[Unionfs] 2.3.2 serious file corruption

[Paul Bender]

bkauler at goosee.com wrote:
> Guys,
> I have a serious problem with unionfs 2.3.2 and kernel 2.6.25.
> It's a vanilla kernel except for squashfs patch.
> 
> After switch_root from the initramfs, Puppy Linux has a unionfs on '/'
> and two or more layers. The top is a tmpfs in ram, below that a
> squashfs file with all the Puppy files.
> 
> At bootup, if a Smartlink module is loaded, a bootup script has this
> code:
> 
>  echo 'alias char-major-242 slamr' >> /etc/modprobe.conf
>  echo 'install slamr modprobe --ignore-install ungrab-winmodem ; modprobe
> --ignore-install slamr' >> /etc/modprobe.conf
>  echo 'alias char-major-243 slusb' >> /etc/modprobe.conf
> 
> Ok, /etc/modprobe.conf is on the second layer, so unionfs should
> copy it up then append to it. But it goes horribly wrong.
> The first echo writes before the file is copied up, the next echo
> appends after the copyup. The file ends up a horrible mess.
> 
> The original /etc/modprobe.conf:
> 
> alias binfmt-204 binfmt_aout
> alias binfmt-263 binfmt_aout
> alias binfmt-264 binfmt_aout
> ....
> remove sdhci /sbin/modprobe -r --ignore-remove sdhci ; /sbin/modprobe -r
> mmc_block
> install tifm_7xx1 /sbin/modprobe --ignore-install tifm_7xx1 ;
> /sbin/modprobe mmc_block ; /sbin/modprobe tifm_sd
> 
> After the echoes:
> 
> alias char-major-242 slamr
> t
> alias binfmt-263 binfmt_aout
> alias binfmt-264 binfmt_aout
> .....
> remove sdhci /sbin/modprobe -r --ignore-remove sdhci ; /sbin/modprobe -r
> mmc_block
> install tifm_7xx1 /sbin/modprobe --ignore-install tifm_7xx1 ;
> /sbin/modprobe mmc_block ; /sbin/modprobe tifm_sd
> install slamr modprobe --ignore-install ungrab-winmodem ; modprobe
> --ignore-install slamr
> alias char-major-243 slusb
> 
> You see, a disaster. I tested with 'touch /etc/modprobe.conf' before
> doing the echoes, and that fixes it. But it doesn't solve the
> fundamental problem.
> 
> I see that 2.3.3 release announcement has a comment about some bug
> with delayed copyup. But, does it fix the problem that I have described
> above?
> 
> If not, would you mind providing a patch for 2.6.25 without the
> copyup delay "optimisation"?
> 
> Regards,
> Barry Kauler

I posted what appears to be the same problem to the mail list recently 
and a fix was provided: 
<http://www.fsl.cs.sunysb.edu/pipermail/unionfs/2008-April/005814.html>.

Also, the problem is fixed in 2.3.3