Skip to contentSkip to conversation

xfs: ensure truncate forces zeroed blocks to disk

PATCH In review Dave Chinner 3 participants 4 messages Feb 18, 2015 – Feb 23, 2015 XFS filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xfs: ensure truncate forces zeroed blocks to disk

From: Dave Chinner <[email protected]>

A new fsync vs power fail test in xfstests indicated that XFS can have unreliable data consistency when doing extending truncates that require block zeroing. The blocks beyond EOF get zeroed in memory, but we never force those changes to disk before we run the transaction that extends the file size and exposes those blocks to userspace. This can result in the blocks not being correctly zeroed after a crash.

Because in-memory behaviour is correct, tools like fsx don't pick up any coherency problems - it's not until the filesystem is shutdown or the system crashes after writing the truncate transaction to the journal but before the zeroed data in the page cache is flushed that the issue is exposed.

Fix this by also flushing the dirty data in memory region between the old size and new size when we've found blocks that need zeroing in the truncate process.

---
 fs/xfs/xfs_file.c  | 14 ++++++++++----
 fs/xfs/xfs_inode.h |  9 +++++----
 fs/xfs/xfs_iops.c  | 36 ++++++++++++++----------------------
 3 files changed, 29 insertions(+), 30 deletions(-)
fs/xfs/xfs_file.c +10 -4
index ce615d1..a2e1cb8 100644
@@ -397,7 +397,8 @@ STATIC int				/* error (positive) */
 xfs_zero_last_block(
 	struct xfs_inode	*ip,
 	xfs_fsize_t		offset,
-	xfs_fsize_t		isize)
+	xfs_fsize_t		isize,
+	bool			*did_zeroing)
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	xfs_fileoff_t		last_fsb = XFS_B_TO_FSBT(mp, isize);
@@ -425,6 +426,7 @@ xfs_zero_last_block(
 	zero_len = mp->m_sb.sb_blocksize - zero_offset;
 	if (isize + zero_len > offset)
 		zero_len = offset - isize;
+	*did_zeroing = true;
 	return xfs_iozero(ip, isize, zero_len);
 }
 
@@ -443,7 +445,8 @@ int					/* error (positive) */
 xfs_zero_eof(
 	struct xfs_inode	*ip,
 	xfs_off_t		offset,		/* starting I/O offset */
-	xfs_fsize_t		isize)		/* current inode size */
+	xfs_fsize_t		isize,		/* current inode size */
+	bool			*did_zeroing)
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	xfs_fileoff_t		start_zero_fsb;
@@ -465,7 +468,7 @@ xfs_zero_eof(
 	 * We only zero a part of that block so it is handled specially.
 	 */
 	if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
-		error = xfs_zero_last_block(ip, offset, isize);
+		error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
 		if (error)
 			return error;
 	}
@@ -525,6 +528,7 @@ xfs_zero_eof(
 		if (error)
 			return error;
 
+		*did_zeroing = true;
 		start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 		ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 	}
@@ -567,13 +571,15 @@ restart:
 	 * having to redo all checks before.
 	 */
 	if (*pos > i_size_read(inode)) {
+		bool	zero = false;
+
 		if (*iolock == XFS_IOLOCK_SHARED) {
 			xfs_rw_iunlock(ip, *iolock);
 			*iolock = XFS_IOLOCK_EXCL;
 			xfs_rw_ilock(ip, *iolock);
 			goto restart;
 		}
-		error = xfs_zero_eof(ip, *pos, i_size_read(inode));
+		error = xfs_zero_eof(ip, *pos, i_size_read(inode), &zero);
 		if (error)
 			return error;
 	}
fs/xfs/xfs_inode.h +5 -4
index 8e82b41..c73b63d 100644
@@ -384,10 +384,11 @@ enum xfs_prealloc_flags {
 	XFS_PREALLOC_INVISIBLE	= (1 << 4),
 };
 
-int		xfs_update_prealloc_flags(struct xfs_inode *,
-			enum xfs_prealloc_flags);
-int		xfs_zero_eof(struct xfs_inode *, xfs_off_t, xfs_fsize_t);
-int		xfs_iozero(struct xfs_inode *, loff_t, size_t);
+int	xfs_update_prealloc_flags(struct xfs_inode *ip,
+				  enum xfs_prealloc_flags flags);
+int	xfs_zero_eof(struct xfs_inode *ip, xfs_off_t offset,
+		     xfs_fsize_t isize, bool *did_zeroing);
+int	xfs_iozero(struct xfs_inode *ip, loff_t pos, size_t count);
 
 
 /* from xfs_iops.c */
fs/xfs/xfs_iops.c +14 -22
index d7782ae..3ccc28e 100644
@@ -756,6 +756,7 @@ xfs_setattr_size(
 	int			error;
 	uint			lock_flags = 0;
 	uint			commit_flags = 0;
+	bool			did_zeroing = false;
 
 	trace_xfs_setattr(ip);
 
@@ -799,20 +800,16 @@ xfs_setattr_size(
 		return error;
 
 	/*
-	 * Now we can make the changes.  Before we join the inode to the
-	 * transaction, take care of the part of the truncation that must be
-	 * done without the inode lock.  This needs to be done before joining
-	 * the inode to the transaction, because the inode cannot be unlocked
-	 * once it is a part of the transaction.
+	 * File data changes must be complete before we start the transaction to
+	 * modify the inode.  This needs to be done before joining the inode to
+	 * the transaction because the inode cannot be unlocked once it is a
+	 * part of the transaction.
+	 *
+	 * Start with zeroing any data block beyond EOF that we may expose on
+	 * file extension.
 	 */
 	if (newsize > oldsize) {
-		/*
-		 * Do the first part of growing a file: zero any data in the
-		 * last block that is beyond the old EOF.  We need to do this
-		 * before the inode is joined to the transaction to modify
-		 * i_size.
-		 */
-		error = xfs_zero_eof(ip, newsize, oldsize);
+		error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
 		if (error)
 			return error;
 	}
@@ -822,23 +819,18 @@ xfs_setattr_size(
 	 * any previous writes that are beyond the on disk EOF and the new
 	 * EOF that have not been written out need to be written here.  If we
 	 * do not write the data out, we expose ourselves to the null files
-	 * problem.
-	 *
-	 * Only flush from the on disk size to the smaller of the in memory
-	 * file size or the new size as that's the range we really care about
-	 * here and prevents waiting for other data not within the range we
-	 * care about here.
+	 * problem. Note that this includes any block zeroing we did above;
+	 * otherwise those blocks may not be zeroed after a crash.
 	 */
-	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
+	if (newsize > ip->i_d.di_size &&
+	    (oldsize != ip->i_d.di_size || did_zeroing)) {
 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 						      ip->i_d.di_size, newsize);
 		if (error)
 			return error;
 	}
 
-	/*
-	 * Wait for all direct I/O to complete.
-	 */
+	/* Now wait for all direct I/O to complete. */
 	inode_dio_wait(inode);
 
 	/*
Signature · 6 lines
2.0.0

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Reviewed-by on v1 1/1
#1 lore raw
Quoted 16 lines · Dave Chinner

From: Dave Chinner <[email protected]>

A new fsync vs power fail test in xfstests indicated that XFS can have unreliable data consistency when doing extending truncates that require block zeroing. The blocks beyond EOF get zeroed in memory, but we never force those changes to disk before we run the transaction that extends the file size and exposes those blocks to userspace. This can result in the blocks not being correctly zeroed after a crash.

Because in-memory behaviour is correct, tools like fsx don't pick up any coherency problems - it's not until the filesystem is shutdown or the system crashes after writing the truncate transaction to the journal but before the zeroed data in the page cache is flushed that the issue is exposed.

Quoted 8 lines

Fix this by also flushing the dirty data in memory region between the old size and new size when we've found blocks that need zeroing in the truncate process.

---

Quoted 168 lines
 fs/xfs/xfs_file.c  | 14 ++++++++++----
 fs/xfs/xfs_inode.h |  9 +++++----
 fs/xfs/xfs_iops.c  | 36 ++++++++++++++----------------------
 3 files changed, 29 insertions(+), 30 deletions(-)
fs/xfs/xfs_file.c +10 -4
index ce615d1..a2e1cb8 100644
@@ -397,7 +397,8 @@ STATIC int				/* error (positive) */
 xfs_zero_last_block(
 	struct xfs_inode	*ip,
 	xfs_fsize_t		offset,
-	xfs_fsize_t		isize)
+	xfs_fsize_t		isize,
+	bool			*did_zeroing)
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	xfs_fileoff_t		last_fsb = XFS_B_TO_FSBT(mp, isize);
@@ -425,6 +426,7 @@ xfs_zero_last_block(
 	zero_len = mp->m_sb.sb_blocksize - zero_offset;
 	if (isize + zero_len > offset)
 		zero_len = offset - isize;
+	*did_zeroing = true;
 	return xfs_iozero(ip, isize, zero_len);
 }
 
@@ -443,7 +445,8 @@ int					/* error (positive) */
 xfs_zero_eof(
 	struct xfs_inode	*ip,
 	xfs_off_t		offset,		/* starting I/O offset */
-	xfs_fsize_t		isize)		/* current inode size */
+	xfs_fsize_t		isize,		/* current inode size */
+	bool			*did_zeroing)
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	xfs_fileoff_t		start_zero_fsb;
@@ -465,7 +468,7 @@ xfs_zero_eof(
 	 * We only zero a part of that block so it is handled specially.
 	 */
 	if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
-		error = xfs_zero_last_block(ip, offset, isize);
+		error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
 		if (error)
 			return error;
 	}
@@ -525,6 +528,7 @@ xfs_zero_eof(
 		if (error)
 			return error;
 
+		*did_zeroing = true;
 		start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 		ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 	}
@@ -567,13 +571,15 @@ restart:
 	 * having to redo all checks before.
 	 */
 	if (*pos > i_size_read(inode)) {
+		bool	zero = false;
+
 		if (*iolock == XFS_IOLOCK_SHARED) {
 			xfs_rw_iunlock(ip, *iolock);
 			*iolock = XFS_IOLOCK_EXCL;
 			xfs_rw_ilock(ip, *iolock);
 			goto restart;
 		}
-		error = xfs_zero_eof(ip, *pos, i_size_read(inode));
+		error = xfs_zero_eof(ip, *pos, i_size_read(inode), &zero);
 		if (error)
 			return error;
 	}
fs/xfs/xfs_inode.h +5 -4
index 8e82b41..c73b63d 100644
@@ -384,10 +384,11 @@ enum xfs_prealloc_flags {
 	XFS_PREALLOC_INVISIBLE	= (1 << 4),
 };
 
-int		xfs_update_prealloc_flags(struct xfs_inode *,
-			enum xfs_prealloc_flags);
-int		xfs_zero_eof(struct xfs_inode *, xfs_off_t, xfs_fsize_t);
-int		xfs_iozero(struct xfs_inode *, loff_t, size_t);
+int	xfs_update_prealloc_flags(struct xfs_inode *ip,
+				  enum xfs_prealloc_flags flags);
+int	xfs_zero_eof(struct xfs_inode *ip, xfs_off_t offset,
+		     xfs_fsize_t isize, bool *did_zeroing);
+int	xfs_iozero(struct xfs_inode *ip, loff_t pos, size_t count);
 
 
 /* from xfs_iops.c */
fs/xfs/xfs_iops.c +14 -22
index d7782ae..3ccc28e 100644
@@ -756,6 +756,7 @@ xfs_setattr_size(
 	int			error;
 	uint			lock_flags = 0;
 	uint			commit_flags = 0;
+	bool			did_zeroing = false;
 
 	trace_xfs_setattr(ip);
 
@@ -799,20 +800,16 @@ xfs_setattr_size(
 		return error;
 
 	/*
-	 * Now we can make the changes.  Before we join the inode to the
-	 * transaction, take care of the part of the truncation that must be
-	 * done without the inode lock.  This needs to be done before joining
-	 * the inode to the transaction, because the inode cannot be unlocked
-	 * once it is a part of the transaction.
+	 * File data changes must be complete before we start the transaction to
+	 * modify the inode.  This needs to be done before joining the inode to
+	 * the transaction because the inode cannot be unlocked once it is a
+	 * part of the transaction.
+	 *
+	 * Start with zeroing any data block beyond EOF that we may expose on
+	 * file extension.
 	 */
 	if (newsize > oldsize) {
-		/*
-		 * Do the first part of growing a file: zero any data in the
-		 * last block that is beyond the old EOF.  We need to do this
-		 * before the inode is joined to the transaction to modify
-		 * i_size.
-		 */
-		error = xfs_zero_eof(ip, newsize, oldsize);
+		error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
 		if (error)
 			return error;
 	}
@@ -822,23 +819,18 @@ xfs_setattr_size(
 	 * any previous writes that are beyond the on disk EOF and the new
 	 * EOF that have not been written out need to be written here.  If we
 	 * do not write the data out, we expose ourselves to the null files
-	 * problem.
-	 *
-	 * Only flush from the on disk size to the smaller of the in memory
-	 * file size or the new size as that's the range we really care about
-	 * here and prevents waiting for other data not within the range we
-	 * care about here.
+	 * problem. Note that this includes any block zeroing we did above;
+	 * otherwise those blocks may not be zeroed after a crash.
 	 */
-	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
+	if (newsize > ip->i_d.di_size &&
+	    (oldsize != ip->i_d.di_size || did_zeroing)) {
 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 						      ip->i_d.di_size, newsize);
 		if (error)
 			return error;
 	}
 
-	/*
-	 * Wait for all direct I/O to complete.
-	 */
+	/* Now wait for all direct I/O to complete. */
 	inode_dio_wait(inode);
 
 	/*
Signature · 6 lines
2.0.0

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

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

#3 lore raw
Quoted 2 lines · Christoph Hellwig

This looks correct, but is there a good (performance) reason against simply unconditionally flushing and waiting?

No idea. All I am concerned about is correctness - getting the partial block flushed in the case that the on-disk size is the same as the in-memory size is the fix needed here, otherwise the behaviour should be unchanged.

Given that I'm not sure what the effect of an unconditional flush is going to be, I'm not going to mix such a change with an otherwise obvious data corruption fix that we need to backport to other kernels. If you have the time to determine there is no performance impact from an unconditional flush, then I'll happily take the change. ;)

Cheers,

Dave.

Signature · 7 lines
Dave Chinner
[email protected]

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