#include "util.h"
#include "hashmap.h"
#include "macro.h"
+#include "siphash24.h"
-#define NBUCKETS 127
+#define INITIAL_N_BUCKETS 31
struct hashmap_entry {
const void *key;
compare_func_t compare_func;
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;
static struct pool *first_entry_pool = NULL;
static void *first_entry_tile = NULL;
-static void* allocate_tile(struct pool **first_pool, void **first_tile, size_t tile_size) {
+static void* allocate_tile(struct pool **first_pool, void **first_tile, size_t tile_size, unsigned at_least) {
unsigned i;
+ /* When a tile is released we add it to the list and simply
+ * place the next pointer at its offset 0. */
+
+ assert(tile_size >= sizeof(void*));
+ assert(at_least > 0);
+
if (*first_tile) {
void *r;
struct pool *p;
n = *first_pool ? (*first_pool)->n_tiles : 0;
- n = MAX(512U, n * 2);
+ n = MAX(at_least, n * 2);
size = PAGE_ALIGN(ALIGN(sizeof(struct pool)) + n*tile_size);
n = (size - ALIGN(sizeof(struct pool))) / tile_size;
*first_tile = p;
}
-#ifndef __OPTIMIZE__
+#ifdef VALGRIND
static void drop_pool(struct pool *p) {
while (p) {
#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;
+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;
}
int string_compare_func(const void *a, const void *b) {
return strcmp(a, b);
}
-unsigned trivial_hash_func(const void *p) {
- return PTR_TO_UINT(p);
+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;
}
int trivial_compare_func(const void *a, const void *b) {
return a < b ? -1 : (a > b ? 1 : 0);
}
+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;
+}
+
+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);
+}
+
+static unsigned bucket_hash(Hashmap *h, const void *p) {
+ return (unsigned) (h->hash_func(p, h->hash_key) % h->n_buckets);
+}
+
+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;
+
+ /* 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 */
+
+ 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) {
bool b;
Hashmap *h;
b = is_main_thread();
- size = ALIGN(sizeof(Hashmap)) + NBUCKETS * sizeof(struct hashmap_entry*);
+ size = ALIGN(sizeof(Hashmap)) + INITIAL_N_BUCKETS * sizeof(struct hashmap_entry*);
if (b) {
- h = allocate_tile(&first_hashmap_pool, &first_hashmap_tile, size);
+ h = allocate_tile(&first_hashmap_pool, &first_hashmap_tile, size, 8);
if (!h)
return NULL;
- memset(h, 0, size);
+ memzero(h, size);
} else {
h = malloc0(size);
h->hash_func = hash_func ? hash_func : trivial_hash_func;
h->compare_func = compare_func ? compare_func : trivial_compare_func;
+ h->n_buckets = INITIAL_N_BUCKETS;
h->n_entries = 0;
h->iterate_list_head = h->iterate_list_tail = NULL;
+ h->buckets = (struct hashmap_entry**) ((uint8_t*) h + ALIGN(sizeof(Hashmap)));
+
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) {
+ Hashmap *q;
+
assert(h);
if (*h)
return 0;
- if (!(*h = hashmap_new(hash_func, compare_func)))
+ q = hashmap_new(hash_func, compare_func);
+ if (!q)
return -ENOMEM;
+ *h = q;
return 0;
}
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;
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--;
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)
void hashmap_free(Hashmap*h) {
+ /* Free the hashmap, but nothing in it */
+
if (!h)
return;
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);
else
}
void hashmap_free_free(Hashmap *h) {
+
+ /* Free the hashmap and all data objects in it, but not the
+ * keys */
+
+ if (!h)
+ return;
+
hashmap_clear_free(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;
void hashmap_clear_free(Hashmap *h) {
void *p;
- assert(h);
+ if (!h)
+ return;
while ((p = hashmap_steal_first(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)
+ for (e = h->buckets[hash]; e; e = e->bucket_next)
if (h->compare_func(e->key, key) == 0)
return e;
