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/btf.h>
15 #include <linux/jhash.h>
16 #include <linux/filter.h>
17 #include <linux/rculist_nulls.h>
18 #include <linux/random.h>
19 #include <uapi/linux/btf.h>
20 #include "percpu_freelist.h"
21 #include "bpf_lru_list.h"
22 #include "map_in_map.h"
24 #define HTAB_CREATE_FLAG_MASK \
25 (BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE | \
26 BPF_F_RDONLY | BPF_F_WRONLY | BPF_F_ZERO_SEED)
29 struct hlist_nulls_head head
;
35 struct bucket
*buckets
;
38 struct pcpu_freelist freelist
;
41 struct htab_elem
*__percpu
*extra_elems
;
42 atomic_t count
; /* number of elements in this hashtable */
43 u32 n_buckets
; /* number of hash buckets */
44 u32 elem_size
; /* size of each element in bytes */
48 /* each htab element is struct htab_elem + key + value */
51 struct hlist_nulls_node hash_node
;
55 struct bpf_htab
*htab
;
56 struct pcpu_freelist_node fnode
;
62 struct bpf_lru_node lru_node
;
65 char key
[0] __aligned(8);
68 static bool htab_lru_map_delete_node(void *arg
, struct bpf_lru_node
*node
);
70 static bool htab_is_lru(const struct bpf_htab
*htab
)
72 return htab
->map
.map_type
== BPF_MAP_TYPE_LRU_HASH
||
73 htab
->map
.map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
;
76 static bool htab_is_percpu(const struct bpf_htab
*htab
)
78 return htab
->map
.map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
79 htab
->map
.map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
;
82 static bool htab_is_prealloc(const struct bpf_htab
*htab
)
84 return !(htab
->map
.map_flags
& BPF_F_NO_PREALLOC
);
87 static inline void htab_elem_set_ptr(struct htab_elem
*l
, u32 key_size
,
90 *(void __percpu
**)(l
->key
+ key_size
) = pptr
;
93 static inline void __percpu
*htab_elem_get_ptr(struct htab_elem
*l
, u32 key_size
)
95 return *(void __percpu
**)(l
->key
+ key_size
);
98 static void *fd_htab_map_get_ptr(const struct bpf_map
*map
, struct htab_elem
*l
)
100 return *(void **)(l
->key
+ roundup(map
->key_size
, 8));
103 static struct htab_elem
*get_htab_elem(struct bpf_htab
*htab
, int i
)
105 return (struct htab_elem
*) (htab
->elems
+ i
* htab
->elem_size
);
108 static void htab_free_elems(struct bpf_htab
*htab
)
112 if (!htab_is_percpu(htab
))
115 for (i
= 0; i
< htab
->map
.max_entries
; i
++) {
118 pptr
= htab_elem_get_ptr(get_htab_elem(htab
, i
),
124 bpf_map_area_free(htab
->elems
);
127 static struct htab_elem
*prealloc_lru_pop(struct bpf_htab
*htab
, void *key
,
130 struct bpf_lru_node
*node
= bpf_lru_pop_free(&htab
->lru
, hash
);
134 l
= container_of(node
, struct htab_elem
, lru_node
);
135 memcpy(l
->key
, key
, htab
->map
.key_size
);
142 static int prealloc_init(struct bpf_htab
*htab
)
144 u32 num_entries
= htab
->map
.max_entries
;
145 int err
= -ENOMEM
, i
;
147 if (!htab_is_percpu(htab
) && !htab_is_lru(htab
))
148 num_entries
+= num_possible_cpus();
150 htab
->elems
= bpf_map_area_alloc(htab
->elem_size
* num_entries
,
151 htab
->map
.numa_node
);
155 if (!htab_is_percpu(htab
))
156 goto skip_percpu_elems
;
158 for (i
= 0; i
< num_entries
; i
++) {
159 u32 size
= round_up(htab
->map
.value_size
, 8);
162 pptr
= __alloc_percpu_gfp(size
, 8, GFP_USER
| __GFP_NOWARN
);
165 htab_elem_set_ptr(get_htab_elem(htab
, i
), htab
->map
.key_size
,
171 if (htab_is_lru(htab
))
172 err
= bpf_lru_init(&htab
->lru
,
173 htab
->map
.map_flags
& BPF_F_NO_COMMON_LRU
,
174 offsetof(struct htab_elem
, hash
) -
175 offsetof(struct htab_elem
, lru_node
),
176 htab_lru_map_delete_node
,
179 err
= pcpu_freelist_init(&htab
->freelist
);
184 if (htab_is_lru(htab
))
185 bpf_lru_populate(&htab
->lru
, htab
->elems
,
186 offsetof(struct htab_elem
, lru_node
),
187 htab
->elem_size
, num_entries
);
189 pcpu_freelist_populate(&htab
->freelist
,
190 htab
->elems
+ offsetof(struct htab_elem
, fnode
),
191 htab
->elem_size
, num_entries
);
196 htab_free_elems(htab
);
200 static void prealloc_destroy(struct bpf_htab
*htab
)
202 htab_free_elems(htab
);
204 if (htab_is_lru(htab
))
205 bpf_lru_destroy(&htab
->lru
);
207 pcpu_freelist_destroy(&htab
->freelist
);
210 static int alloc_extra_elems(struct bpf_htab
*htab
)
212 struct htab_elem
*__percpu
*pptr
, *l_new
;
213 struct pcpu_freelist_node
*l
;
216 pptr
= __alloc_percpu_gfp(sizeof(struct htab_elem
*), 8,
217 GFP_USER
| __GFP_NOWARN
);
221 for_each_possible_cpu(cpu
) {
222 l
= pcpu_freelist_pop(&htab
->freelist
);
223 /* pop will succeed, since prealloc_init()
224 * preallocated extra num_possible_cpus elements
226 l_new
= container_of(l
, struct htab_elem
, fnode
);
227 *per_cpu_ptr(pptr
, cpu
) = l_new
;
229 htab
->extra_elems
= pptr
;
233 /* Called from syscall */
234 static int htab_map_alloc_check(union bpf_attr
*attr
)
236 bool percpu
= (attr
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
237 attr
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
);
238 bool lru
= (attr
->map_type
== BPF_MAP_TYPE_LRU_HASH
||
239 attr
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
);
240 /* percpu_lru means each cpu has its own LRU list.
241 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
242 * the map's value itself is percpu. percpu_lru has
243 * nothing to do with the map's value.
245 bool percpu_lru
= (attr
->map_flags
& BPF_F_NO_COMMON_LRU
);
246 bool prealloc
= !(attr
->map_flags
& BPF_F_NO_PREALLOC
);
247 bool zero_seed
= (attr
->map_flags
& BPF_F_ZERO_SEED
);
248 int numa_node
= bpf_map_attr_numa_node(attr
);
250 BUILD_BUG_ON(offsetof(struct htab_elem
, htab
) !=
251 offsetof(struct htab_elem
, hash_node
.pprev
));
252 BUILD_BUG_ON(offsetof(struct htab_elem
, fnode
.next
) !=
253 offsetof(struct htab_elem
, hash_node
.pprev
));
255 if (lru
&& !capable(CAP_SYS_ADMIN
))
256 /* LRU implementation is much complicated than other
257 * maps. Hence, limit to CAP_SYS_ADMIN for now.
261 if (zero_seed
&& !capable(CAP_SYS_ADMIN
))
262 /* Guard against local DoS, and discourage production use. */
265 if (attr
->map_flags
& ~HTAB_CREATE_FLAG_MASK
)
266 /* reserved bits should not be used */
269 if (!lru
&& percpu_lru
)
272 if (lru
&& !prealloc
)
275 if (numa_node
!= NUMA_NO_NODE
&& (percpu
|| percpu_lru
))
278 /* check sanity of attributes.
279 * value_size == 0 may be allowed in the future to use map as a set
281 if (attr
->max_entries
== 0 || attr
->key_size
== 0 ||
282 attr
->value_size
== 0)
285 if (attr
->key_size
> MAX_BPF_STACK
)
286 /* eBPF programs initialize keys on stack, so they cannot be
287 * larger than max stack size
291 if (attr
->value_size
>= KMALLOC_MAX_SIZE
-
292 MAX_BPF_STACK
- sizeof(struct htab_elem
))
293 /* if value_size is bigger, the user space won't be able to
294 * access the elements via bpf syscall. This check also makes
295 * sure that the elem_size doesn't overflow and it's
296 * kmalloc-able later in htab_map_update_elem()
303 static struct bpf_map
*htab_map_alloc(union bpf_attr
*attr
)
305 bool percpu
= (attr
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
306 attr
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
);
307 bool lru
= (attr
->map_type
== BPF_MAP_TYPE_LRU_HASH
||
308 attr
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
);
309 /* percpu_lru means each cpu has its own LRU list.
310 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
311 * the map's value itself is percpu. percpu_lru has
312 * nothing to do with the map's value.
314 bool percpu_lru
= (attr
->map_flags
& BPF_F_NO_COMMON_LRU
);
315 bool prealloc
= !(attr
->map_flags
& BPF_F_NO_PREALLOC
);
316 struct bpf_htab
*htab
;
320 htab
= kzalloc(sizeof(*htab
), GFP_USER
);
322 return ERR_PTR(-ENOMEM
);
324 bpf_map_init_from_attr(&htab
->map
, attr
);
327 /* ensure each CPU's lru list has >=1 elements.
328 * since we are at it, make each lru list has the same
329 * number of elements.
331 htab
->map
.max_entries
= roundup(attr
->max_entries
,
332 num_possible_cpus());
333 if (htab
->map
.max_entries
< attr
->max_entries
)
334 htab
->map
.max_entries
= rounddown(attr
->max_entries
,
335 num_possible_cpus());
338 /* hash table size must be power of 2 */
339 htab
->n_buckets
= roundup_pow_of_two(htab
->map
.max_entries
);
341 htab
->elem_size
= sizeof(struct htab_elem
) +
342 round_up(htab
->map
.key_size
, 8);
344 htab
->elem_size
+= sizeof(void *);
346 htab
->elem_size
+= round_up(htab
->map
.value_size
, 8);
349 /* prevent zero size kmalloc and check for u32 overflow */
350 if (htab
->n_buckets
== 0 ||
351 htab
->n_buckets
> U32_MAX
/ sizeof(struct bucket
))
354 cost
= (u64
) htab
->n_buckets
* sizeof(struct bucket
) +
355 (u64
) htab
->elem_size
* htab
->map
.max_entries
;
358 cost
+= (u64
) round_up(htab
->map
.value_size
, 8) *
359 num_possible_cpus() * htab
->map
.max_entries
;
361 cost
+= (u64
) htab
->elem_size
* num_possible_cpus();
363 if (cost
>= U32_MAX
- PAGE_SIZE
)
364 /* make sure page count doesn't overflow */
367 htab
->map
.pages
= round_up(cost
, PAGE_SIZE
) >> PAGE_SHIFT
;
369 /* if map size is larger than memlock limit, reject it early */
370 err
= bpf_map_precharge_memlock(htab
->map
.pages
);
375 htab
->buckets
= bpf_map_area_alloc(htab
->n_buckets
*
376 sizeof(struct bucket
),
377 htab
->map
.numa_node
);
381 if (htab
->map
.map_flags
& BPF_F_ZERO_SEED
)
384 htab
->hashrnd
= get_random_int();
386 for (i
= 0; i
< htab
->n_buckets
; i
++) {
387 INIT_HLIST_NULLS_HEAD(&htab
->buckets
[i
].head
, i
);
388 raw_spin_lock_init(&htab
->buckets
[i
].lock
);
392 err
= prealloc_init(htab
);
396 if (!percpu
&& !lru
) {
397 /* lru itself can remove the least used element, so
398 * there is no need for an extra elem during map_update.
