xref: /openbmc/linux/kernel/bpf/hashtab.c (revision 71de0a05)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  * Copyright (c) 2016 Facebook
4  */
5 #include <linux/bpf.h>
6 #include <linux/btf.h>
7 #include <linux/jhash.h>
8 #include <linux/filter.h>
9 #include <linux/rculist_nulls.h>
10 #include <linux/random.h>
11 #include <uapi/linux/btf.h>
12 #include <linux/rcupdate_trace.h>
13 #include <linux/btf_ids.h>
14 #include "percpu_freelist.h"
15 #include "bpf_lru_list.h"
16 #include "map_in_map.h"
17 #include <linux/bpf_mem_alloc.h>
18 
19 #define HTAB_CREATE_FLAG_MASK						\
20 	(BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE |	\
21 	 BPF_F_ACCESS_MASK | BPF_F_ZERO_SEED)
22 
23 #define BATCH_OPS(_name)			\
24 	.map_lookup_batch =			\
25 	_name##_map_lookup_batch,		\
26 	.map_lookup_and_delete_batch =		\
27 	_name##_map_lookup_and_delete_batch,	\
28 	.map_update_batch =			\
29 	generic_map_update_batch,		\
30 	.map_delete_batch =			\
31 	generic_map_delete_batch
32 
33 /*
34  * The bucket lock has two protection scopes:
35  *
36  * 1) Serializing concurrent operations from BPF programs on different
37  *    CPUs
38  *
39  * 2) Serializing concurrent operations from BPF programs and sys_bpf()
40  *
41  * BPF programs can execute in any context including perf, kprobes and
42  * tracing. As there are almost no limits where perf, kprobes and tracing
43  * can be invoked from the lock operations need to be protected against
44  * deadlocks. Deadlocks can be caused by recursion and by an invocation in
45  * the lock held section when functions which acquire this lock are invoked
46  * from sys_bpf(). BPF recursion is prevented by incrementing the per CPU
47  * variable bpf_prog_active, which prevents BPF programs attached to perf
48  * events, kprobes and tracing to be invoked before the prior invocation
49  * from one of these contexts completed. sys_bpf() uses the same mechanism
50  * by pinning the task to the current CPU and incrementing the recursion
51  * protection across the map operation.
52  *
53  * This has subtle implications on PREEMPT_RT. PREEMPT_RT forbids certain
54  * operations like memory allocations (even with GFP_ATOMIC) from atomic
55  * contexts. This is required because even with GFP_ATOMIC the memory
56  * allocator calls into code paths which acquire locks with long held lock
57  * sections. To ensure the deterministic behaviour these locks are regular
58  * spinlocks, which are converted to 'sleepable' spinlocks on RT. The only
59  * true atomic contexts on an RT kernel are the low level hardware
60  * handling, scheduling, low level interrupt handling, NMIs etc. None of
61  * these contexts should ever do memory allocations.
62  *
63  * As regular device interrupt handlers and soft interrupts are forced into
64  * thread context, the existing code which does
65  *   spin_lock*(); alloc(GFP_ATOMIC); spin_unlock*();
66  * just works.
67  *
68  * In theory the BPF locks could be converted to regular spinlocks as well,
69  * but the bucket locks and percpu_freelist locks can be taken from
70  * arbitrary contexts (perf, kprobes, tracepoints) which are required to be
71  * atomic contexts even on RT. Before the introduction of bpf_mem_alloc,
72  * it is only safe to use raw spinlock for preallocated hash map on a RT kernel,
73  * because there is no memory allocation within the lock held sections. However
74  * after hash map was fully converted to use bpf_mem_alloc, there will be
75  * non-synchronous memory allocation for non-preallocated hash map, so it is
76  * safe to always use raw spinlock for bucket lock.
77  */
78 struct bucket {
79 	struct hlist_nulls_head head;
80 	raw_spinlock_t raw_lock;
81 };
82 
83 #define HASHTAB_MAP_LOCK_COUNT 8
84 #define HASHTAB_MAP_LOCK_MASK (HASHTAB_MAP_LOCK_COUNT - 1)
85 
86 struct bpf_htab {
87 	struct bpf_map map;
88 	struct bpf_mem_alloc ma;
89 	struct bpf_mem_alloc pcpu_ma;
90 	struct bucket *buckets;
91 	void *elems;
92 	union {
93 		struct pcpu_freelist freelist;
94 		struct bpf_lru lru;
95 	};
96 	struct htab_elem *__percpu *extra_elems;
97 	/* number of elements in non-preallocated hashtable are kept
98 	 * in either pcount or count
99 	 */
100 	struct percpu_counter pcount;
101 	atomic_t count;
102 	bool use_percpu_counter;
103 	u32 n_buckets;	/* number of hash buckets */
104 	u32 elem_size;	/* size of each element in bytes */
105 	u32 hashrnd;
106 	struct lock_class_key lockdep_key;
107 	int __percpu *map_locked[HASHTAB_MAP_LOCK_COUNT];
108 };
109 
110 /* each htab element is struct htab_elem + key + value */
111 struct htab_elem {
112 	union {
113 		struct hlist_nulls_node hash_node;
114 		struct {
115 			void *padding;
116 			union {
117 				struct pcpu_freelist_node fnode;
118 				struct htab_elem *batch_flink;
119 			};
120 		};
121 	};
122 	union {
123 		/* pointer to per-cpu pointer */
124 		void *ptr_to_pptr;
125 		struct bpf_lru_node lru_node;
126 	};
127 	u32 hash;
128 	char key[] __aligned(8);
129 };
130 
131 static inline bool htab_is_prealloc(const struct bpf_htab *htab)
132 {
133 	return !(htab->map.map_flags & BPF_F_NO_PREALLOC);
134 }
135 
136 static void htab_init_buckets(struct bpf_htab *htab)
137 {
138 	unsigned int i;
139 
140 	for (i = 0; i < htab->n_buckets; i++) {
141 		INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
142 		raw_spin_lock_init(&htab->buckets[i].raw_lock);
143 		lockdep_set_class(&htab->buckets[i].raw_lock,
144 					  &htab->lockdep_key);
145 		cond_resched();
146 	}
147 }
148 
149 static inline int htab_lock_bucket(const struct bpf_htab *htab,
150 				   struct bucket *b, u32 hash,
151 				   unsigned long *pflags)
152 {
153 	unsigned long flags;
154 
155 	hash = hash & min_t(u32, HASHTAB_MAP_LOCK_MASK, htab->n_buckets - 1);
156 
157 	preempt_disable();
158 	if (unlikely(__this_cpu_inc_return(*(htab->map_locked[hash])) != 1)) {
159 		__this_cpu_dec(*(htab->map_locked[hash]));
160 		preempt_enable();
161 		return -EBUSY;
162 	}
163 
164 	raw_spin_lock_irqsave(&b->raw_lock, flags);
165 	*pflags = flags;
166 
167 	return 0;
168 }
169 
170 static inline void htab_unlock_bucket(const struct bpf_htab *htab,
171 				      struct bucket *b, u32 hash,
172 				      unsigned long flags)
173 {
174 	hash = hash & min_t(u32, HASHTAB_MAP_LOCK_MASK, htab->n_buckets - 1);
175 	raw_spin_unlock_irqrestore(&b->raw_lock, flags);
176 	__this_cpu_dec(*(htab->map_locked[hash]));
177 	preempt_enable();
178 }
179 
180 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
181 
182 static bool htab_is_lru(const struct bpf_htab *htab)
183 {
184 	return htab->map.map_type == BPF_MAP_TYPE_LRU_HASH ||
185 		htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
186 }
187 
188 static bool htab_is_percpu(const struct bpf_htab *htab)
189 {
190 	return htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH ||
191 		htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
192 }
193 
194 static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
195 				     void __percpu *pptr)
196 {
197 	*(void __percpu **)(l->key + key_size) = pptr;
198 }
199 
200 static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
201 {
202 	return *(void __percpu **)(l->key + key_size);
203 }
204 
205 static void *fd_htab_map_get_ptr(const struct bpf_map *map, struct htab_elem *l)
206 {
207 	return *(void **)(l->key + roundup(map->key_size, 8));
208 }
209 
210 static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
211 {
212 	return (struct htab_elem *) (htab->elems + i * (u64)htab->elem_size);
213 }
214 
215 static bool htab_has_extra_elems(struct bpf_htab *htab)
216 {
217 	return !htab_is_percpu(htab) && !htab_is_lru(htab);
218 }
219 
220 static void htab_free_prealloced_timers(struct bpf_htab *htab)
221 {
222 	u32 num_entries = htab->map.max_entries;
223 	int i;
224 
225 	if (!btf_record_has_field(htab->map.record, BPF_TIMER))
226 		return;
227 	if (htab_has_extra_elems(htab))
228 		num_entries += num_possible_cpus();
229 
230 	for (i = 0; i < num_entries; i++) {
231 		struct htab_elem *elem;
232 
233 		elem = get_htab_elem(htab, i);
234 		bpf_obj_free_timer(htab->map.record, elem->key + round_up(htab->map.key_size, 8));
235 		cond_resched();
236 	}
237 }
238 
239 static void htab_free_prealloced_fields(struct bpf_htab *htab)
240 {
241 	u32 num_entries = htab->map.max_entries;
242 	int i;
243 
244 	if (IS_ERR_OR_NULL(htab->map.record))
245 		return;
246 	if (htab_has_extra_elems(htab))
247 		num_entries += num_possible_cpus();
248 	for (i = 0; i < num_entries; i++) {
249 		struct htab_elem *elem;
250 
251 		elem = get_htab_elem(htab, i);
252 		bpf_obj_free_fields(htab->map.record, elem->key + round_up(htab->map.key_size, 8));
253 		cond_resched();
254 	}
255 }
256 
257 static void htab_free_elems(struct bpf_htab *htab)
258 {
259 	int i;
260 
261 	if (!htab_is_percpu(htab))
262 		goto free_elems;
263 
264 	for (i = 0; i < htab->map.max_entries; i++) {
265 		void __percpu *pptr;
266 
267 		pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
268 					 htab->map.key_size);
269 		free_percpu(pptr);
270 		cond_resched();
271 	}
272 free_elems:
273 	bpf_map_area_free(htab->elems);
274 }
275 
276 /* The LRU list has a lock (lru_lock). Each htab bucket has a lock
277  * (bucket_lock). If both locks need to be acquired together, the lock
278  * order is always lru_lock -> bucket_lock and this only happens in
279  * bpf_lru_list.c logic. For example, certain code path of
280  * bpf_lru_pop_free(), which is called by function prealloc_lru_pop(),
281  * will acquire lru_lock first followed by acquiring bucket_lock.
282  *
283  * In hashtab.c, to avoid deadlock, lock acquisition of
284  * bucket_lock followed by lru_lock is not allowed. In such cases,
285  * bucket_lock needs to be released first before acquiring lru_lock.
