ASSERT in xlog_find_verify_log_record during xfs/098

ASSERT in xlog_find_verify_log_record during xfs/098

[Eric Sandeen]

Al mentioned that he hit an ASSERT during xfs/098, a log fuzzer
test, in xlog_find_verify_log_record():


        for (i = (*last_blk) - 1; i >= 0; i--) {
                if (i < start_blk) {
                        /* valid log record not found */
                        xfs_warn(log->l_mp,
                "Log inconsistent (didn't find previous header)");
                        ASSERT(0);
                        error = -EIO;
                        goto out;
                }

Should that assert even be there?  Looks like it's gracefully
handled, I don't see offhand that anything should have caught
this corruption earlier, and we don't really want to bug on disk
corruption.  Am I missing something?

Thanks,
-Eric

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: ASSERT in xlog_find_verify_log_record during xfs/098

[Christoph Hellwig]

On Tue, Mar 01, 2016 at 01:04:58PM -0600, Eric Sandeen wrote:
> Should that assert even be there?  Looks like it's gracefully
> handled, I don't see offhand that anything should have caught
> this corruption earlier, and we don't really want to bug on disk
> corruption.  Am I missing something?

We used to have a couple of these, especially in log recovery - whoever
wrote this code probably though trapping on these corruptions for
a debug kernel makes sense.

I'm fine with dropping them, and a little audit for more of them
might not hurt.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs