On Tue, Aug 2, 2011 at 11:41 AM, Ben Hutchings <b...@decadent.org.uk> wrote: > On Tue, 2011-08-02 at 03:51 +0100, Luke Kenneth Casson Leighton wrote: >> On Tue, Aug 2, 2011 at 1:54 AM, Ben Hutchings <b...@decadent.org.uk> wrote: >> > On Tue, 2011-08-02 at 01:29 +0100, lkcl wrote: >> >> Package: linux-2.6 >> >> Version: 2.6.32-5 >> > >> > Not an actual version number. >> > >> >> Severity: important >> >> >> >> >> >> running 2.6.32-5-amd64, after running fsck on a corrupted ext4 >> >> filesystem which is an LVM partition on top of a RAID1 mirror with 3 >> >> drives and is 1tb in size, there are *still* errors after the fsck, >> >> as detected by running fsck a 2nd time. >> > >> > That would be a bug in e2fsck, not the kernel. But I would be more >> > inclined to suspect some sort of hardware problem. >> >> ... across three drives?? all three physical drives have a hardware >> problem? yes i know they're WD 1.5Tb external USB2 drives, but even >> so. > > RAID1 doesn't protect against data corruption, only total failure of a > drive. Therefore only one drive would need to be faulty. And I wasn't > particularly thinking of the drives being faulty, anyway.
yeah - i've found what the problem is: it's that the bitmap blocks were corrupted, and i surmise that the bitmap block "fixing" resulted in fsck re-introducing filesystem corruption. so, yes, i believe you're right about the suspicions falling on e2fsck. i'll look up how to reassign this to e2fsck. >> >> i'd say this is fairly serious, and am not going to hang about: will >> >> be moving this data onto ext3 as quickly as possible. >> > >> > And yet ext4 works fine for other people. >> >> yehh... are they using 1Tb logical volumes on triple-RAID1'd 1.5Tb >> drives, though - that's the question. > > I expect some other people are. If I were to guess at the least > well-tested part of that, though, it would be RAID1 over 3 rather than 2 > drives. that's valuable insight [for the e2fsck team]. thanks ben. l. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
