1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/bpf.h>
14 #include <linux/jhash.h>
15 #include <linux/filter.h>
16 #include <linux/vmalloc.h>
17 #include "percpu_freelist.h"
18 #include "bpf_lru_list.h"
21 struct hlist_head head;
27 struct bucket *buckets;
30 struct pcpu_freelist freelist;
33 void __percpu *extra_elems;
34 atomic_t count; /* number of elements in this hashtable */
35 u32 n_buckets; /* number of hash buckets */
36 u32 elem_size; /* size of each element in bytes */
39 enum extra_elem_state {
40 HTAB_NOT_AN_EXTRA_ELEM = 0,
45 /* each htab element is struct htab_elem + key + value */
48 struct hlist_node hash_node;
49 struct bpf_htab *htab;
50 struct pcpu_freelist_node fnode;
54 enum extra_elem_state state;
55 struct bpf_lru_node lru_node;
58 char key[0] __aligned(8);
61 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
63 static bool htab_is_lru(const struct bpf_htab *htab)
65 return htab->map.map_type == BPF_MAP_TYPE_LRU_HASH ||
66 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
69 static bool htab_is_percpu(const struct bpf_htab *htab)
71 return htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH ||
72 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
75 static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
78 *(void __percpu **)(l->key + key_size) = pptr;
81 static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
83 return *(void __percpu **)(l->key + key_size);
86 static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
88 return (struct htab_elem *) (htab->elems + i * htab->elem_size);
91 static void htab_free_elems(struct bpf_htab *htab)
95 if (!htab_is_percpu(htab))
98 for (i = 0; i < htab->map.max_entries; i++) {
101 pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
109 static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
112 struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
116 l = container_of(node, struct htab_elem, lru_node);
117 memcpy(l->key, key, htab->map.key_size);
124 static int prealloc_init(struct bpf_htab *htab)
126 int err = -ENOMEM, i;
128 htab->elems = vzalloc(htab->elem_size * htab->map.max_entries);
132 if (!htab_is_percpu(htab))
133 goto skip_percpu_elems;
135 for (i = 0; i < htab->map.max_entries; i++) {
136 u32 size = round_up(htab->map.value_size, 8);
139 pptr = __alloc_percpu_gfp(size, 8, GFP_USER | __GFP_NOWARN);
142 htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
147 if (htab_is_lru(htab))
148 err = bpf_lru_init(&htab->lru,
149 htab->map.map_flags & BPF_F_NO_COMMON_LRU,
150 offsetof(struct htab_elem, hash) -
151 offsetof(struct htab_elem, lru_node),
152 htab_lru_map_delete_node,
155 err = pcpu_freelist_init(&htab->freelist);
160 if (htab_is_lru(htab))
161 bpf_lru_populate(&htab->lru, htab->elems,
162 offsetof(struct htab_elem, lru_node),
163 htab->elem_size, htab->map.max_entries);
165 pcpu_freelist_populate(&htab->freelist, htab->elems,
166 htab->elem_size, htab->map.max_entries);
171 htab_free_elems(htab);
175 static void prealloc_destroy(struct bpf_htab *htab)
177 htab_free_elems(htab);
179 if (htab_is_lru(htab))
180 bpf_lru_destroy(&htab->lru);
182 pcpu_freelist_destroy(&htab->freelist);
185 static int alloc_extra_elems(struct bpf_htab *htab)
190 pptr = __alloc_percpu_gfp(htab->elem_size, 8, GFP_USER | __GFP_NOWARN);
194 for_each_possible_cpu(cpu) {
195 ((struct htab_elem *)per_cpu_ptr(pptr, cpu))->state =
196 HTAB_EXTRA_ELEM_FREE;
198 htab->extra_elems = pptr;
202 /* Called from syscall */
203 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
205 bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
206 attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
207 bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
208 attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
209 /* percpu_lru means each cpu has its own LRU list.
210 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
211 * the map's value itself is percpu. percpu_lru has
212 * nothing to do with the map's value.
214 bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
215 bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
216 struct bpf_htab *htab;
220 if (lru && !capable(CAP_SYS_ADMIN))
221 /* LRU implementation is much complicated than other
222 * maps. Hence, limit to CAP_SYS_ADMIN for now.
224 return ERR_PTR(-EPERM);
226 if (attr->map_flags & ~(BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU))
227 /* reserved bits should not be used */
228 return ERR_PTR(-EINVAL);
230 if (!lru && percpu_lru)
231 return ERR_PTR(-EINVAL);
233 if (lru && !prealloc)
234 return ERR_PTR(-ENOTSUPP);
236 htab = kzalloc(sizeof(*htab), GFP_USER);
238 return ERR_PTR(-ENOMEM);
240 /* mandatory map attributes */
241 htab->map.map_type = attr->map_type;
242 htab->map.key_size = attr->key_size;
243 htab->map.value_size = attr->value_size;
244 htab->map.max_entries = attr->max_entries;
245 htab->map.map_flags = attr->map_flags;
247 /* check sanity of attributes.
248 * value_size == 0 may be allowed in the future to use map as a set
251 if (htab->map.max_entries == 0 || htab->map.key_size == 0 ||
252 htab->map.value_size == 0)
256 /* ensure each CPU's lru list has >=1 elements.
257 * since we are at it, make each lru list has the same
258 * number of elements.
260 htab->map.max_entries = roundup(attr->max_entries,
261 num_possible_cpus());
262 if (htab->map.max_entries < attr->max_entries)
263 htab->map.max_entries = rounddown(attr->max_entries,
264 num_possible_cpus());
267 /* hash table size must be power of 2 */
268 htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
271 if (htab->map.key_size > MAX_BPF_STACK)
272 /* eBPF programs initialize keys on stack, so they cannot be
273 * larger than max stack size
277 if (htab->map.value_size >= KMALLOC_MAX_SIZE -
278 MAX_BPF_STACK - sizeof(struct htab_elem))
279 /* if value_size is bigger, the user space won't be able to
280 * access the elements via bpf syscall. This check also makes
281 * sure that the elem_size doesn't overflow and it's
282 * kmalloc-able later in htab_map_update_elem()
286 if (percpu && round_up(htab->map.value_size, 8) > PCPU_MIN_UNIT_SIZE)
287 /* make sure the size for pcpu_alloc() is reasonable */
290 htab->elem_size = sizeof(struct htab_elem) +
291 round_up(htab->map.key_size, 8);
293 htab->elem_size += sizeof(void *);
295 htab->elem_size += round_up(htab->map.value_size, 8);
297 /* prevent zero size kmalloc and check for u32 overflow */
298 if (htab->n_buckets == 0 ||
299 htab->n_buckets > U32_MAX / sizeof(struct bucket))
302 cost = (u64) htab->n_buckets * sizeof(struct bucket) +
303 (u64) htab->elem_size * htab->map.max_entries;
306 cost += (u64) round_up(htab->map.value_size, 8) *
307 num_possible_cpus() * htab->map.max_entries;
309 cost += (u64) htab->elem_size * num_possible_cpus();
311 if (cost >= U32_MAX - PAGE_SIZE)
312 /* make sure page count doesn't overflow */
315 htab->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
317 /* if map size is larger than memlock limit, reject it early */
318 err = bpf_map_precharge_memlock(htab->map.pages);
323 htab->buckets = kmalloc_array(htab->n_buckets, sizeof(struct bucket),
324 GFP_USER | __GFP_NOWARN);
326 if (!htab->buckets) {
327 htab->buckets = vmalloc(htab->n_buckets * sizeof(struct bucket));
332 for (i = 0; i < htab->n_buckets; i++) {
333 INIT_HLIST_HEAD(&htab->buckets[i].head);
334 raw_spin_lock_init(&htab->buckets[i].lock);
337 if (!percpu && !lru) {
338 /* lru itself can remove the least used element, so
339 * there is no need for an extra elem during map_update.
341 err = alloc_extra_elems(htab);
347 err = prealloc_init(htab);
349 goto free_extra_elems;
355 free_percpu(htab->extra_elems);
357 kvfree(htab->buckets);
363 static inline u32 htab_map_hash(const void *key, u32 key_len)
365 return jhash(key, key_len, 0);
368 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
370 return &htab->buckets[hash & (htab->n_buckets - 1)];
373 static inline struct hlist_head *select_bucket(struct bpf_htab *htab, u32 hash)
375 return &__select_bucket(htab, hash)->head;
378 static struct htab_elem *lookup_elem_raw(struct hlist_head *head, u32 hash,
379 void *key, u32 key_size)
383 hlist_for_each_entry_rcu(l, head, hash_node)
384 if (l->hash == hash && !memcmp(&l->key, key, key_size))
390 /* Called from syscall or from eBPF program */
391 static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
393 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
394 struct hlist_head *head;
398 /* Must be called with rcu_read_lock. */
399 WARN_ON_ONCE(!rcu_read_lock_held());
401 key_size = map->key_size;
403 hash = htab_map_hash(key, key_size);
405 head = select_bucket(htab, hash);
407 l = lookup_elem_raw(head, hash, key, key_size);
412 static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
414 struct htab_elem *l = __htab_map_lookup_elem(map, key);
417 return l->key + round_up(map->key_size, 8);
422 static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
424 struct htab_elem *l = __htab_map_lookup_elem(map, key);
427 bpf_lru_node_set_ref(&l->lru_node);
428 return l->key + round_up(map->key_size, 8);
434 /* It is called from the bpf_lru_list when the LRU needs to delete
435 * older elements from the htab.
437 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
439 struct bpf_htab *htab = (struct bpf_htab *)arg;
440 struct htab_elem *l, *tgt_l;
441 struct hlist_head *head;
445 tgt_l = container_of(node, struct htab_elem, lru_node);
446 b = __select_bucket(htab, tgt_l->hash);
449 raw_spin_lock_irqsave(&b->lock, flags);
451 hlist_for_each_entry_rcu(l, head, hash_node)
453 hlist_del_rcu(&l->hash_node);
457 raw_spin_unlock_irqrestore(&b->lock, flags);
462 /* Called from syscall */
463 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
465 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
466 struct hlist_head *head;
467 struct htab_elem *l, *next_l;
471 WARN_ON_ONCE(!rcu_read_lock_held());
473 key_size = map->key_size;
475 hash = htab_map_hash(key, key_size);
477 head = select_bucket(htab, hash);
480 l = lookup_elem_raw(head, hash, key, key_size);
484 goto find_first_elem;
487 /* key was found, get next key in the same bucket */
488 next_l = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&l->hash_node)),
489 struct htab_elem, hash_node);
492 /* if next elem in this hash list is non-zero, just return it */
493 memcpy(next_key, next_l->key, key_size);
497 /* no more elements in this hash list, go to the next bucket */
498 i = hash & (htab->n_buckets - 1);
502 /* iterate over buckets */
503 for (; i < htab->n_buckets; i++) {
504 head = select_bucket(htab, i);
506 /* pick first element in the bucket */
507 next_l = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
508 struct htab_elem, hash_node);
510 /* if it's not empty, just return it */
511 memcpy(next_key, next_l->key, key_size);
516 /* iterated over all buckets and all elements */
520 static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
522 if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
523 free_percpu(htab_elem_get_ptr(l, htab->map.key_size));
527 static void htab_elem_free_rcu(struct rcu_head *head)
529 struct htab_elem *l = container_of(head, struct htab_elem, rcu);
530 struct bpf_htab *htab = l->htab;
532 /* must increment bpf_prog_active to avoid kprobe+bpf triggering while
533 * we're calling kfree, otherwise deadlock is possible if kprobes
534 * are placed somewhere inside of slub
537 __this_cpu_inc(bpf_prog_active);