286  */
287 static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
288 					  u32 hash)
289 {
290 	struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
291 	struct htab_elem *l;
292 
293 	if (node) {
294 		l = container_of(node, struct htab_elem, lru_node);
295 		memcpy(l->key, key, htab->map.key_size);
296 		return l;
297 	}
298 
299 	return NULL;
300 }
301 
302 static int prealloc_init(struct bpf_htab *htab)
303 {
304 	u32 num_entries = htab->map.max_entries;
305 	int err = -ENOMEM, i;
306 
307 	if (htab_has_extra_elems(htab))
308 		num_entries += num_possible_cpus();
309 
310 	htab->elems = bpf_map_area_alloc((u64)htab->elem_size * num_entries,
311 					 htab->map.numa_node);
312 	if (!htab->elems)
313 		return -ENOMEM;
314 
315 	if (!htab_is_percpu(htab))
316 		goto skip_percpu_elems;
317 
318 	for (i = 0; i < num_entries; i++) {
319 		u32 size = round_up(htab->map.value_size, 8);
320 		void __percpu *pptr;
321 
322 		pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
323 					    GFP_USER | __GFP_NOWARN);
324 		if (!pptr)
325 			goto free_elems;
326 		htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
327 				  pptr);
328 		cond_resched();
329 	}
330 
331 skip_percpu_elems:
332 	if (htab_is_lru(htab))
333 		err = bpf_lru_init(&htab->lru,
334 				   htab->map.map_flags & BPF_F_NO_COMMON_LRU,
335 				   offsetof(struct htab_elem, hash) -
336 				   offsetof(struct htab_elem, lru_node),
337 				   htab_lru_map_delete_node,
338 				   htab);
339 	else
340 		err = pcpu_freelist_init(&htab->freelist);
341 
342 	if (err)
343 		goto free_elems;
344 
345 	if (htab_is_lru(htab))
346 		bpf_lru_populate(&htab->lru, htab->elems,
347 				 offsetof(struct htab_elem, lru_node),
348 				 htab->elem_size, num_entries);
349 	else
350 		pcpu_freelist_populate(&htab->freelist,
351 				       htab->elems + offsetof(struct htab_elem, fnode),
352 				       htab->elem_size, num_entries);
353 
354 	return 0;
355 
356 free_elems:
357 	htab_free_elems(htab);
358 	return err;
359 }
360 
361 static void prealloc_destroy(struct bpf_htab *htab)
362 {
363 	htab_free_elems(htab);
364 
365 	if (htab_is_lru(htab))
366 		bpf_lru_destroy(&htab->lru);
367 	else
368 		pcpu_freelist_destroy(&htab->freelist);
369 }
370 
371 static int alloc_extra_elems(struct bpf_htab *htab)
372 {
373 	struct htab_elem *__percpu *pptr, *l_new;
374 	struct pcpu_freelist_node *l;
375 	int cpu;
376 
377 	pptr = bpf_map_alloc_percpu(&htab->map, sizeof(struct htab_elem *), 8,
378 				    GFP_USER | __GFP_NOWARN);
379 	if (!pptr)
380 		return -ENOMEM;
381 
382 	for_each_possible_cpu(cpu) {
383 		l = pcpu_freelist_pop(&htab->freelist);
384 		/* pop will succeed, since prealloc_init()
385 		 * preallocated extra num_possible_cpus elements
386 		 */
387 		l_new = container_of(l, struct htab_elem, fnode);
388 		*per_cpu_ptr(pptr, cpu) = l_new;
389 	}
390 	htab->extra_elems = pptr;
391 	return 0;
392 }
393 
394 /* Called from syscall */
395 static int htab_map_alloc_check(union bpf_attr *attr)
396 {
397 	bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
398 		       attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
399 	bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
400 		    attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
401 	/* percpu_lru means each cpu has its own LRU list.
402 	 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
403 	 * the map's value itself is percpu.  percpu_lru has
404 	 * nothing to do with the map's value.
405 	 */
406 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
407 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
408 	bool zero_seed = (attr->map_flags & BPF_F_ZERO_SEED);
409 	int numa_node = bpf_map_attr_numa_node(attr);
410 
411 	BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) !=
412 		     offsetof(struct htab_elem, hash_node.pprev));
413 
414 	if (lru && !bpf_capable())
415 		/* LRU implementation is much complicated than other
416 		 * maps.  Hence, limit to CAP_BPF.
417 		 */
418 		return -EPERM;
419 
420 	if (zero_seed && !capable(CAP_SYS_ADMIN))
421 		/* Guard against local DoS, and discourage production use. */
422 		return -EPERM;
423 
424 	if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK ||
425 	    !bpf_map_flags_access_ok(attr->map_flags))
426 		return -EINVAL;
427 
428 	if (!lru && percpu_lru)
429 		return -EINVAL;
430 
431 	if (lru && !prealloc)
432 		return -ENOTSUPP;
433 
434 	if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
435 		return -EINVAL;
436 
437 	/* check sanity of attributes.
438 	 * value_size == 0 may be allowed in the future to use map as a set
439 	 */
440 	if (attr->max_entries == 0 || attr->key_size == 0 ||
441 	    attr->value_size == 0)
442 		return -EINVAL;
443 
444 	if ((u64)attr->key_size + attr->value_size >= KMALLOC_MAX_SIZE -
445 	   sizeof(struct htab_elem))
446 		/* if key_size + value_size is bigger, the user space won't be
447 		 * able to access the elements via bpf syscall. This check
448 		 * also makes sure that the elem_size doesn't overflow and it's
449 		 * kmalloc-able later in htab_map_update_elem()
450 		 */
451 		return -E2BIG;
452 
453 	return 0;
454 }
455 
456 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
457 {
458 	bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
459 		       attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
460 	bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
461 		    attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
462 	/* percpu_lru means each cpu has its own LRU list.
463 	 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
464 	 * the map's value itself is percpu.  percpu_lru has
465 	 * nothing to do with the map's value.
466 	 */
467 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
468 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
469 	struct bpf_htab *htab;
470 	int err, i;
471 
472 	htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
473 	if (!htab)
474 		return ERR_PTR(-ENOMEM);
475 
476 	lockdep_register_key(&htab->lockdep_key);
477 
478 	bpf_map_init_from_attr(&htab->map, attr);
479 
480 	if (percpu_lru) {
481 		/* ensure each CPU's lru list has >=1 elements.
482 		 * since we are at it, make each lru list has the same
483 		 * number of elements.
484 		 */
485 		htab->map.max_entries = roundup(attr->max_entries,
486 						num_possible_cpus());
487 		if (htab->map.max_entries < attr->max_entries)
488 			htab->map.max_entries = rounddown(attr->max_entries,
489 							  num_possible_cpus());
490 	}
491 
492 	/* hash table size must be power of 2 */
493 	htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
494 
495 	htab->elem_size = sizeof(struct htab_elem) +
496 			  round_up(htab->map.key_size, 8);
497 	if (percpu)
498 		htab->elem_size += sizeof(void *);
499 	else
500 		htab->elem_size += round_up(htab->map.value_size, 8);
501 
502 	err = -E2BIG;
503 	/* prevent zero size kmalloc and check for u32 overflow */
504 	if (htab->n_buckets == 0 ||
505 	    htab->n_buckets > U32_MAX / sizeof(struct bucket))
506 		goto free_htab;
507 
508 	err = -ENOMEM;
509 	htab->buckets = bpf_map_area_alloc(htab->n_buckets *
510 					   sizeof(struct bucket),
511 					   htab->map.numa_node);
512 	if (!htab->buckets)
513 		goto free_htab;
514 
515 	for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++) {
516 		htab->map_locked[i] = bpf_map_alloc_percpu(&htab->map,
517 							   sizeof(int),
518 							   sizeof(int),
519 							   GFP_USER);
520 		if (!htab->map_locked[i])
521 			goto free_map_locked;
522 	}
523 
524 	if (htab->map.map_flags & BPF_F_ZERO_SEED)
525 		htab->hashrnd = 0;
526 	else
527 		htab->hashrnd = get_random_u32();
528 
529 	htab_init_buckets(htab);
530 
531 /* compute_batch_value() computes batch value as num_online_cpus() * 2
532  * and __percpu_counter_compare() needs
533  * htab->max_entries - cur_number_of_elems to be more than batch * num_online_cpus()
534  * for percpu_counter to be faster than atomic_t. In practice the average bpf
535  * hash map size is 10k, which means that a system with 64 cpus will fill
536  * hashmap to 20% of 10k before percpu_counter becomes ineffective. Therefore
537  * define our own batch count as 32 then 10k hash map can be filled up to 80%:
538  * 10k - 8k > 32 _batch_ * 64 _cpus_
539  * and __percpu_counter_compare() will still be fast. At that point hash map
540  * collisions will dominate its performance anyway. Assume that hash map filled
541  * to 50+% isn't going to be O(1) and use the following formula to choose
542  * between percpu_counter and atomic_t.
543  */
544 #define PERCPU_COUNTER_BATCH 32
545 	if (attr->max_entries / 2 > num_online_cpus() * PERCPU_COUNTER_BATCH)
546 		htab->use_percpu_counter = true;
547 
548 	if (htab->use_percpu_counter) {
549 		err = percpu_counter_init(&htab->pcount, 0, GFP_KERNEL);
550 		if (err)
551 			goto free_map_locked;
552 	}
553 
554 	if (prealloc) {
555 		err = prealloc_init(htab);
556 		if (err)
557 			goto free_map_locked;
558 
559 		if (!percpu && !lru) {
560 			/* lru itself can remove the least used element, so
561 			 * there is no need for an extra elem during map_update.
562 			 */
563 			err = alloc_extra_elems(htab);
564 			if (err)
565 				goto free_prealloc;
566 		}
567 	} else {
568 		err = bpf_mem_alloc_init(&htab->ma, htab->elem_size, false);
569 		if (err)
570 			goto free_map_locked;
571 		if (percpu) {
572 			err = bpf_mem_alloc_init(&htab->pcpu_ma,
573 						 round_up(htab->map.value_size, 8), true);
574 			if (err)
575 				goto free_map_locked;
576 		}
577 	}
578 
579 	return &htab->map;
580 
581 free_prealloc:
582 	prealloc_destroy(htab);
583 free_map_locked:
584 	if (htab->use_percpu_counter)
585 		percpu_counter_destroy(&htab->pcount);
586 	for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
587 		free_percpu(htab->map_locked[i]);
588 	bpf_map_area_free(htab->buckets);
589 	bpf_mem_alloc_destroy(&htab->pcpu_ma);
590 	bpf_mem_alloc_destroy(&htab->ma);
591 free_htab:
592 	lockdep_unregister_key(&htab->lockdep_key);
593 	bpf_map_area_free(htab);
594 	return ERR_PTR(err);
595 }
596 
597 static inline u32 htab_map_hash(const void *key, u32 key_len, u32 hashrnd)
598 {
599 	return jhash(key, key_len, hashrnd);
600 }
601 
602 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
603 {
604 	return &htab->buckets[hash & (htab->n_buckets - 1)];
605 }
606 
607 static inline struct hlist_nulls_head *select_bucket(struct bpf_htab *htab, u32 hash)
608 {
609 	return &__select_bucket(htab, hash)->head;
610 }
611 
612 /* this lookup function can only be called with bucket lock taken */
613 static struct htab_elem *lookup_elem_raw(struct hlist_nulls_head *head, u32 hash,
614 					 void *key, u32 key_size)
615 {
616 	struct hlist_nulls_node *n;
617 	struct htab_elem *l;
618 
619 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
620 		if (l->hash == hash && !memcmp(&l->key, key, key_size))
621 			return l;
622 
623 	return NULL;
624 }
625 
626 /* can be called without bucket lock. it will repeat the loop in
627  * the unlikely event when elements moved from one bucket into another
628  * while link list is being walked
629  */
630 static struct htab_elem *lookup_nulls_elem_raw(struct hlist_nulls_head *head,
631 					       u32 hash, void *key,
632 					       u32 key_size, u32 n_buckets)
633 {
634 	struct hlist_nulls_node *n;
635 	struct htab_elem *l;
636 
637 again:
638 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
639 		if (l->hash == hash && !memcmp(&l->key, key, key_size))
640 			return l;
641 
642 	if (unlikely(get_nulls_value(n) != (hash & (n_buckets - 1))))
643 		goto again;
644 
645 	return NULL;
646 }
647 
648 /* Called from syscall or from eBPF program directly, so
649  * arguments have to match bpf_map_lookup_elem() exactly.
650  * The return value is adjusted by BPF instructions
651  * in htab_map_gen_lookup().
652  */
653 static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
654 {
655 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
656 	struct hlist_nulls_head *head;
657 	struct htab_elem *l;
658 	u32 hash, key_size;
659 
660 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
661 		     !rcu_read_lock_bh_held());
662 
663 	key_size = map->key_size;
664 
665 	hash = htab_map_hash(key, key_size, htab->hashrnd);
666 
667 	head = select_bucket(htab, hash);
668 
669 	l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
670 
671 	return l;
672 }
673 
674 static void *htab_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 l->key + round_up(map->key_size, 8);
680 
681 	return NULL;
682 }
683 
684 /* inline bpf_map_lookup_elem() call.
685  * Instead of:
686  * bpf_prog
687  *   bpf_map_lookup_elem
688  *     map->ops->map_lookup_elem
689  *       htab_map_lookup_elem
690  *         __htab_map_lookup_elem
691  * do:
692  * bpf_prog
693  *   __htab_map_lookup_elem
694  */
695 static int htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
696 {
697 	struct bpf_insn *insn = insn_buf;
698 	const int ret = BPF_REG_0;
699 
700 	BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
701 		     (void *(*)(struct bpf_map *map, void *key))NULL));
702 	*insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
703 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
704 	*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
705 				offsetof(struct htab_elem, key) +
706 				round_up(map->key_size, 8));
707 	return insn - insn_buf;
708 }
709 
710 static __always_inline void *__htab_lru_map_lookup_elem(struct bpf_map *map,
711 							void *key, const bool mark)
712 {
713 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
714 
715 	if (l) {
716 		if (mark)
717 			bpf_lru_node_set_ref(&l->lru_node);
718 		return l->key + round_up(map->key_size, 8);
719 	}
720 
721 	return NULL;
722 }
723 
724 static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
725 {
726 	return __htab_lru_map_lookup_elem(map, key, true);
727 }
728 
729 static void *htab_lru_map_lookup_elem_sys(struct bpf_map *map, void *key)
730 {
731 	return __htab_lru_map_lookup_elem(map, key, false);
732 }
733 
734 static int htab_lru_map_gen_lookup(struct bpf_map *map,
735 				   struct bpf_insn *insn_buf)
736 {
737 	struct bpf_insn *insn = insn_buf;
738 	const int ret = BPF_REG_0;
739 	const int ref_reg = BPF_REG_1;
740 
741 	BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
742 		     (void *(*)(struct bpf_map *map, void *key))NULL));
743 	*insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
744 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 4);
745 	*insn++ = BPF_LDX_MEM(BPF_B, ref_reg, ret,
746 			      offsetof(struct htab_elem, lru_node) +
747 			      offsetof(struct bpf_lru_node, ref));
748 	*insn++ = BPF_JMP_IMM(BPF_JNE, ref_reg, 0, 1);
749 	*insn++ = BPF_ST_MEM(BPF_B, ret,
750 			     offsetof(struct htab_elem, lru_node) +
751 			     offsetof(struct bpf_lru_node, ref),
752 			     1);
753 	*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
754 				offsetof(struct htab_elem, key) +
755 				round_up(map->key_size, 8));
756 	return insn - insn_buf;
757 }
758 
759 static void check_and_free_fields(struct bpf_htab *htab,
760 				  struct htab_elem *elem)
761 {
762 	void *map_value = elem->key + round_up(htab->map.key_size, 8);
763 
764 	bpf_obj_free_fields(htab->map.record, map_value);
765 }
766 
767 /* It is called from the bpf_lru_list when the LRU needs to delete
768  * older elements from the htab.
769  */
770 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
771 {
772 	struct bpf_htab *htab = arg;
773 	struct htab_elem *l = NULL, *tgt_l;
774 	struct hlist_nulls_head *head;
775 	struct hlist_nulls_node *n;
776 	unsigned long flags;
777 	struct bucket *b;
778 	int ret;
779 
780 	tgt_l = container_of(node, struct htab_elem, lru_node);
781 	b = __select_bucket(htab, tgt_l->hash);
782 	head = &b->head;
783 
784 	ret = htab_lock_bucket(htab, b, tgt_l->hash, &flags);
785 	if (ret)
786 		return false;
787 
788 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
789 		if (l == tgt_l) {
790 			hlist_nulls_del_rcu(&l->hash_node);
791 			check_and_free_fields(htab, l);
792 			break;
793 		}
794 
795 	htab_unlock_bucket(htab, b, tgt_l->hash, flags);
796 
797 	return l == tgt_l;
798 }
799 
800 /* Called from syscall */
801 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
802 {
803 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
804 	struct hlist_nulls_head *head;
805 	struct htab_elem *l, *next_l;
806 	u32 hash, key_size;
807 	int i = 0;
808 
809 	WARN_ON_ONCE(!rcu_read_lock_held());
810 
811 	key_size = map->key_size;
812 
813 	if (!key)
814 		goto find_first_elem;
815 
816 	hash = htab_map_hash(key, key_size, htab->hashrnd);
817 
818 	head = select_bucket(htab, hash);
819 
820 	/* lookup the key */
821 	l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
822 
823 	if (!l)
824 		goto find_first_elem;
825 
826 	/* key was found, get next key in the same bucket */
827 	next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l->hash_node)),
828 				  struct htab_elem, hash_node);
829 
830 	if (next_l) {
831 		/* if next elem in this hash list is non-zero, just return it */
832 		memcpy(next_key, next_l->key, key_size);
833 		return 0;
834 	}
835 
836 	/* no more elements in this hash list, go to the next bucket */
837 	i = hash & (htab->n_buckets - 1);
838 	i++;
839 
840 find_first_elem:
841 	/* iterate over buckets */
842 	for (; i < htab->n_buckets; i++) {
843 		head = select_bucket(htab, i);
844 
845 		/* pick first element in the bucket */
846 		next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head)),
847 					  struct htab_elem, hash_node);
848 		if (next_l) {
849 			/* if it's not empty, just return it */
850 			memcpy(next_key, next_l->key, key_size);
851 			return 0;
852 		}
853 	}
854 
855 	/* iterated over all buckets and all elements */
856 	return -ENOENT;
857 }
858 
859 static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
860 {
861 	if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
862 		bpf_mem_cache_free(&htab->pcpu_ma, l->ptr_to_pptr);
863 	check_and_free_fields(htab, l);
864 	bpf_mem_cache_free(&htab->ma, l);
865 }
866 
867 static void htab_put_fd_value(struct bpf_htab *htab, struct htab_elem *l)
868 {
869 	struct bpf_map *map = &htab->map;
870 	void *ptr;
871 
872 	if (map->ops->map_fd_put_ptr) {
873 		ptr = fd_htab_map_get_ptr(map, l);
874 		map->ops->map_fd_put_ptr(ptr);
875 	}
876 }
877 
878 static bool is_map_full(struct bpf_htab *htab)
879 {
880 	if (htab->use_percpu_counter)
881 		return __percpu_counter_compare(&htab->pcount, htab->map.max_entries,
882 						PERCPU_COUNTER_BATCH) >= 0;
883 	return atomic_read(&htab->count) >= htab->map.max_entries;
884 }
885 
886 static void inc_elem_count(struct bpf_htab *htab)
887 {
888 	if (htab->use_percpu_counter)
889 		percpu_counter_add_batch(&htab->pcount, 1, PERCPU_COUNTER_BATCH);
890 	else
891 		atomic_inc(&htab->count);
892 }
893 
894 static void dec_elem_count(struct bpf_htab *htab)
895 {
896 	if (htab->use_percpu_counter)
897 		percpu_counter_add_batch(&htab->pcount, -1, PERCPU_COUNTER_BATCH);
898 	else
899 		atomic_dec(&htab->count);
900 }
901 
902 
903 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
904 {
905 	htab_put_fd_value(htab, l);
906 
907 	if (htab_is_prealloc(htab)) {
908 		check_and_free_fields(htab, l);
909 		__pcpu_freelist_push(&htab->freelist, &l->fnode);
910 	} else {
911 		dec_elem_count(htab);
912 		htab_elem_free(htab, l);
913 	}
914 }
915 
916 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
917 			    void *value, bool onallcpus)
918 {
919 	if (!onallcpus) {
920 		/* copy true value_size bytes */
921 		memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
922 	} else {
923 		u32 size = round_up(htab->map.value_size, 8);
924 		int off = 0, cpu;
925 
926 		for_each_possible_cpu(cpu) {
927 			bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
928 					value + off, size);
929 			off += size;
930 		}
931 	}
932 }
933 
934 static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
935 			    void *value, bool onallcpus)
936 {
937 	/* When not setting the initial value on all cpus, zero-fill element
938 	 * values for other cpus. Otherwise, bpf program has no way to ensure
939 	 * known initial values for cpus other than current one
940 	 * (onallcpus=false always when coming from bpf prog).
941 	 */
942 	if (!onallcpus) {
943 		u32 size = round_up(htab->map.value_size, 8);
944 		int current_cpu = raw_smp_processor_id();
945 		int cpu;
946 
947 		for_each_possible_cpu(cpu) {
948 			if (cpu == current_cpu)
949 				bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value,
950 						size);
951 			else
952 				memset(per_cpu_ptr(pptr, cpu), 0, size);
953 		}
954 	} else {
955 		pcpu_copy_value(htab, pptr, value, onallcpus);
956 	}
957 }
958 
959 static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
960 {
961 	return htab->map.map_type == BPF_MAP_TYPE_HASH_OF_MAPS &&
962 	       BITS_PER_LONG == 64;
963 }
964 
965 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
966 					 void *value, u32 key_size, u32 hash,
967 					 bool percpu, bool onallcpus,
968 					 struct htab_elem *old_elem)
969 {
970 	u32 size = htab->map.value_size;
971 	bool prealloc = htab_is_prealloc(htab);
972 	struct htab_elem *l_new, **pl_new;
973 	void __percpu *pptr;
974 
975 	if (prealloc) {
976 		if (old_elem) {
977 			/* if we're updating the existing element,
978 			 * use per-cpu extra elems to avoid freelist_pop/push
979 			 */
980 			pl_new = this_cpu_ptr(htab->extra_elems);
981 			l_new = *pl_new;
982 			htab_put_fd_value(htab, old_elem);
983 			*pl_new = old_elem;
984 		} else {
985 			struct pcpu_freelist_node *l;
986 
987 			l = __pcpu_freelist_pop(&htab->freelist);
988 			if (!l)
989 				return ERR_PTR(-E2BIG);
990 			l_new = container_of(l, struct htab_elem, fnode);
991 		}
992 	} else {
993 		if (is_map_full(htab))
994 			if (!old_elem)
995 				/* when map is full and update() is replacing
996 				 * old element, it's ok to allocate, since
997 				 * old element will be freed immediately.
998 				 * Otherwise return an error
999 				 */
1000 				return ERR_PTR(-E2BIG);
1001 		inc_elem_count(htab);
1002 		l_new = bpf_mem_cache_alloc(&htab->ma);
1003 		if (!l_new) {
1004 			l_new = ERR_PTR(-ENOMEM);
1005 			goto dec_count;
1006 		}
1007 	}
1008 
1009 	memcpy(l_new->key, key, key_size);
1010 	if (percpu) {
1011 		if (prealloc) {
1012 			pptr = htab_elem_get_ptr(l_new, key_size);
1013 		} else {
1014 			/* alloc_percpu zero-fills */
1015 			pptr = bpf_mem_cache_alloc(&htab->pcpu_ma);
1016 			if (!pptr) {
1017 				bpf_mem_cache_free(&htab->ma, l_new);
1018 				l_new = ERR_PTR(-ENOMEM);
1019 				goto dec_count;
1020 			}
1021 			l_new->ptr_to_pptr = pptr;
1022 			pptr = *(void **)pptr;
1023 		}
1024 
1025 		pcpu_init_value(htab, pptr, value, onallcpus);
1026 
1027 		if (!prealloc)
1028 			htab_elem_set_ptr(l_new, key_size, pptr);
1029 	} else if (fd_htab_map_needs_adjust(htab)) {
1030 		size = round_up(size, 8);
1031 		memcpy(l_new->key + round_up(key_size, 8), value, size);
1032 	} else {
1033 		copy_map_value(&htab->map,
1034 			       l_new->key + round_up(key_size, 8),
1035 			       value);
1036 	}
1037 
1038 	l_new->hash = hash;
1039 	return l_new;
1040 dec_count:
1041 	dec_elem_count(htab);
1042 	return l_new;
1043 }
1044 
1045 static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
1046 		       u64 map_flags)
1047 {
1048 	if (l_old && (map_flags & ~BPF_F_LOCK) == BPF_NOEXIST)
1049 		/* elem already exists */
1050 		return -EEXIST;
1051 
1052 	if (!l_old && (map_flags & ~BPF_F_LOCK) == BPF_EXIST)
1053 		/* elem doesn't exist, cannot update it */
1054 		return -ENOENT;
1055 
1056 	return 0;
1057 }
1058 
1059 /* Called from syscall or from eBPF program */
1060 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
1061 				u64 map_flags)
1062 {
1063 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1064 	struct htab_elem *l_new = NULL, *l_old;
1065 	struct hlist_nulls_head *head;
1066 	unsigned long flags;
1067 	struct bucket *b;
1068 	u32 key_size, hash;
1069 	int ret;
1070 
1071 	if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
1072 		/* unknown flags */
1073 		return -EINVAL;
1074 
1075 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1076 		     !rcu_read_lock_bh_held());
1077 
1078 	key_size = map->key_size;
1079 
1080 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1081 
1082 	b = __select_bucket(htab, hash);
1083 	head = &b->head;
1084 
1085 	if (unlikely(map_flags & BPF_F_LOCK)) {
1086 		if (unlikely(!btf_record_has_field(map->record, BPF_SPIN_LOCK)))
1087 			return -EINVAL;
1088 		/* find an element without taking the bucket lock */
1089 		l_old = lookup_nulls_elem_raw(head, hash, key, key_size,
1090 					      htab->n_buckets);
1091 		ret = check_flags(htab, l_old, map_flags);
1092 		if (ret)
1093 			return ret;
1094 		if (l_old) {
1095 			/* grab the element lock and update value in place */
1096 			copy_map_value_locked(map,
1097 					      l_old->key + round_up(key_size, 8),
1098 					      value, false);
1099 			return 0;
1100 		}
1101 		/* fall through, grab the bucket lock and lookup again.
1102 		 * 99.9% chance that the element won't be found,
1103 		 * but second lookup under lock has to be done.
1104 		 */
1105 	}
1106 
1107 	ret = htab_lock_bucket(htab, b, hash, &flags);
1108 	if (ret)
1109 		return ret;
1110 
1111 	l_old = lookup_elem_raw(head, hash, key, key_size);
1112 
1113 	ret = check_flags(htab, l_old, map_flags);
1114 	if (ret)
1115 		goto err;
1116 
1117 	if (unlikely(l_old && (map_flags & BPF_F_LOCK))) {
1118 		/* first lookup without the bucket lock didn't find the element,
1119 		 * but second lookup with the bucket lock found it.
1120 		 * This case is highly unlikely, but has to be dealt with:
1121 		 * grab the element lock in addition to the bucket lock
1122 		 * and update element in place
1123 		 */
1124 		copy_map_value_locked(map,
1125 				      l_old->key + round_up(key_size, 8),
1126 				      value, false);
1127 		ret = 0;
1128 		goto err;
1129 	}
1130 
1131 	l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
1132 				l_old);
1133 	if (IS_ERR(l_new)) {
1134 		/* all pre-allocated elements are in use or memory exhausted */
1135 		ret = PTR_ERR(l_new);
1136 		goto err;
1137 	}
1138 
1139 	/* add new element to the head of the list, so that
1140 	 * concurrent search will find it before old elem
1141 	 */
1142 	hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1143 	if (l_old) {
1144 		hlist_nulls_del_rcu(&l_old->hash_node);
1145 		if (!htab_is_prealloc(htab))
1146 			free_htab_elem(htab, l_old);
1147 		else
1148 			check_and_free_fields(htab, l_old);
1149 	}
1150 	ret = 0;
1151 err:
1152 	htab_unlock_bucket(htab, b, hash, flags);
1153 	return ret;
1154 }
1155 
1156 static void htab_lru_push_free(struct bpf_htab *htab, struct htab_elem *elem)
1157 {
1158 	check_and_free_fields(htab, elem);
1159 	bpf_lru_push_free(&htab->lru, &elem->lru_node);
1160 }
1161 
1162 static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
1163 				    u64 map_flags)
1164 {
1165 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1166 	struct htab_elem *l_new, *l_old = NULL;
1167 	struct hlist_nulls_head *head;
1168 	unsigned long flags;
1169 	struct bucket *b;
1170 	u32 key_size, hash;
1171 	int ret;
1172 
1173 	if (unlikely(map_flags > BPF_EXIST))
1174 		/* unknown flags */
1175 		return -EINVAL;
1176 
1177 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1178 		     !rcu_read_lock_bh_held());
1179 
1180 	key_size = map->key_size;
1181 
1182 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1183 
1184 	b = __select_bucket(htab, hash);
1185 	head = &b->head;
1186 
1187 	/* For LRU, we need to alloc before taking bucket's
1188 	 * spinlock because getting free nodes from LRU may need
1189 	 * to remove older elements from htab and this removal
1190 	 * operation will need a bucket lock.
1191 	 */
1192 	l_new = prealloc_lru_pop(htab, key, hash);
1193 	if (!l_new)
1194 		return -ENOMEM;
1195 	copy_map_value(&htab->map,
1196 		       l_new->key + round_up(map->key_size, 8), value);
1197 
1198 	ret = htab_lock_bucket(htab, b, hash, &flags);
1199 	if (ret)
1200 		return ret;
1201 
1202 	l_old = lookup_elem_raw(head, hash, key, key_size);
1203 
1204 	ret = check_flags(htab, l_old, map_flags);
1205 	if (ret)
1206 		goto err;
1207 
1208 	/* add new element to the head of the list, so that
1209 	 * concurrent search will find it before old elem
1210 	 */
1211 	hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1212 	if (l_old) {
1213 		bpf_lru_node_set_ref(&l_new->lru_node);
1214 		hlist_nulls_del_rcu(&l_old->hash_node);
1215 	}
1216 	ret = 0;
1217 
1218 err:
1219 	htab_unlock_bucket(htab, b, hash, flags);
1220 
1221 	if (ret)
1222 		htab_lru_push_free(htab, l_new);
1223 	else if (l_old)
1224 		htab_lru_push_free(htab, l_old);
1225 
1226 	return ret;
1227 }
1228 
1229 static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1230 					 void *value, u64 map_flags,
1231 					 bool onallcpus)
1232 {
1233 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1234 	struct htab_elem *l_new = NULL, *l_old;
1235 	struct hlist_nulls_head *head;
1236 	unsigned long flags;
1237 	struct bucket *b;
1238 	u32 key_size, hash;
1239 	int ret;
1240 
1241 	if (unlikely(map_flags > BPF_EXIST))
1242 		/* unknown flags */
1243 		return -EINVAL;
1244 
1245 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1246 		     !rcu_read_lock_bh_held());
1247 
1248 	key_size = map->key_size;
1249 
1250 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1251 
1252 	b = __select_bucket(htab, hash);
1253 	head = &b->head;
1254 
1255 	ret = htab_lock_bucket(htab, b, hash, &flags);
1256 	if (ret)
1257 		return ret;
1258 
1259 	l_old = lookup_elem_raw(head, hash, key, key_size);
1260 
1261 	ret = check_flags(htab, l_old, map_flags);
1262 	if (ret)
1263 		goto err;
1264 
1265 	if (l_old) {
1266 		/* per-cpu hash map can update value in-place */
1267 		pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1268 				value, onallcpus);
1269 	} else {
1270 		l_new = alloc_htab_elem(htab, key, value, key_size,
1271 					hash, true, onallcpus, NULL);
1272 		if (IS_ERR(l_new)) {
1273 			ret = PTR_ERR(l_new);
1274 			goto err;
1275 		}
1276 		hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1277 	}
1278 	ret = 0;
1279 err:
1280 	htab_unlock_bucket(htab, b, hash, flags);
1281 	return ret;
1282 }
1283 
1284 static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1285 					     void *value, u64 map_flags,
1286 					     bool onallcpus)
1287 {
1288 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1289 	struct htab_elem *l_new = NULL, *l_old;
1290 	struct hlist_nulls_head *head;
1291 	unsigned long flags;
1292 	struct bucket *b;
1293 	u32 key_size, hash;
1294 	int ret;
1295 
1296 	if (unlikely(map_flags > BPF_EXIST))
1297 		/* unknown flags */
1298 		return -EINVAL;
1299 
1300 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1301 		     !rcu_read_lock_bh_held());
1302 
1303 	key_size = map->key_size;
1304 
1305 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1306 
1307 	b = __select_bucket(htab, hash);
1308 	head = &b->head;
1309 
1310 	/* For LRU, we need to alloc before taking bucket's
1311 	 * spinlock because LRU's elem alloc may need
1312 	 * to remove older elem from htab and this removal
1313 	 * operation will need a bucket lock.
1314 	 */
1315 	if (map_flags != BPF_EXIST) {
1316 		l_new = prealloc_lru_pop(htab, key, hash);
1317 		if (!l_new)
1318 			return -ENOMEM;
1319 	}
1320 
1321 	ret = htab_lock_bucket(htab, b, hash, &flags);
1322 	if (ret)
1323 		return ret;
1324 
1325 	l_old = lookup_elem_raw(head, hash, key, key_size);
1326 
1327 	ret = check_flags(htab, l_old, map_flags);
1328 	if (ret)
1329 		goto err;
1330 
1331 	if (l_old) {
1332 		bpf_lru_node_set_ref(&l_old->lru_node);
1333 
1334 		/* per-cpu hash map can update value in-place */
1335 		pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1336 				value, onallcpus);
1337 	} else {
1338 		pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
1339 				value, onallcpus);
1340 		hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1341 		l_new = NULL;
1342 	}
1343 	ret = 0;
1344 err:
1345 	htab_unlock_bucket(htab, b, hash, flags);
1346 	if (l_new)
1347 		bpf_lru_push_free(&htab->lru, &l_new->lru_node);
1348 	return ret;
1349 }
1350 
1351 static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1352 				       void *value, u64 map_flags)
1353 {
1354 	return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
1355 }
1356 
1357 static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1358 					   void *value, u64 map_flags)
1359 {
1360 	return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
1361 						 false);
1362 }
1363 
1364 /* Called from syscall or from eBPF program */
1365 static int htab_map_delete_elem(struct bpf_map *map, void *key)
1366 {
1367 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1368 	struct hlist_nulls_head *head;
1369 	struct bucket *b;
1370 	struct htab_elem *l;
1371 	unsigned long flags;
1372 	u32 hash, key_size;
1373 	int ret;
1374 
1375 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1376 		     !rcu_read_lock_bh_held());
1377 
1378 	key_size = map->key_size;
1379 
1380 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1381 	b = __select_bucket(htab, hash);
1382 	head = &b->head;
1383 
1384 	ret = htab_lock_bucket(htab, b, hash, &flags);
1385 	if (ret)
1386 		return ret;
1387 
1388 	l = lookup_elem_raw(head, hash, key, key_size);
1389 
1390 	if (l) {
1391 		hlist_nulls_del_rcu(&l->hash_node);
1392 		free_htab_elem(htab, l);
1393 	} else {
1394 		ret = -ENOENT;
1395 	}
1396 
1397 	htab_unlock_bucket(htab, b, hash, flags);
1398 	return ret;
1399 }
1400 
1401 static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
1402 {
1403 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1404 	struct hlist_nulls_head *head;
1405 	struct bucket *b;
1406 	struct htab_elem *l;
1407 	unsigned long flags;
1408 	u32 hash, key_size;
1409 	int ret;
1410 
1411 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1412 		     !rcu_read_lock_bh_held());
1413 
1414 	key_size = map->key_size;
1415 
1416 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1417 	b = __select_bucket(htab, hash);
1418 	head = &b->head;
1419 
1420 	ret = htab_lock_bucket(htab, b, hash, &flags);
1421 	if (ret)
1422 		return ret;
1423 
1424 	l = lookup_elem_raw(head, hash, key, key_size);
1425 
1426 	if (l)
1427 		hlist_nulls_del_rcu(&l->hash_node);
1428 	else
1429 		ret = -ENOENT;
1430 
1431 	htab_unlock_bucket(htab, b, hash, flags);
1432 	if (l)
1433 		htab_lru_push_free(htab, l);
1434 	return ret;
1435 }
1436 
1437 static void delete_all_elements(struct bpf_htab *htab)
1438 {
1439 	int i;
1440 
1441 	/* It's called from a worker thread, so disable migration here,
1442 	 * since bpf_mem_cache_free() relies on that.
1443 	 */
1444 	migrate_disable();
1445 	for (i = 0; i < htab->n_buckets; i++) {
1446 		struct hlist_nulls_head *head = select_bucket(htab, i);
1447 		struct hlist_nulls_node *n;
1448 		struct htab_elem *l;
1449 
1450 		hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1451 			hlist_nulls_del_rcu(&l->hash_node);
1452 			htab_elem_free(htab, l);
1453 		}
1454 	}
1455 	migrate_enable();
1456 }
1457 
1458 static void htab_free_malloced_timers(struct bpf_htab *htab)
1459 {
1460 	int i;
1461 
1462 	rcu_read_lock();
1463 	for (i = 0; i < htab->n_buckets; i++) {
1464 		struct hlist_nulls_head *head = select_bucket(htab, i);
1465 		struct hlist_nulls_node *n;
1466 		struct htab_elem *l;
1467 
1468 		hlist_nulls_for_each_entry(l, n, head, hash_node) {
1469 			/* We only free timer on uref dropping to zero */
1470 			bpf_obj_free_timer(htab->map.record, l->key + round_up(htab->map.key_size, 8));
1471 		}
1472 		cond_resched_rcu();
1473 	}
1474 	rcu_read_unlock();
1475 }
1476 
1477 static void htab_map_free_timers(struct bpf_map *map)
1478 {
1479 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1480 
1481 	/* We only free timer on uref dropping to zero */
1482 	if (!btf_record_has_field(htab->map.record, BPF_TIMER))
1483 		return;
1484 	if (!htab_is_prealloc(htab))
1485 		htab_free_malloced_timers(htab);
1486 	else
1487 		htab_free_prealloced_timers(htab);
1488 }
1489 
1490 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
1491 static void htab_map_free(struct bpf_map *map)
1492 {
1493 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1494 	int i;
1495 
1496 	/* bpf_free_used_maps() or close(map_fd) will trigger this map_free callback.
1497 	 * bpf_free_used_maps() is called after bpf prog is no longer executing.
1498 	 * There is no need to synchronize_rcu() here to protect map elements.
1499 	 */
1500 
1501 	/* htab no longer uses call_rcu() directly. bpf_mem_alloc does it
1502 	 * underneath and is reponsible for waiting for callbacks to finish
1503 	 * during bpf_mem_alloc_destroy().
1504 	 */
1505 	if (!htab_is_prealloc(htab)) {
1506 		delete_all_elements(htab);
1507 	} else {
1508 		htab_free_prealloced_fields(htab);
1509 		prealloc_destroy(htab);
1510 	}
1511 
1512 	free_percpu(htab->extra_elems);
1513 	bpf_map_area_free(htab->buckets);
1514 	bpf_mem_alloc_destroy(&htab->pcpu_ma);
1515 	bpf_mem_alloc_destroy(&htab->ma);
1516 	if (htab->use_percpu_counter)
1517 		percpu_counter_destroy(&htab->pcount);
1518 	for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
1519 		free_percpu(htab->map_locked[i]);
1520 	lockdep_unregister_key(&htab->lockdep_key);
1521 	bpf_map_area_free(htab);
1522 }
1523 
1524 static void htab_map_seq_show_elem(struct bpf_map *map, void *key,
1525 				   struct seq_file *m)
1526 {
1527 	void *value;
1528 
1529 	rcu_read_lock();
1530 
1531 	value = htab_map_lookup_elem(map, key);
1532 	if (!value) {
1533 		rcu_read_unlock();
1534 		return;
1535 	}
1536 
1537 	btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
1538 	seq_puts(m, ": ");
1539 	btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
1540 	seq_puts(m, "\n");
1541 
1542 	rcu_read_unlock();
1543 }
1544 
1545 static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1546 					     void *value, bool is_lru_map,
1547 					     bool is_percpu, u64 flags)
1548 {
1549 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1550 	struct hlist_nulls_head *head;
1551 	unsigned long bflags;
1552 	struct htab_elem *l;
1553 	u32 hash, key_size;
1554 	struct bucket *b;
1555 	int ret;
1556 
1557 	key_size = map->key_size;
1558 
1559 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1560 	b = __select_bucket(htab, hash);
1561 	head = &b->head;
1562 
1563 	ret = htab_lock_bucket(htab, b, hash, &bflags);
1564 	if (ret)
1565 		return ret;
1566 
1567 	l = lookup_elem_raw(head, hash, key, key_size);
1568 	if (!l) {
1569 		ret = -ENOENT;
1570 	} else {
1571 		if (is_percpu) {
1572 			u32 roundup_value_size = round_up(map->value_size, 8);
1573 			void __percpu *pptr;
1574 			int off = 0, cpu;
1575 
1576 			pptr = htab_elem_get_ptr(l, key_size);
1577 			for_each_possible_cpu(cpu) {
1578 				bpf_long_memcpy(value + off,
1579 						per_cpu_ptr(pptr, cpu),
1580 						roundup_value_size);
1581 				off += roundup_value_size;
1582 			}
1583 		} else {
1584 			u32 roundup_key_size = round_up(map->key_size, 8);
1585 
1586 			if (flags & BPF_F_LOCK)
1587 				copy_map_value_locked(map, value, l->key +
1588 						      roundup_key_size,
1589 						      true);
1590 			else
1591 				copy_map_value(map, value, l->key +
1592 					       roundup_key_size);
1593 			/* Zeroing special fields in the temp buffer */
1594 			check_and_init_map_value(map, value);
1595 		}
1596 
1597 		hlist_nulls_del_rcu(&l->hash_node);
1598 		if (!is_lru_map)
1599 			free_htab_elem(htab, l);
1600 	}
1601 
1602 	htab_unlock_bucket(htab, b, hash, bflags);
1603 
1604 	if (is_lru_map && l)
1605 		htab_lru_push_free(htab, l);
1606 
1607 	return ret;
1608 }
1609 
1610 static int htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1611 					   void *value, u64 flags)
1612 {
1613 	return __htab_map_lookup_and_delete_elem(map, key, value, false, false,
1614 						 flags);
1615 }
1616 
1617 static int htab_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1618 						  void *key, void *value,
1619 						  u64 flags)
1620 {
1621 	return __htab_map_lookup_and_delete_elem(map, key, value, false, true,
1622 						 flags);
1623 }
1624 
1625 static int htab_lru_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1626 					       void *value, u64 flags)
1627 {
1628 	return __htab_map_lookup_and_delete_elem(map, key, value, true, false,
1629 						 flags);
1630 }
1631 
1632 static int htab_lru_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1633 						      void *key, void *value,
1634 						      u64 flags)
1635 {
1636 	return __htab_map_lookup_and_delete_elem(map, key, value, true, true,
1637 						 flags);
1638 }
1639 
1640 static int
1641 __htab_map_lookup_and_delete_batch(struct bpf_map *map,
1642 				   const union bpf_attr *attr,
1643 				   union bpf_attr __user *uattr,
1644 				   bool do_delete, bool is_lru_map,
1645 				   bool is_percpu)
1646 {
1647 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1648 	u32 bucket_cnt, total, key_size, value_size, roundup_key_size;
1649 	void *keys = NULL, *values = NULL, *value, *dst_key, *dst_val;
1650 	void __user *uvalues = u64_to_user_ptr(attr->batch.values);
1651 	void __user *ukeys = u64_to_user_ptr(attr->batch.keys);
1652 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1653 	u32 batch, max_count, size, bucket_size, map_id;
1654 	struct htab_elem *node_to_free = NULL;
1655 	u64 elem_map_flags, map_flags;
1656 	struct hlist_nulls_head *head;
1657 	struct hlist_nulls_node *n;
1658 	unsigned long flags = 0;
1659 	bool locked = false;
1660 	struct htab_elem *l;
1661 	struct bucket *b;
1662 	int ret = 0;
1663 
1664 	elem_map_flags = attr->batch.elem_flags;
1665 	if ((elem_map_flags & ~BPF_F_LOCK) ||
1666 	    ((elem_map_flags & BPF_F_LOCK) && !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
1667 		return -EINVAL;
1668 
1669 	map_flags = attr->batch.flags;
1670 	if (map_flags)
1671 		return -EINVAL;
1672 
1673 	max_count = attr->batch.count;
1674 	if (!max_count)
1675 		return 0;
1676 
1677 	if (put_user(0, &uattr->batch.count))
1678 		return -EFAULT;
1679 
1680 	batch = 0;
1681 	if (ubatch && copy_from_user(&batch, ubatch, sizeof(batch)))
1682 		return -EFAULT;
1683 
1684 	if (batch >= htab->n_buckets)
1685 		return -ENOENT;
1686 
1687 	key_size = htab->map.key_size;
1688 	roundup_key_size = round_up(htab->map.key_size, 8);
1689 	value_size = htab->map.value_size;
1690 	size = round_up(value_size, 8);
1691 	if (is_percpu)
1692 		value_size = size * num_possible_cpus();
1693 	total = 0;
1694 	/* while experimenting with hash tables with sizes ranging from 10 to
1695 	 * 1000, it was observed that a bucket can have up to 5 entries.
1696 	 */
1697 	bucket_size = 5;
1698 
1699 alloc:
1700 	/* We cannot do copy_from_user or copy_to_user inside
1701 	 * the rcu_read_lock. Allocate enough space here.
1702 	 */
1703 	keys = kvmalloc_array(key_size, bucket_size, GFP_USER | __GFP_NOWARN);
1704 	values = kvmalloc_array(value_size, bucket_size, GFP_USER | __GFP_NOWARN);
1705 	if (!keys || !values) {
1706 		ret = -ENOMEM;
1707 		goto after_loop;
1708 	}
1709 
1710 again:
1711 	bpf_disable_instrumentation();
1712 	rcu_read_lock();
1713 again_nocopy:
1714 	dst_key = keys;
1715 	dst_val = values;
1716 	b = &htab->buckets[batch];
1717 	head = &b->head;
1718 	/* do not grab the lock unless need it (bucket_cnt > 0). */
1719 	if (locked) {
1720 		ret = htab_lock_bucket(htab, b, batch, &flags);
1721 		if (ret) {
1722 			rcu_read_unlock();
1723 			bpf_enable_instrumentation();
1724 			goto after_loop;
1725 		}
1726 	}
1727 
1728 	bucket_cnt = 0;
1729 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
1730 		bucket_cnt++;
1731 
1732 	if (bucket_cnt && !locked) {
1733 		locked = true;
1734 		goto again_nocopy;
1735 	}
1736 
1737 	if (bucket_cnt > (max_count - total)) {
1738 		if (total == 0)
1739 			ret = -ENOSPC;
1740 		/* Note that since bucket_cnt > 0 here, it is implicit
1741 		 * that the locked was grabbed, so release it.
1742 		 */
1743 		htab_unlock_bucket(htab, b, batch, flags);
1744 		rcu_read_unlock();
1745 		bpf_enable_instrumentation();
1746 		goto after_loop;
1747 	}
1748 
1749 	if (bucket_cnt > bucket_size) {
1750 		bucket_size = bucket_cnt;
1751 		/* Note that since bucket_cnt > 0 here, it is implicit
1752 		 * that the locked was grabbed, so release it.
1753 		 */
1754 		htab_unlock_bucket(htab, b, batch, flags);
1755 		rcu_read_unlock();
1756 		bpf_enable_instrumentation();
1757 		kvfree(keys);
1758 		kvfree(values);
1759 		goto alloc;
1760 	}
1761 
1762 	/* Next block is only safe to run if you have grabbed the lock */
1763 	if (!locked)
1764 		goto next_batch;
1765 
1766 	hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1767 		memcpy(dst_key, l->key, key_size);
1768 
1769 		if (is_percpu) {
1770 			int off = 0, cpu;
1771 			void __percpu *pptr;
1772 
1773 			pptr = htab_elem_get_ptr(l, map->key_size);
1774 			for_each_possible_cpu(cpu) {
1775 				bpf_long_memcpy(dst_val + off,
1776 						per_cpu_ptr(pptr, cpu), size);
1777 				off += size;
1778 			}
1779 		} else {
1780 			value = l->key + roundup_key_size;
1781 			if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1782 				struct bpf_map **inner_map = value;
1783 
1784 				 /* Actual value is the id of the inner map */
1785 				map_id = map->ops->map_fd_sys_lookup_elem(*inner_map);
1786 				value = &map_id;
1787 			}
1788 
1789 			if (elem_map_flags & BPF_F_LOCK)
1790 				copy_map_value_locked(map, dst_val, value,
1791 						      true);
1792 			else
1793 				copy_map_value(map, dst_val, value);
1794 			/* Zeroing special fields in the temp buffer */
1795 			check_and_init_map_value(map, dst_val);
1796 		}
1797 		if (do_delete) {
1798 			hlist_nulls_del_rcu(&l->hash_node);
1799 
1800 			/* bpf_lru_push_free() will acquire lru_lock, which
1801 			 * may cause deadlock. See comments in function
1802 			 * prealloc_lru_pop(). Let us do bpf_lru_push_free()
1803 			 * after releasing the bucket lock.
1804 			 */
1805 			if (is_lru_map) {
1806 				l->batch_flink = node_to_free;
1807 				node_to_free = l;
1808 			} else {
1809 				free_htab_elem(htab, l);
1810 			}
1811 		}
1812 		dst_key += key_size;
1813 		dst_val += value_size;
1814 	}
1815 
1816 	htab_unlock_bucket(htab, b, batch, flags);
1817 	locked = false;
1818 
1819 	while (node_to_free) {
1820 		l = node_to_free;
1821 		node_to_free = node_to_free->batch_flink;
1822 		htab_lru_push_free(htab, l);
1823 	}
1824 
1825 next_batch:
1826 	/* If we are not copying data, we can go to next bucket and avoid
1827 	 * unlocking the rcu.
1828 	 */
1829 	if (!bucket_cnt && (batch + 1 < htab->n_buckets)) {
1830 		batch++;
1831 		goto again_nocopy;
1832 	}
1833 
1834 	rcu_read_unlock();
1835 	bpf_enable_instrumentation();
1836 	if (bucket_cnt && (copy_to_user(ukeys + total * key_size, keys,
1837 	    key_size * bucket_cnt) ||
1838 	    copy_to_user(uvalues + total * value_size, values,
1839 	    value_size * bucket_cnt))) {
1840 		ret = -EFAULT;
1841 		goto after_loop;
1842 	}
1843 
1844 	total += bucket_cnt;
1845 	batch++;
1846 	if (batch >= htab->n_buckets) {
1847 		ret = -ENOENT;
1848 		goto after_loop;
1849 	}
1850 	goto again;
1851 
1852 after_loop:
1853 	if (ret == -EFAULT)
1854 		goto out;
1855 
1856 	/* copy # of entries and next batch */
1857 	ubatch = u64_to_user_ptr(attr->batch.out_batch);
1858 	if (copy_to_user(ubatch, &batch, sizeof(batch)) ||
1859 	    put_user(total, &uattr->batch.count))
1860 		ret = -EFAULT;
1861 
1862 out:
1863 	kvfree(keys);
1864 	kvfree(values);
1865 	return ret;
1866 }
1867 
1868 static int
1869 htab_percpu_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1870 			     union bpf_attr __user *uattr)
1871 {
1872 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1873 						  false, true);
1874 }
1875 
1876 static int
1877 htab_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1878 					const union bpf_attr *attr,
1879 					union bpf_attr __user *uattr)
1880 {
1881 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1882 						  false, true);
1883 }
1884 
1885 static int
1886 htab_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1887 		      union bpf_attr __user *uattr)
1888 {
1889 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1890 						  false, false);
1891 }
1892 
1893 static int
1894 htab_map_lookup_and_delete_batch(struct bpf_map *map,
1895 				 const union bpf_attr *attr,
1896 				 union bpf_attr __user *uattr)
1897 {
1898 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1899 						  false, false);
1900 }
1901 
1902 static int
1903 htab_lru_percpu_map_lookup_batch(struct bpf_map *map,
1904 				 const union bpf_attr *attr,
1905 				 union bpf_attr __user *uattr)
1906 {
1907 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1908 						  true, true);
1909 }
1910 
1911 static int
1912 htab_lru_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1913 					    const union bpf_attr *attr,
1914 					    union bpf_attr __user *uattr)
1915 {
1916 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1917 						  true, true);
1918 }
1919 
1920 static int
1921 htab_lru_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1922 			  union bpf_attr __user *uattr)
1923 {
1924 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1925 						  true, false);
1926 }
1927 
1928 static int
1929 htab_lru_map_lookup_and_delete_batch(struct bpf_map *map,
1930 				     const union bpf_attr *attr,
1931 				     union bpf_attr __user *uattr)
1932 {
1933 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1934 						  true, false);
1935 }
1936 
1937 struct bpf_iter_seq_hash_map_info {
1938 	struct bpf_map *map;
1939 	struct bpf_htab *htab;
1940 	void *percpu_value_buf; // non-zero means percpu hash
1941 	u32 bucket_id;
1942 	u32 skip_elems;
1943 };
1944 
1945 static struct htab_elem *
1946 bpf_hash_map_seq_find_next(struct bpf_iter_seq_hash_map_info *info,
1947 			   struct htab_elem *prev_elem)
1948 {
1949 	const struct bpf_htab *htab = info->htab;
1950 	u32 skip_elems = info->skip_elems;
1951 	u32 bucket_id = info->bucket_id;
1952 	struct hlist_nulls_head *head;
1953 	struct hlist_nulls_node *n;
1954 	struct htab_elem *elem;
1955 	struct bucket *b;
1956 	u32 i, count;
1957 
1958 	if (bucket_id >= htab->n_buckets)
1959 		return NULL;
1960 
1961 	/* try to find next elem in the same bucket */
1962 	if (prev_elem) {
1963 		/* no update/deletion on this bucket, prev_elem should be still valid
1964 		 * and we won't skip elements.
