size_t tdb_mutex_size(struct tdb_context *tdb)
{
size_t mutex_size;
if (!tdb_have_mutexes(tdb)) {
return 0;
}
mutex_size = sizeof(struct tdb_mutexes);
mutex_size += tdb->hash_size * sizeof(pthread_mutex_t);
return TDB_ALIGN(mutex_size, tdb->page_size);
}
/* expand the database at least size bytes by expanding the underlying
file and doing the mmap again if necessary */
int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
{
struct tdb_record rec;
tdb_off_t offset, new_size, top_size, map_size;
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
return -1;
}
/* must know about any previous expansions by another process */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
/* limit size in order to avoid using up huge amounts of memory for
* in memory tdbs if an oddball huge record creeps in */
if (size > 100 * 1024) {
top_size = tdb->map_size + size * 2;
} else {
top_size = tdb->map_size + size * 100;
}
/* always make room for at least top_size more records, and at
least 25% more space. if the DB is smaller than 100MiB,
otherwise grow it by 10% only. */
if (tdb->map_size > 100 * 1024 * 1024) {
map_size = tdb->map_size * 1.10;
} else {
map_size = tdb->map_size * 1.25;
}
/* Round the database up to a multiple of the page size */
new_size = MAX(top_size, map_size);
size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
if (!(tdb->flags & TDB_INTERNAL))
tdb_munmap(tdb);
/*
* We must ensure the file is unmapped before doing this
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* expand the file itself */
if (!(tdb->flags & TDB_INTERNAL)) {
if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
goto fail;
}
tdb->map_size += size;
if (tdb->flags & TDB_INTERNAL) {
char *new_map_ptr = (char *)realloc(tdb->map_ptr,
tdb->map_size);
if (!new_map_ptr) {
tdb->map_size -= size;
goto fail;
}
tdb->map_ptr = new_map_ptr;
} else {
/*
* We must ensure the file is remapped before adding the space
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* We're ok if the mmap fails as we'll fallback to read/write */
tdb_mmap(tdb);
}
/* form a new freelist record */
memset(&rec,'\0',sizeof(rec));
rec.rec_len = size - sizeof(rec);
/* link it into the free list */
offset = tdb->map_size - size;
if (tdb_free(tdb, offset, &rec) == -1)
goto fail;
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
/* allocate some space from the free list. The offset returned points
to a unconnected tdb_record within the database with room for at
least length bytes of total data
0 is returned if the space could not be allocated
*/
tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec)
{
tdb_off_t rec_ptr, last_ptr, newrec_ptr;
struct {
tdb_off_t rec_ptr, last_ptr;
tdb_len_t rec_len;
} bestfit;
float multiplier = 1.0;
if (tdb_lock(tdb, -1, F_WRLCK) == -1)
return 0;
/* over-allocate to reduce fragmentation */
length *= 1.25;
/* Extra bytes required for tailer */
length += sizeof(tdb_off_t);
length = TDB_ALIGN(length, TDB_ALIGNMENT);
again:
last_ptr = FREELIST_TOP;
/* read in the freelist top */
if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
goto fail;
bestfit.rec_ptr = 0;
bestfit.last_ptr = 0;
bestfit.rec_len = 0;
/*
this is a best fit allocation strategy. Originally we used
a first fit strategy, but it suffered from massive fragmentation
issues when faced with a slowly increasing record size.
*/
while (rec_ptr) {
if (tdb_rec_free_read(tdb, rec_ptr, rec) == -1) {
goto fail;
}
if (rec->rec_len >= length) {
if (bestfit.rec_ptr == 0 ||
rec->rec_len < bestfit.rec_len) {
bestfit.rec_len = rec->rec_len;
bestfit.rec_ptr = rec_ptr;
bestfit.last_ptr = last_ptr;
}
}
/* move to the next record */
last_ptr = rec_ptr;
rec_ptr = rec->next;
/* if we've found a record that is big enough, then
stop searching if its also not too big. The
definition of 'too big' changes as we scan
through */
if (bestfit.rec_len > 0 &&
bestfit.rec_len < length * multiplier) {
break;
}
/* this multiplier means we only extremely rarely
search more than 50 or so records. At 50 records we
accept records up to 11 times larger than what we
want */
multiplier *= 1.05;
}
if (bestfit.rec_ptr != 0) {
if (tdb_rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
goto fail;
}
newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr,
rec, bestfit.last_ptr);
tdb_unlock(tdb, -1, F_WRLCK);
return newrec_ptr;
}
/* we didn't find enough space. See if we can expand the
database and if we can then try again */
if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
goto again;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
}
/* expand the database at least size bytes by expanding the underlying
file and doing the mmap again if necessary */
int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
{
struct list_struct rec;
tdb_off_t offset;
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
return -1;
}
/* must know about any previous expansions by another process */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
/* always make room for at least 10 more records, and round
the database up to a multiple of the page size */
size = TDB_ALIGN(tdb->map_size + size*10, tdb->page_size) - tdb->map_size;
if (!(tdb->flags & TDB_INTERNAL))
tdb_munmap(tdb);
/*
* We must ensure the file is unmapped before doing this
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* expand the file itself */
if (!(tdb->flags & TDB_INTERNAL)) {
if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
goto fail;
}
tdb->map_size += size;
if (tdb->flags & TDB_INTERNAL) {
char *new_map_ptr = (char *)realloc(tdb->map_ptr,
tdb->map_size);
if (!new_map_ptr) {
tdb->map_size -= size;
goto fail;
}
tdb->map_ptr = new_map_ptr;
} else {
/*
* We must ensure the file is remapped before adding the space
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* We're ok if the mmap fails as we'll fallback to read/write */
tdb_mmap(tdb);
}
/* form a new freelist record */
memset(&rec,'\0',sizeof(rec));
rec.rec_len = size - sizeof(rec);
/* link it into the free list */
offset = tdb->map_size - size;
if (tdb_free(tdb, offset, &rec) == -1)
goto fail;
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}