 return root_mem_cgroup;
439 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
442 struct mem_cgroup *memcg, *old_memcg;
445 memcg = bpf_map_get_memcg(map);
446 old_memcg = set_active_memcg(memcg);
447 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
448 set_active_memcg(old_memcg);
449 mem_cgroup_put(memcg);
454 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
456 struct mem_cgroup *memcg, *old_memcg;
459 memcg = bpf_map_get_memcg(map);
460 old_memcg = set_active_memcg(memcg);
461 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
462 set_active_memcg(old_memcg);
463 mem_cgroup_put(memcg);
468 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
471 struct mem_cgroup *memcg, *old_memcg;
474 memcg = bpf_map_get_memcg(map);
475 old_memcg = set_active_memcg(memcg);
476 ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
477 set_active_memcg(old_memcg);
478 mem_cgroup_put(memcg);
483 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
484 size_t align, gfp_t flags)
486 struct mem_cgroup *memcg, *old_memcg;
489 memcg = bpf_map_get_memcg(map);
490 old_memcg = set_active_memcg(memcg);
491 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
492 set_active_memcg(old_memcg);
493 mem_cgroup_put(memcg);
499 static void bpf_map_save_memcg(struct bpf_map *map)
503 static void bpf_map_release_memcg(struct bpf_map *map)
508 static int btf_field_cmp(const void *a, const void *b)
510 const struct btf_field *f1 = a, *f2 = b;
512 if (f1->offset < f2->offset)
514 else if (f1->offset > f2->offset)
519 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
520 enum btf_field_type type)
522 struct btf_field *field;
524 if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & type))
526 field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
527 if (!field || !(field->type & type))
532 void btf_record_free(struct btf_record *rec)
536 if (IS_ERR_OR_NULL(rec))
538 for (i = 0; i < rec->cnt; i++) {
539 switch (rec->fields[i].type) {
542 if (rec->fields[i].kptr.module)
543 module_put(rec->fields[i].kptr.module);
544 btf_put(rec->fields[i].kptr.btf);
552 /* Nothing to release */
562 void bpf_map_free_record(struct bpf_map *map)
564 btf_record_free(map->record);
568 struct btf_record *btf_record_dup(const struct btf_record *rec)
570 const struct btf_field *fields;
571 struct btf_record *new_rec;
574 if (IS_ERR_OR_NULL(rec))
576 size = offsetof(struct btf_record, fields[rec->cnt]);
577 new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
579 return ERR_PTR(-ENOMEM);
580 /* Do a deep copy of the btf_record */
581 fields = rec->fields;
583 for (i = 0; i < rec->cnt; i++) {
584 switch (fields[i].type) {
587 btf_get(fields[i].kptr.btf);
588 if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
599 /* Nothing to acquire */
610 btf_record_free(new_rec);
614 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
616 bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
619 if (!a_has_fields && !b_has_fields)
621 if (a_has_fields != b_has_fields)
623 if (rec_a->cnt != rec_b->cnt)
625 size = offsetof(struct btf_record, fields[rec_a->cnt]);
626 /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
627 * members are zeroed out. So memcmp is safe to do without worrying
628 * about padding/unused fields.
630 * While spin_lock, timer, and kptr have no relation to map BTF,
631 * list_head metadata is specific to map BTF, the btf and value_rec
632 * members in particular. btf is the map BTF, while value_rec points to
633 * btf_record in that map BTF.
635 * So while by default, we don't rely on the map BTF (which the records
636 * were parsed from) matching for both records, which is not backwards
637 * compatible, in case list_head is part of it, we implicitly rely on
638 * that by way of depending on memcmp succeeding for it.
640 return !memcmp(rec_a, rec_b, size);
643 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
645 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
647 bpf_timer_cancel_and_free(obj + rec->timer_off);
650 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
652 const struct btf_field *fields;
655 if (IS_ERR_OR_NULL(rec))
657 fields = rec->fields;
658 for (i = 0; i < rec->cnt; i++) {
659 const struct btf_field *field = &fields[i];
660 void *field_ptr = obj + field->offset;
662 switch (fields[i].type) {
666 bpf_timer_cancel_and_free(field_ptr);
669 WRITE_ONCE(*(u64 *)field_ptr, 0);
672 field->kptr.dtor((void *)xchg((unsigned long *)field_ptr, 0));
675 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
677 bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
680 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
682 bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
694 /* called from workqueue */
695 static void bpf_map_free_deferred(struct work_struct *work)
697 struct bpf_map *map = container_of(work, struct bpf_map, work);
698 struct btf_field_offs *foffs = map->field_offs;
699 struct btf_record *rec = map->record;
701 security_bpf_map_free(map);
702 bpf_map_release_memcg(map);
703 /* implementation dependent freeing */
704 map->ops->map_free(map);
705 /* Delay freeing of field_offs and btf_record for maps, as map_free
706 * callback usually needs access to them. It is better to do it here
707 * than require each callback to do the free itself manually.
709 * Note that the btf_record stashed in map->inner_map_meta->record was
710 * already freed using the map_free callback for map in map case which
711 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
712 * template bpf_map struct used during verification.
715 btf_record_free(rec);
718 static void bpf_map_put_uref(struct bpf_map *map)
720 if (atomic64_dec_and_test(&map->usercnt)) {
721 if (map->ops->map_release_uref)
722 map->ops->map_release_uref(map);
726 /* decrement map refcnt and schedule it for freeing via workqueue
727 * (underlying map implementation ops->map_free() might sleep)
729 void bpf_map_put(struct bpf_map *map)
731 if (atomic64_dec_and_test(&map->refcnt)) {
732 /* bpf_map_free_id() must be called first */
733 bpf_map_free_id(map);
735 INIT_WORK(&map->work, bpf_map_free_deferred);
736 /* Avoid spawning kworkers, since they all might contend
737 * for the same mutex like slab_mutex.
739 queue_work(system_unbound_wq, &map->work);
742 EXPORT_SYMBOL_GPL(bpf_map_put);
744 void bpf_map_put_with_uref(struct bpf_map *map)
746 bpf_map_put_uref(map);
750 static int bpf_map_release(struct inode *inode, struct file *filp)
752 struct bpf_map *map = filp->private_data;
754 if (map->ops->map_release)
755 map->ops->map_release(map, filp);
757 bpf_map_put_with_uref(map);
761 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
763 fmode_t mode = f.file->f_mode;
765 /* Our file permissions may have been overridden by global
766 * map permissions facing syscall side.
768 if (READ_ONCE(map->frozen))
769 mode &= ~FMODE_CAN_WRITE;
773 #ifdef CONFIG_PROC_FS
774 /* Provides an approximation of the map's memory footprint.
775 * Used only to provide a backward compatibility and display
776 * a reasonable "memlock" info.
778 static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
782 size = round_up(map->key_size + bpf_map_value_size(map), 8);
784 return round_up(map->max_entries * size, PAGE_SIZE);
787 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
789 struct bpf_map *map = filp->private_data;
790 u32 type = 0, jited = 0;
792 if (map_type_contains_progs(map)) {
793 spin_lock(&map->owner.lock);
794 type = map->owner.type;
795 jited = map->owner.jited;
796 spin_unlock(&map->owner.lock);
805 "map_extra:\t%#llx\n"
814 (unsigned long long)map->map_extra,
815 bpf_map_memory_footprint(map),
817 READ_ONCE(map->frozen));
819 seq_printf(m, "owner_prog_type:\t%u\n", type);
820 seq_printf(m, "owner_jited:\t%u\n", jited);
825 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
828 /* We need this handler such that alloc_file() enables
829 * f_mode with FMODE_CAN_READ.
834 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
835 size_t siz, loff_t *ppos)
837 /* We need this handler such that alloc_file() enables
838 * f_mode with FMODE_CAN_WRITE.
843 /* called for any extra memory-mapped regions (except initial) */
844 static void bpf_map_mmap_open(struct vm_area_struct *vma)
846 struct bpf_map *map = vma->vm_file->private_data;
848 if (vma->vm_flags & VM_MAYWRITE)
849 bpf_map_write_active_inc(map);
852 /* called for all unmapped memory region (including initial) */
853 static void bpf_map_mmap_close(struct vm_area_struct *vma)
855 struct bpf_map *map = vma->vm_file->private_data;
857 if (vma->vm_flags & VM_MAYWRITE)
858 bpf_map_write_active_dec(map);
861 static const struct vm_operations_struct bpf_map_default_vmops = {
862 .open = bpf_map_mmap_open,
863 .close = bpf_map_mmap_close,
866 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
868 struct bpf_map *map = filp->private_data;
871 if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
874 if (!(vma->vm_flags & VM_SHARED))
877 mutex_lock(&map->freeze_mutex);
879 if (vma->vm_flags & VM_WRITE) {
884 /* map is meant to be read-only, so do not allow mapping as
885 * writable, because it's possible to leak a writable page
886 * reference and allows user-space to still modify it after
887 * freezing, while verifier will assume contents do not change
889 if (map->map_flags & BPF_F_RDONLY_PROG) {
895 /* set default open/close callbacks */
896 vma->vm_ops = &bpf_map_default_vmops;
897 vma->vm_private_data = map;
898 vm_flags_clear(vma, VM_MAYEXEC);
899 if (!(vma->vm_flags & VM_WRITE))
900 /* disallow re-mapping with PROT_WRITE */
901 vm_flags_clear(vma, VM_MAYWRITE);
903 err = map->ops->map_mmap(map, vma);
907 if (vma->vm_flags & VM_MAYWRITE)
908 bpf_map_write_active_inc(map);
910 mutex_unlock(&map->freeze_mutex);
914 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
916 struct bpf_map *map = filp->private_data;
918 if (map->ops->map_poll)
919 return map->ops->map_poll(map, filp, pts);
924 const struct file_operations bpf_map_fops = {
925 #ifdef CONFIG_PROC_FS
926 .show_fdinfo = bpf_map_show_fdinfo,
928 .release = bpf_map_release,
929 .read = bpf_dummy_read,
930 .write = bpf_dummy_write,
931 .mmap = bpf_map_mmap,
932 .poll = bpf_map_poll,
935 int bpf_map_new_fd(struct bpf_map *map, int flags)
939 ret = security_bpf_map(map, OPEN_FMODE(flags));
943 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
947 int bpf_get_file_flag(int flags)
949 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
951 if (flags & BPF_F_RDONLY)
953 if (flags & BPF_F_WRONLY)
958 /* helper macro to check that unused fields 'union bpf_attr' are zero */
959 #define CHECK_ATTR(CMD) \
960 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
961 sizeof(attr->CMD##_LAST_FIELD), 0, \
963 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
964 sizeof(attr->CMD##_LAST_FIELD)) != NULL
966 /* dst and src must have at least "size" number of bytes.
967 * Return strlen on success and < 0 on error.
969 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
971 const char *end = src + size;
972 const char *orig_src = src;
974 memset(dst, 0, size);
975 /* Copy all isalnum(), '_' and '.' chars. */
976 while (src < end && *src) {
977 if (!isalnum(*src) &&
978 *src != '_' && *src != '.')
983 /* No '\0' found in "size" number of bytes */
987 return src - orig_src;
990 int map_check_no_btf(const struct bpf_map *map,
991 const struct btf *btf,
992 const struct btf_type *key_type,
993 const struct btf_type *value_type)
998 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
999 u32 btf_key_id, u32 btf_value_id)
1001 const struct btf_type *key_type, *value_type;
1002 u32 key_size, value_size;
1005 /* Some maps allow key to be unspecified. */
1007 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1008 if (!key_type || key_size != map->key_size)
1011 key_type = btf_type_by_id(btf, 0);
1012 if (!map->ops->map_check_btf)
1016 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1017 if (!value_type || value_size != map->value_size)
1020 map->record = btf_parse_fields(btf, value_type,
1021 BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1024 if (!IS_ERR_OR_NULL(map->record)) {
1027 if (!bpf_capable()) {
1031 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1035 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1036 switch (map->record->field_mask & (1 << i)) {
1040 if (map->map_type != BPF_MAP_TYPE_HASH &&
1041 map->map_type != BPF_MAP_TYPE_ARRAY &&
1042 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1043 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1044 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1045 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1046 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1052 if (map->map_type != BPF_MAP_TYPE_HASH &&
1053 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1054 map->map_type != BPF_MAP_TYPE_ARRAY) {
1059 case BPF_KPTR_UNREF:
1061 if (map->map_type != BPF_MAP_TYPE_HASH &&
1062 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1063 map->map_type != BPF_MAP_TYPE_ARRAY &&
1064 map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY) {
1071 if (map->map_type != BPF_MAP_TYPE_HASH &&
1072 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1073 map->map_type != BPF_MAP_TYPE_ARRAY) {
1079 /* Fail if map_type checks are missing for a field type */
1086 ret = btf_check_and_fixup_fields(btf, map->record);
1090 if (map->ops->map_check_btf) {
1091 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1098 bpf_map_free_record(map);
1102 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1103 /* called via syscall */
1104 static int map_create(union bpf_attr *attr)
1106 int numa_node = bpf_map_attr_numa_node(attr);
1107 struct btf_field_offs *foffs;
1108 struct bpf_map *map;
1112 err = CHECK_ATTR(BPF_MAP_CREATE);
1116 if (attr->btf_vmlinux_value_type_id) {
1117 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1118 attr->btf_key_type_id || attr->btf_value_type_id)
1120 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1124 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1125 attr->map_extra != 0)
1128 f_flags = bpf_get_file_flag(attr->map_flags);
1132 if (numa_node != NUMA_NO_NODE &&
1133 ((unsigned int)numa_node >= nr_node_ids ||
1134 !node_online(numa_node)))
1137 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1138 map = find_and_alloc_map(attr);
1140 return PTR_ERR(map);
1142 err = bpf_obj_name_cpy(map->name, attr->map_name,
1143 sizeof(attr->map_name));
1147 atomic64_set(&map->refcnt, 1);
1148 atomic64_set(&map->usercnt, 1);
1149 mutex_init(&map->freeze_mutex);
1150 spin_lock_init(&map->owner.lock);
1152 if (attr->btf_key_type_id || attr->btf_value_type_id ||
1153 /* Even the map's value is a kernel's struct,
1154 * the bpf_prog.o must have BTF to begin with
1155 * to figure out the corresponding kernel's
1156 * counter part. Thus, attr->btf_fd has
1159 attr->btf_vmlinux_value_type_id) {
1162 btf = btf_get_by_fd(attr->btf_fd);
1167 if (btf_is_kernel(btf)) {
1174 if (attr->btf_value_type_id) {
1175 err = map_check_btf(map, btf, attr->btf_key_type_id,
1176 attr->btf_value_type_id);
1181 map->btf_key_type_id = attr->btf_key_type_id;
1182 map->btf_value_type_id = attr->btf_value_type_id;
1183 map->btf_vmlinux_value_type_id =
1184 attr->btf_vmlinux_value_type_id;
1188 foffs = btf_parse_field_offs(map->record);
1189 if (IS_ERR(foffs)) {
1190 err = PTR_ERR(foffs);
1193 map->field_offs = foffs;
1195 err = security_bpf_map_alloc(map);
1197 goto free_map_field_offs;
1199 err = bpf_map_alloc_id(map);
1203 bpf_map_save_memcg(map);
1205 err = bpf_map_new_fd(map, f_flags);
1207 /* failed to allocate fd.
1208 * bpf_map_put_with_uref() is needed because the above
1209 * bpf_map_alloc_id() has published the map
1210 * to the userspace and the userspace may
1211 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1213 bpf_map_put_with_uref(map);
1220 security_bpf_map_free(map);
1221 free_map_field_offs:
1222 kfree(map->field_offs);
1225 map->ops->map_free(map);
1229 /* if error is returned, fd is released.
1230 * On success caller should complete fd access with matching fdput()
1232 struct bpf_map *__bpf_map_get(struct fd f)
1235 return ERR_PTR(-EBADF);
1236 if (f.file->f_op != &bpf_map_fops) {
1238 return ERR_PTR(-EINVAL);
1241 return f.file->private_data;
1244 void bpf_map_inc(struct bpf_map *map)
1246 atomic64_inc(&map->refcnt);
1248 EXPORT_SYMBOL_GPL(bpf_map_inc);
1250 void bpf_map_inc_with_uref(struct bpf_map *map)
1252 atomic64_inc(&map->refcnt);
1253 atomic64_inc(&map->usercnt);
1255 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1257 struct bpf_map *bpf_map_get(u32 ufd)
1259 struct fd f = fdget(ufd);
1260 struct bpf_map *map;
1262 map = __bpf_map_get(f);
1271 EXPORT_SYMBOL(bpf_map_get);
1273 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1275 struct fd f = fdget(ufd);
1276 struct bpf_map *map;
1278 map = __bpf_map_get(f);
1282 bpf_map_inc_with_uref(map);
1288 /* map_idr_lock should have been held */
1289 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1293 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1295 return ERR_PTR(-ENOENT);
1297 atomic64_inc(&map->usercnt);
1302 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1304 spin_lock_bh(&map_idr_lock);
1305 map = __bpf_map_inc_not_zero(map, false);
1306 spin_unlock_bh(&map_idr_lock);
1310 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1312 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1317 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1320 return vmemdup_user(ukey, key_size);
1323 return ERR_PTR(-EINVAL);
1328 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1331 return kvmemdup_bpfptr(ukey, key_size);
1333 if (!bpfptr_is_null(ukey))
1334 return ERR_PTR(-EINVAL);
1339 /* last field in 'union bpf_attr' used by this command */
1340 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1342 static int map_lookup_elem(union bpf_attr *attr)
1344 void __user *ukey = u64_to_user_ptr(attr->key);
1345 void __user *uvalue = u64_to_user_ptr(attr->value);
1346 int ufd = attr->map_fd;
1347 struct bpf_map *map;
1353 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1356 if (attr->flags & ~BPF_F_LOCK)
1360 map = __bpf_map_get(f);
1362 return PTR_ERR(map);
1363 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1368 if ((attr->flags & BPF_F_LOCK) &&
1369 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1374 key = __bpf_copy_key(ukey, map->key_size);
1380 value_size = bpf_map_value_size(map);
1383 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1387 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1388 if (copy_from_user(value, uvalue, value_size))
1391 err = bpf_map_copy_value(map, key, value, attr->flags);
1395 err = bpf_map_copy_value(map, key, value, attr->flags);
1400 if (copy_to_user(uvalue, value, value_size) != 0)
1415 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1417 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1419 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1420 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1421 int ufd = attr->map_fd;
1422 struct bpf_map *map;
1428 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1432 map = __bpf_map_get(f);
1434 return PTR_ERR(map);
1435 bpf_map_write_active_inc(map);
1436 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1441 if ((attr->flags & BPF_F_LOCK) &&
1442 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1447 key = ___bpf_copy_key(ukey, map->key_size);
1453 value_size = bpf_map_value_size(map);
1454 value = kvmemdup_bpfptr(uvalue, value_size);
1455 if (IS_ERR(value)) {
1456 err = PTR_ERR(value);
1460 err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1466 bpf_map_write_active_dec(map);
1471 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1473 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1475 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1476 int ufd = attr->map_fd;
1477 struct bpf_map *map;
1482 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1486 map = __bpf_map_get(f);
1488 return PTR_ERR(map);
1489 bpf_map_write_active_inc(map);
1490 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1495 key = ___bpf_copy_key(ukey, map->key_size);
1501 if (bpf_map_is_offloaded(map)) {
1502 err = bpf_map_offload_delete_elem(map, key);
1504 } else if (IS_FD_PROG_ARRAY(map) ||
1505 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1506 /* These maps require sleepable context */
1507 err = map->ops->map_delete_elem(map, key);
1511 bpf_disable_instrumentation();
1513 err = map->ops->map_delete_elem(map, key);
1515 bpf_enable_instrumentation();
1516 maybe_wait_bpf_programs(map);
1520 bpf_map_write_active_dec(map);
1525 /* last field in 'union bpf_attr' used by this command */
1526 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1528 static int map_get_next_key(union bpf_attr *attr)
1530 void __user *ukey = u64_to_user_ptr(attr->key);
1531 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1532 int ufd = attr->map_fd;
1533 struct bpf_map *map;
1534 void *key, *next_key;
1538 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1542 map = __bpf_map_get(f);
1544 return PTR_ERR(map);
1545 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1551 key = __bpf_copy_key(ukey, map->key_size);
1561 next_key = kvmalloc(map->key_size, GFP_USER);
1565 if (bpf_map_is_offloaded(map)) {
1566 err = bpf_map_offload_get_next_key(map, key, next_key);
1571 err = map->ops->map_get_next_key(map, key, next_key);
1578 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1592 int generic_map_delete_batch(struct bpf_map *map,
1593 const union bpf_attr *attr,
1594 union bpf_attr __user *uattr)
1596 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1601 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1604 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1605 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1609 max_count = attr->batch.count;
1613 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1617 for (cp = 0; cp < max_count; cp++) {
1619 if (copy_from_user(key, keys + cp * map->key_size,
1623 if (bpf_map_is_offloaded(map)) {
1624 err = bpf_map_offload_delete_elem(map, key);
1628 bpf_disable_instrumentation();
1630 err = map->ops->map_delete_elem(map, key);
1632 bpf_enable_instrumentation();
1637 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1642 maybe_wait_bpf_programs(map);
1646 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1647 const union bpf_attr *attr,
1648 union bpf_attr __user *uattr)
1650 void __user *values = u64_to_user_ptr(attr->batch.values);
1651 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1652 u32 value_size, cp, max_count;
1656 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1659 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1660 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1664 value_size = bpf_map_value_size(map);
1666 max_count = attr->batch.count;
1670 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1674 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1680 for (cp = 0; cp < max_count; cp++) {
1682 if (copy_from_user(key, keys + cp * map->key_size,
1684 copy_from_user(value, values + cp * value_size, value_size))
1687 err = bpf_map_update_value(map, map_file, key, value,
1688 attr->batch.elem_flags);
1695 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1703 #define MAP_LOOKUP_RETRIES 3
1705 int generic_map_lookup_batch(struct bpf_map *map,
1706 const union bpf_attr *attr,
1707 union bpf_attr __user *uattr)
1709 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1710 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1711 void __user *values = u64_to_user_ptr(attr->batch.values);
1712 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1713 void *buf, *buf_prevkey, *prev_key, *key, *value;
1714 int err, retry = MAP_LOOKUP_RETRIES;
1715 u32 value_size, cp, max_count;
1717 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1720 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1721 !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1724 value_size = bpf_map_value_size(map);
1726 max_count = attr->batch.count;
1730 if (put_user(0, &uattr->batch.count))
1733 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1737 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1739 kvfree(buf_prevkey);
1745 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1748 value = key + map->key_size;
1750 prev_key = buf_prevkey;
1752 for (cp = 0; cp < max_count;) {
1754 err = map->ops->map_get_next_key(map, prev_key, key);
1758 err = bpf_map_copy_value(map, key, value,
1759 attr->batch.elem_flags);
1761 if (err == -ENOENT) {
1773 if (copy_to_user(keys + cp * map->key_size, key,
1778 if (copy_to_user(values + cp * value_size, value, value_size)) {
1784 prev_key = buf_prevkey;
1786 swap(prev_key, key);
1787 retry = MAP_LOOKUP_RETRIES;
1795 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1796 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1800 kvfree(buf_prevkey);
1805 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1807 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1809 void __user *ukey = u64_to_user_ptr(attr->key);
1810 void __user *uvalue = u64_to_user_ptr(attr->value);
1811 int ufd = attr->map_fd;
1812 struct bpf_map *map;
1818 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1821 if (attr->flags & ~BPF_F_LOCK)
1825 map = __bpf_map_get(f);
1827 return PTR_ERR(map);
1828 bpf_map_write_active_inc(map);
1829 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1830 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1836 (map->map_type == BPF_MAP_TYPE_QUEUE ||
1837 map->map_type == BPF_MAP_TYPE_STACK)) {
1842 if ((attr->flags & BPF_F_LOCK) &&
1843 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1848 key = __bpf_copy_key(ukey, map->key_size);
1854 value_size = bpf_map_value_size(map);
1857 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1862 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1863 map->map_type == BPF_MAP_TYPE_STACK) {
1864 err = map->ops->map_pop_elem(map, value);
1865 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1866 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1867 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1868 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1869 if (!bpf_map_is_offloaded(map)) {
1870 bpf_disable_instrumentation();
1872 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1874 bpf_enable_instrumentation();
1881 if (copy_to_user(uvalue, value, value_size) != 0) {
1893 bpf_map_write_active_dec(map);
1898 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1900 static int map_freeze(const union bpf_attr *attr)
1902 int err = 0, ufd = attr->map_fd;
1903 struct bpf_map *map;
1906 if (CHECK_ATTR(BPF_MAP_FREEZE))
1910 map = __bpf_map_get(f);
1912 return PTR_ERR(map);
1914 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
1919 mutex_lock(&map->freeze_mutex);
1920 if (bpf_map_write_active(map)) {
1924 if (READ_ONCE(map->frozen)) {
1928 if (!bpf_capable()) {
1933 WRITE_ONCE(map->frozen, true);
1935 mutex_unlock(&map->freeze_mutex);
1940 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1941 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1942 [_id] = & _name ## _prog_ops,
1943 #define BPF_MAP_TYPE(_id, _ops)
1944 #define BPF_LINK_TYPE(_id, _name)
1945 #include <linux/bpf_types.h>
1946 #undef BPF_PROG_TYPE
1948 #undef BPF_LINK_TYPE
1951 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1953 const struct bpf_prog_ops *ops;
1955 if (type >= ARRAY_SIZE(bpf_prog_types))
1957 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1958 ops = bpf_prog_types[type];
1962 if (!bpf_prog_is_offloaded(prog->aux))
1963 prog->aux->ops = ops;
1965 prog->aux->ops = &bpf_offload_prog_ops;
1976 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1977 [BPF_AUDIT_LOAD] = "LOAD",
1978 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1981 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1983 struct audit_context *ctx = NULL;
1984 struct audit_buffer *ab;
1986 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1988 if (audit_enabled == AUDIT_OFF)
1990 if (!in_irq() && !irqs_disabled())
1991 ctx = audit_context();
1992 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1995 audit_log_format(ab, "prog-id=%u op=%s",
1996 prog->aux->id, bpf_audit_str[op]);
2000 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2004 idr_preload(GFP_KERNEL);
2005 spin_lock_bh(&prog_idr_lock);
2006 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2009 spin_unlock_bh(&prog_idr_lock);
2012 /* id is in [1, INT_MAX) */
2013 if (WARN_ON_ONCE(!id))
2016 return id > 0 ? 0 : id;
2019 void bpf_prog_free_id(struct bpf_prog *prog)
2021 unsigned long flags;
2023 /* cBPF to eBPF migrations are currently not in the idr store.
2024 * Offloaded programs are removed from the store when their device
2025 * disappears - even if someone grabs an fd to them they are unusable,
2026 * simply waiting for refcnt to drop to be freed.
2031 spin_lock_irqsave(&prog_idr_lock, flags);
2032 idr_remove(&prog_idr, prog->aux->id);
2034 spin_unlock_irqrestore(&prog_idr_lock, flags);
2037 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2039 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2041 kvfree(aux->func_info);
2042 kfree(aux->func_info_aux);
2043 free_uid(aux->user);
2044 security_bpf_prog_free(aux);
2045 bpf_prog_free(aux->prog);
2048 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2050 bpf_prog_kallsyms_del_all(prog);
2051 btf_put(prog->aux->btf);
2052 kvfree(prog->aux->jited_linfo);
2053 kvfree(prog->aux->linfo);
2054 kfree(prog->aux->kfunc_tab);
2055 if (prog->aux->attach_btf)
2056 btf_put(prog->aux->attach_btf);
2059 if (prog->aux->sleepable)
2060 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2062 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2064 __bpf_prog_put_rcu(&prog->aux->rcu);
2068 static void bpf_prog_put_deferred(struct work_struct *work)
2070 struct bpf_prog_aux *aux;
2071 struct bpf_prog *prog;
2073 aux = container_of(work, struct bpf_prog_aux, work);
2075 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2076 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2077 bpf_prog_free_id(prog);
2078 __bpf_prog_put_noref(prog, true);
2081 static void __bpf_prog_put(struct bpf_prog *prog)
2083 struct bpf_prog_aux *aux = prog->aux;
2085 if (atomic64_dec_and_test(&aux->refcnt)) {
2086 if (in_irq() || irqs_disabled()) {
2087 INIT_WORK(&aux->work, bpf_prog_put_deferred);
2088 schedule_work(&aux->work);
2090 bpf_prog_put_deferred(&aux->work);
2095 void bpf_prog_put(struct bpf_prog *prog)
2097 __bpf_prog_put(prog);
2099 EXPORT_SYMBOL_GPL(bpf_prog_put);
2101 static int bpf_prog_release(struct inode *inode, struct file *filp)
2103 struct bpf_prog *prog = filp->private_data;
2109 struct bpf_prog_kstats {
2115 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2117 struct bpf_prog_stats *stats;
2120 stats = this_cpu_ptr(prog->stats);
2121 flags = u64_stats_update_begin_irqsave(&stats->syncp);
2122 u64_stats_inc(&stats->misses);
2123 u64_stats_update_end_irqrestore(&stats->syncp, flags);
2126 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2127 struct bpf_prog_kstats *stats)
2129 u64 nsecs = 0, cnt = 0, misses = 0;
2132 for_each_possible_cpu(cpu) {
2133 const struct bpf_prog_stats *st;
2135 u64 tnsecs, tcnt, tmisses;
2137 st = per_cpu_ptr(prog->stats, cpu);
2139 start = u64_stats_fetch_begin(&st->syncp);
2140 tnsecs = u64_stats_read(&st->nsecs);
2141 tcnt = u64_stats_read(&st->cnt);
2142 tmisses = u64_stats_read(&st->misses);
2143 } while (u64_stats_fetch_retry(&st->syncp, start));
2148 stats->nsecs = nsecs;
2150 stats->misses = misses;
2153 #ifdef CONFIG_PROC_FS
2154 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2156 const struct bpf_prog *prog = filp->private_data;
2157 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2158 struct bpf_prog_kstats stats;
2160 bpf_prog_get_stats(prog, &stats);
2161 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2168 "run_time_ns:\t%llu\n"
2170 "recursion_misses:\t%llu\n"
2171 "verified_insns:\t%u\n",
2175 prog->pages * 1ULL << PAGE_SHIFT,
2180 prog->aux->verified_insns);
2184 const struct file_operations bpf_prog_fops = {
2185 #ifdef CONFIG_PROC_FS
2186 .show_fdinfo = bpf_prog_show_fdinfo,
2188 .release = bpf_prog_release,
2189 .read = bpf_dummy_read,
2190 .write = bpf_dummy_write,
2193 int bpf_prog_new_fd(struct bpf_prog *prog)
2197 ret = security_bpf_prog(prog);
2201 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2202 O_RDWR | O_CLOEXEC);
2205 static struct bpf_prog *____bpf_prog_get(struct fd f)
2208 return ERR_PTR(-EBADF);
2209 if (f.file->f_op != &bpf_prog_fops) {
2211 return ERR_PTR(-EINVAL);
2214 return f.file->private_data;
2217 void bpf_prog_add(struct bpf_prog *prog, int i)
2219 atomic64_add(i, &prog->aux->refcnt);
2221 EXPORT_SYMBOL_GPL(bpf_prog_add);
2223 void bpf_prog_sub(struct bpf_prog *prog, int i)
2225 /* Only to be used for undoing previous bpf_prog_add() in some
2226 * error path. We still know that another entity in our call
2227 * path holds a reference to the program, thus atomic_sub() can
2228 * be safely used in such cases!
2230 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2232 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2234 void bpf_prog_inc(struct bpf_prog *prog)
2236 atomic64_inc(&prog->aux->refcnt);
2238 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2240 /* prog_idr_lock should have been held */
2241 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2245 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2248 return ERR_PTR(-ENOENT);
2252 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2254 bool bpf_prog_get_ok(struct bpf_prog *prog,
2255 enum bpf_prog_type *attach_type, bool attach_drv)
2257 /* not an attachment, just a refcount inc, always allow */
2261 if (prog->type != *attach_type)
2263 if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2269 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2272 struct fd f = fdget(ufd);
2273 struct bpf_prog *prog;
2275 prog = ____bpf_prog_get(f);
2278 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2279 prog = ERR_PTR(-EINVAL);
2289 struct bpf_prog *bpf_prog_get(u32 ufd)
2291 return __bpf_prog_get(ufd, NULL, false);
2294 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2297 return __bpf_prog_get(ufd, &type, attach_drv);
2299 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2301 /* Initially all BPF programs could be loaded w/o specifying
2302 * expected_attach_type. Later for some of them specifying expected_attach_type
2303 * at load time became required so that program could be validated properly.
2304 * Programs of types that are allowed to be loaded both w/ and w/o (for
2305 * backward compatibility) expected_attach_type, should have the default attach
2306 * type assigned to expected_attach_type for the latter case, so that it can be
2307 * validated later at attach time.
2309 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2310 * prog type requires it but has some attach types that have to be backward
2313 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2315 switch (attr->prog_type) {
2316 case BPF_PROG_TYPE_CGROUP_SOCK:
2317 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2318 * exist so checking for non-zero is the way to go here.
2320 if (!attr->expected_attach_type)
2321 attr->expected_attach_type =
2322 BPF_CGROUP_INET_SOCK_CREATE;
2324 case BPF_PROG_TYPE_SK_REUSEPORT:
2325 if (!attr->expected_attach_type)
2326 attr->expected_attach_type =
2327 BPF_SK_REUSEPORT_SELECT;
2333 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2334 enum bpf_attach_type expected_attach_type,
2335 struct btf *attach_btf, u32 btf_id,
2336 struct bpf_prog *dst_prog)
2339 if (btf_id > BTF_MAX_TYPE)
2342 if (!attach_btf && !dst_prog)
2345 switch (prog_type) {
2346 case BPF_PROG_TYPE_TRACING:
2347 case BPF_PROG_TYPE_LSM:
2348 case BPF_PROG_TYPE_STRUCT_OPS:
2349 case BPF_PROG_TYPE_EXT:
2356 if (attach_btf && (!btf_id || dst_prog))
2359 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2360 prog_type != BPF_PROG_TYPE_EXT)
2363 switch (prog_type) {
2364 case BPF_PROG_TYPE_CGROUP_SOCK:
2365 switch (expected_attach_type) {
2366 case BPF_CGROUP_INET_SOCK_CREATE:
2367 case BPF_CGROUP_INET_SOCK_RELEASE:
2368 case BPF_CGROUP_INET4_POST_BIND:
2369 case BPF_CGROUP_INET6_POST_BIND:
2374 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2375 switch (expected_attach_type) {
2376 case BPF_CGROUP_INET4_BIND:
2377 case BPF_CGROUP_INET6_BIND:
2378 case BPF_CGROUP_INET4_CONNECT:
2379 case BPF_CGROUP_INET6_CONNECT:
2380 case BPF_CGROUP_INET4_GETPEERNAME:
2381 case BPF_CGROUP_INET6_GETPEERNAME:
2382 case BPF_CGROUP_INET4_GETSOCKNAME:
2383 case BPF_CGROUP_INET6_GETSOCKNAME:
2384 case BPF_CGROUP_UDP4_SENDMSG:
2385 case BPF_CGROUP_UDP6_SENDMSG:
2386 case BPF_CGROUP_UDP4_RECVMSG:
2387 case BPF_CGROUP_UDP6_RECVMSG:
2392 case BPF_PROG_TYPE_CGROUP_SKB:
2393 switch (expected_attach_type) {
2394 case BPF_CGROUP_INET_INGRESS:
2395 case BPF_CGROUP_INET_EGRESS:
2400 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2401 switch (expected_attach_type) {
2402 case BPF_CGROUP_SETSOCKOPT:
2403 case BPF_CGROUP_GETSOCKOPT:
2408 case BPF_PROG_TYPE_SK_LOOKUP:
2409 if (expected_attach_type == BPF_SK_LOOKUP)
2412 case BPF_PROG_TYPE_SK_REUSEPORT:
2413 switch (expected_attach_type) {
2414 case BPF_SK_REUSEPORT_SELECT:
2415 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2420 case BPF_PROG_TYPE_SYSCALL:
2421 case BPF_PROG_TYPE_EXT:
2422 if (expected_attach_type)
2430 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2432 switch (prog_type) {
2433 case BPF_PROG_TYPE_SCHED_CLS:
2434 case BPF_PROG_TYPE_SCHED_ACT:
2435 case BPF_PROG_TYPE_XDP:
2436 case BPF_PROG_TYPE_LWT_IN:
2437 case BPF_PROG_TYPE_LWT_OUT:
2438 case BPF_PROG_TYPE_LWT_XMIT:
2439 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2440 case BPF_PROG_TYPE_SK_SKB:
2441 case BPF_PROG_TYPE_SK_MSG:
2442 case BPF_PROG_TYPE_LIRC_MODE2:
2443 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2444 case BPF_PROG_TYPE_CGROUP_DEVICE:
2445 case BPF_PROG_TYPE_CGROUP_SOCK:
2446 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2447 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2448 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2449 case BPF_PROG_TYPE_SOCK_OPS:
2450 case BPF_PROG_TYPE_EXT: /* extends any prog */
2452 case BPF_PROG_TYPE_CGROUP_SKB:
2454 case BPF_PROG_TYPE_SK_REUSEPORT:
2455 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2461 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2463 switch (prog_type) {
2464 case BPF_PROG_TYPE_KPROBE:
2465 case BPF_PROG_TYPE_TRACEPOINT:
2466 case BPF_PROG_TYPE_PERF_EVENT:
2467 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2468 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2469 case BPF_PROG_TYPE_TRACING:
2470 case BPF_PROG_TYPE_LSM:
2471 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2472 case BPF_PROG_TYPE_EXT: /* extends any prog */
2479 /* last field in 'union bpf_attr' used by this command */
2480 #define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
2482 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
2484 enum bpf_prog_type type = attr->prog_type;
2485 struct bpf_prog *prog, *dst_prog = NULL;
2486 struct btf *attach_btf = NULL;
2491 if (CHECK_ATTR(BPF_PROG_LOAD))
2494 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2495 BPF_F_ANY_ALIGNMENT |
2496 BPF_F_TEST_STATE_FREQ |
2498 BPF_F_TEST_RND_HI32 |
2499 BPF_F_XDP_HAS_FRAGS |
2500 BPF_F_XDP_DEV_BOUND_ONLY))
2503 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2504 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2508 /* copy eBPF program license from user space */
2509 if (strncpy_from_bpfptr(license,
2510 make_bpfptr(attr->license, uattr.is_kernel),
2511 sizeof(license) - 1) < 0)
2513 license[sizeof(license) - 1] = 0;
2515 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2516 is_gpl = license_is_gpl_compatible(license);
2518 if (attr->insn_cnt == 0 ||
2519 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2521 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2522 type != BPF_PROG_TYPE_CGROUP_SKB &&
2526 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2528 if (is_perfmon_prog_type(type) && !perfmon_capable())
2531 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2532 * or btf, we need to check which one it is
2534 if (attr->attach_prog_fd) {
2535 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2536 if (IS_ERR(dst_prog)) {
2538 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2539 if (IS_ERR(attach_btf))
2541 if (!btf_is_kernel(attach_btf)) {
2542 /* attaching through specifying bpf_prog's BTF
2543 * objects directly might be supported eventually
2545 btf_put(attach_btf);
2549 } else if (attr->attach_btf_id) {
2550 /* fall back to vmlinux BTF, if BTF type ID is specified */
2551 attach_btf = bpf_get_btf_vmlinux();
2552 if (IS_ERR(attach_btf))
2553 return PTR_ERR(attach_btf);
2556 btf_get(attach_btf);
2559 bpf_prog_load_fixup_attach_type(attr);
2560 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2561 attach_btf, attr->attach_btf_id,
2564 bpf_prog_put(dst_prog);
2566 btf_put(attach_btf);
2570 /* plain bpf_prog allocation */
2571 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2574 bpf_prog_put(dst_prog);
2576 btf_put(attach_btf);
2580 prog->expected_attach_type = attr->expected_attach_type;
2581 prog->aux->attach_btf = attach_btf;
2582 prog->aux->attach_btf_id = attr->attach_btf_id;
2583 prog->aux->dst_prog = dst_prog;
2584 prog->aux->dev_bound = !!attr->prog_ifindex;
2585 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2586 prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2588 err = security_bpf_prog_alloc(prog->aux);
2592 prog->aux->user = get_current_user();
2593 prog->len = attr->insn_cnt;
2596 if (copy_from_bpfptr(prog->insns,
2597 make_bpfptr(attr->insns, uattr.is_kernel),
2598 bpf_prog_insn_size(prog)) != 0)
2601 prog->orig_prog = NULL;
2604 atomic64_set(&prog->aux->refcnt, 1);
2605 prog->gpl_compatible = is_gpl ? 1 : 0;
2607 if (bpf_prog_is_dev_bound(prog->aux)) {
2608 err = bpf_prog_dev_bound_init(prog, attr);
2613 if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2614 bpf_prog_is_dev_bound(dst_prog->aux)) {
2615 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2620 /* find program type: socket_filter vs tracing_filter */
2621 err = find_prog_type(type, prog);
2625 prog->aux->load_time = ktime_get_boottime_ns();
2626 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2627 sizeof(attr->prog_name));
2631 /* run eBPF verifier */
2632 err = bpf_check(&prog, attr, uattr);
2634 goto free_used_maps;
2636 prog = bpf_prog_select_runtime(prog, &err);
2638 goto free_used_maps;
2640 err = bpf_prog_alloc_id(prog);
2642 goto free_used_maps;
2644 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2645 * effectively publicly exposed. However, retrieving via
2646 * bpf_prog_get_fd_by_id() will take another reference,
2647 * therefore it cannot be gone underneath us.
2649 * Only for the time /after/ successful bpf_prog_new_fd()
2650 * and before returning to userspace, we might just hold
2651 * one reference and any parallel close on that fd could
2652 * rip everything out. Hence, below notifications must
2653 * happen before bpf_prog_new_fd().
2655 * Also, any failure handling from this point onwards must
2656 * be using bpf_prog_put() given the program is exposed.
2658 bpf_prog_kallsyms_add(prog);
2659 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2660 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2662 err = bpf_prog_new_fd(prog);
2668 /* In case we have subprogs, we need to wait for a grace
2669 * period before we can tear down JIT memory since symbols
2670 * are already exposed under kallsyms.
2672 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2675 free_uid(prog->aux->user);
2676 security_bpf_prog_free(prog->aux);
2678 if (prog->aux->attach_btf)
2679 btf_put(prog->aux->attach_btf);
2680 bpf_prog_free(prog);
2684 #define BPF_OBJ_LAST_FIELD file_flags
2686 static int bpf_obj_pin(const union bpf_attr *attr)
2688 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2691 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2694 static int bpf_obj_get(const union bpf_attr *attr)
2696 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2697 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2700 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2704 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2705 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2707 atomic64_set(&link->refcnt, 1);
2714 static void bpf_link_free_id(int id)
2719 spin_lock_bh(&link_idr_lock);
2720 idr_remove(&link_idr, id);
2721 spin_unlock_bh(&link_idr_lock);
2724 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2725 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2726 * anon_inode's release() call. This helper marksbpf_link as
2727 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2728 * is not decremented, it's the responsibility of a calling code that failed
2729 * to complete bpf_link initialization.
2731 void bpf_link_cleanup(struct bpf_link_primer *primer)
2733 primer->link->prog = NULL;
2734 bpf_link_free_id(primer->id);
2736 put_unused_fd(primer->fd);
2739 void bpf_link_inc(struct bpf_link *link)
2741 atomic64_inc(&link->refcnt);
2744 /* bpf_link_free is guaranteed to be called from process context */
2745 static void bpf_link_free(struct bpf_link *link)
2747 bpf_link_free_id(link->id);
2749 /* detach BPF program, clean up used resources */
2750 link->ops->release(link);
2751 bpf_prog_put(link->prog);
2753 /* free bpf_link and its containing memory */
2754 link->ops->dealloc(link);
2757 static void bpf_link_put_deferred(struct work_struct *work)
2759 struct bpf_link *link = container_of(work, struct bpf_link, work);
2761 bpf_link_free(link);
2764 /* bpf_link_put can be called from atomic context, but ensures that resources
2765 * are freed from process context
2767 void bpf_link_put(struct bpf_link *link)
2769 if (!atomic64_dec_and_test(&link->refcnt))
2773 INIT_WORK(&link->work, bpf_link_put_deferred);
2774 schedule_work(&link->work);
2776 bpf_link_free(link);
2779 EXPORT_SYMBOL(bpf_link_put);
2781 static int bpf_link_release(struct inode *inode, struct file *filp)
2783 struct bpf_link *link = filp->private_data;
2789 #ifdef CONFIG_PROC_FS
2790 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2791 #define BPF_MAP_TYPE(_id, _ops)
2792 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2793 static const char *bpf_link_type_strs[] = {
2794 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2795 #include <linux/bpf_types.h>
2797 #undef BPF_PROG_TYPE
2799 #undef BPF_LINK_TYPE
2801 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2803 const struct bpf_link *link = filp->private_data;
2804 const struct bpf_prog *prog = link->prog;
2805 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2807 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2813 bpf_link_type_strs[link->type],
2817 if (link->ops->show_fdinfo)
2818 link->ops->show_fdinfo(link, m);
2822 static const struct file_operations bpf_link_fops = {
2823 #ifdef CONFIG_PROC_FS
2824 .show_fdinfo = bpf_link_show_fdinfo,
2826 .release = bpf_link_release,
2827 .read = bpf_dummy_read,
2828 .write = bpf_dummy_write,
2831 static int bpf_link_alloc_id(struct bpf_link *link)
2835 idr_preload(GFP_KERNEL);
2836 spin_lock_bh(&link_idr_lock);
2837 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2838 spin_unlock_bh(&link_idr_lock);
2844 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2845 * reserving unused FD and allocating ID from link_idr. This is to be paired
2846 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2847 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2848 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2849 * transient state is passed around in struct bpf_link_primer.
2850 * This is preferred way to create and initialize bpf_link, especially when
2851 * there are complicated and expensive operations in between creating bpf_link
2852 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2853 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2854 * expensive (and potentially failing) roll back operations in a rare case
2855 * that file, FD, or ID can't be allocated.
2857 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2862 fd = get_unused_fd_flags(O_CLOEXEC);
2867 id = bpf_link_alloc_id(link);
2873 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2875 bpf_link_free_id(id);
2877 return PTR_ERR(file);
2880 primer->link = link;
2881 primer->file = file;
2887 int bpf_link_settle(struct bpf_link_primer *primer)
2889 /* make bpf_link fetchable by ID */
2890 spin_lock_bh(&link_idr_lock);
2891 primer->link->id = primer->id;
2892 spin_unlock_bh(&link_idr_lock);
2893 /* make bpf_link fetchable by FD */
2894 fd_install(primer->fd, primer->file);
2895 /* pass through installed FD */
2899 int bpf_link_new_fd(struct bpf_link *link)
2901 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2904 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2906 struct fd f = fdget(ufd);
2907 struct bpf_link *link;
2910 return ERR_PTR(-EBADF);
2911 if (f.file->f_op != &bpf_link_fops) {
2913 return ERR_PTR(-EINVAL);
2916 link = f.file->private_data;
2922 EXPORT_SYMBOL(bpf_link_get_from_fd);
2924 static void bpf_tracing_link_release(struct bpf_link *link)
2926 struct bpf_tracing_link *tr_link =
2927 container_of(link, struct bpf_tracing_link, link.link);
2929 WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
2930 tr_link->trampoline));
2932 bpf_trampoline_put(tr_link->trampoline);
2934 /* tgt_prog is NULL if target is a kernel function */
2935 if (tr_link->tgt_prog)
2936 bpf_prog_put(tr_link->tgt_prog);
2939 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2941 struct bpf_tracing_link *tr_link =
2942 container_of(link, struct bpf_tracing_link, link.link);
2947 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2948 struct seq_file *seq)
2950 struct bpf_tracing_link *tr_link =
2951 container_of(link, struct bpf_tracing_link, link.link);
2954 "attach_type:\t%d\n",
2955 tr_link->attach_type);
2958 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2959 struct bpf_link_info *info)
2961 struct bpf_tracing_link *tr_link =
2962 container_of(link, struct bpf_tracing_link, link.link);
2964 info->tracing.attach_type = tr_link->attach_type;
2965 bpf_trampoline_unpack_key(tr_link->trampoline->key,
2966 &info->tracing.target_obj_id,
2967 &info->tracing.target_btf_id);
2972 static const struct bpf_link_ops bpf_tracing_link_lops = {
2973 .release = bpf_tracing_link_release,
2974 .dealloc = bpf_tracing_link_dealloc,
2975 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2976 .fill_link_info = bpf_tracing_link_fill_link_info,
2979 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2984 struct bpf_link_primer link_primer;
2985 struct bpf_prog *tgt_prog = NULL;
2986 struct bpf_trampoline *tr = NULL;
2987 struct bpf_tracing_link *link;
2991 switch (prog->type) {
2992 case BPF_PROG_TYPE_TRACING:
2993 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2994 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2995 prog->expected_attach_type != BPF_MODIFY_RETURN) {
3000 case BPF_PROG_TYPE_EXT:
3001 if (prog->expected_attach_type != 0) {
3006 case BPF_PROG_TYPE_LSM:
3007 if (prog->expected_attach_type != BPF_LSM_MAC) {
3017 if (!!tgt_prog_fd != !!btf_id) {
3023 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3024 if (prog->type != BPF_PROG_TYPE_EXT) {
3029 tgt_prog = bpf_prog_get(tgt_prog_fd);
3030 if (IS_ERR(tgt_prog)) {
3031 err = PTR_ERR(tgt_prog);
3036 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3039 link = kzalloc(sizeof(*link), GFP_USER);
3044 bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3045 &bpf_tracing_link_lops, prog);
3046 link->attach_type = prog->expected_attach_type;
3047 link->link.cookie = bpf_cookie;
3049 mutex_lock(&prog->aux->dst_mutex);
3051 /* There are a few possible cases here:
3053 * - if prog->aux->dst_trampoline is set, the program was just loaded
3054 * and not yet attached to anything, so we can use the values stored
3057 * - if prog->aux->dst_trampoline is NULL, the program has already been
3058 * attached to a target and its initial target was cleared (below)
3060 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3061 * target_btf_id using the link_create API.
3063 * - if tgt_prog == NULL when this function was called using the old
3064 * raw_tracepoint_open API, and we need a target from prog->aux
3066 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3067 * was detached and is going for re-attachment.
3069 if (!prog->aux->dst_trampoline && !tgt_prog) {
3071 * Allow re-attach for TRACING and LSM programs. If it's
3072 * currently linked, bpf_trampoline_link_prog will fail.
3073 * EXT programs need to specify tgt_prog_fd, so they
3074 * re-attach in separate code path.
3076 if (prog->type != BPF_PROG_TYPE_TRACING &&
3077 prog->type != BPF_PROG_TYPE_LSM) {
3081 btf_id = prog->aux->attach_btf_id;
3082 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3085 if (!prog->aux->dst_trampoline ||
3086 (key && key != prog->aux->dst_trampoline->key)) {
3087 /* If there is no saved target, or the specified target is
3088 * different from the destination specified at load time, we
3089 * need a new trampoline and a check for compatibility
3091 struct bpf_attach_target_info tgt_info = {};
3093 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3098 tr = bpf_trampoline_get(key, &tgt_info);
3104 /* The caller didn't specify a target, or the target was the
3105 * same as the destination supplied during program load. This
3106 * means we can reuse the trampoline and reference from program
3107 * load time, and there is no need to allocate a new one. This
3108 * can only happen once for any program, as the saved values in
3109 * prog->aux are cleared below.
3111 tr = prog->aux->dst_trampoline;
3112 tgt_prog = prog->aux->dst_prog;
3115 err = bpf_link_prime(&link->link.link, &link_primer);
3119 err = bpf_trampoline_link_prog(&link->link, tr);
3121 bpf_link_cleanup(&link_primer);
3126 link->tgt_prog = tgt_prog;
3127 link->trampoline = tr;
3129 /* Always clear the trampoline and target prog from prog->aux to make
3130 * sure the original attach destination is not kept alive after a
3131 * program is (re-)attached to another target.
3133 if (prog->aux->dst_prog &&
3134 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3135 /* got extra prog ref from syscall, or attaching to different prog */
3136 bpf_prog_put(prog->aux->dst_prog);
3137 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3138 /* we allocated a new trampoline, so free the old one */
3139 bpf_trampoline_put(prog->aux->dst_trampoline);
3141 prog->aux->dst_prog = NULL;
3142 prog->aux->dst_trampoline = NULL;
3143 mutex_unlock(&prog->aux->dst_mutex);
3145 return bpf_link_settle(&link_primer);
3147 if (tr && tr != prog->aux->dst_trampoline)
3148 bpf_trampoline_put(tr);
3149 mutex_unlock(&prog->aux->dst_mutex);
3152 if (tgt_prog_fd && tgt_prog)
3153 bpf_prog_put(tgt_prog);
3157 struct bpf_raw_tp_link {
3158 struct bpf_link link;
3159 struct bpf_raw_event_map *btp;
3162 static void bpf_raw_tp_link_release(struct bpf_link *link)
3164 struct bpf_raw_tp_link *raw_tp =
3165 container_of(link, struct bpf_raw_tp_link, link);
3167 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3168 bpf_put_raw_tracepoint(raw_tp->btp);
3171 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3173 struct bpf_raw_tp_link *raw_tp =
3174 container_of(link, struct bpf_raw_tp_link, link);
3179 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3180 struct seq_file *seq)
3182 struct bpf_raw_tp_link *raw_tp_link =
3183 container_of(link, struct bpf_raw_tp_link, link);
3187 raw_tp_link->btp->tp->name);
3190 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3191 struct bpf_link_info *info)
3193 struct bpf_raw_tp_link *raw_tp_link =
3194 container_of(link, struct bpf_raw_tp_link, link);
3195 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3196 const char *tp_name = raw_tp_link->btp->tp->name;
3197 u32 ulen = info->raw_tracepoint.tp_name_len;
3198 size_t tp_len = strlen(tp_name);
3203 info->raw_tracepoint.tp_name_len = tp_len + 1;
3208 if (ulen >= tp_len + 1) {
3209 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3214 if (copy_to_user(ubuf, tp_name, ulen - 1))
3216 if (put_user(zero, ubuf + ulen - 1))
3224 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3225 .release = bpf_raw_tp_link_release,
3226 .dealloc = bpf_raw_tp_link_dealloc,
3227 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3228 .fill_link_info = bpf_raw_tp_link_fill_link_info,
3231 #ifdef CONFIG_PERF_EVENTS
3232 struct bpf_perf_link {
3233 struct bpf_link link;
3234 struct file *perf_file;
3237 static void bpf_perf_link_release(struct bpf_link *link)
3239 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3240 struct perf_event *event = perf_link->perf_file->private_data;
3242 perf_event_free_bpf_prog(event);
3243 fput(perf_link->perf_file);
3246 static void bpf_perf_link_dealloc(struct bpf_link *link)
3248 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3253 static const struct bpf_link_ops bpf_perf_link_lops = {
3254 .release = bpf_perf_link_release,
3255 .dealloc = bpf_perf_link_dealloc,
3258 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3260 struct bpf_link_primer link_primer;
3261 struct bpf_perf_link *link;
3262 struct perf_event *event;
3263 struct file *perf_file;
3266 if (attr->link_create.flags)
3269 perf_file = perf_event_get(attr->link_create.target_fd);
3270 if (IS_ERR(perf_file))
3271 return PTR_ERR(perf_file);
3273 link = kzalloc(sizeof(*link), GFP_USER);
3278 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3279 link->perf_file = perf_file;
3281 err = bpf_link_prime(&link->link, &link_primer);
3287 event = perf_file->private_data;
3288 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3290 bpf_link_cleanup(&link_primer);
3293 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3296 return bpf_link_settle(&link_primer);
3303 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3307 #endif /* CONFIG_PERF_EVENTS */
3309 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3310 const char __user *user_tp_name)
3312 struct bpf_link_primer link_primer;
3313 struct bpf_raw_tp_link *link;
3314 struct bpf_raw_event_map *btp;
3315 const char *tp_name;
3319 switch (prog->type) {
3320 case BPF_PROG_TYPE_TRACING:
3321 case BPF_PROG_TYPE_EXT:
3322 case BPF_PROG_TYPE_LSM:
3324 /* The attach point for this category of programs
3325 * should be specified via btf_id during program load.
3328 if (prog->type == BPF_PROG_TYPE_TRACING &&
3329 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3330 tp_name = prog->aux->attach_func_name;
3333 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3334 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3335 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3336 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3338 buf[sizeof(buf) - 1] = 0;
3345 btp = bpf_get_raw_tracepoint(tp_name);
3349 link = kzalloc(sizeof(*link), GFP_USER);
3354 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3355 &bpf_raw_tp_link_lops, prog);
3358 err = bpf_link_prime(&link->link, &link_primer);
3364 err = bpf_probe_register(link->btp, prog);
3366 bpf_link_cleanup(&link_primer);
3370 return bpf_link_settle(&link_primer);
3373 bpf_put_raw_tracepoint(btp);
3377 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3379 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3381 struct bpf_prog *prog;
3384 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3387 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3389 return PTR_ERR(prog);
3391 fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3397 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3398 enum bpf_attach_type attach_type)
3400 switch (prog->type) {
3401 case BPF_PROG_TYPE_CGROUP_SOCK:
3402 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3403 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3404 case BPF_PROG_TYPE_SK_LOOKUP:
3405 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3406 case BPF_PROG_TYPE_CGROUP_SKB:
3407 if (!capable(CAP_NET_ADMIN))
3408 /* cg-skb progs can be loaded by unpriv user.
3409 * check permissions at attach time.
3412 return prog->enforce_expected_attach_type &&
3413 prog->expected_attach_type != attach_type ?
3420 static enum bpf_prog_type
3421 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3423 switch (attach_type) {
3424 case BPF_CGROUP_INET_INGRESS:
3425 case BPF_CGROUP_INET_EGRESS:
3426 return BPF_PROG_TYPE_CGROUP_SKB;
3427 case BPF_CGROUP_INET_SOCK_CREATE:
3428 case BPF_CGROUP_INET_SOCK_RELEASE:
3429 case BPF_CGROUP_INET4_POST_BIND:
3430 case BPF_CGROUP_INET6_POST_BIND:
3431 return BPF_PROG_TYPE_CGROUP_SOCK;
3432 case BPF_CGROUP_INET4_BIND:
3433 case BPF_CGROUP_INET6_BIND:
3434 case BPF_CGROUP_INET4_CONNECT:
3435 case BPF_CGROUP_INET6_CONNECT:
3436 case BPF_CGROUP_INET4_GETPEERNAME:
3437 case BPF_CGROUP_INET6_GETPEERNAME:
3438 case BPF_CGROUP_INET4_GETSOCKNAME:
3439 case BPF_CGROUP_INET6_GETSOCKNAME:
3440 case BPF_CGROUP_UDP4_SENDMSG:
3441 case BPF_CGROUP_UDP6_SENDMSG:
3442 case BPF_CGROUP_UDP4_RECVMSG:
3443 case BPF_CGROUP_UDP6_RECVMSG:
3444 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3445 case BPF_CGROUP_SOCK_OPS:
3446 return BPF_PROG_TYPE_SOCK_OPS;
3447 case BPF_CGROUP_DEVICE:
3448 return BPF_PROG_TYPE_CGROUP_DEVICE;
3449 case BPF_SK_MSG_VERDICT:
3450 return BPF_PROG_TYPE_SK_MSG;
3451 case BPF_SK_SKB_STREAM_PARSER:
3452 case BPF_SK_SKB_STREAM_VERDICT:
3453 case BPF_SK_SKB_VERDICT:
3454 return BPF_PROG_TYPE_SK_SKB;
3455 case BPF_LIRC_MODE2:
3456 return BPF_PROG_TYPE_LIRC_MODE2;
3457 case BPF_FLOW_DISSECTOR:
3458 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3459 case BPF_CGROUP_SYSCTL:
3460 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3461 case BPF_CGROUP_GETSOCKOPT:
3462 case BPF_CGROUP_SETSOCKOPT:
3463 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3464 case BPF_TRACE_ITER:
3465 case BPF_TRACE_RAW_TP:
3466 case BPF_TRACE_FENTRY:
3467 case BPF_TRACE_FEXIT:
3468 case BPF_MODIFY_RETURN:
3469 return BPF_PROG_TYPE_TRACING;
3471 return BPF_PROG_TYPE_LSM;
3473 return BPF_PROG_TYPE_SK_LOOKUP;
3475 return BPF_PROG_TYPE_XDP;
3476 case BPF_LSM_CGROUP:
3477 return BPF_PROG_TYPE_LSM;
3479 return BPF_PROG_TYPE_UNSPEC;
3483 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3485 #define BPF_F_ATTACH_MASK \
3486 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3488 static int bpf_prog_attach(const union bpf_attr *attr)
3490 enum bpf_prog_type ptype;
3491 struct bpf_prog *prog;
3494 if (CHECK_ATTR(BPF_PROG_ATTACH))
3497 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3500 ptype = attach_type_to_prog_type(attr->attach_type);
3501 if (ptype == BPF_PROG_TYPE_UNSPEC)
3504 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3506 return PTR_ERR(prog);
3508 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3514 case BPF_PROG_TYPE_SK_SKB:
3515 case BPF_PROG_TYPE_SK_MSG:
3516 ret = sock_map_get_from_fd(attr, prog);
3518 case BPF_PROG_TYPE_LIRC_MODE2:
3519 ret = lirc_prog_attach(attr, prog);
3521 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3522 ret = netns_bpf_prog_attach(attr, prog);
3524 case BPF_PROG_TYPE_CGROUP_DEVICE:
3525 case BPF_PROG_TYPE_CGROUP_SKB:
3526 case BPF_PROG_TYPE_CGROUP_SOCK:
3527 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3528 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3529 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3530 case BPF_PROG_TYPE_SOCK_OPS:
3531 case BPF_PROG_TYPE_LSM:
3532 if (ptype == BPF_PROG_TYPE_LSM &&
3533 prog->expected_attach_type != BPF_LSM_CGROUP)
3536 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3547 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3549 static int bpf_prog_detach(const union bpf_attr *attr)
3551 enum bpf_prog_type ptype;
3553 if (CHECK_ATTR(BPF_PROG_DETACH))
3556 ptype = attach_type_to_prog_type(attr->attach_type);
3559 case BPF_PROG_TYPE_SK_MSG:
3560 case BPF_PROG_TYPE_SK_SKB:
3561 return sock_map_prog_detach(attr, ptype);
3562 case BPF_PROG_TYPE_LIRC_MODE2:
3563 return lirc_prog_detach(attr);
3564 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3565 return netns_bpf_prog_detach(attr, ptype);
3566 case BPF_PROG_TYPE_CGROUP_DEVICE:
3567 case BPF_PROG_TYPE_CGROUP_SKB:
3568 case BPF_PROG_TYPE_CGROUP_SOCK:
3569 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3570 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3571 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3572 case BPF_PROG_TYPE_SOCK_OPS:
3573 case BPF_PROG_TYPE_LSM:
3574 return cgroup_bpf_prog_detach(attr, ptype);
3580 #define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3582 static int bpf_prog_query(const union bpf_attr *attr,
3583 union bpf_attr __user *uattr)
3585 if (!capable(CAP_NET_ADMIN))
3587 if (CHECK_ATTR(BPF_PROG_QUERY))
3589 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3592 switch (attr->query.attach_type) {
3593 case BPF_CGROUP_INET_INGRESS:
3594 case BPF_CGROUP_INET_EGRESS:
3595 case BPF_CGROUP_INET_SOCK_CREATE:
3596 case BPF_CGROUP_INET_SOCK_RELEASE:
3597 case BPF_CGROUP_INET4_BIND:
3598 case BPF_CGROUP_INET6_BIND:
3599 case BPF_CGROUP_INET4_POST_BIND:
3600 case BPF_CGROUP_INET6_POST_BIND:
3601 case BPF_CGROUP_INET4_CONNECT:
3602 case BPF_CGROUP_INET6_CONNECT:
3603 case BPF_CGROUP_INET4_GETPEERNAME:
3604 case BPF_CGROUP_INET6_GETPEERNAME:
3605 case BPF_CGROUP_INET4_GETSOCKNAME:
3606 case BPF_CGROUP_INET6_GETSOCKNAME:
3607 case BPF_CGROUP_UDP4_SENDMSG:
3608 case BPF_CGROUP_UDP6_SENDMSG:
3609 case BPF_CGROUP_UDP4_RECVMSG:
3610 case BPF_CGROUP_UDP6_RECVMSG:
3611 case BPF_CGROUP_SOCK_OPS:
3612 case BPF_CGROUP_DEVICE:
3613 case BPF_CGROUP_SYSCTL:
3614 case BPF_CGROUP_GETSOCKOPT:
3615 case BPF_CGROUP_SETSOCKOPT:
3616 case BPF_LSM_CGROUP:
3617 return cgroup_bpf_prog_query(attr, uattr);
3618 case BPF_LIRC_MODE2:
3619 return lirc_prog_query(attr, uattr);
3620 case BPF_FLOW_DISSECTOR:
3622 return netns_bpf_prog_query(attr, uattr);
3623 case BPF_SK_SKB_STREAM_PARSER:
3624 case BPF_SK_SKB_STREAM_VERDICT:
3625 case BPF_SK_MSG_VERDICT:
3626 case BPF_SK_SKB_VERDICT:
3627 return sock_map_bpf_prog_query(attr, uattr);
3633 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3635 static int bpf_prog_test_run(const union bpf_attr *attr,
3636 union bpf_attr __user *uattr)
3638 struct bpf_prog *prog;
3639 int ret = -ENOTSUPP;
3641 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3644 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3645 (!attr->test.ctx_size_in && attr->test.ctx_in))
3648 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3649 (!attr->test.ctx_size_out && attr->test.ctx_out))
3652 prog = bpf_prog_get(attr->test.prog_fd);
3654 return PTR_ERR(prog);
3656 if (prog->aux->ops->test_run)
3657 ret = prog->aux->ops->test_run(prog, attr, uattr);
3663 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3665 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3666 union bpf_attr __user *uattr,
3670 u32 next_id = attr->start_id;
3673 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3676 if (!capable(CAP_SYS_ADMIN))
3681 if (!idr_get_next(idr, &next_id))
3683 spin_unlock_bh(lock);
3686 err = put_user(next_id, &uattr->next_id);
3691 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3693 struct bpf_map *map;
3695 spin_lock_bh(&map_idr_lock);
3697 map = idr_get_next(&map_idr, id);
3699 map = __bpf_map_inc_not_zero(map, false);
3705 spin_unlock_bh(&map_idr_lock);
3710 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3712 struct bpf_prog *prog;
3714 spin_lock_bh(&prog_idr_lock);
3716 prog = idr_get_next(&prog_idr, id);
3718 prog = bpf_prog_inc_not_zero(prog);
3724 spin_unlock_bh(&prog_idr_lock);
3729 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3731 struct bpf_prog *bpf_prog_by_id(u32 id)
3733 struct bpf_prog *prog;
3736 return ERR_PTR(-ENOENT);
3738 spin_lock_bh(&prog_idr_lock);
3739 prog = idr_find(&prog_idr, id);
3741 prog = bpf_prog_inc_not_zero(prog);
3743 prog = ERR_PTR(-ENOENT);
3744 spin_unlock_bh(&prog_idr_lock);
3748 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3750 struct bpf_prog *prog;
3751 u32 id = attr->prog_id;
3754 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3757 if (!capable(CAP_SYS_ADMIN))
3760 prog = bpf_prog_by_id(id);
3762 return PTR_ERR(prog);
3764 fd = bpf_prog_new_fd(prog);
3771 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3773 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3775 struct bpf_map *map;
3776 u32 id = attr->map_id;
3780 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3781 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3784 if (!capable(CAP_SYS_ADMIN))
3787 f_flags = bpf_get_file_flag(attr->open_flags);
3791 spin_lock_bh(&map_idr_lock);
3792 map = idr_find(&map_idr, id);
3794 map = __bpf_map_inc_not_zero(map, true);
3796 map = ERR_PTR(-ENOENT);
3797 spin_unlock_bh(&map_idr_lock);
3800 return PTR_ERR(map);
3802 fd = bpf_map_new_fd(map, f_flags);
3804 bpf_map_put_with_uref(map);
3809 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3810 unsigned long addr, u32 *off,
3813 const struct bpf_map *map;
3816 mutex_lock(&prog->aux->used_maps_mutex);
3817 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3818 map = prog->aux->used_maps[i];
3819 if (map == (void *)addr) {
3820 *type = BPF_PSEUDO_MAP_FD;
3823 if (!map->ops->map_direct_value_meta)
3825 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3826 *type = BPF_PSEUDO_MAP_VALUE;
3833 mutex_unlock(&prog->aux->used_maps_mutex);
3837 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3838 const struct cred *f_cred)
3840 const struct bpf_map *map;
3841 struct bpf_insn *insns;
3847 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3852 for (i = 0; i < prog->len; i++) {
3853 code = insns[i].code;
3855 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3856 insns[i].code = BPF_JMP | BPF_CALL;
3857 insns[i].imm = BPF_FUNC_tail_call;
3860 if (code == (BPF_JMP | BPF_CALL) ||
3861 code == (BPF_JMP | BPF_CALL_ARGS)) {
3862 if (code == (BPF_JMP | BPF_CALL_ARGS))
3863 insns[i].code = BPF_JMP | BPF_CALL;
3864 if (!bpf_dump_raw_ok(f_cred))
3868 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3869 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3873 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3876 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3877 map = bpf_map_from_imm(prog, imm, &off, &type);
3879 insns[i].src_reg = type;
3880 insns[i].imm = map->id;
3881 insns[i + 1].imm = off;
3889 static int set_info_rec_size(struct bpf_prog_info *info)
3892 * Ensure info.*_rec_size is the same as kernel expected size
3896 * Only allow zero *_rec_size if both _rec_size and _cnt are
3897 * zero. In this case, the kernel will set the expected
3898 * _rec_size back to the info.
3901 if ((info->nr_func_info || info->func_info_rec_size) &&
3902 info->func_info_rec_size != sizeof(struct bpf_func_info))
3905 if ((info->nr_line_info || info->line_info_rec_size) &&
3906 info->line_info_rec_size != sizeof(struct bpf_line_info))
3909 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3910 info->jited_line_info_rec_size != sizeof(__u64))
3913 info->func_info_rec_size = sizeof(struct bpf_func_info);
3914 info->line_info_rec_size = sizeof(struct bpf_line_info);
3915 info->jited_line_info_rec_size = sizeof(__u64);
3920 static int bpf_prog_get_info_by_fd(struct file *file,
3921 struct bpf_prog *prog,
3922 const union bpf_attr *attr,
3923 union bpf_attr __user *uattr)
3925 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3926 struct btf *attach_btf = bpf_prog_get_target_btf(prog);
3927 struct bpf_prog_info info;
3928 u32 info_len = attr->info.info_len;
3929 struct bpf_prog_kstats stats;
3930 char __user *uinsns;
3934 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3937 info_len = min_t(u32, sizeof(info), info_len);
3939 memset(&info, 0, sizeof(info));
3940 if (copy_from_user(&info, uinfo, info_len))
3943 info.type = prog->type;
3944 info.id = prog->aux->id;
3945 info.load_time = prog->aux->load_time;
3946 info.created_by_uid = from_kuid_munged(current_user_ns(),
3947 prog->aux->user->uid);
3948 info.gpl_compatible = prog->gpl_compatible;
3950 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3951 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3953 mutex_lock(&prog->aux->used_maps_mutex);
3954 ulen = info.nr_map_ids;
3955 info.nr_map_ids = prog->aux->used_map_cnt;
3956 ulen = min_t(u32, info.nr_map_ids, ulen);
3958 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3961 for (i = 0; i < ulen; i++)
3962 if (put_user(prog->aux->used_maps[i]->id,
3963 &user_map_ids[i])) {
3964 mutex_unlock(&prog->aux->used_maps_mutex);
3968 mutex_unlock(&prog->aux->used_maps_mutex);
3970 err = set_info_rec_size(&info);
3974 bpf_prog_get_stats(prog, &stats);
3975 info.run_time_ns = stats.nsecs;
3976 info.run_cnt = stats.cnt;
3977 info.recursion_misses = stats.misses;
3979 info.verified_insns = prog->aux->verified_insns;
3981 if (!bpf_capable()) {
3982 info.jited_prog_len = 0;
3983 info.xlated_prog_len = 0;
3984 info.nr_jited_ksyms = 0;
3985 info.nr_jited_func_lens = 0;
3986 info.nr_func_info = 0;
3987 info.nr_line_info = 0;
3988 info.nr_jited_line_info = 0;
3992 ulen = info.xlated_prog_len;
3993 info.xlated_prog_len = bpf_prog_insn_size(prog);
3994 if (info.xlated_prog_len && ulen) {
3995 struct bpf_insn *insns_sanitized;
3998 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3999 info.xlated_prog_insns = 0;
4002 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4003 if (!insns_sanitized)
4005 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4006 ulen = min_t(u32, info.xlated_prog_len, ulen);
4007 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4008 kfree(insns_sanitized);
4013 if (bpf_prog_is_offloaded(prog->aux)) {
4014 err = bpf_prog_offload_info_fill(&info, prog);
4020 /* NOTE: the following code is supposed to be skipped for offload.
4021 * bpf_prog_offload_info_fill() is the place to fill similar fields
4024 ulen = info.jited_prog_len;
4025 if (prog->aux->func_cnt) {
4028 info.jited_prog_len = 0;
4029 for (i = 0; i < prog->aux->func_cnt; i++)
4030 info.jited_prog_len += prog->aux->func[i]->jited_len;
4032 info.jited_prog_len = prog->jited_len;
4035 if (info.jited_prog_len && ulen) {
4036 if (bpf_dump_raw_ok(file->f_cred)) {
4037 uinsns = u64_to_user_ptr(info.jited_prog_insns);
4038 ulen = min_t(u32, info.jited_prog_len, ulen);
4040 /* for multi-function programs, copy the JITed
4041 * instructions for all the functions
4043 if (prog->aux->func_cnt) {
4048 for (i = 0; i < prog->aux->func_cnt; i++) {
4049 len = prog->aux->func[i]->jited_len;
4050 len = min_t(u32, len, free);
4051 img = (u8 *) prog->aux->func[i]->bpf_func;
4052 if (copy_to_user(uinsns, img, len))
4060 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4064 info.jited_prog_insns = 0;
4068 ulen = info.nr_jited_ksyms;
4069 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4071 if (bpf_dump_raw_ok(file->f_cred)) {
4072 unsigned long ksym_addr;
4073 u64 __user *user_ksyms;
4076 /* copy the address of the kernel symbol
4077 * corresponding to each function
4079 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4080 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4081 if (prog->aux->func_cnt) {
4082 for (i = 0; i < ulen; i++) {
4083 ksym_addr = (unsigned long)
4084 prog->aux->func[i]->bpf_func;
4085 if (put_user((u64) ksym_addr,
4090 ksym_addr = (unsigned long) prog->bpf_func;
4091 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4095 info.jited_ksyms = 0;
4099 ulen = info.nr_jited_func_lens;
4100 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4102 if (bpf_dump_raw_ok(file->f_cred)) {
4103 u32 __user *user_lens;
4106 /* copy the JITed image lengths for each function */
4107 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4108 user_lens = u64_to_user_ptr(info.jited_func_lens);
4109 if (prog->aux->func_cnt) {
4110 for (i = 0; i < ulen; i++) {
4112 prog->aux->func[i]->jited_len;
4113 if (put_user(func_len, &user_lens[i]))
4117 func_len = prog->jited_len;
4118 if (put_user(func_len, &user_lens[0]))
4122 info.jited_func_lens = 0;
4127 info.btf_id = btf_obj_id(prog->aux->btf);
4128 info.attach_btf_id = prog->aux->attach_btf_id;
4130 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4132 ulen = info.nr_func_info;
4133 info.nr_func_info = prog->aux->func_info_cnt;
4134 if (info.nr_func_info && ulen) {
4135 char __user *user_finfo;
4137 user_finfo = u64_to_user_ptr(info.func_info);
4138 ulen = min_t(u32, info.nr_func_info, ulen);
4139 if (copy_to_user(user_finfo, prog->aux->func_info,
4140 info.func_info_rec_size * ulen))
4144 ulen = info.nr_line_info;
4145 info.nr_line_info = prog->aux->nr_linfo;
4146 if (info.nr_line_info && ulen) {
4147 __u8 __user *user_linfo;
4149 user_linfo = u64_to_user_ptr(info.line_info);
4150 ulen = min_t(u32, info.nr_line_info, ulen);
4151 if (copy_to_user(user_linfo, prog->aux->linfo,
4152 info.line_info_rec_size * ulen))
4156 ulen = info.nr_jited_line_info;
4157 if (prog->aux->jited_linfo)
4158 info.nr_jited_line_info = prog->aux->nr_linfo;
4160 info.nr_jited_line_info = 0;
4161 if (info.nr_jited_line_info && ulen) {
4162 if (bpf_dump_raw_ok(file->f_cred)) {
4163 unsigned long line_addr;
4164 __u64 __user *user_linfo;
4167 user_linfo = u64_to_user_ptr(info.jited_line_info);
4168 ulen = min_t(u32, info.nr_jited_line_info, ulen);
4169 for (i = 0; i < ulen; i++) {
4170 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4171 if (put_user((__u64)line_addr, &user_linfo[i]))
4175 info.jited_line_info = 0;
4179 ulen = info.nr_prog_tags;
4180 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4182 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4185 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4186 ulen = min_t(u32, info.nr_prog_tags, ulen);
4187 if (prog->aux->func_cnt) {
4188 for (i = 0; i < ulen; i++) {
4189 if (copy_to_user(user_prog_tags[i],
4190 prog->aux->func[i]->tag,
4195 if (copy_to_user(user_prog_tags[0],
4196 prog->tag, BPF_TAG_SIZE))
4202 if (copy_to_user(uinfo, &info, info_len) ||
4203 put_user(info_len, &uattr->info.info_len))
4209 static int bpf_map_get_info_by_fd(struct file *file,
4210 struct bpf_map *map,
4211 const union bpf_attr *attr,
4212 union bpf_attr __user *uattr)
4214 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4215 struct bpf_map_info info;
4216 u32 info_len = attr->info.info_len;
4219 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4222 info_len = min_t(u32, sizeof(info), info_len);
4224 memset(&info, 0, sizeof(info));
4225 info.type = map->map_type;
4227 info.key_size = map->key_size;
4228 info.value_size = map->value_size;
4229 info.max_entries = map->max_entries;
4230 info.map_flags = map->map_flags;
4231 info.map_extra = map->map_extra;
4232 memcpy(info.name, map->name, sizeof(map->name));
4235 info.btf_id = btf_obj_id(map->btf);
4236 info.btf_key_type_id = map->btf_key_type_id;
4237 info.btf_value_type_id = map->btf_value_type_id;
4239 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4241 if (bpf_map_is_offloaded(map)) {
4242 err = bpf_map_offload_info_fill(&info, map);
4247 if (copy_to_user(uinfo, &info, info_len) ||
4248 put_user(info_len, &uattr->info.info_len))
4254 static int bpf_btf_get_info_by_fd(struct file *file,
4256 const union bpf_attr *attr,
4257 union bpf_attr __user *uattr)
4259 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4260 u32 info_len = attr->info.info_len;
4263 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4267 return btf_get_info_by_fd(btf, attr, uattr);
4270 static int bpf_link_get_info_by_fd(struct file *file,
4271 struct bpf_link *link,
4272 const union bpf_attr *attr,
4273 union bpf_attr __user *uattr)
4275 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4276 struct bpf_link_info info;
4277 u32 info_len = attr->info.info_len;
4280 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4283 info_len = min_t(u32, sizeof(info), info_len);
4285 memset(&info, 0, sizeof(info));
4286 if (copy_from_user(&info, uinfo, info_len))
4289 info.type = link->type;
4291 info.prog_id = link->prog->aux->id;
4293 if (link->ops->fill_link_info) {
4294 err = link->ops->fill_link_info(link, &info);
4299 if (copy_to_user(uinfo, &info, info_len) ||
4300 put_user(info_len, &uattr->info.info_len))
4307 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4309 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4310 union bpf_attr __user *uattr)
4312 int ufd = attr->info.bpf_fd;
4316 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4323 if (f.file->f_op == &bpf_prog_fops)
4324 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4326 else if (f.file->f_op == &bpf_map_fops)
4327 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4329 else if (f.file->f_op == &btf_fops)
4330 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4331 else if (f.file->f_op == &bpf_link_fops)
4332 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4341 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
4343 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
4345 if (CHECK_ATTR(BPF_BTF_LOAD))
4351 return btf_new_fd(attr, uattr);
4354 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4356 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4358 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4361 if (!capable(CAP_SYS_ADMIN))
4364 return btf_get_fd_by_id(attr->btf_id);
4367 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4368 union bpf_attr __user *uattr,
4369 u32 prog_id, u32 fd_type,
4370 const char *buf, u64 probe_offset,
4373 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4374 u32 len = buf ? strlen(buf) : 0, input_len;
4377 if (put_user(len, &uattr->task_fd_query.buf_len))
4379 input_len = attr->task_fd_query.buf_len;
4380 if (input_len && ubuf) {
4382 /* nothing to copy, just make ubuf NULL terminated */
4385 if (put_user(zero, ubuf))
4387 } else if (input_len >= len + 1) {
4388 /* ubuf can hold the string with NULL terminator */
4389 if (copy_to_user(ubuf, buf, len + 1))
4392 /* ubuf cannot hold the string with NULL terminator,
4393 * do a partial copy with NULL terminator.
4398 if (copy_to_user(ubuf, buf, input_len - 1))
4400 if (put_user(zero, ubuf + input_len - 1))
4405 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4406 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4407 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4408 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4414 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4416 static int bpf_task_fd_query(const union bpf_attr *attr,
4417 union bpf_attr __user *uattr)
4419 pid_t pid = attr->task_fd_query.pid;
4420 u32 fd = attr->task_fd_query.fd;
4421 const struct perf_event *event;
4422 struct task_struct *task;
4426 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4429 if (!capable(CAP_SYS_ADMIN))
4432 if (attr->task_fd_query.flags != 0)
4436 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4442 file = fget_task(task, fd);
4443 put_task_struct(task);
4447 if (file->f_op == &bpf_link_fops) {
4448 struct bpf_link *link = file->private_data;
4450 if (link->ops == &bpf_raw_tp_link_lops) {
4451 struct bpf_raw_tp_link *raw_tp =
4452 container_of(link, struct bpf_raw_tp_link, link);
4453 struct bpf_raw_event_map *btp = raw_tp->btp;
4455 err = bpf_task_fd_query_copy(attr, uattr,
4456 raw_tp->link.prog->aux->id,
4457 BPF_FD_TYPE_RAW_TRACEPOINT,
4458 btp->tp->name, 0, 0);
4464 event = perf_get_event(file);
4465 if (!IS_ERR(event)) {
4466 u64 probe_offset, probe_addr;
4467 u32 prog_id, fd_type;
4470 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4471 &buf, &probe_offset,
4474 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4488 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4490 #define BPF_DO_BATCH(fn, ...) \
4496 err = fn(__VA_ARGS__); \
4499 static int bpf_map_do_batch(const union bpf_attr *attr,
4500 union bpf_attr __user *uattr,
4503 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
4504 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4505 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4506 struct bpf_map *map;
4510 if (CHECK_ATTR(BPF_MAP_BATCH))
4513 ufd = attr->batch.map_fd;
4515 map = __bpf_map_get(f);
4517 return PTR_ERR(map);
4519 bpf_map_write_active_inc(map);
4520 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4524 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4529 if (cmd == BPF_MAP_LOOKUP_BATCH)
4530 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4531 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4532 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4533 else if (cmd == BPF_MAP_UPDATE_BATCH)
4534 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4536 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4539 bpf_map_write_active_dec(map);
4544 #define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4545 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4547 enum bpf_prog_type ptype;
4548 struct bpf_prog *prog;
4551 if (CHECK_ATTR(BPF_LINK_CREATE))
4554 prog = bpf_prog_get(attr->link_create.prog_fd);
4556 return PTR_ERR(prog);
4558 ret = bpf_prog_attach_check_attach_type(prog,
4559 attr->link_create.attach_type);
4563 switch (prog->type) {
4564 case BPF_PROG_TYPE_EXT:
4566 case BPF_PROG_TYPE_PERF_EVENT:
4567 case BPF_PROG_TYPE_TRACEPOINT:
4568 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4573 case BPF_PROG_TYPE_KPROBE:
4574 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4575 attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4581 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4582 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4589 switch (prog->type) {
4590 case BPF_PROG_TYPE_CGROUP_SKB:
4591 case BPF_PROG_TYPE_CGROUP_SOCK:
4592 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4593 case BPF_PROG_TYPE_SOCK_OPS:
4594 case BPF_PROG_TYPE_CGROUP_DEVICE:
4595 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4596 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4597 ret = cgroup_bpf_link_attach(attr, prog);
4599 case BPF_PROG_TYPE_EXT:
4600 ret = bpf_tracing_prog_attach(prog,
4601 attr->link_create.target_fd,
4602 attr->link_create.target_btf_id,
4603 attr->link_create.tracing.cookie);
4605 case BPF_PROG_TYPE_LSM:
4606 case BPF_PROG_TYPE_TRACING:
4607 if (attr->link_create.attach_type != prog->expected_attach_type) {
4611 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4612 ret = bpf_raw_tp_link_attach(prog, NULL);
4613 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4614 ret = bpf_iter_link_attach(attr, uattr, prog);
4615 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4616 ret = cgroup_bpf_link_attach(attr, prog);
4618 ret = bpf_tracing_prog_attach(prog,
4619 attr->link_create.target_fd,
4620 attr->link_create.target_btf_id,
4621 attr->link_create.tracing.cookie);
4623 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4624 case BPF_PROG_TYPE_SK_LOOKUP:
4625 ret = netns_bpf_link_create(attr, prog);
4628 case BPF_PROG_TYPE_XDP:
4629 ret = bpf_xdp_link_attach(attr, prog);
4632 case BPF_PROG_TYPE_PERF_EVENT:
4633 case BPF_PROG_TYPE_TRACEPOINT:
4634 ret = bpf_perf_link_attach(attr, prog);
4636 case BPF_PROG_TYPE_KPROBE:
4637 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4638 ret = bpf_perf_link_attach(attr, prog);
4640 ret = bpf_kprobe_multi_link_attach(attr, prog);
4652 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4654 static int link_update(union bpf_attr *attr)
4656 struct bpf_prog *old_prog = NULL, *new_prog;
4657 struct bpf_link *link;
4661 if (CHECK_ATTR(BPF_LINK_UPDATE))
4664 flags = attr->link_update.flags;
4665 if (flags & ~BPF_F_REPLACE)
4668 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4670 return PTR_ERR(link);
4672 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4673 if (IS_ERR(new_prog)) {
4674 ret = PTR_ERR(new_prog);
4678 if (flags & BPF_F_REPLACE) {
4679 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4680 if (IS_ERR(old_prog)) {
4681 ret = PTR_ERR(old_prog);
4685 } else if (attr->link_update.old_prog_fd) {
4690 if (link->ops->update_prog)
4691 ret = link->ops->update_prog(link, new_prog, old_prog);
4697 bpf_prog_put(old_prog);
4699 bpf_prog_put(new_prog);
4705 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4707 static int link_detach(union bpf_attr *attr)
4709 struct bpf_link *link;
4712 if (CHECK_ATTR(BPF_LINK_DETACH))
4715 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4717 return PTR_ERR(link);
4719 if (link->ops->detach)
4720 ret = link->ops->detach(link);
4728 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4730 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4733 struct bpf_link *bpf_link_by_id(u32 id)
4735 struct bpf_link *link;
4738 return ERR_PTR(-ENOENT);
4740 spin_lock_bh(&link_idr_lock);
4741 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4742 link = idr_find(&link_idr, id);
4745 link = bpf_link_inc_not_zero(link);
4747 link = ERR_PTR(-EAGAIN);
4749 link = ERR_PTR(-ENOENT);
4751 spin_unlock_bh(&link_idr_lock);
4755 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4757 struct bpf_link *link;
4759 spin_lock_bh(&link_idr_lock);
4761 link = idr_get_next(&link_idr, id);
4763 link = bpf_link_inc_not_zero(link);
4769 spin_unlock_bh(&link_idr_lock);
4774 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4776 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4778 struct bpf_link *link;
4779 u32 id = attr->link_id;
4782 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4785 if (!capable(CAP_SYS_ADMIN))
4788 link = bpf_link_by_id(id);
4790 return PTR_ERR(link);
4792 fd = bpf_link_new_fd(link);
4799 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4801 static int bpf_stats_release(struct inode *inode, struct file *file)
4803 mutex_lock(&bpf_stats_enabled_mutex);
4804 static_key_slow_dec(&bpf_stats_enabled_key.key);
4805 mutex_unlock(&bpf_stats_enabled_mutex);
4809 static const struct file_operations bpf_stats_fops = {
4810 .release = bpf_stats_release,
4813 static int bpf_enable_runtime_stats(void)
4817 mutex_lock(&bpf_stats_enabled_mutex);
4819 /* Set a very high limit to avoid overflow */
4820 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4821 mutex_unlock(&bpf_stats_enabled_mutex);
4825 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4827 static_key_slow_inc(&bpf_stats_enabled_key.key);
4829 mutex_unlock(&bpf_stats_enabled_mutex);
4833 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4835 static int bpf_enable_stats(union bpf_attr *attr)
4838 if (CHECK_ATTR(BPF_ENABLE_STATS))
4841 if (!capable(CAP_SYS_ADMIN))
4844 switch (attr->enable_stats.type) {
4845 case BPF_STATS_RUN_TIME:
4846 return bpf_enable_runtime_stats();
4853 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4855 static int bpf_iter_create(union bpf_attr *attr)
4857 struct bpf_link *link;
4860 if (CHECK_ATTR(BPF_ITER_CREATE))
4863 if (attr->iter_create.flags)
4866 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4868 return PTR_ERR(link);
4870 err = bpf_iter_new_fd(link);
4876 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4878 static int bpf_prog_bind_map(union bpf_attr *attr)
4880 struct bpf_prog *prog;
4881 struct bpf_map *map;
4882 struct bpf_map **used_maps_old, **used_maps_new;
4885 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4888 if (attr->prog_bind_map.flags)
4891 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4893 return PTR_ERR(prog);
4895 map = bpf_map_get(attr->prog_bind_map.map_fd);
4901 mutex_lock(&prog->aux->used_maps_mutex);
4903 used_maps_old = prog->aux->used_maps;
4905 for (i = 0; i < prog->aux->used_map_cnt; i++)
4906 if (used_maps_old[i] == map) {
4911 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4912 sizeof(used_maps_new[0]),
4914 if (!used_maps_new) {
4919 memcpy(used_maps_new, used_maps_old,
4920 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4921 used_maps_new[prog->aux->used_map_cnt] = map;
4923 prog->aux->used_map_cnt++;
4924 prog->aux->used_maps = used_maps_new;
4926 kfree(used_maps_old);
4929 mutex_unlock(&prog->aux->used_maps_mutex);
4938 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
4940 union bpf_attr attr;
4944 capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
4946 /* Intent here is for unprivileged_bpf_disabled to block key object
4947 * creation commands for unprivileged users; other actions depend
4948 * of fd availability and access to bpffs, so are dependent on
4949 * object creation success. Capabilities are later verified for
4950 * operations such as load and map create, so even with unprivileged
4951 * BPF disabled, capability checks are still carried out for these
4952 * and other operations.
4955 (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
4958 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4961 size = min_t(u32, size, sizeof(attr));
4963 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4964 memset(&attr, 0, sizeof(attr));
4965 if (copy_from_bpfptr(&attr, uattr, size) != 0)
4968 err = security_bpf(cmd, &attr, size);
4973 case BPF_MAP_CREATE:
4974 err = map_create(&attr);
4976 case BPF_MAP_LOOKUP_ELEM:
4977 err = map_lookup_elem(&attr);
4979 case BPF_MAP_UPDATE_ELEM:
4980 err = map_update_elem(&attr, uattr);
4982 case BPF_MAP_DELETE_ELEM:
4983 err = map_delete_elem(&attr, uattr);
4985 case BPF_MAP_GET_NEXT_KEY:
4986 err = map_get_next_key(&attr);
4988 case BPF_MAP_FREEZE:
4989 err = map_freeze(&attr);
4992 err = bpf_prog_load(&attr, uattr);
4995 err = bpf_obj_pin(&attr);
4998 err = bpf_obj_get(&attr);
5000 case BPF_PROG_ATTACH:
5001 err = bpf_prog_attach(&attr);
5003 case BPF_PROG_DETACH:
5004 err = bpf_prog_detach(&attr);
5006 case BPF_PROG_QUERY:
5007 err = bpf_prog_query(&attr, uattr.user);
5009 case BPF_PROG_TEST_RUN:
5010 err = bpf_prog_test_run(&attr, uattr.user);
5012 case BPF_PROG_GET_NEXT_ID:
5013 err = bpf_obj_get_next_id(&attr, uattr.user,
5014 &prog_idr, &prog_idr_lock);
5016 case BPF_MAP_GET_NEXT_ID:
5017 err = bpf_obj_get_next_id(&attr, uattr.user,
5018 &map_idr, &map_idr_lock);
5020 case BPF_BTF_GET_NEXT_ID:
5021 err = bpf_obj_get_next_id(&attr, uattr.user,
5022 &btf_idr, &btf_idr_lock);
5024 case BPF_PROG_GET_FD_BY_ID:
5025 err = bpf_prog_get_fd_by_id(&attr);
5027 case BPF_MAP_GET_FD_BY_ID:
5028 err = bpf_map_get_fd_by_id(&attr);
5030 case BPF_OBJ_GET_INFO_BY_FD:
5031 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5033 case BPF_RAW_TRACEPOINT_OPEN:
5034 err = bpf_raw_tracepoint_open(&attr);
5037 err = bpf_btf_load(&attr, uattr);
5039 case BPF_BTF_GET_FD_BY_ID:
5040 err = bpf_btf_get_fd_by_id(&attr);
5042 case BPF_TASK_FD_QUERY:
5043 err = bpf_task_fd_query(&attr, uattr.user);
5045 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5046 err = map_lookup_and_delete_elem(&attr);
5048 case BPF_MAP_LOOKUP_BATCH:
5049 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5051 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5052 err = bpf_map_do_batch(&attr, uattr.user,
5053 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5055 case BPF_MAP_UPDATE_BATCH:
5056 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5058 case BPF_MAP_DELETE_BATCH:
5059 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5061 case BPF_LINK_CREATE:
5062 err = link_create(&attr, uattr);
5064 case BPF_LINK_UPDATE:
5065 err = link_update(&attr);
5067 case BPF_LINK_GET_FD_BY_ID:
5068 err = bpf_link_get_fd_by_id(&attr);
5070 case BPF_LINK_GET_NEXT_ID:
5071 err = bpf_obj_get_next_id(&attr, uattr.user,
5072 &link_idr, &link_idr_lock);
5074 case BPF_ENABLE_STATS:
5075 err = bpf_enable_stats(&attr);
5077 case BPF_ITER_CREATE:
5078 err = bpf_iter_create(&attr);
5080 case BPF_LINK_DETACH:
5081 err = link_detach(&attr);
5083 case BPF_PROG_BIND_MAP:
5084 err = bpf_prog_bind_map(&attr);
5094 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5096 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5099 static bool syscall_prog_is_valid_access(int off, int size,
5100 enum bpf_access_type type,
5101 const struct bpf_prog *prog,
5102 struct bpf_insn_access_aux *info)
5104 if (off < 0 || off >= U16_MAX)
5106 if (off % size != 0)
5111 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5114 case BPF_MAP_CREATE:
5115 case BPF_MAP_DELETE_ELEM:
5116 case BPF_MAP_UPDATE_ELEM:
5117 case BPF_MAP_FREEZE:
5118 case BPF_MAP_GET_FD_BY_ID:
5121 case BPF_LINK_CREATE:
5122 case BPF_RAW_TRACEPOINT_OPEN:
5127 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5131 /* To shut up -Wmissing-prototypes.
5132 * This function is used by the kernel light skeleton
5133 * to load bpf programs when modules are loaded or during kernel boot.
5134 * See tools/lib/bpf/skel_internal.h
5136 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5138 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5140 struct bpf_prog * __maybe_unused prog;
5141 struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5144 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5145 case BPF_PROG_TEST_RUN:
5146 if (attr->test.data_in || attr->test.data_out ||
5147 attr->test.ctx_out || attr->test.duration ||
5148 attr->test.repeat || attr->test.flags)
5151 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5153 return PTR_ERR(prog);
5155 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5156 attr->test.ctx_size_in > U16_MAX) {
5161 run_ctx.bpf_cookie = 0;
5162 run_ctx.saved_run_ctx = NULL;
5163 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5164 /* recursion detected */
5168 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5169 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5175 return ____bpf_sys_bpf(cmd, attr, size);
5178 EXPORT_SYMBOL(kern_sys_bpf);
5180 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5181 .func = bpf_sys_bpf,
5183 .ret_type = RET_INTEGER,
5184 .arg1_type = ARG_ANYTHING,
5185 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5186 .arg3_type = ARG_CONST_SIZE,
5189 const struct bpf_func_proto * __weak
5190 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5192 return bpf_base_func_proto(func_id);
5195 BPF_CALL_1(bpf_sys_close, u32, fd)
5197 /* When bpf program calls this helper there should not be
5198 * an fdget() without matching completed fdput().
5199 * This helper is allowed in the following callchain only:
5200 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5202 return close_fd(fd);
5205 static const struct bpf_func_proto bpf_sys_close_proto = {
5206 .func = bpf_sys_close,
5208 .ret_type = RET_INTEGER,
5209 .arg1_type = ARG_ANYTHING,
5212 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5217 if (name_sz <= 1 || name[name_sz - 1])
5220 if (!bpf_dump_raw_ok(current_cred()))
5223 *res = kallsyms_lookup_name(name);
5224 return *res ? 0 : -ENOENT;
5227 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5228 .func = bpf_kallsyms_lookup_name,
5230 .ret_type = RET_INTEGER,
5231 .arg1_type = ARG_PTR_TO_MEM,
5232 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
5233 .arg3_type = ARG_ANYTHING,
5234 .arg4_type = ARG_PTR_TO_LONG,
5237 static const struct bpf_func_proto *
5238 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5241 case BPF_FUNC_sys_bpf:
5242 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5243 case BPF_FUNC_btf_find_by_name_kind:
5244 return &bpf_btf_find_by_name_kind_proto;
5245 case BPF_FUNC_sys_close:
5246 return &bpf_sys_close_proto;
5247 case BPF_FUNC_kallsyms_lookup_name:
5248 return &bpf_kallsyms_lookup_name_proto;
5250 return tracing_prog_func_proto(func_id, prog);
5254 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5255 .get_func_proto = syscall_prog_func_proto,
5256 .is_valid_access = syscall_prog_is_valid_access,
5259 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5260 .test_run = bpf_prog_test_run_syscall,
5263 #ifdef CONFIG_SYSCTL
5264 static int bpf_stats_handler(struct ctl_table *table, int write,
5265 void *buffer, size_t *lenp, loff_t *ppos)
5267 struct static_key *key = (struct static_key *)table->data;
5268 static int saved_val;
5270 struct ctl_table tmp = {
5272 .maxlen = sizeof(val),
5273 .mode = table->mode,
5274 .extra1 = SYSCTL_ZERO,
5275 .extra2 = SYSCTL_ONE,
5278 if (write && !capable(CAP_SYS_ADMIN))
5281 mutex_lock(&bpf_stats_enabled_mutex);
5283 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5284 if (write && !ret && val != saved_val) {
5286 static_key_slow_inc(key);
5288 static_key_slow_dec(key);
5291 mutex_unlock(&bpf_stats_enabled_mutex);
5295 void __weak unpriv_ebpf_notify(int new_state)
5299 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5300 void *buffer, size_t *lenp, loff_t *ppos)
5302 int ret, unpriv_enable = *(int *)table->data;
5303 bool locked_state = unpriv_enable == 1;
5304 struct ctl_table tmp = *table;
5306 if (write && !capable(CAP_SYS_ADMIN))
5309 tmp.data = &unpriv_enable;
5310 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5311 if (write && !ret) {
5312 if (locked_state && unpriv_enable != 1)
5314 *(int *)table->data = unpriv_enable;
5317 unpriv_ebpf_notify(unpriv_enable);
5322 static struct ctl_table bpf_syscall_table[] = {
5324 .procname = "unprivileged_bpf_disabled",
5325 .data = &sysctl_unprivileged_bpf_disabled,
5326 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
5328 .proc_handler = bpf_unpriv_handler,
5329 .extra1 = SYSCTL_ZERO,
5330 .extra2 = SYSCTL_TWO,
5333 .procname = "bpf_stats_enabled",
5334 .data = &bpf_stats_enabled_key.key,
5336 .proc_handler = bpf_stats_handler,
5341 static int __init bpf_syscall_sysctl_init(void)
5343 register_sysctl_init("kernel", bpf_syscall_table);
5346 late_initcall(bpf_syscall_sysctl_init);
5347 #endif /* CONFIG_SYSCTL */