Commit 850c5cfe authored by Rusty Russell's avatar Rusty Russell

tdb2: rework lock.c functions to return enum TDB_ERROR.

Make the other functions in lock.s return an error code, too.
parent 50b15fbd
......@@ -673,11 +673,17 @@ int tdb_check(struct tdb_context *tdb,
{
tdb_off_t *fr = NULL, *used = NULL, ft, recovery;
size_t num_free = 0, num_used = 0, num_found = 0, num_ftables = 0;
enum TDB_ERROR ecode;
if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false) != 0)
ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
if (ecpde != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
if (tdb_lock_expand(tdb, F_RDLCK) != 0) {
ecode = tdb_lock_expand(tdb, F_RDLCK);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
tdb_allrecord_unlock(tdb, F_RDLCK);
return -1;
}
......
......@@ -204,14 +204,18 @@ int add_free_record(struct tdb_context *tdb,
tdb_off_t b_off;
tdb_len_t len;
int ret;
enum TDB_ERROR ecode;
assert(len_with_header >= sizeof(struct tdb_free_record));
len = len_with_header - sizeof(struct tdb_used_record);
b_off = bucket_off(tdb->ftable_off, size_to_bucket(len));
if (tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) != 0)
ecode = tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
ret = enqueue_in_free(tdb, b_off, off, len);
tdb_unlock_free_bucket(tdb, b_off);
......@@ -290,7 +294,8 @@ static int coalesce(struct tdb_context *tdb,
tdb_access_release(tdb, r);
/* We may be violating lock order here, so best effort. */
if (tdb_lock_free_bucket(tdb, nb_off, TDB_LOCK_NOWAIT) == -1) {
if (tdb_lock_free_bucket(tdb, nb_off, TDB_LOCK_NOWAIT)
!= TDB_SUCCESS) {
add_stat(tdb, alloc_coalesce_lockfail, 1);
break;
}
......@@ -377,6 +382,7 @@ static tdb_off_t lock_and_alloc(struct tdb_context *tdb,
struct tdb_free_record best = { 0 };
double multiplier;
size_t size = adjust_size(keylen, datalen);
enum TDB_ERROR ecode;
add_stat(tdb, allocs, 1);
again:
......@@ -384,7 +390,9 @@ again:
/* FIXME: Try non-blocking wait first, to measure contention. */
/* Lock this bucket. */
if (tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) == -1) {
ecode = tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return TDB_OFF_ERR;
}
......@@ -581,6 +589,7 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size)
{
uint64_t old_size;
tdb_len_t wanted;
enum TDB_ERROR ecode;
/* We need room for the record header too. */
wanted = sizeof(struct tdb_used_record) + size;
......@@ -602,8 +611,11 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size)
wanted = adjust_size(0, wanted);
/* Only one person can expand file at a time. */
if (tdb_lock_expand(tdb, F_WRLCK) != 0)
ecode = tdb_lock_expand(tdb, F_WRLCK);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
/* Someone else may have expanded the file, so retry. */
old_size = tdb->map_size;
......
......@@ -229,6 +229,7 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
{
uint32_t i, group;
tdb_off_t hashtable;
enum TDB_ERROR ecode;
h->h = tdb_hash(tdb, key.dptr, key.dsize);
h->hash_used = 0;
......@@ -236,9 +237,12 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
h->home_bucket = use_bits(h, TDB_HASH_GROUP_BITS);
h->hlock_start = hlock_range(group, &h->hlock_range);
if (tdb_lock_hashes(tdb, h->hlock_start, h->hlock_range, ltype,
TDB_LOCK_WAIT))
ecode = tdb_lock_hashes(tdb, h->hlock_start, h->hlock_range, ltype,
TDB_LOCK_WAIT);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return TDB_OFF_ERR;
}
hashtable = offsetof(struct tdb_header, hashtable);
if (tinfo) {
......@@ -684,14 +688,18 @@ int next_in_hash(struct tdb_context *tdb,
{
const unsigned group_bits = TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS;
tdb_off_t hl_start, hl_range, off;
enum TDB_ERROR ecode;
while (tinfo->toplevel_group < (1 << group_bits)) {
hl_start = (tdb_off_t)tinfo->toplevel_group
<< (64 - group_bits);
hl_range = 1ULL << group_bits;
if (tdb_lock_hashes(tdb, hl_start, hl_range, F_RDLCK,
TDB_LOCK_WAIT) != 0)
ecode = tdb_lock_hashes(tdb, hl_start, hl_range, F_RDLCK,
TDB_LOCK_WAIT);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
off = iterate_hash(tdb, tinfo);
if (off) {
......@@ -760,7 +768,7 @@ static int chainlock(struct tdb_context *tdb, const TDB_DATA *key,
int ltype, enum tdb_lock_flags waitflag,
const char *func)
{
int ret;
enum TDB_ERROR ecode;
uint64_t h = tdb_hash(tdb, key->dptr, key->dsize);
tdb_off_t lockstart, locksize;
unsigned int group, gbits;
......@@ -770,9 +778,13 @@ static int chainlock(struct tdb_context *tdb, const TDB_DATA *key,
lockstart = hlock_range(group, &locksize);
ret = tdb_lock_hashes(tdb, lockstart, locksize, ltype, waitflag);
ecode = tdb_lock_hashes(tdb, lockstart, locksize, ltype, waitflag);
tdb_trace_1rec(tdb, func, *key);
return ret;
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
return 0;
}
/* lock/unlock one hash chain. This is meant to be used to reduce
......@@ -787,6 +799,7 @@ int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
tdb_off_t lockstart, locksize;
unsigned int group, gbits;
enum TDB_ERROR ecode;
gbits = TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS;
group = bits_from(h, 64 - gbits, gbits);
......@@ -794,5 +807,10 @@ int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
lockstart = hlock_range(group, &locksize);
tdb_trace_1rec(tdb, "tdb_chainunlock", key);
return tdb_unlock_hashes(tdb, lockstart, locksize, F_WRLCK);
ecode = tdb_unlock_hashes(tdb, lockstart, locksize, F_WRLCK);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
return 0;
}
......@@ -70,6 +70,7 @@ void tdb_mmap(struct tdb_context *tdb)
static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
{
struct stat st;
enum TDB_ERROR ecode;
/* We can't hold pointers during this: we could unmap! */
assert(!tdb->direct_access
......@@ -89,8 +90,11 @@ static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
return -1;
}
if (tdb_lock_expand(tdb, F_RDLCK) != 0)
ecode = tdb_lock_expand(tdb, F_RDLCK);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
if (fstat(tdb->fd, &st) != 0) {
tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
......
......@@ -198,21 +198,21 @@ static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
deadlock detection and claim a deadlock when progress can be
made. For those OSes we may loop for a while.
*/
int tdb_allrecord_upgrade(struct tdb_context *tdb)
enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
{
int count = 1000;
if (tdb->allrecord_lock.count != 1) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_upgrade failed: count %u too high",
tdb->allrecord_lock.count);
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_upgrade failed:"
" count %u too high",
tdb->allrecord_lock.count);
}
if (tdb->allrecord_lock.off != 1) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_upgrade failed: already upgraded?");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_upgrade failed:"
" already upgraded?");
}
while (count--) {
......@@ -222,7 +222,7 @@ int tdb_allrecord_upgrade(struct tdb_context *tdb)
TDB_LOCK_WAIT|TDB_LOCK_PROBE) == TDB_SUCCESS) {
tdb->allrecord_lock.ltype = F_WRLCK;
tdb->allrecord_lock.off = 0;
return 0;
return TDB_SUCCESS;
}
if (errno != EDEADLK) {
break;
......@@ -232,9 +232,8 @@ int tdb_allrecord_upgrade(struct tdb_context *tdb)
tv.tv_usec = 1;
select(0, NULL, NULL, NULL, &tv);
}
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_upgrade failed");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_upgrade failed");
}
static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
......@@ -250,85 +249,86 @@ static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
