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 
htab_is_prealloc(const struct bpf_htab * htab)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 
htab_init_buckets(struct bpf_htab * htab)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 
htab_lock_bucket(const struct bpf_htab * htab,struct bucket * b,u32 hash,unsigned long * pflags)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 
htab_unlock_bucket(const struct bpf_htab * htab,struct bucket * b,u32 hash,unsigned long flags)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 
htab_is_lru(const struct bpf_htab * htab)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 
htab_is_percpu(const struct bpf_htab * htab)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 
htab_elem_set_ptr(struct htab_elem * l,u32 key_size,void __percpu * pptr)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 
htab_elem_get_ptr(struct htab_elem * l,u32 key_size)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 
fd_htab_map_get_ptr(const struct bpf_map * map,struct htab_elem * l)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 
get_htab_elem(struct bpf_htab * htab,int i)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 
htab_has_extra_elems(struct bpf_htab * htab)215 static bool htab_has_extra_elems(struct bpf_htab *htab)
216 {
217 	return !htab_is_percpu(htab) && !htab_is_lru(htab);
218 }
219 
htab_free_prealloced_timers(struct bpf_htab * htab)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 (!map_value_has_timer(&htab->map))
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_timer_cancel_and_free(elem->key +
235 					  round_up(htab->map.key_size, 8) +
236 					  htab->map.timer_off);
237 		cond_resched();
238 	}
239 }
240 
htab_free_prealloced_kptrs(struct bpf_htab * htab)241 static void htab_free_prealloced_kptrs(struct bpf_htab *htab)
242 {
243 	u32 num_entries = htab->map.max_entries;
244 	int i;
245 
246 	if (!map_value_has_kptrs(&htab->map))
247 		return;
248 	if (htab_has_extra_elems(htab))
249 		num_entries += num_possible_cpus();
250 
251 	for (i = 0; i < num_entries; i++) {
252 		struct htab_elem *elem;
253 
254 		elem = get_htab_elem(htab, i);
255 		bpf_map_free_kptrs(&htab->map, elem->key + round_up(htab->map.key_size, 8));
256 		cond_resched();
257 	}
258 }
259 
htab_free_elems(struct bpf_htab * htab)260 static void htab_free_elems(struct bpf_htab *htab)
261 {
262 	int i;
263 
264 	if (!htab_is_percpu(htab))
265 		goto free_elems;
266 
267 	for (i = 0; i < htab->map.max_entries; i++) {
268 		void __percpu *pptr;
269 
270 		pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
271 					 htab->map.key_size);
272 		free_percpu(pptr);
273 		cond_resched();
274 	}
275 free_elems:
276 	bpf_map_area_free(htab->elems);
277 }
278 
279 /* The LRU list has a lock (lru_lock). Each htab bucket has a lock
280  * (bucket_lock). If both locks need to be acquired together, the lock
281  * order is always lru_lock -> bucket_lock and this only happens in
282  * bpf_lru_list.c logic. For example, certain code path of
283  * bpf_lru_pop_free(), which is called by function prealloc_lru_pop(),
284  * will acquire lru_lock first followed by acquiring bucket_lock.
285  *
286  * In hashtab.c, to avoid deadlock, lock acquisition of
287  * bucket_lock followed by lru_lock is not allowed. In such cases,
288  * bucket_lock needs to be released first before acquiring lru_lock.
289  */
prealloc_lru_pop(struct bpf_htab * htab,void * key,u32 hash)290 static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
291 					  u32 hash)
292 {
293 	struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
294 	struct htab_elem *l;
295 
296 	if (node) {
297 		l = container_of(node, struct htab_elem, lru_node);
298 		memcpy(l->key, key, htab->map.key_size);
299 		return l;
300 	}
301 
302 	return NULL;
303 }
304 
prealloc_init(struct bpf_htab * htab)305 static int prealloc_init(struct bpf_htab *htab)
306 {
307 	u32 num_entries = htab->map.max_entries;
308 	int err = -ENOMEM, i;
309 
310 	if (htab_has_extra_elems(htab))
311 		num_entries += num_possible_cpus();
312 
313 	htab->elems = bpf_map_area_alloc((u64)htab->elem_size * num_entries,
314 					 htab->map.numa_node);
315 	if (!htab->elems)
316 		return -ENOMEM;
317 
318 	if (!htab_is_percpu(htab))
319 		goto skip_percpu_elems;
320 
321 	for (i = 0; i < num_entries; i++) {
322 		u32 size = round_up(htab->map.value_size, 8);
323 		void __percpu *pptr;
324 
325 		pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
326 					    GFP_USER | __GFP_NOWARN);
327 		if (!pptr)
328 			goto free_elems;
329 		htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
330 				  pptr);
331 		cond_resched();
332 	}
333 
334 skip_percpu_elems:
335 	if (htab_is_lru(htab))
336 		err = bpf_lru_init(&htab->lru,
337 				   htab->map.map_flags & BPF_F_NO_COMMON_LRU,
338 				   offsetof(struct htab_elem, hash) -
339 				   offsetof(struct htab_elem, lru_node),
340 				   htab_lru_map_delete_node,
341 				   htab);
342 	else
343 		err = pcpu_freelist_init(&htab->freelist);
344 
345 	if (err)
346 		goto free_elems;
347 
348 	if (htab_is_lru(htab))
349 		bpf_lru_populate(&htab->lru, htab->elems,
350 				 offsetof(struct htab_elem, lru_node),
351 				 htab->elem_size, num_entries);
352 	else
353 		pcpu_freelist_populate(&htab->freelist,
354 				       htab->elems + offsetof(struct htab_elem, fnode),
355 				       htab->elem_size, num_entries);
356 
357 	return 0;
358 
359 free_elems:
360 	htab_free_elems(htab);
361 	return err;
362 }
363 
prealloc_destroy(struct bpf_htab * htab)364 static void prealloc_destroy(struct bpf_htab *htab)
365 {
366 	htab_free_elems(htab);
367 
368 	if (htab_is_lru(htab))
369 		bpf_lru_destroy(&htab->lru);
370 	else
371 		pcpu_freelist_destroy(&htab->freelist);
372 }
373 
alloc_extra_elems(struct bpf_htab * htab)374 static int alloc_extra_elems(struct bpf_htab *htab)
375 {
376 	struct htab_elem *__percpu *pptr, *l_new;
377 	struct pcpu_freelist_node *l;
378 	int cpu;
379 
380 	pptr = bpf_map_alloc_percpu(&htab->map, sizeof(struct htab_elem *), 8,
381 				    GFP_USER | __GFP_NOWARN);
382 	if (!pptr)
383 		return -ENOMEM;
384 
385 	for_each_possible_cpu(cpu) {
386 		l = pcpu_freelist_pop(&htab->freelist);
387 		/* pop will succeed, since prealloc_init()
388 		 * preallocated extra num_possible_cpus elements
389 		 */
390 		l_new = container_of(l, struct htab_elem, fnode);
391 		*per_cpu_ptr(pptr, cpu) = l_new;
392 	}
393 	htab->extra_elems = pptr;
394 	return 0;
395 }
396 
397 /* Called from syscall */
htab_map_alloc_check(union bpf_attr * attr)398 static int htab_map_alloc_check(union bpf_attr *attr)
399 {
400 	bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
401 		       attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
402 	bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
403 		    attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
404 	/* percpu_lru means each cpu has its own LRU list.
405 	 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
406 	 * the map's value itself is percpu.  percpu_lru has
407 	 * nothing to do with the map's value.
408 	 */
409 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
410 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
411 	bool zero_seed = (attr->map_flags & BPF_F_ZERO_SEED);
412 	int numa_node = bpf_map_attr_numa_node(attr);
413 
414 	BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) !=
415 		     offsetof(struct htab_elem, hash_node.pprev));
416 
417 	if (lru && !bpf_capable())
418 		/* LRU implementation is much complicated than other
419 		 * maps.  Hence, limit to CAP_BPF.
420 		 */
421 		return -EPERM;
422 
423 	if (zero_seed && !capable(CAP_SYS_ADMIN))
424 		/* Guard against local DoS, and discourage production use. */
425 		return -EPERM;
426 
427 	if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK ||
428 	    !bpf_map_flags_access_ok(attr->map_flags))
429 		return -EINVAL;
430 
431 	if (!lru && percpu_lru)
432 		return -EINVAL;
433 
434 	if (lru && !prealloc)
435 		return -ENOTSUPP;
436 
437 	if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
438 		return -EINVAL;
439 
440 	/* check sanity of attributes.
441 	 * value_size == 0 may be allowed in the future to use map as a set
442 	 */
443 	if (attr->max_entries == 0 || attr->key_size == 0 ||
444 	    attr->value_size == 0)
445 		return -EINVAL;
446 
447 	if ((u64)attr->key_size + attr->value_size >= KMALLOC_MAX_SIZE -
448 	   sizeof(struct htab_elem))
449 		/* if key_size + value_size is bigger, the user space won't be
450 		 * able to access the elements via bpf syscall. This check
451 		 * also makes sure that the elem_size doesn't overflow and it's
452 		 * kmalloc-able later in htab_map_update_elem()
453 		 */
454 		return -E2BIG;
455 
456 	return 0;
457 }
458 
htab_map_alloc(union bpf_attr * attr)459 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
460 {
461 	bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
462 		       attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
463 	bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
464 		    attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
465 	/* percpu_lru means each cpu has its own LRU list.
466 	 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
467 	 * the map's value itself is percpu.  percpu_lru has
468 	 * nothing to do with the map's value.
469 	 */
470 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
471 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
472 	struct bpf_htab *htab;
473 	int err, i;
474 
475 	htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
476 	if (!htab)
477 		return ERR_PTR(-ENOMEM);
478 
479 	lockdep_register_key(&htab->lockdep_key);
480 
481 	bpf_map_init_from_attr(&htab->map, attr);
482 
483 	if (percpu_lru) {
484 		/* ensure each CPU's lru list has >=1 elements.
485 		 * since we are at it, make each lru list has the same
486 		 * number of elements.
487 		 */
488 		htab->map.max_entries = roundup(attr->max_entries,
489 						num_possible_cpus());
490 		if (htab->map.max_entries < attr->max_entries)
491 			htab->map.max_entries = rounddown(attr->max_entries,
492 							  num_possible_cpus());
493 	}
494 
495 	/* hash table size must be power of 2 */
496 	htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
497 
498 	htab->elem_size = sizeof(struct htab_elem) +
499 			  round_up(htab->map.key_size, 8);
500 	if (percpu)
501 		htab->elem_size += sizeof(void *);
502 	else
503 		htab->elem_size += round_up(htab->map.value_size, 8);
504 
505 	err = -E2BIG;
506 	/* prevent zero size kmalloc and check for u32 overflow */
507 	if (htab->n_buckets == 0 ||
508 	    htab->n_buckets > U32_MAX / sizeof(struct bucket))
509 		goto free_htab;
510 
511 	err = -ENOMEM;
512 	htab->buckets = bpf_map_area_alloc(htab->n_buckets *
513 					   sizeof(struct bucket),
514 					   htab->map.numa_node);
515 	if (!htab->buckets)
516 		goto free_htab;
517 
518 	for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++) {
519 		htab->map_locked[i] = bpf_map_alloc_percpu(&htab->map,
520 							   sizeof(int),
521 							   sizeof(int),
522 							   GFP_USER);
523 		if (!htab->map_locked[i])
524 			goto free_map_locked;
525 	}
526 
527 	if (htab->map.map_flags & BPF_F_ZERO_SEED)
528 		htab->hashrnd = 0;
529 	else
530 		htab->hashrnd = get_random_u32();
531 
532 	htab_init_buckets(htab);
533 
534 /* compute_batch_value() computes batch value as num_online_cpus() * 2
535  * and __percpu_counter_compare() needs
536  * htab->max_entries - cur_number_of_elems to be more than batch * num_online_cpus()
537  * for percpu_counter to be faster than atomic_t. In practice the average bpf
538  * hash map size is 10k, which means that a system with 64 cpus will fill
539  * hashmap to 20% of 10k before percpu_counter becomes ineffective. Therefore
540  * define our own batch count as 32 then 10k hash map can be filled up to 80%:
541  * 10k - 8k > 32 _batch_ * 64 _cpus_
542  * and __percpu_counter_compare() will still be fast. At that point hash map
543  * collisions will dominate its performance anyway. Assume that hash map filled
544  * to 50+% isn't going to be O(1) and use the following formula to choose
545  * between percpu_counter and atomic_t.
546  */
547 #define PERCPU_COUNTER_BATCH 32
548 	if (attr->max_entries / 2 > num_online_cpus() * PERCPU_COUNTER_BATCH)
549 		htab->use_percpu_counter = true;
550 
551 	if (htab->use_percpu_counter) {
552 		err = percpu_counter_init(&htab->pcount, 0, GFP_KERNEL);
553 		if (err)
554 			goto free_map_locked;
555 	}
556 
557 	if (prealloc) {
558 		err = prealloc_init(htab);
559 		if (err)
560 			goto free_map_locked;
561 
562 		if (!percpu && !lru) {
563 			/* lru itself can remove the least used element, so
564 			 * there is no need for an extra elem during map_update.
565 			 */
566 			err = alloc_extra_elems(htab);
567 			if (err)
568 				goto free_prealloc;
569 		}
570 	} else {
571 		err = bpf_mem_alloc_init(&htab->ma, htab->elem_size, false);
572 		if (err)
573 			goto free_map_locked;
574 		if (percpu) {
575 			err = bpf_mem_alloc_init(&htab->pcpu_ma,
576 						 round_up(htab->map.value_size, 8), true);
577 			if (err)
578 				goto free_map_locked;
579 		}
580 	}
581 
582 	return &htab->map;
583 
584 free_prealloc:
585 	prealloc_destroy(htab);
586 free_map_locked:
587 	if (htab->use_percpu_counter)
588 		percpu_counter_destroy(&htab->pcount);
589 	for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
590 		free_percpu(htab->map_locked[i]);
591 	bpf_map_area_free(htab->buckets);
592 	bpf_mem_alloc_destroy(&htab->pcpu_ma);
593 	bpf_mem_alloc_destroy(&htab->ma);
594 free_htab:
595 	lockdep_unregister_key(&htab->lockdep_key);
596 	bpf_map_area_free(htab);
597 	return ERR_PTR(err);
598 }
599 
htab_map_hash(const void * key,u32 key_len,u32 hashrnd)600 static inline u32 htab_map_hash(const void *key, u32 key_len, u32 hashrnd)
601 {
602 	return jhash(key, key_len, hashrnd);
603 }
604 
__select_bucket(struct bpf_htab * htab,u32 hash)605 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
606 {
607 	return &htab->buckets[hash & (htab->n_buckets - 1)];
608 }
609 
select_bucket(struct bpf_htab * htab,u32 hash)610 static inline struct hlist_nulls_head *select_bucket(struct bpf_htab *htab, u32 hash)
611 {
612 	return &__select_bucket(htab, hash)->head;
613 }
614 
615 /* this lookup function can only be called with bucket lock taken */
lookup_elem_raw(struct hlist_nulls_head * head,u32 hash,void * key,u32 key_size)616 static struct htab_elem *lookup_elem_raw(struct hlist_nulls_head *head, u32 hash,
617 					 void *key, u32 key_size)
618 {
619 	struct hlist_nulls_node *n;
620 	struct htab_elem *l;
621 
622 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
623 		if (l->hash == hash && !memcmp(&l->key, key, key_size))
624 			return l;
625 
626 	return NULL;
627 }
628 
629 /* can be called without bucket lock. it will repeat the loop in
630  * the unlikely event when elements moved from one bucket into another
631  * while link list is being walked
632  */
lookup_nulls_elem_raw(struct hlist_nulls_head * head,u32 hash,void * key,u32 key_size,u32 n_buckets)633 static struct htab_elem *lookup_nulls_elem_raw(struct hlist_nulls_head *head,
634 					       u32 hash, void *key,
635 					       u32 key_size, u32 n_buckets)
636 {
637 	struct hlist_nulls_node *n;
638 	struct htab_elem *l;
639 
640 again:
641 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
642 		if (l->hash == hash && !memcmp(&l->key, key, key_size))
643 			return l;
644 
645 	if (unlikely(get_nulls_value(n) != (hash & (n_buckets - 1))))
646 		goto again;
647 
648 	return NULL;
649 }
650 
651 /* Called from syscall or from eBPF program directly, so
652  * arguments have to match bpf_map_lookup_elem() exactly.
653  * The return value is adjusted by BPF instructions
654  * in htab_map_gen_lookup().
655  */
__htab_map_lookup_elem(struct bpf_map * map,void * key)656 static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
657 {
658 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
659 	struct hlist_nulls_head *head;
660 	struct htab_elem *l;
661 	u32 hash, key_size;
662 
663 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
664 		     !rcu_read_lock_bh_held());
665 
666 	key_size = map->key_size;
667 
668 	hash = htab_map_hash(key, key_size, htab->hashrnd);
669 
670 	head = select_bucket(htab, hash);
671 
672 	l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
673 
674 	return l;
675 }
676 
htab_map_lookup_elem(struct bpf_map * map,void * key)677 static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
678 {
679 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
680 
681 	if (l)
682 		return l->key + round_up(map->key_size, 8);
683 
684 	return NULL;
685 }
686 
687 /* inline bpf_map_lookup_elem() call.
688  * Instead of:
689  * bpf_prog
690  *   bpf_map_lookup_elem
691  *     map->ops->map_lookup_elem
692  *       htab_map_lookup_elem
693  *         __htab_map_lookup_elem
694  * do:
695  * bpf_prog
696  *   __htab_map_lookup_elem
697  */
htab_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)698 static int htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
699 {
700 	struct bpf_insn *insn = insn_buf;
701 	const int ret = BPF_REG_0;
702 
703 	BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
704 		     (void *(*)(struct bpf_map *map, void *key))NULL));
705 	*insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
706 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
707 	*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
708 				offsetof(struct htab_elem, key) +
709 				round_up(map->key_size, 8));
710 	return insn - insn_buf;
711 }
712 
__htab_lru_map_lookup_elem(struct bpf_map * map,void * key,const bool mark)713 static __always_inline void *__htab_lru_map_lookup_elem(struct bpf_map *map,
714 							void *key, const bool mark)
715 {
716 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
717 
718 	if (l) {
719 		if (mark)
720 			bpf_lru_node_set_ref(&l->lru_node);
721 		return l->key + round_up(map->key_size, 8);
722 	}
723 
724 	return NULL;
725 }
726 
htab_lru_map_lookup_elem(struct bpf_map * map,void * key)727 static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
728 {
729 	return __htab_lru_map_lookup_elem(map, key, true);
730 }
731 
htab_lru_map_lookup_elem_sys(struct bpf_map * map,void * key)732 static void *htab_lru_map_lookup_elem_sys(struct bpf_map *map, void *key)
733 {
734 	return __htab_lru_map_lookup_elem(map, key, false);
735 }
736 
htab_lru_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)737 static int htab_lru_map_gen_lookup(struct bpf_map *map,
738 				   struct bpf_insn *insn_buf)
739 {
740 	struct bpf_insn *insn = insn_buf;
741 	const int ret = BPF_REG_0;
742 	const int ref_reg = BPF_REG_1;
743 
744 	BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
745 		     (void *(*)(struct bpf_map *map, void *key))NULL));
746 	*insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
747 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 4);
748 	*insn++ = BPF_LDX_MEM(BPF_B, ref_reg, ret,
749 			      offsetof(struct htab_elem, lru_node) +
750 			      offsetof(struct bpf_lru_node, ref));
751 	*insn++ = BPF_JMP_IMM(BPF_JNE, ref_reg, 0, 1);
752 	*insn++ = BPF_ST_MEM(BPF_B, ret,
753 			     offsetof(struct htab_elem, lru_node) +
754 			     offsetof(struct bpf_lru_node, ref),
755 			     1);
756 	*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
757 				offsetof(struct htab_elem, key) +
758 				round_up(map->key_size, 8));
759 	return insn - insn_buf;
760 }
761 
check_and_free_fields(struct bpf_htab * htab,struct htab_elem * elem)762 static void check_and_free_fields(struct bpf_htab *htab,
763 				  struct htab_elem *elem)
764 {
765 	void *map_value = elem->key + round_up(htab->map.key_size, 8);
766 
767 	if (map_value_has_timer(&htab->map))
768 		bpf_timer_cancel_and_free(map_value + htab->map.timer_off);
769 	if (map_value_has_kptrs(&htab->map))
770 		bpf_map_free_kptrs(&htab->map, map_value);
771 }
772 
773 /* It is called from the bpf_lru_list when the LRU needs to delete
774  * older elements from the htab.
775  */
htab_lru_map_delete_node(void * arg,struct bpf_lru_node * node)776 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
777 {
778 	struct bpf_htab *htab = arg;
779 	struct htab_elem *l = NULL, *tgt_l;
780 	struct hlist_nulls_head *head;
781 	struct hlist_nulls_node *n;
782 	unsigned long flags;
783 	struct bucket *b;
784 	int ret;
785 
786 	tgt_l = container_of(node, struct htab_elem, lru_node);
787 	b = __select_bucket(htab, tgt_l->hash);
788 	head = &b->head;
789 
790 	ret = htab_lock_bucket(htab, b, tgt_l->hash, &flags);
791 	if (ret)
792 		return false;
793 
794 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
795 		if (l == tgt_l) {
796 			hlist_nulls_del_rcu(&l->hash_node);
797 			check_and_free_fields(htab, l);
798 			break;
799 		}
800 
801 	htab_unlock_bucket(htab, b, tgt_l->hash, flags);
802 
803 	return l == tgt_l;
804 }
805 
806 /* Called from syscall */
htab_map_get_next_key(struct bpf_map * map,void * key,void * next_key)807 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
808 {
809 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
810 	struct hlist_nulls_head *head;
811 	struct htab_elem *l, *next_l;
812 	u32 hash, key_size;
813 	int i = 0;
814 
815 	WARN_ON_ONCE(!rcu_read_lock_held());
816 
817 	key_size = map->key_size;
818 
819 	if (!key)
820 		goto find_first_elem;
821 
822 	hash = htab_map_hash(key, key_size, htab->hashrnd);
823 
824 	head = select_bucket(htab, hash);
825 
826 	/* lookup the key */
827 	l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
828 
829 	if (!l)
830 		goto find_first_elem;
831 
832 	/* key was found, get next key in the same bucket */
833 	next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l->hash_node)),
834 				  struct htab_elem, hash_node);
835 
836 	if (next_l) {
837 		/* if next elem in this hash list is non-zero, just return it */
838 		memcpy(next_key, next_l->key, key_size);
839 		return 0;
840 	}
841 
842 	/* no more elements in this hash list, go to the next bucket */
843 	i = hash & (htab->n_buckets - 1);
844 	i++;
845 
846 find_first_elem:
847 	/* iterate over buckets */
848 	for (; i < htab->n_buckets; i++) {
849 		head = select_bucket(htab, i);
850 
851 		/* pick first element in the bucket */
852 		next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head)),
853 					  struct htab_elem, hash_node);
854 		if (next_l) {
855 			/* if it's not empty, just return it */
856 			memcpy(next_key, next_l->key, key_size);
857 			return 0;
858 		}
859 	}
860 
861 	/* iterated over all buckets and all elements */
862 	return -ENOENT;
863 }
864 
htab_elem_free(struct bpf_htab * htab,struct htab_elem * l)865 static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
866 {
867 	if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
868 		bpf_mem_cache_free(&htab->pcpu_ma, l->ptr_to_pptr);
869 	check_and_free_fields(htab, l);
870 	bpf_mem_cache_free(&htab->ma, l);
871 }
872 
htab_put_fd_value(struct bpf_htab * htab,struct htab_elem * l)873 static void htab_put_fd_value(struct bpf_htab *htab, struct htab_elem *l)
874 {
875 	struct bpf_map *map = &htab->map;
876 	void *ptr;
877 
878 	if (map->ops->map_fd_put_ptr) {
879 		ptr = fd_htab_map_get_ptr(map, l);
880 		map->ops->map_fd_put_ptr(ptr);
881 	}
882 }
883 
is_map_full(struct bpf_htab * htab)884 static bool is_map_full(struct bpf_htab *htab)
885 {
886 	if (htab->use_percpu_counter)
887 		return __percpu_counter_compare(&htab->pcount, htab->map.max_entries,
888 						PERCPU_COUNTER_BATCH) >= 0;
889 	return atomic_read(&htab->count) >= htab->map.max_entries;
890 }
891 
inc_elem_count(struct bpf_htab * htab)892 static void inc_elem_count(struct bpf_htab *htab)
893 {
894 	if (htab->use_percpu_counter)
895 		percpu_counter_add_batch(&htab->pcount, 1, PERCPU_COUNTER_BATCH);
896 	else
897 		atomic_inc(&htab->count);
898 }
899 
dec_elem_count(struct bpf_htab * htab)900 static void dec_elem_count(struct bpf_htab *htab)
901 {
902 	if (htab->use_percpu_counter)
903 		percpu_counter_add_batch(&htab->pcount, -1, PERCPU_COUNTER_BATCH);
904 	else
905 		atomic_dec(&htab->count);
906 }
907 
908 
free_htab_elem(struct bpf_htab * htab,struct htab_elem * l)909 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
910 {
911 	htab_put_fd_value(htab, l);
912 
913 	if (htab_is_prealloc(htab)) {
914 		check_and_free_fields(htab, l);
915 		__pcpu_freelist_push(&htab->freelist, &l->fnode);
916 	} else {
917 		dec_elem_count(htab);
918 		htab_elem_free(htab, l);
919 	}
920 }
921 
pcpu_copy_value(struct bpf_htab * htab,void __percpu * pptr,void * value,bool onallcpus)922 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
923 			    void *value, bool onallcpus)
924 {
925 	if (!onallcpus) {
926 		/* copy true value_size bytes */
927 		memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
928 	} else {
929 		u32 size = round_up(htab->map.value_size, 8);
930 		int off = 0, cpu;
931 
932 		for_each_possible_cpu(cpu) {
933 			bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
934 					value + off, size);
935 			off += size;
936 		}
937 	}
938 }
939 
pcpu_init_value(struct bpf_htab * htab,void __percpu * pptr,void * value,bool onallcpus)940 static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
941 			    void *value, bool onallcpus)
942 {
943 	/* When not setting the initial value on all cpus, zero-fill element
944 	 * values for other cpus. Otherwise, bpf program has no way to ensure
945 	 * known initial values for cpus other than current one
946 	 * (onallcpus=false always when coming from bpf prog).
947 	 */
948 	if (!onallcpus) {
949 		u32 size = round_up(htab->map.value_size, 8);
950 		int current_cpu = raw_smp_processor_id();
951 		int cpu;
952 
953 		for_each_possible_cpu(cpu) {
954 			if (cpu == current_cpu)
955 				bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value,
956 						size);
957 			else
958 				memset(per_cpu_ptr(pptr, cpu), 0, size);
959 		}
960 	} else {
961 		pcpu_copy_value(htab, pptr, value, onallcpus);
962 	}
963 }
964 
fd_htab_map_needs_adjust(const struct bpf_htab * htab)965 static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
966 {
967 	return htab->map.map_type == BPF_MAP_TYPE_HASH_OF_MAPS &&
968 	       BITS_PER_LONG == 64;
969 }
970 
alloc_htab_elem(struct bpf_htab * htab,void * key,void * value,u32 key_size,u32 hash,bool percpu,bool onallcpus,struct htab_elem * old_elem)971 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
972 					 void *value, u32 key_size, u32 hash,
973 					 bool percpu, bool onallcpus,
974 					 struct htab_elem *old_elem)
975 {
976 	u32 size = htab->map.value_size;
977 	bool prealloc = htab_is_prealloc(htab);
978 	struct htab_elem *l_new, **pl_new;
979 	void __percpu *pptr;
980 
981 	if (prealloc) {
982 		if (old_elem) {
983 			/* if we're updating the existing element,
984 			 * use per-cpu extra elems to avoid freelist_pop/push
985 			 */
986 			pl_new = this_cpu_ptr(htab->extra_elems);
987 			l_new = *pl_new;
988 			htab_put_fd_value(htab, old_elem);
989 			*pl_new = old_elem;
990 		} else {
991 			struct pcpu_freelist_node *l;
992 
993 			l = __pcpu_freelist_pop(&htab->freelist);
994 			if (!l)
995 				return ERR_PTR(-E2BIG);
996 			l_new = container_of(l, struct htab_elem, fnode);
997 		}
998 	} else {
999 		if (is_map_full(htab))
1000 			if (!old_elem)
1001 				/* when map is full and update() is replacing
1002 				 * old element, it's ok to allocate, since
1003 				 * old element will be freed immediately.
1004 				 * Otherwise return an error
1005 				 */
1006 				return ERR_PTR(-E2BIG);
1007 		inc_elem_count(htab);
1008 		l_new = bpf_mem_cache_alloc(&htab->ma);
1009 		if (!l_new) {
1010 			l_new = ERR_PTR(-ENOMEM);
1011 			goto dec_count;
1012 		}
1013 		check_and_init_map_value(&htab->map,
1014 					 l_new->key + round_up(key_size, 8));
1015 	}
1016 
1017 	memcpy(l_new->key, key, key_size);
1018 	if (percpu) {
1019 		if (prealloc) {
1020 			pptr = htab_elem_get_ptr(l_new, key_size);
1021 		} else {
1022 			/* alloc_percpu zero-fills */
1023 			pptr = bpf_mem_cache_alloc(&htab->pcpu_ma);
1024 			if (!pptr) {
1025 				bpf_mem_cache_free(&htab->ma, l_new);
1026 				l_new = ERR_PTR(-ENOMEM);
1027 				goto dec_count;
1028 			}
1029 			l_new->ptr_to_pptr = pptr;
1030 			pptr = *(void **)pptr;
1031 		}
1032 
1033 		pcpu_init_value(htab, pptr, value, onallcpus);
1034 
1035 		if (!prealloc)
1036 			htab_elem_set_ptr(l_new, key_size, pptr);
1037 	} else if (fd_htab_map_needs_adjust(htab)) {
1038 		size = round_up(size, 8);
1039 		memcpy(l_new->key + round_up(key_size, 8), value, size);
1040 	} else {
1041 		copy_map_value(&htab->map,
1042 			       l_new->key + round_up(key_size, 8),
1043 			       value);
1044 	}
1045 
1046 	l_new->hash = hash;
1047 	return l_new;
1048 dec_count:
1049 	dec_elem_count(htab);
1050 	return l_new;
1051 }
1052 
check_flags(struct bpf_htab * htab,struct htab_elem * l_old,u64 map_flags)1053 static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
1054 		       u64 map_flags)
1055 {
1056 	if (l_old && (map_flags & ~BPF_F_LOCK) == BPF_NOEXIST)
1057 		/* elem already exists */
1058 		return -EEXIST;
1059 
1060 	if (!l_old && (map_flags & ~BPF_F_LOCK) == BPF_EXIST)
1061 		/* elem doesn't exist, cannot update it */
1062 		return -ENOENT;
1063 
1064 	return 0;
1065 }
1066 
1067 /* Called from syscall or from eBPF program */
htab_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)1068 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
1069 				u64 map_flags)
1070 {
1071 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1072 	struct htab_elem *l_new = NULL, *l_old;
1073 	struct hlist_nulls_head *head;
1074 	unsigned long flags;
1075 	struct bucket *b;
1076 	u32 key_size, hash;
1077 	int ret;
1078 
1079 	if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
1080 		/* unknown flags */
1081 		return -EINVAL;
1082 
1083 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1084 		     !rcu_read_lock_bh_held());
1085 
1086 	key_size = map->key_size;
1087 
1088 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1089 
1090 	b = __select_bucket(htab, hash);
1091 	head = &b->head;
1092 
1093 	if (unlikely(map_flags & BPF_F_LOCK)) {
1094 		if (unlikely(!map_value_has_spin_lock(map)))
1095 			return -EINVAL;
1096 		/* find an element without taking the bucket lock */
1097 		l_old = lookup_nulls_elem_raw(head, hash, key, key_size,
1098 					      htab->n_buckets);
1099 		ret = check_flags(htab, l_old, map_flags);
1100 		if (ret)
1101 			return ret;
1102 		if (l_old) {
1103 			/* grab the element lock and update value in place */
1104 			copy_map_value_locked(map,
1105 					      l_old->key + round_up(key_size, 8),
1106 					      value, false);
1107 			return 0;
1108 		}
1109 		/* fall through, grab the bucket lock and lookup again.