400 err
= alloc_extra_elems(htab
);
409 prealloc_destroy(htab
);
411 bpf_map_area_free(htab
->buckets
);
417 static inline u32
htab_map_hash(const void *key
, u32 key_len
, u32 hashrnd
)
419 return jhash(key
, key_len
, hashrnd
);
422 static inline struct bucket
*__select_bucket(struct bpf_htab
*htab
, u32 hash
)
424 return &htab
->buckets
[hash
& (htab
->n_buckets
- 1)];
427 static inline struct hlist_nulls_head
*select_bucket(struct bpf_htab
*htab
, u32 hash
)
429 return &__select_bucket(htab
, hash
)->head
;
432 /* this lookup function can only be called with bucket lock taken */
433 static struct htab_elem
*lookup_elem_raw(struct hlist_nulls_head
*head
, u32 hash
,
434 void *key
, u32 key_size
)
436 struct hlist_nulls_node
*n
;
439 hlist_nulls_for_each_entry_rcu(l
, n
, head
, hash_node
)
440 if (l
->hash
== hash
&& !memcmp(&l
->key
, key
, key_size
))
446 /* can be called without bucket lock. it will repeat the loop in
447 * the unlikely event when elements moved from one bucket into another
448 * while link list is being walked
450 static struct htab_elem
*lookup_nulls_elem_raw(struct hlist_nulls_head
*head
,
452 u32 key_size
, u32 n_buckets
)
454 struct hlist_nulls_node
*n
;
458 hlist_nulls_for_each_entry_rcu(l
, n
, head
, hash_node
)
459 if (l
->hash
== hash
&& !memcmp(&l
->key
, key
, key_size
))
462 if (unlikely(get_nulls_value(n
) != (hash
& (n_buckets
- 1))))
468 /* Called from syscall or from eBPF program directly, so
469 * arguments have to match bpf_map_lookup_elem() exactly.
470 * The return value is adjusted by BPF instructions
471 * in htab_map_gen_lookup().
473 static void *__htab_map_lookup_elem(struct bpf_map
*map
, void *key
)
475 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
476 struct hlist_nulls_head
*head
;
480 /* Must be called with rcu_read_lock. */
481 WARN_ON_ONCE(!rcu_read_lock_held());
483 key_size
= map
->key_size
;
485 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
487 head
= select_bucket(htab
, hash
);
489 l
= lookup_nulls_elem_raw(head
, hash
, key
, key_size
, htab
->n_buckets
);
494 static void *htab_map_lookup_elem(struct bpf_map
*map
, void *key
)
496 struct htab_elem
*l
= __htab_map_lookup_elem(map
, key
);
499 return l
->key
+ round_up(map
->key_size
, 8);
504 /* inline bpf_map_lookup_elem() call.
507 * bpf_map_lookup_elem
508 * map->ops->map_lookup_elem
509 * htab_map_lookup_elem
510 * __htab_map_lookup_elem
513 * __htab_map_lookup_elem
515 static u32
htab_map_gen_lookup(struct bpf_map
*map
, struct bpf_insn
*insn_buf
)
517 struct bpf_insn
*insn
= insn_buf
;
518 const int ret
= BPF_REG_0
;
520 BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem
,
521 (void *(*)(struct bpf_map
*map
, void *key
))NULL
));
522 *insn
++ = BPF_EMIT_CALL(BPF_CAST_CALL(__htab_map_lookup_elem
));
523 *insn
++ = BPF_JMP_IMM(BPF_JEQ
, ret
, 0, 1);
524 *insn
++ = BPF_ALU64_IMM(BPF_ADD
, ret
,
525 offsetof(struct htab_elem
, key
) +
526 round_up(map
->key_size
, 8));
527 return insn
- insn_buf
;
530 static void *htab_lru_map_lookup_elem(struct bpf_map
*map
, void *key
)
532 struct htab_elem
*l
= __htab_map_lookup_elem(map
, key
);
535 bpf_lru_node_set_ref(&l
->lru_node
);
536 return l
->key
+ round_up(map
->key_size
, 8);
542 static u32
htab_lru_map_gen_lookup(struct bpf_map
*map
,
543 struct bpf_insn
*insn_buf
)
545 struct bpf_insn
*insn
= insn_buf
;
546 const int ret
= BPF_REG_0
;
547 const int ref_reg
= BPF_REG_1
;
549 BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem
,
550 (void *(*)(struct bpf_map
*map
, void *key
))NULL
));
551 *insn
++ = BPF_EMIT_CALL(BPF_CAST_CALL(__htab_map_lookup_elem
));
552 *insn
++ = BPF_JMP_IMM(BPF_JEQ
, ret
, 0, 4);
553 *insn
++ = BPF_LDX_MEM(BPF_B
, ref_reg
, ret
,
554 offsetof(struct htab_elem
, lru_node
) +
555 offsetof(struct bpf_lru_node
, ref
));
556 *insn
++ = BPF_JMP_IMM(BPF_JNE
, ref_reg
, 0, 1);
557 *insn
++ = BPF_ST_MEM(BPF_B
, ret
,
558 offsetof(struct htab_elem
, lru_node
) +
559 offsetof(struct bpf_lru_node
, ref
),
561 *insn
++ = BPF_ALU64_IMM(BPF_ADD
, ret
,
562 offsetof(struct htab_elem
, key
) +
563 round_up(map
->key_size
, 8));
564 return insn
- insn_buf
;
567 /* It is called from the bpf_lru_list when the LRU needs to delete
568 * older elements from the htab.
570 static bool htab_lru_map_delete_node(void *arg
, struct bpf_lru_node
*node
)
572 struct bpf_htab
*htab
= (struct bpf_htab
*)arg
;
573 struct htab_elem
*l
= NULL
, *tgt_l
;
574 struct hlist_nulls_head
*head
;
575 struct hlist_nulls_node
*n
;
579 tgt_l
= container_of(node
, struct htab_elem
, lru_node
);
580 b
= __select_bucket(htab
, tgt_l
->hash
);
583 raw_spin_lock_irqsave(&b
->lock
, flags
);
585 hlist_nulls_for_each_entry_rcu(l
, n
, head
, hash_node
)
587 hlist_nulls_del_rcu(&l
->hash_node
);
591 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
596 /* Called from syscall */
597 static int htab_map_get_next_key(struct bpf_map
*map
, void *key
, void *next_key
)
599 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
600 struct hlist_nulls_head
*head
;
601 struct htab_elem
*l
, *next_l
;
605 WARN_ON_ONCE(!rcu_read_lock_held());
607 key_size
= map
->key_size
;
610 goto find_first_elem
;
612 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
614 head
= select_bucket(htab
, hash
);
617 l
= lookup_nulls_elem_raw(head
, hash
, key
, key_size
, htab
->n_buckets
);
620 goto find_first_elem
;
622 /* key was found, get next key in the same bucket */
623 next_l
= hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l
->hash_node
)),
624 struct htab_elem
, hash_node
);
627 /* if next elem in this hash list is non-zero, just return it */
628 memcpy(next_key
, next_l
->key
, key_size
);
632 /* no more elements in this hash list, go to the next bucket */
633 i
= hash
& (htab
->n_buckets
- 1);
637 /* iterate over buckets */
638 for (; i
< htab
->n_buckets
; i
++) {
639 head
= select_bucket(htab
, i
);
641 /* pick first element in the bucket */
642 next_l
= hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head
)),
643 struct htab_elem
, hash_node
);
645 /* if it's not empty, just return it */
646 memcpy(next_key
, next_l
->key
, key_size
);
651 /* iterated over all buckets and all elements */
655 static void htab_elem_free(struct bpf_htab
*htab
, struct htab_elem
*l
)
657 if (htab
->map
.map_type
== BPF_MAP_TYPE_PERCPU_HASH
)
658 free_percpu(htab_elem_get_ptr(l
, htab
->map
.key_size
));
662 static void htab_elem_free_rcu(struct rcu_head
*head
)
664 struct htab_elem
*l
= container_of(head
, struct htab_elem
, rcu
);
665 struct bpf_htab
*htab
= l
->htab
;
667 /* must increment bpf_prog_active to avoid kprobe+bpf triggering while
668 * we're calling kfree, otherwise deadlock is possible if kprobes
669 * are placed somewhere inside of slub
672 __this_cpu_inc(bpf_prog_active
);
673 htab_elem_free(htab
, l
);
674 __this_cpu_dec(bpf_prog_active
);
678 static void free_htab_elem(struct bpf_htab
*htab
, struct htab_elem
*l
)
680 struct bpf_map
*map
= &htab
->map
;
682 if (map
->ops
->map_fd_put_ptr
) {
683 void *ptr
= fd_htab_map_get_ptr(map
, l
);
685 map
->ops
->map_fd_put_ptr(ptr
);
688 if (htab_is_prealloc(htab
)) {
689 __pcpu_freelist_push(&htab
->freelist
, &l
->fnode
);
691 atomic_dec(&htab
->count
);
693 call_rcu(&l
->rcu
, htab_elem_free_rcu
);
697 static void pcpu_copy_value(struct bpf_htab
*htab
, void __percpu
*pptr
,
698 void *value
, bool onallcpus
)
701 /* copy true value_size bytes */
702 memcpy(this_cpu_ptr(pptr
), value
, htab
->map
.value_size
);
704 u32 size
= round_up(htab
->map
.value_size
, 8);
707 for_each_possible_cpu(cpu
) {
708 bpf_long_memcpy(per_cpu_ptr(pptr
, cpu
),
715 static bool fd_htab_map_needs_adjust(const struct bpf_htab
*htab
)
717 return htab
->map
.map_type
== BPF_MAP_TYPE_HASH_OF_MAPS
&&
721 static u32
htab_size_value(const struct bpf_htab
*htab
, bool percpu
)
723 u32 size
= htab
->map
.value_size
;
725 if (percpu
|| fd_htab_map_needs_adjust(htab
))
726 size
= round_up(size
, 8);
730 static struct htab_elem
*alloc_htab_elem(struct bpf_htab
*htab
, void *key
,
731 void *value
, u32 key_size
, u32 hash
,
732 bool percpu
, bool onallcpus
,
733 struct htab_elem
*old_elem
)
735 u32 size
= htab_size_value(htab
, percpu
);
736 bool prealloc
= htab_is_prealloc(htab
);
737 struct htab_elem
*l_new
, **pl_new
;
742 /* if we're updating the existing element,
743 * use per-cpu extra elems to avoid freelist_pop/push
745 pl_new
= this_cpu_ptr(htab
->extra_elems
);
749 struct pcpu_freelist_node
*l
;
751 l
= __pcpu_freelist_pop(&htab
->freelist
);
753 return ERR_PTR(-E2BIG
);
754 l_new
= container_of(l
, struct htab_elem
, fnode
);
757 if (atomic_inc_return(&htab
->count
) > htab
->map
.max_entries
)
759 /* when map is full and update() is replacing
760 * old element, it's ok to allocate, since
761 * old element will be freed immediately.
