X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=src%2Fshared%2Fhashmap.c;h=8225b8ebccd5e742eea1f90b42091c1c48238254;hb=dd0124f6cb6ff7da1ce8be23ec7e1137fe15958d;hp=633079277d758f7961a2c854d84c0e861d6ffafa;hpb=45fa9e29f8c9759c8f2f4238fed956f695c73dc3;p=elogind.git

diff --git a/src/shared/hashmap.c b/src/shared/hashmap.c
index 633079277..8225b8ebc 100644
--- a/src/shared/hashmap.c
+++ b/src/shared/hashmap.c
@@ -27,6 +27,7 @@
 #include "util.h"
 #include "hashmap.h"
 #include "macro.h"
+#include "siphash24.h"
 
 #define INITIAL_N_BUCKETS 31
 
@@ -38,15 +39,15 @@ 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;
 
         struct hashmap_entry ** buckets;
         unsigned n_buckets, n_entries;
 
-        bool from_pool;
+        uint8_t hash_key[HASH_KEY_SIZE];
+        bool from_pool:1;
 };
 
 struct pool {
@@ -61,13 +62,14 @@ static void *first_hashmap_tile = NULL;
 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;
@@ -83,7 +85,7 @@ static void* allocate_tile(struct pool **first_pool, void **first_tile, size_t t
                 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;
 
@@ -128,50 +130,98 @@ __attribute__((destructor)) static void cleanup_pool(void) {
 
 #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);
+const struct hash_ops string_hash_ops = {
+        .hash = string_hash_func,
+        .compare = string_compare_func
+};
+
+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) {
-        uint64_t u;
-
-        assert_cc(sizeof(uint64_t) == 2*sizeof(unsigned));
-
-        u = *(const uint64_t*) p;
+const struct hash_ops trivial_hash_ops = {
+        .hash = trivial_hash_func,
+        .compare = trivial_compare_func
+};
 
-        return (unsigned) ((u >> 32) ^ u);
+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);
+}
 
+const struct hash_ops uint64_hash_ops = {
+        .hash = uint64_hash_func,
+        .compare = uint64_compare_func
+};
+
+#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;
+}
+
+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);
 }
 
-Hashmap *hashmap_new(hash_func_t hash_func, compare_func_t compare_func) {
+const struct hash_ops devt_hash_ops = {
+        .hash = devt_hash_func,
+        .compare = devt_compare_func
+};
+#endif
+
+static unsigned bucket_hash(Hashmap *h, const void *p) {
+        return (unsigned) (h->hash_ops->hash(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(const struct hash_ops *hash_ops) {
         bool b;
         Hashmap *h;
         size_t size;
@@ -181,11 +231,11 @@ Hashmap *hashmap_new(hash_func_t hash_func, compare_func_t compare_func) {
         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);
 
@@ -193,8 +243,7 @@ Hashmap *hashmap_new(hash_func_t hash_func, compare_func_t compare_func) {
                         return NULL;
         }
 
-        h->hash_func = hash_func ? hash_func : trivial_hash_func;
-        h->compare_func = compare_func ? compare_func : trivial_compare_func;
+        h->hash_ops = hash_ops ? hash_ops : &trivial_hash_ops;
 
         h->n_buckets = INITIAL_N_BUCKETS;
         h->n_entries = 0;
@@ -204,10 +253,12 @@ Hashmap *hashmap_new(hash_func_t hash_func, compare_func_t compare_func) {
 
         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);
@@ -215,7 +266,7 @@ int hashmap_ensure_allocated(Hashmap **h, hash_func_t hash_func, compare_func_t
         if (*h)
                 return 0;
 
-        q = hashmap_new(hash_func, compare_func);
+        q = hashmap_new(hash_ops);
         if (!q)
                 return -ENOMEM;
 
@@ -284,8 +335,7 @@ static void remove_entry(Hashmap *h, struct hashmap_entry *e) {
         assert(h);
         assert(e);
 
-        hash = h->hash_func(e->key) % h->n_buckets;
-
+        hash = bucket_hash(h, e->key);
         unlink_entry(h, e, hash);
 
         if (h->from_pool)
@@ -368,22 +418,22 @@ void hashmap_clear_free_free(Hashmap *h) {
         }
 }
 
-
 static struct hashmap_entry *hash_scan(Hashmap *h, unsigned hash, const void *key) {
         struct hashmap_entry *e;
         assert(h);
         assert(hash < h->n_buckets);
 
         for (e = h->buckets[hash]; e; e = e->bucket_next)
-                if (h->compare_func(e->key, key) == 0)
+                if (h->hash_ops->compare(e->key, key) == 0)
                         return e;
 
         return NULL;
 }
 
 static bool resize_buckets(Hashmap *h) {
-        unsigned m;
         struct hashmap_entry **n, *i;
+        unsigned m;
+        uint8_t nkey[HASH_KEY_SIZE];
 
         assert(h);
 
@@ -398,10 +448,15 @@ static bool resize_buckets(Hashmap *h) {
         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 hash, x;
+                unsigned long old_bucket, new_bucket;
 
-                hash = h->hash_func(i->key);
+                old_bucket = h->hash_ops->hash(i->key, h->hash_key) % h->n_buckets;
 
                 /* First, drop from old bucket table */
                 if (i->bucket_next)
@@ -410,16 +465,16 @@ static bool resize_buckets(Hashmap *h) {
                 if (i->bucket_previous)
                         i->bucket_previous->bucket_next = i->bucket_next;
                 else
-                        h->buckets[hash % h->n_buckets] = i->bucket_next;
+                        h->buckets[old_bucket] = i->bucket_next;
 
