1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
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"
18 #define HTAB_CREATE_FLAG_MASK \
19 (BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE | \
20 BPF_F_ACCESS_MASK | BPF_F_ZERO_SEED)
22 #define BATCH_OPS(_name) \
24 _name##_map_lookup_batch, \
25 .map_lookup_and_delete_batch = \
26 _name##_map_lookup_and_delete_batch, \
28 generic_map_update_batch, \
30 generic_map_delete_batch
33 * The bucket lock has two protection scopes:
35 * 1) Serializing concurrent operations from BPF programs on different
38 * 2) Serializing concurrent operations from BPF programs and sys_bpf()
40 * BPF programs can execute in any context including perf, kprobes and
41 * tracing. As there are almost no limits where perf, kprobes and tracing
42 * can be invoked from the lock operations need to be protected against
43 * deadlocks. Deadlocks can be caused by recursion and by an invocation in
44 * the lock held section when functions which acquire this lock are invoked
45 * from sys_bpf(). BPF recursion is prevented by incrementing the per CPU
46 * variable bpf_prog_active, which prevents BPF programs attached to perf
47 * events, kprobes and tracing to be invoked before the prior invocation
48 * from one of these contexts completed. sys_bpf() uses the same mechanism
49 * by pinning the task to the current CPU and incrementing the recursion
50 * protection across the map operation.
52 * This has subtle implications on PREEMPT_RT. PREEMPT_RT forbids certain
53 * operations like memory allocations (even with GFP_ATOMIC) from atomic
54 * contexts. This is required because even with GFP_ATOMIC the memory
55 * allocator calls into code paths which acquire locks with long held lock
56 * sections. To ensure the deterministic behaviour these locks are regular
57 * spinlocks, which are converted to 'sleepable' spinlocks on RT. The only
58 * true atomic contexts on an RT kernel are the low level hardware
59 * handling, scheduling, low level interrupt handling, NMIs etc. None of
60 * these contexts should ever do memory allocations.
62 * As regular device interrupt handlers and soft interrupts are forced into
63 * thread context, the existing code which does
64 * spin_lock*(); alloc(GFP_ATOMIC); spin_unlock*();
67 * In theory the BPF locks could be converted to regular spinlocks as well,
68 * but the bucket locks and percpu_freelist locks can be taken from
69 * arbitrary contexts (perf, kprobes, tracepoints) which are required to be
70 * atomic contexts even on RT. These mechanisms require preallocated maps,
71 * so there is no need to invoke memory allocations within the lock held
74 * BPF maps which need dynamic allocation are only used from (forced)
75 * thread context on RT and can therefore use regular spinlocks which in
76 * turn allows to invoke memory allocations from the lock held section.
78 * On a non RT kernel this distinction is neither possible nor required.
79 * spinlock maps to raw_spinlock and the extra code is optimized out by the
83 struct hlist_nulls_head head;
85 raw_spinlock_t raw_lock;
90 #define HASHTAB_MAP_LOCK_COUNT 8
91 #define HASHTAB_MAP_LOCK_MASK (HASHTAB_MAP_LOCK_COUNT - 1)
95 struct bucket *buckets;
98 struct pcpu_freelist freelist;
101 struct htab_elem *__percpu *extra_elems;
102 atomic_t count; /* number of elements in this hashtable */
103 u32 n_buckets; /* number of hash buckets */
104 u32 elem_size; /* size of each element in bytes */
106 struct lock_class_key lockdep_key;
107 int __percpu *map_locked[HASHTAB_MAP_LOCK_COUNT];
110 /* each htab element is struct htab_elem + key + value */
113 struct hlist_nulls_node hash_node;
117 struct bpf_htab *htab;
118 struct pcpu_freelist_node fnode;
119 struct htab_elem *batch_flink;
125 struct bpf_lru_node lru_node;
128 char key[] __aligned(8);
131 static inline bool htab_is_prealloc(const struct bpf_htab *htab)
133 return !(htab->map.map_flags & BPF_F_NO_PREALLOC);
136 static inline bool htab_use_raw_lock(const struct bpf_htab *htab)
138 return (!IS_ENABLED(CONFIG_PREEMPT_RT) || htab_is_prealloc(htab));
141 static void htab_init_buckets(struct bpf_htab *htab)
145 for (i = 0; i < htab->n_buckets; i++) {
146 INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
147 if (htab_use_raw_lock(htab)) {
148 raw_spin_lock_init(&htab->buckets[i].raw_lock);
149 lockdep_set_class(&htab->buckets[i].raw_lock,
152 spin_lock_init(&htab->buckets[i].lock);
153 lockdep_set_class(&htab->buckets[i].lock,
160 static inline int htab_lock_bucket(const struct bpf_htab *htab,
161 struct bucket *b, u32 hash,
162 unsigned long *pflags)
166 hash = hash & HASHTAB_MAP_LOCK_MASK;
169 if (unlikely(__this_cpu_inc_return(*(htab->map_locked[hash])) != 1)) {
170 __this_cpu_dec(*(htab->map_locked[hash]));
175 if (htab_use_raw_lock(htab))
176 raw_spin_lock_irqsave(&b->raw_lock, flags);
178 spin_lock_irqsave(&b->lock, flags);
184 static inline void htab_unlock_bucket(const struct bpf_htab *htab,
185 struct bucket *b, u32 hash,
188 hash = hash & HASHTAB_MAP_LOCK_MASK;
189 if (htab_use_raw_lock(htab))
190 raw_spin_unlock_irqrestore(&b->raw_lock, flags);
192 spin_unlock_irqrestore(&b->lock, flags);
193 __this_cpu_dec(*(htab->map_locked[hash]));
197 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
199 static bool htab_is_lru(const struct bpf_htab *htab)
201 return htab->map.map_type == BPF_MAP_TYPE_LRU_HASH ||
202 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
205 static bool htab_is_percpu(const struct bpf_htab *htab)
207 return htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH ||
208 htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
211 static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
214 *(void __percpu **)(l->key + key_size) = pptr;
217 static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
219 return *(void __percpu **)(l->key + key_size);
222 static void *fd_htab_map_get_ptr(const struct bpf_map *map, struct htab_elem *l)
224 return *(void **)(l->key + roundup(map->key_size, 8));
227 static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
229 return (struct htab_elem *) (htab->elems + i * (u64)htab->elem_size);
232 static bool htab_has_extra_elems(struct bpf_htab *htab)
234 return !htab_is_percpu(htab) && !htab_is_lru(htab);
237 static void htab_free_prealloced_timers(struct bpf_htab *htab)
239 u32 num_entries = htab->map.max_entries;
242 if (!map_value_has_timer(&htab->map))
244 if (htab_has_extra_elems(htab))
245 num_entries += num_possible_cpus();
247 for (i = 0; i < num_entries; i++) {
248 struct htab_elem *elem;
250 elem = get_htab_elem(htab, i);
251 bpf_timer_cancel_and_free(elem->key +
252 round_up(htab->map.key_size, 8) +
253 htab->map.timer_off);
258 static void htab_free_prealloced_kptrs(struct bpf_htab *htab)
260 u32 num_entries = htab->map.max_entries;
263 if (!map_value_has_kptrs(&htab->map))
265 if (htab_has_extra_elems(htab))
266 num_entries += num_possible_cpus();
268 for (i = 0; i < num_entries; i++) {
269 struct htab_elem *elem;
271 elem = get_htab_elem(htab, i);
272 bpf_map_free_kptrs(&htab->map, elem->key + round_up(htab->map.key_size, 8));
277 static void htab_free_elems(struct bpf_htab *htab)
281 if (!htab_is_percpu(htab))
284 for (i = 0; i < htab->map.max_entries; i++) {
287 pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
293 bpf_map_area_free(htab->elems);
296 /* The LRU list has a lock (lru_lock). Each htab bucket has a lock
297 * (bucket_lock). If both locks need to be acquired together, the lock
298 * order is always lru_lock -> bucket_lock and this only happens in
299 * bpf_lru_list.c logic. For example, certain code path of
300 * bpf_lru_pop_free(), which is called by function prealloc_lru_pop(),
301 * will acquire lru_lock first followed by acquiring bucket_lock.
303 * In hashtab.c, to avoid deadlock, lock acquisition of
304 * bucket_lock followed by lru_lock is not allowed. In such cases,
305 * bucket_lock needs to be released first before acquiring lru_lock.
307 static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
310 struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
314 u32 key_size = htab->map.key_size;
316 l = container_of(node, struct htab_elem, lru_node);
317 memcpy(l->key, key, key_size);
318 check_and_init_map_value(&htab->map,
319 l->key + round_up(key_size, 8));
326 static int prealloc_init(struct bpf_htab *htab)
328 u32 num_entries = htab->map.max_entries;
329 int err = -ENOMEM, i;
331 if (htab_has_extra_elems(htab))
332 num_entries += num_possible_cpus();
334 htab->elems = bpf_map_area_alloc((u64)htab->elem_size * num_entries,
335 htab->map.numa_node);
339 if (!htab_is_percpu(htab))
340 goto skip_percpu_elems;
342 for (i = 0; i < num_entries; i++) {
343 u32 size = round_up(htab->map.value_size, 8);
346 pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
347 GFP_USER | __GFP_NOWARN);
350 htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
356 if (htab_is_lru(htab))
357 err = bpf_lru_init(&htab->lru,
358 htab->map.map_flags & BPF_F_NO_COMMON_LRU,
359 offsetof(struct htab_elem, hash) -
360 offsetof(struct htab_elem, lru_node),
361 htab_lru_map_delete_node,
364 err = pcpu_freelist_init(&htab->freelist);
369 if (htab_is_lru(htab))
370 bpf_lru_populate(&htab->lru, htab->elems,
371 offsetof(struct htab_elem, lru_node),
372 htab->elem_size, num_entries);
374 pcpu_freelist_populate(&htab->freelist,
375 htab->elems + offsetof(struct htab_elem, fnode),
376 htab->elem_size, num_entries);
381 htab_free_elems(htab);
385 static void prealloc_destroy(struct bpf_htab *htab)
387 htab_free_elems(htab);
389 if (htab_is_lru(htab))
390 bpf_lru_destroy(&htab->lru);
392 pcpu_freelist_destroy(&htab->freelist);
395 static int alloc_extra_elems(struct bpf_htab *htab)
397 struct htab_elem *__percpu *pptr, *l_new;
398 struct pcpu_freelist_node *l;
401 pptr = bpf_map_alloc_percpu(&htab->map, sizeof(struct htab_elem *), 8,
402 GFP_USER | __GFP_NOWARN);
406 for_each_possible_cpu(cpu) {
407 l = pcpu_freelist_pop(&htab->freelist);
408 /* pop will succeed, since prealloc_init()
409 * preallocated extra num_possible_cpus elements
411 l_new = container_of(l, struct htab_elem, fnode);
412 *per_cpu_ptr(pptr, cpu) = l_new;
414 htab->extra_elems = pptr;
418 /* Called from syscall */
419 static int htab_map_alloc_check(union bpf_attr *attr)
421 bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
422 attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
423 bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
424 attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
425 /* percpu_lru means each cpu has its own LRU list.
426 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
427 * the map's value itself is percpu. percpu_lru has
428 * nothing to do with the map's value.
430 bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
431 bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
432 bool zero_seed = (attr->map_flags & BPF_F_ZERO_SEED);
433 int numa_node = bpf_map_attr_numa_node(attr);
435 BUILD_BUG_ON(offsetof(struct htab_elem, htab) !=
436 offsetof(struct htab_elem, hash_node.pprev));
437 BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) !=
438 offsetof(struct htab_elem, hash_node.pprev));
440 if (lru && !bpf_capable())
441 /* LRU implementation is much complicated than other
442 * maps. Hence, limit to CAP_BPF.
446 if (zero_seed && !capable(CAP_SYS_ADMIN))
447 /* Guard against local DoS, and discourage production use. */
450 if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK ||
451 !bpf_map_flags_access_ok(attr->map_flags))
454 if (!lru && percpu_lru)
457 if (lru && !prealloc)
460 if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
463 /* check sanity of attributes.
464 * value_size == 0 may be allowed in the future to use map as a set
466 if (attr->max_entries == 0 || attr->key_size == 0 ||
467 attr->value_size == 0)
470 if ((u64)attr->key_size + attr->value_size >= KMALLOC_MAX_SIZE -
471 sizeof(struct htab_elem))
472 /* if key_size + value_size is bigger, the user space won't be
473 * able to access the elements via bpf syscall. This check
474 * also makes sure that the elem_size doesn't overflow and it's
475 * kmalloc-able later in htab_map_update_elem()
482 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
484 bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
485 attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
486 bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
487 attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
488 /* percpu_lru means each cpu has its own LRU list.
489 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
490 * the map's value itself is percpu. percpu_lru has
491 * nothing to do with the map's value.
493 bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
494 bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
495 struct bpf_htab *htab;
498 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
500 return ERR_PTR(-ENOMEM);
502 lockdep_register_key(&htab->lockdep_key);
504 bpf_map_init_from_attr(&htab->map, attr);
507 /* ensure each CPU's lru list has >=1 elements.
508 * since we are at it, make each lru list has the same
509 * number of elements.
511 htab->map.max_entries = roundup(attr->max_entries,
512 num_possible_cpus());
513 if (htab->map.max_entries < attr->max_entries)
514 htab->map.max_entries = rounddown(attr->max_entries,
515 num_possible_cpus());
518 /* hash table size must be power of 2 */
519 htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
521 htab->elem_size = sizeof(struct htab_elem) +
522 round_up(htab->map.key_size, 8);
524 htab->elem_size += sizeof(void *);
526 htab->elem_size += round_up(htab->map.value_size, 8);
529 /* prevent zero size kmalloc and check for u32 overflow */
530 if (htab->n_buckets == 0 ||
531 htab->n_buckets > U32_MAX / sizeof(struct bucket))
535 htab->buckets = bpf_map_area_alloc(htab->n_buckets *
536 sizeof(struct bucket),
537 htab->map.numa_node);
541 for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++) {
542 htab->map_locked[i] = bpf_map_alloc_percpu(&htab->map,
546 if (!htab->map_locked[i])
547 goto free_map_locked;
550 if (htab->map.map_flags & BPF_F_ZERO_SEED)
553 htab->hashrnd = get_random_int();
555 htab_init_buckets(htab);
558 err = prealloc_init(htab);
560 goto free_map_locked;
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.
