Re: Reauthentication implementation flaw due to EINTR

[Carl Fredrik Hammar]

On Tue, Dec 29, 2009 at 11:34:08PM +0100, Samuel Thibault wrote:
> Carl Fredrik Hammar, le Tue 29 Dec 2009 23:29:35 +0100, a écrit :
> > On Tue, Dec 29, 2009 at 10:55:41PM +0100, Samuel Thibault wrote:
> > > Carl Fredrik Hammar, le Tue 29 Dec 2009 22:48:26 +0100, a écrit :
> > > > It's still a mystery to me why ext2fs gets an EINTR without auth
> > > > explicitly returning EINTR...
> > > 
> > > AIUI, the RPC gets aborted and ext2fs receives MACH_RCV_PORT_DIED, which
> > > _hurd_intr_rpc_mach_msg turns into EINTR.
> > 
> > Wait, isn't the receive port in this case the auth port?  This could
> > be bad...
> 
> Oops, I meant MACH_RCV_TIMED_OUT above.

Phew!  :-)

This might make sense if auth's reply is interrupted.  It is still a
bit surprising though, considering that auth_server_authenticate does
not set a timeout.

> > > BTW, reproducing the bug is really easy:
> > > 
> > > while sudo date ; do : ; done
> > 
> > OK, I can use that for the client side.  I don't really have a setup
> > for testing the auth server though.  Do you use a sub-Hurd or something?
> 
> I've never managed to find time to set up a sub-hurd, so I just debug
> with writes on the console :)

Ah, that's a classic technique.  But how do you replace the auth server?
Do you install a modified one and just reboot?  Seems kinda risky.

Regards,
  Fredrik