X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Flock.c;h=4033d0afe813e59bf4dee4b8bcc9e5d34757c298;hp=3230b25e383d3b2a3ead9a1ac7ca156b4030f303;hb=dc9da1e34fe6a9d113fd57e116ebbc6d5bd54819;hpb=4916bc0a12ede1bf81c6895be9e7df39199f3290

diff --git a/ccan/tdb2/lock.c b/ccan/tdb2/lock.c
index 3230b25e..4033d0af 100644
--- a/ccan/tdb2/lock.c
+++ b/ccan/tdb2/lock.c
@@ -1,4 +1,4 @@
- /* 
+ /*
    Unix SMB/CIFS implementation.
 
    trivial database library
@@ -26,27 +26,93 @@
 */
 
 #include "private.h"
+#include <assert.h>
+#include <ccan/build_assert/build_assert.h>
 
-static int fcntl_lock(struct tdb_context *tdb,
-		      int rw, off_t off, off_t len, bool waitflag)
+/* If we were threaded, we could wait for unlock, but we're not, so fail. */
+static enum TDB_ERROR owner_conflict(struct tdb_context *tdb, const char *call)
+{
+	return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+			  "%s: lock owned by another tdb in this process.",
+			  call);
+}
+
+/* If we fork, we no longer really own locks: preserves errno */
+static bool check_lock_pid(struct tdb_context *tdb,
+			   const char *call, bool log)
+{
+	/* No locks?  No problem! */
+	if (tdb->file->allrecord_lock.count == 0
+	    && tdb->file->num_lockrecs == 0) {
+		return true;
+	}
+
+	/* No fork?  No problem! */
+	if (tdb->file->locker == getpid()) {
+		return true;
+	}
+
+	if (log) {
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+			   "%s: fork() detected after lock acquisition!"
+			   " (%u vs %u)", call, tdb->file->locker, getpid());
+	}
+	return false;
+}
+
+int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag,
+		   void *unused)
 {
 	struct flock fl;
+	int ret;
 
-	fl.l_type = rw;
-	fl.l_whence = SEEK_SET;
-	fl.l_start = off;
-	fl.l_len = len;
-	fl.l_pid = 0;
+	do {
+		fl.l_type = rw;
+		fl.l_whence = SEEK_SET;
+		fl.l_start = off;
+		fl.l_len = len;
 
-	if (waitflag)
-		return fcntl(tdb->fd, F_SETLKW, &fl);
-	else
-		return fcntl(tdb->fd, F_SETLK, &fl);
+		if (waitflag)
+			ret = fcntl(fd, F_SETLKW, &fl);
+		else
+			ret = fcntl(fd, F_SETLK, &fl);
+	} while (ret != 0 && errno == EINTR);
+	return ret;
 }
 
-static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
+int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *unused)
 {
 	struct flock fl;
+	int ret;
+
+	do {
+		fl.l_type = F_UNLCK;
+		fl.l_whence = SEEK_SET;
+		fl.l_start = off;
+		fl.l_len = len;
+
+		ret = fcntl(fd, F_SETLKW, &fl);
+	} while (ret != 0 && errno == EINTR);
+	return ret;
+}
+
+static int lock(struct tdb_context *tdb,
+		      int rw, off_t off, off_t len, bool waitflag)
+{
+	if (tdb->file->allrecord_lock.count == 0
+	    && tdb->file->num_lockrecs == 0) {
+		tdb->file->locker = getpid();
+	}
+
+	tdb->stats.lock_lowlevel++;
+	if (!waitflag)
+		tdb->stats.lock_nonblock++;
+	return tdb->lock_fn(tdb->file->fd, rw, off, len, waitflag,
+			    tdb->lock_data);
+}
+
+static int unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
+{
 #if 0 /* Check they matched up locks and unlocks correctly. */
 	char line[80];
 	FILE *locks;
@@ -97,7 +163,7 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
 	}
 
 	if (!found) {
-		fprintf(stderr, "Unlock on %u@%u not found!\n",
+		fprintf(stderr, "Unlock on %u@%u not found!",
 			(int)off, (int)len);
 		abort();
 	}
@@ -105,122 +171,116 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
 	fclose(locks);
 #endif
 
-	fl.l_type = F_UNLCK;
-	fl.l_whence = SEEK_SET;
-	fl.l_start = off;
-	fl.l_len = len;
-	fl.l_pid = 0;
-
-	return fcntl(tdb->fd, F_SETLKW, &fl);
+	return tdb->unlock_fn(tdb->file->fd, rw, off, len, tdb->lock_data);
 }
 
