_PUBLIC_ enum NTDB_ERROR ntdb_fetch(struct ntdb_context *ntdb, NTDB_DATA key,
NTDB_DATA *data)
{
ntdb_off_t off;
struct ntdb_used_record rec;
struct hash_info h;
enum NTDB_ERROR ecode;
const char *keyp;
off = find_and_lock(ntdb, key, F_RDLCK, &h, &rec, &keyp);
if (NTDB_OFF_IS_ERR(off)) {
return NTDB_OFF_TO_ERR(off);
}
if (!off) {
ecode = NTDB_ERR_NOEXIST;
} else {
data->dsize = rec_data_length(&rec);
data->dptr = ntdb->alloc_fn(ntdb, data->dsize, ntdb->alloc_data);
if (unlikely(!data->dptr)) {
ecode = NTDB_ERR_OOM;
} else {
memcpy(data->dptr, keyp + key.dsize, data->dsize);
ecode = NTDB_SUCCESS;
}
ntdb_access_release(ntdb, keyp);
}
ntdb_unlock_hash(ntdb, h.h, F_RDLCK);
return 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;
}
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;
}
int main(int argc, char *argv[])
{
unsigned int i;
struct ntdb_used_record rec;
struct ntdb_context ntdb = { .log_fn = tap_log_fn };
plan_tests(64 + 32 + 48*5 + 1);
/* We should be able to encode any data value. */
for (i = 0; i < 64; i++)
ok1(set_header(&ntdb, &rec, NTDB_USED_MAGIC, 0, 1ULL << i,
1ULL << i) == 0);
/* And any key and data with < 64 bits between them. */
for (i = 0; i < 32; i++) {
ntdb_len_t dlen = 1ULL >> (63 - i), klen = 1ULL << i;
ok1(set_header(&ntdb, &rec, NTDB_USED_MAGIC, klen, dlen,
klen + dlen) == 0);
}
/* We should neatly encode all values. */
for (i = 0; i < 48; i++) {
uint64_t klen = 1ULL << (i < 16 ? i : 15);
uint64_t dlen = 1ULL << i;
uint64_t xlen = 1ULL << (i < 32 ? i : 31);
ok1(set_header(&ntdb, &rec, NTDB_USED_MAGIC, klen, dlen,
klen+dlen+xlen) == 0);
ok1(rec_key_length(&rec) == klen);
ok1(rec_data_length(&rec) == dlen);
ok1(rec_extra_padding(&rec) == xlen);
ok1(rec_magic(&rec) == NTDB_USED_MAGIC);
}
ok1(tap_log_messages == 0);
return exit_status();
}
_PUBLIC_ enum NTDB_ERROR ntdb_store(struct ntdb_context *ntdb,
NTDB_DATA key, NTDB_DATA dbuf, int flag)
{
struct hash_info h;
ntdb_off_t off;
ntdb_len_t old_room = 0;
struct ntdb_used_record rec;
enum NTDB_ERROR ecode;
off = find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL);
if (NTDB_OFF_IS_ERR(off)) {
return NTDB_OFF_TO_ERR(off);
}
/* Now we have lock on this hash bucket. */
if (flag == NTDB_INSERT) {
if (off) {
ecode = NTDB_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(ntdb, off,
key.dsize, dbuf.dsize,
&rec);
if (ecode != NTDB_SUCCESS) {
goto out;
}
ecode = update_data(ntdb,
off + sizeof(rec)
+ key.dsize, dbuf,
old_room - dbuf.dsize);
if (ecode != NTDB_SUCCESS) {
goto out;
}
ntdb_unlock_hash(ntdb, h.h, F_WRLCK);
return NTDB_SUCCESS;
}
} else {
if (flag == NTDB_MODIFY) {
/* if the record doesn't exist and we
are in NTDB_MODIFY mode then we should fail
the store */
ecode = NTDB_ERR_NOEXIST;
goto out;
}
}
}
/* If we didn't use the old record, this implies we're growing. */
ecode = replace_data(ntdb, &h, key, dbuf, off, old_room, off);
out:
ntdb_unlock_hash(ntdb, h.h, F_WRLCK);
return 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;
}
static enum NTDB_ERROR update_rec_hdr(struct ntdb_context *ntdb,
ntdb_off_t off,
ntdb_len_t keylen,
ntdb_len_t datalen,
struct ntdb_used_record *rec)
{
uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
enum NTDB_ERROR ecode;
ecode = set_header(ntdb, rec, NTDB_USED_MAGIC, keylen, datalen,
keylen + dataroom);
if (ecode == NTDB_SUCCESS) {
ecode = ntdb_write_convert(ntdb, off, rec, sizeof(*rec));
}
return ecode;
}
_PUBLIC_ enum NTDB_ERROR ntdb_parse_record_(struct ntdb_context *ntdb,
NTDB_DATA key,
enum NTDB_ERROR (*parse)(NTDB_DATA k,
NTDB_DATA d,
void *data),
void *data)
{
ntdb_off_t off;
struct ntdb_used_record rec;
struct hash_info h;
enum NTDB_ERROR ecode;
const char *keyp;
off = find_and_lock(ntdb, key, F_RDLCK, &h, &rec, &keyp);
if (NTDB_OFF_IS_ERR(off)) {
return NTDB_OFF_TO_ERR(off);
}
if (!off) {
ecode = NTDB_ERR_NOEXIST;
} else {
unsigned int old_flags;
NTDB_DATA d = ntdb_mkdata(keyp + key.dsize,
rec_data_length(&rec));
/*
* Make sure they don't try to write db, since they
* have read lock! They can if they've done
* ntdb_lockall(): if it was ntdb_lockall_read, that'll
* stop them doing a write operation anyway.
