AIO, SMB2 and Samba 3.6

[Volker Lendecke]

On Wed, Feb 22, 2012 at 10:16:14PM +1100, Andrew Bartlett wrote:
> On Wed, 2012-02-22 at 10:54 +0100, Volker Lendecke wrote:
> > On Wed, Feb 22, 2012 at 03:56:52PM +1100, Andrew Bartlett wrote:
> > > I've now reproduced this on the NAS with a level 10 log, using
> > > smbclient4 from master:
> > > 
> > > bin/smbclient //192.168.1.167/media -Uuser%password
> > > smb: \> put /data/kvm/cdroms/FreeBSD-8.2-RELEASE-amd64-bootonly.iso
> > > freebsd7.iso
> > > Error writing file: socket_error
> > > 
> > > It seems to me that the underlying AIO layer isn't returning promptly or
> > > correctly.  It appears that after the socket error, smbclient still
> > > tries to close the file, and this then triggers finally an AIO error. 
> > 
> > The reference to freebsd makes me wonder if the NAS box
> > might be FreeBSD based. 
> 
> Sorry for the red herring.  That was just a 50MB file I had lying
> around.  It is Linux 2.6.37.6.RNx86_64.2.2 and glibc 2.7-18lenny7.
> 
> > I think I've seen problems with
> > Samba setting up the RT signal we choose for AIO. It should
> > have disabled using aio. Can you check further up in the log
> > whether that might be the case? 
> 
> This certainly matches the symptoms: what would the message be in that
> case?  The log is the full log from the point of the share connection. 
> 
> > Also, you might try the
> > aio_fork and aio_pthread modules.
> 
> Indeed, given the well known performance issues with glibc AIO
> aio_pthread has already been recommended.  This task was just to first
> sort out (if possible) why the default aio didn't work at all.

Can you run strace -ttT -o /tmp/smbd.out -f smbd -i

and run the test against that?

Thanks,

Volker

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen
http://www.sernet.de, mailto:kontakt at sernet.de