If EXT4, don't use 3.5.7 / 3.6.2 kernels [OBSOLETE]

aCOSwt · Last edited by aCOSwt on Wed Oct 31, 2012 8:51 am; edited 1 time in total

ext4 data corruption regression
_________________

gerard27 · Posted: Wed Oct 24, 2012 5:47 pm Post subject:

Thanks aCOSwt.
I installed 3.5.7 two days ago.
No problems so far but to be on the safe side I switched back to 3.4.9.
Gerard.
_________________
To install Gentoo I use sysrescuecd.Based on Gentoo,has firefox to browse Gentoo docs and mc to browse (and edit) files.
The same disk can be used for 32 and 64 bit installs.
You can follow the Handbook verbatim.
http://www.sysresccd.org/Download

bandreabis · Posted: Thu Oct 25, 2012 6:17 am Post subject:

Affected kernels have been Hard Masked!
At least yesterday they were.

aCOSwt · Posted: Thu Oct 25, 2012 6:41 am Post subject:

Could be some poorly tested specific option's fault :

c00l.wave · Apprentice Joined: 24 Aug 2003 Posts: 268

Note that there's a null-pointer dereference occuring when large files are being deleted from ext4 filesystems in 3.4.9, which was my main reason to upgrade to and stay at 3.5.7 on the systems I maintain (I've actually hit the null-pointer bug when moving backups of many tens of gigabytes across disks). It's much more probable to hit that bug than hitting the journal bug that led into masking panics - for what I read, the current bug is considered to occur only under very specific circumstances that require having created and mounted the filesystem with uncommon options.

The null-pointer bug seems to have been missed by Gentoo devs but now has a bug report as well (at least it reads like the same one I encountered).

So I would add "don't use 3.4.9 either" or "but run 3.5.7/3.6.2 anyway if you run default filesystems" (without warranty, you'd better have backups either way).
_________________
nohup nice -n -20 cp /dev/urandom /dev/null &

platojones · Posted: Sat Oct 27, 2012 12:51 pm Post subject:

Mask has been lifted for 3.6.2.

ppurka · Advocate Joined: 26 Dec 2004 Posts: 3256

platojones · Posted: Sat Oct 27, 2012 11:37 pm Post subject:

Tony0945 · Posted: Fri Nov 02, 2012 8:15 am Post subject:

As a user, I'm thoroughly confused. I run a stable system as much as possible. I was on 3.4.9, a routine update installed 3.5.7. I rebuilt the kernel and removed 3.4.9, then 3.5.7 was masked and I re-emerged 3.4.9, rebuilt the kernel, and removed 3.5.7, now emerge -auvND world wants to re-install 3.5.7

I don't understand the mount option problem. My /etc/fstab has the following line:

c00l.wave · Apprentice Joined: 24 Aug 2003 Posts: 268

I don't think the 3.4.9 bug causes random data loss - loss should happen only to files that were being written/deleted at that time. After the null-pointer dereference occured, I found the backup files I copied to be randomly 0 byte size on either target or source and one file cut off. Having upgraded to 3.5.7 I compared file sizes and copied the larger file to the destination drive but I haven't tried if the data is still ok (those were only old backups moved to make space for newer ones). I'm not a kernel developer but the effect of the 3.4.9 bug does not appear to be worse than simply cutting power while writing to disk - the journal will revert any pending transactions and fsck will check for structural conistency.

If you don't remember having set any fancy options for your ext4 partitions, I wouldn't mind the bug in 3.5.7. However, it would be much more severe if it stroke. It's your own choice but I stayed with 3.5.7 so far.

To be completely safe, you could also choose a kernel older than 3.4. I wouldn't want to "downgrade" to ext3, though.

(Your "noatime" mount option is nothing special, it just disables the usually unnecessary "access time" logging.)
_________________
nohup nice -n -20 cp /dev/urandom /dev/null &

aCOSwt · Posted: Fri Nov 02, 2012 8:51 am Post subject:

Tony0945 · Posted: Fri Nov 02, 2012 8:35 pm Post subject:

Thanks for the prompt response.