krb5_sendauth vs NAGLE vs DelayedAck

John Hascall john at iastate.edu
Thu Jan 15 15:49:31 EST 2009



> On Jan 15, 2009, at 09:41, John Hascall wrote:
> > I think you might want to also change krb5_sendauth() to
> > call krb5int_net_writev() directly (or yet another helper
> > function!) since krb5_sendauth() will still be doing two
> > consecutive writes where it does the two version strings:

> Yeah, I missed that, thanks.  I've just checked in another fix.

   Great.

> There are also multiple calls possible in recvauth but it's not  
> unconditionally doing two successive writes, so some rearranging of  
> code would be needed there; the hardest bit would probably be making  
> sure multiple paths through the code get tested.  If you feel like  
> tackling it, that'd be great, but I'm going to get back to my current  
> project for a bit.

   The only place I could see where a double-write could happen
   was a successful mutual auth.  So, here's my attempt at a
   minimal rewrite to handle that:

      http://john.public.iastate.edu/recvauth.c

   (basically I just made outbuf into a array[2] and some
   logic around calling krb5int_write_messages() with
   1 or 2 entries from it).

John



More information about the Kerberos mailing list