Skip to contentSkip to conversation

xfs: avoid direct I/O write vs buffered I/O race

PATCH In review Christoph Hellwig 3 participants 3 messages Aug 27, 2011 – Sep 12, 2011 XFS filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xfs: avoid direct I/O write vs buffered I/O race

Currently a buffered reader or writer can add pages to the pagecache while we are waiting for the iolock in xfs_file_dio_aio_write. Prevent this by re-checking mapping->nrpages after we got the iolock, and if nessecary upgrade the lock to exclusive mode. To simplify this a bit only take the ilock inside of xfs_file_aio_write_checks.

fs/xfs/xfs_file.c +14 -3
Index: xfs/fs/xfs/xfs_file.c
===================================================================
@@ -680,6 +680,7 @@ xfs_file_aio_write_checks(
 	xfs_fsize_t		new_size;
 	int			error = 0;
 
+	xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
 	*new_sizep = 0;
 restart:
 	error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
@@ -802,14 +803,24 @@ xfs_file_dio_aio_write(
 		*iolock = XFS_IOLOCK_EXCL;
 	else
 		*iolock = XFS_IOLOCK_SHARED;
-	xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
+	xfs_rw_ilock(ip, *iolock);
+
+	/*
+	 * Recheck if there are cached pages that need invalidate after we got
+	 * the iolock to protect against other threads adding new pages while
+	 * we were waiting for the iolock.
+	 */
+	if (mapping->nrpages && *iolock == XFS_IOLOCK_SHARED) {
+		xfs_rw_iunlock(ip, *iolock);
+		*iolock = XFS_IOLOCK_EXCL;
+		xfs_rw_ilock(ip, *iolock);
+	}
 
 	ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
 	if (ret)
 		return ret;
 
 	if (mapping->nrpages) {
-		WARN_ON(*iolock != XFS_IOLOCK_EXCL);
 		ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
 							FI_REMAPF_LOCKED);
 		if (ret)
@@ -855,7 +866,7 @@ xfs_file_buffered_aio_write(
 	size_t			count = ocount;
 
 	*iolock = XFS_IOLOCK_EXCL;
-	xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
+	xfs_rw_ilock(ip, *iolock);
 
 	ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
 	if (ret)

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Reviewed-by on v1 1/1
#1 lore raw
Quoted 7 lines · Christoph Hellwig

Currently a buffered reader or writer can add pages to the pagecache while we are waiting for the iolock in xfs_file_dio_aio_write. Prevent this by re-checking mapping->nrpages after we got the iolock, and if nessecary upgrade the lock to exclusive mode. To simplify this a bit only take the ilock inside of xfs_file_aio_write_checks.

Looks sane. Pushing the ILOCK completely into xfs_file_aio_write_checks() is a nice segregation of locking responsibilities.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Reviewed-by on v1 1/1
#2 lore raw
In reply to Christoph Hellwig
Quoted 7 lines · Christoph Hellwig

Currently a buffered reader or writer can add pages to the pagecache while we are waiting for the iolock in xfs_file_dio_aio_write. Prevent this by re-checking mapping->nrpages after we got the iolock, and if nessecary upgrade the lock to exclusive mode. To simplify this a bit only take the ilock inside of xfs_file_aio_write_checks.

I agree with Dave, moving the ILOCK acquisition into xfs_file_aio_write_checks() is very nice.

In any case the core change looks good.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs