#include <string.h>
#include <errno.h>
-#ifdef HAVE_SYS_AUXV_H
-#include <sys/auxv.h>
-#endif
-
#include "util.h"
#include "hashmap.h"
#include "macro.h"
+#include "siphash24.h"
#define INITIAL_N_BUCKETS 31
struct hashmap_entry ** buckets;
unsigned n_buckets, n_entries;
- unsigned random_xor;
- bool from_pool;
+ uint8_t hash_key[HASH_KEY_SIZE];
+ bool from_pool:1;
};
struct pool {
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;
#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 uint64_hash_func(const void *p) {
+unsigned long uint64_hash_func(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]) {
uint64_t u;
-
- assert_cc(sizeof(uint64_t) == 2*sizeof(unsigned));
-
- u = *(const uint64_t*) p;
-
- return (unsigned) ((u >> 32) ^ 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 (h->hash_func(p) ^ h->random_xor) % h->n_buckets;
+ 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;
size_t size;
- void *auxv;
b = is_main_thread();
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->from_pool = b;
- /* Let's randomize our hash functions a bit so that they are
- * harder to guess for clients. For this, start out by cheaply
- * using some bits the kernel passed into the process using
- * the auxiliary vector. If the hashmap grows later on we will
- * rehash everything using a new random XOR mask from
- * /dev/random. */
-#ifdef HAVE_SYS_AUXV_H
- auxv = (void*) getauxval(AT_RANDOM);
- h->random_xor = auxv ? *(unsigned*) auxv : random_u();
-#else
- h->random_xor = random_u();
-#endif
+ get_hash_key(h->hash_key, true);
return h;
}
static bool resize_buckets(Hashmap *h) {
struct hashmap_entry **n, *i;
- unsigned m, nxor;
+ unsigned m;
+ uint8_t nkey[HASH_KEY_SIZE];
assert(h);
if (!n)
return false;
- /* Let's use a different randomized xor value for the
+ /* Let's use a different randomized hash key for the
* extension, so that people cannot guess what we are using
* here forever */
- nxor = random_u();
+ get_hash_key(nkey, false);
for (i = h->iterate_list_head; i; i = i->iterate_next) {
- unsigned hash, x;
+ unsigned long old_bucket, new_bucket;
- hash = h->hash_func(i->key);
+ old_bucket = h->hash_func(i->key, h->hash_key) % h->n_buckets;
/* First, drop from old bucket table */
if (i->bucket_next)
if (i->bucket_previous)
i->bucket_previous->bucket_next = i->bucket_next;
else
- h->buckets[(hash ^ h->random_xor) % h->n_buckets] = i->bucket_next;
+ h->buckets[old_bucket] = i->bucket_next;
/* Then, add to new backet table */
- x = (hash ^ nxor) % m;
+ new_bucket = h->hash_func(i->key, nkey) % m;
- i->bucket_next = n[x];
+ i->bucket_next = n[new_bucket];
i->bucket_previous = NULL;
- if (n[x])
- n[x]->bucket_previous = i;
- n[x] = i;
+ 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))))
h->buckets = n;
h->n_buckets = m;
- h->random_xor = nxor;
+
+ memcpy(h->hash_key, nkey, HASH_KEY_SIZE);
return true;
}
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);
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;
+ }
+
+ 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;
+}
+
int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, void *value) {
struct hashmap_entry *e;
unsigned old_hash, new_hash;