X-Git-Url: http://git.samba.org/samba.git/?a=blobdiff_plain;f=source3%2Flocking%2Fposix.c;h=606f2086732513a693b99259f51d116827e5ba6d;hb=5e54558c6dea67b56bbfaba5698f3a434d3dffb6;hp=6a621200ca0f504e9f883325ffb33f77c5a9377a;hpb=4a3f07456d0a04b84916a99af6ce1cbe65e0370c;p=vlendec%2Fsamba-autobuild%2F.git

diff --git a/source3/locking/posix.c b/source3/locking/posix.c
index 6a621200ca0..606f2086732 100644
--- a/source3/locking/posix.c
+++ b/source3/locking/posix.c
@@ -1,12 +1,11 @@
 /* 
-   Unix SMB/Netbios implementation.
-   Version 3.0
+   Unix SMB/CIFS implementation.
    Locking functions
-   Copyright (C) Jeremy Allison 1992-2000
+   Copyright (C) Jeremy Allison 1992-2006
    
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
-   the Free Software Foundation; either version 2 of the License, or
+   the Free Software Foundation; either version 3 of the License, or
    (at your option) any later version.
    
    This program is distributed in the hope that it will be useful,
@@ -15,8 +14,7 @@
    GNU General Public License for more details.
    
    You should have received a copy of the GNU General Public License
-   along with this program; if not, write to the Free Software
-   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
    Revision History:
 
@@ -24,13 +22,9 @@
 */
 
 #include "includes.h"
-extern int DEBUGLEVEL;
 
-/*
- * The POSIX locking database handle.
- */
-
-static TDB_CONTEXT *posix_lock_tdb;
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_LOCKING
 
 /*
  * The pending close database handle.
@@ -38,776 +32,693 @@ static TDB_CONTEXT *posix_lock_tdb;
 
 static TDB_CONTEXT *posix_pending_close_tdb;
 
-/*
- * The data in POSIX lock records is an unsorted linear array of these
- * records.  It is unnecessary to store the count as tdb provides the
- * size of the record.
- */
-
-struct posix_lock {
-	int fd;
-	SMB_OFF_T start;
-	SMB_OFF_T size;
-	int lock_type;
-};
-
-/*
- * The data in POSIX pending close records is an unsorted linear array of int
- * records.  It is unnecessary to store the count as tdb provides the
- * size of the record.
- */
-
-/* The key used in both the POSIX databases. */
-
-struct posix_lock_key {
-	SMB_DEV_T device;
-	SMB_INO_T inode;
-}; 
-
-/*******************************************************************
- Form a static locking key for a dev/inode pair.
-******************************************************************/
-
-static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
-{
-	static struct posix_lock_key key;
-	TDB_DATA kbuf;
-
-	memset(&key, '\0', sizeof(key));
-	key.device = dev;
-	key.inode = inode;
-	kbuf.dptr = (char *)&key;
-	kbuf.dsize = sizeof(key);
-	return kbuf;
-}
-
-/*******************************************************************
- Convenience function to get a key from an fsp.
-******************************************************************/
-
-static TDB_DATA locking_key_fsp(files_struct *fsp)
-{
-	return locking_key(fsp->dev, fsp->inode);
-}
+/****************************************************************************
+ First - the functions that deal with the underlying system locks - these
+ functions are used no matter if we're mapping CIFS Windows locks or CIFS
+ POSIX locks onto POSIX.
+****************************************************************************/
 
 /****************************************************************************
- Add an fd to the pending close tdb.
+ Utility function to map a lock type correctly depending on the open
+ mode of a file.
 ****************************************************************************/
 
-static BOOL add_fd_to_close_entry(files_struct *fsp)
+static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
 {
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	char *tp;
-
-	dbuf.dptr = NULL;
-
-	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
-
-	tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(int));
-	if (!tp) {
-		DEBUG(0,("add_fd_to_close_entry: Realloc fail !\n"));
-		if (dbuf.dptr)
-			free(dbuf.dptr);
-		return False;
-	} else
-		dbuf.dptr = tp;
-
-	memcpy(dbuf.dptr + dbuf.dsize, &fsp->fd, sizeof(int));
-	dbuf.dsize += sizeof(int);
-
-	if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
-		DEBUG(0,("add_fd_to_close_entry: tdb_store fail !\n"));
+	if((lock_type == WRITE_LOCK) && !fsp->can_write) {
+		/*
+		 * Many UNIX's cannot get a write lock on a file opened read-only.
+		 * Win32 locking semantics allow this.
+		 * Do the best we can and attempt a read-only lock.
+		 */
+		DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
+		return F_RDLCK;
 	}
 
-	free(dbuf.dptr);
-	return True;
+	/*
+	 * This return should be the most normal, as we attempt
+	 * to always open files read/write.
+	 */
+
+	return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
 }
 
 /****************************************************************************
- Remove all fd entries for a specific dev/inode pair from the tdb.
+ Debugging aid :-).
 ****************************************************************************/
 
-static void delete_close_entries(files_struct *fsp)
+static const char *posix_lock_type_name(int lock_type)
 {
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-
-	if (tdb_delete(posix_pending_close_tdb, kbuf) == -1)
-		DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
+	return (lock_type == F_RDLCK) ? "READ" : "WRITE";
 }
 
 /****************************************************************************
- Get the array of POSIX pending close records for an open fsp. Caller must
- free. Returns number of entries.
+ Check to see if the given unsigned lock range is within the possible POSIX
+ range. Modifies the given args to be in range if possible, just returns
+ False if not.
 ****************************************************************************/
 
-static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
+static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
+				SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
 {
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	size_t count = 0;
-
-	*entries = NULL;
-	dbuf.dptr = NULL;
-
-	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
-
-    if (!dbuf.dptr) {
-		return 0;
-	}
-
-	*entries = (int *)dbuf.dptr;
-	count = (size_t)(dbuf.dsize / sizeof(int));
+	SMB_OFF_T offset = (SMB_OFF_T)u_offset;
+	SMB_OFF_T count = (SMB_OFF_T)u_count;
 
-	return count;
-}
+	/*
+	 * For the type of system we are, attempt to
+	 * find the maximum positive lock offset as an SMB_OFF_T.
+	 */
 
-/****************************************************************************
- Get the array of POSIX locks for an fsp. Caller must free. Returns
- number of entries.
-****************************************************************************/
+#if defined(MAX_POSITIVE_LOCK_OFFSET) /* Some systems have arbitrary limits. */
 
-static size_t get_posix_lock_entries(files_struct *fsp, struct posix_lock **entries)
-{
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	size_t count = 0;
+	SMB_OFF_T max_positive_lock_offset = (MAX_POSITIVE_LOCK_OFFSET);
 
-	*entries = NULL;
+#elif defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
 
-	dbuf.dptr = NULL;
+	/*
+	 * In this case SMB_OFF_T is 64 bits,
+	 * and the underlying system can handle 64 bit signed locks.
+	 */
 
-	dbuf = tdb_fetch(posix_lock_tdb, kbuf);
+	SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
+	SMB_OFF_T mask = (mask2<<1);
+	SMB_OFF_T max_positive_lock_offset = ~mask;
 
-    if (!dbuf.dptr) {
-		return 0;
-	}
+#else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
 
-	*entries = (struct posix_lock *)dbuf.dptr;
-	count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
+	/*
+	 * In this case either SMB_OFF_T is 32 bits,
+	 * or the underlying system cannot handle 64 bit signed locks.
+	 * All offsets & counts must be 2^31 or less.
+	 */
 
-	return count;
-}
+	SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
 
-/****************************************************************************
- Deal with pending closes needed by POSIX locking support.
- Note that posix_locking_close_file() is expected to have been called
- to delete all locks on this fsp before this function is called.
-****************************************************************************/
+#endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
 
