Loading...
Note: File does not exist in v3.15.
1/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
3 *
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.
7 *
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.
12 */
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
19struct bucket {
20 struct hlist_head head;
21 raw_spinlock_t lock;
22};
23
24struct bpf_htab {
25 struct bpf_map map;
26 struct bucket *buckets;
27 void *elems;
28 struct pcpu_freelist freelist;
29 atomic_t count; /* number of elements in this hashtable */
30 u32 n_buckets; /* number of hash buckets */
31 u32 elem_size; /* size of each element in bytes */
32};
33
34/* each htab element is struct htab_elem + key + value */
35struct htab_elem {
36 union {
37 struct hlist_node hash_node;
38 struct bpf_htab *htab;
39 struct pcpu_freelist_node fnode;
40 };
41 struct rcu_head rcu;
42 u32 hash;
43 char key[0] __aligned(8);
44};
45
46static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
47 void __percpu *pptr)
48{
49 *(void __percpu **)(l->key + key_size) = pptr;
50}
51
52static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
53{
54 return *(void __percpu **)(l->key + key_size);
55}
56
57static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
58{
59 return (struct htab_elem *) (htab->elems + i * htab->elem_size);
60}
61
62static void htab_free_elems(struct bpf_htab *htab)
63{
64 int i;
65
66 if (htab->map.map_type != BPF_MAP_TYPE_PERCPU_HASH)
67 goto free_elems;
68
69 for (i = 0; i < htab->map.max_entries; i++) {
70 void __percpu *pptr;
71
72 pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
73 htab->map.key_size);
74 free_percpu(pptr);
75 }
76free_elems:
77 vfree(htab->elems);
78}
79
80static int prealloc_elems_and_freelist(struct bpf_htab *htab)
81{
82 int err = -ENOMEM, i;
83
84 htab->elems = vzalloc(htab->elem_size * htab->map.max_entries);
85 if (!htab->elems)
86 return -ENOMEM;
87
88 if (htab->map.map_type != BPF_MAP_TYPE_PERCPU_HASH)
89 goto skip_percpu_elems;
90
91 for (i = 0; i < htab->map.max_entries; i++) {
92 u32 size = round_up(htab->map.value_size, 8);
93 void __percpu *pptr;
94
95 pptr = __alloc_percpu_gfp(size, 8, GFP_USER | __GFP_NOWARN);
96 if (!pptr)
97 goto free_elems;
98 htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
99 pptr);
100 }
101
102skip_percpu_elems:
103 err = pcpu_freelist_init(&htab->freelist);
104 if (err)
105 goto free_elems;
106
107 pcpu_freelist_populate(&htab->freelist, htab->elems, htab->elem_size,
108 htab->map.max_entries);
109 return 0;
110
111free_elems:
112 htab_free_elems(htab);
113 return err;
114}
115
116/* Called from syscall */
117static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
118{
119 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_HASH;
120 struct bpf_htab *htab;
121 int err, i;
122 u64 cost;
123
124 if (attr->map_flags & ~BPF_F_NO_PREALLOC)
125 /* reserved bits should not be used */
126 return ERR_PTR(-EINVAL);
127
128 htab = kzalloc(sizeof(*htab), GFP_USER);
129 if (!htab)
130 return ERR_PTR(-ENOMEM);
131
132 /* mandatory map attributes */
133 htab->map.map_type = attr->map_type;
134 htab->map.key_size = attr->key_size;
135 htab->map.value_size = attr->value_size;
136 htab->map.max_entries = attr->max_entries;
137 htab->map.map_flags = attr->map_flags;
138
139 /* check sanity of attributes.
140 * value_size == 0 may be allowed in the future to use map as a set
141 */
142 err = -EINVAL;
143 if (htab->map.max_entries == 0 || htab->map.key_size == 0 ||
144 htab->map.value_size == 0)
145 goto free_htab;
146
147 /* hash table size must be power of 2 */
148 htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
149
150 err = -E2BIG;
151 if (htab->map.key_size > MAX_BPF_STACK)
152 /* eBPF programs initialize keys on stack, so they cannot be
153 * larger than max stack size
154 */
155 goto free_htab;
156
157 if (htab->map.value_size >= (1 << (KMALLOC_SHIFT_MAX - 1)) -
158 MAX_BPF_STACK - sizeof(struct htab_elem))
159 /* if value_size is bigger, the user space won't be able to
160 * access the elements via bpf syscall. This check also makes
161 * sure that the elem_size doesn't overflow and it's
162 * kmalloc-able later in htab_map_update_elem()
163 */
164 goto free_htab;
165
166 if (percpu && round_up(htab->map.value_size, 8) > PCPU_MIN_UNIT_SIZE)
167 /* make sure the size for pcpu_alloc() is reasonable */
168 goto free_htab;
169
170 htab->elem_size = sizeof(struct htab_elem) +
171 round_up(htab->map.key_size, 8);
172 if (percpu)
173 htab->elem_size += sizeof(void *);
174 else
175 htab->elem_size += round_up(htab->map.value_size, 8);
176
177 /* prevent zero size kmalloc and check for u32 overflow */
178 if (htab->n_buckets == 0 ||
179 htab->n_buckets > U32_MAX / sizeof(struct bucket))
180 goto free_htab;
181
182 cost = (u64) htab->n_buckets * sizeof(struct bucket) +
183 (u64) htab->elem_size * htab->map.max_entries;
184
185 if (percpu)
186 cost += (u64) round_up(htab->map.value_size, 8) *
187 num_possible_cpus() * htab->map.max_entries;
188
189 if (cost >= U32_MAX - PAGE_SIZE)
190 /* make sure page count doesn't overflow */
191 goto free_htab;
192
193 htab->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
194
195 /* if map size is larger than memlock limit, reject it early */
196 err = bpf_map_precharge_memlock(htab->map.pages);
197 if (err)
198 goto free_htab;
199
200 err = -ENOMEM;
201 htab->buckets = kmalloc_array(htab->n_buckets, sizeof(struct bucket),
202 GFP_USER | __GFP_NOWARN);
203
204 if (!htab->buckets) {
205 htab->buckets = vmalloc(htab->n_buckets * sizeof(struct bucket));
206 if (!htab->buckets)
207 goto free_htab;
208 }
209
210 for (i = 0; i < htab->n_buckets; i++) {
211 INIT_HLIST_HEAD(&htab->buckets[i].head);
212 raw_spin_lock_init(&htab->buckets[i].lock);
213 }
214
215 if (!(attr->map_flags & BPF_F_NO_PREALLOC)) {
216 err = prealloc_elems_and_freelist(htab);
217 if (err)
218 goto free_buckets;
219 }
220
221 return &htab->map;
222
223free_buckets:
224 kvfree(htab->buckets);
225free_htab:
226 kfree(htab);
227 return ERR_PTR(err);
228}
229
230static inline u32 htab_map_hash(const void *key, u32 key_len)
231{
232 return jhash(key, key_len, 0);
233}
234
235static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
236{
237 return &htab->buckets[hash & (htab->n_buckets - 1)];
238}
239
240static inline struct hlist_head *select_bucket(struct bpf_htab *htab, u32 hash)
241{
242 return &__select_bucket(htab, hash)->head;
243}
244
245static struct htab_elem *lookup_elem_raw(struct hlist_head *head, u32 hash,
246 void *key, u32 key_size)
247{
248 struct htab_elem *l;
249
250 hlist_for_each_entry_rcu(l, head, hash_node)
251 if (l->hash == hash && !memcmp(&l->key, key, key_size))
252 return l;
253
254 return NULL;
255}
256
257/* Called from syscall or from eBPF program */
258static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
259{
260 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
261 struct hlist_head *head;
262 struct htab_elem *l;
263 u32 hash, key_size;
264
265 /* Must be called with rcu_read_lock. */
266 WARN_ON_ONCE(!rcu_read_lock_held());
267
268 key_size = map->key_size;
269
270 hash = htab_map_hash(key, key_size);
271
272 head = select_bucket(htab, hash);
273
274 l = lookup_elem_raw(head, hash, key, key_size);
275
276 return l;
277}
278
279static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
280{
281 struct htab_elem *l = __htab_map_lookup_elem(map, key);
282
283 if (l)
284 return l->key + round_up(map->key_size, 8);
285
286 return NULL;
287}
288
289/* Called from syscall */
290static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
291{
292 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
293 struct hlist_head *head;
294 struct htab_elem *l, *next_l;
295 u32 hash, key_size;
296 int i;
297
298 WARN_ON_ONCE(!rcu_read_lock_held());
299
300 key_size = map->key_size;
301
302 hash = htab_map_hash(key, key_size);
303
304 head = select_bucket(htab, hash);
305
306 /* lookup the key */
307 l = lookup_elem_raw(head, hash, key, key_size);
308
309 if (!l) {
310 i = 0;
311 goto find_first_elem;
312 }
313
314 /* key was found, get next key in the same bucket */
315 next_l = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&l->hash_node)),
316 struct htab_elem, hash_node);
317
318 if (next_l) {
319 /* if next elem in this hash list is non-zero, just return it */
320 memcpy(next_key, next_l->key, key_size);
321 return 0;
322 }
323
324 /* no more elements in this hash list, go to the next bucket */
325 i = hash & (htab->n_buckets - 1);
326 i++;
327
328find_first_elem:
329 /* iterate over buckets */
330 for (; i < htab->n_buckets; i++) {
331 head = select_bucket(htab, i);
332
333 /* pick first element in the bucket */
334 next_l = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
335 struct htab_elem, hash_node);
336 if (next_l) {
337 /* if it's not empty, just return it */
338 memcpy(next_key, next_l->key, key_size);
339 return 0;
340 }
341 }
342
343 /* iterated over all buckets and all elements */
344 return -ENOENT;
345}
346
347static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
348{
349 if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
350 free_percpu(htab_elem_get_ptr(l, htab->map.key_size));
351 kfree(l);
352
353}
354
355static void htab_elem_free_rcu(struct rcu_head *head)
356{
357 struct htab_elem *l = container_of(head, struct htab_elem, rcu);
358 struct bpf_htab *htab = l->htab;
359
360 /* must increment bpf_prog_active to avoid kprobe+bpf triggering while
361 * we're calling kfree, otherwise deadlock is possible if kprobes
362 * are placed somewhere inside of slub
363 */
364 preempt_disable();
365 __this_cpu_inc(bpf_prog_active);
366 htab_elem_free(htab, l);
367 __this_cpu_dec(bpf_prog_active);
368 preempt_enable();
369}
370
371static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
372{
373 if (!(htab->map.map_flags & BPF_F_NO_PREALLOC)) {
374 pcpu_freelist_push(&htab->freelist, &l->fnode);
375 } else {
376 atomic_dec(&htab->count);
377 l->htab = htab;
378 call_rcu(&l->rcu, htab_elem_free_rcu);
379 }
380}
381
382static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
383 void *value, u32 key_size, u32 hash,
384 bool percpu, bool onallcpus)
385{
386 u32 size = htab->map.value_size;
387 bool prealloc = !(htab->map.map_flags & BPF_F_NO_PREALLOC);
388 struct htab_elem *l_new;
389 void __percpu *pptr;
390
391 if (prealloc) {
392 l_new = (struct htab_elem *)pcpu_freelist_pop(&htab->freelist);
393 if (!l_new)
394 return ERR_PTR(-E2BIG);
395 } else {
396 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
397 atomic_dec(&htab->count);
398 return ERR_PTR(-E2BIG);
399 }
400 l_new = kmalloc(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN);
401 if (!l_new)
402 return ERR_PTR(-ENOMEM);
403 }
404
405 memcpy(l_new->key, key, key_size);
406 if (percpu) {
407 /* round up value_size to 8 bytes */
408 size = round_up(size, 8);
409
410 if (prealloc) {
411 pptr = htab_elem_get_ptr(l_new, key_size);
412 } else {
413 /* alloc_percpu zero-fills */
414 pptr = __alloc_percpu_gfp(size, 8,
415 GFP_ATOMIC | __GFP_NOWARN);
416 if (!pptr) {
417 kfree(l_new);
418 return ERR_PTR(-ENOMEM);
419 }
420 }
421
422 if (!onallcpus) {
423 /* copy true value_size bytes */
424 memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
425 } else {
426 int off = 0, cpu;
427
428 for_each_possible_cpu(cpu) {
429 bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
430 value + off, size);