 /* a byte range locking function - return 0 on success
-   this functions locks/unlocks 1 byte at the specified offset.
+   this functions locks len bytes at the specified offset.
 
    note that a len of zero means lock to end of file
 */
-static int tdb_brlock(struct tdb_context *tdb,
-		      int rw_type, tdb_off_t offset, tdb_off_t len,
-		      enum tdb_lock_flags flags)
+static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
+				 int rw_type, tdb_off_t offset, tdb_off_t len,
+				 enum tdb_lock_flags flags)
 {
 	int ret;
 
 	if (tdb->flags & TDB_NOLOCK) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	if (rw_type == F_WRLCK && tdb->read_only) {
-		tdb->ecode = TDB_ERR_RDONLY;
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
+				  "Write lock attempted on read-only database");
 	}
 
 	/* A 32 bit system cannot open a 64-bit file, but it could have
 	 * expanded since then: check here. */
 	if ((size_t)(offset + len) != offset + len) {
-		tdb->ecode = TDB_ERR_IO;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_brlock: lock on giant offset %llu\n",
-			 (long long)(offset + len));
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				  "tdb_brlock: lock on giant offset %llu",
+				  (long long)(offset + len));
 	}
 
-	do {
-		ret = fcntl_lock(tdb, rw_type, offset, len,
-				 flags & TDB_LOCK_WAIT);
-	} while (ret == -1 && errno == EINTR);
-
-	if (ret == -1) {
-		tdb->ecode = TDB_ERR_LOCK;
+	ret = lock(tdb, rw_type, offset, len, flags & TDB_LOCK_WAIT);
+	if (ret != 0) {
 		/* Generic lock error. errno set by fcntl.
 		 * EAGAIN is an expected return from non-blocking
 		 * locks. */
-		if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
-			tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-				 "tdb_brlock failed (fd=%d) at"
-				 " offset %llu rw_type=%d flags=%d len=%llu\n",
-				 tdb->fd, (long long)offset, rw_type,
-				 flags, (long long)len);
+		if (!(flags & TDB_LOCK_PROBE)
+		    && (errno != EAGAIN && errno != EINTR)) {
+			tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				   "tdb_brlock failed (fd=%d) at"
+				   " offset %zu rw_type=%d flags=%d len=%zu:"
+				   " %s",
+				   tdb->file->fd, (size_t)offset, rw_type,
+				   flags, (size_t)len, strerror(errno));
 		}
-		return -1;
+		return TDB_ERR_LOCK;
 	}
-	return 0;
+	return TDB_SUCCESS;
 }
 
-static int tdb_brunlock(struct tdb_context *tdb,
-			int rw_type, tdb_off_t offset, size_t len)
+static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
+				   int rw_type, tdb_off_t offset, size_t len)
 {
 	int ret;
 
 	if (tdb->flags & TDB_NOLOCK) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
-	do {
-		ret = fcntl_unlock(tdb, rw_type, offset, len);
-	} while (ret == -1 && errno == EINTR);
+	ret = unlock(tdb, rw_type, offset, len);
 
-	if (ret == -1) {
-		tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
-			 "tdb_brunlock failed (fd=%d) at offset %llu"
-			 " rw_type=%d len=%llu\n",
-			 tdb->fd, (long long)offset, rw_type, (long long)len);
+	/* If we fail, *then* we verify that we owned the lock.  If not, ok. */
+	if (ret == -1 && check_lock_pid(tdb, "tdb_brunlock", false)) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_brunlock failed (fd=%d) at offset %zu"
+				  " rw_type=%d len=%zu: %s",
+				  tdb->file->fd, (size_t)offset, rw_type,
+				  (size_t)len, strerror(errno));
 	}
-	return ret;
+	return TDB_SUCCESS;
 }
 
-#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
   deadlock detection and claim a deadlock when progress can be
-  made. For those OSes we may loop for a while.  
+  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->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_upgrade failed: count %u too high\n",
-			 tdb->allrecord_lock.count);
-		return -1;
+	if (!check_lock_pid(tdb, "tdb_transaction_prepare_commit", true))
+		return TDB_ERR_LOCK;
+
+	if (tdb->file->allrecord_lock.count != 1) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_allrecord_upgrade failed:"
+				  " count %u too high",
+				  tdb->file->allrecord_lock.count);
 	}
 
-	if (tdb->allrecord_lock.off != 1) {
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_upgrade failed: already upgraded?\n");
-		return -1;
+	if (tdb->file->allrecord_lock.off != 1) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_allrecord_upgrade failed:"
+				  " already upgraded?");
+	}
+
+	if (tdb->file->allrecord_lock.owner != tdb) {
+		return owner_conflict(tdb, "tdb_allrecord_upgrade");
 	}
 
 	while (count--) {
 		struct timeval tv;
 		if (tdb_brlock(tdb, F_WRLCK,
-			       TDB_HASH_LOCK_START
-			       + (1ULL << tdb->header.v.hash_bits), 0,
-			       TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
-			tdb->allrecord_lock.ltype = F_WRLCK;
-			tdb->allrecord_lock.off = 0;
-			return 0;
+			       TDB_HASH_LOCK_START, 0,
+			       TDB_LOCK_WAIT|TDB_LOCK_PROBE) == TDB_SUCCESS) {
+			tdb->file->allrecord_lock.ltype = F_WRLCK;
+			tdb->file->allrecord_lock.off = 0;
+			return TDB_SUCCESS;
 		}
 		if (errno != EDEADLK) {
 			break;
@@ -230,132 +290,169 @@ int tdb_allrecord_upgrade(struct tdb_context *tdb)
 		tv.tv_usec = 1;
 		select(0, NULL, NULL, NULL, &tv);
 	}
-	tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
-		 "tdb_allrecord_upgrade failed\n");
-	return -1;
+
+	if (errno != EAGAIN && errno != EINTR)
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+			   "tdb_allrecord_upgrade failed");
+	return TDB_ERR_LOCK;
 }
-#endif
 
-static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
-					   tdb_off_t offset)
+static struct tdb_lock *find_nestlock(struct tdb_context *tdb, tdb_off_t offset,
+				      const struct tdb_context *owner)
 {
 	unsigned int i;
 
-	for (i=0; i<tdb->num_lockrecs; i++) {
-		if (tdb->lockrecs[i].off == offset) {
-			return &tdb->lockrecs[i];
+	for (i=0; i<tdb->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].off == offset) {
+			if (owner && tdb->file->lockrecs[i].owner != owner)
+				return NULL;
+			return &tdb->file->lockrecs[i];
 		}
 	}
 	return NULL;
 }
 
-/* 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)
+enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb)
 {
-	struct tdb_lock_type *new_lck;
+	enum TDB_ERROR ecode;
 
-	if (offset >= TDB_HASH_LOCK_START + (1ULL << tdb->header.v.hash_bits)
-	    + (tdb->header.v.num_zones * (tdb->header.v.free_buckets+1))) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_lock: invalid offset %llu for ltype=%d\n",
-			 (long long)offset, ltype);
-		return -1;
-	}
-	if (tdb->flags & TDB_NOLOCK)
-		return 0;
+	if (!check_lock_pid(tdb, "tdb_transaction_prepare_commit", true))
+		return TDB_ERR_LOCK;
 
-	new_lck = find_nestlock(tdb, offset);
-	if (new_lck) {
-		/*
-		 * Just increment the in-memory struct, posix locks
-		 * don't stack.
-		 */
-		new_lck->count++;
-		return 0;
+	ecode = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
+				   false);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
 	}
 
-	new_lck = (struct tdb_lock_type *)realloc(
-		tdb->lockrecs,
-		sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
-	if (new_lck == NULL) {
-		tdb->ecode = TDB_ERR_OOM;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_lock: unable to allocate %llu lock structure",
-			 (long long)(tdb->num_lockrecs + 1));
-		errno = ENOMEM;
-		return -1;
+	ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
+	if (ecode != TDB_SUCCESS) {
+		tdb_allrecord_unlock(tdb, F_WRLCK);
+		return ecode;
 	}
-	tdb->lockrecs = new_lck;
+	ecode = tdb_transaction_recover(tdb);
+	tdb_unlock_open(tdb);
+	tdb_allrecord_unlock(tdb, F_WRLCK);
 
-	/* Since fcntl locks don't nest, we do a lock for the first one,
-	   and simply bump the count for future ones */
-	if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
-		return -1;
+	return ecode;
+}
+
+/* lock an offset in the database. */
+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 *new_lck;
+	enum TDB_ERROR ecode;
+
+	if (offset > (TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
+		      + tdb->file->map_size / 8)) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_nest_lock: invalid offset %zu ltype=%d",
+				  (size_t)offset, ltype);
 	}
 
-	tdb->lockrecs[tdb->num_lockrecs].off = offset;
-	tdb->lockrecs[tdb->num_lockrecs].count = 1;
-	tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
-	tdb->num_lockrecs++;
+	if (tdb->flags & TDB_NOLOCK)
+		return TDB_SUCCESS;
 
-	return 0;
-}
+	if (!check_lock_pid(tdb, "tdb_nest_lock", true)) {
+		return TDB_ERR_LOCK;
+	}
 
