On Mon, 11 Aug 2003, Heiko wrote:
> I've already posted on the help-forum at ntfs-sourceforge about an NTFS bad
> sector problem. My drive (ntfs-volume) seems to freeze and only then when it
Well, if the hardware freezes then that sounds really very bad ...
> reads a cluster it has written on before in a bad way. I don't know what
> causes this bad writing : a bad physical sector/track, the last breath of my
> harddrive, maybe a too fragmented drive ?
If the hardware freeze then it's close to death (supposed it worked
earlier).
> ... but my problem started when w2k
> created some new inodes/files/cluster allocations? with writing on bad
> sectors this time needed by booting procedure (past my grub bootloader disk).
You could check the bad sectors with the 'badblocks' util (e2fsprogs
package). If it hangs then it's bad hardware (but could be driver problem
as well).
> Using the ntfsdriver I was able to mount my primary w2k-ntfs-drive from my
> secondary linux-system-drive. But any look into the root directory or the
> seeking for other subdirectories with tab or the direct path to the winnt or
> programs subdirectories would cause the same read freezing I've experienced
> under windows.
The disk freezes or the NTFS driver?
> 1) ntfs is transaction based - Is the transaction writing to clusters closed
> before the clusters are reread ?
I think so. Most of the time if one can write then he can read as well.
> 2) Could this be a pure over-fragmentation problem ?
I don't think so.
> 3) Is it a typical harddrive problem
It's relative what is typical. These issues do happen all the time.
> and is it useful to give the ntfs-driver or libs an
> emergency-cluster-bypass option ?
Well, I have code that can mark the bad sectors on NTFS. Tools should then
just consider these. The problem, if the disk freezes, then it's impossible
to collect this information in a short time (reboot, scan again, reboot,
scan again, reboot, etc)
> 4) As I'm new I'd like to ask Szaka about the programs/procedures ntfsmeta and
ntfsmeta is at http://mlf.linux.rulez.org/mlf/ezaz/ntfsmeta.gz
> ntfscat
Not written yet, but a 'ntfscat --inode <number>' util could be written in
a very short time (less than a hour or a few hours).
> he mentioned on the public board - are they part of the ntfsprogs
> under cvs ?
ntfsmeta wasn't needed, still I've debug/solved quite a lot of issues with
it. E.g. you can check how many clusters are marked bad by doing
ntfsmeta -di8 partition
and the runlist of the last entry (attr name: $Bad) shows this. A disk
without bad sectors would look something like this (but different numbers):
VCN LCN Run length
0 LCN_HOLE 61f6f9
61f6f9 LCN_ENOENT 0 (runlist end)
Doing a
ntfsmeta -a partition | egrep 'MFT record number|Filename '
you can go through all the MFT records ("files"), getting their inode
number (MFT record number) and choose what you want to save with ntfscat
--inode <number> or some other way (if they satisfy the mentioned
emergency-cluster-bypass option ;)
Szaka
|
