libsepol/cil: use DJB2a string hash function

The hash table implementation uses `& (h->size - 1)` to truncate
generated hashes to the number of buckets.  This operation is equal to
`% h->size` if and only if the size is a power of two (which seems to be
always the case).  One property of the binary and with a power of two
(and probably a small one <=2048) is all higher bits are discarded.
Thus a hash function is needed with a good avalanche effect, which the
current one is not.

Benchmark of building dssp5:

    # Current
    Time (mean ± σ):      1.347 s ±  0.065 s    [User: 1.207 s, System: 0.138 s]
    Range (min … max):    1.274 s …  1.436 s    10 runs

    # Patch
    Time (mean ± σ):      1.336 s ±  0.029 s    [User: 1.195 s, System: 0.140 s]
    Range (min … max):    1.303 s …  1.376 s    10 runs

Signed-off-by: Christian Göttsche <cgzones@googlemail.com>
Acked-by: James Carter <jwcart2@gmail.com>
This commit is contained in:
Christian Göttsche 2023-08-16 14:38:43 +02:00 committed by James Carter
parent d03d506a30
commit 08be63572e
1 changed files with 6 additions and 9 deletions

View File

@ -47,16 +47,13 @@ static hashtab_t cil_strpool_tab = NULL;
static unsigned int cil_strpool_hash(hashtab_t h, const_hashtab_key_t key)
{
const char *p;
size_t size;
unsigned int val;
unsigned int hash = 5381;
unsigned char c;
val = 0;
size = strlen(key);
for (p = key; ((size_t) (p - key)) < size; p++)
val =
(val << 4 | (val >> (8 * sizeof(unsigned int) - 4))) ^ (*p);
return val & (h->size - 1);
while ((c = *(unsigned const char *)key++))
hash = ((hash << 5) + hash) ^ c;
return hash & (h->size - 1);
}
static int cil_strpool_compare(hashtab_t h __attribute__ ((unused)), const_hashtab_key_t key1, const_hashtab_key_t key2)