566 err = alloc_extra_elems(htab);
575 prealloc_destroy(htab);
577 for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
578 free_percpu(htab->map_locked[i]);
579 bpf_map_area_free(htab->buckets);
581 lockdep_unregister_key(&htab->lockdep_key);
586 static inline u32 htab_map_hash(const void *key, u32 key_len, u32 hashrnd)
588 return jhash(key, key_len, hashrnd);
591 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
593 return &htab->buckets[hash & (htab->n_buckets - 1)];
596 static inline struct hlist_nulls_head *select_bucket(struct bpf_htab *htab, u32 hash)
598 return &__select_bucket(htab, hash)->head;
601 /* this lookup function can only be called with bucket lock taken */
602 static struct htab_elem *lookup_elem_raw(struct hlist_nulls_head *head, u32 hash,
603 void *key, u32 key_size)
605 struct hlist_nulls_node *n;
608 hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
609 if (l->hash == hash && !memcmp(&l->key, key, key_size))
615 /* can be called without bucket lock. it will repeat the loop in
616 * the unlikely event when elements moved from one bucket into another
617 * while link list is being walked
619 static struct htab_elem *lookup_nulls_elem_raw(struct hlist_nulls_head *head,
621 u32 key_size, u32 n_buckets)
623 struct hlist_nulls_node *n;
627 hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
628 if (l->hash == hash && !memcmp(&l->key, key, key_size))
631 if (unlikely(get_nulls_value(n) != (hash & (n_buckets - 1))))
637 /* Called from syscall or from eBPF program directly, so
638 * arguments have to match bpf_map_lookup_elem() exactly.
639 * The return value is adjusted by BPF instructions
640 * in htab_map_gen_lookup().
642 static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
644 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
645 struct hlist_nulls_head *head;
649 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
650 !rcu_read_lock_bh_held());
652 key_size = map->key_size;
654 hash = htab_map_hash(key, key_size, htab->hashrnd);
656 head = select_bucket(htab, hash);
658 l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
663 static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
665 struct htab_elem *l = __htab_map_lookup_elem(map, key);
668 return l->key + round_up(map->key_size, 8);
673 /* inline bpf_map_lookup_elem() call.
676 * bpf_map_lookup_elem
677 * map->ops->map_lookup_elem
678 * htab_map_lookup_elem
679 * __htab_map_lookup_elem
682 * __htab_map_lookup_elem
684 static int htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
686 struct bpf_insn *insn = insn_buf;
687 const int ret = BPF_REG_0;
689 BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
690 (void *(*)(struct bpf_map *map, void *key))NULL));
691 *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
692 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
693 *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
694 offsetof(struct htab_elem, key) +
695 round_up(map->key_size, 8));
696 return insn - insn_buf;
699 static __always_inline void *__htab_lru_map_lookup_elem(struct bpf_map *map,
700 void *key, const bool mark)
702 struct htab_elem *l = __htab_map_lookup_elem(map, key);
706 bpf_lru_node_set_ref(&l->lru_node);
707 return l->key + round_up(map->key_size, 8);
713 static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
715 return __htab_lru_map_lookup_elem(map, key, true);
718 static void *htab_lru_map_lookup_elem_sys(struct bpf_map *map, void *key)
720 return __htab_lru_map_lookup_elem(map, key, false);
723 static int htab_lru_map_gen_lookup(struct bpf_map *map,
724 struct bpf_insn *insn_buf)
726 struct bpf_insn *insn = insn_buf;
727 const int ret = BPF_REG_0;
728 const int ref_reg = BPF_REG_1;
730 BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
731 (void *(*)(struct bpf_map *map, void *key))NULL));
732 *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
733 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 4);
734 *insn++ = BPF_LDX_MEM(BPF_B, ref_reg, ret,
735 offsetof(struct htab_elem, lru_node) +
736 offsetof(struct bpf_lru_node, ref));
737 *insn++ = BPF_JMP_IMM(BPF_JNE, ref_reg, 0, 1);
738 *insn++ = BPF_ST_MEM(BPF_B, ret,
739 offsetof(struct htab_elem, lru_node) +
740 offsetof(struct bpf_lru_node, ref),
742 *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
743 offsetof(struct htab_elem, key) +
744 round_up(map->key_size, 8));
745 return insn - insn_buf;
748 static void check_and_free_fields(struct bpf_htab *htab,
749 struct htab_elem *elem)
751 void *map_value = elem->key + round_up(htab->map.key_size, 8);
753 if (map_value_has_timer(&htab->map))
754 bpf_timer_cancel_and_free(map_value + htab->map.timer_off);
755 if (map_value_has_kptrs(&htab->map))
756 bpf_map_free_kptrs(&htab->map, map_value);
759 /* It is called from the bpf_lru_list when the LRU needs to delete
760 * older elements from the htab.
762 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
764 struct bpf_htab *htab = arg;
765 struct htab_elem *l = NULL, *tgt_l;
766 struct hlist_nulls_head *head;
767 struct hlist_nulls_node *n;
772 tgt_l = container_of(node, struct htab_elem, lru_node);
773 b = __select_bucket(htab, tgt_l->hash);
776 ret = htab_lock_bucket(htab, b, tgt_l->hash, &flags);
780 hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
782 hlist_nulls_del_rcu(&l->hash_node);
783 check_and_free_fields(htab, l);
787 htab_unlock_bucket(htab, b, tgt_l->hash, flags);
792 /* Called from syscall */
793 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
795 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
796 struct hlist_nulls_head *head;
797 struct htab_elem *l, *next_l;
801 WARN_ON_ONCE(!rcu_read_lock_held());
803 key_size = map->key_size;
806 goto find_first_elem;
808 hash = htab_map_hash(key, key_size, htab->hashrnd);
810 head = select_bucket(htab, hash);
813 l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
816 goto find_first_elem;
818 /* key was found, get next key in the same bucket */
819 next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l->hash_node)),
820 struct htab_elem, hash_node);
823 /* if next elem in this hash list is non-zero, just return it */
824 memcpy(next_key, next_l->key, key_size);
828 /* no more elements in this hash list, go to the next bucket */
829 i = hash & (htab->n_buckets - 1);
833 /* iterate over buckets */
834 for (; i < htab->n_buckets; i++) {
835 head = select_bucket(htab, i);
837 /* pick first element in the bucket */
838 next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head)),
839 struct htab_elem, hash_node);
841 /* if it's not empty, just return it */
842 memcpy(next_key, next_l->key, key_size);
847 /* iterated over all buckets and all elements */
851 static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
853 if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
854 free_percpu(htab_elem_get_ptr(l, htab->map.key_size));
855 check_and_free_fields(htab, l);
859 static void htab_elem_free_rcu(struct rcu_head *head)
861 struct htab_elem *l = container_of(head, struct htab_elem, rcu);
862 struct bpf_htab *htab = l->htab;
864 htab_elem_free(htab, l);
867 static void htab_put_fd_value(struct bpf_htab *htab, struct htab_elem *l)
869 struct bpf_map *map = &htab->map;
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);
878 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
880 htab_put_fd_value(htab, l);
882 if (htab_is_prealloc(htab)) {
883 check_and_free_fields(htab, l);
884 __pcpu_freelist_push(&htab->freelist, &l->fnode);
886 atomic_dec(&htab->count);
888 call_rcu(&l->rcu, htab_elem_free_rcu);
892 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
893 void *value, bool onallcpus)
896 /* copy true value_size bytes */
897 memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
899 u32 size = round_up(htab->map.value_size, 8);
902 for_each_possible_cpu(cpu) {
903 bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
910 static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
911 void *value, bool onallcpus)
913 /* When using prealloc and not setting the initial value on all cpus,
914 * zero-fill element values for other cpus (just as what happens when
915 * not using prealloc). Otherwise, bpf program has no way to ensure
916 * known initial values for cpus other than current one
917 * (onallcpus=false always when coming from bpf prog).
919 if (htab_is_prealloc(htab) && !onallcpus) {
920 u32 size = round_up(htab->map.value_size, 8);
921 int current_cpu = raw_smp_processor_id();
924 for_each_possible_cpu(cpu) {
925 if (cpu == current_cpu)
926 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value,
929 memset(per_cpu_ptr(pptr, cpu), 0, size);
932 pcpu_copy_value(htab, pptr, value, onallcpus);
936 static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
938 return htab->map.map_type == BPF_MAP_TYPE_HASH_OF_MAPS &&
942 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
943 void *value, u32 key_size, u32 hash,
944 bool percpu, bool onallcpus,
945 struct htab_elem *old_elem)
947 u32 size = htab->map.value_size;
948 bool prealloc = htab_is_prealloc(htab);
949 struct htab_elem *l_new, **pl_new;
954 /* if we're updating the existing element,
955 * use per-cpu extra elems to avoid freelist_pop/push
957 pl_new = this_cpu_ptr(htab->extra_elems);
959 htab_put_fd_value(htab, old_elem);
962 struct pcpu_freelist_node *l;
964 l = __pcpu_freelist_pop(&htab->freelist);
966 return ERR_PTR(-E2BIG);
967 l_new = container_of(l, struct htab_elem, fnode);
970 if (atomic_inc_return(&htab->count) > htab->map.max_entries)
972 /* when map is full and update() is replacing
973 * old element, it's ok to allocate, since
974 * old element will be freed immediately.
975 * Otherwise return an error
977 l_new = ERR_PTR(-E2BIG);
980 l_new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
981 GFP_NOWAIT | __GFP_NOWARN,
982 htab->map.numa_node);
984 l_new = ERR_PTR(-ENOMEM);
987 check_and_init_map_value(&htab->map,
988 l_new->key + round_up(key_size, 8));
991 memcpy(l_new->key, key, key_size);
993 size = round_up(size, 8);
995 pptr = htab_elem_get_ptr(l_new, key_size);
997 /* alloc_percpu zero-fills */
998 pptr = bpf_map_alloc_percpu(&htab->map, size, 8,
999 GFP_NOWAIT | __GFP_NOWARN);
1002 l_new = ERR_PTR(-ENOMEM);
1007 pcpu_init_value(htab, pptr, value, onallcpus);
1010 htab_elem_set_ptr(l_new, key_size, pptr);
1011 } else if (fd_htab_map_needs_adjust(htab)) {
1012 size = round_up(size, 8);
1013 memcpy(l_new->key + round_up(key_size, 8), value, size);
1015 copy_map_value(&htab->map,
1016 l_new->key + round_up(key_size, 8),
1023 atomic_dec(&htab->count);
1027 static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
1030 if (l_old && (map_flags & ~BPF_F_LOCK) == BPF_NOEXIST)
1031 /* elem already exists */
1034 if (!l_old && (map_flags & ~BPF_F_LOCK) == BPF_EXIST)
1035 /* elem doesn't exist, cannot update it */
1041 /* Called from syscall or from eBPF program */
1042 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
1045 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1046 struct htab_elem *l_new = NULL, *l_old;
1047 struct hlist_nulls_head *head;
1048 unsigned long flags;
1053 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
1057 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1058 !rcu_read_lock_bh_held());
1060 key_size = map->key_size;
1062 hash = htab_map_hash(key, key_size, htab->hashrnd);
1064 b = __select_bucket(htab, hash);
1067 if (unlikely(map_flags & BPF_F_LOCK)) {
1068 if (unlikely(!map_value_has_spin_lock(map)))
1070 /* find an element without taking the bucket lock */
1071 l_old = lookup_nulls_elem_raw(head, hash, key, key_size,
1073 ret = check_flags(htab, l_old, map_flags);
1077 /* grab the element lock and update value in place */
1078 copy_map_value_locked(map,
1079 l_old->key + round_up(key_size, 8),
1083 /* fall through, grab the bucket lock and lookup again.
1084 * 99.9% chance that the element won't be found,
1085 * but second lookup under lock has to be done.
1089 ret = htab_lock_bucket(htab, b, hash, &flags);
1093 l_old = lookup_elem_raw(head, hash, key, key_size);
1095 ret = check_flags(htab, l_old, map_flags);
1099 if (unlikely(l_old && (map_flags & BPF_F_LOCK))) {
1100 /* first lookup without the bucket lock didn't find the element,
1101 * but second lookup with the bucket lock found it.
1102 * This case is highly unlikely, but has to be dealt with:
1103 * grab the element lock in addition to the bucket lock
1104 * and update element in place
1106 copy_map_value_locked(map,
1107 l_old->key + round_up(key_size, 8),
1113 l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
1115 if (IS_ERR(l_new)) {
1116 /* all pre-allocated elements are in use or memory exhausted */
1117 ret = PTR_ERR(l_new);
1121 /* add new element to the head of the list, so that
1122 * concurrent search will find it before old elem
1124 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1126 hlist_nulls_del_rcu(&l_old->hash_node);
1127 if (!htab_is_prealloc(htab))
1128 free_htab_elem(htab, l_old);
1130 check_and_free_fields(htab, l_old);
1134 htab_unlock_bucket(htab, b, hash, flags);
1138 static void htab_lru_push_free(struct bpf_htab *htab, struct htab_elem *elem)
1140 check_and_free_fields(htab, elem);
1141 bpf_lru_push_free(&htab->lru, &elem->lru_node);
1144 static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
1147 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1148 struct htab_elem *l_new, *l_old = NULL;
1149 struct hlist_nulls_head *head;
1150 unsigned long flags;
1155 if (unlikely(map_flags > BPF_EXIST))
1159 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1160 !rcu_read_lock_bh_held());
1162 key_size = map->key_size;
1164 hash = htab_map_hash(key, key_size, htab->hashrnd);
1166 b = __select_bucket(htab, hash);
1169 /* For LRU, we need to alloc before taking bucket's
1170 * spinlock because getting free nodes from LRU may need
1171 * to remove older elements from htab and this removal
1172 * operation will need a bucket lock.
1174 l_new = prealloc_lru_pop(htab, key, hash);
1177 copy_map_value(&htab->map,
1178 l_new->key + round_up(map->key_size, 8), value);
1180 ret = htab_lock_bucket(htab, b, hash, &flags);
1184 l_old = lookup_elem_raw(head, hash, key, key_size);
1186 ret = check_flags(htab, l_old, map_flags);
1190 /* add new element to the head of the list, so that
1191 * concurrent search will find it before old elem
1193 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1195 bpf_lru_node_set_ref(&l_new->lru_node);
1196 hlist_nulls_del_rcu(&l_old->hash_node);
1201 htab_unlock_bucket(htab, b, hash, flags);
1204 htab_lru_push_free(htab, l_new);
1206 htab_lru_push_free(htab, l_old);
1211 static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1212 void *value, u64 map_flags,
1215 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1216 struct htab_elem *l_new = NULL, *l_old;
1217 struct hlist_nulls_head *head;
1218 unsigned long flags;
1223 if (unlikely(map_flags > BPF_EXIST))
1227 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1228 !rcu_read_lock_bh_held());
1230 key_size = map->key_size;
1232 hash = htab_map_hash(key, key_size, htab->hashrnd);
1234 b = __select_bucket(htab, hash);
1237 ret = htab_lock_bucket(htab, b, hash, &flags);
1241 l_old = lookup_elem_raw(head, hash, key, key_size);
1243 ret = check_flags(htab, l_old, map_flags);
1248 /* per-cpu hash map can update value in-place */
1249 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1252 l_new = alloc_htab_elem(htab, key, value, key_size,
1253 hash, true, onallcpus, NULL);
1254 if (IS_ERR(l_new)) {
1255 ret = PTR_ERR(l_new);
1258 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1262 htab_unlock_bucket(htab, b, hash, flags);
1266 static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1267 void *value, u64 map_flags,
1270 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1271 struct htab_elem *l_new = NULL, *l_old;
1272 struct hlist_nulls_head *head;
1273 unsigned long flags;
1278 if (unlikely(map_flags > BPF_EXIST))
1282 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1283 !rcu_read_lock_bh_held());
1285 key_size = map->key_size;
1287 hash = htab_map_hash(key, key_size, htab->hashrnd);
1289 b = __select_bucket(htab, hash);
1292 /* For LRU, we need to alloc before taking bucket's
1293 * spinlock because LRU's elem alloc may need
1294 * to remove older elem from htab and this removal
1295 * operation will need a bucket lock.