return NULL;
}
int tdb_lock_and_recover(struct tdb_context *tdb)
enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb)
{
int ret;
enum TDB_ERROR ecode;
if (tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
false) == -1) {
return -1;
ecode = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
false);
if (ecode != TDB_SUCCESS) {
return ecode;
}
if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
if (ecode != TDB_SUCCESS) {
tdb_allrecord_unlock(tdb, F_WRLCK);
return -1;
return ecode;
}
if (tdb_transaction_recover(tdb) == -1) {
ecode = tdb->ecode;
}
ret = tdb_transaction_recover(tdb);
tdb_unlock_open(tdb);
tdb_allrecord_unlock(tdb, F_WRLCK);
return ret;
return ecode;
}
/* lock an offset in the database. */
static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
enum tdb_lock_flags flags)
static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
tdb_off_t offset, int ltype,
enum tdb_lock_flags flags)
{
struct tdb_lock_type *new_lck;
enum TDB_ERROR ecode;
if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_lock: invalid offset %zu ltype=%d",
(size_t)offset, ltype);
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_lock: invalid offset %zu ltype=%d",
(size_t)offset, ltype);
}
if (tdb->flags & TDB_NOLOCK)
return 0;
return TDB_SUCCESS;
add_stat(tdb, locks, 1);
new_lck = find_nestlock(tdb, offset);
if (new_lck) {
if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_lock: offset %zu has read lock",
(size_t)offset);
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_lock:"
" offset %zu has read lock",
(size_t)offset);
}
/* Just increment the struct, posix locks don't stack. */
new_lck->count++;
return 0;
return TDB_SUCCESS;
}
if (tdb->num_lockrecs
&& offset >= TDB_HASH_LOCK_START
&& offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_lock: already have a hash lock?");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_lock: already have a hash lock?");
}
new_lck = (struct tdb_lock_type *)realloc(
tdb->lockrecs,
sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
if (new_lck == NULL) {
tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
"tdb_nest_lock: unable to allocate %zu lock struct",
tdb->num_lockrecs + 1);
errno = ENOMEM;
return -1;
return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
"tdb_nest_lock:"
" unable to allocate %zu lock struct",
tdb->num_lockrecs + 1);
}
tdb->lockrecs = new_lck;
/* Since fcntl locks don't nest, we do a lock for the first one,
and simply bump the count for future ones */
ecode = tdb_brlock(tdb, ltype, offset, 1, flags);
if (ecode) {
tdb->ecode = ecode;
return -1;
if (ecode != TDB_SUCCESS) {
return ecode;
}
/* First time we grab a lock, perhaps someone died in commit? */
......@@ -337,14 +337,12 @@ static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
&& unlikely(tdb_needs_recovery(tdb))) {
tdb_brunlock(tdb, ltype, offset, 1);
if (tdb_lock_and_recover(tdb) == -1) {
return -1;
ecode = tdb_lock_and_recover(tdb);
if (ecode == TDB_SUCCESS) {
ecode = tdb_brlock(tdb, ltype, offset, 1, flags);
}
ecode = tdb_brlock(tdb, ltype, offset, 1, flags);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
return ecode;
}
}
......@@ -353,27 +351,28 @@ static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
tdb->num_lockrecs++;
return 0;
return TDB_SUCCESS;
}
static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
static enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
tdb_off_t off, int ltype)
{
struct tdb_lock_type *lck;
enum TDB_ERROR ecode;
if (tdb->flags & TDB_NOLOCK)
return 0;
return TDB_SUCCESS;
lck = find_nestlock(tdb, off);
if ((lck == NULL) || (lck->count == 0)) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_unlock: no lock for %zu", (size_t)off);
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_nest_unlock: no lock for %zu",
(size_t)off);
}
if (lck->count > 1) {
lck->count--;
return 0;
return TDB_SUCCESS;
}
/*
......@@ -390,17 +389,13 @@ static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
*/
*lck = tdb->lockrecs[--tdb->num_lockrecs];
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return -1;
}
return 0;
return ecode;
}
/*
get the transaction lock
*/
int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype)
{
return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
}
......@@ -408,9 +403,9 @@ int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
/*
release the transaction lock
*/
int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
void tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
{
return tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
}
/* We only need to lock individual bytes, but Linux merges consecutive locks
......@@ -449,45 +444,43 @@ static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
/* lock/unlock entire database. It can only be upgradable if you have some
