[Sheflug] Heads up -- 2.4.15 could cause ext2 problems

["Chris J/#6" <xxxxx [at] xxxxxxxxxxxx>]

Hmm...I think I could do with taking my finger off the pulse at some point 
... I'm posting a lot today.

There's a post in LKML that reads as follows:

---------------------------------------------------------------------------
From: Russell King <xxxxx [at] arm.linux.org.uk>
Date: Fri Nov 23 20:37:07 GMT 2001
To: linux-kernel [at] vger.kernel.org
Subject: 2.4.15: FS corruption on EXT2

Hi,

I've just been testing 2.4.15 on my NetWinder (which was running 2.4.15-pre5
quite happily - its built several kernels and been through a fair number of
reboot cycles with that version).

With 2.4.15, I can now 100% reproduce every time filesystem corruption of
an ext2 filesystem, specifically in /var/lock/subsys/.

/var/lock/subsys contains a load of 0 byte files, one for each daemon that
is started by the redhat-like boot scripts.  On shutdown they're removed.

After a second boot with 2.4.15, all the names are still present in the
directory, rm complains with:

        rm: cannot remove `/var/lock/subsys/xyz': Input/output error

stracing rm reveals that lstat of a file in /var/lock/subsys/ returns -EIO.
Trying to get a directory listing results in every single file giving an
input/output error.

This is just a heads up - several people are already looking into it.

--
Russell King (rmk [at] arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

---------------------------------------------------------------------------

Another one is from Chris Siebenmann <cks [at] utcc.utoronto.ca>, with the 
subject: "2.4.15 problem: deleted inodes still present in ext2 directories 
after shutdown" which details the same problem.

To follow-on from this, Al Viro has said that -pre9 broke inode.c ... here's 
his post:

---------------------------------------------------------------------------
From: Alexander Viro <xxxxx [at] math.psu.edu>
Date: Fri Nov 23 21:24:27 GMT 2001
To: linux-kernel [at] vger.kernel.org
cc: Linus Torvalds <torvalds [at] transmeta.com>,
+   Marcelo Tosatti <marcelo [at] conectiva.com.br>
Subject: 2.4.15-pre9 breakage (inode.c)

        Sigh...  Supposed fix to problems with stale inodes was completely
broken.

        What we need is "if we are doing last iput() on fs that is getting
shut, sync it and don't leave it in cache".  And yes, we have a similar
path in iput().  Similar, but not quite the same.

        Fix is
* new fs flag: "MS_ACTIVE".
* set after normal ->read_super().
* reset after we are done with fsync_super() in kill_super().
* iput() checking that and if it's set - doing write_inode_now() and kicking
it out of hash.

I'll send patch in ~10 minutes.

----------------------------------------------------------------------------

Expect a 2.4.16-pre patch or a 2.4.16 release in the not too distant future :)

Anyone on 2.4.15 want the patch when I see it arrive, shout, as I'm going to 
2.4.15 myself to test but I'm going to apply this patch first once Al's 
written it and it's arrived in my news spool (LKML much easier to follow in 
fa.linux.kernel :-)

Chris...


-- 
\ Chris Johnson           \
 \ cej [at] nccnet.co.uk        \
  \ www.nccnet.co.uk/~cej/  ~-----------------------------------------,
   \ Redclaw chat - http://redclaw.org.uk - telnet redclaw.org.uk 2000 \____


___________________________________________________________________

Sheffield Linux User's Group - http://www.sheflug.co.uk . 
To unsubscribe from this list send mail to 
shef-lug-request@list.sheflug.org.uk with the word
"unsubscribe" in the body of the message. 

  GNU the choice of a complete generation.