1110 		 * 99.9% chance that the element won't be found,
1111 		 * but second lookup under lock has to be done.
1112 		 */
1113 	}
1114 
1115 	ret = htab_lock_bucket(htab, b, hash, &flags);
1116 	if (ret)
1117 		return ret;
1118 
1119 	l_old = lookup_elem_raw(head, hash, key, key_size);
1120 
1121 	ret = check_flags(htab, l_old, map_flags);
1122 	if (ret)
1123 		goto err;
1124 
1125 	if (unlikely(l_old && (map_flags & BPF_F_LOCK))) {
1126 		/* first lookup without the bucket lock didn't find the element,
1127 		 * but second lookup with the bucket lock found it.
1128 		 * This case is highly unlikely, but has to be dealt with:
1129 		 * grab the element lock in addition to the bucket lock
1130 		 * and update element in place
1131 		 */
1132 		copy_map_value_locked(map,
1133 				      l_old->key + round_up(key_size, 8),
1134 				      value, false);
1135 		ret = 0;
1136 		goto err;
1137 	}
1138 
1139 	l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
1140 				l_old);
1141 	if (IS_ERR(l_new)) {
1142 		/* all pre-allocated elements are in use or memory exhausted */
1143 		ret = PTR_ERR(l_new);
1144 		goto err;
1145 	}
1146 
1147 	/* add new element to the head of the list, so that
1148 	 * concurrent search will find it before old elem
1149 	 */
1150 	hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1151 	if (l_old) {
1152 		hlist_nulls_del_rcu(&l_old->hash_node);
1153 		if (!htab_is_prealloc(htab))
1154 			free_htab_elem(htab, l_old);
1155 		else
1156 			check_and_free_fields(htab, l_old);
1157 	}
1158 	ret = 0;
1159 err:
1160 	htab_unlock_bucket(htab, b, hash, flags);
1161 	return ret;
1162 }
1163 
htab_lru_push_free(struct bpf_htab * htab,struct htab_elem * elem)1164 static void htab_lru_push_free(struct bpf_htab *htab, struct htab_elem *elem)
1165 {
1166 	check_and_free_fields(htab, elem);
1167 	bpf_lru_push_free(&htab->lru, &elem->lru_node);
1168 }
1169 
htab_lru_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)1170 static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
1171 				    u64 map_flags)
1172 {
1173 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1174 	struct htab_elem *l_new, *l_old = NULL;
1175 	struct hlist_nulls_head *head;
1176 	unsigned long flags;
1177 	struct bucket *b;
1178 	u32 key_size, hash;
1179 	int ret;
1180 
1181 	if (unlikely(map_flags > BPF_EXIST))
1182 		/* unknown flags */
1183 		return -EINVAL;
1184 
1185 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1186 		     !rcu_read_lock_bh_held());
1187 
1188 	key_size = map->key_size;
1189 
1190 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1191 
1192 	b = __select_bucket(htab, hash);
1193 	head = &b->head;
1194 
1195 	/* For LRU, we need to alloc before taking bucket's
1196 	 * spinlock because getting free nodes from LRU may need
1197 	 * to remove older elements from htab and this removal
1198 	 * operation will need a bucket lock.
1199 	 */
1200 	l_new = prealloc_lru_pop(htab, key, hash);
1201 	if (!l_new)
1202 		return -ENOMEM;
1203 	copy_map_value(&htab->map,
1204 		       l_new->key + round_up(map->key_size, 8), value);
1205 
1206 	ret = htab_lock_bucket(htab, b, hash, &flags);
1207 	if (ret)
1208 		return ret;
1209 
1210 	l_old = lookup_elem_raw(head, hash, key, key_size);
1211 
1212 	ret = check_flags(htab, l_old, map_flags);
1213 	if (ret)
1214 		goto err;
1215 
1216 	/* add new element to the head of the list, so that
1217 	 * concurrent search will find it before old elem
1218 	 */
1219 	hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1220 	if (l_old) {
1221 		bpf_lru_node_set_ref(&l_new->lru_node);
1222 		hlist_nulls_del_rcu(&l_old->hash_node);
1223 	}
1224 	ret = 0;
1225 
1226 err:
1227 	htab_unlock_bucket(htab, b, hash, flags);
1228 
1229 	if (ret)
1230 		htab_lru_push_free(htab, l_new);
1231 	else if (l_old)
1232 		htab_lru_push_free(htab, l_old);
1233 
1234 	return ret;
1235 }
1236 
__htab_percpu_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags,bool onallcpus)1237 static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1238 					 void *value, u64 map_flags,
1239 					 bool onallcpus)
1240 {
1241 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1242 	struct htab_elem *l_new = NULL, *l_old;
1243 	struct hlist_nulls_head *head;
1244 	unsigned long flags;
1245 	struct bucket *b;
1246 	u32 key_size, hash;
1247 	int ret;
1248 
1249 	if (unlikely(map_flags > BPF_EXIST))
1250 		/* unknown flags */
1251 		return -EINVAL;
1252 
1253 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1254 		     !rcu_read_lock_bh_held());
1255 
1256 	key_size = map->key_size;
1257 
1258 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1259 
1260 	b = __select_bucket(htab, hash);
1261 	head = &b->head;
1262 
1263 	ret = htab_lock_bucket(htab, b, hash, &flags);
1264 	if (ret)
1265 		return ret;
1266 
1267 	l_old = lookup_elem_raw(head, hash, key, key_size);
1268 
1269 	ret = check_flags(htab, l_old, map_flags);
1270 	if (ret)
1271 		goto err;
1272 
1273 	if (l_old) {
1274 		/* per-cpu hash map can update value in-place */
1275 		pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1276 				value, onallcpus);
1277 	} else {
1278 		l_new = alloc_htab_elem(htab, key, value, key_size,
1279 					hash, true, onallcpus, NULL);
1280 		if (IS_ERR(l_new)) {
1281 			ret = PTR_ERR(l_new);
1282 			goto err;
1283 		}
1284 		hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1285 	}
1286 	ret = 0;
1287 err:
1288 	htab_unlock_bucket(htab, b, hash, flags);
1289 	return ret;
1290 }
1291 
__htab_lru_percpu_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags,bool onallcpus)1292 static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1293 					     void *value, u64 map_flags,
1294 					     bool onallcpus)
1295 {
1296 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1297 	struct htab_elem *l_new = NULL, *l_old;
1298 	struct hlist_nulls_head *head;
1299 	unsigned long flags;
1300 	struct bucket *b;
1301 	u32 key_size, hash;
1302 	int ret;
1303 
1304 	if (unlikely(map_flags > BPF_EXIST))
1305 		/* unknown flags */
1306 		return -EINVAL;
1307 
1308 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1309 		     !rcu_read_lock_bh_held());
1310 
1311 	key_size = map->key_size;
1312 
1313 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1314 
1315 	b = __select_bucket(htab, hash);
1316 	head = &b->head;
1317 
1318 	/* For LRU, we need to alloc before taking bucket's
1319 	 * spinlock because LRU's elem alloc may need
1320 	 * to remove older elem from htab and this removal
1321 	 * operation will need a bucket lock.
1322 	 */
1323 	if (map_flags != BPF_EXIST) {
1324 		l_new = prealloc_lru_pop(htab, key, hash);
1325 		if (!l_new)
1326 			return -ENOMEM;
1327 	}
1328 
1329 	ret = htab_lock_bucket(htab, b, hash, &flags);
1330 	if (ret)
1331 		return ret;
1332 
1333 	l_old = lookup_elem_raw(head, hash, key, key_size);
1334 
1335 	ret = check_flags(htab, l_old, map_flags);
1336 	if (ret)
1337 		goto err;
1338 
1339 	if (l_old) {
1340 		bpf_lru_node_set_ref(&l_old->lru_node);
1341 
1342 		/* per-cpu hash map can update value in-place */
1343 		pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1344 				value, onallcpus);
1345 	} else {
1346 		pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
1347 				value, onallcpus);
1348 		hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1349 		l_new = NULL;
1350 	}
1351 	ret = 0;
1352 err:
1353 	htab_unlock_bucket(htab, b, hash, flags);
1354 	if (l_new)
1355 		bpf_lru_push_free(&htab->lru, &l_new->lru_node);
1356 	return ret;
1357 }
1358 
htab_percpu_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)1359 static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1360 				       void *value, u64 map_flags)
1361 {
1362 	return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
1363 }
1364 
htab_lru_percpu_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)1365 static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1366 					   void *value, u64 map_flags)
1367 {
1368 	return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
1369 						 false);
1370 }
1371 
1372 /* Called from syscall or from eBPF program */
htab_map_delete_elem(struct bpf_map * map,void * key)1373 static int htab_map_delete_elem(struct bpf_map *map, void *key)
1374 {
1375 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1376 	struct hlist_nulls_head *head;
1377 	struct bucket *b;
1378 	struct htab_elem *l;
1379 	unsigned long flags;
1380 	u32 hash, key_size;
1381 	int ret;
1382 
1383 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1384 		     !rcu_read_lock_bh_held());
1385 
1386 	key_size = map->key_size;
1387 
1388 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1389 	b = __select_bucket(htab, hash);
1390 	head = &b->head;
1391 
1392 	ret = htab_lock_bucket(htab, b, hash, &flags);
1393 	if (ret)
1394 		return ret;
1395 
1396 	l = lookup_elem_raw(head, hash, key, key_size);
1397 
1398 	if (l) {
1399 		hlist_nulls_del_rcu(&l->hash_node);
1400 		free_htab_elem(htab, l);
1401 	} else {
1402 		ret = -ENOENT;
1403 	}
1404 
1405 	htab_unlock_bucket(htab, b, hash, flags);
1406 	return ret;
1407 }
1408 
htab_lru_map_delete_elem(struct bpf_map * map,void * key)1409 static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
1410 {
1411 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1412 	struct hlist_nulls_head *head;
1413 	struct bucket *b;
1414 	struct htab_elem *l;
1415 	unsigned long flags;
1416 	u32 hash, key_size;
1417 	int ret;
1418 
1419 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1420 		     !rcu_read_lock_bh_held());
1421 
1422 	key_size = map->key_size;
1423 
1424 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1425 	b = __select_bucket(htab, hash);
1426 	head = &b->head;
1427 
1428 	ret = htab_lock_bucket(htab, b, hash, &flags);
1429 	if (ret)
1430 		return ret;
1431 
1432 	l = lookup_elem_raw(head, hash, key, key_size);
1433 
1434 	if (l)
1435 		hlist_nulls_del_rcu(&l->hash_node);
1436 	else
1437 		ret = -ENOENT;
1438 
1439 	htab_unlock_bucket(htab, b, hash, flags);
1440 	if (l)
1441 		htab_lru_push_free(htab, l);
1442 	return ret;
1443 }
1444 
delete_all_elements(struct bpf_htab * htab)1445 static void delete_all_elements(struct bpf_htab *htab)
1446 {
1447 	int i;
1448 
1449 	/* It's called from a worker thread, so disable migration here,
1450 	 * since bpf_mem_cache_free() relies on that.
