1 //SPDX-License-Identifier: GPL-2.0
2 #include <linux/bpf-cgroup.h>
6 #include <linux/filter.h>
8 #include <linux/rbtree.h>
9 #include <linux/slab.h>
10 #include <uapi/linux/btf.h>
12 DEFINE_PER_CPU(struct bpf_cgroup_storage*, bpf_cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]);
14 #ifdef CONFIG_CGROUP_BPF
16 #define LOCAL_STORAGE_CREATE_FLAG_MASK \
17 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
19 struct bpf_cgroup_storage_map {
23 struct bpf_prog *prog;
25 struct list_head list;
28 static struct bpf_cgroup_storage_map *map_to_storage(struct bpf_map *map)
30 return container_of(map, struct bpf_cgroup_storage_map, map);
33 static int bpf_cgroup_storage_key_cmp(
34 const struct bpf_cgroup_storage_key *key1,
35 const struct bpf_cgroup_storage_key *key2)
37 if (key1->cgroup_inode_id < key2->cgroup_inode_id)
39 else if (key1->cgroup_inode_id > key2->cgroup_inode_id)
41 else if (key1->attach_type < key2->attach_type)
43 else if (key1->attach_type > key2->attach_type)
48 static struct bpf_cgroup_storage *cgroup_storage_lookup(
49 struct bpf_cgroup_storage_map *map, struct bpf_cgroup_storage_key *key,
52 struct rb_root *root = &map->root;
56 spin_lock_bh(&map->lock);
60 struct bpf_cgroup_storage *storage;
62 storage = container_of(node, struct bpf_cgroup_storage, node);
64 switch (bpf_cgroup_storage_key_cmp(key, &storage->key)) {
69 node = node->rb_right;
73 spin_unlock_bh(&map->lock);
79 spin_unlock_bh(&map->lock);
84 static int cgroup_storage_insert(struct bpf_cgroup_storage_map *map,
85 struct bpf_cgroup_storage *storage)
87 struct rb_root *root = &map->root;
88 struct rb_node **new = &(root->rb_node), *parent = NULL;
91 struct bpf_cgroup_storage *this;
93 this = container_of(*new, struct bpf_cgroup_storage, node);
96 switch (bpf_cgroup_storage_key_cmp(&storage->key, &this->key)) {
98 new = &((*new)->rb_left);
101 new = &((*new)->rb_right);
108 rb_link_node(&storage->node, parent, new);
109 rb_insert_color(&storage->node, root);
114 static void *cgroup_storage_lookup_elem(struct bpf_map *_map, void *_key)
116 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
117 struct bpf_cgroup_storage_key *key = _key;
118 struct bpf_cgroup_storage *storage;
120 storage = cgroup_storage_lookup(map, key, false);
124 return &READ_ONCE(storage->buf)->data[0];
127 static int cgroup_storage_update_elem(struct bpf_map *map, void *_key,
128 void *value, u64 flags)
130 struct bpf_cgroup_storage_key *key = _key;
131 struct bpf_cgroup_storage *storage;
132 struct bpf_storage_buffer *new;
134 if (unlikely(flags & ~(BPF_F_LOCK | BPF_EXIST | BPF_NOEXIST)))
137 if (unlikely(flags & BPF_NOEXIST))
140 if (unlikely((flags & BPF_F_LOCK) &&
141 !map_value_has_spin_lock(map)))
144 storage = cgroup_storage_lookup((struct bpf_cgroup_storage_map *)map,
149 if (flags & BPF_F_LOCK) {
150 copy_map_value_locked(map, storage->buf->data, value, false);
154 new = kmalloc_node(sizeof(struct bpf_storage_buffer) +
156 __GFP_ZERO | GFP_ATOMIC | __GFP_NOWARN,
161 memcpy(&new->data[0], value, map->value_size);
162 check_and_init_map_lock(map, new->data);
164 new = xchg(&storage->buf, new);
170 int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *_key,
173 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
174 struct bpf_cgroup_storage_key *key = _key;
175 struct bpf_cgroup_storage *storage;
180 storage = cgroup_storage_lookup(map, key, false);
186 /* per_cpu areas are zero-filled and bpf programs can only
187 * access 'value_size' of them, so copying rounded areas
188 * will not leak any kernel data
190 size = round_up(_map->value_size, 8);
191 for_each_possible_cpu(cpu) {
192 bpf_long_memcpy(value + off,
193 per_cpu_ptr(storage->percpu_buf, cpu), size);
200 int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *_key,
201 void *value, u64 map_flags)
203 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
204 struct bpf_cgroup_storage_key *key = _key;
205 struct bpf_cgroup_storage *storage;
209 if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
213 storage = cgroup_storage_lookup(map, key, false);
219 /* the user space will provide round_up(value_size, 8) bytes that
220 * will be copied into per-cpu area. bpf programs can only access
221 * value_size of it. During lookup the same extra bytes will be
222 * returned or zeros which were zero-filled by percpu_alloc,
223 * so no kernel data leaks possible
225 size = round_up(_map->value_size, 8);
226 for_each_possible_cpu(cpu) {
227 bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
235 static int cgroup_storage_get_next_key(struct bpf_map *_map, void *_key,
238 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
239 struct bpf_cgroup_storage_key *key = _key;
240 struct bpf_cgroup_storage_key *next = _next_key;
241 struct bpf_cgroup_storage *storage;
243 spin_lock_bh(&map->lock);
245 if (list_empty(&map->list))
249 storage = cgroup_storage_lookup(map, key, true);
253 storage = list_next_entry(storage, list);
257 storage = list_first_entry(&map->list,
258 struct bpf_cgroup_storage, list);
261 spin_unlock_bh(&map->lock);
262 next->attach_type = storage->key.attach_type;
263 next->cgroup_inode_id = storage->key.cgroup_inode_id;
267 spin_unlock_bh(&map->lock);
271 static struct bpf_map *cgroup_storage_map_alloc(union bpf_attr *attr)
273 int numa_node = bpf_map_attr_numa_node(attr);
274 struct bpf_cgroup_storage_map *map;
275 struct bpf_map_memory mem;
278 if (attr->key_size != sizeof(struct bpf_cgroup_storage_key))
279 return ERR_PTR(-EINVAL);
281 if (attr->value_size == 0)
282 return ERR_PTR(-EINVAL);
284 if (attr->value_size > PAGE_SIZE)
285 return ERR_PTR(-E2BIG);
287 if (attr->map_flags & ~LOCAL_STORAGE_CREATE_FLAG_MASK ||
288 !bpf_map_flags_access_ok(attr->map_flags))
289 return ERR_PTR(-EINVAL);
291 if (attr->max_entries)
292 /* max_entries is not used and enforced to be 0 */
293 return ERR_PTR(-EINVAL);
295 ret = bpf_map_charge_init(&mem, sizeof(struct bpf_cgroup_storage_map));
299 map = kmalloc_node(sizeof(struct bpf_cgroup_storage_map),
300 __GFP_ZERO | GFP_USER, numa_node);
302 bpf_map_charge_finish(&mem);
303 return ERR_PTR(-ENOMEM);
306 bpf_map_charge_move(&map->map.memory, &mem);
308 /* copy mandatory map attributes */
309 bpf_map_init_from_attr(&map->map, attr);
311 spin_lock_init(&map->lock);
313 INIT_LIST_HEAD(&map->list);
318 static void cgroup_storage_map_free(struct bpf_map *_map)
320 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
322 WARN_ON(!RB_EMPTY_ROOT(&map->root));
323 WARN_ON(!list_empty(&map->list));
328 static int cgroup_storage_delete_elem(struct bpf_map *map, void *key)
333 static int cgroup_storage_check_btf(const struct bpf_map *map,
334 const struct btf *btf,
335 const struct btf_type *key_type,
336 const struct btf_type *value_type)
338 struct btf_member *m;
341 /* Key is expected to be of struct bpf_cgroup_storage_key type,
343 * struct bpf_cgroup_storage_key {
344 * __u64 cgroup_inode_id;
350 * Key_type must be a structure with two fields.
352 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_STRUCT ||
353 BTF_INFO_VLEN(key_type->info) != 2)
357 * The first field must be a 64 bit integer at 0 offset.
359 m = (struct btf_member *)(key_type + 1);
360 size = FIELD_SIZEOF(struct bpf_cgroup_storage_key, cgroup_inode_id);
361 if (!btf_member_is_reg_int(btf, key_type, m, 0, size))
365 * The second field must be a 32 bit integer at 64 bit offset.
368 offset = offsetof(struct bpf_cgroup_storage_key, attach_type);
369 size = FIELD_SIZEOF(struct bpf_cgroup_storage_key, attach_type);
370 if (!btf_member_is_reg_int(btf, key_type, m, offset, size))
376 static void cgroup_storage_seq_show_elem(struct bpf_map *map, void *_key,
379 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
380 struct bpf_cgroup_storage_key *key = _key;
381 struct bpf_cgroup_storage *storage;
385 storage = cgroup_storage_lookup(map_to_storage(map), key, false);
391 btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
392 stype = cgroup_storage_type(map);
393 if (stype == BPF_CGROUP_STORAGE_SHARED) {
395 btf_type_seq_show(map->btf, map->btf_value_type_id,
396 &READ_ONCE(storage->buf)->data[0], m);
399 seq_puts(m, ": {\n");
400 for_each_possible_cpu(cpu) {
401 seq_printf(m, "\tcpu%d: ", cpu);
402 btf_type_seq_show(map->btf, map->btf_value_type_id,
403 per_cpu_ptr(storage->percpu_buf, cpu),
412 const struct bpf_map_ops cgroup_storage_map_ops = {
413 .map_alloc = cgroup_storage_map_alloc,
414 .map_free = cgroup_storage_map_free,
415 .map_get_next_key = cgroup_storage_get_next_key,
416 .map_lookup_elem = cgroup_storage_lookup_elem,
417 .map_update_elem = cgroup_storage_update_elem,
418 .map_delete_elem = cgroup_storage_delete_elem,
419 .map_check_btf = cgroup_storage_check_btf,
420 .map_seq_show_elem = cgroup_storage_seq_show_elem,
423 int bpf_cgroup_storage_assign(struct bpf_prog *prog, struct bpf_map *_map)
425 enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
426 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
429 spin_lock_bh(&map->lock);
431 if (map->prog && map->prog != prog)
433 if (prog->aux->cgroup_storage[stype] &&
434 prog->aux->cgroup_storage[stype] != _map)
438 prog->aux->cgroup_storage[stype] = _map;
441 spin_unlock_bh(&map->lock);
446 void bpf_cgroup_storage_release(struct bpf_prog *prog, struct bpf_map *_map)
448 enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
449 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
451 spin_lock_bh(&map->lock);
452 if (map->prog == prog) {
453 WARN_ON(prog->aux->cgroup_storage[stype] != _map);
455 prog->aux->cgroup_storage[stype] = NULL;
457 spin_unlock_bh(&map->lock);
460 static size_t bpf_cgroup_storage_calculate_size(struct bpf_map *map, u32 *pages)
464 if (cgroup_storage_type(map) == BPF_CGROUP_STORAGE_SHARED) {
465 size = sizeof(struct bpf_storage_buffer) + map->value_size;
466 *pages = round_up(sizeof(struct bpf_cgroup_storage) + size,
467 PAGE_SIZE) >> PAGE_SHIFT;
469 size = map->value_size;
470 *pages = round_up(round_up(size, 8) * num_possible_cpus(),
471 PAGE_SIZE) >> PAGE_SHIFT;
477 struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
478 enum bpf_cgroup_storage_type stype)
480 struct bpf_cgroup_storage *storage;
486 map = prog->aux->cgroup_storage[stype];
490 size = bpf_cgroup_storage_calculate_size(map, &pages);
492 if (bpf_map_charge_memlock(map, pages))
493 return ERR_PTR(-EPERM);
495 storage = kmalloc_node(sizeof(struct bpf_cgroup_storage),
496 __GFP_ZERO | GFP_USER, map->numa_node);
500 flags = __GFP_ZERO | GFP_USER;
502 if (stype == BPF_CGROUP_STORAGE_SHARED) {
503 storage->buf = kmalloc_node(size, flags, map->numa_node);
506 check_and_init_map_lock(map, storage->buf->data);
508 storage->percpu_buf = __alloc_percpu_gfp(size, 8, flags);
509 if (!storage->percpu_buf)
513 storage->map = (struct bpf_cgroup_storage_map *)map;
518 bpf_map_uncharge_memlock(map, pages);
520 return ERR_PTR(-ENOMEM);
523 static void free_shared_cgroup_storage_rcu(struct rcu_head *rcu)
525 struct bpf_cgroup_storage *storage =
526 container_of(rcu, struct bpf_cgroup_storage, rcu);
532 static void free_percpu_cgroup_storage_rcu(struct rcu_head *rcu)
534 struct bpf_cgroup_storage *storage =
535 container_of(rcu, struct bpf_cgroup_storage, rcu);
537 free_percpu(storage->percpu_buf);
541 void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage)
543 enum bpf_cgroup_storage_type stype;
550 map = &storage->map->map;
552 bpf_cgroup_storage_calculate_size(map, &pages);
553 bpf_map_uncharge_memlock(map, pages);
555 stype = cgroup_storage_type(map);
556 if (stype == BPF_CGROUP_STORAGE_SHARED)
557 call_rcu(&storage->rcu, free_shared_cgroup_storage_rcu);
559 call_rcu(&storage->rcu, free_percpu_cgroup_storage_rcu);
562 void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
563 struct cgroup *cgroup,
564 enum bpf_attach_type type)
566 struct bpf_cgroup_storage_map *map;
571 storage->key.attach_type = type;
572 storage->key.cgroup_inode_id = cgroup->kn->id.id;
576 spin_lock_bh(&map->lock);
577 WARN_ON(cgroup_storage_insert(map, storage));
578 list_add(&storage->list, &map->list);
579 spin_unlock_bh(&map->lock);
582 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage)
584 struct bpf_cgroup_storage_map *map;
585 struct rb_root *root;
592 spin_lock_bh(&map->lock);
594 rb_erase(&storage->node, root);
596 list_del(&storage->list);
597 spin_unlock_bh(&map->lock);