1297 if (map_flags != BPF_EXIST) {
1298 l_new = prealloc_lru_pop(htab, key, hash);
1303 ret = htab_lock_bucket(htab, b, hash, &flags);
1307 l_old = lookup_elem_raw(head, hash, key, key_size);
1309 ret = check_flags(htab, l_old, map_flags);
1314 bpf_lru_node_set_ref(&l_old->lru_node);
1316 /* per-cpu hash map can update value in-place */
1317 pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
1320 pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
1322 hlist_nulls_add_head_rcu(&l_new->hash_node, head);
1327 htab_unlock_bucket(htab, b, hash, flags);
1329 bpf_lru_push_free(&htab->lru, &l_new->lru_node);
1333 static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
1334 void *value, u64 map_flags)
1336 return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
1339 static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
1340 void *value, u64 map_flags)
1342 return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
1346 /* Called from syscall or from eBPF program */
1347 static int htab_map_delete_elem(struct bpf_map *map, void *key)
1349 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1350 struct hlist_nulls_head *head;
1352 struct htab_elem *l;
1353 unsigned long flags;
1357 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1358 !rcu_read_lock_bh_held());
1360 key_size = map->key_size;
1362 hash = htab_map_hash(key, key_size, htab->hashrnd);
1363 b = __select_bucket(htab, hash);
1366 ret = htab_lock_bucket(htab, b, hash, &flags);
1370 l = lookup_elem_raw(head, hash, key, key_size);
1373 hlist_nulls_del_rcu(&l->hash_node);
1374 free_htab_elem(htab, l);
1379 htab_unlock_bucket(htab, b, hash, flags);
1383 static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
1385 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1386 struct hlist_nulls_head *head;
1388 struct htab_elem *l;
1389 unsigned long flags;
1393 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
1394 !rcu_read_lock_bh_held());
1396 key_size = map->key_size;
1398 hash = htab_map_hash(key, key_size, htab->hashrnd);
1399 b = __select_bucket(htab, hash);
1402 ret = htab_lock_bucket(htab, b, hash, &flags);
1406 l = lookup_elem_raw(head, hash, key, key_size);
1409 hlist_nulls_del_rcu(&l->hash_node);
1413 htab_unlock_bucket(htab, b, hash, flags);
1415 htab_lru_push_free(htab, l);
1419 static void delete_all_elements(struct bpf_htab *htab)
1423 for (i = 0; i < htab->n_buckets; i++) {
1424 struct hlist_nulls_head *head = select_bucket(htab, i);
1425 struct hlist_nulls_node *n;
1426 struct htab_elem *l;
1428 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1429 hlist_nulls_del_rcu(&l->hash_node);
1430 htab_elem_free(htab, l);
1435 static void htab_free_malloced_timers(struct bpf_htab *htab)
1440 for (i = 0; i < htab->n_buckets; i++) {
1441 struct hlist_nulls_head *head = select_bucket(htab, i);
1442 struct hlist_nulls_node *n;
1443 struct htab_elem *l;
1445 hlist_nulls_for_each_entry(l, n, head, hash_node) {
1446 /* We don't reset or free kptr on uref dropping to zero,
1447 * hence just free timer.
1449 bpf_timer_cancel_and_free(l->key +
1450 round_up(htab->map.key_size, 8) +
1451 htab->map.timer_off);
1458 static void htab_map_free_timers(struct bpf_map *map)
1460 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1462 /* We don't reset or free kptr on uref dropping to zero. */
1463 if (!map_value_has_timer(&htab->map))
1465 if (!htab_is_prealloc(htab))
1466 htab_free_malloced_timers(htab);
1468 htab_free_prealloced_timers(htab);
1471 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
1472 static void htab_map_free(struct bpf_map *map)
1474 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1477 /* bpf_free_used_maps() or close(map_fd) will trigger this map_free callback.
1478 * bpf_free_used_maps() is called after bpf prog is no longer executing.
1479 * There is no need to synchronize_rcu() here to protect map elements.
1482 /* some of free_htab_elem() callbacks for elements of this map may
1483 * not have executed. Wait for them.
1486 if (!htab_is_prealloc(htab)) {
1487 delete_all_elements(htab);
1489 htab_free_prealloced_kptrs(htab);
1490 prealloc_destroy(htab);
1493 bpf_map_free_kptr_off_tab(map);
1494 free_percpu(htab->extra_elems);
1495 bpf_map_area_free(htab->buckets);
1496 for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
1497 free_percpu(htab->map_locked[i]);
1498 lockdep_unregister_key(&htab->lockdep_key);
1502 static void htab_map_seq_show_elem(struct bpf_map *map, void *key,
1509 value = htab_map_lookup_elem(map, key);
1515 btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
1517 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
1523 static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1524 void *value, bool is_lru_map,
1525 bool is_percpu, u64 flags)
1527 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1528 struct hlist_nulls_head *head;
1529 unsigned long bflags;
1530 struct htab_elem *l;
1535 key_size = map->key_size;
1537 hash = htab_map_hash(key, key_size, htab->hashrnd);
1538 b = __select_bucket(htab, hash);
1541 ret = htab_lock_bucket(htab, b, hash, &bflags);
1545 l = lookup_elem_raw(head, hash, key, key_size);
1550 u32 roundup_value_size = round_up(map->value_size, 8);
1551 void __percpu *pptr;
1554 pptr = htab_elem_get_ptr(l, key_size);
1555 for_each_possible_cpu(cpu) {
1556 bpf_long_memcpy(value + off,
1557 per_cpu_ptr(pptr, cpu),
1558 roundup_value_size);
1559 off += roundup_value_size;
1562 u32 roundup_key_size = round_up(map->key_size, 8);
1564 if (flags & BPF_F_LOCK)
1565 copy_map_value_locked(map, value, l->key +
1569 copy_map_value(map, value, l->key +
1571 check_and_init_map_value(map, value);
1574 hlist_nulls_del_rcu(&l->hash_node);
1576 free_htab_elem(htab, l);
1579 htab_unlock_bucket(htab, b, hash, bflags);
1581 if (is_lru_map && l)
1582 htab_lru_push_free(htab, l);
1587 static int htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1588 void *value, u64 flags)
1590 return __htab_map_lookup_and_delete_elem(map, key, value, false, false,
1594 static int htab_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1595 void *key, void *value,
1598 return __htab_map_lookup_and_delete_elem(map, key, value, false, true,
1602 static int htab_lru_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
1603 void *value, u64 flags)
1605 return __htab_map_lookup_and_delete_elem(map, key, value, true, false,
1609 static int htab_lru_percpu_map_lookup_and_delete_elem(struct bpf_map *map,
1610 void *key, void *value,
1613 return __htab_map_lookup_and_delete_elem(map, key, value, true, true,
1618 __htab_map_lookup_and_delete_batch(struct bpf_map *map,
1619 const union bpf_attr *attr,
1620 union bpf_attr __user *uattr,
1621 bool do_delete, bool is_lru_map,
1624 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1625 u32 bucket_cnt, total, key_size, value_size, roundup_key_size;
1626 void *keys = NULL, *values = NULL, *value, *dst_key, *dst_val;
1627 void __user *uvalues = u64_to_user_ptr(attr->batch.values);
1628 void __user *ukeys = u64_to_user_ptr(attr->batch.keys);
1629 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1630 u32 batch, max_count, size, bucket_size, map_id;
1631 struct htab_elem *node_to_free = NULL;
1632 u64 elem_map_flags, map_flags;
1633 struct hlist_nulls_head *head;
1634 struct hlist_nulls_node *n;
1635 unsigned long flags = 0;
1636 bool locked = false;
1637 struct htab_elem *l;
1641 elem_map_flags = attr->batch.elem_flags;
1642 if ((elem_map_flags & ~BPF_F_LOCK) ||
1643 ((elem_map_flags & BPF_F_LOCK) && !map_value_has_spin_lock(map)))
1646 map_flags = attr->batch.flags;
1650 max_count = attr->batch.count;
1654 if (put_user(0, &uattr->batch.count))
1658 if (ubatch && copy_from_user(&batch, ubatch, sizeof(batch)))
1661 if (batch >= htab->n_buckets)
1664 key_size = htab->map.key_size;
1665 roundup_key_size = round_up(htab->map.key_size, 8);
1666 value_size = htab->map.value_size;
1667 size = round_up(value_size, 8);
1669 value_size = size * num_possible_cpus();
1671 /* while experimenting with hash tables with sizes ranging from 10 to
1672 * 1000, it was observed that a bucket can have up to 5 entries.
1677 /* We cannot do copy_from_user or copy_to_user inside
1678 * the rcu_read_lock. Allocate enough space here.
1680 keys = kvmalloc_array(key_size, bucket_size, GFP_USER | __GFP_NOWARN);
1681 values = kvmalloc_array(value_size, bucket_size, GFP_USER | __GFP_NOWARN);
1682 if (!keys || !values) {
1688 bpf_disable_instrumentation();
1693 b = &htab->buckets[batch];
1695 /* do not grab the lock unless need it (bucket_cnt > 0). */
1697 ret = htab_lock_bucket(htab, b, batch, &flags);
1703 hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
1706 if (bucket_cnt && !locked) {
1711 if (bucket_cnt > (max_count - total)) {
1714 /* Note that since bucket_cnt > 0 here, it is implicit
1715 * that the locked was grabbed, so release it.
1717 htab_unlock_bucket(htab, b, batch, flags);
1719 bpf_enable_instrumentation();
1723 if (bucket_cnt > bucket_size) {
1724 bucket_size = bucket_cnt;
1725 /* Note that since bucket_cnt > 0 here, it is implicit
1726 * that the locked was grabbed, so release it.
1728 htab_unlock_bucket(htab, b, batch, flags);
1730 bpf_enable_instrumentation();
1736 /* Next block is only safe to run if you have grabbed the lock */
1740 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
1741 memcpy(dst_key, l->key, key_size);
1745 void __percpu *pptr;
1747 pptr = htab_elem_get_ptr(l, map->key_size);
1748 for_each_possible_cpu(cpu) {
1749 bpf_long_memcpy(dst_val + off,
1750 per_cpu_ptr(pptr, cpu), size);
1754 value = l->key + roundup_key_size;
1755 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1756 struct bpf_map **inner_map = value;
1758 /* Actual value is the id of the inner map */
1759 map_id = map->ops->map_fd_sys_lookup_elem(*inner_map);
1763 if (elem_map_flags & BPF_F_LOCK)
1764 copy_map_value_locked(map, dst_val, value,
1767 copy_map_value(map, dst_val, value);
1768 check_and_init_map_value(map, dst_val);
1771 hlist_nulls_del_rcu(&l->hash_node);
1773 /* bpf_lru_push_free() will acquire lru_lock, which
1774 * may cause deadlock. See comments in function
1775 * prealloc_lru_pop(). Let us do bpf_lru_push_free()
1776 * after releasing the bucket lock.
1779 l->batch_flink = node_to_free;
1782 free_htab_elem(htab, l);
1785 dst_key += key_size;
1786 dst_val += value_size;
1789 htab_unlock_bucket(htab, b, batch, flags);
1792 while (node_to_free) {
1794 node_to_free = node_to_free->batch_flink;
1795 htab_lru_push_free(htab, l);
1799 /* If we are not copying data, we can go to next bucket and avoid
1800 * unlocking the rcu.
1802 if (!bucket_cnt && (batch + 1 < htab->n_buckets)) {
1808 bpf_enable_instrumentation();
1809 if (bucket_cnt && (copy_to_user(ukeys + total * key_size, keys,
1810 key_size * bucket_cnt) ||
1811 copy_to_user(uvalues + total * value_size, values,
1812 value_size * bucket_cnt))) {
1817 total += bucket_cnt;
1819 if (batch >= htab->n_buckets) {
1829 /* copy # of entries and next batch */
1830 ubatch = u64_to_user_ptr(attr->batch.out_batch);
1831 if (copy_to_user(ubatch, &batch, sizeof(batch)) ||
1832 put_user(total, &uattr->batch.count))
1842 htab_percpu_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1843 union bpf_attr __user *uattr)
1845 return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1850 htab_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1851 const union bpf_attr *attr,
1852 union bpf_attr __user *uattr)
1854 return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1859 htab_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1860 union bpf_attr __user *uattr)
1862 return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1867 htab_map_lookup_and_delete_batch(struct bpf_map *map,
1868 const union bpf_attr *attr,
1869 union bpf_attr __user *uattr)
1871 return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1876 htab_lru_percpu_map_lookup_batch(struct bpf_map *map,
1877 const union bpf_attr *attr,
1878 union bpf_attr __user *uattr)
1880 return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1885 htab_lru_percpu_map_lookup_and_delete_batch(struct bpf_map *map,
1886 const union bpf_attr *attr,
1887 union bpf_attr __user *uattr)
1889 return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1894 htab_lru_map_lookup_batch(struct bpf_map *map, const union bpf_attr *attr,
1895 union bpf_attr __user *uattr)
1897 return __htab_map_lookup_and_delete_batch(map, attr, uattr, false,
1902 htab_lru_map_lookup_and_delete_batch(struct bpf_map *map,
1903 const union bpf_attr *attr,
1904 union bpf_attr __user *uattr)
1906 return __htab_map_lookup_and_delete_batch(map, attr, uattr, true,
1910 struct bpf_iter_seq_hash_map_info {
1911 struct bpf_map *map;
1912 struct bpf_htab *htab;
1913 void *percpu_value_buf; // non-zero means percpu hash
1918 static struct htab_elem *
1919 bpf_hash_map_seq_find_next(struct bpf_iter_seq_hash_map_info *info,
1920 struct htab_elem *prev_elem)
1922 const struct bpf_htab *htab = info->htab;
1923 u32 skip_elems = info->skip_elems;
1924 u32 bucket_id = info->bucket_id;
1925 struct hlist_nulls_head *head;
1926 struct hlist_nulls_node *n;
1927 struct htab_elem *elem;
1931 if (bucket_id >= htab->n_buckets)
1934 /* try to find next elem in the same bucket */
1936 /* no update/deletion on this bucket, prev_elem should be still valid
1937 * and we won't skip elements.