1451 	 */
1452 	migrate_disable();
1453 	for (i = 0; i < htab->n_buckets; i++) {
1454 		struct hlist_nulls_head *head = select_bucket(htab, i);
1455 		struct hlist_nulls_node *n;
1456 		struct htab_elem *l;
1457 
1458 		hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1459 			hlist_nulls_del_rcu(&l->hash_node);
1460 			htab_elem_free(htab, l);
1461 		}
1462 	}
1463 	migrate_enable();
1464 }
1465 
htab_free_malloced_timers(struct bpf_htab * htab)1466 static void htab_free_malloced_timers(struct bpf_htab *htab)
1467 {
1468 	int i;
1469 
1470 	rcu_read_lock();
1471 	for (i = 0; i < htab->n_buckets; i++) {
1472 		struct hlist_nulls_head *head = select_bucket(htab, i);
1473 		struct hlist_nulls_node *n;
1474 		struct htab_elem *l;
1475 
1476 		hlist_nulls_for_each_entry(l, n, head, hash_node) {
1477 			/* We don't reset or free kptr on uref dropping to zero,
1478 			 * hence just free timer.
1479 			 */
1480 			bpf_timer_cancel_and_free(l->key +
1481 						  round_up(htab->map.key_size, 8) +
1482 						  htab->map.timer_off);
1483 		}
1484 		cond_resched_rcu();
1485 	}
1486 	rcu_read_unlock();
1487 }
1488 
htab_map_free_timers(struct bpf_map * map)1489 static void htab_map_free_timers(struct bpf_map *map)
1490 {
1491 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1492 
1493 	/* We don't reset or free kptr on uref dropping to zero. */
1494 	if (!map_value_has_timer(&htab->map))
1495 		return;
1496 	if (!htab_is_prealloc(htab))
1497 		htab_free_malloced_timers(htab);
1498 	else
1499 		htab_free_prealloced_timers(htab);
1500 }
1501 
1502 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
htab_map_free(struct bpf_map * map)1503 static void htab_map_free(struct bpf_map *map)
1504 {
1505 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1506 	int i;
1507 
1508 	/* bpf_free_used_maps() or close(map_fd) will trigger this map_free callback.
1509 	 * bpf_free_used_maps() is called after bpf prog is no longer executing.
1510 	 * There is no need to synchronize_rcu() here to protect map elements.
1511 	 */
1512 
1513 	/* htab no longer uses call_rcu() directly. bpf_mem_alloc does it
1514 	 * underneath and is reponsible for waiting for callbacks to finish
1515 	 * during bpf_mem_alloc_destroy().
1516 	 */
1517 	if (!htab_is_prealloc(htab)) {
1518 		delete_all_elements(htab);
1519 	} else {
1520 		htab_free_prealloced_kptrs(htab);
1521 		prealloc_destroy(htab);
1522 	}
1523 
1524 	bpf_map_free_kptr_off_tab(map);
1525 	free_percpu(htab->extra_elems);
1526 	bpf_map_area_free(htab->buckets);
1527 	bpf_mem_alloc_destroy(&htab->pcpu_ma);
1528 	bpf_mem_alloc_destroy(&htab->ma);
1529 	if (htab->use_percpu_counter)
1530 		percpu_counter_destroy(&htab->pcount);
1531 	for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
1532 		free_percpu(htab->map_locked[i]);
1533 	lockdep_unregister_key(&htab->lockdep_key);
1534 	bpf_map_area_free(htab);
1535 }
1536 
htab_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)1537 static void htab_map_seq_show_elem(struct bpf_map *map, void *key,
1538 				   struct seq_file *m)
1539 {
1540 	void *value;
1541 
1542 	rcu_read_lock();
1543 
1544 	value = htab_map_lookup_elem(map, key);
1545 	if (!value) {
1546 		rcu_read_unlock();
1547 		return;
1548 	}
1549 
1550 	btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
1551 	seq_puts(m, ": ");
1552 	btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
1553 	seq_puts(m, "\n");
1554 
1555 	rcu_read_unlock();
1556 }
1557 
__htab_map_lookup_and_delete_elem(struct bpf_map * map,void * key,void * value,bool is_lru_map,bool is_percpu,u64 flags)1558 static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1559 					     void *value, bool is_lru_map,
1560 					     bool is_percpu, u64 flags)
1561 {
1562 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1563 	struct hlist_nulls_head *head;
1564 	unsigned long bflags;
1565 	struct htab_elem *l;
1566 	u32 hash, key_size;
1567 	struct bucket *b;
1568 	int ret;
1569 
1570 	key_size = map->key_size;
1571 
1572 	hash = htab_map_hash(key, key_size, htab->hashrnd);
1573 	b = __select_bucket(htab, hash);
1574 	head = &b->head;
1575 
1576 	ret = htab_lock_bucket(htab, b, hash, &bflags);
1577 	if (ret)
1578 		return ret;
1579 
1580 	l = lookup_elem_raw(head, hash, key, key_size);
1581 	if (!l) {
1582 		ret = -ENOENT;
1583 	} else {
1584 		if (is_percpu) {
1585 			u32 roundup_value_size = round_up(map->value_size, 8);
1586 			void __percpu *pptr;
1587 			int off = 0, cpu;
1588 
1589 			pptr = htab_elem_get_ptr(l, key_size);
1590 			for_each_possible_cpu(cpu) {
1591 				bpf_long_memcpy(value + off,
1592 						per_cpu_ptr(pptr, cpu),
1593 						roundup_value_size);
1594 				off += roundup_value_size;
1595 			}
1596 		} else {
1597 			u32 roundup_key_size = round_up(map->key_size, 8);
1598 
1599 			if (flags & BPF_F_LOCK)
1600 				copy_map_value_locked(map, value, l->key +
1601 						      roundup_key_size,
1602 						      true);
1603 			else
1604 				copy_map_value(map, value, l->key +
1605 					       roundup_key_size);
1606 			check_and_init_map_value(map, value);
1607 		}
1608 
1609 		hlist_nulls_del_rcu(&l->hash_node);
1610 		if (!is_lru_map)
1611 			free_htab_elem(htab, l);
1612 	}
1613 
1614 	htab_unlock_bucket(htab, b, hash, bflags);
1615 
1616 	if (is_lru_map && l)
1617 		htab_lru_push_free(htab, l);
1618 
1619 	return ret;
1620 }
1621 
htab_map_lookup_and_delete_elem(struct bpf_map * map,void * key,void * value,u64 flags)1622 static int htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1623 					   void *value, u64 flags)
1624 {
1625 	return __htab_map_lookup_and_delete_elem(map, key, value, false, false,
1626 						 flags);
1627 }
1628 
htab_percpu_map_lookup_and_delete_elem(struct bpf_map * map,void * key,void * value,u64 flags)1629 static int htab_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1630 						  void *key, void *value,
1631 						  u64 flags)
1632 {
1633 	return __htab_map_lookup_and_delete_elem(map, key, value, false, true,
1634 						 flags);
1635 }
1636 
htab_lru_map_lookup_and_delete_elem(struct bpf_map * map,void * key,void * value,u64 flags)1637 static int htab_lru_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1638 					       void *value, u64 flags)
1639 {
1640 	return __htab_map_lookup_and_delete_elem(map, key, value, true, false,
1641 						 flags);
1642 }
1643 
htab_lru_percpu_map_lookup_and_delete_elem(struct bpf_map * map,void * key,void * value,u64 flags)1644 static int htab_lru_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1645 						      void *key, void *value,
1646 						      u64 flags)
1647 {
1648 	return __htab_map_lookup_and_delete_elem(map, key, value, true, true,
1649 						 flags);
1650 }
1651 
1652 static int
__htab_map_lookup_and_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr,bool do_delete,bool is_lru_map,bool is_percpu)1653 __htab_map_lookup_and_delete_batch(struct bpf_map *map,
1654 				   const union bpf_attr *attr,
1655 				   union bpf_attr __user *uattr,
1656 				   bool do_delete, bool is_lru_map,
1657 				   bool is_percpu)
1658 {
1659 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1660 	u32 bucket_cnt, total, key_size, value_size, roundup_key_size;
1661 	void *keys = NULL, *values = NULL, *value, *dst_key, *dst_val;
1662 	void __user *uvalues = u64_to_user_ptr(attr->batch.values);
1663 	void __user *ukeys = u64_to_user_ptr(attr->batch.keys);
1664 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1665 	u32 batch, max_count, size, bucket_size, map_id;
1666 	struct htab_elem *node_to_free = NULL;
1667 	u64 elem_map_flags, map_flags;
1668 	struct hlist_nulls_head *head;
1669 	struct hlist_nulls_node *n;
1670 	unsigned long flags = 0;
1671 	bool locked = false;
1672 	struct htab_elem *l;
1673 	struct bucket *b;
1674 	int ret = 0;
1675 
1676 	elem_map_flags = attr->batch.elem_flags;
1677 	if ((elem_map_flags & ~BPF_F_LOCK) ||
1678 	    ((elem_map_flags & BPF_F_LOCK) && !map_value_has_spin_lock(map)))
1679 		return -EINVAL;
1680 
1681 	map_flags = attr->batch.flags;
1682 	if (map_flags)
1683 		return -EINVAL;
1684 
1685 	max_count = attr->batch.count;
1686 	if (!max_count)
1687 		return 0;
1688 
1689 	if (put_user(0, &uattr->batch.count))
1690 		return -EFAULT;
1691 
1692 	batch = 0;
1693 	if (ubatch && copy_from_user(&batch, ubatch, sizeof(batch)))
1694 		return -EFAULT;
1695 
1696 	if (batch >= htab->n_buckets)
1697 		return -ENOENT;
1698 
1699 	key_size = htab->map.key_size;
1700 	roundup_key_size = round_up(htab->map.key_size, 8);
1701 	value_size = htab->map.value_size;
1702 	size = round_up(value_size, 8);
1703 	if (is_percpu)
1704 		value_size = size * num_possible_cpus();
1705 	total = 0;
1706 	/* while experimenting with hash tables with sizes ranging from 10 to
1707 	 * 1000, it was observed that a bucket can have up to 5 entries.