-int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
-{
-	int saved_errno = 0;
-	int ret;
-	size_t count, i;
-	struct posix_lock *entries = NULL;
-	int *fd_array = NULL;
-	BOOL locks_on_other_fds = False;
+	/*
+	 * POSIX locks of length zero mean lock to end-of-file.
+	 * Win32 locks of length zero are point probes. Ignore
+	 * any Win32 locks of length zero. JRA.
+	 */
 
-	if (!lp_posix_locking(SNUM(conn))) {
-		/*
-		 * No POSIX to worry about, just close.
-		 */
-		ret = conn->vfs_ops.close(fsp,fsp->fd);
-		fsp->fd = -1;
-		return ret;
+	if (count == (SMB_OFF_T)0) {
+		DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n"));
+		return False;
 	}
 
 	/*
-	 * Get the number of outstanding POSIX locks on this dev/inode pair.
+	 * If the given offset was > max_positive_lock_offset then we cannot map this at all
+	 * ignore this lock.
 	 */
 
-	count = get_posix_lock_entries(fsp, &entries);
+	if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
+		DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
+				(double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
+		return False;
+	}
 
 	/*
-	 * Check if there are any outstanding locks belonging to
-	 * other fd's. This should never be the case if posix_locking_close_file()
-	 * has been called first, but it never hurts to be *sure*.
+	 * We must truncate the count to less than max_positive_lock_offset.
 	 */
 
-	for (i = 0; i < count; i++) {
-		if (entries[i].fd != fsp->fd) {
-			locks_on_other_fds = True;
-			break;
-		}
+	if (u_count & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
+		count = max_positive_lock_offset;
 	}
 
-	if (locks_on_other_fds) {
+	/*
+	 * Truncate count to end at max lock offset.
+	 */
 
-		/*
-		 * There are outstanding locks on this dev/inode pair on other fds.
-		 * Add our fd to the pending close tdb and set fsp->fd to -1.
-		 */
+	if (offset + count < 0 || offset + count > max_positive_lock_offset) {
+		count = max_positive_lock_offset - offset;
+	}
 
-		if (!add_fd_to_close_entry(fsp)) {
-			free((char *)entries);
-			return False;
-		}
+	/*
+	 * If we ate all the count, ignore this lock.
+	 */
 
-		free((char *)entries);
-		fsp->fd = -1;
-		return 0;
+	if (count == 0) {
+		DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
+				(double)u_offset, (double)u_count ));
+		return False;
 	}
 
-	if(entries)
-		free((char *)entries);
-
 	/*
-	 * No outstanding POSIX locks. Get the pending close fd's
-	 * from the tdb and close them all.
+	 * The mapping was successful.
 	 */
 
-	count = get_posix_pending_close_entries(fsp, &fd_array);
+	DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
+			(double)offset, (double)count ));
 
-	if (count) {
-		DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
+	*offset_out = offset;
+	*count_out = count;
+	
+	return True;
+}
 
-		for(i = 0; i < count; i++) {
-			if (conn->vfs_ops.close(fsp,fd_array[i]) == -1) {
-				saved_errno = errno;
-			}
-		}
+/****************************************************************************
+ Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
+ broken NFS implementations.
+****************************************************************************/
 
-		/*
-		 * Delete all fd's stored in the tdb
-		 * for this dev/inode pair.
-		 */
+static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
+{
+	BOOL ret;
 
-		delete_close_entries(fsp);
-	}
+	DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fh->fd,op,(double)offset,(double)count,type));
 
-	if (fd_array)
-		free((char *)fd_array);
+	ret = SMB_VFS_LOCK(fsp,fsp->fh->fd,op,offset,count,type);
 
-	/*
-	 * Finally close the fd associated with this fsp.
-	 */
+	if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno ==  EINVAL))) {
 
-	ret = conn->vfs_ops.close(fsp,fsp->fd);
+		DEBUG(0,("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n",
+					(double)offset,(double)count));
+		DEBUGADD(0,("an %s error. This can happen when using 64 bit lock offsets\n", strerror(errno)));
+		DEBUGADD(0,("on 32 bit NFS mounted file systems.\n"));
 
-	if (saved_errno != 0) {
-        errno = saved_errno;
-		ret = -1;
-    } 
+		/*
+		 * If the offset is > 0x7FFFFFFF then this will cause problems on
+		 * 32 bit NFS mounted filesystems. Just ignore it.
+		 */
 
-	fsp->fd = -1;
+		if (offset & ~((SMB_OFF_T)0x7fffffff)) {
+			DEBUG(0,("Offset greater than 31 bits. Returning success.\n"));
+			return True;
+		}
+
+		if (count & ~((SMB_OFF_T)0x7fffffff)) {
+			/* 32 bit NFS file system, retry with smaller offset */
+			DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n"));
+			errno = 0;
+			count &= 0x7fffffff;
+			ret = SMB_VFS_LOCK(fsp,fsp->fh->fd,op,offset,count,type);
+		}
+	}
 
+	DEBUG(8,("posix_fcntl_lock: Lock call %s\n", ret ? "successful" : "failed"));
 	return ret;
 }
 
 /****************************************************************************
- Debugging aid :-).
+ Actual function that gets POSIX locks. Copes with 64 -> 32 bit cruft and
+ broken NFS implementations.
 ****************************************************************************/
 
-static const char *posix_lock_type_name(int lock_type)
+static BOOL posix_fcntl_getlock(files_struct *fsp, SMB_OFF_T *poffset, SMB_OFF_T *pcount, int *ptype)
 {
-	return (lock_type == F_RDLCK) ? "READ" : "WRITE";
-}
+	pid_t pid;
+	BOOL ret;
 
-/****************************************************************************
- Delete a POSIX lock entry by index number. Used if the tdb add succeeds, but
- then the POSIX fcntl lock fails.
-****************************************************************************/
+	DEBUG(8,("posix_fcntl_getlock %d %.0f %.0f %d\n",
+		fsp->fh->fd,(double)*poffset,(double)*pcount,*ptype));
 
-static BOOL delete_posix_lock_entry_by_index(files_struct *fsp, size_t entry)
-{
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	struct posix_lock *locks;
-	size_t count;
+	ret = SMB_VFS_GETLOCK(fsp,fsp->fh->fd,poffset,pcount,ptype,&pid);
 
-	dbuf.dptr = NULL;
-	
-	dbuf = tdb_fetch(posix_lock_tdb, kbuf);
+	if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno ==  EINVAL))) {
 
-	if (!dbuf.dptr) {
-		DEBUG(10,("delete_posix_lock_entry_by_index: tdb_fetch failed !\n"));
-		goto fail;
-	}
+		DEBUG(0,("posix_fcntl_getlock: WARNING: lock request at offset %.0f, length %.0f returned\n",
+					(double)*poffset,(double)*pcount));
+		DEBUGADD(0,("an %s error. This can happen when using 64 bit lock offsets\n", strerror(errno)));
+		DEBUGADD(0,("on 32 bit NFS mounted file systems.\n"));
 
-	count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
-	locks = (struct posix_lock *)dbuf.dptr;
+		/*
+		 * If the offset is > 0x7FFFFFFF then this will cause problems on
+		 * 32 bit NFS mounted filesystems. Just ignore it.
+		 */
 
-	if (count == 1) {
-		tdb_delete(posix_lock_tdb, kbuf);
-	} else {
-		if (entry < count-1) {
-			memmove(&locks[entry], &locks[entry+1], sizeof(*locks)*((count-1) - entry));
+		if (*poffset & ~((SMB_OFF_T)0x7fffffff)) {
+			DEBUG(0,("Offset greater than 31 bits. Returning success.\n"));
+			return True;
 		}
-		dbuf.dsize -= sizeof(*locks);
-		tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
-	}
-
-	free(dbuf.dptr);
 
-	return True;
+		if (*pcount & ~((SMB_OFF_T)0x7fffffff)) {
+			/* 32 bit NFS file system, retry with smaller offset */
+			DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n"));
+			errno = 0;
+			*pcount &= 0x7fffffff;
+			ret = SMB_VFS_GETLOCK(fsp,fsp->fh->fd,poffset,pcount,ptype,&pid);
+		}
+	}
 
- fail:
-    if (dbuf.dptr)
-		free(dbuf.dptr);
-    return False;
+	DEBUG(8,("posix_fcntl_getlock: Lock query call %s\n", ret ? "successful" : "failed"));
+	return ret;
 }
 
 /****************************************************************************
- Add an entry into the POSIX locking tdb. We return the index number of the
- added lock (used in case we need to delete *exactly* this entry). Returns
- False on fail, True on success.
+ POSIX function to see if a file region is locked. Returns True if the
+ region is locked, False otherwise.
 ****************************************************************************/
 
-static BOOL add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *pentry_num)
+BOOL is_posix_locked(files_struct *fsp,
+			SMB_BIG_UINT *pu_offset,
+			SMB_BIG_UINT *pu_count,
+			enum brl_type *plock_type,
+			enum brl_flavour lock_flav)
 {
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	struct posix_lock pl;
-	char *tp;
-
-	dbuf.dptr = NULL;
-
-	dbuf = tdb_fetch(posix_lock_tdb, kbuf);
+	SMB_OFF_T offset;
+	SMB_OFF_T count;
+	int posix_lock_type = map_posix_lock_type(fsp,*plock_type);
 
-	*pentry_num = (size_t)(dbuf.dsize / sizeof(pl));
+	DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
+		fsp->fsp_name, (double)*pu_offset, (double)*pu_count, posix_lock_type_name(*plock_type) ));
 
 	/*
-	 * Add new record.
+	 * If the requested lock won't fit in the POSIX range, we will
+	 * never set it, so presume it is not locked.
 	 */
 
-	pl.fd = fsp->fd;
-	pl.start = start;
-	pl.size = size;
-	pl.lock_type = lock_type;
+	if(!posix_lock_in_range(&offset, &count, *pu_offset, *pu_count)) {
+		return False;
+	}
 
-	tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(pl));
-	if (!tp) {
-		DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
-		goto fail;
-	} else
-		dbuf.dptr = tp;
+	if (!posix_fcntl_getlock(fsp,&offset,&count,&posix_lock_type)) {
+		return False;
+	}
 
-	memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
-	dbuf.dsize += sizeof(pl);
+	if (posix_lock_type == F_UNLCK) {
+		return False;
+	}
 
-	if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
-		DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
-		goto fail;
+	if (lock_flav == POSIX_LOCK) {
+		/* Only POSIX lock queries need to know the details. */
+		*pu_offset = (SMB_BIG_UINT)offset;
+		*pu_count = (SMB_BIG_UINT)count;
+		*plock_type = (posix_lock_type == F_RDLCK) ? READ_LOCK : WRITE_LOCK;
 	}
+	return True;
+}
 
-    free(dbuf.dptr);
+/****************************************************************************
+ Next - the functions that deal with in memory database storing representations
+ of either Windows CIFS locks or POSIX CIFS locks.
+****************************************************************************/
 
-	DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: dev=%.0f inode=%.0f\n",
-			fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size,
-			(double)fsp->dev, (double)fsp->inode ));
+/* The key used in the in-memory POSIX databases. */
 
-    return True;
+struct lock_ref_count_key {
+	struct file_id id;
+	char r;
+}; 
+
+/*******************************************************************
+ Form a static locking key for a dev/inode pair for the fd array.
+******************************************************************/
 
- fail:
-    if (dbuf.dptr)
-		free(dbuf.dptr);
-    return False;
+static TDB_DATA fd_array_key(struct file_id id)
+{
+	static struct file_id key;
+	TDB_DATA kbuf;
+	key = id;
+	kbuf.dptr = (uint8 *)&key;
+	kbuf.dsize = sizeof(key);
+	return kbuf;
 }
 
-/****************************************************************************
- Calculate if locks have any overlap at all.
-****************************************************************************/
+/*******************************************************************
+ Form a static locking key for a dev/inode pair for the lock ref count
+******************************************************************/
 
-static BOOL does_lock_overlap(SMB_OFF_T start1, SMB_OFF_T size1, SMB_OFF_T start2, SMB_OFF_T size2)
+static TDB_DATA locking_ref_count_key(struct file_id id)
 {
-	if (start1 >= start2 && start1 <= start2 + size2)
-		return True;
+	static struct lock_ref_count_key key;
+	TDB_DATA kbuf;
 
-	if (start1 < start2 && start1 + size1 > start2)
-		return True;
+	memset(&key, '\0', sizeof(key));
+	key.id = id;
+	key.r = 'r';
+	kbuf.dptr = (uint8 *)&key;
+	kbuf.dsize = sizeof(key);
+	return kbuf;
+}
+
+/*******************************************************************
+ Convenience function to get an fd_array key from an fsp.
+******************************************************************/
 
-	return False;
+static TDB_DATA fd_array_key_fsp(files_struct *fsp)
+{
+	return fd_array_key(fsp->file_id);
 }
 
-/****************************************************************************
- Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
- deleted and the number of records that are overlapped by this one, or -1 on error.
-****************************************************************************/
+/*******************************************************************
+ Convenience function to get a lock ref count key from an fsp.
+******************************************************************/
 
-static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
+static TDB_DATA locking_ref_count_key_fsp(files_struct *fsp)
 {
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	struct posix_lock *locks;
-	size_t i, count;
-	BOOL found = False;
-	int num_overlapping_records = 0;
+	return locking_ref_count_key(fsp->file_id);
+}
 
-	dbuf.dptr = NULL;
-	
-	dbuf = tdb_fetch(posix_lock_tdb, kbuf);
+/*******************************************************************
+ Create the in-memory POSIX lock databases.
+********************************************************************/
 
-	if (!dbuf.dptr) {
-		DEBUG(10,("delete_posix_lock_entry: tdb_fetch failed !\n"));
-		goto fail;
+BOOL posix_locking_init(int read_only)
+{
+	if (posix_pending_close_tdb) {
+		return True;
+	}
+	
+	if (!posix_pending_close_tdb) {
+		posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
+						   read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
+	}
+	if (!posix_pending_close_tdb) {
+		DEBUG(0,("Failed to open POSIX pending close database.\n"));
+		return False;
 	}
 
-	/* There are existing locks - find a match. */
-	locks = (struct posix_lock *)dbuf.dptr;
-	count = (size_t)(dbuf.dsize / sizeof(*locks));
+	return True;
+}
 
-	/*
-	 * Search for and delete the first record that matches the
-	 * unlock criteria.
-	 */
+/*******************************************************************
+ Delete the in-memory POSIX lock databases.
+********************************************************************/
+
+BOOL posix_locking_end(void)
+{
+	if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0) {
+		return False;
+	}
+	return True;
+}
 
-	for (i=0; i<count; i++) { 
-		struct posix_lock *entry = &locks[i];
+/****************************************************************************
+ Next - the functions that deal with storing fd's that have outstanding
+ POSIX locks when closed.
+****************************************************************************/
 
-		if (entry->fd == fsp->fd &&
-			entry->start == start &&
-			entry->size == size) {
+/****************************************************************************
+ The records in posix_pending_close_tdb are composed of an array of ints
+ keyed by dev/ino pair.
+ The first int is a reference count of the number of outstanding locks on
+ all open fd's on this dev/ino pair. Any subsequent ints are the fd's that
+ were open on this dev/ino pair that should have been closed, but can't as
+ the lock ref count is non zero.
+****************************************************************************/
+
+/****************************************************************************
+ Keep a reference count of the number of Windows locks open on this dev/ino
+ pair. Creates entry if it doesn't exist.
+****************************************************************************/
 
-			/* Make a copy if requested. */
-			if (pl)
-				*pl = *entry;
+static void increment_windows_lock_ref_count(files_struct *fsp)
+{
+	TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
+	TDB_DATA dbuf;
+	int lock_ref_count;
 
-			/* Found it - delete it. */
-			if (count == 1) {
-				tdb_delete(posix_lock_tdb, kbuf);
-			} else {
-				if (i < count-1) {
-					memmove(&locks[i], &locks[i+1], sizeof(*locks)*((count-1) - i));
-				}
-				dbuf.dsize -= sizeof(*locks);
-				tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
-			}
-			count--;
-			found = True;
-			break;
+	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
+	if (dbuf.dptr == NULL) {
+		dbuf.dptr = (uint8 *)SMB_MALLOC_P(int);
+		if (!dbuf.dptr) {
+			smb_panic("increment_windows_lock_ref_count: malloc fail");
 		}
+		memset(dbuf.dptr, '\0', sizeof(int));
+		dbuf.dsize = sizeof(int);
 	}
 
-	if (!found)
-		goto fail;
+	memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
+	lock_ref_count++;
+	memcpy(dbuf.dptr, &lock_ref_count, sizeof(int));
+	
+	if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
+		smb_panic("increment_windows_lock_ref_count: tdb_store_fail");
+	}
+	SAFE_FREE(dbuf.dptr);
 
-	/*
-	 * Count the number of entries that are
-	 * overlapped by this unlock request.
-	 */
+	DEBUG(10,("increment_windows_lock_ref_count for file now %s = %d\n",
+		fsp->fsp_name, lock_ref_count ));
+}
 
-	for (i = 0; i < count; i++) {
-		struct posix_lock *entry = &locks[i];
+static void decrement_windows_lock_ref_count(files_struct *fsp)
+{
+	TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
+	TDB_DATA dbuf;
+	int lock_ref_count;
 
-		if (fsp->fd == entry->fd &&
-			does_lock_overlap( start, size, entry->start, entry->size))
-				num_overlapping_records++;
+	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
+	if (!dbuf.dptr) {
+		smb_panic("decrement_windows_lock_ref_count: logic error");
 	}
 
-	DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
-			posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
-				(unsigned int)num_overlapping_records ));
+	memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
+	lock_ref_count--;
+	memcpy(dbuf.dptr, &lock_ref_count, sizeof(int));
 
-    if (dbuf.dptr)
-		free(dbuf.dptr);
+	if (lock_ref_count < 0) {
+		smb_panic("decrement_windows_lock_ref_count: lock_count logic error");
+	}
 
-	return num_overlapping_records;
+	if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
+		smb_panic("decrement_windows_lock_ref_count: tdb_store_fail");
+	}
+	SAFE_FREE(dbuf.dptr);
 
- fail:
-    if (dbuf.dptr)
-		free(dbuf.dptr);
-    return -1;
+	DEBUG(10,("decrement_windows_lock_ref_count for file now %s = %d\n",
+		fsp->fsp_name, lock_ref_count ));
 }
 
 /****************************************************************************
- Utility function to map a lock type correctly depending on the open
- mode of a file.
+ Bulk delete - subtract as many locks as we've just deleted.
 ****************************************************************************/
 
-static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
+void reduce_windows_lock_ref_count(files_struct *fsp, unsigned int dcount)
 {
-	if((lock_type == WRITE_LOCK) && !fsp->can_write) {
-		/*
-		 * Many UNIX's cannot get a write lock on a file opened read-only.
-		 * Win32 locking semantics allow this.
-		 * Do the best we can and attempt a read-only lock.
-		 */
-		DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
-		return F_RDLCK;
-	} else if((lock_type == READ_LOCK) && !fsp->can_read) {
-		/*
-		 * Ditto for read locks on write only files.
-		 */
-		DEBUG(10,("map_posix_lock_type: Changing read lock to write due to write-only file.\n"));
-		return F_WRLCK;
+	TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
+	TDB_DATA dbuf;
+	int lock_ref_count;
+
+	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
+	if (!dbuf.dptr) {
+		return;
 	}
 
-  /*
-   * This return should be the most normal, as we attempt
-   * to always open files read/write.
-   */
+	memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
+	lock_ref_count -= dcount;
+
+	if (lock_ref_count < 0) {
+		smb_panic("reduce_windows_lock_ref_count: lock_count logic error");
+	}
+	memcpy(dbuf.dptr, &lock_ref_count, sizeof(int));
+	
+	if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
+		smb_panic("reduce_windows_lock_ref_count: tdb_store_fail");
+	}
+	SAFE_FREE(dbuf.dptr);
 
-  return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
+	DEBUG(10,("reduce_windows_lock_ref_count for file now %s = %d\n",
+		fsp->fsp_name, lock_ref_count ));
 }
 
 /****************************************************************************
- Check to see if the given unsigned lock range is within the possible POSIX
- range. Modifies the given args to be in range if possible, just returns
- False if not.
+ Fetch the lock ref count.
 ****************************************************************************/
 
-static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
-								SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
+static int get_windows_lock_ref_count(files_struct *fsp)
 {
-	SMB_OFF_T offset = (SMB_OFF_T)u_offset;
-	SMB_OFF_T count = (SMB_OFF_T)u_count;
-
-	/*
-	 * For the type of system we are, attempt to
-	 * find the maximum positive lock offset as an SMB_OFF_T.
-	 */
-
-#if defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
-
-	/*
-	 * In this case SMB_OFF_T is 64 bits,
-	 * and the underlying system can handle 64 bit signed locks.
-	 */
-
-    SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
-    SMB_OFF_T mask = (mask2<<1);
-    SMB_OFF_T max_positive_lock_offset = ~mask;
-
-#else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
-
-	/*
-	 * In this case either SMB_OFF_T is 32 bits,
-	 * or the underlying system cannot handle 64 bit signed locks.
-	 * All offsets & counts must be 2^31 or less.
-	 */
-
-    SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
-
-#endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
-
-	/*
-	 * POSIX locks of length zero mean lock to end-of-file.
-	 * Win32 locks of length zero are point probes. Ignore
-	 * any Win32 locks of length zero. JRA.
-	 */
-
-	if (count == (SMB_OFF_T)0) {
-		DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n"));
-		return False;
-	}
-
-	/*
-	 * If the given offset was > max_positive_lock_offset then we cannot map this at all
-	 * ignore this lock.
-	 */
+	TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
+	TDB_DATA dbuf;
+	int lock_ref_count;
 
-	if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
-		DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
-				(double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
-		return False;
+	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
+	if (!dbuf.dptr) {
+		lock_ref_count = 0;
+	} else {
+		memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
 	}
+	SAFE_FREE(dbuf.dptr);
 
-	/*
-	 * We must truncate the offset and count to less than max_positive_lock_offset.
-	 */
+	DEBUG(10,("get_windows_lock_count for file %s = %d\n",
+		fsp->fsp_name, lock_ref_count ));
+	return lock_ref_count;
+}
 
-	offset &= max_positive_lock_offset;
-	count &= max_positive_lock_offset;
+/****************************************************************************
+ Delete a lock_ref_count entry.
+****************************************************************************/
 
+static void delete_windows_lock_ref_count(files_struct *fsp)
+{
+	TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
 
-	/*
-	 * Deal with a very common case of count of all ones.
-	 * (lock entire file).
-	 */
+	/* Not a bug if it doesn't exist - no locks were ever granted. */
+	tdb_delete(posix_pending_close_tdb, kbuf);
+	DEBUG(10,("delete_windows_lock_ref_count for file %s\n", fsp->fsp_name));
+}
 
-	if(count == (SMB_OFF_T)-1)
-		count = max_positive_lock_offset;
+/****************************************************************************
+ Add an fd to the pending close tdb.
+****************************************************************************/
 
-	/*
-	 * Truncate count to end at max lock offset.
-	 */
+static void add_fd_to_close_entry(files_struct *fsp)
+{
+	TDB_DATA kbuf = fd_array_key_fsp(fsp);
+	TDB_DATA dbuf;
 
-	if (offset + count < 0 || offset + count > max_positive_lock_offset)
-		count = max_positive_lock_offset - offset;
+	dbuf.dptr = NULL;
+	dbuf.dsize = 0;
 
-	/*
-	 * If we ate all the count, ignore this lock.
-	 */
+	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
 
-	if (count == 0) {
-		DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
-				(double)u_offset, (double)u_count ));
-		return False;
+	dbuf.dptr = (uint8 *)SMB_REALLOC(dbuf.dptr, dbuf.dsize + sizeof(int));
+	if (!dbuf.dptr) {
+		smb_panic("add_fd_to_close_entry: SMB_REALLOC failed");
 	}
 
-	/*
-	 * The mapping was successful.
-	 */
+	memcpy(dbuf.dptr + dbuf.dsize, &fsp->fh->fd, sizeof(int));
+	dbuf.dsize += sizeof(int);
+
+	if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
+		smb_panic("add_fd_to_close_entry: tdb_store_fail");
+	}
 
-	DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
-			(double)offset, (double)count ));
+	DEBUG(10,("add_fd_to_close_entry: added fd %d file %s\n",
+		fsp->fh->fd, fsp->fsp_name ));
 
-	*offset_out = offset;
-	*count_out = count;
-	
-	return True;
+	SAFE_FREE(dbuf.dptr);
 }
 
 /****************************************************************************
- Pathetically try and map a 64 bit lock offset into 31 bits. I hate Windows :-).
+ Remove all fd entries for a specific dev/inode pair from the tdb.
 ****************************************************************************/
 
-uint32 map_lock_offset(uint32 high, uint32 low)
+static void delete_close_entries(files_struct *fsp)
 {
-	unsigned int i;
-	uint32 mask = 0;
-	uint32 highcopy = high;
+	TDB_DATA kbuf = fd_array_key_fsp(fsp);
 
-	/*
-	 * Try and find out how many significant bits there are in high.
-	 */
+	if (tdb_delete(posix_pending_close_tdb, kbuf) == -1) {
+		smb_panic("delete_close_entries: tdb_delete failed");
+	}
+}
+
+/****************************************************************************
+ Get the array of POSIX pending close records for an open fsp. Caller must
+ free. Returns number of entries.
+****************************************************************************/
 
-	for(i = 0; highcopy; i++)
-		highcopy >>= 1;
+static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
+{
+	TDB_DATA kbuf = fd_array_key_fsp(fsp);
+	TDB_DATA dbuf;
+	size_t count = 0;
 
-	/*
-	 * We use 31 bits not 32 here as POSIX
-	 * lock offsets may not be negative.
-	 */
+	*entries = NULL;
+	dbuf.dptr = NULL;
 
-	mask = (~0) << (31 - i);
+	dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
 
-	if(low & mask)
-		return 0; /* Fail. */
+	if (!dbuf.dptr) {
+		return 0;
+	}
 
-	high <<= (31 - i);
+	*entries = (int *)dbuf.dptr;
+	count = (size_t)(dbuf.dsize / sizeof(int));
 
-	return (high|low);
+	return count;
 }
 
 /****************************************************************************
- Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
- broken NFS implementations.
+ Deal with pending closes needed by POSIX locking support.
+ Note that posix_locking_close_file() is expected to have been called
+ to delete all locks on this fsp before this function is called.
 ****************************************************************************/
 
-static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
+NTSTATUS fd_close_posix(struct connection_struct *conn, files_struct *fsp)
 {
+	int saved_errno = 0;
 	int ret;
-	struct connection_struct *conn = fsp->conn;
+	int *fd_array = NULL;
+	size_t count, i;
 
-#if defined(LARGE_SMB_OFF_T)
-	/*
-	 * In the 64 bit locking case we store the original
-	 * values in case we have to map to a 32 bit lock on
-	 * a filesystem that doesn't support 64 bit locks.
-	 */
-	SMB_OFF_T orig_offset = offset;
-	SMB_OFF_T orig_count = count;
-#endif /* LARGE_SMB_OFF_T */
+	if (!lp_locking(fsp->conn->params) || !lp_posix_locking(conn->params)) {
+		/*
+		 * No locking or POSIX to worry about or we want POSIX semantics
+		 * which will lose all locks on all fd's open on this dev/inode,
+		 * just close.
+		 */
+		ret = SMB_VFS_CLOSE(fsp,fsp->fh->fd);
+		fsp->fh->fd = -1;
+		return map_nt_error_from_unix(errno);
+	}
 
-	DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
+	if (get_windows_lock_ref_count(fsp)) {
 
-	ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
+		/*
+		 * There are outstanding locks on this dev/inode pair on other fds.
+		 * Add our fd to the pending close tdb and set fsp->fh->fd to -1.
+		 */
 
-	if (!ret && (errno == EFBIG)) {
-		if( DEBUGLVL( 0 )) {
-			dbgtext("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n", (double)offset,(double)count);
-			dbgtext("a 'file too large' error. This can happen when using 64 bit lock offsets\n");
-			dbgtext("on 32 bit NFS mounted file systems. Retrying with 32 bit truncated length.\n");
-		}
-		/* 32 bit NFS file system, retry with smaller offset */
-		errno = 0;
-		count &= 0x7fffffff;
-		ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
+		add_fd_to_close_entry(fsp);
+		fsp->fh->fd = -1;
+		return NT_STATUS_OK;
 	}
 
-	/* A lock query - just return. */
-	if (op == SMB_F_GETLK)
-		return ret;
+	/*
+	 * No outstanding locks. Get the pending close fd's
+	 * from the tdb and close them all.
+	 */
 
-	/* A lock set or unset. */
-	if (!ret) {
-		DEBUG(3,("posix_fcntl_lock: lock failed at offset %.0f count %.0f op %d type %d (%s)\n",
-				(double)offset,(double)count,op,type,strerror(errno)));
+	count = get_posix_pending_close_entries(fsp, &fd_array);
 
-		/* Perhaps it doesn't support this sort of locking ? */
-		if (errno == EINVAL) {
-#if defined(LARGE_SMB_OFF_T)
-			{
-				/*
-				 * Ok - if we get here then we have a 64 bit lock request
-				 * that has returned EINVAL. Try and map to 31 bits for offset
-				 * and length and try again. This may happen if a filesystem
-				 * doesn't support 64 bit offsets (efs/ufs) although the underlying
-				 * OS does.
-				 */
-				uint32 off_low = (orig_offset & 0xFFFFFFFF);
-				uint32 off_high = ((orig_offset >> 32) & 0xFFFFFFFF);
-
-				count = (orig_count & 0x7FFFFFFF);
-				offset = (SMB_OFF_T)map_lock_offset(off_high, off_low);
-				ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
-				if (!ret) {
-					if (errno == EINVAL) {
-						DEBUG(3,("posix_fcntl_lock: locking not supported? returning True\n"));
-						return(True);
-					}
-					return False;
-				}
-				DEBUG(3,("posix_fcntl_lock: 64 -> 32 bit modified lock call successful\n"));
-				return True;
+	if (count) {
+		DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
+
+		for(i = 0; i < count; i++) {
+			if (SMB_VFS_CLOSE(fsp,fd_array[i]) == -1) {
+				saved_errno = errno;
 			}
-#else /* LARGE_SMB_OFF_T */
-			DEBUG(3,("locking not supported? returning True\n"));
-			return(True);
-#endif /* LARGE_SMB_OFF_T */
 		}
 
-		return(False);
-	}
-
-	DEBUG(8,("posix_fcntl_lock: Lock call successful\n"));
-
-	return(True);
-}
+		/*
+		 * Delete all fd's stored in the tdb
+		 * for this dev/inode pair.
+		 */
 
-/****************************************************************************
- POSIX function to see if a file region is locked. Returns True if the
- region is locked, False otherwise.
-****************************************************************************/
+		delete_close_entries(fsp);
+	}
 
-BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
-{
-	SMB_OFF_T offset;
-	SMB_OFF_T count;
-	int posix_lock_type = map_posix_lock_type(fsp,lock_type);
+	SAFE_FREE(fd_array);
 