431 off += size;
432 }
433 }
434 if (!prealloc)
435 htab_elem_set_ptr(l_new, key_size, pptr);
436 } else {
437 memcpy(l_new->key + round_up(key_size, 8), value, size);
438 }
439
440 l_new->hash = hash;
441 return l_new;
442}
443
444static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
445 u64 map_flags)
446{
447 if (l_old && map_flags == BPF_NOEXIST)
448 /* elem already exists */
449 return -EEXIST;
450
451 if (!l_old && map_flags == BPF_EXIST)
452 /* elem doesn't exist, cannot update it */
453 return -ENOENT;
454
455 return 0;
456}
457
458/* Called from syscall or from eBPF program */
459static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
460 u64 map_flags)
461{
462 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
463 struct htab_elem *l_new = NULL, *l_old;
464 struct hlist_head *head;
465 unsigned long flags;
466 struct bucket *b;
467 u32 key_size, hash;
468 int ret;
469
470 if (unlikely(map_flags > BPF_EXIST))
471 /* unknown flags */
472 return -EINVAL;
473
474 WARN_ON_ONCE(!rcu_read_lock_held());
475
476 key_size = map->key_size;
477
478 hash = htab_map_hash(key, key_size);
479
480 b = __select_bucket(htab, hash);
481 head = &b->head;
482
483 /* bpf_map_update_elem() can be called in_irq() */
484 raw_spin_lock_irqsave(&b->lock, flags);
485
486 l_old = lookup_elem_raw(head, hash, key, key_size);
487
488 ret = check_flags(htab, l_old, map_flags);
489 if (ret)
490 goto err;
491
492 l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false);
493 if (IS_ERR(l_new)) {
494 /* all pre-allocated elements are in use or memory exhausted */
495 ret = PTR_ERR(l_new);
496 goto err;
497 }
498
499 /* add new element to the head of the list, so that
500 * concurrent search will find it before old elem
501 */
502 hlist_add_head_rcu(&l_new->hash_node, head);
503 if (l_old) {
504 hlist_del_rcu(&l_old->hash_node);
505 free_htab_elem(htab, l_old);
506 }
507 ret = 0;
508err:
509 raw_spin_unlock_irqrestore(&b->lock, flags);
510 return ret;
511}
512
513static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
514 void *value, u64 map_flags,
515 bool onallcpus)
516{
517 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
518 struct htab_elem *l_new = NULL, *l_old;
519 struct hlist_head *head;
520 unsigned long flags;
521 struct bucket *b;
522 u32 key_size, hash;
523 int ret;
524
525 if (unlikely(map_flags > BPF_EXIST))
526 /* unknown flags */
527 return -EINVAL;
528
529 WARN_ON_ONCE(!rcu_read_lock_held());
530
531 key_size = map->key_size;
532
533 hash = htab_map_hash(key, key_size);
534
535 b = __select_bucket(htab, hash);
536 head = &b->head;
537
538 /* bpf_map_update_elem() can be called in_irq() */
539 raw_spin_lock_irqsave(&b->lock, flags);
540
541 l_old = lookup_elem_raw(head, hash, key, key_size);
542
543 ret = check_flags(htab, l_old, map_flags);
544 if (ret)
545 goto err;
546
547 if (l_old) {
548 void __percpu *pptr = htab_elem_get_ptr(l_old, key_size);
549 u32 size = htab->map.value_size;
550
551 /* per-cpu hash map can update value in-place */
552 if (!onallcpus) {
553 memcpy(this_cpu_ptr(pptr), value, size);
554 } else {
555 int off = 0, cpu;
556
557 size = round_up(size, 8);
558 for_each_possible_cpu(cpu) {
559 bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
560 value + off, size);
561 off += size;
562 }
563 }
564 } else {
565 l_new = alloc_htab_elem(htab, key, value, key_size,
566 hash, true, onallcpus);
567 if (IS_ERR(l_new)) {
568 ret = PTR_ERR(l_new);
569 goto err;
570 }
571 hlist_add_head_rcu(&l_new->hash_node, head);
572 }
573 ret = 0;
574err:
575 raw_spin_unlock_irqrestore(&b->lock, flags);
576 return ret;
577}
578
579static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
580 void *value, u64 map_flags)
581{
582 return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
583}
584
585/* Called from syscall or from eBPF program */
586static int htab_map_delete_elem(struct bpf_map *map, void *key)
587{
588 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
589 struct hlist_head *head;
590 struct bucket *b;
591 struct htab_elem *l;