1708 	 */
1709 	bucket_size = 5;
1710 
1711 alloc:
1712 	/* We cannot do copy_from_user or copy_to_user inside
1713 	 * the rcu_read_lock. Allocate enough space here.
1714 	 */
1715 	keys = kvmalloc_array(key_size, bucket_size, GFP_USER | __GFP_NOWARN);
1716 	values = kvmalloc_array(value_size, bucket_size, GFP_USER | __GFP_NOWARN);
1717 	if (!keys || !values) {
1718 		ret = -ENOMEM;
1719 		goto after_loop;
1720 	}
1721 
1722 again:
1723 	bpf_disable_instrumentation();
1724 	rcu_read_lock();
1725 again_nocopy:
1726 	dst_key = keys;
1727 	dst_val = values;
1728 	b = &htab->buckets[batch];
1729 	head = &b->head;
1730 	/* do not grab the lock unless need it (bucket_cnt > 0). */
1731 	if (locked) {
1732 		ret = htab_lock_bucket(htab, b, batch, &flags);
1733 		if (ret) {
1734 			rcu_read_unlock();
1735 			bpf_enable_instrumentation();
1736 			goto after_loop;
1737 		}
1738 	}
1739 
1740 	bucket_cnt = 0;
1741 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
1742 		bucket_cnt++;
1743 
1744 	if (bucket_cnt && !locked) {
1745 		locked = true;
1746 		goto again_nocopy;
1747 	}
1748 
1749 	if (bucket_cnt > (max_count - total)) {
1750 		if (total == 0)
1751 			ret = -ENOSPC;
1752 		/* Note that since bucket_cnt > 0 here, it is implicit
1753 		 * that the locked was grabbed, so release it.
1754 		 */
1755 		htab_unlock_bucket(htab, b, batch, flags);
1756 		rcu_read_unlock();
1757 		bpf_enable_instrumentation();
1758 		goto after_loop;
1759 	}
1760 
1761 	if (bucket_cnt > bucket_size) {
1762 		bucket_size = bucket_cnt;
1763 		/* Note that since bucket_cnt > 0 here, it is implicit
1764 		 * that the locked was grabbed, so release it.
1765 		 */
1766 		htab_unlock_bucket(htab, b, batch, flags);
1767 		rcu_read_unlock();
1768 		bpf_enable_instrumentation();
1769 		kvfree(keys);
1770 		kvfree(values);
1771 		goto alloc;
1772 	}
1773 
1774 	/* Next block is only safe to run if you have grabbed the lock */
1775 	if (!locked)
1776 		goto next_batch;
1777 
1778 	hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1779 		memcpy(dst_key, l->key, key_size);
1780 
1781 		if (is_percpu) {
1782 			int off = 0, cpu;
1783 			void __percpu *pptr;
1784 
1785 			pptr = htab_elem_get_ptr(l, map->key_size);
1786 			for_each_possible_cpu(cpu) {
1787 				bpf_long_memcpy(dst_val + off,
1788 						per_cpu_ptr(pptr, cpu), size);
1789 				off += size;
1790 			}
1791 		} else {
1792 			value = l->key + roundup_key_size;
1793 			if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1794 				struct bpf_map **inner_map = value;
1795 
1796 				 /* Actual value is the id of the inner map */
1797 				map_id = map->ops->map_fd_sys_lookup_elem(*inner_map);
1798 				value = &map_id;
1799 			}
1800 
1801 			if (elem_map_flags & BPF_F_LOCK)
1802 				copy_map_value_locked(map, dst_val, value,
1803 						      true);
1804 			else
1805 				copy_map_value(map, dst_val, value);
1806 			check_and_init_map_value(map, dst_val);
1807 		}
1808 		if (do_delete) {
1809 			hlist_nulls_del_rcu(&l->hash_node);
1810 
1811 			/* bpf_lru_push_free() will acquire lru_lock, which
1812 			 * may cause deadlock. See comments in function
1813 			 * prealloc_lru_pop(). Let us do bpf_lru_push_free()
1814 			 * after releasing the bucket lock.
1815 			 */
1816 			if (is_lru_map) {
1817 				l->batch_flink = node_to_free;
1818 				node_to_free = l;
1819 			} else {
1820 				free_htab_elem(htab, l);
1821 			}
1822 		}
1823 		dst_key += key_size;
1824 		dst_val += value_size;
1825 	}
1826 
1827 	htab_unlock_bucket(htab, b, batch, flags);
1828 	locked = false;
1829 
1830 	while (node_to_free) {
1831 		l = node_to_free;
1832 		node_to_free = node_to_free->batch_flink;
1833 		htab_lru_push_free(htab, l);
1834 	}
1835 
1836 next_batch:
1837 	/* If we are not copying data, we can go to next bucket and avoid
1838 	 * unlocking the rcu.
1839 	 */
1840 	if (!bucket_cnt && (batch + 1 < htab->n_buckets)) {
1841 		batch++;
1842 		goto again_nocopy;
1843 	}
1844 
1845 	rcu_read_unlock();
1846 	bpf_enable_instrumentation();
1847 	if (bucket_cnt && (copy_to_user(ukeys + total * key_size, keys,
1848 	    key_size * bucket_cnt) ||
1849 	    copy_to_user(uvalues + total * value_size, values,
1850 	    value_size * bucket_cnt))) {
1851 		ret = -EFAULT;
1852 		goto after_loop;
1853 	}
1854 
1855 	total += bucket_cnt;
1856 	batch++;
1857 	if (batch >= htab->n_buckets) {
1858 		ret = -ENOENT;
1859 		goto after_loop;
1860 	}
1861 	goto again;
1862 
1863 after_loop:
1864 	if (ret == -EFAULT)
1865 		goto out;
1866 
1867 	/* copy # of entries and next batch */
1868 	ubatch = u64_to_user_ptr(attr->batch.out_batch);
1869 	if (copy_to_user(ubatch, &batch, sizeof(batch)) ||
1870 	    put_user(total, &uattr->batch.count))
1871 		ret = -EFAULT;
1872 
1873 out:
1874 	kvfree(keys);
1875 	kvfree(values);
1876 	return ret;
1877 }
1878 
1879 static int
htab_percpu_map_lookup_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1880 htab_percpu_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1881 			     union bpf_attr __user *uattr)
1882 {
1883 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1884 						  false, true);
1885 }
1886 
1887 static int
htab_percpu_map_lookup_and_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1888 htab_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1889 					const union bpf_attr *attr,
1890 					union bpf_attr __user *uattr)
1891 {
1892 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1893 						  false, true);
1894 }
1895 
1896 static int
htab_map_lookup_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1897 htab_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1898 		      union bpf_attr __user *uattr)
1899 {
1900 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1901 						  false, false);
1902 }
1903 
1904 static int
htab_map_lookup_and_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1905 htab_map_lookup_and_delete_batch(struct bpf_map *map,
1906 				 const union bpf_attr *attr,
1907 				 union bpf_attr __user *uattr)
1908 {
1909 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1910 						  false, false);
1911 }
1912 
1913 static int
htab_lru_percpu_map_lookup_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1914 htab_lru_percpu_map_lookup_batch(struct bpf_map *map,
1915 				 const union bpf_attr *attr,
1916 				 union bpf_attr __user *uattr)
1917 {
1918 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1919 						  true, true);
1920 }
1921 
1922 static int
htab_lru_percpu_map_lookup_and_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1923 htab_lru_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1924 					    const union bpf_attr *attr,
1925 					    union bpf_attr __user *uattr)
1926 {
1927 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1928 						  true, true);
1929 }
1930 
1931 static int
htab_lru_map_lookup_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1932 htab_lru_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1933 			  union bpf_attr __user *uattr)
1934 {
1935 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1936 						  true, false);
1937 }
1938 
1939 static int
htab_lru_map_lookup_and_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1940 htab_lru_map_lookup_and_delete_batch(struct bpf_map *map,
1941 				     const union bpf_attr *attr,
1942 				     union bpf_attr __user *uattr)
1943 {
1944 	return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1945 						  true, false);
1946 }
1947 
1948 struct bpf_iter_seq_hash_map_info {
1949 	struct bpf_map *map;
1950 	struct bpf_htab *htab;
1951 	void *percpu_value_buf; // non-zero means percpu hash
1952 	u32 bucket_id;
1953 	u32 skip_elems;
1954 };
1955 
1956 static struct htab_elem *
bpf_hash_map_seq_find_next(struct bpf_iter_seq_hash_map_info * info,struct htab_elem * prev_elem)1957 bpf_hash_map_seq_find_next(struct bpf_iter_seq_hash_map_info *info,
1958 			   struct htab_elem *prev_elem)
1959 {
1960 	const struct bpf_htab *htab = info->htab;
1961 	u32 skip_elems = info->skip_elems;
1962 	u32 bucket_id = info->bucket_id;
1963 	struct hlist_nulls_head *head;
1964 	struct hlist_nulls_node *n;
1965 	struct htab_elem *elem;
1966 	struct bucket *b;
1967 	u32 i, count;
1968 
1969 	if (bucket_id >= htab->n_buckets)
1970 		return NULL;
1971 
1972 	/* try to find next elem in the same bucket */
1973 	if (prev_elem) {
1974 		/* no update/deletion on this bucket, prev_elem should be still valid
1975 		 * and we won't skip elements.