762 * Otherwise return an error
764 l_new
= ERR_PTR(-E2BIG
);
767 l_new
= kmalloc_node(htab
->elem_size
, GFP_ATOMIC
| __GFP_NOWARN
,
768 htab
->map
.numa_node
);
770 l_new
= ERR_PTR(-ENOMEM
);
775 memcpy(l_new
->key
, key
, key_size
);
778 pptr
= htab_elem_get_ptr(l_new
, key_size
);
780 /* alloc_percpu zero-fills */
781 pptr
= __alloc_percpu_gfp(size
, 8,
782 GFP_ATOMIC
| __GFP_NOWARN
);
785 l_new
= ERR_PTR(-ENOMEM
);
790 pcpu_copy_value(htab
, pptr
, value
, onallcpus
);
793 htab_elem_set_ptr(l_new
, key_size
, pptr
);
795 memcpy(l_new
->key
+ round_up(key_size
, 8), value
, size
);
801 atomic_dec(&htab
->count
);
805 static int check_flags(struct bpf_htab
*htab
, struct htab_elem
*l_old
,
808 if (l_old
&& map_flags
== BPF_NOEXIST
)
809 /* elem already exists */
812 if (!l_old
&& map_flags
== BPF_EXIST
)
813 /* elem doesn't exist, cannot update it */
819 /* Called from syscall or from eBPF program */
820 static int htab_map_update_elem(struct bpf_map
*map
, void *key
, void *value
,
823 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
824 struct htab_elem
*l_new
= NULL
, *l_old
;
825 struct hlist_nulls_head
*head
;
831 if (unlikely(map_flags
> BPF_EXIST
))
835 WARN_ON_ONCE(!rcu_read_lock_held());
837 key_size
= map
->key_size
;
839 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
841 b
= __select_bucket(htab
, hash
);
844 /* bpf_map_update_elem() can be called in_irq() */
845 raw_spin_lock_irqsave(&b
->lock
, flags
);
847 l_old
= lookup_elem_raw(head
, hash
, key
, key_size
);
849 ret
= check_flags(htab
, l_old
, map_flags
);
853 l_new
= alloc_htab_elem(htab
, key
, value
, key_size
, hash
, false, false,
856 /* all pre-allocated elements are in use or memory exhausted */
857 ret
= PTR_ERR(l_new
);
861 /* add new element to the head of the list, so that
862 * concurrent search will find it before old elem
864 hlist_nulls_add_head_rcu(&l_new
->hash_node
, head
);
866 hlist_nulls_del_rcu(&l_old
->hash_node
);
867 if (!htab_is_prealloc(htab
))
868 free_htab_elem(htab
, l_old
);
872 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
876 static int htab_lru_map_update_elem(struct bpf_map
*map
, void *key
, void *value
,
879 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
880 struct htab_elem
*l_new
, *l_old
= NULL
;
881 struct hlist_nulls_head
*head
;
887 if (unlikely(map_flags
> BPF_EXIST
))
891 WARN_ON_ONCE(!rcu_read_lock_held());
893 key_size
= map
->key_size
;
895 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
897 b
= __select_bucket(htab
, hash
);
900 /* For LRU, we need to alloc before taking bucket's
901 * spinlock because getting free nodes from LRU may need
902 * to remove older elements from htab and this removal
903 * operation will need a bucket lock.
905 l_new
= prealloc_lru_pop(htab
, key
, hash
);
908 memcpy(l_new
->key
+ round_up(map
->key_size
, 8), value
, map
->value_size
);
910 /* bpf_map_update_elem() can be called in_irq() */
911 raw_spin_lock_irqsave(&b
->lock
, flags
);
913 l_old
= lookup_elem_raw(head
, hash
, key
, key_size
);
915 ret
= check_flags(htab
, l_old
, map_flags
);
919 /* add new element to the head of the list, so that
920 * concurrent search will find it before old elem
922 hlist_nulls_add_head_rcu(&l_new
->hash_node
, head
);
924 bpf_lru_node_set_ref(&l_new
->lru_node
);
925 hlist_nulls_del_rcu(&l_old
->hash_node
);
930 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
933 bpf_lru_push_free(&htab
->lru
, &l_new
->lru_node
);
935 bpf_lru_push_free(&htab
->lru
, &l_old
->lru_node
);
940 static int __htab_percpu_map_update_elem(struct bpf_map
*map
, void *key
,
941 void *value
, u64 map_flags
,
944 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
945 struct htab_elem
*l_new
= NULL
, *l_old
;
946 struct hlist_nulls_head
*head
;
952 if (unlikely(map_flags
> BPF_EXIST
))
956 WARN_ON_ONCE(!rcu_read_lock_held());
958 key_size
= map
->key_size
;
960 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
962 b
= __select_bucket(htab
, hash
);
965 /* bpf_map_update_elem() can be called in_irq() */
966 raw_spin_lock_irqsave(&b
->lock
, flags
);
968 l_old
= lookup_elem_raw(head
, hash
, key
, key_size
);
970 ret
= check_flags(htab
, l_old
, map_flags
);
975 /* per-cpu hash map can update value in-place */
976 pcpu_copy_value(htab
, htab_elem_get_ptr(l_old
, key_size
),
979 l_new
= alloc_htab_elem(htab
, key
, value
, key_size
,
980 hash
, true, onallcpus
, NULL
);
982 ret
= PTR_ERR(l_new
);
985 hlist_nulls_add_head_rcu(&l_new
->hash_node
, head
);
989 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
993 static int __htab_lru_percpu_map_update_elem(struct bpf_map
*map
, void *key
,
994 void *value
, u64 map_flags
,
997 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
998 struct htab_elem
*l_new
= NULL
, *l_old
;
999 struct hlist_nulls_head
*head
;
1000 unsigned long flags
;
1005 if (unlikely(map_flags
> BPF_EXIST
))
1009 WARN_ON_ONCE(!rcu_read_lock_held());
1011 key_size
= map
->key_size
;
1013 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
1015 b
= __select_bucket(htab
, hash
);
1018 /* For LRU, we need to alloc before taking bucket's
1019 * spinlock because LRU's elem alloc may need
1020 * to remove older elem from htab and this removal
1021 * operation will need a bucket lock.
1023 if (map_flags
!= BPF_EXIST
) {
1024 l_new
= prealloc_lru_pop(htab
, key
, hash
);
1029 /* bpf_map_update_elem() can be called in_irq() */
1030 raw_spin_lock_irqsave(&b
->lock
, flags
);
1032 l_old
= lookup_elem_raw(head
, hash
, key
, key_size
);
1034 ret
= check_flags(htab
, l_old
, map_flags
);
1039 bpf_lru_node_set_ref(&l_old
->lru_node
);
1041 /* per-cpu hash map can update value in-place */
1042 pcpu_copy_value(htab
, htab_elem_get_ptr(l_old
, key_size
),
1045 pcpu_copy_value(htab
, htab_elem_get_ptr(l_new
, key_size
),
1047 hlist_nulls_add_head_rcu(&l_new
->hash_node
, head
);
1052 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
1054 bpf_lru_push_free(&htab
->lru
, &l_new
->lru_node
);
1058 static int htab_percpu_map_update_elem(struct bpf_map
*map
, void *key
,
1059 void *value
, u64 map_flags
)
1061 return __htab_percpu_map_update_elem(map
, key
, value
, map_flags
, false);
1064 static int htab_lru_percpu_map_update_elem(struct bpf_map
*map
, void *key
,
1065 void *value
, u64 map_flags
)
1067 return __htab_lru_percpu_map_update_elem(map
, key
, value
, map_flags
,
1071 /* Called from syscall or from eBPF program */
1072 static int htab_map_delete_elem(struct bpf_map
*map
, void *key
)
1074 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
1075 struct hlist_nulls_head
*head
;
1077 struct htab_elem
*l
;
1078 unsigned long flags
;
1082 WARN_ON_ONCE(!rcu_read_lock_held());
1084 key_size
= map
->key_size
;
1086 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
1087 b
= __select_bucket(htab
, hash
);
1090 raw_spin_lock_irqsave(&b
->lock
, flags
);
1092 l
= lookup_elem_raw(head
, hash
, key
, key_size
);
1095 hlist_nulls_del_rcu(&l
->hash_node
);
1096 free_htab_elem(htab
, l
);
1100 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
1104 static int htab_lru_map_delete_elem(struct bpf_map
*map
, void *key
)
1106 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
1107 struct hlist_nulls_head
*head
;
1109 struct htab_elem
*l
;
1110 unsigned long flags
;
1114 WARN_ON_ONCE(!rcu_read_lock_held());
1116 key_size
= map
->key_size
;
1118 hash
= htab_map_hash(key
, key_size
, htab
->hashrnd
);
1119 b
= __select_bucket(htab
, hash
);
1122 raw_spin_lock_irqsave(&b
->lock
, flags
);
1124 l
= lookup_elem_raw(head
, hash
, key
, key_size
);
1127 hlist_nulls_del_rcu(&l
->hash_node
);
1131 raw_spin_unlock_irqrestore(&b
->lock
, flags
);
1133 bpf_lru_push_free(&htab
->lru
, &l
->lru_node
);
1137 static void delete_all_elements(struct bpf_htab
*htab
)
1141 for (i
= 0; i
< htab
->n_buckets
; i
++) {
1142 struct hlist_nulls_head
*head
= select_bucket(htab
, i
);
1143 struct hlist_nulls_node
*n
;
1144 struct htab_elem
*l
;
1146 hlist_nulls_for_each_entry_safe(l
, n
, head
, hash_node
) {
1147 hlist_nulls_del_rcu(&l
->hash_node
);
1148 htab_elem_free(htab
, l
);
1153 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
1154 static void htab_map_free(struct bpf_map
*map
)
1156 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
1158 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
1159 * so the programs (can be more than one that used this map) were
1160 * disconnected from events. Wait for outstanding critical sections in
1161 * these programs to complete
1165 /* some of free_htab_elem() callbacks for elements of this map may
1166 * not have executed. Wait for them.
