X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fshared%2Fhashmap.c;h=6f5f8204dd3dd0d2c733862edaa0b48662af3595;hp=ef78070f4c0854cd767e7a2803317fccb51b7691;hb=7ad63f57b6ce7ae9e3cc19dcb441f0a4494fa3f2;hpb=d99ae53a7327e1520ea4b9a3408c2d7f938c4b37 diff --git a/src/shared/hashmap.c b/src/shared/hashmap.c index ef78070f4..6f5f8204d 100644 --- a/src/shared/hashmap.c +++ b/src/shared/hashmap.c @@ -27,8 +27,10 @@ #include "util.h" #include "hashmap.h" #include "macro.h" +#include "siphash24.h" +#include "mempool.h" -#define NBUCKETS 127 +#define INITIAL_N_BUCKETS 31 struct hashmap_entry { const void *key; @@ -38,157 +40,176 @@ struct hashmap_entry { }; struct Hashmap { - hash_func_t hash_func; - compare_func_t compare_func; + const struct hash_ops *hash_ops; struct hashmap_entry *iterate_list_head, *iterate_list_tail; - unsigned n_entries; - bool from_pool; -}; + struct hashmap_entry ** buckets; + unsigned n_buckets, n_entries; -#define BY_HASH(h) ((struct hashmap_entry**) ((uint8_t*) (h) + ALIGN(sizeof(Hashmap)))) + uint8_t hash_key[HASH_KEY_SIZE]; + bool from_pool:1; +}; -struct pool { - struct pool *next; - unsigned n_tiles; - unsigned n_used; +struct hashmap_tile { + Hashmap h; + struct hashmap_entry *initial_buckets[INITIAL_N_BUCKETS]; }; -static struct pool *first_hashmap_pool = NULL; -static void *first_hashmap_tile = NULL; +static DEFINE_MEMPOOL(hashmap_pool, struct hashmap_tile, 8); +static DEFINE_MEMPOOL(hashmap_entry_pool, struct hashmap_entry, 64); -static struct pool *first_entry_pool = NULL; -static void *first_entry_tile = NULL; +#ifdef VALGRIND -static void* allocate_tile(struct pool **first_pool, void **first_tile, size_t tile_size) { - unsigned i; +__attribute__((destructor)) static void cleanup_pools(void) { + /* Be nice to valgrind */ - if (*first_tile) { - void *r; + mempool_drop(&hashmap_entry_pool); + mempool_drop(&hashmap_pool); +} - r = *first_tile; - *first_tile = * (void**) (*first_tile); - return r; - } +#endif - if (_unlikely_(!*first_pool) || _unlikely_((*first_pool)->n_used >= (*first_pool)->n_tiles)) { - unsigned n; - size_t size; - struct pool *p; +unsigned long string_hash_func(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]) { + uint64_t u; + siphash24((uint8_t*) &u, p, strlen(p), hash_key); + return (unsigned long) u; +} - n = *first_pool ? (*first_pool)->n_tiles : 0; - n = MAX(512U, n * 2); - size = PAGE_ALIGN(ALIGN(sizeof(struct pool)) + n*tile_size); - n = (size - ALIGN(sizeof(struct pool))) / tile_size; +int string_compare_func(const void *a, const void *b) { + return strcmp(a, b); +} - p = malloc(size); - if (!p) - return NULL; +const struct hash_ops string_hash_ops = { + .hash = string_hash_func, + .compare = string_compare_func +}; - p->next = *first_pool; - p->n_tiles = n; - p->n_used = 0; +unsigned long trivial_hash_func(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]) { + uint64_t u; + siphash24((uint8_t*) &u, &p, sizeof(p), hash_key); + return (unsigned long) u; +} - *first_pool = p; - } +int trivial_compare_func(const void *a, const void *b) { + return a < b ? -1 : (a > b ? 1 : 0); +} - i = (*first_pool)->n_used++; +const struct hash_ops trivial_hash_ops = { + .hash = trivial_hash_func, + .compare = trivial_compare_func +}; - return ((uint8_t*) (*first_pool)) + ALIGN(sizeof(struct pool)) + i*tile_size; +unsigned long uint64_hash_func(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]) { + uint64_t u; + siphash24((uint8_t*) &u, p, sizeof(uint64_t), hash_key); + return (unsigned long) u; } -static void deallocate_tile(void **first_tile, void *p) { - * (void**) p = *first_tile; - *first_tile = p; +int uint64_compare_func(const void *_a, const void *_b) { + uint64_t a, b; + a = *(const uint64_t*) _a; + b = *(const uint64_t*) _b; + return a < b ? -1 : (a > b ? 1 : 0); } -#ifndef __OPTIMIZE__ +const struct hash_ops uint64_hash_ops = { + .hash = uint64_hash_func, + .compare = uint64_compare_func +}; -static void drop_pool(struct pool *p) { - while (p) { - struct pool *n; - n = p->next; - free(p); - p = n; - } +#if SIZEOF_DEV_T != 8 +unsigned long devt_hash_func(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]) { + uint64_t u; + siphash24((uint8_t*) &u, p, sizeof(dev_t), hash_key); + return (unsigned long) u; } -__attribute__((destructor)) static void cleanup_pool(void) { - /* Be nice to valgrind */ - - drop_pool(first_hashmap_pool); - drop_pool(first_entry_pool); +int devt_compare_func(const void *_a, const void *_b) { + dev_t a, b; + a = *(const dev_t*) _a; + b = *(const dev_t*) _b; + return a < b ? -1 : (a > b ? 1 : 0); } +const struct hash_ops devt_hash_ops = { + .hash = devt_hash_func, + .compare = devt_compare_func +}; #endif -unsigned string_hash_func(const void *p) { - unsigned hash = 5381; - const signed char *c; - - /* DJB's hash function */ - - for (c = p; *c; c++) - hash = (hash << 5) + hash + (unsigned) *c; - - return hash; +static unsigned bucket_hash(Hashmap *h, const void *p) { + return (unsigned) (h->hash_ops->hash(p, h->hash_key) % h->n_buckets); } -int string_compare_func(const void *a, const void *b) { - return strcmp(a, b); -} +static void get_hash_key(uint8_t hash_key[HASH_KEY_SIZE], bool reuse_is_ok) { + static uint8_t current[HASH_KEY_SIZE]; + static bool current_initialized = false; -unsigned trivial_hash_func(const void *p) { - return PTR_TO_UINT(p); -} + /* Returns a hash function key to use. In order to keep things + * fast we will not generate a new key each time we allocate a + * new hash table. Instead, we'll just reuse the most recently + * generated one, except if we never generated one or when we + * are rehashing an entire hash table because we reached a + * fill level */ -int trivial_compare_func(const void *a, const void *b) { - return a < b ? -1 : (a > b ? 1 : 0); + if (!current_initialized || !reuse_is_ok) { + random_bytes(current, sizeof(current)); + current_initialized = true; + } + + memcpy(hash_key, current, sizeof(current)); } -Hashmap *hashmap_new(hash_func_t hash_func, compare_func_t compare_func) { +Hashmap *hashmap_new(const struct hash_ops *hash_ops) { bool b; + struct hashmap_tile *ht; Hashmap *h; - size_t size; b = is_main_thread(); - size = ALIGN(sizeof(Hashmap)) + NBUCKETS * sizeof(struct hashmap_entry*); - if (b) { - h = allocate_tile(&first_hashmap_pool, &first_hashmap_tile, size); - if (!h) + ht = mempool_alloc_tile(&hashmap_pool); + if (!ht) return NULL; - memset(h, 0, size); + memzero(ht, sizeof(struct hashmap_tile)); } else { - h = malloc0(size); + ht = malloc0(sizeof(struct hashmap_tile)); - if (!h) + if (!ht) return