-static int tdb_lock_and_recover(struct tdb_context *tdb)
-{
-#if 0 /* FIXME */
+	tdb->stats.locks++;
 
-	int ret;
+	new_lck = find_nestlock(tdb, offset, NULL);
+	if (new_lck) {
+		if (new_lck->owner != tdb) {
+			return owner_conflict(tdb, "tdb_nest_lock");
+		}
 
-	/* We need to match locking order in transaction commit. */
-	if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
-		return -1;
+		if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
+			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 TDB_SUCCESS;
 	}
 
-	if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
-		tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
-		return -1;
+#if 0
+	if (tdb->file->num_lockrecs
+	    && offset >= TDB_HASH_LOCK_START
+	    && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_nest_lock: already have a hash lock?");
 	}
+#endif
 
-	ret = tdb_transaction_recover(tdb);
+	new_lck = (struct tdb_lock *)realloc(
+		tdb->file->lockrecs,
+		sizeof(*tdb->file->lockrecs) * (tdb->file->num_lockrecs+1));
+	if (new_lck == NULL) {
+		return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+				  "tdb_nest_lock:"
+				  " unable to allocate %zu lock struct",
+				  tdb->file->num_lockrecs + 1);
+	}
+	tdb->file->lockrecs = new_lck;
 
-	tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
-	tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
+	/* 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_SUCCESS) {
+		return ecode;
+	}
+
+	/* First time we grab a lock, perhaps someone died in commit? */
+	if (!(flags & TDB_LOCK_NOCHECK)
+	    && tdb->file->num_lockrecs == 0) {
+		tdb_bool_err berr = tdb_needs_recovery(tdb);
+		if (berr != false) {
+			tdb_brunlock(tdb, ltype, offset, 1);
+
+			if (berr < 0)
+				return berr;
+			ecode = tdb_lock_and_recover(tdb);
+			if (ecode == TDB_SUCCESS) {
+				ecode = tdb_brlock(tdb, ltype, offset, 1,
+						   flags);
+			}
+			if (ecode != TDB_SUCCESS) {
+				return ecode;
+			}
+		}
+	}
 
-	return ret;
-#else
-	abort();
-	return -1;
-#endif
-}
+	tdb->file->lockrecs[tdb->file->num_lockrecs].owner = tdb;
+	tdb->file->lockrecs[tdb->file->num_lockrecs].off = offset;
+	tdb->file->lockrecs[tdb->file->num_lockrecs].count = 1;
+	tdb->file->lockrecs[tdb->file->num_lockrecs].ltype = ltype;
+	tdb->file->num_lockrecs++;
 
-static bool tdb_needs_recovery(struct tdb_context *tdb)
-{
-	/* FIXME */
-	return false;
+	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)
 {
-	int ret = -1;
-	struct tdb_lock_type *lck;
+	struct tdb_lock *lck;
+	enum TDB_ERROR ecode;
 
 	if (tdb->flags & TDB_NOLOCK)
-		return 0;
+		return TDB_SUCCESS;
 
-	lck = find_nestlock(tdb, off);
+	lck = find_nestlock(tdb, off, tdb);
 	if ((lck == NULL) || (lck->count == 0)) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_unlock: no lock for %llu\n", (long long)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;
 	}
 
 	/*
@@ -364,164 +461,160 @@ static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
 	 * element, we're about to overwrite it with the last array element
 	 * anyway.
 	 */
-	ret = tdb_brunlock(tdb, ltype, off, 1);
+	ecode = tdb_brunlock(tdb, ltype, off, 1);
 
 	/*
 	 * Shrink the array by overwriting the element just unlocked with the
 	 * last array element.
 	 */
-	*lck = tdb->lockrecs[--tdb->num_lockrecs];
+	*lck = tdb->file->lockrecs[--tdb->file->num_lockrecs];
 
-	if (tdb->num_lockrecs == 0) {
-		/* If we're not holding any locks, header can change. */
-		tdb->header_uptodate = false;
-	}
-
-	return ret;
+	return ecode;
 }
 
-#if 0
 /*
   get the transaction lock
  */
-int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
-			 enum tdb_lock_flags lockflags)
+enum TDB_ERROR 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);
 }
 
 /*
   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, TRANSACTION_LOCK, ltype, false);
+	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. */
-static int tdb_lock_gradual(struct tdb_context *tdb,
-			    int ltype, enum tdb_lock_flags flags,
-			    tdb_off_t off, tdb_off_t len)
+static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
+				       int ltype, enum tdb_lock_flags flags,
+				       tdb_off_t off, tdb_off_t len)
 {
-	int ret;
+	enum TDB_ERROR ecode;
 	enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
 
