[linux-cifs-client] CIFS mounted files are corrupt when passed to apache web clients

[Jeff Layton]

On Wed, 28 Nov 2007 13:37:01 -0500
Jeff Layton <jlayton at redhat.com> wrote:

> On Wed, 28 Nov 2007 10:06:54 -0800 (PST)
> JNaujok <Jester at devista.org> wrote:
> 
> > 
> > 
> > Jeremy Allison wrote:
> > > 
> > > Get wireshark traces and post them please.
> > > 
> > 
> > Attached are the source and destination filtered files from
> > wireshark. Sorry, I don't know the filter well enough to get them
> > both in the same file.
> > 
> > 192.168.0.49 is the Windows machine, 192.168.0.230 is the apache
> > box.
> > 
> > Jeff
> > 
> > http://www.nabble.com/file/p13998819/Source Source  
> > http://www.nabble.com/file/p13998819/Destination Destination 
> > 
> 
> The captures show that after 128B.txt is opened that the client tries
> to read 4k from it and gets 128 bytes back. It then closes the file.
> 
> After opening 256B.txt, it just does an immediate close on the file
> (never actually attempting a read). That may be because the file data
> is cached (or the client thinks it is), or some other bug.
> 
> strace of apache would be the best thing at this point so we can see
> what it might be doing differently than "standard" user access.
> 

Actually...an even easier quick test than stracing might be to try
disabling sendfile() in apache:

http://httpd.apache.org/docs/2.0/mod/core.html#enablesendfile

...its just a hunch, but I suspect that that might be the difference
between apache and "normal" users are doing. If that works around the
problem, then we can start having a close look at what's broken there
(and maybe come up with a simpler reproducer).

Thanks,
-- 
Jeff Layton <jlayton at redhat.com>