* other way of guaranteeing exclusivity (ie. transaction write lock). */
int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable)
enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable)
{
enum TDB_ERROR ecode;
/* FIXME: There are no locks on read-only dbs */
if (tdb->read_only) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_lock: read-only");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_lock: read-only");
}
if (tdb->allrecord_lock.count
&& (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
tdb->allrecord_lock.count++;
return 0;
return TDB_SUCCESS;
}
if (tdb->allrecord_lock.count) {
/* a global lock of a different type exists */
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_lock: already have %s lock",
tdb->allrecord_lock.ltype == F_RDLCK
? "read" : "write");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_lock: already have %s lock",
tdb->allrecord_lock.ltype == F_RDLCK
? "read" : "write");
}
if (tdb_has_hash_locks(tdb)) {
/* can't combine global and chain locks */
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_lock: already have chain lock");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_lock:"
" already have chain lock");
}
if (upgradable && ltype != F_RDLCK) {
/* tdb error: you can't upgrade a write lock! */
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_lock: can't upgrade a write lock");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_lock:"
" can't upgrade a write lock");
}
add_stat(tdb, locks, 1);
......@@ -496,12 +489,11 @@ again:
ecode = tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
TDB_HASH_LOCK_RANGE);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
if (!(flags & TDB_LOCK_PROBE)) {
tdb_logerr(tdb, tdb->ecode, TDB_LOG_ERROR,
tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
"tdb_allrecord_lock hashes failed");
}
return -1;
return ecode;
}
/* Lock free tables: there to end of file. */
......@@ -509,14 +501,13 @@ again:
TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
0, flags);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
if (!(flags & TDB_LOCK_PROBE)) {
tdb_logerr(tdb, tdb->ecode, TDB_LOG_ERROR,
tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
"tdb_allrecord_lock freetables failed");
}
tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
TDB_HASH_LOCK_RANGE);
return -1;
return ecode;
}
tdb->allrecord_lock.count = 1;
......@@ -528,16 +519,17 @@ again:
/* Now check for needing recovery. */
if (!(flags & TDB_LOCK_NOCHECK) && unlikely(tdb_needs_recovery(tdb))) {
tdb_allrecord_unlock(tdb, ltype);
if (tdb_lock_and_recover(tdb) == -1) {
return -1;
ecode = tdb_lock_and_recover(tdb);
if (ecode != TDB_SUCCESS) {
return ecode;
}
goto again;
}
return 0;
return TDB_SUCCESS;
}
int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
{
return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
}
......@@ -552,7 +544,7 @@ bool tdb_has_open_lock(struct tdb_context *tdb)
return find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
}
int tdb_lock_expand(struct tdb_context *tdb, int ltype)
enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype)
{
/* Lock doesn't protect data, so don't check (we recurse if we do!) */
return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype,
......@@ -565,33 +557,33 @@ void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
}
/* unlock entire db */
int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
{
if (tdb->allrecord_lock.count == 0) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_allrecord_unlock: not locked!");
return -1;
return;
}
/* Upgradable locks are marked as write locks. */
if (tdb->allrecord_lock.ltype != ltype
&& (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_allrecord_unlock: have %s lock",
"tdb_allrecord_unlock: have %s lock",
tdb->allrecord_lock.ltype == F_RDLCK
? "read" : "write");
return -1;
return;
}
if (tdb->allrecord_lock.count > 1) {
tdb->allrecord_lock.count--;
return 0;
return;
}
tdb->allrecord_lock.count = 0;
tdb->allrecord_lock.ltype = 0;
return tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
}
bool tdb_has_expansion_lock(struct tdb_context *tdb)
......@@ -624,10 +616,10 @@ static bool tdb_has_free_lock(struct tdb_context *tdb)
return false;
}
int tdb_lock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock,
tdb_len_t hash_range,
int ltype, enum tdb_lock_flags waitflag)
enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock,
tdb_len_t hash_range,
int ltype, enum tdb_lock_flags waitflag)
{
/* FIXME: Do this properly, using hlock_range */
unsigned lock = TDB_HASH_LOCK_START
......@@ -636,35 +628,34 @@ int tdb_lock_hashes(struct tdb_context *tdb,
/* a allrecord lock allows us to avoid per chain locks */
if (tdb->allrecord_lock.count &&
(ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
return 0;
return TDB_SUCCESS;
}
if (tdb->allrecord_lock.count) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_lock_hashes: already have %s allrecordlock",
tdb->allrecord_lock.ltype == F_RDLCK
? "read" : "write");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
"tdb_lock_hashes:"
" already have %s allrecordlock",
tdb->allrecord_lock.ltype == F_RDLCK
? "read" : "write");
}
if (tdb_has_free_lock(tdb)) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_hashes: already have free lock");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_hashes: already have free lock");
}
if (tdb_has_expansion_lock(tdb)) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_hashes: already have expansion lock");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_hashes:"
" already have expansion lock");
}
return tdb_nest_lock(tdb, lock, ltype, waitflag);
}
int tdb_unlock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock,
tdb_len_t hash_range, int ltype)
enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock,
tdb_len_t hash_range, int ltype)
{
unsigned lock = TDB_HASH_LOCK_START
+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
......@@ -673,11 +664,10 @@ int tdb_unlock_hashes(struct tdb_context *tdb,
if (tdb->allrecord_lock.count) {
if (tdb->allrecord_lock.ltype == F_RDLCK
&& ltype == F_WRLCK) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_unlock_hashes RO allrecord!");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_unlock_hashes RO allrecord!");
}
return 0;
return TDB_SUCCESS;
}
return tdb_nest_unlock(tdb, lock, ltype);
......@@ -694,8 +684,8 @@ static tdb_off_t free_lock_off(tdb_off_t b_off)
+ b_off / sizeof(tdb_off_t);
}
int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
enum tdb_lock_flags waitflag)
enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
enum tdb_lock_flags waitflag)
{
assert(b_off >= sizeof(struct tdb_header));
......@@ -703,16 +693,16 @@ int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
if (tdb->allrecord_lock.count) {
if (tdb->allrecord_lock.ltype == F_WRLCK)
return 0;
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_free_bucket with RO allrecordlock!");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_free_bucket with"
" read-only allrecordlock!");
}
#if 0 /* FIXME */
if (tdb_has_expansion_lock(tdb)) {
tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_free_bucket: already have expansion lock");
return -1;
return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
"tdb_lock_free_bucket:"
" already have expansion lock");
}
#endif
......
......@@ -518,43 +518,44 @@ void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
void tdb_lock_init(struct tdb_context *tdb);
/* Lock/unlock a range of hashes. */
int tdb_lock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock, tdb_len_t hash_range,
int ltype, enum tdb_lock_flags waitflag);
int tdb_unlock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock,
tdb_len_t hash_range, int ltype);
enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock, tdb_len_t hash_range,
int ltype, enum tdb_lock_flags waitflag);
enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
tdb_off_t hash_lock,
tdb_len_t hash_range, int ltype);
/* Lock/unlock a particular free bucket. */
int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
enum tdb_lock_flags waitflag);
enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
enum tdb_lock_flags waitflag);
void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
/* Serialize transaction start. */
int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
/* Do we have any hash locks (ie. via tdb_chainlock) ? */
bool tdb_has_hash_locks(struct tdb_context *tdb);
/* Lock entire database. */
int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable);
int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
int tdb_allrecord_upgrade(struct tdb_context *tdb);
enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable);
void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb);
/* Serialize db open. */
int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags);
enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
enum tdb_lock_flags flags);
void tdb_unlock_open(struct tdb_context *tdb);
bool tdb_has_open_lock(struct tdb_context *tdb);
/* Serialize db expand. */
int tdb_lock_expand(struct tdb_context *tdb, int ltype);
enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
bool tdb_has_expansion_lock(struct tdb_context *tdb);
/* If it needs recovery, grab all the locks and do it. */
int tdb_lock_and_recover(struct tdb_context *tdb);
enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
/* transaction.c: */
int tdb_transaction_recover(struct tdb_context *tdb);
......