-	if (len <= 4) {
-		/* Single record.  Just do blocking lock. */
+	if (len <= 1) {
+		/* 0 would mean to end-of-file... */
+		assert(len != 0);
+		/* Single hash.  Just do blocking lock. */
 		return tdb_brlock(tdb, ltype, off, len, flags);
 	}
 
 	/* First we try non-blocking. */
-	ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
-	if (ret == 0) {
-		return 0;
+	if (tdb_brlock(tdb, ltype, off, len, nb_flags) == TDB_SUCCESS) {
+		return TDB_SUCCESS;
 	}
 
 	/* Try locking first half, then second. */
-	ret = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
-	if (ret == -1)
-		return -1;
+	ecode = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
+	if (ecode != TDB_SUCCESS)
+		return ecode;
 
-	ret = tdb_lock_gradual(tdb, ltype, flags,
-				    off + len / 2, len - len / 2);
-	if (ret == -1) {
+	ecode = tdb_lock_gradual(tdb, ltype, flags,
+				 off + len / 2, len - len / 2);
+	if (ecode != TDB_SUCCESS) {
 		tdb_brunlock(tdb, ltype, off, len / 2);
-		return -1;
 	}
-	return 0;
+	return ecode;
 }
 
 /* lock/unlock entire database.  It can only be upgradable if you have some
- * other way of guaranteeing exclusivity (ie. transaction write lock).
- * Note that we don't lock the free chains: noone can get those locks
- * without a hash chain lock first.
- * The header *will be* up to date once this returns success. */
-int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
-		       enum tdb_lock_flags flags, bool upgradable)
+ * other way of guaranteeing exclusivity (ie. transaction write lock). */
+enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
+				  enum tdb_lock_flags flags, bool upgradable)
 {
-	tdb_off_t hash_size;
+	enum TDB_ERROR ecode;
+	tdb_bool_err berr;
 
-	/* FIXME: There are no locks on read-only dbs */
-	if (tdb->read_only) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_lock: read-only\n");
-		return -1;
-	}
+	if (tdb->flags & TDB_NOLOCK)
+		return TDB_SUCCESS;
 
-	if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
-		tdb->allrecord_lock.count++;
-		return 0;
+	if (!check_lock_pid(tdb, "tdb_allrecord_lock", true)) {
+		return TDB_ERR_LOCK;
 	}
 
-	if (tdb->allrecord_lock.count) {
+	if (tdb->file->allrecord_lock.count) {
+		if (tdb->file->allrecord_lock.owner != tdb) {
+			return owner_conflict(tdb, "tdb_allrecord_lock");
+		}
+
+		if (ltype == F_RDLCK
+		    || tdb->file->allrecord_lock.ltype == F_WRLCK) {
+			tdb->file->allrecord_lock.count++;
+			return TDB_SUCCESS;
+		}
+
 		/* a global lock of a different type exists */
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_lock: already have %s lock\n",
-			 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->file->allrecord_lock.ltype == F_RDLCK
+				  ? "read" : "write");
 	}
 
-	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,
-			 "tdb_allrecord_lock: already have chain lock\n");
-		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->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_lock: can't upgrade a write lock\n");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_allrecord_lock:"
+				  " can't upgrade a write lock");
 	}
 
-	/* Lock all the hash buckets. */
+	tdb->stats.locks++;
 again:
-	hash_size = (1ULL << tdb->header.v.hash_bits);
-	if (tdb_lock_gradual(tdb, ltype, TDB_HASH_LOCK_START,
-			     1ULL << tdb->header.v.hash_bits, flags)) {
-		if (!(flags & TDB_LOCK_PROBE)) {
-			tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-				 "tdb_lockall hashes failed (%s)\n",
-				 strerror(errno));
-		}
-		return -1;
-	}
-
-	tdb->allrecord_lock.count = 1;
+	/* Lock hashes, gradually. */
+	ecode = tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
+				 TDB_HASH_LOCK_RANGE);
+	if (ecode != TDB_SUCCESS)
+		return ecode;
+
+	/* Lock free tables: there to end of file. */
+	ecode = tdb_brlock(tdb, ltype,
+			   TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
+			   0, flags);
+	if (ecode != TDB_SUCCESS) {
+		tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
+			     TDB_HASH_LOCK_RANGE);
+		return ecode;
+	}
+
+	tdb->file->allrecord_lock.owner = tdb;
+	tdb->file->allrecord_lock.count = 1;
 	/* If it's upgradable, it's actually exclusive so we can treat
 	 * it as a write lock. */
-	tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
-	tdb->allrecord_lock.off = upgradable;
-
-	/* Now we re-check header, holding lock. */
-	if (unlikely(header_changed(tdb))) {
-		tdb_allrecord_unlock(tdb, ltype);
-		goto again;
-	}
+	tdb->file->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
+	tdb->file->allrecord_lock.off = upgradable;
 
 	/* Now check for needing recovery. */
-	if (unlikely(tdb_needs_recovery(tdb))) {
-		tdb_allrecord_unlock(tdb, ltype);
-		if (tdb_lock_and_recover(tdb) == -1) {
-			return -1;
-		}		
-		goto again;
-	}
+	if (flags & TDB_LOCK_NOCHECK)
+		return TDB_SUCCESS;
 
-	return 0;
+	berr = tdb_needs_recovery(tdb);
+	if (likely(berr == false))
+		return TDB_SUCCESS;
+
+	tdb_allrecord_unlock(tdb, ltype);
+	if (berr < 0)
+		return berr;
+	ecode = tdb_lock_and_recover(tdb);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
+	}
+	goto again;
 }
 
-int tdb_lock_open(struct tdb_context *tdb)
+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, TDB_LOCK_WAIT);
+	return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
 }
 
 void tdb_unlock_open(struct tdb_context *tdb)
@@ -529,335 +622,246 @@ 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 !(tdb->flags & TDB_NOLOCK)
+		&& find_nestlock(tdb, TDB_OPEN_LOCK, tdb) != NULL;
+}
+
+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,
+			     TDB_LOCK_WAIT | TDB_LOCK_NOCHECK);
+}
+
+void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
+{
+	tdb_nest_unlock(tdb, TDB_EXPANSION_LOCK, ltype);
+}
+
 /* unlock entire db */
-int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
+void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
 {
-	tdb_off_t hash_size;
+	if (tdb->flags & TDB_NOLOCK)
+		return;
 
-	/* FIXME: There are no locks on read-only dbs */
-	if (tdb->read_only) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_unlock: read-only\n");
-		return -1;
+	if (tdb->file->allrecord_lock.count == 0) {
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+			   "tdb_allrecord_unlock: not locked!");
+		return;
 	}
 
-	if (tdb->allrecord_lock.count == 0) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_unlock: not locked!\n");
-		return -1;
+	if (tdb->file->allrecord_lock.owner != tdb) {
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+			   "tdb_allrecord_unlock: not locked by us!");
+		return;
 	}
 
 	/* Upgradable locks are marked as write locks. */
-	if (tdb->allrecord_lock.ltype != ltype
-	    && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_allrecord_unlock: have %s lock\n",
-			 tdb->allrecord_lock.ltype == F_RDLCK
-			 ? "read" : "write");
-		return -1;
-	}
-
-	if (tdb->allrecord_lock.count > 1) {
-		tdb->allrecord_lock.count--;
-		return 0;
+	if (tdb->file->allrecord_lock.ltype != ltype
+	    && (!tdb->file->allrecord_lock.off || ltype != F_RDLCK)) {
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+			   "tdb_allrecord_unlock: have %s lock",
+			   tdb->file->allrecord_lock.ltype == F_RDLCK
+			   ? "read" : "write");
+		return;
 	}
 
-	tdb->allrecord_lock.count = 0;
-	tdb->allrecord_lock.ltype = 0;
-	tdb->header_uptodate = false;
+	if (tdb->file->allrecord_lock.count > 1) {
+		tdb->file->allrecord_lock.count--;
+		return;
+	}
 
-	hash_size = (1ULL << tdb->header.v.hash_bits);
+	tdb->file->allrecord_lock.count = 0;
+	tdb->file->allrecord_lock.ltype = 0;
 
-	return tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, hash_size);
+	tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
 }
 
-bool tdb_has_locks(struct tdb_context *tdb)
+bool tdb_has_expansion_lock(struct tdb_context *tdb)
 {
-	return tdb->allrecord_lock.count || tdb->num_lockrecs;
+	return find_nestlock(tdb, TDB_EXPANSION_LOCK, tdb) != NULL;
 }
 
-#if 0
-/* lock entire database with write lock */
-int tdb_lockall(struct tdb_context *tdb)
+bool tdb_has_hash_locks(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;
-}
+	unsigned int i;
 
-/* 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);
+	for (i=0; i<tdb->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].off >= TDB_HASH_LOCK_START
+		    && tdb->file->lockrecs[i].off < (TDB_HASH_LOCK_START
+						     + TDB_HASH_LOCK_RANGE))
+			return true;
+	}
+	return false;
 }
 
-/* lock entire database with read lock */
-int tdb_lockall_read(struct tdb_context *tdb)
+static bool tdb_has_free_lock(struct tdb_context *tdb)
 {
-	tdb_trace(tdb, "tdb_lockall_read");
-	return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
-}
+	unsigned int i;
 
-/* lock entire database with read lock - nonblock varient */
-int tdb_lockall_read_nonblock(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;
-}
+	if (tdb->flags & TDB_NOLOCK)
+		return false;
 
-/* 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->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].off
+		    > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE)
+			return true;
+	}
+	return false;
 }
-#endif
 
-/* Returns the list we actually locked. */
-tdb_off_t tdb_lock_list(struct tdb_context *tdb, uint64_t hash,
-			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)
 {
-	tdb_off_t list = hash & ((1ULL << tdb->header.v.hash_bits) - 1);
-	/* Header can change ONLY if we had no locks before. */
-	bool can_change = tdb->num_lockrecs == 0;
+	/* FIXME: Do this properly, using hlock_range */
+	unsigned l = TDB_HASH_LOCK_START
+		+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
 
 	/* a allrecord lock allows us to avoid per chain locks */
-	if (tdb->allrecord_lock.count &&
-	    (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
-		return list;
+	if (tdb->file->allrecord_lock.count) {
+		if (!check_lock_pid(tdb, "tdb_lock_hashes", true))
+			return TDB_ERR_LOCK;
+
+		if (tdb->file->allrecord_lock.owner != tdb)
+			return owner_conflict(tdb, "tdb_lock_hashes");
+		if (ltype == tdb->file->allrecord_lock.ltype
+		    || ltype == F_RDLCK) {
+			return TDB_SUCCESS;
+		}
+
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+				  "tdb_lock_hashes:"
+				  " already have %s allrecordlock",
+				  tdb->file->allrecord_lock.ltype == F_RDLCK
+				  ? "read" : "write");
 	}
 
-	if (tdb->allrecord_lock.count) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_lock_list: have %s allrecordlock\n",
-			 tdb->allrecord_lock.ltype == F_RDLCK
-			 ? "read" : "write");
-		return TDB_OFF_ERR;
+	if (tdb_has_free_lock(tdb)) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_lock_hashes: already have free lock");
 	}
 
-again:
-	if (tdb_nest_lock(tdb, TDB_HASH_LOCK_START + list, ltype, waitflag))
-		return TDB_OFF_ERR;
-
-	if (can_change && unlikely(header_changed(tdb))) {
-		tdb_off_t new = hash & ((1ULL << tdb->header.v.hash_bits) - 1);
-		if (new != list) {
-			tdb_nest_unlock(tdb, TDB_HASH_LOCK_START+list, ltype);
-			list = new;
-			goto again;
-		}
+	if (tdb_has_expansion_lock(tdb)) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_lock_hashes:"
+				  " already have expansion lock");
 	}
-	return list;
+
+	return tdb_nest_lock(tdb, l, ltype, waitflag);
 }
 
-int tdb_unlock_list(struct tdb_context *tdb, tdb_off_t list, 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 l = TDB_HASH_LOCK_START
+		+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
+
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
 	/* a allrecord lock allows us to avoid per chain locks */
-	if (tdb->allrecord_lock.count) {
-		if (tdb->allrecord_lock.ltype == F_RDLCK
+	if (tdb->file->allrecord_lock.count) {
+		if (tdb->file->allrecord_lock.ltype == F_RDLCK
 		    && ltype == F_WRLCK) {
-			tdb->ecode = TDB_ERR_LOCK;
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-				 "tdb_unlock_list RO allrecord!\n");
-			return -1;
+			return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+					  "tdb_unlock_hashes RO allrecord!");
 		}
-		return 0;
-	} else {
-		return tdb_nest_unlock(tdb, TDB_HASH_LOCK_START + list, ltype);
+		return TDB_SUCCESS;
 	}
+
+	return tdb_nest_unlock(tdb, l, ltype);
 }
 
-/* Free list locks come after hash locks */
-int tdb_lock_free_list(struct tdb_context *tdb, tdb_off_t flist,
-		       enum tdb_lock_flags waitflag)
+/* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
+ * Then we begin; bucket offsets are sizeof(tdb_len_t) apart, so we divide.
+ * The result is that on 32 bit systems we don't use lock values > 2^31 on
+ * files that are less than 4GB.
+ */
+static tdb_off_t free_lock_off(tdb_off_t b_off)
 {
-	/* You're supposed to have a hash lock first! */
-	if (!(tdb->flags & TDB_NOLOCK) && !tdb_has_locks(tdb)) {
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_lock_free_list without lock!\n");
-		return -1;
-	}
-
-	/* a allrecord lock allows us to avoid per chain locks */
-	if (tdb->allrecord_lock.count) {
-		if (tdb->allrecord_lock.ltype == F_WRLCK)
-			return 0;
-		tdb->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_lock_free_list with RO allrecordlock!\n");
-		return -1;
-	}
-
-	return tdb_nest_lock(tdb, TDB_HASH_LOCK_START
-			     + (1ULL << tdb->header.v.hash_bits)
-			     + flist, F_WRLCK, waitflag);
+	return TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
+		+ b_off / sizeof(tdb_off_t);
 }
 
-void tdb_unlock_free_list(struct tdb_context *tdb, tdb_off_t flist)
+enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
+				    enum tdb_lock_flags waitflag)
 {
-	if (tdb->allrecord_lock.count)
-		return;
+	assert(b_off >= sizeof(struct tdb_header));
 
-	tdb_nest_unlock(tdb, TDB_HASH_LOCK_START
-			+ (1ULL << tdb->header.v.hash_bits)
-			+ flist, F_WRLCK);
-}
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
 
-#if 0
-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);
+	/* a allrecord lock allows us to avoid per chain locks */
+	if (tdb->file->allrecord_lock.count) {
+		if (!check_lock_pid(tdb, "tdb_lock_free_bucket", true))
+			return TDB_ERR_LOCK;
 
-	ret = tdb_lock_list(tdb, h, ltype, waitflag) == TDB_OFF_ERR ? -1 : 0;
-	tdb_trace_1rec(tdb, func, *key);
-	return ret;
-}
+		if (tdb->file->allrecord_lock.ltype == F_WRLCK)
+			return 0;
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_lock_free_bucket with"
+				  " read-only allrecordlock!");
+	}
 
-/* 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");
-}
+#if 0 /* FIXME */
+	if (tdb_has_expansion_lock(tdb)) {
+		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+				  "tdb_lock_free_bucket:"
+				  " already have expansion lock");
+	}
+#endif
 
-/* 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");
+	return tdb_nest_lock(tdb, free_lock_off(b_off), F_WRLCK, waitflag);
 }
 
-int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
+void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
 {
-	uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
-	tdb_trace_1rec(tdb, "tdb_chainunlock", key);
-	return tdb_unlock_list(tdb, h & ((1ULL << tdb->header.v.hash_bits)-1),
-			       F_WRLCK);
-}
+	if (tdb->file->allrecord_lock.count)
+		return;
 
-int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
-{
-	return chainlock(tdb, &key, F_RDLCK, TDB_LOCK_WAIT,
-			 "tdb_chainlock_read");
+	tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
 }
 
-int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
+enum TDB_ERROR tdb_lockall(struct tdb_context *tdb)
 {
-	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);
+	return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
 }
 
-/* record lock stops delete underneath */
-int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
+void tdb_unlockall(struct tdb_context *tdb)
 {
-	if (tdb->allrecord_lock.count) {
-		return 0;
-	}
-	return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
+	tdb_allrecord_unlock(tdb, F_WRLCK);
 }
 
-/*
-  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)
+enum TDB_ERROR tdb_lockall_read(struct tdb_context *tdb)
 {
-	if (tdb->allrecord_lock.count) {
-		return 0;
-	}
-	return tdb_brunlock(tdb, F_WRLCK, off, 1);
+	return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
 }
 
-/* fcntl locks don't stack: avoid unlocking someone else's */
-int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+void tdb_unlockall_read(struct tdb_context *tdb)
 {
-	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);
+	tdb_allrecord_unlock(tdb, F_RDLCK);
 }
 
-/* The transaction code uses this to remove all locks. */
-void tdb_release_transaction_locks(struct tdb_context *tdb)
+void tdb_lock_cleanup(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;
+	while (tdb->file->allrecord_lock.count
+	       && tdb->file->allrecord_lock.owner == tdb) {
+		tdb_allrecord_unlock(tdb, tdb->file->allrecord_lock.ltype);
 	}
 
-	for (i = 0; i<tdb->num_lockrecs; i++) {
-		struct tdb_lock_type *lck = &tdb->lockrecs[i];
-
-		tdb_brunlock(tdb, lck->ltype, lck->off, 1);
+	for (i=0; i<tdb->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].owner == tdb) {
+			tdb_nest_unlock(tdb,
+					tdb->file->lockrecs[i].off,
+					tdb->file->lockrecs[i].ltype);
+			i--;
+		}
 	}
-	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;
-	tdb->lockrecs = NULL;
-	tdb->allrecord_lock.count = 0;
 }