-	DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
-			fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
+	/* Don't need a lock ref count on this dev/ino anymore. */
+	delete_windows_lock_ref_count(fsp);
 
 	/*
-	 * If the requested lock won't fit in the POSIX range, we will
-	 * never set it, so presume it is not locked.
+	 * Finally close the fd associated with this fsp.
 	 */
 
-	if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
-		return False;
+	ret = SMB_VFS_CLOSE(fsp,fsp->fh->fd);
 
-	/*
-	 * Note that most UNIX's can *test* for a write lock on
-	 * a read-only fd, just not *set* a write lock on a read-only
-	 * fd. So we don't need to use map_lock_type here.
-	 */ 
+	if (ret == 0 && saved_errno != 0) {
+		errno = saved_errno;
+		ret = -1;
+	} 
+
+	fsp->fh->fd = -1;
+
+	if (ret == -1) {
+		return map_nt_error_from_unix(errno);
+	}
 
-	return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
+	return NT_STATUS_OK;
 }
 
+/****************************************************************************
+ Next - the functions that deal with the mapping CIFS Windows locks onto
+ the underlying system POSIX locks.
+****************************************************************************/
+
 /*
  * Structure used when splitting a lock range
  * into a POSIX lock range. Doubly linked list.
  */
 
 struct lock_list {
-    struct lock_list *next;
-    struct lock_list *prev;
-    SMB_OFF_T start;
-    SMB_OFF_T size;
+	struct lock_list *next;
+	struct lock_list *prev;
+	SMB_OFF_T start;
+	SMB_OFF_T size;
 };
 
 /****************************************************************************
@@ -816,22 +727,14 @@ struct lock_list {
  understand it :-).
 ****************************************************************************/
 
-static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
+static struct lock_list *posix_lock_list(TALLOC_CTX *ctx,
+						struct lock_list *lhead,
+						const struct lock_context *lock_ctx, /* Lock context lhead belongs to. */
+						files_struct *fsp,
+						const struct lock_struct *plocks,
+						int num_locks)
 {
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-	TDB_DATA dbuf;
-	struct posix_lock *locks;
-	size_t num_locks, i;
-
-	dbuf.dptr = NULL;
-
-	dbuf = tdb_fetch(posix_lock_tdb, kbuf);
-
-	if (!dbuf.dptr)
-		return lhead;
-	
-	locks = (struct posix_lock *)dbuf.dptr;
-	num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
+	int i;
 
 	/*
 	 * Check the current lock list on this dev/inode pair.
@@ -842,10 +745,19 @@ static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhea
 		(double)lhead->start, (double)lhead->size ));
 
 	for (i=0; i<num_locks && lhead; i++) {
-
-		struct posix_lock *lock = &locks[i];
+		const struct lock_struct *lock = &plocks[i];
 		struct lock_list *l_curr;
 
+		/* Ignore all but read/write locks. */
+		if (lock->lock_type != READ_LOCK && lock->lock_type != WRITE_LOCK) {
+			continue;
+		}
+
+		/* Ignore locks not owned by this process. */
+		if (!procid_equal(&lock->context.pid, &lock_ctx->pid)) {
+			continue;
+		}
+
 		/*
 		 * Walk the lock list, checking for overlaps. Note that
 		 * the lock list can expand within this loop if the current
@@ -854,13 +766,13 @@ static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhea
 
 		for (l_curr = lhead; l_curr;) {
 
-			DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
+			DEBUG(10,("posix_lock_list: lock: fnum=%d: start=%.0f,size=%.0f:type=%s", lock->fnum,
 				(double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
 
 			if ( (l_curr->start >= (lock->start + lock->size)) ||
 				 (lock->start >= (l_curr->start + l_curr->size))) {
 
-				/* No overlap with this lock - leave this range alone. */
+				/* No overlap with existing lock - leave this range alone. */
 /*********************************************
                                              +---------+
                                              | l_curr  |
@@ -874,7 +786,7 @@ OR....
              +---------+
 **********************************************/
 
-				DEBUG(10,("no overlap case.\n" ));
+				DEBUG(10,(" no overlap case.\n" ));
 
 				l_curr = l_curr->next;
 
@@ -882,8 +794,8 @@ OR....
 						(l_curr->start + l_curr->size <= lock->start + lock->size) ) {
 
 				/*
-				 * This unlock is completely overlapped by this existing lock range
-				 * and thus should have no effect (not be unlocked). Delete it from the list.
+				 * This range is completely overlapped by this existing lock range
+				 * and thus should have no effect. Delete it from the list.
 				 */
 /*********************************************
                 +---------+
@@ -896,11 +808,12 @@ OR....
 				/* Save the next pointer */
 				struct lock_list *ul_next = l_curr->next;
 
-				DEBUG(10,("delete case.\n" ));
+				DEBUG(10,(" delete case.\n" ));
 
 				DLIST_REMOVE(lhead, l_curr);
-				if(lhead == NULL)
+				if(lhead == NULL) {
 					break; /* No more list... */
+				}
 
 				l_curr = ul_next;
 				
@@ -909,7 +822,7 @@ OR....
 						(l_curr->start + l_curr->size > lock->start + lock->size) ) {
 
 				/*
-				 * This unlock overlaps the existing lock range at the high end.
+				 * This range overlaps the existing lock range at the high end.
 				 * Truncate by moving start to existing range end and reducing size.
 				 */
 /*********************************************
@@ -928,7 +841,7 @@ BECOMES....
 				l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
 				l_curr->start = lock->start + lock->size;
 
-				DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
+				DEBUG(10,(" truncate high case: start=%.0f,size=%.0f\n",
 								(double)l_curr->start, (double)l_curr->size ));
 
 				l_curr = l_curr->next;
@@ -938,7 +851,7 @@ BECOMES....
 						(l_curr->start + l_curr->size <= lock->start + lock->size) ) {
 
 				/*
-				 * This unlock overlaps the existing lock range at the low end.
+				 * This range overlaps the existing lock range at the low end.
 				 * Truncate by reducing size.
 				 */
 /*********************************************
@@ -956,7 +869,7 @@ BECOMES....
 
 				l_curr->size = lock->start - l_curr->start;
 
-				DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
+				DEBUG(10,(" truncate low case: start=%.0f,size=%.0f\n",
 								(double)l_curr->start, (double)l_curr->size ));
 
 				l_curr = l_curr->next;
@@ -964,10 +877,10 @@ BECOMES....
 			} else if ( (l_curr->start < lock->start) &&
 						(l_curr->start + l_curr->size > lock->start + lock->size) ) {
 				/*
-				 * Worst case scenario. Unlock request completely overlaps an existing
+				 * Worst case scenario. Range completely overlaps an existing
 				 * lock range. Split the request into two, push the new (upper) request
-				 * into the dlink list, and continue with the entry after ul_new (as we
-				 * know that ul_new will not overlap with this lock).
+				 * into the dlink list, and continue with the entry after l_new (as we
+				 * know that l_new will not overlap with this lock).
 				 */
 /*********************************************
         +---------------------------+
@@ -981,8 +894,7 @@ BECOMES.....
         | l_curr|         | l_new   |
         +-------+         +---------+
 **********************************************/
-				struct lock_list *l_new = (struct lock_list *)talloc(ctx,
-													sizeof(struct lock_list));
+				struct lock_list *l_new = TALLOC_P(ctx, struct lock_list);
 
 				if(l_new == NULL) {
 					DEBUG(0,("posix_lock_list: talloc fail.\n"));
@@ -996,7 +908,7 @@ BECOMES.....
 				/* Truncate the l_curr. */
 				l_curr->size = lock->start - l_curr->start;
 
-				DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
+				DEBUG(10,(" split case: curr: start=%.0f,size=%.0f \
 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
 								(double)l_new->start, (double)l_new->size ));
 
@@ -1021,16 +933,13 @@ new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
 				pstring msg;
 
 				slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
-lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
+lock: start = %.0f, size = %.0f", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
 
 				smb_panic(msg);
 			}
 		} /* end for ( l_curr = lhead; l_curr;) */
 	} /* end for (i=0; i<num_locks && ul_head; i++) */
 
-	if (dbuf.dptr)
-		free(dbuf.dptr);
-	
 	return lhead;
 }
 
@@ -1039,19 +948,25 @@ lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size,
  lock could be granted, False if not.
 ****************************************************************************/
 
-BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
+BOOL set_posix_lock_windows_flavour(files_struct *fsp,
+			SMB_BIG_UINT u_offset,
+			SMB_BIG_UINT u_count,
+			enum brl_type lock_type,
+			const struct lock_context *lock_ctx,
+			const struct lock_struct *plocks,
+			int num_locks,
+			int *errno_ret)
 {
 	SMB_OFF_T offset;
 	SMB_OFF_T count;
+	int posix_lock_type = map_posix_lock_type(fsp,lock_type);
 	BOOL ret = True;
-	size_t entry_num = 0;
 	size_t lock_count;
 	TALLOC_CTX *l_ctx = NULL;
 	struct lock_list *llist = NULL;
 	struct lock_list *ll = NULL;
-	int posix_lock_type = map_posix_lock_type(fsp,lock_type);
 
-	DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
+	DEBUG(5,("set_posix_lock_windows_flavour: File %s, offset = %.0f, count = %.0f, type = %s\n",
 			fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
 
 	/*
@@ -1059,8 +974,10 @@ BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_cou
 	 * pretend it was successful.
 	 */
 
-	if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
+	if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
+		increment_windows_lock_ref_count(fsp);
 		return True;
+	}
 
 	/*
 	 * Windows is very strange. It allows read locks to be overlayed
@@ -1077,21 +994,18 @@ BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_cou
 	 *                                            READ LOCK: start =0, len = 10 - OK
 	 *
 	 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
-	 * would leave a single read lock over the 0-14 region. In order to
-	 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
-	 * entries, one for each overlayed lock request. We are guarenteed by the brlock
-	 * semantics that if a write lock is added, then it will be first in the array.
+	 * would leave a single read lock over the 0-14 region.
 	 */
 	
-	if ((l_ctx = talloc_init()) == NULL) {
-		DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
-		return True; /* Not a fatal error. */
+	if ((l_ctx = talloc_init("set_posix_lock")) == NULL) {
+		DEBUG(0,("set_posix_lock_windows_flavour: unable to init talloc context.\n"));
+		return False;
 	}
 
-	if ((ll = (struct lock_list *)talloc(l_ctx, sizeof(struct lock_list))) == NULL) {
-		DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
+	if ((ll = TALLOC_P(l_ctx, struct lock_list)) == NULL) {
+		DEBUG(0,("set_posix_lock_windows_flavour: unable to talloc unlock list.\n"));
 		talloc_destroy(l_ctx);
-		return True; /* Not a fatal error. */
+		return False;
 	}
 
 	/*
@@ -1113,19 +1027,12 @@ BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_cou
 	 * POSIX locks.
 	 */
 
-	llist = posix_lock_list(l_ctx, llist, fsp);
-
-	/*
-	 * Now we have the list of ranges to lock it is safe to add the
-	 * entry into the POSIX lock tdb. We take note of the entry we
-	 * added here in case we have to remove it on POSIX lock fail.
-	 */
-
-	if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
-		DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
-		talloc_destroy(l_ctx);
-		return False;
-	}
+	llist = posix_lock_list(l_ctx,
+				llist,
+				lock_ctx, /* Lock context llist belongs to. */
+				fsp,
+				plocks,
+				num_locks);
 
 	/*
 	 * Add the POSIX locks on the list of ranges returned.
@@ -1137,11 +1044,12 @@ BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_cou
 		offset = ll->start;
 		count = ll->size;
 
-		DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
+		DEBUG(5,("set_posix_lock_windows_flavour: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
 			posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
 
 		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
-			DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
+			*errno_ret = errno;
+			DEBUG(5,("set_posix_lock_windows_flavour: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
 				posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
 			ret = False;
 			break;
@@ -1158,17 +1066,14 @@ BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_cou
 			offset = ll->start;
 			count = ll->size;
 
-			DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
+			DEBUG(5,("set_posix_lock_windows_flavour: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
 				posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
 
 			posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
 		}
-
-		/*
-		 * Remove the tdb entry for this lock.
-		 */
-
-		delete_posix_lock_entry_by_index(fsp,entry_num);
+	} else {
+		/* Remember the number of Windows locks we have on this dev/ino pair. */
+		increment_windows_lock_ref_count(fsp);
 	}
 
 	talloc_destroy(l_ctx);
@@ -1180,7 +1085,13 @@ BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_cou
  lock could be released, False if not.
 ****************************************************************************/
 
-BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
+BOOL release_posix_lock_windows_flavour(files_struct *fsp,
+				SMB_BIG_UINT u_offset,
+				SMB_BIG_UINT u_count,
+				enum brl_type deleted_lock_type,
+				const struct lock_context *lock_ctx,
+				const struct lock_struct *plocks,
+				int num_locks)
 {
 	SMB_OFF_T offset;
 	SMB_OFF_T count;
@@ -1188,56 +1099,31 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
 	TALLOC_CTX *ul_ctx = NULL;
 	struct lock_list *ulist = NULL;
 	struct lock_list *ul = NULL;
-	struct posix_lock deleted_lock;
-	int num_overlapped_entries;
 
-	DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
+	DEBUG(5,("release_posix_lock_windows_flavour: File %s, offset = %.0f, count = %.0f\n",
 		fsp->fsp_name, (double)u_offset, (double)u_count ));
 
+	/* Remember the number of Windows locks we have on this dev/ino pair. */
+	decrement_windows_lock_ref_count(fsp);
+
 	/*
 	 * If the requested lock won't fit in the POSIX range, we will
 	 * pretend it was successful.
 	 */
 
-	if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
+	if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
 		return True;
-
-	/*
-	 * We treat this as one unlock request for POSIX accounting purposes even
-	 * if it may later be split into multiple smaller POSIX unlock ranges.
-	 * num_overlapped_entries is the number of existing locks that have any
-	 * overlap with this unlock request.
-	 */ 
-
-	num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
-
-	if (num_overlapped_entries == -1) {
-        smb_panic("release_posix_lock: unable find entry to delete !\n");
-	}
-
-	/*
-	 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
-	 * a POSIX write lock, then before doing the unlock we need to downgrade
-	 * the POSIX lock to a read lock. This allows any overlapping read locks
-	 * to be atomically maintained.
-	 */
-
-	if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
-		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
-			DEBUG(0,("release_posix_lock: downgrade of lock failed with error %s !\n", strerror(errno) ));
-			return False;
-		}
 	}
 
-	if ((ul_ctx = talloc_init()) == NULL) {
-		DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
-		return True; /* Not a fatal error. */
+	if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
+		DEBUG(0,("release_posix_lock_windows_flavour: unable to init talloc context.\n"));
+		return False;
 	}
 
-	if ((ul = (struct lock_list *)talloc(ul_ctx, sizeof(struct lock_list))) == NULL) {
-		DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
+	if ((ul = TALLOC_P(ul_ctx, struct lock_list)) == NULL) {
+		DEBUG(0,("release_posix_lock_windows_flavour: unable to talloc unlock list.\n"));
 		talloc_destroy(ul_ctx);
-		return True; /* Not a fatal error. */
+		return False;
 	}
 
 	/*
@@ -1260,7 +1146,33 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
 	 * unlocks are performed.
 	 */
 
-	ulist = posix_lock_list(ul_ctx, ulist, fsp);
+	ulist = posix_lock_list(ul_ctx,
+				ulist,
+				lock_ctx, /* Lock context ulist belongs to. */
+				fsp,
+				plocks,
+				num_locks);
+
+	/*
+	 * If there were any overlapped entries (list is > 1 or size or start have changed),
+	 * and the lock_type we just deleted from
+	 * the upper layer tdb was a write lock, then before doing the unlock we need to downgrade
+	 * the POSIX lock to a read lock. This allows any overlapping read locks
+	 * to be atomically maintained.
+	 */
+
+	if (deleted_lock_type == WRITE_LOCK &&
+			(!ulist || ulist->next != NULL || ulist->start != offset || ulist->size != count)) {
+
+		DEBUG(5,("release_posix_lock_windows_flavour: downgrading lock to READ: offset = %.0f, count = %.0f\n",
+			(double)offset, (double)count ));
+
+		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
+			DEBUG(0,("release_posix_lock_windows_flavour: downgrade of lock failed with error %s !\n", strerror(errno) ));
+			talloc_destroy(ul_ctx);
+			return False;
+		}
+	}
 
 	/*
 	 * Release the POSIX locks on the list of ranges returned.
@@ -1270,129 +1182,148 @@ BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u
 		offset = ulist->start;
 		count = ulist->size;
 
-		DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
+		DEBUG(5,("release_posix_lock_windows_flavour: Real unlock: offset = %.0f, count = %.0f\n",
 			(double)offset, (double)count ));
 
-		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
+		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK)) {
 			ret = False;
+		}
 	}
 
 	talloc_destroy(ul_ctx);
-
 	return ret;
 }
 
 /****************************************************************************
- Remove all lock entries for a specific dev/inode pair from the tdb.
+ Next - the functions that deal with mapping CIFS POSIX locks onto
+ the underlying system POSIX locks.
 ****************************************************************************/
 
-static void delete_posix_lock_entries(files_struct *fsp)
-{
-	TDB_DATA kbuf = locking_key_fsp(fsp);
-
-	if (tdb_delete(posix_lock_tdb, kbuf) == -1)
-		DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
-}
-
 /****************************************************************************
- Debug function.
+ POSIX function to acquire a lock. Returns True if the
+ lock could be granted, False if not.
+ As POSIX locks don't stack or conflict (they just overwrite)
+ we can map the requested lock directly onto a system one. We
+ know it doesn't conflict with locks on other contexts as the
+ upper layer would have refused it.
 ****************************************************************************/
 
-static void dump_entry(struct posix_lock *pl)
+BOOL set_posix_lock_posix_flavour(files_struct *fsp,
+			SMB_BIG_UINT u_offset,
+			SMB_BIG_UINT u_count,
+			enum brl_type lock_type,
+			int *errno_ret)
 {
-	DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
-		(double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
+	SMB_OFF_T offset;
+	SMB_OFF_T count;
+	int posix_lock_type = map_posix_lock_type(fsp,lock_type);
+
+	DEBUG(5,("set_posix_lock_posix_flavour: File %s, offset = %.0f, count = %.0f, type = %s\n",
+			fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
+
+	/*
+	 * If the requested lock won't fit in the POSIX range, we will
+	 * pretend it was successful.
+	 */
+
+	if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
+		return True;
+	}
+
+	if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
+		*errno_ret = errno;
+		DEBUG(5,("set_posix_lock_posix_flavour: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
+			posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
+		return False;
+	}
+	return True;
 }
 
 /****************************************************************************
- Remove any locks on this fd. Called from file_close().
+ POSIX function to release a lock. Returns True if the
+ lock could be released, False if not.
+ We are given a complete lock state from the upper layer which is what the lock
+ state should be after the unlock has already been done, so what
+ we do is punch out holes in the unlock range where locks owned by this process
+ have a different lock context.
 ****************************************************************************/
 
-void posix_locking_close_file(files_struct *fsp)
+BOOL release_posix_lock_posix_flavour(files_struct *fsp,
+				SMB_BIG_UINT u_offset,
+				SMB_BIG_UINT u_count,
+				const struct lock_context *lock_ctx,
+				const struct lock_struct *plocks,
+				int num_locks)
 {
-	struct posix_lock *entries = NULL;
-	size_t count, i;
+	BOOL ret = True;
+	SMB_OFF_T offset;
+	SMB_OFF_T count;
+	TALLOC_CTX *ul_ctx = NULL;
+	struct lock_list *ulist = NULL;
+	struct lock_list *ul = NULL;
+
+	DEBUG(5,("release_posix_lock_posix_flavour: File %s, offset = %.0f, count = %.0f\n",
+		fsp->fsp_name, (double)u_offset, (double)u_count ));
 
 	/*
-	 * Optimization for the common case where we are the only
-	 * opener of a file. If all fd entries are our own, we don't
-	 * need to explicitly release all the locks via the POSIX functions,
-	 * we can just remove all the entries in the tdb and allow the
-	 * close to remove the real locks.
+	 * If the requested lock won't fit in the POSIX range, we will
+	 * pretend it was successful.
 	 */
 
-	count = get_posix_lock_entries(fsp, &entries);
-
-	if (count == 0) {
-		DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
-		return;
+	if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
+		return True;
 	}
 
-	for (i = 0; i < count; i++) {
-		if (entries[i].fd != fsp->fd )
-			break;
-
-		dump_entry(&entries[i]);
+	if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
+		DEBUG(0,("release_posix_lock_windows_flavour: unable to init talloc context.\n"));
+		return False;
 	}
 
-	if (i == count) {
-		/* All locks are ours. */
-		DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n", 
-			fsp->fsp_name, (unsigned int)count ));
-		free((char *)entries);
-		delete_posix_lock_entries(fsp);
-		return;
+	if ((ul = TALLOC_P(ul_ctx, struct lock_list)) == NULL) {
+		DEBUG(0,("release_posix_lock_windows_flavour: unable to talloc unlock list.\n"));
+		talloc_destroy(ul_ctx);
+		return False;
 	}
 
 	/*
-	 * Difficult case. We need to delete all our locks, whilst leaving
-	 * all other POSIX locks in place.
+	 * Create the initial list entry containing the
+	 * lock we want to remove.
 	 */
 
-	for (i = 0; i < count; i++) {
-		struct posix_lock *pl = &entries[i];
-		if (pl->fd == fsp->fd)
-			release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
-	}
-	free((char *)entries);
-}
+	ZERO_STRUCTP(ul);
+	ul->start = offset;
+	ul->size = count;
 
-/*******************************************************************
- Create the in-memory POSIX lock databases.
-********************************************************************/
+	DLIST_ADD(ulist, ul);
 
-BOOL posix_locking_init(int read_only)
-{
-	if (posix_lock_tdb && posix_pending_close_tdb)
-		return True;
-	
-	if (!posix_lock_tdb)
-		posix_lock_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
-					  read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
-	if (!posix_lock_tdb) {
-		DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
-		return False;
-	}
-	if (!posix_pending_close_tdb)
-		posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
-						   read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
-	if (!posix_pending_close_tdb) {
-		DEBUG(0,("Failed to open POSIX pending close database.\n"));
-		return False;
-	}
+	/*
+	 * Walk the given array creating a linked list
+	 * of unlock requests.
+	 */
 
-	return True;
-}
+	ulist = posix_lock_list(ul_ctx,
+				ulist,
+				lock_ctx, /* Lock context ulist belongs to. */
+				fsp,
+				plocks,
+				num_locks);
 
-/*******************************************************************
- Delete the in-memory POSIX lock databases.
-********************************************************************/
+	/*
+	 * Release the POSIX locks on the list of ranges returned.
+	 */
 
-BOOL posix_locking_end(void)
-{
-    if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
-		return False;
-    if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)
-		return False;
-	return True;
+	for(; ulist; ulist = ulist->next) {
+		offset = ulist->start;
+		count = ulist->size;
+
+		DEBUG(5,("release_posix_lock_posix_flavour: Real unlock: offset = %.0f, count = %.0f\n",
+			(double)offset, (double)count ));
+
+		if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK)) {
+			ret = False;
+		}
+	}
+
+	talloc_destroy(ul_ctx);
+	return ret;
 }