On Mon, 21 May 2012 13:34:05 +0800, Miao Xie wrote:

On Fri, 18 May 2012 14:52:07 +0200, David Sterba wrote:

quoted

On Thu, May 17, 2012 at 07:58:21PM +0800, Miao Xie wrote:

quoted

--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c

@@ -1151,6 +1151,8 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 		/* pause restriper - we want to resume on remount to r/w */
 		btrfs_pause_balance(root->fs_info);
 
+		btrfs_scrub_cancel(root);

Can we possibly switch scrub to readonly instead ? I'm not sure what's
the 'least surprise here', whether to cancel everything on the
filesystem upon ro-remount or just the minimal set of operations (and
leave the rest running if possible).

I don't think it is better to switch scrub to readonly soundlessly, because
it is not the operation that the users expect if they don't choose the readonly
mode.

Since David put me on CC for this thread, I state my humble opinion:
If I manually remount read-only, I would expect that scrub is switched
into read-only mode as well, not canceled. Plus a log message about this
change.

BTW, I think we needn't cancel the readonly scrub on the filesystem upon ro-remount.

Thanks
Miao

quoted

Looking at the scrub code, if dev->readonly is set, no repairs are done,
so the only concern is to wait for any outstanding IOs and then switch
to RO.


david