X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Flock.c;h=fe6072439f1c00c20ddf8005698c7608f7de2390;hp=14f9c0ae307b348b1a7401b6001bbd419a8e0c69;hb=6520c8318b9b8f2d92ae4b9f062788b7e4035aac;hpb=a0ac303da08672d4cfe5629b794b2651e1e98399

diff --git a/ccan/tdb2/lock.c b/ccan/tdb2/lock.c
index 14f9c0ae..fe607243 100644
--- a/ccan/tdb2/lock.c
+++ b/ccan/tdb2/lock.c
@@ -190,7 +190,6 @@ static int tdb_brunlock(struct tdb_context *tdb,
 	return ret;
 }
 
-#if 0
 /*
   upgrade a read lock to a write lock. This needs to be handled in a
   special way as some OSes (such as solaris) have too conservative
@@ -217,8 +216,7 @@ int tdb_allrecord_upgrade(struct tdb_context *tdb)
 	while (count--) {
 		struct timeval tv;
 		if (tdb_brlock(tdb, F_WRLCK,
-			       TDB_HASH_LOCK_START
-			       + (1ULL << tdb->header.v.hash_bits), 0,
+			       TDB_HASH_LOCK_START, 0,
 			       TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
 			tdb->allrecord_lock.ltype = F_WRLCK;
 			tdb->allrecord_lock.off = 0;
@@ -236,7 +234,6 @@ int tdb_allrecord_upgrade(struct tdb_context *tdb)
 		 "tdb_allrecord_upgrade failed\n");
 	return -1;
 }
-#endif
 
 static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
 					   tdb_off_t offset)
@@ -251,13 +248,34 @@ static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
 	return NULL;
 }
 
+int tdb_lock_and_recover(struct tdb_context *tdb)
+{
+	int ret;
+
+	if (tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
+			       false) == -1) {
+		return -1;
+	}
+
+	if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
+		tdb_allrecord_unlock(tdb, F_WRLCK);
+		return -1;
+	}
+	ret = tdb_transaction_recover(tdb);
+
+	tdb_unlock_open(tdb);
+	tdb_allrecord_unlock(tdb, F_WRLCK);
+
+	return ret;
+}
+
 /* 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)
 {
 	struct tdb_lock_type *new_lck;
 
-	if (offset >= TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
+	if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
 		tdb->ecode = TDB_ERR_LOCK;
 		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 			 "tdb_nest_lock: invalid offset %llu ltype=%d\n",
@@ -270,14 +288,27 @@ static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
 
 	new_lck = find_nestlock(tdb, offset);
 	if (new_lck) {
-		/*
-		 * Just increment the in-memory struct, posix locks
-		 * don't stack.
-		 */
+		if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
+			tdb->ecode = TDB_ERR_LOCK;
+			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+				 "tdb_nest_lock: offset %llu has read lock\n",
+				 (long long)offset);
+			return -1;
+		}
+		/* Just increment the struct, posix locks don't stack. */
 		new_lck->count++;
 		return 0;
 	}
 
+	if (tdb->num_lockrecs
+	    && offset >= TDB_HASH_LOCK_START
+	    && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
+		tdb->ecode = TDB_ERR_LOCK;
+		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+			 "tdb_nest_lock: already have a hash lock?\n");
+		return -1;
+	}
+
 	new_lck = (struct tdb_lock_type *)realloc(
 		tdb->lockrecs,
 		sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
@@ -297,6 +328,21 @@ static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
 		return -1;
 	}
 
+	/* First time we grab a lock, perhaps someone died in commit? */
+	if (!(flags & TDB_LOCK_NOCHECK)
+	    && tdb->num_lockrecs == 0
+	    && unlikely(tdb_needs_recovery(tdb))) {
+		tdb_brunlock(tdb, ltype, offset, 1);
+
+		if (tdb_lock_and_recover(tdb) == -1) {
+			return -1;
+		}
+
+		if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
+			return -1;
+		}
+	}
+
 	tdb->lockrecs[tdb->num_lockrecs].off = offset;
 	tdb->lockrecs[tdb->num_lockrecs].count = 1;
 	tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
@@ -305,40 +351,6 @@ static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
 	return 0;
 }
 
-static int tdb_lock_and_recover(struct tdb_context *tdb)
-{
-#if 0 /* FIXME */
-
-	int ret;
-
-	/* We need to match locking order in transaction commit. */
-	if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
-		return -1;
-	}
-
-	if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
-		tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
-		return -1;
-	}
-
-	ret = tdb_transaction_recover(tdb);
-
-	tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
-	tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
-
-	return ret;
-#else
-	abort();
-	return -1;
-#endif
-}
-
-static bool tdb_needs_recovery(struct tdb_context *tdb)
-{
-	/* FIXME */
-	return false;
-}
-
 static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
 {
 	int ret = -1;
@@ -377,14 +389,12 @@ static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
 	return ret;
 }
 
-#if 0
 /*
   get the transaction lock
  */
-int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
-			 enum tdb_lock_flags lockflags)
+int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
 {
-	return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, lockflags);
+	return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
 }
 
 /*
@@ -392,9 +402,8 @@ int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
  */
 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
 {
-	return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
+	return tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
 }
-#endif
 
 /* We only need to lock individual bytes, but Linux merges consecutive locks
  * so we lock in contiguous ranges. */
@@ -445,7 +454,8 @@ int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
 		return -1;
 	}
 
-	if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
+	if (tdb->allrecord_lock.count
+	    && (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
 		tdb->allrecord_lock.count++;
 		return 0;
 	}
@@ -460,7 +470,7 @@ int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
 		return -1;
 	}
 
-	if (tdb_has_locks(tdb)) {
+	if (tdb_has_hash_locks(tdb)) {
 		/* can't combine global and chain locks */
 		tdb->ecode = TDB_ERR_LOCK;
 		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
@@ -508,7 +518,7 @@ again:
 	tdb->allrecord_lock.off = upgradable;
 
 	/* Now check for needing recovery. */
-	if (unlikely(tdb_needs_recovery(tdb))) {
+	if (!(flags & TDB_LOCK_NOCHECK) && unlikely(tdb_needs_recovery(tdb))) {
 		tdb_allrecord_unlock(tdb, ltype);
 		if (tdb_lock_and_recover(tdb) == -1) {
 			return -1;
@@ -519,9 +529,9 @@ again:
 	return 0;
 }
 
-int tdb_lock_open(struct tdb_context *tdb)
+int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
 {
-	return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT);
+	return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
 }
 
 void tdb_unlock_open(struct tdb_context *tdb)
@@ -529,9 +539,16 @@ void tdb_unlock_open(struct tdb_context *tdb)
 	tdb_nest_unlock(tdb, TDB_OPEN_LOCK, F_WRLCK);
 }
 
+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)
 {
-	return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype, TDB_LOCK_WAIT);
+	/* Lock doesn't protect data, so don't check (we recurse if we do!) */
+	return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype,
+			     TDB_LOCK_WAIT | TDB_LOCK_NOCHECK);
 }
 
 void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
@@ -584,56 +601,18 @@ bool tdb_has_expansion_lock(struct tdb_context *tdb)
 	return find_nestlock(tdb, TDB_EXPANSION_LOCK) != NULL;
 }
 
-bool tdb_has_locks(struct tdb_context *tdb)
-{
-	return tdb->allrecord_lock.count || tdb->num_lockrecs;
-}
-
-#if 0
-/* lock entire database with write lock */
-int tdb_lockall(struct tdb_context *tdb)
-{
-	tdb_trace(tdb, "tdb_lockall");
-	return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
-}
-
-/* lock entire database with write lock - nonblocking varient */
-int tdb_lockall_nonblock(struct tdb_context *tdb)
-{
-	int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
-	tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
-	return ret;
-}
-
-/* unlock entire database with write lock */
-int tdb_unlockall(struct tdb_context *tdb)
-{
-	tdb_trace(tdb, "tdb_unlockall");
-	return tdb_allrecord_unlock(tdb, F_WRLCK);
-}
-
-/* lock entire database with read lock */
-int tdb_lockall_read(struct tdb_context *tdb)
-{
-	tdb_trace(tdb, "tdb_lockall_read");
-	return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
-}
-
-/* lock entire database with read lock - nonblock varient */
-int tdb_lockall_read_nonblock(struct tdb_context *tdb)
+bool tdb_has_hash_locks(struct tdb_context *tdb)
 {
-	int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
-	tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
-	return ret;
-}
+	unsigned int i;
 
-/* unlock entire database with read lock */
-int tdb_unlockall_read(struct tdb_context *tdb)
-{
-	tdb_trace(tdb, "tdb_unlockall_read");
-	return tdb_allrecord_unlock(tdb, F_RDLCK);
+	for (i=0; i<tdb->num_lockrecs; i++) {
+		if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
+		    && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
+					       + TDB_HASH_LOCK_RANGE))
+			return true;
+	}
+	return false;
 }
-#endif
 
 static bool tdb_has_free_lock(struct tdb_context *tdb)
 {
@@ -756,145 +735,6 @@ void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
 	tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
 }
 
