On Mon, 2013-01-28 at 13:39 +0100, Sebastian Riemer wrote:
>
> On 28.01.2013 11:32, Sebastian Riemer wrote:
> > O.K., then I hope Neil applies the attached patch. I've changed the
> > return value to success.
> >
> > This is also something for linux-stable and should apply to many kernel
> > versions without an issue.
> >
>
> I've tried to race with continuous fsyncs against continuous "mdadm -o
> /dev/md0; mdadm -w /dev/md0;" in parallel but couldn't break it.
> Therefore, no additional locking is required and this part can be put
> directly before the RCU locking stuff and the suspended handling.
>
> I've attached version 3 of the patch.
> From fe0357344877c9b9cc623fd582a4e0670e448317 Mon Sep 17 00:00:00 2001
> From: Sebastian Riemer <sebastian.riemer@profitbricks.com>
> Date: Fri, 25 Jan 2013 12:46:59 +0100
> Subject: [PATCH v3] md: protect against crash upon fsync on ro array
>
> If an fsync occurrs on a read-only array, we need to send a
> completion for the IO and may not increment the active IO count.
> Otherwise, we hit a bug trace and can't stop the MD array anymore.
>
> By advice of Christoph Hellwig we silently return success.
>
> Cc: Christoph Hellwig <hch@infradead.org>
> Cc: Ben Hutchings <ben@decadent.org.uk>
> Cc: NeilBrown <neilb@suse.de>
> Signed-off-by: Sebastian Riemer <sebastian.riemer@profitbricks.com>
> Reported-by: Ben Hutchings <ben@decadent.org.uk>
>
> ---
> drivers/md/md.c | 5 +++++
> 1 files changed, 5 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/md/md.c b/drivers/md/md.c
> index 3db3d1b..6ba20f7 100644
> --- a/drivers/md/md.c
> +++ b/drivers/md/md.c
> @@ -307,6 +307,10 @@ static void md_make_request(struct request_queue
> *q, struct bio *bio)
> bio_io_error(bio);
> return;
> }
> + if (mddev->ro == 1 && unlikely(rw == WRITE)) {
> + bio_endio(bio, 0);
> + return;
> + }
> smp_rmb(); /* Ensure implications of 'active' are visible */
> rcu_read_lock();
> if (mddev->suspended) {
I'm slightly uneasy about returning 0 for all writes to a read-only
deivce, because if someone ever fails to check the read-only flag
elsewhere this could turn into silent data loss. Does this work:
BUG_ON(bio_segments(bio));
Or should it be:
bio_endio(bio, bio_segments(bio) == 0 ? 0 : -EROFS);
to make the error survivable?
Ben.
--
Ben Hutchings
If more than one person is responsible for a bug, no one is at fault.
Attachment:
signature.asc
Description: This is a digitally signed message part