1965 		 */
1966 		n = rcu_dereference_raw(hlist_nulls_next_rcu(&prev_elem->hash_node));
1967 		elem = hlist_nulls_entry_safe(n, struct htab_elem, hash_node);
1968 		if (elem)
1969 			return elem;
1970 
1971 		/* not found, unlock and go to the next bucket */
1972 		b = &htab->buckets[bucket_id++];
1973 		rcu_read_unlock();
1974 		skip_elems = 0;
1975 	}
1976 
1977 	for (i = bucket_id; i < htab->n_buckets; i++) {
1978 		b = &htab->buckets[i];
1979 		rcu_read_lock();
1980 
1981 		count = 0;
1982 		head = &b->head;
1983 		hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
1984 			if (count >= skip_elems) {
1985 				info->bucket_id = i;
1986 				info->skip_elems = count;
1987 				return elem;
1988 			}
1989 			count++;
1990 		}
1991 
1992 		rcu_read_unlock();
1993 		skip_elems = 0;
1994 	}
1995 
1996 	info->bucket_id = i;
1997 	info->skip_elems = 0;
1998 	return NULL;
1999 }
2000 
2001 static void *bpf_hash_map_seq_start(struct seq_file *seq, loff_t *pos)
2002 {
2003 	struct bpf_iter_seq_hash_map_info *info = seq->private;
2004 	struct htab_elem *elem;
2005 
2006 	elem = bpf_hash_map_seq_find_next(info, NULL);
2007 	if (!elem)
2008 		return NULL;
2009 
2010 	if (*pos == 0)
2011 		++*pos;
2012 	return elem;
2013 }
2014 
2015 static void *bpf_hash_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2016 {
2017 	struct bpf_iter_seq_hash_map_info *info = seq->private;
2018 
2019 	++*pos;
2020 	++info->skip_elems;
2021 	return bpf_hash_map_seq_find_next(info, v);
2022 }
2023 
2024 static int __bpf_hash_map_seq_show(struct seq_file *seq, struct htab_elem *elem)
2025 {
2026 	struct bpf_iter_seq_hash_map_info *info = seq->private;
2027 	u32 roundup_key_size, roundup_value_size;
2028 	struct bpf_iter__bpf_map_elem ctx = {};
2029 	struct bpf_map *map = info->map;
2030 	struct bpf_iter_meta meta;
2031 	int ret = 0, off = 0, cpu;
2032 	struct bpf_prog *prog;
2033 	void __percpu *pptr;
2034 
2035 	meta.seq = seq;
2036 	prog = bpf_iter_get_info(&meta, elem == NULL);
2037 	if (prog) {
2038 		ctx.meta = &meta;
2039 		ctx.map = info->map;
2040 		if (elem) {
2041 			roundup_key_size = round_up(map->key_size, 8);
2042 			ctx.key = elem->key;
2043 			if (!info->percpu_value_buf) {
2044 				ctx.value = elem->key + roundup_key_size;
2045 			} else {
2046 				roundup_value_size = round_up(map->value_size, 8);
2047 				pptr = htab_elem_get_ptr(elem, map->key_size);
2048 				for_each_possible_cpu(cpu) {
2049 					bpf_long_memcpy(info->percpu_value_buf + off,
2050 							per_cpu_ptr(pptr, cpu),
2051 							roundup_value_size);
2052 					off += roundup_value_size;
2053 				}
2054 				ctx.value = info->percpu_value_buf;
2055 			}
2056 		}
2057 		ret = bpf_iter_run_prog(prog, &ctx);
2058 	}
2059 
2060 	return ret;
2061 }
2062 
2063 static int bpf_hash_map_seq_show(struct seq_file *seq, void *v)
2064 {
2065 	return __bpf_hash_map_seq_show(seq, v);
2066 }
2067 
2068 static void bpf_hash_map_seq_stop(struct seq_file *seq, void *v)
2069 {
2070 	if (!v)
2071 		(void)__bpf_hash_map_seq_show(seq, NULL);
2072 	else
2073 		rcu_read_unlock();
2074 }
2075 
2076 static int bpf_iter_init_hash_map(void *priv_data,
2077 				  struct bpf_iter_aux_info *aux)
2078 {
2079 	struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2080 	struct bpf_map *map = aux->map;
2081 	void *value_buf;
2082 	u32 buf_size;
2083 
2084 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2085 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2086 		buf_size = round_up(map->value_size, 8) * num_possible_cpus();
2087 		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
2088 		if (!value_buf)
2089 			return -ENOMEM;
2090 
2091 		seq_info->percpu_value_buf = value_buf;
2092 	}
2093 
2094 	bpf_map_inc_with_uref(map);
2095 	seq_info->map = map;
2096 	seq_info->htab = container_of(map, struct bpf_htab, map);
2097 	return 0;
2098 }
2099 
2100 static void bpf_iter_fini_hash_map(void *priv_data)
2101 {
2102 	struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2103 
2104 	bpf_map_put_with_uref(seq_info->map);
2105 	kfree(seq_info->percpu_value_buf);
2106 }
2107 
2108 static const struct seq_operations bpf_hash_map_seq_ops = {
2109 	.start	= bpf_hash_map_seq_start,
2110 	.next	= bpf_hash_map_seq_next,
2111 	.stop	= bpf_hash_map_seq_stop,
2112 	.show	= bpf_hash_map_seq_show,
2113 };
2114 
2115 static const struct bpf_iter_seq_info iter_seq_info = {
2116 	.seq_ops		= &bpf_hash_map_seq_ops,
2117 	.init_seq_private	= bpf_iter_init_hash_map,
2118 	.fini_seq_private	= bpf_iter_fini_hash_map,
2119 	.seq_priv_size		= sizeof(struct bpf_iter_seq_hash_map_info),
2120 };
2121 
2122 static int bpf_for_each_hash_elem(struct bpf_map *map, bpf_callback_t callback_fn,
2123 				  void *callback_ctx, u64 flags)
2124 {
2125 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2126 	struct hlist_nulls_head *head;
2127 	struct hlist_nulls_node *n;
2128 	struct htab_elem *elem;
2129 	u32 roundup_key_size;
2130 	int i, num_elems = 0;
2131 	void __percpu *pptr;
2132 	struct bucket *b;
2133 	void *key, *val;
2134 	bool is_percpu;
2135 	u64 ret = 0;
2136 
2137 	if (flags != 0)
2138 		return -EINVAL;
2139 
2140 	is_percpu = htab_is_percpu(htab);
2141 
2142 	roundup_key_size = round_up(map->key_size, 8);
2143 	/* disable migration so percpu value prepared here will be the
2144 	 * same as the one seen by the bpf program with bpf_map_lookup_elem().
2145 	 */
2146 	if (is_percpu)
2147 		migrate_disable();
2148 	for (i = 0; i < htab->n_buckets; i++) {
2149 		b = &htab->buckets[i];
2150 		rcu_read_lock();
2151 		head = &b->head;
2152 		hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
2153 			key = elem->key;
2154 			if (is_percpu) {
2155 				/* current cpu value for percpu map */
2156 				pptr = htab_elem_get_ptr(elem, map->key_size);
2157 				val = this_cpu_ptr(pptr);
2158 			} else {
2159 				val = elem->key + roundup_key_size;
2160 			}
2161 			num_elems++;
2162 			ret = callback_fn((u64)(long)map, (u64)(long)key,
2163 					  (u64)(long)val, (u64)(long)callback_ctx, 0);
2164 			/* return value: 0 - continue, 1 - stop and return */
2165 			if (ret) {
2166 				rcu_read_unlock();
2167 				goto out;
2168 			}
2169 		}
2170 		rcu_read_unlock();
2171 	}
2172 out:
2173 	if (is_percpu)
2174 		migrate_enable();
2175 	return num_elems;
2176 }
2177 
2178 BTF_ID_LIST_SINGLE(htab_map_btf_ids, struct, bpf_htab)
2179 const struct bpf_map_ops htab_map_ops = {
2180 	.map_meta_equal = bpf_map_meta_equal,
2181 	.map_alloc_check = htab_map_alloc_check,
2182 	.map_alloc = htab_map_alloc,
2183 	.map_free = htab_map_free,
2184 	.map_get_next_key = htab_map_get_next_key,
2185 	.map_release_uref = htab_map_free_timers,
2186 	.map_lookup_elem = htab_map_lookup_elem,
2187 	.map_lookup_and_delete_elem = htab_map_lookup_and_delete_elem,
2188 	.map_update_elem = htab_map_update_elem,
2189 	.map_delete_elem = htab_map_delete_elem,
2190 	.map_gen_lookup = htab_map_gen_lookup,
2191 	.map_seq_show_elem = htab_map_seq_show_elem,
2192 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2193 	.map_for_each_callback = bpf_for_each_hash_elem,
2194 	BATCH_OPS(htab),
2195 	.map_btf_id = &htab_map_btf_ids[0],
2196 	.iter_seq_info = &iter_seq_info,
2197 };
2198 
2199 const struct bpf_map_ops htab_lru_map_ops = {
2200 	.map_meta_equal = bpf_map_meta_equal,
2201 	.map_alloc_check = htab_map_alloc_check,
2202 	.map_alloc = htab_map_alloc,
2203 	.map_free = htab_map_free,
2204 	.map_get_next_key = htab_map_get_next_key,
2205 	.map_release_uref = htab_map_free_timers,
2206 	.map_lookup_elem = htab_lru_map_lookup_elem,
2207 	.map_lookup_and_delete_elem = htab_lru_map_lookup_and_delete_elem,
2208 	.map_lookup_elem_sys_only = htab_lru_map_lookup_elem_sys,
2209 	.map_update_elem = htab_lru_map_update_elem,
2210 	.map_delete_elem = htab_lru_map_delete_elem,
2211 	.map_gen_lookup = htab_lru_map_gen_lookup,
2212 	.map_seq_show_elem = htab_map_seq_show_elem,
2213 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2214 	.map_for_each_callback = bpf_for_each_hash_elem,
2215 	BATCH_OPS(htab_lru),
2216 	.map_btf_id = &htab_map_btf_ids[0],
2217 	.iter_seq_info = &iter_seq_info,
2218 };
2219 
2220 /* Called from eBPF program */
2221 static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2222 {
2223 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
2224 
2225 	if (l)
2226 		return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2227 	else
2228 		return NULL;
2229 }
2230 
2231 static void *htab_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2232 {
2233 	struct htab_elem *l;
2234 
2235 	if (cpu >= nr_cpu_ids)
2236 		return NULL;
2237 
2238 	l = __htab_map_lookup_elem(map, key);
2239 	if (l)
2240 		return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2241 	else
2242 		return NULL;
2243 }
2244 
2245 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2246 {
2247 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
2248 
2249 	if (l) {
2250 		bpf_lru_node_set_ref(&l->lru_node);
2251 		return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2252 	}
2253 
2254 	return NULL;
2255 }
2256 
2257 static void *htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2258 {
2259 	struct htab_elem *l;
2260 
2261 	if (cpu >= nr_cpu_ids)
2262 		return NULL;
2263 
2264 	l = __htab_map_lookup_elem(map, key);
2265 	if (l) {
2266 		bpf_lru_node_set_ref(&l->lru_node);
2267 		return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2268 	}
2269 
2270 	return NULL;
2271 }
2272 
2273 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
2274 {
2275 	struct htab_elem *l;
2276 	void __percpu *pptr;
2277 	int ret = -ENOENT;
2278 	int cpu, off = 0;
2279 	u32 size;
2280 
2281 	/* per_cpu areas are zero-filled and bpf programs can only
2282 	 * access 'value_size' of them, so copying rounded areas
2283 	 * will not leak any kernel data
2284 	 */
2285 	size = round_up(map->value_size, 8);
2286 	rcu_read_lock();
2287 	l = __htab_map_lookup_elem(map, key);
2288 	if (!l)
2289 		goto out;
2290 	/* We do not mark LRU map element here in order to not mess up
2291 	 * eviction heuristics when user space does a map walk.
2292 	 */
2293 	pptr = htab_elem_get_ptr(l, map->key_size);
2294 	for_each_possible_cpu(cpu) {
2295 		bpf_long_memcpy(value + off,
2296 				per_cpu_ptr(pptr, cpu), size);
2297 		off += size;
2298 	}
2299 	ret = 0;
2300 out:
2301 	rcu_read_unlock();
2302 	return ret;
2303 }
2304 
2305 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
2306 			   u64 map_flags)
2307 {
2308 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2309 	int ret;
2310 
2311 	rcu_read_lock();
2312 	if (htab_is_lru(htab))
2313 		ret = __htab_lru_percpu_map_update_elem(map, key, value,
2314 							map_flags, true);
2315 	else
2316 		ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
2317 						    true);
2318 	rcu_read_unlock();
2319 
2320 	return ret;
2321 }
2322 
2323 static void htab_percpu_map_seq_show_elem(struct bpf_map *map, void *key,
2324 					  struct seq_file *m)
2325 {
2326 	struct htab_elem *l;
2327 	void __percpu *pptr;
2328 	int cpu;
2329 
2330 	rcu_read_lock();
2331 
2332 	l = __htab_map_lookup_elem(map, key);
2333 	if (!l) {
2334 		rcu_read_unlock();
2335 		return;
2336 	}
2337 
2338 	btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
2339 	seq_puts(m, ": {\n");
2340 	pptr = htab_elem_get_ptr(l, map->key_size);
2341 	for_each_possible_cpu(cpu) {
2342 		seq_printf(m, "\tcpu%d: ", cpu);
2343 		btf_type_seq_show(map->btf, map->btf_value_type_id,
2344 				  per_cpu_ptr(pptr, cpu), m);
2345 		seq_puts(m, "\n");
2346 	}
2347 	seq_puts(m, "}\n");
2348 
2349 	rcu_read_unlock();
2350 }
2351 
2352 const struct bpf_map_ops htab_percpu_map_ops = {
2353 	.map_meta_equal = bpf_map_meta_equal,
2354 	.map_alloc_check = htab_map_alloc_check,
2355 	.map_alloc = htab_map_alloc,
2356 	.map_free = htab_map_free,
2357 	.map_get_next_key = htab_map_get_next_key,
2358 	.map_lookup_elem = htab_percpu_map_lookup_elem,
2359 	.map_lookup_and_delete_elem = htab_percpu_map_lookup_and_delete_elem,
2360 	.map_update_elem = htab_percpu_map_update_elem,
2361 	.map_delete_elem = htab_map_delete_elem,
2362 	.map_lookup_percpu_elem = htab_percpu_map_lookup_percpu_elem,
2363 	.map_seq_show_elem = htab_percpu_map_seq_show_elem,
2364 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2365 	.map_for_each_callback = bpf_for_each_hash_elem,
2366 	BATCH_OPS(htab_percpu),
2367 	.map_btf_id = &htab_map_btf_ids[0],
2368 	.iter_seq_info = &iter_seq_info,
2369 };
2370 
2371 const struct bpf_map_ops htab_lru_percpu_map_ops = {
2372 	.map_meta_equal = bpf_map_meta_equal,
2373 	.map_alloc_check = htab_map_alloc_check,
2374 	.map_alloc = htab_map_alloc,
2375 	.map_free = htab_map_free,
2376 	.map_get_next_key = htab_map_get_next_key,
2377 	.map_lookup_elem = htab_lru_percpu_map_lookup_elem,
2378 	.map_lookup_and_delete_elem = htab_lru_percpu_map_lookup_and_delete_elem,
2379 	.map_update_elem = htab_lru_percpu_map_update_elem,
2380 	.map_delete_elem = htab_lru_map_delete_elem,
2381 	.map_lookup_percpu_elem = htab_lru_percpu_map_lookup_percpu_elem,
2382 	.map_seq_show_elem = htab_percpu_map_seq_show_elem,
2383 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2384 	.map_for_each_callback = bpf_for_each_hash_elem,
2385 	BATCH_OPS(htab_lru_percpu),
2386 	.map_btf_id = &htab_map_btf_ids[0],
2387 	.iter_seq_info = &iter_seq_info,
2388 };
2389 
2390 static int fd_htab_map_alloc_check(union bpf_attr *attr)
2391 {
2392 	if (attr->value_size != sizeof(u32))
2393 		return -EINVAL;
2394 	return htab_map_alloc_check(attr);
2395 }
2396 
2397 static void fd_htab_map_free(struct bpf_map *map)
2398 {
2399 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2400 	struct hlist_nulls_node *n;
2401 	struct hlist_nulls_head *head;
2402 	struct htab_elem *l;
2403 	int i;
2404 
2405 	for (i = 0; i < htab->n_buckets; i++) {
2406 		head = select_bucket(htab, i);
2407 
2408 		hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
2409 			void *ptr = fd_htab_map_get_ptr(map, l);
2410 
2411 			map->ops->map_fd_put_ptr(ptr);
2412 		}
2413 	}
2414 
2415 	htab_map_free(map);
2416 }
2417 
2418 /* only called from syscall */
2419 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
2420 {
2421 	void **ptr;
2422 	int ret = 0;
2423 
2424 	if (!map->ops->map_fd_sys_lookup_elem)
2425 		return -ENOTSUPP;
2426 
2427 	rcu_read_lock();
2428 	ptr = htab_map_lookup_elem(map, key);
2429 	if (ptr)
2430 		*value = map->ops->map_fd_sys_lookup_elem(READ_ONCE(*ptr));
2431 	else
2432 		ret = -ENOENT;
2433 	rcu_read_unlock();
2434 
2435 	return ret;
2436 }
2437 
2438 /* only called from syscall */
2439 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
2440 				void *key, void *value, u64 map_flags)
2441 {
2442 	void *ptr;
2443 	int ret;
2444 	u32 ufd = *(u32 *)value;
2445 
2446 	ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
2447 	if (IS_ERR(ptr))
2448 		return PTR_ERR(ptr);
2449 
2450 	ret = htab_map_update_elem(map, key, &ptr, map_flags);
2451 	if (ret)
2452 		map->ops->map_fd_put_ptr(ptr);
2453 
2454 	return ret;
2455 }
2456 
2457 static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
2458 {
2459 	struct bpf_map *map, *inner_map_meta;
2460 
2461 	inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
2462 	if (IS_ERR(inner_map_meta))
2463 		return inner_map_meta;
2464 
2465 	map = htab_map_alloc(attr);
2466 	if (IS_ERR(map)) {
2467 		bpf_map_meta_free(inner_map_meta);
2468 		return map;
2469 	}
2470 
2471 	map->inner_map_meta = inner_map_meta;
2472 
2473 	return map;
2474 }
2475 
2476 static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
2477 {
2478 	struct bpf_map **inner_map  = htab_map_lookup_elem(map, key);
2479 
2480 	if (!inner_map)
2481 		return NULL;
2482 
2483 	return READ_ONCE(*inner_map);
2484 }
2485 
2486 static int htab_of_map_gen_lookup(struct bpf_map *map,
2487 				  struct bpf_insn *insn_buf)
2488 {
2489 	struct bpf_insn *insn = insn_buf;
2490 	const int ret = BPF_REG_0;
2491 
2492 	BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
2493 		     (void *(*)(struct bpf_map *map, void *key))NULL));
2494 	*insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
2495 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 2);
2496 	*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
2497 				offsetof(struct htab_elem, key) +
2498 				round_up(map->key_size, 8));
2499 	*insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
2500 
2501 	return insn - insn_buf;
2502 }
2503 
2504 static void htab_of_map_free(struct bpf_map *map)
2505 {
2506 	bpf_map_meta_free(map->inner_map_meta);
2507 	fd_htab_map_free(map);
2508 }
2509 
2510 const struct bpf_map_ops htab_of_maps_map_ops = {
2511 	.map_alloc_check = fd_htab_map_alloc_check,
2512 	.map_alloc = htab_of_map_alloc,
2513 	.map_free = htab_of_map_free,
2514 	.map_get_next_key = htab_map_get_next_key,
2515 	.map_lookup_elem = htab_of_map_lookup_elem,
2516 	.map_delete_elem = htab_map_delete_elem,
2517 	.map_fd_get_ptr = bpf_map_fd_get_ptr,
2518 	.map_fd_put_ptr = bpf_map_fd_put_ptr,
2519 	.map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
2520 	.map_gen_lookup = htab_of_map_gen_lookup,
2521 	.map_check_btf = map_check_no_btf,
2522 	BATCH_OPS(htab),
2523 	.map_btf_id = &htab_map_btf_ids[0],
2524 };
2525