[OpenAFS-devel] Re: Windows error ERROR_REQ_NOT_ACCEP

[Ben Creech]

To refresh your memories, this past semester, NC State experienced a 
problem wherein the OpenAFS Client for Windows would go into a zombie state 
and stop responding to traffic on the SMB interface until restarted.  With 
some help, we tracked down what we think was the bug in the code, and 
installed a patched binary in our labs in order to test it.

We ran a 200-machine experiment in our labs over the past three weeks.  Of 
those, 150 or so were running a patched afsd_service.exe.  Out of the 150, 
there was a single recurrence of the KTC_NOCM problem out of hundreds of 
logins.  Unfortunately, we weren't running with TraceOption set, so I can't 
look at the logs and figure out exactly what caused it.  I can say that 
Weird Things were going on with that machine at that time (someone's 
roaming Novell profile was repeatedly failing to copy back to the server).

However, there were about 50 machines that were running unpatched client. 
Of those, there were 4 occurrences of the problem.  This would indicate 
that we fixed one bug, although another may well exist.

I might point out that, while 5 occurrences of a problem over one month and 
50 machines does not seem like a very big deal, we had a much higher error 
rate over the spring semester, when our labs saw heavier usage.

To further investigate, we are currently setting all our lab machines with 
TraceOption=1 so that we can see exactly what goes wrong if and when it 
does.  Either way, looking at the code (smb_Listener in smb.c), it's fairly 
obviously that we should do the following:

*** src/WINNT/afsd/smb.c-1.2.9a Wed Jun 25 14:13:36 2003
--- src/WINNT/afsd/smb.c        Wed Jun 25 14:14:00 2003
***************
*** 5902,5907 ****
--- 5902,5908 ----

        while (1) {
                memset(ncbp, 0, sizeof(NCB));
+               flags = 0;
  #ifdef DJGPP
               /* terminate if shutdown flag is set */
               if (smbShutdownFlag == 1)

Thanks,
Ben Creech
NCSU ITECS