1976 		 */
1977 		n = rcu_dereference_raw(hlist_nulls_next_rcu(&prev_elem->hash_node));
1978 		elem = hlist_nulls_entry_safe(n, struct htab_elem, hash_node);
1979 		if (elem)
1980 			return elem;
1981 
1982 		/* not found, unlock and go to the next bucket */
1983 		b = &htab->buckets[bucket_id++];
1984 		rcu_read_unlock();
1985 		skip_elems = 0;
1986 	}
1987 
1988 	for (i = bucket_id; i < htab->n_buckets; i++) {
1989 		b = &htab->buckets[i];
1990 		rcu_read_lock();
1991 
1992 		count = 0;
1993 		head = &b->head;
1994 		hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
1995 			if (count >= skip_elems) {
1996 				info->bucket_id = i;
1997 				info->skip_elems = count;
1998 				return elem;
1999 			}
2000 			count++;
2001 		}
2002 
2003 		rcu_read_unlock();
2004 		skip_elems = 0;
2005 	}
2006 
2007 	info->bucket_id = i;
2008 	info->skip_elems = 0;
2009 	return NULL;
2010 }
2011 
bpf_hash_map_seq_start(struct seq_file * seq,loff_t * pos)2012 static void *bpf_hash_map_seq_start(struct seq_file *seq, loff_t *pos)
2013 {
2014 	struct bpf_iter_seq_hash_map_info *info = seq->private;
2015 	struct htab_elem *elem;
2016 
2017 	elem = bpf_hash_map_seq_find_next(info, NULL);
2018 	if (!elem)
2019 		return NULL;
2020 
2021 	if (*pos == 0)
2022 		++*pos;
2023 	return elem;
2024 }
2025 
bpf_hash_map_seq_next(struct seq_file * seq,void * v,loff_t * pos)2026 static void *bpf_hash_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2027 {
2028 	struct bpf_iter_seq_hash_map_info *info = seq->private;
2029 
2030 	++*pos;
2031 	++info->skip_elems;
2032 	return bpf_hash_map_seq_find_next(info, v);
2033 }
2034 
__bpf_hash_map_seq_show(struct seq_file * seq,struct htab_elem * elem)2035 static int __bpf_hash_map_seq_show(struct seq_file *seq, struct htab_elem *elem)
2036 {
2037 	struct bpf_iter_seq_hash_map_info *info = seq->private;
2038 	u32 roundup_key_size, roundup_value_size;
2039 	struct bpf_iter__bpf_map_elem ctx = {};
2040 	struct bpf_map *map = info->map;
2041 	struct bpf_iter_meta meta;
2042 	int ret = 0, off = 0, cpu;
2043 	struct bpf_prog *prog;
2044 	void __percpu *pptr;
2045 
2046 	meta.seq = seq;
2047 	prog = bpf_iter_get_info(&meta, elem == NULL);
2048 	if (prog) {
2049 		ctx.meta = &meta;
2050 		ctx.map = info->map;
2051 		if (elem) {
2052 			roundup_key_size = round_up(map->key_size, 8);
2053 			ctx.key = elem->key;
2054 			if (!info->percpu_value_buf) {
2055 				ctx.value = elem->key + roundup_key_size;
2056 			} else {
2057 				roundup_value_size = round_up(map->value_size, 8);
2058 				pptr = htab_elem_get_ptr(elem, map->key_size);
2059 				for_each_possible_cpu(cpu) {
2060 					bpf_long_memcpy(info->percpu_value_buf + off,
2061 							per_cpu_ptr(pptr, cpu),
2062 							roundup_value_size);
2063 					off += roundup_value_size;
2064 				}
2065 				ctx.value = info->percpu_value_buf;
2066 			}
2067 		}
2068 		ret = bpf_iter_run_prog(prog, &ctx);
2069 	}
2070 
2071 	return ret;
2072 }
2073 
bpf_hash_map_seq_show(struct seq_file * seq,void * v)2074 static int bpf_hash_map_seq_show(struct seq_file *seq, void *v)
2075 {
2076 	return __bpf_hash_map_seq_show(seq, v);
2077 }
2078 
bpf_hash_map_seq_stop(struct seq_file * seq,void * v)2079 static void bpf_hash_map_seq_stop(struct seq_file *seq, void *v)
2080 {
2081 	if (!v)
2082 		(void)__bpf_hash_map_seq_show(seq, NULL);
2083 	else
2084 		rcu_read_unlock();
2085 }
2086 
bpf_iter_init_hash_map(void * priv_data,struct bpf_iter_aux_info * aux)2087 static int bpf_iter_init_hash_map(void *priv_data,
2088 				  struct bpf_iter_aux_info *aux)
2089 {
2090 	struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2091 	struct bpf_map *map = aux->map;
2092 	void *value_buf;
2093 	u32 buf_size;
2094 
2095 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2096 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2097 		buf_size = round_up(map->value_size, 8) * num_possible_cpus();
2098 		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
2099 		if (!value_buf)
2100 			return -ENOMEM;
2101 
2102 		seq_info->percpu_value_buf = value_buf;
2103 	}
2104 
2105 	bpf_map_inc_with_uref(map);
2106 	seq_info->map = map;
2107 	seq_info->htab = container_of(map, struct bpf_htab, map);
2108 	return 0;
2109 }
2110 
bpf_iter_fini_hash_map(void * priv_data)2111 static void bpf_iter_fini_hash_map(void *priv_data)
2112 {
2113 	struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2114 
2115 	bpf_map_put_with_uref(seq_info->map);
2116 	kfree(seq_info->percpu_value_buf);
2117 }
2118 
2119 static const struct seq_operations bpf_hash_map_seq_ops = {
2120 	.start	= bpf_hash_map_seq_start,
2121 	.next	= bpf_hash_map_seq_next,
2122 	.stop	= bpf_hash_map_seq_stop,
2123 	.show	= bpf_hash_map_seq_show,
2124 };
2125 
2126 static const struct bpf_iter_seq_info iter_seq_info = {
2127 	.seq_ops		= &bpf_hash_map_seq_ops,
2128 	.init_seq_private	= bpf_iter_init_hash_map,
2129 	.fini_seq_private	= bpf_iter_fini_hash_map,
2130 	.seq_priv_size		= sizeof(struct bpf_iter_seq_hash_map_info),
2131 };
2132 
bpf_for_each_hash_elem(struct bpf_map * map,bpf_callback_t callback_fn,void * callback_ctx,u64 flags)2133 static int bpf_for_each_hash_elem(struct bpf_map *map, bpf_callback_t callback_fn,
2134 				  void *callback_ctx, u64 flags)
2135 {
2136 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2137 	struct hlist_nulls_head *head;
2138 	struct hlist_nulls_node *n;
2139 	struct htab_elem *elem;
2140 	u32 roundup_key_size;
2141 	int i, num_elems = 0;
2142 	void __percpu *pptr;
2143 	struct bucket *b;
2144 	void *key, *val;
2145 	bool is_percpu;
2146 	u64 ret = 0;
2147 
2148 	if (flags != 0)
2149 		return -EINVAL;
2150 
2151 	is_percpu = htab_is_percpu(htab);
2152 
2153 	roundup_key_size = round_up(map->key_size, 8);
2154 	/* disable migration so percpu value prepared here will be the
2155 	 * same as the one seen by the bpf program with bpf_map_lookup_elem().
2156 	 */
2157 	if (is_percpu)
2158 		migrate_disable();
2159 	for (i = 0; i < htab->n_buckets; i++) {
2160 		b = &htab->buckets[i];
2161 		rcu_read_lock();
2162 		head = &b->head;
2163 		hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
2164 			key = elem->key;
2165 			if (is_percpu) {
2166 				/* current cpu value for percpu map */
2167 				pptr = htab_elem_get_ptr(elem, map->key_size);
2168 				val = this_cpu_ptr(pptr);
2169 			} else {
2170 				val = elem->key + roundup_key_size;
2171 			}
2172 			num_elems++;
2173 			ret = callback_fn((u64)(long)map, (u64)(long)key,
2174 					  (u64)(long)val, (u64)(long)callback_ctx, 0);
2175 			/* return value: 0 - continue, 1 - stop and return */
2176 			if (ret) {
2177 				rcu_read_unlock();
2178 				goto out;
2179 			}
2180 		}
2181 		rcu_read_unlock();
2182 	}
2183 out:
2184 	if (is_percpu)
2185 		migrate_enable();
2186 	return num_elems;
2187 }
2188 
2189 BTF_ID_LIST_SINGLE(htab_map_btf_ids, struct, bpf_htab)
2190 const struct bpf_map_ops htab_map_ops = {
2191 	.map_meta_equal = bpf_map_meta_equal,
2192 	.map_alloc_check = htab_map_alloc_check,
2193 	.map_alloc = htab_map_alloc,
2194 	.map_free = htab_map_free,
2195 	.map_get_next_key = htab_map_get_next_key,
2196 	.map_release_uref = htab_map_free_timers,
2197 	.map_lookup_elem = htab_map_lookup_elem,
2198 	.map_lookup_and_delete_elem = htab_map_lookup_and_delete_elem,
2199 	.map_update_elem = htab_map_update_elem,
2200 	.map_delete_elem = htab_map_delete_elem,
2201 	.map_gen_lookup = htab_map_gen_lookup,
2202 	.map_seq_show_elem = htab_map_seq_show_elem,
2203 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2204 	.map_for_each_callback = bpf_for_each_hash_elem,
2205 	BATCH_OPS(htab),
2206 	.map_btf_id = &htab_map_btf_ids[0],
2207 	.iter_seq_info = &iter_seq_info,
2208 };
2209 
2210 const struct bpf_map_ops htab_lru_map_ops = {
2211 	.map_meta_equal = bpf_map_meta_equal,
2212 	.map_alloc_check = htab_map_alloc_check,
2213 	.map_alloc = htab_map_alloc,
2214 	.map_free = htab_map_free,
2215 	.map_get_next_key = htab_map_get_next_key,
2216 	.map_release_uref = htab_map_free_timers,
2217 	.map_lookup_elem = htab_lru_map_lookup_elem,
2218 	.map_lookup_and_delete_elem = htab_lru_map_lookup_and_delete_elem,
2219 	.map_lookup_elem_sys_only = htab_lru_map_lookup_elem_sys,
2220 	.map_update_elem = htab_lru_map_update_elem,
2221 	.map_delete_elem = htab_lru_map_delete_elem,
2222 	.map_gen_lookup = htab_lru_map_gen_lookup,
2223 	.map_seq_show_elem = htab_map_seq_show_elem,
2224 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2225 	.map_for_each_callback = bpf_for_each_hash_elem,
2226 	BATCH_OPS(htab_lru),
2227 	.map_btf_id = &htab_map_btf_ids[0],
2228 	.iter_seq_info = &iter_seq_info,
2229 };
2230 
2231 /* Called from eBPF program */
htab_percpu_map_lookup_elem(struct bpf_map * map,void * key)2232 static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2233 {
2234 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
2235 
2236 	if (l)
2237 		return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2238 	else
2239 		return NULL;
2240 }
2241 
htab_percpu_map_lookup_percpu_elem(struct bpf_map * map,void * key,u32 cpu)2242 static void *htab_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2243 {
2244 	struct htab_elem *l;
2245 
2246 	if (cpu >= nr_cpu_ids)
2247 		return NULL;
2248 
2249 	l = __htab_map_lookup_elem(map, key);
2250 	if (l)
2251 		return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2252 	else
2253 		return NULL;
2254 }
2255 
htab_lru_percpu_map_lookup_elem(struct bpf_map * map,void * key)2256 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2257 {
2258 	struct htab_elem *l = __htab_map_lookup_elem(map, key);
2259 
2260 	if (l) {
2261 		bpf_lru_node_set_ref(&l->lru_node);
2262 		return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2263 	}
2264 
2265 	return NULL;
2266 }
2267 
htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map * map,void * key,u32 cpu)2268 static void *htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2269 {
2270 	struct htab_elem *l;
2271 
2272 	if (cpu >= nr_cpu_ids)
2273 		return NULL;
2274 
2275 	l = __htab_map_lookup_elem(map, key);
2276 	if (l) {
2277 		bpf_lru_node_set_ref(&l->lru_node);
2278 		return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2279 	}
2280 
2281 	return NULL;
2282 }
2283 
bpf_percpu_hash_copy(struct bpf_map * map,void * key,void * value)2284 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
2285 {
2286 	struct htab_elem *l;
2287 	void __percpu *pptr;
2288 	int ret = -ENOENT;
2289 	int cpu, off = 0;
2290 	u32 size;
2291 
2292 	/* per_cpu areas are zero-filled and bpf programs can only
2293 	 * access 'value_size' of them, so copying rounded areas
2294 	 * will not leak any kernel data
2295 	 */
2296 	size = round_up(map->value_size, 8);
2297 	rcu_read_lock();
2298 	l = __htab_map_lookup_elem(map, key);
2299 	if (!l)
2300 		goto out;
2301 	/* We do not mark LRU map element here in order to not mess up
2302 	 * eviction heuristics when user space does a map walk.
