[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: File System crash

david.lozano wrote:

Hi all,

I've got a new PC and I've just installed that:

On a 160 Gbytes HD, three partitions:

hda1 with Windows 2000 (119 Gbytes), NTFS
hda2 with a Linux swap partition (1 Gbyte)
hda3 with an ext2 file system (40 Gbytes)

My linux installation is a Debian Woody (just downloaded the image files
using jigdo a month ago).

My problem: after rebooting from Linux (ctrl+alt+del) the FS crashed. I
was told to run fsck on /dev/hda3 manually, so I did. I had to use the -y
option because the huge amount of question I was asked (several
thousands). Windows 2000 worked fine.

After running fsck Debian started fine again, but Windows 2000 FS didn't.
It seemed as fsck had done something wrong on hda1... very strange.

I had to install Windows 2000 from scratch again.

Then, two days later I got the same error. Debian startup didn't work, I
was told again to run fsck manually. Just to check, I rebooted and started
Windows 2000: it worked fine. I run then fsck on /dev/hda3, rebooted, and
hda1 was broken again (hda3 was clean).

It seems clear to me that the fsck utility does something on hda1, even
when run on hda3. Is it a known bug?

<snip>
That cannot be due to a bug in fsck, because if it opens hda3 then it only has access to that partition and not to hda1. I think the problem is at a lower level. I'm pretty sure that changing the filesystem type will *not* fix it.
The LBA mode which has been the normal mode of addressing IDE disks for some years uses 28 address bits and can only address 128 GiB (2^28 sectors * 512 bytes/sector). There is a newer mode called LBA-48 that uses 48 address bits, but older kernels and IDE interfaces do not support it. I suspect that at some level disk addresses beyond 128 GiB are wrapping around so that when the kernel intends to write to the last part of the disk (the second half of hda3) it actually writes to the first part of the disk (hda1).
Since you say it's a "new PC" I would imagine that the IDE interface is compatible with LBA-48, so I would check your kernel version. LBA-48 was added to the various IDE drivers during the 2.4 series so if you're using the 2.2 kernel you must upgrade. 

Ben.
Reply to: