Re: ide errors in 7-rc1-mm1 and later

[Chris Mason <mason@suse.com>]

On Wed, 2004-06-09 at 20:38, Andrew Morton wrote:
> Chris Mason <mason@suse.com> wrote:
> >
> > On Wed, 2004-06-09 at 19:50, Andrew Morton wrote:
> > > Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl> wrote:
> > > >
> > > > Does journal has checksum or some other protection against failure during
> > > > writing journal to a disk?  If not than it still can be screwed even with
> > > > ordered writes if we are unfortunate enough.  ;-)
> > > 
> > > A transaction is written to disk as two synchronous operations: write all
> > > the data, wait on it, write the single commit block, wait on that.
> > > 
> > > If the commit block were to hit disk before the data then we have a window
> > > in which poweroff+recovery would replay garbage into the filesystem.
> > > 
> > > So I think we have a bug in the current ext3 barrier implementation - we
> > > need a blk_issue_flush() before submitting the buffer_ordered commit block.
> > 
> > The IDE barriers are both a pre and post flush.  If the commit block is
> > ordered, before the commit block hits the disk we know all the blocks
> > previously submitted are also on disk.
> > 
> 
> Oh, OK.  Will the same apply to (for example) scsi?

For scsi the general expectation is that write cache will be off unless
it is battery backed.  blkdev_issue_flush does go down to scsi, but I'm
not sure about the regular WRITE_BARRIER stuff.  Jens?

It's true that we need an extra step for external journals in both ext3
and reiser.  We need extra flushes for O_SYNC and O_DIRECT as well, I
wanted to get the core basics working and API fixed before we sprinkling
flushes all over the kernel for complete coverage.

I just did some benchmarking of the two BH_Eopnotsupp patches I sent,
and for synctest -t 20 -f -n 1 dir, there's not enough difference
between barriers on and off for ext3. (1-2% at most).  It doesn't look
like ext3_sync_file is triggering commits all the time, I think we need
extra flushes there too.

Andrew, both O_SYNC and ext3 fsync rely on inode->i_state & I_DIRTY to
decide when to call write_inode(wait = 1).  What happens when a
background writeout clears I_DIRTY without triggering the commit?  Looks
like we won't wait on the transaction to complete in this case.

-chris