NULL; } - h->hash_func = hash_func ? hash_func : trivial_hash_func; - h->compare_func = compare_func ? compare_func : trivial_compare_func; + h = &ht->h; + h->hash_ops = hash_ops ? hash_ops : &trivial_hash_ops; + h->n_buckets = INITIAL_N_BUCKETS; h->n_entries = 0; h->iterate_list_head = h->iterate_list_tail = NULL; + h->buckets = ht->initial_buckets; + h->from_pool = b; + get_hash_key(h->hash_key, true); + return h; } -int hashmap_ensure_allocated(Hashmap **h, hash_func_t hash_func, compare_func_t compare_func) { +int hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops) { + Hashmap *q; + assert(h); if (*h) return 0; - if (!(*h = hashmap_new(hash_func, compare_func))) + q = hashmap_new(hash_ops); + if (!q) return -ENOMEM; + *h = q; return 0; } @@ -197,11 +218,11 @@ static void link_entry(Hashmap *h, struct hashmap_entry *e, unsigned hash) { assert(e); /* Insert into hash table */ - e->bucket_next = BY_HASH(h)[hash]; + e->bucket_next = h->buckets[hash]; e->bucket_previous = NULL; - if (BY_HASH(h)[hash]) - BY_HASH(h)[hash]->bucket_previous = e; - BY_HASH(h)[hash] = e; + if (h->buckets[hash]) + h->buckets[hash]->bucket_previous = e; + h->buckets[hash] = e; /* Insert into iteration list */ e->iterate_previous = h->iterate_list_tail; @@ -241,7 +262,7 @@ static void unlink_entry(Hashmap *h, struct hashmap_entry *e, unsigned hash) { if (e->bucket_previous) e->bucket_previous->bucket_next = e->bucket_next; else - BY_HASH(h)[hash] = e->bucket_next; + h->buckets[hash] = e->bucket_next; assert(h->n_entries >= 1); h->n_entries--; @@ -253,12 +274,11 @@ static void remove_entry(Hashmap *h, struct hashmap_entry *e) { assert(h); assert(e); - hash = h->hash_func(e->key) % NBUCKETS; - + hash = bucket_hash(h, e->key); unlink_entry(h, e, hash); if (h->from_pool) - deallocate_tile(&first_entry_tile, e); + mempool_free_tile(&hashmap_entry_pool, e); else free(e); } @@ -272,8 +292,11 @@ void hashmap_free(Hashmap*h) { hashmap_clear(h); + if (h->buckets != (struct hashmap_entry**) ((uint8_t*) h + ALIGN(sizeof(Hashmap)))) + free(h->buckets); + if (h->from_pool) - deallocate_tile(&first_hashmap_tile, h); + mempool_free_tile(&hashmap_pool, container_of(h, struct hashmap_tile, h)); else free(h); } @@ -290,6 +313,17 @@ void hashmap_free_free(Hashmap *h) { hashmap_free(h); } +void hashmap_free_free_free(Hashmap *h) { + + /* Free the hashmap and all data and key objects in it */ + + if (!h) + return; + + hashmap_clear_free_free(h); + hashmap_free(h); +} + void hashmap_clear(Hashmap *h) { if (!h) return; @@ -308,37 +342,109 @@ void hashmap_clear_free(Hashmap *h) { free(p); } +void hashmap_clear_free_free(Hashmap *h) { + if (!h) + return; + + while (h->iterate_list_head) { + void *a, *b; + + a = h->iterate_list_head->value; + b = (void*) h->iterate_list_head->key; + remove_entry(h, h->iterate_list_head); + free(a); + free(b); + } +} + static struct hashmap_entry *hash_scan(Hashmap *h, unsigned hash, const void *key) { struct hashmap_entry *e; assert(h); - assert(hash < NBUCKETS); + assert(hash < h->n_buckets); - for (e = BY_HASH(h)[hash]; e; e = e->bucket_next) - if (h->compare_func(e->key, key) == 0) + for (e = h->buckets[hash]; e; e = e->bucket_next) + if (h->hash_ops->compare(e->key, key) == 0) return e; return NULL; } -int hashmap_put(Hashmap *h, const void *key, void *value) { - struct hashmap_entry *e; - unsigned hash; +static int resize_buckets(Hashmap *h, unsigned entries_add) { + struct hashmap_entry **n, *i; + unsigned m, new_n_entries, new_n_buckets; + uint8_t nkey[HASH_KEY_SIZE]; assert(h); - hash = h->hash_func(key) % NBUCKETS; + new_n_entries = h->n_entries + entries_add; - e = hash_scan(h, hash, key); - if (e) { + /* overflow? */ + if (_unlikely_(new_n_entries < entries_add || new_n_entries > UINT_MAX / 4)) + return -ENOMEM; - if (e->value == value) - return 0; + new_n_buckets = new_n_entries * 4 / 3; - return -EEXIST; + if (_likely_(new_n_buckets <= h->n_buckets)) + return 0; + + /* Increase by four at least */ + m = MAX((h->n_entries+1)*4-1, new_n_buckets); + + /* If we hit OOM we simply risk packed hashmaps... */ + n = new0(struct hashmap_entry*, m); + if (!n) + return -ENOMEM; + + /* Let's use a different randomized hash key for the + * extension, so that people cannot guess what we are using + * here forever */ + get_hash_key(nkey, false); + + for (i = h->iterate_list_head; i; i = i->iterate_next) { + unsigned long old_bucket, new_bucket; + + old_bucket = h->hash_ops->hash(i->key, h->hash_key) % h->n_buckets; + + /* First, drop from old bucket table */ + if (i->bucket_next) + i->bucket_next->bucket_previous = i->bucket_previous; + + if (i->bucket_previous) + i->bucket_previous->bucket_next = i->bucket_next; + else + h->buckets[old_bucket] = i->bucket_next; + + /* Then, add to new backet table */ + new_bucket = h->hash_ops->hash(i->key, nkey) % m; + + i->bucket_next = n[new_bucket]; + i->bucket_previous = NULL; + if (n[new_bucket]) + n[new_bucket]->bucket_previous = i; + n[new_bucket] = i; } + if (h->buckets != (struct hashmap_entry**) ((uint8_t*) h + ALIGN(sizeof(Hashmap)))) + free(h->buckets); + + h->buckets = n; + h->n_buckets = m; + + memcpy(h->hash_key, nkey, HASH_KEY_SIZE); + + return 1; +} + +static int __hashmap_put(Hashmap *h, const void *key, void *value, unsigned hash) { + /* For when we know no such entry exists yet */ + + struct hashmap_entry *e; + + if (resize_buckets(h, 1) > 0) + hash = bucket_hash(h, key); + if (h->from_pool) - e = allocate_tile(&first_entry_pool, &first_entry_tile, sizeof(struct hashmap_entry)); + e = mempool_alloc_tile(&hashmap_entry_pool); else e = new(struct hashmap_entry, 1); @@ -353,13 +459,30 @@ int hashmap_put(Hashmap *h, const void *key, void *value) { return 1; } +int hashmap_put(Hashmap *h, const void *key, void *value) { + struct hashmap_entry *e; + unsigned hash; + + assert(h); + + hash = bucket_hash(h, key); + e = hash_scan(h, hash, key); + if (e) { + if (e->value == value) + return 0; + return -EEXIST; + } + + return __hashmap_put(h, key, value, hash); +} + int hashmap_replace(Hashmap *h, const void *key, void *value) { struct hashmap_entry *e; unsigned hash; assert(h); - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); e = hash_scan(h, hash, key); if (e) { e->key = key; @@ -367,7 +490,7 @@ int hashmap_replace(Hashmap *h, const void *key, void *value) { return 0; } - return hashmap_put(h, key, value); + return __hashmap_put(h, key, value, hash); } int hashmap_update(Hashmap *h, const void *key, void *value) { @@ -376,7 +499,7 @@ int hashmap_update(Hashmap *h, const void *key, void *value) { assert(h); - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); e = hash_scan(h, hash, key); if (!e) return -ENOENT; @@ -392,7 +515,7 @@ void* hashmap_get(Hashmap *h, const void *key) { if (!h) return NULL; - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); e = hash_scan(h, hash, key); if (!e) return NULL; @@ -407,7 +530,7 @@ void* hashmap_get2(Hashmap *h, const void *key, void **key2) { if (!h) return NULL; - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); e = hash_scan(h, hash, key); if (!e) return NULL; @@ -424,12 +547,8 @@ bool hashmap_contains(Hashmap *h, const void *key) { if (!h) return false; - hash = h->hash_func(key) % NBUCKETS; - - if (!hash_scan(h, hash, key)) - return false; - - return true; + hash = bucket_hash(h, key); + return !!hash_scan(h, hash, key); } void* hashmap_remove(Hashmap *h, const void *key) { @@ -440,12 +559,40 @@ void* hashmap_remove(Hashmap *h, const void *key) { if (!h) return NULL; - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); + e = hash_scan(h, hash, key); + if (!e) + return NULL; + + data = e->value; + remove_entry(h, e); + + return data; +} + +void* hashmap_remove2(Hashmap *h, const void *key, void **rkey) { + struct hashmap_entry *e; + unsigned hash; + void *data; - if (!(e = hash_scan(h, hash, key))) + if (!h) { + if (rkey) + *rkey = NULL; return NULL; + } + + hash = bucket_hash(h, key); + e = hash_scan(h, hash, key); + if (!e) { + if (rkey) + *rkey = NULL; + return NULL; + } data = e->value; + if (rkey) + *rkey = (void*) e->key; + remove_entry(h, e); return data; @@ -458,11 +605,12 @@ int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, if (!h) return -ENOENT; - old_hash = h->hash_func(old_key) % NBUCKETS; - if (!(e = hash_scan(h, old_hash, old_key))) + old_hash = bucket_hash(h, old_key); + e = hash_scan(h, old_hash, old_key); + if (!e) return -ENOENT; - new_hash = h->hash_func(new_key) % NBUCKETS; + new_hash = bucket_hash(h, new_key); if (hash_scan(h, new_hash, new_key)) return -EEXIST; @@ -483,13 +631,14 @@ int hashmap_remove_and_replace(Hashmap *h, const void *old_key, const void *new_ if (!h) return -ENOENT; - old_hash = h->hash_func(old_key) % NBUCKETS; - if (!(e = hash_scan(h, old_hash, old_key))) + old_hash = bucket_hash(h, old_key); + e = hash_scan(h, old_hash, old_key); + if (!e) return -ENOENT; - new_hash = h->hash_func(new_key) % NBUCKETS; - - if ((k = hash_scan(h, new_hash, new_key))) + new_hash = bucket_hash(h, new_key); + k = hash_scan(h, new_hash, new_key); + if (k) if (e != k) remove_entry(h, k); @@ -510,9 +659,10 @@ void* hashmap_remove_value(Hashmap *h, const void *key, void *value) { if (!h) return NULL; - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); - if (!(e = hash_scan(h, hash, key))) + e = hash_scan(h, hash, key); + if (!e) return NULL; if (e->value != value) @@ -558,58 +708,6 @@ at_end: return NULL; } -void *hashmap_iterate_backwards(Hashmap *h, Iterator *i, const void **key) { - struct hashmap_entry *e; - - assert(i); - - if (!h) - goto at_beginning; - - if (*i == ITERATOR_FIRST) - goto at_beginning; - - if (*i == ITERATOR_LAST && !h->iterate_list_tail) - goto at_beginning; - - e = *i == ITERATOR_LAST ? h->iterate_list_tail : (struct hashmap_entry*) *i; - - if (e->iterate_previous) - *i = (Iterator) e->iterate_previous; - else - *i = ITERATOR_FIRST; - - if (key) - *key = e->key; - - return e->value; - -at_beginning: - *i = ITERATOR_FIRST; - - if (key) - *key = NULL; - - return NULL; -} - -void *hashmap_iterate_skip(Hashmap *h, const void *key, Iterator *i) { - unsigned hash; - struct hashmap_entry *e; - - if (!h) - return NULL; - - hash = h->hash_func(key) % NBUCKETS; - - if (!(e = hash_scan(h, hash, key))) - return NULL; - - *i = (Iterator) e; - - return e->value; -} - void* hashmap_first(Hashmap *h) { if (!h) @@ -632,17 +730,6 @@ void* hashmap_first_key(Hashmap *h) { return (void*) h->iterate_list_head->key; } -void* hashmap_last(Hashmap *h) { - - if (!h) - return NULL; - - if (!h->iterate_list_tail) - return NULL; - - return h->iterate_list_tail->value; -} - void* hashmap_steal_first(Hashmap *h) { void *data; @@ -681,6 +768,14 @@ unsigned hashmap_size(Hashmap *h) { return h->n_entries; } +unsigned hashmap_buckets(Hashmap *h) { + + if (!h) + return 0; + + return h->n_buckets; +} + bool hashmap_isempty(Hashmap *h) { if (!h) @@ -700,57 +795,70 @@ int hashmap_merge(Hashmap *h, Hashmap *other) { for (e = other->iterate_list_head; e; e = e->iterate_next) { int r; - if ((r = hashmap_put(h, e->key, e->value)) < 0) - if (r != -EEXIST) - return r; + r = hashmap_put(h, e->key, e->value); + if (r < 0 && r != -EEXIST) + return r; } return 0; } -void hashmap_move(Hashmap *h, Hashmap *other) { +int hashmap_reserve(Hashmap *h, unsigned entries_add) { + int r; + + assert(h); + + r = resize_buckets(h, entries_add); + if (r < 0) + return r; + + return 0; +} + +int hashmap_move(Hashmap *h, Hashmap *other) { struct hashmap_entry *e, *n; assert(h); /* The same as hashmap_merge(), but every new item from other - * is moved to h. This function is guaranteed to succeed. */ + * is moved to h. */ if (!other) - return; + return 0; for (e = other->iterate_list_head; e; e = n) { unsigned h_hash, other_hash; n = e->iterate_next; - h_hash = h->hash_func(e->key) % NBUCKETS; - + h_hash = bucket_hash(h, e->key); if (hash_scan(h, h_hash, e->key)) continue; - other_hash = other->hash_func(e->key) % NBUCKETS; - + other_hash = bucket_hash(other, e->key); unlink_entry(other, e, other_hash); link_entry(h, e, h_hash); } + + return 0; } int hashmap_move_one(Hashmap *h, Hashmap *other, const void *key) { unsigned h_hash, other_hash; struct hashmap_entry *e; - if (!other) - return 0; - assert(h); - h_hash = h->hash_func(key) % NBUCKETS; + h_hash = bucket_hash(h, key); if (hash_scan(h, h_hash, key)) return -EEXIST; - other_hash = other->hash_func(key) % NBUCKETS; - if (!(e = hash_scan(other, other_hash, key))) + if (!other) + return -ENOENT; + + other_hash = bucket_hash(other, key); + e = hash_scan(other, other_hash, key); + if (!e) return -ENOENT; unlink_entry(other, e, other_hash); @@ -764,7 +872,8 @@ Hashmap *hashmap_copy(Hashmap *h) { assert(h); - if (!(copy = hashmap_new(h->hash_func, h->compare_func))) + copy = hashmap_new(h->hash_ops); + if (!copy) return NULL; if (hashmap_merge(copy, h) < 0) { @@ -797,13 +906,12 @@ void *hashmap_next(Hashmap *h, const void *key) { unsigned hash; struct hashmap_entry *e; - assert(h); assert(key); if (!h) return NULL; - hash = h->hash_func(key) % NBUCKETS; + hash = bucket_hash(h, key); e = hash_scan(h, hash, key); if (!e) return NULL;