538 htab_elem_free(htab, l);
539 __this_cpu_dec(bpf_prog_active);
543 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
545 if (l->state == HTAB_EXTRA_ELEM_USED) {
546 l->state = HTAB_EXTRA_ELEM_FREE;
550 if (!(htab->map.map_flags & BPF_F_NO_PREALLOC)) {
551 pcpu_freelist_push(&htab->freelist, &l->fnode);
553 atomic_dec(&htab->count);
555 call_rcu(&l->rcu, htab_elem_free_rcu);
559 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
560 void *value, bool onallcpus)
563 /* copy true value_size bytes */
564 memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
566 u32 size = round_up(htab->map.value_size, 8);
569 for_each_possible_cpu(cpu) {
570 bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
577 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
578 void *value, u32 key_size, u32 hash,
579 bool percpu, bool onallcpus,
580 bool old_elem_exists)
582 u32 size = htab->map.value_size;
583 bool prealloc = !(htab->map.map_flags & BPF_F_NO_PREALLOC);
584 struct htab_elem *l_new;
589 l_new = (struct htab_elem *)pcpu_freelist_pop(&htab->freelist);
593 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
594 atomic_dec(&htab->count);
597 l_new = kmalloc(htab->elem_size,
598 GFP_ATOMIC | __GFP_NOWARN);
600 return ERR_PTR(-ENOMEM);
605 if (!old_elem_exists)
608 /* if we're updating the existing element and the hash table
609 * is full, use per-cpu extra elems
611 l_new = this_cpu_ptr(htab->extra_elems);
612 if (l_new->state != HTAB_EXTRA_ELEM_FREE)
613 return ERR_PTR(-E2BIG);
614 l_new->state = HTAB_EXTRA_ELEM_USED;
616 l_new->state = HTAB_NOT_AN_EXTRA_ELEM;
619 memcpy(l_new->key, key, key_size);
621 /* round up value_size to 8 bytes */
622 size = round_up(size, 8);
625 pptr = htab_elem_get_ptr(l_new, key_size);
627 /* alloc_percpu zero-fills */
628 pptr = __alloc_percpu_gfp(size, 8,
629 GFP_ATOMIC | __GFP_NOWARN);
632 return ERR_PTR(-ENOMEM);
636 pcpu_copy_value(htab, pptr, value, onallcpus);
639 htab_elem_set_ptr(l_new, key_size, pptr);
641 memcpy(l_new->key + round_up(key_size, 8), value, size);
648 static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
651 if (l_old && map_flags == BPF_NOEXIST)
652 /* elem already exists */
655 if (!l_old && map_flags == BPF_EXIST)
656 /* elem doesn't exist, cannot update it */
662 /* Called from syscall or from eBPF program */
663 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
666 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
667 struct htab_elem *l_new = NULL, *l_old;
668 struct hlist_head *head;
674 if (unlikely(map_flags > BPF_EXIST))
678 WARN_ON_ONCE(!rcu_read_lock_held());
680 key_size = map->key_size;
682 hash = htab_map_hash(key, key_size);
684 b = __select_bucket(htab, hash);
687 /* bpf_map_update_elem() can be called in_irq() */
688 raw_spin_lock_irqsave(&b->lock, flags);
690 l_old = lookup_elem_raw(head, hash, key, key_size);
692 ret = check_flags(htab, l_old, map_flags);
696 l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
699 /* all pre-allocated elements are in use or memory exhausted */
700 ret = PTR_ERR(l_new);
704 /* add new element to the head of the list, so that
705 * concurrent search will find it before old elem
707 hlist_add_head_rcu(&l_new->hash_node, head);
709 hlist_del_rcu(&l_old->hash_node);
710 free_htab_elem(htab, l_old);
714 raw_spin_unlock_irqrestore(&b->lock, flags);
718 static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
721 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
722 struct htab_elem *l_new, *l_old = NULL;
723 struct hlist_head *head;
729 if (unlikely(map_flags > BPF_EXIST))
733 WARN_ON_ONCE(!rcu_read_lock_held());
735 key_size = map->key_size;
737 hash = htab_map_hash(key, key_size);
739 b = __select_bucket(htab, hash);
742 /* For LRU, we need to alloc before taking bucket's
743 * spinlock because getting free nodes from LRU may need
744 * to remove older elements from htab and this removal
745 * operation will need a bucket lock.