1939 n = rcu_dereference_raw(hlist_nulls_next_rcu(&prev_elem->hash_node));
1940 elem = hlist_nulls_entry_safe(n, struct htab_elem, hash_node);
1944 /* not found, unlock and go to the next bucket */
1945 b = &htab->buckets[bucket_id++];
1950 for (i = bucket_id; i < htab->n_buckets; i++) {
1951 b = &htab->buckets[i];
1956 hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
1957 if (count >= skip_elems) {
1958 info->bucket_id = i;
1959 info->skip_elems = count;
1969 info->bucket_id = i;
1970 info->skip_elems = 0;
1974 static void *bpf_hash_map_seq_start(struct seq_file *seq, loff_t *pos)
1976 struct bpf_iter_seq_hash_map_info *info = seq->private;
1977 struct htab_elem *elem;
1979 elem = bpf_hash_map_seq_find_next(info, NULL);
1988 static void *bpf_hash_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1990 struct bpf_iter_seq_hash_map_info *info = seq->private;
1994 return bpf_hash_map_seq_find_next(info, v);
1997 static int __bpf_hash_map_seq_show(struct seq_file *seq, struct htab_elem *elem)
1999 struct bpf_iter_seq_hash_map_info *info = seq->private;
2000 u32 roundup_key_size, roundup_value_size;
2001 struct bpf_iter__bpf_map_elem ctx = {};
2002 struct bpf_map *map = info->map;
2003 struct bpf_iter_meta meta;
2004 int ret = 0, off = 0, cpu;
2005 struct bpf_prog *prog;
2006 void __percpu *pptr;
2009 prog = bpf_iter_get_info(&meta, elem == NULL);
2012 ctx.map = info->map;
2014 roundup_key_size = round_up(map->key_size, 8);
2015 ctx.key = elem->key;
2016 if (!info->percpu_value_buf) {
2017 ctx.value = elem->key + roundup_key_size;
2019 roundup_value_size = round_up(map->value_size, 8);
2020 pptr = htab_elem_get_ptr(elem, map->key_size);
2021 for_each_possible_cpu(cpu) {
2022 bpf_long_memcpy(info->percpu_value_buf + off,
2023 per_cpu_ptr(pptr, cpu),
2024 roundup_value_size);
2025 off += roundup_value_size;
2027 ctx.value = info->percpu_value_buf;
2030 ret = bpf_iter_run_prog(prog, &ctx);
2036 static int bpf_hash_map_seq_show(struct seq_file *seq, void *v)
2038 return __bpf_hash_map_seq_show(seq, v);
2041 static void bpf_hash_map_seq_stop(struct seq_file *seq, void *v)
2044 (void)__bpf_hash_map_seq_show(seq, NULL);
2049 static int bpf_iter_init_hash_map(void *priv_data,
2050 struct bpf_iter_aux_info *aux)
2052 struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2053 struct bpf_map *map = aux->map;
2057 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2058 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2059 buf_size = round_up(map->value_size, 8) * num_possible_cpus();
2060 value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
2064 seq_info->percpu_value_buf = value_buf;
2067 seq_info->map = map;
2068 seq_info->htab = container_of(map, struct bpf_htab, map);
2072 static void bpf_iter_fini_hash_map(void *priv_data)
2074 struct bpf_iter_seq_hash_map_info *seq_info = priv_data;
2076 kfree(seq_info->percpu_value_buf);
2079 static const struct seq_operations bpf_hash_map_seq_ops = {
2080 .start = bpf_hash_map_seq_start,
2081 .next = bpf_hash_map_seq_next,
2082 .stop = bpf_hash_map_seq_stop,
2083 .show = bpf_hash_map_seq_show,
2086 static const struct bpf_iter_seq_info iter_seq_info = {
2087 .seq_ops = &bpf_hash_map_seq_ops,
2088 .init_seq_private = bpf_iter_init_hash_map,
2089 .fini_seq_private = bpf_iter_fini_hash_map,
2090 .seq_priv_size = sizeof(struct bpf_iter_seq_hash_map_info),
2093 static int bpf_for_each_hash_elem(struct bpf_map *map, bpf_callback_t callback_fn,
2094 void *callback_ctx, u64 flags)
2096 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2097 struct hlist_nulls_head *head;
2098 struct hlist_nulls_node *n;
2099 struct htab_elem *elem;
2100 u32 roundup_key_size;
2101 int i, num_elems = 0;
2102 void __percpu *pptr;
2111 is_percpu = htab_is_percpu(htab);
2113 roundup_key_size = round_up(map->key_size, 8);
2114 /* disable migration so percpu value prepared here will be the
2115 * same as the one seen by the bpf program with bpf_map_lookup_elem().
2119 for (i = 0; i < htab->n_buckets; i++) {
2120 b = &htab->buckets[i];
2123 hlist_nulls_for_each_entry_rcu(elem, n, head, hash_node) {
2126 /* current cpu value for percpu map */
2127 pptr = htab_elem_get_ptr(elem, map->key_size);
2128 val = this_cpu_ptr(pptr);
2130 val = elem->key + roundup_key_size;
2133 ret = callback_fn((u64)(long)map, (u64)(long)key,
2134 (u64)(long)val, (u64)(long)callback_ctx, 0);
2135 /* return value: 0 - continue, 1 - stop and return */
2149 BTF_ID_LIST_SINGLE(htab_map_btf_ids, struct, bpf_htab)
2150 const struct bpf_map_ops htab_map_ops = {
2151 .map_meta_equal = bpf_map_meta_equal,
2152 .map_alloc_check = htab_map_alloc_check,
2153 .map_alloc = htab_map_alloc,
2154 .map_free = htab_map_free,
2155 .map_get_next_key = htab_map_get_next_key,
2156 .map_release_uref = htab_map_free_timers,
2157 .map_lookup_elem = htab_map_lookup_elem,
2158 .map_lookup_and_delete_elem = htab_map_lookup_and_delete_elem,
2159 .map_update_elem = htab_map_update_elem,
2160 .map_delete_elem = htab_map_delete_elem,
2161 .map_gen_lookup = htab_map_gen_lookup,
2162 .map_seq_show_elem = htab_map_seq_show_elem,
2163 .map_set_for_each_callback_args = map_set_for_each_callback_args,
2164 .map_for_each_callback = bpf_for_each_hash_elem,
2166 .map_btf_id = &htab_map_btf_ids[0],
2167 .iter_seq_info = &iter_seq_info,
2170 const struct bpf_map_ops htab_lru_map_ops = {
2171 .map_meta_equal = bpf_map_meta_equal,
2172 .map_alloc_check = htab_map_alloc_check,
2173 .map_alloc = htab_map_alloc,
2174 .map_free = htab_map_free,
2175 .map_get_next_key = htab_map_get_next_key,
2176 .map_release_uref = htab_map_free_timers,
2177 .map_lookup_elem = htab_lru_map_lookup_elem,
2178 .map_lookup_and_delete_elem = htab_lru_map_lookup_and_delete_elem,
2179 .map_lookup_elem_sys_only = htab_lru_map_lookup_elem_sys,
2180 .map_update_elem = htab_lru_map_update_elem,
2181 .map_delete_elem = htab_lru_map_delete_elem,
2182 .map_gen_lookup = htab_lru_map_gen_lookup,
2183 .map_seq_show_elem = htab_map_seq_show_elem,
2184 .map_set_for_each_callback_args = map_set_for_each_callback_args,
2185 .map_for_each_callback = bpf_for_each_hash_elem,
2186 BATCH_OPS(htab_lru),
2187 .map_btf_id = &htab_map_btf_ids[0],
2188 .iter_seq_info = &iter_seq_info,
2191 /* Called from eBPF program */
2192 static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2194 struct htab_elem *l = __htab_map_lookup_elem(map, key);
2197 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2202 static void *htab_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2204 struct htab_elem *l;
2206 if (cpu >= nr_cpu_ids)
2209 l = __htab_map_lookup_elem(map, key);
2211 return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2216 static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
2218 struct htab_elem *l = __htab_map_lookup_elem(map, key);
2221 bpf_lru_node_set_ref(&l->lru_node);
2222 return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
2228 static void *htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
2230 struct htab_elem *l;
2232 if (cpu >= nr_cpu_ids)
2235 l = __htab_map_lookup_elem(map, key);
2237 bpf_lru_node_set_ref(&l->lru_node);
2238 return per_cpu_ptr(htab_elem_get_ptr(l, map->key_size), cpu);
2244 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
2246 struct htab_elem *l;
2247 void __percpu *pptr;
2252 /* per_cpu areas are zero-filled and bpf programs can only
2253 * access 'value_size' of them, so copying rounded areas
2254 * will not leak any kernel data
2256 size = round_up(map->value_size, 8);
2258 l = __htab_map_lookup_elem(map, key);
2261 /* We do not mark LRU map element here in order to not mess up
2262 * eviction heuristics when user space does a map walk.