-/* Even if the entry isn't in this hash bucket, you'd have to lock this
- * bucket to find it. */
-static int chainlock(struct tdb_context *tdb, const TDB_DATA *key,
-		     int ltype, enum tdb_lock_flags waitflag,
-		     const char *func)
-{
-	int ret;
-	uint64_t h = tdb_hash(tdb, key->dptr, key->dsize);
-
-	ret = tdb_lock_hashes(tdb, h, 1, ltype, waitflag);
-	tdb_trace_1rec(tdb, func, *key);
-	return ret;
-}
-
-/* lock/unlock one hash chain. This is meant to be used to reduce
-   contention - it cannot guarantee how many records will be locked */
-int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
-{
-	return chainlock(tdb, &key, F_WRLCK, TDB_LOCK_WAIT, "tdb_chainlock");
-}
-
-int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
-{
-	uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
-	tdb_trace_1rec(tdb, "tdb_chainunlock", key);
-	return tdb_unlock_hashes(tdb, h, 1, F_WRLCK);
-}
-
-#if 0
-/* lock/unlock one hash chain, non-blocking. This is meant to be used
-   to reduce contention - it cannot guarantee how many records will be
-   locked */
-int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
-{
-	return chainlock(tdb, &key, F_WRLCK, TDB_LOCK_NOWAIT,
-			 "tdb_chainlock_nonblock");
-}
-
-int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
-{
-	return chainlock(tdb, &key, F_RDLCK, TDB_LOCK_WAIT,
-			 "tdb_chainlock_read");
-}
-
-int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
-{
-	uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
-	tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
-	return tdb_unlock_list(tdb, h & ((1ULL << tdb->header.v.hash_bits)-1),
-			       F_RDLCK);
-}
-
-/* record lock stops delete underneath */
-int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-	if (tdb->allrecord_lock.count) {
-		return 0;
-	}
-	return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
-}
-
-/*
-  Write locks override our own fcntl readlocks, so check it here.
-  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
-  an error to fail to get the lock here.
-*/
-int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-	struct tdb_traverse_lock *i;
-	for (i = &tdb->travlocks; i; i = i->next)
-		if (i->off == off)
-			return -1;
-	if (tdb->allrecord_lock.count) {
-		if (tdb->allrecord_lock.ltype == F_WRLCK) {
-			return 0;
-		}
-		return -1;
-	}
-	return tdb_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
-}
-
-int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-	if (tdb->allrecord_lock.count) {
-		return 0;
-	}
-	return tdb_brunlock(tdb, F_WRLCK, off, 1);
-}
-
-/* fcntl locks don't stack: avoid unlocking someone else's */
-int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-	struct tdb_traverse_lock *i;
-	uint32_t count = 0;
-
-	if (tdb->allrecord_lock.count) {
-		return 0;
-	}
-
-	if (off == 0)
-		return 0;
-	for (i = &tdb->travlocks; i; i = i->next)
-		if (i->off == off)
-			count++;
-	return (count == 1 ? tdb_brunlock(tdb, F_RDLCK, off, 1) : 0);
-}
-
-/* The transaction code uses this to remove all locks. */
-void tdb_release_transaction_locks(struct tdb_context *tdb)
-{
-	unsigned int i;
-
-	if (tdb->allrecord_lock.count != 0) {
-		tdb_off_t hash_size, free_size;
-
-		hash_size = (1ULL << tdb->header.v.hash_bits)
-			* sizeof(tdb_off_t);
-		free_size = tdb->header.v.free_zones 
-			* (tdb->header.v.free_buckets + 1) * sizeof(tdb_off_t);
-
-		tdb_brunlock(tdb, tdb->allrecord_lock.ltype,
-			     tdb->header.v.hash_off, hash_size);
-		tdb_brunlock(tdb, tdb->allrecord_lock.ltype,
-			     tdb->header.v.free_off, free_size);
-		tdb->allrecord_lock.count = 0;
-		tdb->allrecord_lock.ltype = 0;
-	}
-
-	for (i = 0; i<tdb->num_lockrecs; i++) {
-		struct tdb_lock_type *lck = &tdb->lockrecs[i];
-
-		tdb_brunlock(tdb, lck->ltype, lck->off, 1);
-	}
-	tdb->num_lockrecs = 0;
-	SAFE_FREE(tdb->lockrecs);
-	tdb->header_uptodate = false;
-}
-#endif
-
 void tdb_lock_init(struct tdb_context *tdb)
 {
 	tdb->num_lockrecs = 0;