return NULL;
}
+static bool resize_buckets(Hashmap *h) {
+ struct hashmap_entry **n, *i;
+ unsigned m;
+ uint8_t nkey[HASH_KEY_SIZE];
+
+ assert(h);
+
+ if (_likely_(h->n_entries*4 < h->n_buckets*3))
+ return false;
+
+ /* Increase by four */
+ m = (h->n_entries+1)*4-1;
+
+ /* If we hit OOM we simply risk packed hashmaps... */
+ n = new0(struct hashmap_entry*, m);
+ if (!n)
+ return false;
+
+ /* 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_func(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_func(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 true;
+}
+
int hashmap_put(Hashmap *h, const void *key, void *value) {
struct hashmap_entry *e;
unsigned hash;
assert(h);
- hash = h->hash_func(key) % NBUCKETS;
-
- if ((e = hash_scan(h, hash, key))) {
-
+ hash = bucket_hash(h, key);
+ e = hash_scan(h, hash, key);
+ if (e) {
if (e->value == value)
return 0;
-
return -EEXIST;
}
+ if (resize_buckets(h))
+ hash = bucket_hash(h, key);
+
if (h->from_pool)
- e = allocate_tile(&first_entry_pool, &first_entry_tile, sizeof(struct hashmap_entry));
+ e = allocate_tile(&first_entry_pool, &first_entry_tile, sizeof(struct hashmap_entry), 64U);
else
e = new(struct hashmap_entry, 1);
assert(h);
- hash = h->hash_func(key) % NBUCKETS;
-
- if ((e = hash_scan(h, hash, key))) {
+ hash = bucket_hash(h, key);
+ e = hash_scan(h, hash, key);
+ if (e) {
e->key = key;
e->value = value;
return 0;
return hashmap_put(h, key, value);
}
+int hashmap_update(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)
+ return -ENOENT;
+
+ e->value = value;
+ return 0;
+}
+
void* hashmap_get(Hashmap *h, const void *key) {
unsigned hash;
struct hashmap_entry *e;
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;
+
+ return e->value;
+}
+
+void* hashmap_get2(Hashmap *h, const void *key, void **key2) {
+ unsigned hash;
+ struct hashmap_entry *e;
+
+ if (!h)
+ return NULL;
- if (!(e = hash_scan(h, hash, key)))
+ hash = bucket_hash(h, key);
+ e = hash_scan(h, hash, key);
+ if (!e)
return NULL;
+ if (key2)
+ *key2 = (void*) e->key;
+
return e->value;
}
+bool hashmap_contains(Hashmap *h, const void *key) {
+ unsigned hash;
+
+ if (!h)
+ return false;
+
+ hash = bucket_hash(h, key);
+ return !!hash_scan(h, hash, key);
+}
+
void* hashmap_remove(Hashmap *h, const void *key) {
struct hashmap_entry *e;
unsigned hash;
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 (!h) {
+ if (rkey)
+ *rkey = NULL;
+ return NULL;
+ }
- if (!(e = hash_scan(h, hash, key)))
+ 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;
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;
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);
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)
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;
*i = (Iterator) e;
return h->n_entries;
}
+unsigned hashmap_buckets(Hashmap *h) {
+
+ if (!h)
+ return 0;
+
+ return h->n_buckets;
+}
+
bool hashmap_isempty(Hashmap *h) {
if (!h)
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;
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);
}
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)))
+ other_hash = bucket_hash(other, key);
+ e = hash_scan(other, other_hash, key);
+ if (!e)
return -ENOENT;
unlink_entry(other, e, other_hash);
assert(h);
- if (!(copy = hashmap_new(h->hash_func, h->compare_func)))
+ copy = hashmap_new(h->hash_func, h->compare_func);
+ if (!copy)
return NULL;
if (hashmap_merge(copy, h) < 0) {
return sv;
}
+
+void *hashmap_next(Hashmap *h, const void *key) {
+ unsigned hash;
+ struct hashmap_entry *e;
+
+ assert(h);
+ assert(key);
+
+ if (!h)
+ return NULL;
+
+ hash = bucket_hash(h, key);
+ e = hash_scan(h, hash, key);
+ if (!e)
+ return NULL;
+
+ e = e->iterate_next;
+ if (!e)
+ return NULL;
+
+ return e->value;
+}