2303 	 */
2304 	pptr = htab_elem_get_ptr(l, map->key_size);
2305 	for_each_possible_cpu(cpu) {
2306 		bpf_long_memcpy(value + off,
2307 				per_cpu_ptr(pptr, cpu), size);
2308 		off += size;
2309 	}
2310 	ret = 0;
2311 out:
2312 	rcu_read_unlock();
2313 	return ret;
2314 }
2315 
bpf_percpu_hash_update(struct bpf_map * map,void * key,void * value,u64 map_flags)2316 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
2317 			   u64 map_flags)
2318 {
2319 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2320 	int ret;
2321 
2322 	rcu_read_lock();
2323 	if (htab_is_lru(htab))
2324 		ret = __htab_lru_percpu_map_update_elem(map, key, value,
2325 							map_flags, true);
2326 	else
2327 		ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
2328 						    true);
2329 	rcu_read_unlock();
2330 
2331 	return ret;
2332 }
2333 
htab_percpu_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)2334 static void htab_percpu_map_seq_show_elem(struct bpf_map *map, void *key,
2335 					  struct seq_file *m)
2336 {
2337 	struct htab_elem *l;
2338 	void __percpu *pptr;
2339 	int cpu;
2340 
2341 	rcu_read_lock();
2342 
2343 	l = __htab_map_lookup_elem(map, key);
2344 	if (!l) {
2345 		rcu_read_unlock();
2346 		return;
2347 	}
2348 
2349 	btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
2350 	seq_puts(m, ": {\n");
2351 	pptr = htab_elem_get_ptr(l, map->key_size);
2352 	for_each_possible_cpu(cpu) {
2353 		seq_printf(m, "\tcpu%d: ", cpu);
2354 		btf_type_seq_show(map->btf, map->btf_value_type_id,
2355 				  per_cpu_ptr(pptr, cpu), m);
2356 		seq_puts(m, "\n");
2357 	}
2358 	seq_puts(m, "}\n");
2359 
2360 	rcu_read_unlock();
2361 }
2362 
2363 const struct bpf_map_ops htab_percpu_map_ops = {
2364 	.map_meta_equal = bpf_map_meta_equal,
2365 	.map_alloc_check = htab_map_alloc_check,
2366 	.map_alloc = htab_map_alloc,
2367 	.map_free = htab_map_free,
2368 	.map_get_next_key = htab_map_get_next_key,
2369 	.map_lookup_elem = htab_percpu_map_lookup_elem,
2370 	.map_lookup_and_delete_elem = htab_percpu_map_lookup_and_delete_elem,
2371 	.map_update_elem = htab_percpu_map_update_elem,
2372 	.map_delete_elem = htab_map_delete_elem,
2373 	.map_lookup_percpu_elem = htab_percpu_map_lookup_percpu_elem,
2374 	.map_seq_show_elem = htab_percpu_map_seq_show_elem,
2375 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2376 	.map_for_each_callback = bpf_for_each_hash_elem,
2377 	BATCH_OPS(htab_percpu),
2378 	.map_btf_id = &htab_map_btf_ids[0],
2379 	.iter_seq_info = &iter_seq_info,
2380 };
2381 
2382 const struct bpf_map_ops htab_lru_percpu_map_ops = {
2383 	.map_meta_equal = bpf_map_meta_equal,
2384 	.map_alloc_check = htab_map_alloc_check,
2385 	.map_alloc = htab_map_alloc,
2386 	.map_free = htab_map_free,
2387 	.map_get_next_key = htab_map_get_next_key,
2388 	.map_lookup_elem = htab_lru_percpu_map_lookup_elem,
2389 	.map_lookup_and_delete_elem = htab_lru_percpu_map_lookup_and_delete_elem,
2390 	.map_update_elem = htab_lru_percpu_map_update_elem,
2391 	.map_delete_elem = htab_lru_map_delete_elem,
2392 	.map_lookup_percpu_elem = htab_lru_percpu_map_lookup_percpu_elem,
2393 	.map_seq_show_elem = htab_percpu_map_seq_show_elem,
2394 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
2395 	.map_for_each_callback = bpf_for_each_hash_elem,
2396 	BATCH_OPS(htab_lru_percpu),
2397 	.map_btf_id = &htab_map_btf_ids[0],
2398 	.iter_seq_info = &iter_seq_info,
2399 };
2400 
fd_htab_map_alloc_check(union bpf_attr * attr)2401 static int fd_htab_map_alloc_check(union bpf_attr *attr)
2402 {
2403 	if (attr->value_size != sizeof(u32))
2404 		return -EINVAL;
2405 	return htab_map_alloc_check(attr);
2406 }
2407 
fd_htab_map_free(struct bpf_map * map)2408 static void fd_htab_map_free(struct bpf_map *map)
2409 {
2410 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2411 	struct hlist_nulls_node *n;
2412 	struct hlist_nulls_head *head;
2413 	struct htab_elem *l;
2414 	int i;
2415 
2416 	for (i = 0; i < htab->n_buckets; i++) {
2417 		head = select_bucket(htab, i);
2418 
2419 		hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
2420 			void *ptr = fd_htab_map_get_ptr(map, l);
2421 
2422 			map->ops->map_fd_put_ptr(ptr);
2423 		}
2424 	}
2425 
2426 	htab_map_free(map);
2427 }
2428 
2429 /* only called from syscall */
bpf_fd_htab_map_lookup_elem(struct bpf_map * map,void * key,u32 * value)2430 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
2431 {
2432 	void **ptr;
2433 	int ret = 0;
2434 
2435 	if (!map->ops->map_fd_sys_lookup_elem)
2436 		return -ENOTSUPP;
2437 
2438 	rcu_read_lock();
2439 	ptr = htab_map_lookup_elem(map, key);
2440 	if (ptr)
2441 		*value = map->ops->map_fd_sys_lookup_elem(READ_ONCE(*ptr));
2442 	else
2443 		ret = -ENOENT;
2444 	rcu_read_unlock();
2445 
2446 	return ret;
2447 }
2448 
2449 /* only called from syscall */
bpf_fd_htab_map_update_elem(struct bpf_map * map,struct file * map_file,void * key,void * value,u64 map_flags)2450 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
2451 				void *key, void *value, u64 map_flags)
2452 {
2453 	void *ptr;
2454 	int ret;
2455 	u32 ufd = *(u32 *)value;
2456 
2457 	ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
2458 	if (IS_ERR(ptr))
2459 		return PTR_ERR(ptr);
2460 
2461 	ret = htab_map_update_elem(map, key, &ptr, map_flags);
2462 	if (ret)
2463 		map->ops->map_fd_put_ptr(ptr);
2464 
2465 	return ret;
2466 }
2467 
htab_of_map_alloc(union bpf_attr * attr)2468 static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
2469 {
2470 	struct bpf_map *map, *inner_map_meta;
2471 
2472 	inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
2473 	if (IS_ERR(inner_map_meta))
2474 		return inner_map_meta;
2475 
2476 	map = htab_map_alloc(attr);
2477 	if (IS_ERR(map)) {
2478 		bpf_map_meta_free(inner_map_meta);
2479 		return map;
2480 	}
2481 
2482 	map->inner_map_meta = inner_map_meta;
2483 
2484 	return map;
2485 }
2486 
htab_of_map_lookup_elem(struct bpf_map * map,void * key)2487 static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
2488 {
2489 	struct bpf_map **inner_map  = htab_map_lookup_elem(map, key);
2490 
2491 	if (!inner_map)
2492 		return NULL;
2493 
2494 	return READ_ONCE(*inner_map);
2495 }
2496 
htab_of_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)2497 static int htab_of_map_gen_lookup(struct bpf_map *map,
2498 				  struct bpf_insn *insn_buf)
2499 {
2500 	struct bpf_insn *insn = insn_buf;
2501 	const int ret = BPF_REG_0;
2502 
2503 	BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
2504 		     (void *(*)(struct bpf_map *map, void *key))NULL));
2505 	*insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
2506 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 2);
2507 	*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
2508 				offsetof(struct htab_elem, key) +
2509 				round_up(map->key_size, 8));
2510 	*insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
2511 
2512 	return insn - insn_buf;
2513 }
2514 
htab_of_map_free(struct bpf_map * map)2515 static void htab_of_map_free(struct bpf_map *map)
2516 {
2517 	bpf_map_meta_free(map->inner_map_meta);
2518 	fd_htab_map_free(map);
2519 }
2520 
2521 const struct bpf_map_ops htab_of_maps_map_ops = {
2522 	.map_alloc_check = fd_htab_map_alloc_check,
2523 	.map_alloc = htab_of_map_alloc,
2524 	.map_free = htab_of_map_free,
2525 	.map_get_next_key = htab_map_get_next_key,
2526 	.map_lookup_elem = htab_of_map_lookup_elem,
2527 	.map_delete_elem = htab_map_delete_elem,
2528 	.map_fd_get_ptr = bpf_map_fd_get_ptr,
2529 	.map_fd_put_ptr = bpf_map_fd_put_ptr,
2530 	.map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
2531 	.map_gen_lookup = htab_of_map_gen_lookup,
2532 	.map_check_btf = map_check_no_btf,
2533 	BATCH_OPS(htab),
2534 	.map_btf_id = &htab_map_btf_ids[0],
2535 };
2536