2264 pptr = htab_elem_get_ptr(l, map->key_size);
2265 for_each_possible_cpu(cpu) {
2266 bpf_long_memcpy(value + off,
2267 per_cpu_ptr(pptr, cpu), size);
2276 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
2279 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2283 if (htab_is_lru(htab))
2284 ret = __htab_lru_percpu_map_update_elem(map, key, value,
2287 ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
2294 static void htab_percpu_map_seq_show_elem(struct bpf_map *map, void *key,
2297 struct htab_elem *l;
2298 void __percpu *pptr;
2303 l = __htab_map_lookup_elem(map, key);
2309 btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
2310 seq_puts(m, ": {\n");
2311 pptr = htab_elem_get_ptr(l, map->key_size);
2312 for_each_possible_cpu(cpu) {
2313 seq_printf(m, "\tcpu%d: ", cpu);
2314 btf_type_seq_show(map->btf, map->btf_value_type_id,
2315 per_cpu_ptr(pptr, cpu), m);
2323 const struct bpf_map_ops htab_percpu_map_ops = {
2324 .map_meta_equal = bpf_map_meta_equal,
2325 .map_alloc_check = htab_map_alloc_check,
2326 .map_alloc = htab_map_alloc,
2327 .map_free = htab_map_free,
2328 .map_get_next_key = htab_map_get_next_key,
2329 .map_lookup_elem = htab_percpu_map_lookup_elem,
2330 .map_lookup_and_delete_elem = htab_percpu_map_lookup_and_delete_elem,
2331 .map_update_elem = htab_percpu_map_update_elem,
2332 .map_delete_elem = htab_map_delete_elem,
2333 .map_lookup_percpu_elem = htab_percpu_map_lookup_percpu_elem,
2334 .map_seq_show_elem = htab_percpu_map_seq_show_elem,
2335 .map_set_for_each_callback_args = map_set_for_each_callback_args,
2336 .map_for_each_callback = bpf_for_each_hash_elem,
2337 BATCH_OPS(htab_percpu),
2338 .map_btf_id = &htab_map_btf_ids[0],
2339 .iter_seq_info = &iter_seq_info,
2342 const struct bpf_map_ops htab_lru_percpu_map_ops = {
2343 .map_meta_equal = bpf_map_meta_equal,
2344 .map_alloc_check = htab_map_alloc_check,
2345 .map_alloc = htab_map_alloc,
2346 .map_free = htab_map_free,
2347 .map_get_next_key = htab_map_get_next_key,
2348 .map_lookup_elem = htab_lru_percpu_map_lookup_elem,
2349 .map_lookup_and_delete_elem = htab_lru_percpu_map_lookup_and_delete_elem,
2350 .map_update_elem = htab_lru_percpu_map_update_elem,
2351 .map_delete_elem = htab_lru_map_delete_elem,
2352 .map_lookup_percpu_elem = htab_lru_percpu_map_lookup_percpu_elem,
2353 .map_seq_show_elem = htab_percpu_map_seq_show_elem,
2354 .map_set_for_each_callback_args = map_set_for_each_callback_args,
2355 .map_for_each_callback = bpf_for_each_hash_elem,
2356 BATCH_OPS(htab_lru_percpu),
2357 .map_btf_id = &htab_map_btf_ids[0],
2358 .iter_seq_info = &iter_seq_info,
2361 static int fd_htab_map_alloc_check(union bpf_attr *attr)
2363 if (attr->value_size != sizeof(u32))
2365 return htab_map_alloc_check(attr);
2368 static void fd_htab_map_free(struct bpf_map *map)
2370 struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
2371 struct hlist_nulls_node *n;
2372 struct hlist_nulls_head *head;
2373 struct htab_elem *l;
2376 for (i = 0; i < htab->n_buckets; i++) {
2377 head = select_bucket(htab, i);
2379 hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
2380 void *ptr = fd_htab_map_get_ptr(map, l);
2382 map->ops->map_fd_put_ptr(ptr);
2389 /* only called from syscall */
2390 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
2395 if (!map->ops->map_fd_sys_lookup_elem)
2399 ptr = htab_map_lookup_elem(map, key);
2401 *value = map->ops->map_fd_sys_lookup_elem(READ_ONCE(*ptr));
2409 /* only called from syscall */
2410 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
2411 void *key, void *value, u64 map_flags)
2415 u32 ufd = *(u32 *)value;
2417 ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
2419 return PTR_ERR(ptr);
2421 ret = htab_map_update_elem(map, key, &ptr, map_flags);
2423 map->ops->map_fd_put_ptr(ptr);
2428 static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
2430 struct bpf_map *map, *inner_map_meta;
2432 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
2433 if (IS_ERR(inner_map_meta))
2434 return inner_map_meta;
2436 map = htab_map_alloc(attr);
2438 bpf_map_meta_free(inner_map_meta);
2442 map->inner_map_meta = inner_map_meta;
2447 static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
2449 struct bpf_map **inner_map = htab_map_lookup_elem(map, key);
2454 return READ_ONCE(*inner_map);
2457 static int htab_of_map_gen_lookup(struct bpf_map *map,
2458 struct bpf_insn *insn_buf)
2460 struct bpf_insn *insn = insn_buf;
2461 const int ret = BPF_REG_0;
2463 BUILD_BUG_ON(!__same_type(&__htab_map_lookup_elem,
2464 (void *(*)(struct bpf_map *map, void *key))NULL));
2465 *insn++ = BPF_EMIT_CALL(__htab_map_lookup_elem);
2466 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 2);
2467 *insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
2468 offsetof(struct htab_elem, key) +
2469 round_up(map->key_size, 8));
2470 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
2472 return insn - insn_buf;
2475 static void htab_of_map_free(struct bpf_map *map)
2477 bpf_map_meta_free(map->inner_map_meta);
2478 fd_htab_map_free(map);
2481 const struct bpf_map_ops htab_of_maps_map_ops = {
2482 .map_alloc_check = fd_htab_map_alloc_check,
2483 .map_alloc = htab_of_map_alloc,
2484 .map_free = htab_of_map_free,
2485 .map_get_next_key = htab_map_get_next_key,
2486 .map_lookup_elem = htab_of_map_lookup_elem,
2487 .map_delete_elem = htab_map_delete_elem,
2488 .map_fd_get_ptr = bpf_map_fd_get_ptr,
2489 .map_fd_put_ptr = bpf_map_fd_put_ptr,
2490 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
2491 .map_gen_lookup = htab_of_map_gen_lookup,
2492 .map_check_btf = map_check_no_btf,
2494 .map_btf_id = &htab_map_btf_ids[0],