*/
old_flags = ntdb->flags;
if (!ntdb->file->allrecord_lock.count &&
!(ntdb->flags & NTDB_NOLOCK)) {
ntdb->flags |= NTDB_RDONLY;
}
ecode = parse(key, d, data);
ntdb->flags = old_flags;
ntdb_access_release(ntdb, keyp);
}
ntdb_unlock_hash(ntdb, h.h, F_RDLCK);
return ecode;
}
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;
}
_PUBLIC_ enum NTDB_ERROR ntdb_delete(struct ntdb_context *ntdb, NTDB_DATA key)
{
ntdb_off_t off;
struct ntdb_used_record rec;
struct hash_info h;
enum NTDB_ERROR ecode;
off = find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL);
if (NTDB_OFF_IS_ERR(off)) {
return NTDB_OFF_TO_ERR(off);
}
if (!off) {
ecode = NTDB_ERR_NOEXIST;
goto unlock;
}
ecode = delete_from_hash(ntdb, &h);
if (ecode != NTDB_SUCCESS) {
goto unlock;
}
/* Free the deleted entry. */
ntdb->stats.frees++;
ecode = add_free_record(ntdb, off,
sizeof(struct ntdb_used_record)
+ rec_key_length(&rec)
+ rec_data_length(&rec)
+ rec_extra_padding(&rec),
NTDB_LOCK_WAIT, true);
if (ntdb->flags & NTDB_SEQNUM)
ntdb_inc_seqnum(ntdb);
unlock:
ntdb_unlock_hash(ntdb, h.h, F_WRLCK);
return ecode;
}
_PUBLIC_ enum NTDB_ERROR ntdb_append(struct ntdb_context *ntdb,
NTDB_DATA key, NTDB_DATA dbuf)
{
struct hash_info h;
ntdb_off_t off;
struct ntdb_used_record rec;
ntdb_len_t old_room = 0, old_dlen;
unsigned char *newdata;
NTDB_DATA new_dbuf;
enum NTDB_ERROR ecode;
off = find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL);
if (NTDB_OFF_IS_ERR(off)) {
return NTDB_OFF_TO_ERR(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(ntdb, off, key.dsize,
old_dlen + dbuf.dsize, &rec);
if (ecode != NTDB_SUCCESS) {
goto out;
}
off += sizeof(rec) + key.dsize + old_dlen;
ecode = update_data(ntdb, off, dbuf,
rec_extra_padding(&rec));
goto out;
}
/* Slow path. */
newdata = ntdb->alloc_fn(ntdb, key.dsize + old_dlen + dbuf.dsize,
ntdb->alloc_data);
if (!newdata) {
ecode = ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
"ntdb_append:"
" failed to allocate %zu bytes",
(size_t)(key.dsize + old_dlen
+ dbuf.dsize));
goto out;
}
ecode = ntdb->io->tread(ntdb, off + sizeof(rec) + key.dsize,
newdata, old_dlen);
if (ecode != NTDB_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 ntdb_append(), it implies they're growing record. */
ecode = replace_data(ntdb, &h, key, new_dbuf, off, old_room, true);
out_free_newdata:
ntdb->free_fn(newdata, ntdb->alloc_data);
out:
ntdb_unlock_hash(ntdb, h.h, F_WRLCK);
return ecode;
}
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;
}