592 unsigned long flags;
593 u32 hash, key_size;
594 int ret = -ENOENT;
595
596 WARN_ON_ONCE(!rcu_read_lock_held());
597
598 key_size = map->key_size;
599
600 hash = htab_map_hash(key, key_size);
601 b = __select_bucket(htab, hash);
602 head = &b->head;
603
604 raw_spin_lock_irqsave(&b->lock, flags);
605
606 l = lookup_elem_raw(head, hash, key, key_size);
607
608 if (l) {
609 hlist_del_rcu(&l->hash_node);
610 free_htab_elem(htab, l);
611 ret = 0;
612 }
613
614 raw_spin_unlock_irqrestore(&b->lock, flags);
615 return ret;
616}
617
618static void delete_all_elements(struct bpf_htab *htab)
619{
620 int i;
621
622 for (i = 0; i < htab->n_buckets; i++) {
623 struct hlist_head *head = select_bucket(htab, i);
624 struct hlist_node *n;
625 struct htab_elem *l;
626
627 hlist_for_each_entry_safe(l, n, head, hash_node) {
628 hlist_del_rcu(&l->hash_node);
629 htab_elem_free(htab, l);
630 }
631 }
632}
633/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
634static void htab_map_free(struct bpf_map *map)
635{
636 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
637
638 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
639 * so the programs (can be more than one that used this map) were
640 * disconnected from events. Wait for outstanding critical sections in
641 * these programs to complete
642 */
643 synchronize_rcu();
644
645 /* some of free_htab_elem() callbacks for elements of this map may
646 * not have executed. Wait for them.
647 */
648 rcu_barrier();
649 if (htab->map.map_flags & BPF_F_NO_PREALLOC) {
650 delete_all_elements(htab);
651 } else {
652 htab_free_elems(htab);
653 pcpu_freelist_destroy(&htab->freelist);
654 }
655 kvfree(htab->buckets);
656 kfree(htab);
657}
658
659static const struct bpf_map_ops htab_ops = {
660 .map_alloc = htab_map_alloc,
661 .map_free = htab_map_free,
662 .map_get_next_key = htab_map_get_next_key,
663 .map_lookup_elem = htab_map_lookup_elem,
664 .map_update_elem = htab_map_update_elem,
665 .map_delete_elem = htab_map_delete_elem,
666};
667
668static struct bpf_map_type_list htab_type __read_mostly = {
669 .ops = &htab_ops,
670 .type = BPF_MAP_TYPE_HASH,
671};
672
673/* Called from eBPF program */
674static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
675{
676 struct htab_elem *l = __htab_map_lookup_elem(map, key);
677
678 if (l)
679 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
680 else
681 return NULL;
682}
683
684int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
685{
686 struct htab_elem *l;
687 void __percpu *pptr;
688 int ret = -ENOENT;
689 int cpu, off = 0;
690 u32 size;
691
692 /* per_cpu areas are zero-filled and bpf programs can only
693 * access 'value_size' of them, so copying rounded areas
694 * will not leak any kernel data
695 */
696 size = round_up(map->value_size, 8);
697 rcu_read_lock();
698 l = __htab_map_lookup_elem(map, key);
699 if (!l)
700 goto out;
701 pptr = htab_elem_get_ptr(l, map->key_size);
702 for_each_possible_cpu(cpu) {
703 bpf_long_memcpy(value + off,
704 per_cpu_ptr(pptr, cpu), size);
705 off += size;
706 }
707 ret = 0;
708out:
709 rcu_read_unlock();
710 return ret;
711}
712
713int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
714 u64 map_flags)
715{
716 int ret;
717
718 rcu_read_lock();
719 ret = __htab_percpu_map_update_elem(map, key, value, map_flags, true);
720 rcu_read_unlock();
721
722 return ret;
723}
724
725static const struct bpf_map_ops htab_percpu_ops = {
726 .map_alloc = htab_map_alloc,
727 .map_free = htab_map_free,
728 .map_get_next_key = htab_map_get_next_key,
729 .map_lookup_elem = htab_percpu_map_lookup_elem,
730 .map_update_elem = htab_percpu_map_update_elem,
731 .map_delete_elem = htab_map_delete_elem,
732};
733
734static struct bpf_map_type_list htab_percpu_type __read_mostly = {
735 .ops = &htab_percpu_ops,
736 .type = BPF_MAP_TYPE_PERCPU_HASH,
737};
738
739static int __init register_htab_map(void)
740{
741 bpf_register_map_type(&htab_type);
742 bpf_register_map_type(&htab_percpu_type);
743 return 0;
744}
745late_initcall(register_htab_map);