enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
struct tdb_data *data)
{
tdb_off_t off;
struct tdb_used_record rec;
struct hash_info h;
enum TDB_ERROR ecode;
off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
if (TDB_OFF_IS_ERR(off)) {
return tdb->last_error = off;
}
if (!off) {
ecode = TDB_ERR_NOEXIST;
} else {
data->dsize = rec_data_length(&rec);
data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
data->dsize);
if (TDB_PTR_IS_ERR(data->dptr)) {
ecode = TDB_PTR_ERR(data->dptr);
} else
ecode = TDB_SUCCESS;
}
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
return tdb->last_error = ecode;
}
enum TDB_ERROR tdb_store(struct tdb_context *tdb,
struct tdb_data key, struct tdb_data dbuf, int flag)
{
struct hash_info h;
tdb_off_t off;
tdb_len_t old_room = 0;
struct tdb_used_record rec;
enum TDB_ERROR ecode;
off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
if (TDB_OFF_IS_ERR(off)) {
return tdb->last_error = off;
}
/* Now we have lock on this hash bucket. */
if (flag == TDB_INSERT) {
if (off) {
ecode = TDB_ERR_EXISTS;
goto out;
}
} else {
if (off) {
old_room = rec_data_length(&rec)
+ rec_extra_padding(&rec);
if (old_room >= dbuf.dsize) {
/* Can modify in-place. Easy! */
ecode = update_rec_hdr(tdb, off,
key.dsize, dbuf.dsize,
&rec, h.h);
if (ecode != TDB_SUCCESS) {
goto out;
}
ecode = update_data(tdb,
off + sizeof(rec)
+ key.dsize, dbuf,
old_room - dbuf.dsize);
if (ecode != TDB_SUCCESS) {
goto out;
}
tdb_unlock_hashes(tdb, h.hlock_start,
h.hlock_range, F_WRLCK);
return tdb->last_error = TDB_SUCCESS;
}
} else {
if (flag == TDB_MODIFY) {
/* if the record doesn't exist and we
are in TDB_MODIFY mode then we should fail
the store */
ecode = TDB_ERR_NOEXIST;
goto out;
}
}
}
/* If we didn't use the old record, this implies we're growing. */
ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
out:
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
return tdb->last_error = ecode;
}
/* We lock twice, not very efficient. We could keep last key & tinfo cached. */
enum TDB_ERROR tdb_nextkey(struct tdb_context *tdb, struct tdb_data *key)
{
struct traverse_info tinfo;
struct hash_info h;
struct tdb_used_record rec;
if (tdb->flags & TDB_VERSION1) {
struct tdb_data last_key = *key;
tdb->last_error = TDB_SUCCESS;
*key = tdb1_nextkey(tdb, last_key);
free(last_key.dptr);
/* TDB1 didn't set error for last key. */
if (!key->dptr && tdb->last_error == TDB_SUCCESS) {
tdb->last_error = TDB_ERR_NOEXIST;
}
return tdb->last_error;
}
tinfo.prev = find_and_lock(tdb, *key, F_RDLCK, &h, &rec, &tinfo);
free(key->dptr);
if (TDB_OFF_IS_ERR(tinfo.prev)) {
return tdb->last_error = TDB_OFF_TO_ERR(tinfo.prev);
}
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
return tdb->last_error = next_in_hash(tdb, &tinfo, key, NULL);
}
/* We slot in anywhere that's empty in the chain. */
static enum TDB_ERROR COLD add_to_chain(struct tdb_context *tdb,
tdb_off_t subhash,
tdb_off_t new_off)
{
tdb_off_t entry;
enum TDB_ERROR ecode;
entry = tdb_find_zero_off(tdb, subhash, 1<<TDB_HASH_GROUP_BITS);
if (TDB_OFF_IS_ERR(entry)) {
return entry;
}
if (entry == 1 << TDB_HASH_GROUP_BITS) {
tdb_off_t next;
next = tdb_read_off(tdb, subhash
+ offsetof(struct tdb_chain, next));
if (TDB_OFF_IS_ERR(next)) {
return next;
}
if (!next) {
next = alloc(tdb, 0, sizeof(struct tdb_chain), 0,
TDB_CHAIN_MAGIC, false);
if (TDB_OFF_IS_ERR(next))
return next;
ecode = zero_out(tdb,
next+sizeof(struct tdb_used_record),
sizeof(struct tdb_chain));
if (ecode != TDB_SUCCESS) {
return ecode;
}
ecode = tdb_write_off(tdb, subhash
+ offsetof(struct tdb_chain,
next),
next);
if (ecode != TDB_SUCCESS) {
return ecode;
}
}
static enum TDB_ERROR replace_data(struct tdb_context *tdb,
struct hash_info *h,
struct tdb_data key, struct tdb_data dbuf,
tdb_off_t old_off, tdb_len_t old_room,
bool growing)
{
tdb_off_t new_off;
enum TDB_ERROR ecode;
/* Allocate a new record. */
new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
growing);
if (TDB_OFF_IS_ERR(new_off)) {
return new_off;
}
/* We didn't like the existing one: remove it. */
if (old_off) {
tdb->stats.frees++;
ecode = add_free_record(tdb, old_off,
sizeof(struct tdb_used_record)
+ key.dsize + old_room,
TDB_LOCK_WAIT, true);
if (ecode == TDB_SUCCESS)
ecode = replace_in_hash(tdb, h, new_off);
} else {
ecode = add_to_hash(tdb, h, new_off);
}
if (ecode != TDB_SUCCESS) {
return ecode;
}
new_off += sizeof(struct tdb_used_record);
ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
if (ecode != TDB_SUCCESS) {
return ecode;
}
new_off += key.dsize;
ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
if (ecode != TDB_SUCCESS) {
return ecode;
}
if (tdb->flags & TDB_SEQNUM)
tdb_inc_seqnum(tdb);
return TDB_SUCCESS;
}
bool tdb_exists(struct tdb_context *tdb, TDB_DATA key)
{
tdb_off_t off;
struct tdb_used_record rec;
struct hash_info h;
off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
if (TDB_OFF_IS_ERR(off)) {
tdb->last_error = off;
return false;
}
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
tdb->last_error = TDB_SUCCESS;
return off ? true : false;
}
enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
{
tdb_off_t off;
struct tdb_used_record rec;
struct hash_info h;
enum TDB_ERROR ecode;
if (tdb->flags & TDB_VERSION1) {
if (tdb1_delete(tdb, key) == -1)
return tdb->last_error;
return TDB_SUCCESS;
}
off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
if (TDB_OFF_IS_ERR(off)) {
return tdb->last_error = TDB_OFF_TO_ERR(off);
}
if (!off) {
ecode = TDB_ERR_NOEXIST;
goto unlock;
}
ecode = delete_from_hash(tdb, &h);
if (ecode != TDB_SUCCESS) {
goto unlock;
}
/* Free the deleted entry. */
tdb->stats.frees++;
ecode = add_free_record(tdb, off,
sizeof(struct tdb_used_record)
+ rec_key_length(&rec)
+ rec_data_length(&rec)
+ rec_extra_padding(&rec),
TDB_LOCK_WAIT, true);
if (tdb->flags & TDB_SEQNUM)
tdb_inc_seqnum(tdb);
unlock:
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
return tdb->last_error = ecode;
}
enum TDB_ERROR tdb_parse_record_(struct tdb_context *tdb,
TDB_DATA key,
enum TDB_ERROR (*parse)(TDB_DATA k,
TDB_DATA d,
void *data),
void *data)
{
tdb_off_t off;
struct tdb_used_record rec;
struct hash_info h;
enum TDB_ERROR ecode;
if (tdb->flags & TDB_VERSION1) {
return tdb->last_error = tdb1_parse_record(tdb, key, parse,
data);
}
off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
if (TDB_OFF_IS_ERR(off)) {
return tdb->last_error = TDB_OFF_TO_ERR(off);
}
if (!off) {
ecode = TDB_ERR_NOEXIST;
} else {
const void *dptr;
dptr = tdb_access_read(tdb, off + sizeof(rec) + key.dsize,
rec_data_length(&rec), false);
if (TDB_PTR_IS_ERR(dptr)) {
ecode = TDB_PTR_ERR(dptr);
} else {
TDB_DATA d = tdb_mkdata(dptr, rec_data_length(&rec));
ecode = parse(key, d, data);
tdb_access_release(tdb, dptr);
}
}
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
return tdb->last_error = ecode;
}
int64_t tdb_get_seqnum(struct tdb_context *tdb)
{
tdb_off_t off;
if (tdb->flags & TDB_VERSION1) {
tdb1_off_t val;
tdb->last_error = TDB_SUCCESS;
val = tdb1_get_seqnum(tdb);
if (tdb->last_error != TDB_SUCCESS)
return TDB_ERR_TO_OFF(tdb->last_error);
else
return val;
}
off = tdb_read_off(tdb, offsetof(struct tdb_header, seqnum));
if (TDB_OFF_IS_ERR(off))
tdb->last_error = TDB_OFF_TO_ERR(off);
else
tdb->last_error = TDB_SUCCESS;
return off;
}
enum TDB_ERROR tdb_append(struct tdb_context *tdb,
struct tdb_data key, struct tdb_data dbuf)
{
struct hash_info h;
tdb_off_t off;
struct tdb_used_record rec;
tdb_len_t old_room = 0, old_dlen;
unsigned char *newdata;
struct tdb_data new_dbuf;
enum TDB_ERROR ecode;
off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
if (TDB_OFF_IS_ERR(off)) {
return tdb->last_error = off;
}
if (off) {
old_dlen = rec_data_length(&rec);
old_room = old_dlen + rec_extra_padding(&rec);
/* Fast path: can append in place. */
if (rec_extra_padding(&rec) >= dbuf.dsize) {
ecode = update_rec_hdr(tdb, off, key.dsize,
old_dlen + dbuf.dsize, &rec,
h.h);
if (ecode != TDB_SUCCESS) {
goto out;
}
off += sizeof(rec) + key.dsize + old_dlen;
ecode = update_data(tdb, off, dbuf,
rec_extra_padding(&rec));
goto out;
}
/* Slow path. */
newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
if (!newdata) {
ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
"tdb_append:"
" failed to allocate %zu bytes",
(size_t)(key.dsize + old_dlen
+ dbuf.dsize));
goto out;
}
ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
newdata, old_dlen);
if (ecode != TDB_SUCCESS) {
goto out_free_newdata;
}
memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
new_dbuf.dptr = newdata;
new_dbuf.dsize = old_dlen + dbuf.dsize;
} else {
newdata = NULL;
new_dbuf = dbuf;
}
/* If they're using tdb_append(), it implies they're growing record. */
ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
out_free_newdata:
free(newdata);
out:
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
return tdb->last_error = ecode;
}
static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
struct tdb_data key,
tdb_off_t chain,
struct hash_info *h,
struct tdb_used_record *rec,
struct traverse_info *tinfo)
{
tdb_off_t off, next;
enum TDB_ERROR ecode;
/* In case nothing is free, we set these to zero. */
h->home_bucket = h->found_bucket = 0;
for (off = chain; off; off = next) {
unsigned int i;
h->group_start = off;
ecode = tdb_read_convert(tdb, off, h->group, sizeof(h->group));
if (ecode != TDB_SUCCESS) {
return ecode;
}
for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
tdb_off_t recoff;
if (!h->group[i]) {
/* Remember this empty bucket. */
h->home_bucket = h->found_bucket = i;
continue;
}
/* We can insert extra bits via add_to_hash
* empty bucket logic. */
recoff = h->group[i] & TDB_OFF_MASK;
ecode = tdb_read_convert(tdb, recoff, rec,
sizeof(*rec));
if (ecode != TDB_SUCCESS) {
return ecode;
}
ecode = key_matches(tdb, rec, recoff, &key);
if (ecode < 0) {
return ecode;
}
if (ecode == 1) {
h->home_bucket = h->found_bucket = i;
if (tinfo) {
tinfo->levels[tinfo->num_levels]
.hashtable = off;
tinfo->levels[tinfo->num_levels]
.total_buckets
= 1 << TDB_HASH_GROUP_BITS;
tinfo->levels[tinfo->num_levels].entry
= i;
tinfo->num_levels++;
}
return recoff;
}
}
next = tdb_read_off(tdb, off
+ offsetof(struct tdb_chain, next));
if (TDB_OFF_IS_ERR(next)) {
return next;
}
if (next)
next += sizeof(struct tdb_used_record);
}
return 0;
}