1169 if (!htab_is_prealloc(htab
))
1170 delete_all_elements(htab
);
1172 prealloc_destroy(htab
);
1174 free_percpu(htab
->extra_elems
);
1175 bpf_map_area_free(htab
->buckets
);
1179 static void htab_map_seq_show_elem(struct bpf_map
*map
, void *key
,
1186 value
= htab_map_lookup_elem(map
, key
);
1192 btf_type_seq_show(map
->btf
, map
->btf_key_type_id
, key
, m
);
1194 btf_type_seq_show(map
->btf
, map
->btf_value_type_id
, value
, m
);
1200 const struct bpf_map_ops htab_map_ops
= {
1201 .map_alloc_check
= htab_map_alloc_check
,
1202 .map_alloc
= htab_map_alloc
,
1203 .map_free
= htab_map_free
,
1204 .map_get_next_key
= htab_map_get_next_key
,
1205 .map_lookup_elem
= htab_map_lookup_elem
,
1206 .map_update_elem
= htab_map_update_elem
,
1207 .map_delete_elem
= htab_map_delete_elem
,
1208 .map_gen_lookup
= htab_map_gen_lookup
,
1209 .map_seq_show_elem
= htab_map_seq_show_elem
,
1212 const struct bpf_map_ops htab_lru_map_ops
= {
1213 .map_alloc_check
= htab_map_alloc_check
,
1214 .map_alloc
= htab_map_alloc
,
1215 .map_free
= htab_map_free
,
1216 .map_get_next_key
= htab_map_get_next_key
,
1217 .map_lookup_elem
= htab_lru_map_lookup_elem
,
1218 .map_update_elem
= htab_lru_map_update_elem
,
1219 .map_delete_elem
= htab_lru_map_delete_elem
,
1220 .map_gen_lookup
= htab_lru_map_gen_lookup
,
1221 .map_seq_show_elem
= htab_map_seq_show_elem
,
1224 /* Called from eBPF program */
1225 static void *htab_percpu_map_lookup_elem(struct bpf_map
*map
, void *key
)
1227 struct htab_elem
*l
= __htab_map_lookup_elem(map
, key
);
1230 return this_cpu_ptr(htab_elem_get_ptr(l
, map
->key_size
));
1235 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map
*map
, void *key
)
1237 struct htab_elem
*l
= __htab_map_lookup_elem(map
, key
);
1240 bpf_lru_node_set_ref(&l
->lru_node
);
1241 return this_cpu_ptr(htab_elem_get_ptr(l
, map
->key_size
));
1247 int bpf_percpu_hash_copy(struct bpf_map
*map
, void *key
, void *value
)
1249 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
1250 struct htab_elem
*l
;
1251 void __percpu
*pptr
;
1256 /* per_cpu areas are zero-filled and bpf programs can only
1257 * access 'value_size' of them, so copying rounded areas
1258 * will not leak any kernel data
1260 size
= round_up(map
->value_size
, 8);
1262 l
= __htab_map_lookup_elem(map
, key
);
1265 if (htab_is_lru(htab
))
1266 bpf_lru_node_set_ref(&l
->lru_node
);
1267 pptr
= htab_elem_get_ptr(l
, map
->key_size
);
1268 for_each_possible_cpu(cpu
) {
1269 bpf_long_memcpy(value
+ off
,
1270 per_cpu_ptr(pptr
, cpu
), size
);
1279 int bpf_percpu_hash_update(struct bpf_map
*map
, void *key
, void *value
,
1282 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
1286 if (htab_is_lru(htab
))
1287 ret
= __htab_lru_percpu_map_update_elem(map
, key
, value
,
1290 ret
= __htab_percpu_map_update_elem(map
, key
, value
, map_flags
,
1297 static void htab_percpu_map_seq_show_elem(struct bpf_map
*map
, void *key
,
1300 struct htab_elem
*l
;
1301 void __percpu
*pptr
;
1306 l
= __htab_map_lookup_elem(map
, key
);
1312 btf_type_seq_show(map
->btf
, map
->btf_key_type_id
, key
, m
);
1313 seq_puts(m
, ": {\n");
1314 pptr
= htab_elem_get_ptr(l
, map
->key_size
);
1315 for_each_possible_cpu(cpu
) {
1316 seq_printf(m
, "\tcpu%d: ", cpu
);
1317 btf_type_seq_show(map
->btf
, map
->btf_value_type_id
,
1318 per_cpu_ptr(pptr
, cpu
), m
);
1326 const struct bpf_map_ops htab_percpu_map_ops
= {
1327 .map_alloc_check
= htab_map_alloc_check
,
1328 .map_alloc
= htab_map_alloc
,
1329 .map_free
= htab_map_free
,
1330 .map_get_next_key
= htab_map_get_next_key
,
1331 .map_lookup_elem
= htab_percpu_map_lookup_elem
,
1332 .map_update_elem
= htab_percpu_map_update_elem
,
1333 .map_delete_elem
= htab_map_delete_elem
,
1334 .map_seq_show_elem
= htab_percpu_map_seq_show_elem
,
1337 const struct bpf_map_ops htab_lru_percpu_map_ops
= {
1338 .map_alloc_check
= htab_map_alloc_check
,
1339 .map_alloc
= htab_map_alloc
,
1340 .map_free
= htab_map_free
,
1341 .map_get_next_key
= htab_map_get_next_key
,
1342 .map_lookup_elem
= htab_lru_percpu_map_lookup_elem
,
1343 .map_update_elem
= htab_lru_percpu_map_update_elem
,
1344 .map_delete_elem
= htab_lru_map_delete_elem
,
1345 .map_seq_show_elem
= htab_percpu_map_seq_show_elem
,
1348 static int fd_htab_map_alloc_check(union bpf_attr
*attr
)
1350 if (attr
->value_size
!= sizeof(u32
))
1352 return htab_map_alloc_check(attr
);
1355 static void fd_htab_map_free(struct bpf_map
*map
)
1357 struct bpf_htab
*htab
= container_of(map
, struct bpf_htab
, map
);
1358 struct hlist_nulls_node
*n
;
1359 struct hlist_nulls_head
*head
;
1360 struct htab_elem
*l
;
1363 for (i
= 0; i
< htab
->n_buckets
; i
++) {
1364 head
= select_bucket(htab
, i
);
1366 hlist_nulls_for_each_entry_safe(l
, n
, head
, hash_node
) {
1367 void *ptr
= fd_htab_map_get_ptr(map
, l
);
1369 map
->ops
->map_fd_put_ptr(ptr
);
1376 /* only called from syscall */
1377 int bpf_fd_htab_map_lookup_elem(struct bpf_map
*map
, void *key
, u32
*value
)
1382 if (!map
->ops
->map_fd_sys_lookup_elem
)
1386 ptr
= htab_map_lookup_elem(map
, key
);
1388 *value
= map
->ops
->map_fd_sys_lookup_elem(READ_ONCE(*ptr
));
1396 /* only called from syscall */
1397 int bpf_fd_htab_map_update_elem(struct bpf_map
*map
, struct file
*map_file
,
1398 void *key
, void *value
, u64 map_flags
)
1402 u32 ufd
= *(u32
*)value
;
1404 ptr
= map
->ops
->map_fd_get_ptr(map
, map_file
, ufd
);
1406 return PTR_ERR(ptr
);
1408 ret
= htab_map_update_elem(map
, key
, &ptr
, map_flags
);
1410 map
->ops
->map_fd_put_ptr(ptr
);
1415 static struct bpf_map
*htab_of_map_alloc(union bpf_attr
*attr
)
1417 struct bpf_map
*map
, *inner_map_meta
;
1419 inner_map_meta
= bpf_map_meta_alloc(attr
->inner_map_fd
);
1420 if (IS_ERR(inner_map_meta
))
1421 return inner_map_meta
;
1423 map
= htab_map_alloc(attr
);
1425 bpf_map_meta_free(inner_map_meta
);
1429 map
->inner_map_meta
= inner_map_meta
;
1434 static void *htab_of_map_lookup_elem(struct bpf_map
*map
, void *key
)
1436 struct bpf_map
**inner_map
= htab_map_lookup_elem(map
, key
);
1441 return READ_ONCE(*inner_map
);
1444 static u32
htab_of_map_gen_lookup(struct bpf_map
*map
,
1445 struct bpf_insn
*insn_buf
)
1447 struct bpf_insn
*insn
= insn_buf
;
1448 const int ret
= BPF_REG_0
;
1450 BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem
,
1451 (void *(*)(struct bpf_map
*map
, void *key
))NULL
));
1452 *insn
++ = BPF_EMIT_CALL(BPF_CAST_CALL(__htab_map_lookup_elem
));
1453 *insn
++ = BPF_JMP_IMM(BPF_JEQ
, ret
, 0, 2);
1454 *insn
++ = BPF_ALU64_IMM(BPF_ADD
, ret
,
1455 offsetof(struct htab_elem
, key
) +
1456 round_up(map
->key_size
, 8));
1457 *insn
++ = BPF_LDX_MEM(BPF_DW
, ret
, ret
, 0);
1459 return insn
- insn_buf
;
1462 static void htab_of_map_free(struct bpf_map
*map
)
1464 bpf_map_meta_free(map
->inner_map_meta
);
1465 fd_htab_map_free(map
);
1468 const struct bpf_map_ops htab_of_maps_map_ops
= {
1469 .map_alloc_check
= fd_htab_map_alloc_check
,
1470 .map_alloc
= htab_of_map_alloc
,
1471 .map_free
= htab_of_map_free
,
1472 .map_get_next_key
= htab_map_get_next_key
,
1473 .map_lookup_elem
= htab_of_map_lookup_elem
,
1474 .map_delete_elem
= htab_map_delete_elem
,
1475 .map_fd_get_ptr
= bpf_map_fd_get_ptr
,
1476 .map_fd_put_ptr
= bpf_map_fd_put_ptr
,
1477 .map_fd_sys_lookup_elem
= bpf_map_fd_sys_lookup_elem
,
1478 .map_gen_lookup
= htab_of_map_gen_lookup
,
1479 .map_check_btf
= map_check_no_btf
,