                 /* Then, add to new backet table */
-                x = hash % m;
+                new_bucket = h->hash_ops->hash(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))))
@@ -428,28 +483,21 @@ static bool resize_buckets(Hashmap *h) {
         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;
+static int __hashmap_put(Hashmap *h, const void *key, void *value, unsigned hash) {
+        /* For when we know no such entry exists yet */
 
-        assert(h);
-
-        hash = h->hash_func(key) % h->n_buckets;
-        e = hash_scan(h, hash, key);
-        if (e) {
-                if (e->value == value)
-                        return 0;
-                return -EEXIST;
-        }
+        struct hashmap_entry *e;
 
         if (resize_buckets(h))
-                hash = h->hash_func(key) % h->n_buckets;
+                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);
 
@@ -464,13 +512,30 @@ 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) % h->n_buckets;
+        hash = bucket_hash(h, key);
         e = hash_scan(h, hash, key);
         if (e) {
                 e->key = key;
@@ -478,7 +543,7 @@ 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) {
@@ -487,7 +552,7 @@ int hashmap_update(Hashmap *h, const void *key, void *value) {
 
         assert(h);
 
-        hash = h->hash_func(key) % h->n_buckets;
+        hash = bucket_hash(h, key);
         e = hash_scan(h, hash, key);
         if (!e)
                 return -ENOENT;
@@ -503,7 +568,7 @@ void* hashmap_get(Hashmap *h, const void *key) {
         if (!h)
                 return NULL;
 
-        hash = h->hash_func(key) % h->n_buckets;
+        hash = bucket_hash(h, key);
         e = hash_scan(h, hash, key);
         if (!e)
                 return NULL;
@@ -518,7 +583,7 @@ void* hashmap_get2(Hashmap *h, const void *key, void **key2) {
         if (!h)
                 return NULL;
 
-        hash = h->hash_func(key) % h->n_buckets;
+        hash = bucket_hash(h, key);
         e = hash_scan(h, hash, key);
         if (!e)
                 return NULL;
@@ -535,12 +600,8 @@ bool hashmap_contains(Hashmap *h, const void *key) {
         if (!h)
                 return false;
 
-        hash = h->hash_func(key) % h->n_buckets;
-
-        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) {
@@ -551,12 +612,40 @@ void* hashmap_remove(Hashmap *h, const void *key) {
         if (!h)
                 return NULL;
 
-        hash = h->hash_func(key) % h->n_buckets;
+        hash = bucket_hash(h, key);
+        e = hash_scan(h, hash, key);
+        if (!e)
+                return NULL;
+
+        data = e->value;
+        remove_entry(h, e);
+
+        return data;
+}
 
-        if (!(e = hash_scan(h, hash, key)))
+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;
@@ -569,11 +658,12 @@ 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) % h->n_buckets;
-        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) % h->n_buckets;
+        new_hash = bucket_hash(h, new_key);
         if (hash_scan(h, new_hash, new_key))
                 return -EEXIST;
 
@@ -594,12 +684,14 @@ 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) % h->n_buckets;
-        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) % h->n_buckets;
-        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);
 
@@ -620,7 +712,7 @@ void* hashmap_remove_value(Hashmap *h, const void *key, void *value) {
         if (!h)
                 return NULL;
 
-        hash = h->hash_func(key) % h->n_buckets;
+        hash = bucket_hash(h, key);
 
         e = hash_scan(h, hash, key);
         if (!e)
@@ -669,59 +761,6 @@ 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) % h->n_buckets;
-
-        e = hash_scan(h, hash, key);
-        if (!e)
-                return NULL;
-
-        *i = (Iterator) e;
-
-        return e->value;
-}
-
 void* hashmap_first(Hashmap *h) {
 
         if (!h)
@@ -744,17 +783,6 @@ 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;
 
@@ -820,9 +848,9 @@ 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;
@@ -844,13 +872,11 @@ void hashmap_move(Hashmap *h, Hashmap *other) {
 
                 n = e->iterate_next;
 
-                h_hash = h->hash_func(e->key) % h->n_buckets;
-
+                h_hash = bucket_hash(h, e->key);
                 if (hash_scan(h, h_hash, e->key))
                         continue;
 
-                other_hash = other->hash_func(e->key) % other->n_buckets;
-
+                other_hash = bucket_hash(other, e->key);
                 unlink_entry(other, e, other_hash);
                 link_entry(h, e, h_hash);
         }
@@ -860,16 +886,16 @@ 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) % h->n_buckets;
+        h_hash = bucket_hash(h, key);
         if (hash_scan(h, h_hash, key))
                 return -EEXIST;
 
-        other_hash = other->hash_func(key) % other->n_buckets;
+        if (!other)
+                return -ENOENT;
+
+        other_hash = bucket_hash(other, key);
         e = hash_scan(other, other_hash, key);
         if (!e)
                 return -ENOENT;
@@ -885,7 +911,7 @@ Hashmap *hashmap_copy(Hashmap *h) {
 
         assert(h);
 
-        copy = hashmap_new(h->hash_func, h->compare_func);
+        copy = hashmap_new(h->hash_ops);
         if (!copy)
                 return NULL;
 
@@ -919,13 +945,12 @@ 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) % h->n_buckets;
+        hash = bucket_hash(h, key);
         e = hash_scan(h, hash, key);
         if (!e)
                 return NULL;