[linux-cifs-client] CIFS VFS: Send error in read = -12

[Günter Kukkukk]

Am Mittwoch, 4. Juni 2008 schrieb Jeff Layton:
> On Wed, 04 Jun 2008 12:26:21 -0700
> Noah Romer <nromer at arcmailtech.com> wrote:
> 
> > Jeff Layton wrote:
> > > Almost everything is at 1, but there are a few other levels. Generally
> > > I just set it 7. It can get chatty though and can bog down the machine,
> > > so I wouldn't plan to run it that way for long. Ideally, you'll want to
> > > set it to 7 just before reproducing the problem and set it back to 0
> > > just after it occurs.
> > >   
> > Ok. I cranked cifsFYI up to 7 and ran the load that had failed several 
> > times before. No dice. The app ran and completed. It did generate a 
> > healthy amount of debug data, though, including several apparently 
> > non-fatal errors: 112, 11, 32, 104 and 9. The bzip2 compressed log file 
> > is 1.1MB. That seemed a little large to post to the list, so I've put it 
> > up at http://www.eskimo.com/~klevin/messages.bz2 .
> > 
> 
> Ouch -- sounds like cranking up the debugging changed the timing in
> such a way that the problem didn't occur. Looking at the debug data
> isn't likely to be helpful.
> 
> You probably have 2 options:
> 
> You may want to see if you can determine what system call is generating
> this error (probably by stracing the application that's running).
> 
> You can then try to instrument that codepath and determine the cause
> that way...
> 

Hi Noah,

you can also use a network sniffer like wireshark and using "a ring 
buffer with multiple files" - so your sniffs aren't getting to large.
Doing so on the client might also change the timings, so you probably
won't catch the issue.

If your network environment allows, you can also use a third machine
running wireshark in promiscous mode, which won't interfere with
the timings on the two other machines.

Cheers, Günter