4 files changed, 96 insertions, 36 deletions

diff --git a/source4/lib/tdb/common/lock.c b/source4/lib/tdb/common/lock.c
index a5bff2d0b3..8a964371d3 100644
--- a/source4/lib/tdb/common/lock.c
+++ b/source4/lib/tdb/common/lock.c
@@ -107,6 +107,9 @@ int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
 /* lock a list in the database. list -1 is the alloc list */
 int tdb_lock(struct tdb_context *tdb, int list, int ltype)
 {
+	struct tdb_lock_type *new_lck;
+	int i;
+
 	/* a global lock allows us to avoid per chain locks */
 	if (tdb->global_lock.count && 
 	    (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
@@ -125,18 +128,50 @@ int tdb_lock(struct tdb_context *tdb, int list, int ltype)
 	if (tdb->flags & TDB_NOLOCK)
 		return 0;
 
+	for (i=0; i<tdb->num_lockrecs; i++) {
+		if (tdb->lockrecs[i].list == list) {
+			if (tdb->lockrecs[i].count == 0) {
+				/*
+				 * Can't happen, see tdb_unlock(). It should
+				 * be an assert.
+				 */
+				TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock: "
+					 "lck->count == 0 for list %d", list));
+			}
+			/*
+			 * Just increment the in-memory struct, posix locks
+			 * don't stack.
+			 */
+			tdb->lockrecs[i].count++;
+			return 0;
+		}
+	}
+
+	new_lck = (struct tdb_lock_type *)realloc(
+		tdb->lockrecs,
+		sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
+	if (new_lck == NULL) {
+		errno = ENOMEM;
+		return -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 */
-	if (tdb->locked[list+1].count == 0) {
-		if (tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0, 1)) {
-			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d ltype=%d (%s)\n", 
-				 list, ltype, strerror(errno)));
-			return -1;
-		}
-		tdb->locked[list+1].ltype = ltype;
-		tdb->num_locks++;
+	if (tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW,
+				     0, 1)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
+			 "ltype=%d (%s)\n",  list, ltype, strerror(errno)));
+		return -1;
 	}
-	tdb->locked[list+1].count++;
+
+	tdb->num_locks++;
+
+	tdb->lockrecs[tdb->num_lockrecs].list = list;
+	tdb->lockrecs[tdb->num_lockrecs].count = 1;
+	tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
+	tdb->num_lockrecs += 1;
+
 	return 0;
 }
 
@@ -146,6 +181,8 @@ int tdb_lock(struct tdb_context *tdb, int list, int ltype)
 int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
 {
 	int ret = -1;
+	int i;
+	struct tdb_lock_type *lck = NULL;
 
 	/* a global lock allows us to avoid per chain locks */
 	if (tdb->global_lock.count && 
@@ -166,19 +203,52 @@ int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
 		return ret;
 	}
 
-	if (tdb->locked[list+1].count==0) {
+	for (i=0; i<tdb->num_lockrecs; i++) {
+		if (tdb->lockrecs[i].list == list) {
+			lck = &tdb->lockrecs[i];
+			break;
+		}
+	}
+
+	if ((lck == NULL) || (lck->count == 0)) {
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
-		return ret;
+		return -1;
+	}
+
+	if (lck->count > 1) {
+		lck->count--;
+		return 0;
 	}
 
-	if (tdb->locked[list+1].count == 1) {
-		/* Down to last nested lock: unlock underneath */
-		ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0, 1);
-		tdb->num_locks--;
-	} else {
-		ret = 0;
+	/*
+	 * This lock has count==1 left, so we need to unlock it in the
+	 * kernel. We don't bother with decrementing the in-memory array
+	 * element, we're about to overwrite it with the last array element
+	 * anyway.
+	 */
+
+	ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK,
+				       F_SETLKW, 0, 1);
+	tdb->num_locks--;
+
+	/*
+	 * Shrink the array by overwriting the element just unlocked with the
+	 * last array element.
+	 */
+
+	if (tdb->num_lockrecs > 1) {
+		*lck = tdb->lockrecs[tdb->num_lockrecs-1];
+	}
+	tdb->num_lockrecs -= 1;
+
+	/*
+	 * We don't bother with realloc when the array shrinks, but if we have
+	 * a completely idle tdb we should get rid of the locked array.
+	 */
+
+	if (tdb->num_lockrecs == 0) {
+		SAFE_FREE(tdb->lockrecs);
 	}
-	tdb->locked[list+1].count--;
 
 	if (ret)
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n")); 
diff --git a/source4/lib/tdb/common/open.c b/source4/lib/tdb/common/open.c
index e1f21aa856..3d6e222b96 100644
--- a/source4/lib/tdb/common/open.c
+++ b/source4/lib/tdb/common/open.c
@@ -263,15 +263,6 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 	tdb->map_size = st.st_size;
 	tdb->device = st.st_dev;
 	tdb->inode = st.st_ino;
-	tdb->locked = (struct tdb_lock_type *)calloc(tdb->header.hash_size+1,
-						     sizeof(tdb->locked[0]));
-	if (!tdb->locked) {
-		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
-			 "failed to allocate lock structure for %s\n",
-			 name));
-		errno = ENOMEM;
-		goto fail;
-	}
 	tdb_mmap(tdb);
 	if (locked) {
 		if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
@@ -324,7 +315,6 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 	if (tdb->fd != -1)
 		if (close(tdb->fd) != 0)
 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
-	SAFE_FREE(tdb->locked);
 	SAFE_FREE(tdb);
 	errno = save_errno;
 	return NULL;
@@ -354,7 +344,7 @@ int tdb_close(struct tdb_context *tdb)
 	SAFE_FREE(tdb->name);
 	if (tdb->fd != -1)
 		ret = close(tdb->fd);
-	SAFE_FREE(tdb->locked);
+	SAFE_FREE(tdb->lockrecs);
 
 	/* Remove from contexts list */
 	for (i = &tdbs; *i; i = &(*i)->next) {
diff --git a/source4/lib/tdb/common/tdb_private.h b/source4/lib/tdb/common/tdb_private.h
index a3495e42aa..5277857942 100644
--- a/source4/lib/tdb/common/tdb_private.h
+++ b/source4/lib/tdb/common/tdb_private.h
@@ -123,6 +123,7 @@ struct tdb_header {
 };
 
 struct tdb_lock_type {
+	int list;
 	u32 count;
 	u32 ltype;
 };
@@ -152,7 +153,8 @@ struct tdb_context {
 	int read_only; /* opened read-only */
 	int traverse_read; /* read-only traversal */
 	struct tdb_lock_type global_lock;
-	struct tdb_lock_type *locked; /* array of chain locks */
+	int num_lockrecs;
+	struct tdb_lock_type *lockrecs; /* only real locks, all with count>0 */
 	enum TDB_ERROR ecode; /* error code for last tdb error */
 	struct tdb_header header; /* a cached copy of the header */
 	u32 flags; /* the flags passed to tdb_open */
diff --git a/source4/lib/tdb/common/transaction.c b/source4/lib/tdb/common/transaction.c
index 7dff9a95e3..5c94eb0afe 100644
--- a/source4/lib/tdb/common/transaction.c
+++ b/source4/lib/tdb/common/transaction.c
@@ -516,12 +516,10 @@ int tdb_transaction_cancel(struct tdb_context *tdb)
 
 	/* remove any locks created during the transaction */
 	if (tdb->num_locks != 0) {
-		int h;
-		for (h=0;h<tdb->header.hash_size+1;h++) {
-			if (tdb->locked[h].count != 0) {
-				tdb_brlock(tdb,FREELIST_TOP+4*h,F_UNLCK,F_SETLKW, 0, 1);
-				tdb->locked[h].count = 0;
-			}
+		int i;
+		for (i=0;i<tdb->num_lockrecs;i++) {
+			tdb_brlock(tdb,FREELIST_TOP+4*tdb->lockrecs[i].list,
+				   F_UNLCK,F_SETLKW, 0, 1);
 		}
 		tdb->num_locks = 0;
 	}