747 l_new = prealloc_lru_pop(htab, key, hash);
750 memcpy(l_new->key + round_up(map->key_size, 8), value, map->value_size);
752 /* bpf_map_update_elem() can be called in_irq() */
753 raw_spin_lock_irqsave(&b->lock, flags);
755 l_old = lookup_elem_raw(head, hash, key, key_size);
757 ret = check_flags(htab, l_old, map_flags);
761 /* add new element to the head of the list, so that
762 * concurrent search will find it before old elem
764 hlist_add_head_rcu(&l_new->hash_node, head);
766 bpf_lru_node_set_ref(&l_new->lru_node);
767 hlist_del_rcu(&l_old->hash_node);
772 raw_spin_unlock_irqrestore(&b->lock, flags);
775 bpf_lru_push_free(&htab->lru, &l_new->lru_node);
777 bpf_lru_push_free(&htab->lru, &l_old->lru_node);
782 static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
783 void *value, u64 map_flags,
786 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
787 struct htab_elem *l_new = NULL, *l_old;
788 struct hlist_head *head;
794 if (unlikely(map_flags > BPF_EXIST))
798 WARN_ON_ONCE(!rcu_read_lock_held());
800 key_size = map->key_size;
802 hash = htab_map_hash(key, key_size);
804 b = __select_bucket(htab, hash);
807 /* bpf_map_update_elem() can be called in_irq() */
808 raw_spin_lock_irqsave(&b->lock, flags);
810 l_old = lookup_elem_raw(head, hash, key, key_size);
812 ret = check_flags(htab, l_old, map_flags);
817 /* per-cpu hash map can update value in-place */
818 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
821 l_new = alloc_htab_elem(htab, key, value, key_size,
822 hash, true, onallcpus, false);
824 ret = PTR_ERR(l_new);
827 hlist_add_head_rcu(&l_new->hash_node, head);
831 raw_spin_unlock_irqrestore(&b->lock, flags);
835 static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
836 void *value, u64 map_flags,
839 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
840 struct htab_elem *l_new = NULL, *l_old;
841 struct hlist_head *head;
847 if (unlikely(map_flags > BPF_EXIST))
851 WARN_ON_ONCE(!rcu_read_lock_held());
853 key_size = map->key_size;
855 hash = htab_map_hash(key, key_size);
857 b = __select_bucket(htab, hash);
860 /* For LRU, we need to alloc before taking bucket's
861 * spinlock because LRU's elem alloc may need
862 * to remove older elem from htab and this removal
863 * operation will need a bucket lock.
865 if (map_flags != BPF_EXIST) {
866 l_new = prealloc_lru_pop(htab, key, hash);
871 /* bpf_map_update_elem() can be called in_irq() */
872 raw_spin_lock_irqsave(&b->lock, flags);
874 l_old = lookup_elem_raw(head, hash, key, key_size);
876 ret = check_flags(htab, l_old, map_flags);
881 bpf_lru_node_set_ref(&l_old->lru_node);
883 /* per-cpu hash map can update value in-place */
884 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
887 pcpu_copy_value(htab, htab_elem_get_ptr(l_new, key_size),
889 hlist_add_head_rcu(&l_new->hash_node, head);
894 raw_spin_unlock_irqrestore(&b->lock, flags);
896 bpf_lru_push_free(&htab->lru, &l_new->lru_node);
900 static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
901 void *value, u64 map_flags)
903 return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
906 static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
907 void *value, u64 map_flags)
909 return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
913 /* Called from syscall or from eBPF program */
914 static int htab_map_delete_elem(struct bpf_map *map, void *key)
916 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
917 struct hlist_head *head;
924 WARN_ON_ONCE(!rcu_read_lock_held());
926 key_size = map->key_size;
928 hash = htab_map_hash(key, key_size);
929 b = __select_bucket(htab, hash);
932 raw_spin_lock_irqsave(&b->lock, flags);
934 l = lookup_elem_raw(head, hash, key, key_size);
937 hlist_del_rcu(&l->hash_node);
938 free_htab_elem(htab, l);
942 raw_spin_unlock_irqrestore(&b->lock, flags);
946 static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
948 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
949 struct hlist_head *head;
956 WARN_ON_ONCE(!rcu_read_lock_held());
958 key_size = map->key_size;
960 hash = htab_map_hash(key, key_size);
961 b = __select_bucket(htab, hash);
964 raw_spin_lock_irqsave(&b->lock, flags);
966 l = lookup_elem_raw(head, hash, key, key_size);
969 hlist_del_rcu(&l->hash_node);
973 raw_spin_unlock_irqrestore(&b->lock, flags);
975 bpf_lru_push_free(&htab->lru, &l->lru_node);
979 static void delete_all_elements(struct bpf_htab *htab)
983 for (i = 0; i < htab->n_buckets; i++) {
984 struct hlist_head *head = select_bucket(htab, i);
985 struct hlist_node *n;
988 hlist_for_each_entry_safe(l, n, head, hash_node) {
989 hlist_del_rcu(&l->hash_node);
990 if (l->state != HTAB_EXTRA_ELEM_USED)
991 htab_elem_free(htab, l);
995 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
996 static void htab_map_free(struct bpf_map *map)
998 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1000 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
1001 * so the programs (can be more than one that used this map) were
1002 * disconnected from events. Wait for outstanding critical sections in
1003 * these programs to complete
1007 /* some of free_htab_elem() callbacks for elements of this map may
1008 * not have executed. Wait for them.
1011 if (htab->map.map_flags & BPF_F_NO_PREALLOC)
1012 delete_all_elements(htab);
1014 prealloc_destroy(htab);
1016 free_percpu(htab->extra_elems);
1017 kvfree(htab->buckets);
1021 static const struct bpf_map_ops htab_ops = {
1022 .map_alloc = htab_map_alloc,
1023 .map_free = htab_map_free,
1024 .map_get_next_key = htab_map_get_next_key,
1025 .map_lookup_elem = htab_map_lookup_elem,
1026 .map_update_elem = htab_map_update_elem,
1027 .map_delete_elem = htab_map_delete_elem,
1030 static struct bpf_map_type_list htab_type __read_mostly = {
1032 .type = BPF_MAP_TYPE_HASH,
1035 static const struct bpf_map_ops htab_lru_ops = {
1036 .map_alloc = htab_map_alloc,
1037 .map_free = htab_map_free,
1038 .map_get_next_key = htab_map_get_next_key,
1039 .map_lookup_elem = htab_lru_map_lookup_elem,
1040 .map_update_elem = htab_lru_map_update_elem,
1041 .map_delete_elem = htab_lru_map_delete_elem,
1044 static struct bpf_map_type_list htab_lru_type __read_mostly = {
1045 .ops = &htab_lru_ops,
1046 .type = BPF_MAP_TYPE_LRU_HASH,
1049 /* Called from eBPF program */
1050 static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
1052 struct htab_elem *l = __htab_map_lookup_elem(map, key);
1055 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
1060 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
1062 struct htab_elem *l = __htab_map_lookup_elem(map, key);
1065 bpf_lru_node_set_ref(&l->lru_node);
1066 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
1072 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
1074 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1075 struct htab_elem *l;
1076 void __percpu *pptr;
1081 /* per_cpu areas are zero-filled and bpf programs can only
1082 * access 'value_size' of them, so copying rounded areas
1083 * will not leak any kernel data
1085 size = round_up(map->value_size, 8);
1087 l = __htab_map_lookup_elem(map, key);
1090 if (htab_is_lru(htab))
1091 bpf_lru_node_set_ref(&l->lru_node);
1092 pptr = htab_elem_get_ptr(l, map->key_size);
1093 for_each_possible_cpu(cpu) {
1094 bpf_long_memcpy(value + off,
1095 per_cpu_ptr(pptr, cpu), size);
1104 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
1107 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1111 if (htab_is_lru(htab))
1112 ret = __htab_lru_percpu_map_update_elem(map, key, value,
1115 ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
1122 static const struct bpf_map_ops htab_percpu_ops = {
1123 .map_alloc = htab_map_alloc,
1124 .map_free = htab_map_free,
1125 .map_get_next_key = htab_map_get_next_key,
1126 .map_lookup_elem = htab_percpu_map_lookup_elem,
1127 .map_update_elem = htab_percpu_map_update_elem,
1128 .map_delete_elem = htab_map_delete_elem,
1131 static struct bpf_map_type_list htab_percpu_type __read_mostly = {
1132 .ops = &htab_percpu_ops,
1133 .type = BPF_MAP_TYPE_PERCPU_HASH,
1136 static const struct bpf_map_ops htab_lru_percpu_ops = {
1137 .map_alloc = htab_map_alloc,
1138 .map_free = htab_map_free,
1139 .map_get_next_key = htab_map_get_next_key,
1140 .map_lookup_elem = htab_lru_percpu_map_lookup_elem,
1141 .map_update_elem = htab_lru_percpu_map_update_elem,
1142 .map_delete_elem = htab_lru_map_delete_elem,
1145 static struct bpf_map_type_list htab_lru_percpu_type __read_mostly = {
1146 .ops = &htab_lru_percpu_ops,
1147 .type = BPF_MAP_TYPE_LRU_PERCPU_HASH,
1150 static int __init register_htab_map(void)
1152 bpf_register_map_type(&htab_type);
1153 bpf_register_map_type(&htab_percpu_type);
1154 bpf_register_map_type(&htab_lru_type);
1155 bpf_register_map_type(&htab_lru_percpu_type);
1158 late_initcall(register_htab_map);