xref: /openbmc/linux/kernel/bpf/hashtab.c (revision 110e6f26)
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 
19 struct bucket {
20 	struct hlist_head head;
21 	raw_spinlock_t lock;
22 };
23 
24 struct 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 */
35 struct 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 
46 static 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 
52 static 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 
57 static 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 
62 static 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 	}
76 free_elems:
77 	vfree(htab->elems);
78 }
79 
80 static 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 
102 skip_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 
111 free_elems:
112 	htab_free_elems(htab);
113 	return err;
114 }
115 
116 /* Called from syscall */
117 static 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 
223 free_buckets:
224 	kvfree(htab->buckets);
225 free_htab:
226 	kfree(htab);
227 	return ERR_PTR(err);
228 }
229 
230 static inline u32 htab_map_hash(const void *key, u32 key_len)
231 {
232 	return jhash(key, key_len, 0);
233 }
234 
235 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
236 {
237 	return &htab->buckets[hash & (htab->n_buckets - 1)];
238 }
239 
240 static inline struct hlist_head *select_bucket(struct bpf_htab *htab, u32 hash)
241 {
242 	return &__select_bucket(htab, hash)->head;
243 }
244 
245 static 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 */
258 static 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 
279 static 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 */
290 static 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 
328 find_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 
347 static 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 
355 static 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 
371 static 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 
382 static 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 
444 static 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 */
459 static 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;
508 err:
509 	raw_spin_unlock_irqrestore(&b->lock, flags);
510 	return ret;
511 }
512 
513 static 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;
574 err:
575 	raw_spin_unlock_irqrestore(&b->lock, flags);
576 	return ret;
577 }
578 
579 static 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 */
586 static 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 
618 static 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 */
634 static 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 
659 static 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 
668 static 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 */
674 static 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 
684 int 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;
708 out:
709 	rcu_read_unlock();
710 	return ret;
711 }
712 
713 int 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 
725 static 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 
734 static struct bpf_map_type_list htab_percpu_type __read_mostly = {
735 	.ops = &htab_percpu_ops,
736 	.type = BPF_MAP_TYPE_PERCPU_HASH,
737 };
738 
739 static 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 }
745 late_initcall(register_htab_map);
746