......@@ -152,13 +152,19 @@ char *tdb_summary(struct tdb_context *tdb, enum tdb_summary_flags flags)
*buckets, *chains;
char *hashesg, *freeg, *keysg, *datag, *extrag, *uncoalg, *bucketsg;
char *ret = NULL;
enum TDB_ERROR ecode;
hashesg = freeg = keysg = datag = extrag = uncoalg = bucketsg = NULL;
if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false) != 0)
ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
return NULL;
}
if (tdb_lock_expand(tdb, F_RDLCK) != 0) {
ecode = tdb_lock_expand(tdb, F_RDLCK);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
tdb_allrecord_unlock(tdb, F_RDLCK);
return NULL;
}
......
......@@ -173,6 +173,7 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
ssize_t rlen;
struct tdb_header hdr;
struct tdb_attribute_seed *seed = NULL;
enum TDB_ERROR ecode;
tdb = malloc(sizeof(*tdb));
if (!tdb) {
......@@ -265,9 +266,8 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
/* ensure there is only one process initialising at once */
if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
/* errno set by tdb_brlock */
saved_errno = errno;
tdb->ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
if (tdb->ecode != TDB_SUCCESS) {
goto fail;
}
......@@ -346,8 +346,12 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
tdb->methods->oob(tdb, tdb->map_size + 1, true);
/* Now it's fully formed, recover if necessary. */
if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
goto fail;
if (tdb_needs_recovery(tdb)) {
ecode = tdb_lock_and_recover(tdb);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
goto fail;
}
}
if (tdb_ftable_init(tdb) == -1)
......
......@@ -11,7 +11,7 @@
#include "logging.h"
/* FIXME: Check these! */
#define INITIAL_TDB_MALLOC "tdb.c", 177, FAILTEST_MALLOC
#define INITIAL_TDB_MALLOC "tdb.c", 178, FAILTEST_MALLOC
#define LOGGING_MALLOC "tdb.c", 734, FAILTEST_MALLOC
#define URANDOM_OPEN "tdb.c", 49, FAILTEST_OPEN
#define URANDOM_READ "tdb.c", 29, FAILTEST_READ
......
......@@ -499,6 +499,8 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
*/
int tdb_transaction_start(struct tdb_context *tdb)
{
enum TDB_ERROR ecode;
/* some sanity checks */
if (tdb->read_only || (tdb->flags & TDB_INTERNAL)) {
tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
......@@ -536,7 +538,9 @@ int tdb_transaction_start(struct tdb_context *tdb)
/* get the transaction write lock. This is a blocking lock. As
discussed with Volker, there are a number of ways we could
make this async, which we will probably do in the future */
if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
ecode = tdb_transaction_lock(tdb, F_WRLCK);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
SAFE_FREE(tdb->transaction->blocks);
SAFE_FREE(tdb->transaction);
return -1;
......@@ -544,7 +548,9 @@ int tdb_transaction_start(struct tdb_context *tdb)
/* get a read lock over entire file. This is upgraded to a write
lock during the commit */
if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true) == -1) {
ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true);
if (ecode != TDB_SUCCESS) {
tdb->ecode = ecode;
goto fail_allrecord_lock;
}
......@@ -853,6 +859,7 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
{
const struct tdb_methods *methods;
enum TDB_ERROR ecode;
if (tdb->transaction == NULL) {
tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
......@@ -890,8 +897,9 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
methods = tdb->transaction->io_methods;
/* upgrade the main transaction lock region to a write lock */
if (tdb_allrecord_upgrade(tdb) == -1) {
tdb_logerr(tdb, tdb->ecode, TDB_LOG_ERROR,
ecode = tdb_allrecord_upgrade(tdb);
if (ecode != TDB_SUCCESS) {
tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
"tdb_transaction_prepare_commit:"
" failed to upgrade hash locks");
_tdb_transaction_cancel(tdb);
......@@ -900,10 +908,11 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
/* get the open lock - this prevents new users attaching to the database
during the commit */
if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
tdb_logerr(tdb, tdb->ecode, TDB_LOG_ERROR,
"tdb_transaction_prepare_commit:"
" failed to get open lock");
ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
if (ecode != TDB_SUCCESS) {
tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
"tdb_transaction_prepare_commit:"
" failed to get open lock");
_tdb_transaction_cancel(tdb);
return -1;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment