Re: regarding bug #5914 - fs corruption on SATA

[Jens Axboe <axboe@suse.de>]

On Thu, Jan 26 2006, Nicolas Mailhot wrote:
> 
> Le Jeu 26 janvier 2006 06:51, Tejun Heo a écrit :
> > Tejun Heo wrote:
> >> Hello, Nicolas.  Hello, all.
> 
> Hi
> 
> >> Nicolas, I'm probably the guy who broke your filesystem.  :-p This FUA
> >> (forced-unit-access)thing made into the mainline lately, and it seems
> >> that your drive is reporting FUA support but doesn't really do it
> >> properly when it's asked to.
> >>
> >> Can you try the followings to verify the problem?
> >>
> >> 1. make a small partition on the affected drive and do mkfs.ext3 on it.
> >> 2. mount -o barrier new_partition /mnt/tmp
> >
> > This should be 'mount -o barrier=1 new_partition /mnt/tmp'
> >
> >> 3. cd /mnt/tmp; touch asdf; sync
> 
> What parts can be done one a pre-breakage kernel and what parts on a
> problem kernel (I ask this because a problem kernel will corrupt basically
> any file it writes to, even in single login mode the damage is significant
> so I need to limit the corruption window to minimum).

You need a new kernel (after the barrier rework), so 2.6.16-rc1 for
instance.

> Also I have plenty of space to create partitions but that will be
> lvm-on-md-raid1 space (don't know if it matters, if it does I need to
> learn to shrink the lvm/md)

It would be best to exclude lvm/md for now, but I can see it might not
be so easy for you...

> BTW what's FUA in semi-layman terms ?

It stands for Forced Unit Access, basically a way to force the drive to
write through the cache directly to platter even when write back caching
is enabled. Or just bypass the cache on a read, but we use it for writes
with the barrier stuff.

-- 
Jens Axboe