[OpenAFS] Re: 1.3.70 comments?

[Douglas E. Engert]

Horst Birthelmer wrote:

> 
> On Aug 22, 2004, at 7:38 AM, Derrick J Brashear wrote:
> 
>> On Sun, 22 Aug 2004, Derrick J Brashear wrote:
>>
>>> On Fri, 20 Aug 2004, Derrick J Brashear wrote:
>>>
>>>> Only idea I can come up with is that I somehow screwed up rx_Init; 
>>>> If so, checking out the cvs head by date (2 weeks ago) should work. 
>>>> I reproduced the problem, but sadly my AIX machine is 20 miles away 
>>>> and didn't bother to come back up.
>>>
>>>
>>> Nope. Horst did testing and found it's this change:
>>> diff -u -r1.8 -r1.10
>>> --- rxkad.p.h   15 Jul 2003 23:16:42 -0000      1.8
>>> +++ rxkad.p.h   3 Apr 2004 07:59:53 -0000       1.10
>>> @@ -16,7 +16,7 @@
>>>                /* no ticket good for longer than 30 days */
>>> #define MAXKTCTICKETLIFETIME (30*24*3600)
>>> #define MINKTCTICKETLEN              32
>>> -#define        MAXKTCTICKETLEN       344
>>> +#define        MAXKTCTICKETLEN       12000 /* was 344 */
>>> #define        MAXKTCNAMELEN         64        /* name & inst should 
>>> be 256 */
>>> #define MAXKTCREALMLEN       64        /* should be 256 */
>>> #define KTC_TIME_UNCERTAINTY (15*60)   /* max skew bet. machines' 
>>> clocks */
>>>
>>> i don't know what the problem is or how large we can go if that 
>>> problem can be addressed, but currently on aix we cannot go to 12000.
>>

The large ticket size was set to match the size of a ticket that Windows
could send with a PAC. You could try building with a smaller size.

When you say "on aix we cannot go to 12000" What do you mean?  is it failing
somewhere?  The actual size of the ticket is what is sent on the wire,
and transfered by pioctl.

Is this a problem with LWP not alloacting a large enough stack to hold
automatic data that might have allowed for a max ticket?
lwp.h has #define AFS_LWP_MINSTACKSIZE    (48 * 1024)
Is this two small?


>>
>> and, that's the klog problem, i don't know if it's related to the tab 
>> completion problem.
>>
> 
> I think it has to be since not the ticket insertion goes wrong but some 
> calls afterwards are starting with bogus data... (sometimes 
> uninitialized mutexes or things like that)
> So one of this calls can be a tab which lists the current directory or 
> whatever ...
> 
> Horst
> 
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info
> 
> 
> 

-- 

  Douglas E. Engert  <DEEngert@anl.gov>
  Argonne National Laboratory
  9700 South Cass Avenue
  Argonne, Illinois  60439
  (630) 252-5444