[PATCH] Fix ext2 error reporting on fsync

[Jan Kara <jack@suse.cz>]

[-- Attachment #1: Type: text/plain, Size: 380 bytes --]

  Hello,

  attached patch (from Chris Mason) fixes possible bug in error
reporting during fsync on ext2 filesystem. If some previous metadata
async write failed with EIO, it just set AS_EIO in the device's mapping.
We need to check this in ext2_file_write() and act accordingly. The
patch is against 2.6.15. Please apply.

								Honza

-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs

[-- Attachment #2: ext2-sync_check-2.6.15.diff --]
[-- Type: text/plain, Size: 1831 bytes --]

From: Chris Mason <mason@suse.com>
Subject: ext2 should force the FS readonly for metadata write errors
References: 65718

During fsync we should check for write errors to the block device in order to make
sure all metadata writes have been properly written to the disk.  Without this check
writes that happen through the normal async mechanisms might hit errors without
reporting them back to the application.

Signed-off-by: Jan Kara <jack@suse.cz>

diff -rupX /home/jack/.kerndiffexclude linux-2.6.15/fs/ext2/fsync.c linux-2.6.15-1-ext2_sync_check/fs/ext2/fsync.c
--- linux-2.6.15/fs/ext2/fsync.c	2004-10-18 23:55:24.000000000 +0200
+++ linux-2.6.15-1-ext2_sync_check/fs/ext2/fsync.c	2006-01-13 22:56:43.000000000 +0100
@@ -25,6 +25,7 @@
 #include "ext2.h"
 #include <linux/smp_lock.h>
 #include <linux/buffer_head.h>		/* for fsync_inode_buffers() */
+#include <linux/pagemap.h>
 
 
 /*
@@ -35,10 +36,26 @@
 int ext2_sync_file(struct file *file, struct dentry *dentry, int datasync)
 {
 	struct inode *inode = dentry->d_inode;
+	struct super_block *sb = inode->i_sb;
 	int err;
 	int ret;
 
 	ret = sync_mapping_buffers(inode->i_mapping);
+
+	/* it might make more sense to ext2_error on -EIO from
+	 * sync_mapping_buffers as well, but those errors are isolated to just
+	 * this file. We can safely return -EIO to fsync and let the app know
+	 * they have a problem.  
+	 *
+	 * AS_EIO indicates a failure to write a metadata page, but we have no 
+	 * way of knowing which one.  It's best to force readonly and let fsck 
+	 * figure it all out.
+	 */
+	if (test_and_clear_bit(AS_EIO, &sb->s_bdev->bd_inode->i_mapping->flags)) {
+		ext2_error(sb, "ext2_sync_file", "metadata io error");
+		if (!ret)
+			ret = -EIO;
+	}
 	if (!(inode->i_state & I_DIRTY))
 		return ret;
 	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))