[OpenAFS-devel] Windows error ERROR_REQ_NOT_ACCEP

[Ben Creech]

I asked about this problem on openafs-info a month ago, but
unfortunately got no response, so I'm repeating the query here.

Two months ago, the IT department for NC State's College of Engineering
("we") installed the OpenAFS client 1.2.8a on all 150 or so of our Win2k
lab machines.

Since then, we've found that the AFS client will occasionally get into
an error state wherein the CM refuses to communicate with Windows. 
Attempts to mapping drives result in ERROR_REQ_NOT_ACCEP (win32 error
71), which translates to:
"No more connections can be made to this remote computer at this time
because there are already as many connections as the computer can
accept."

Also, when calling ktc_ForgetAllTokens we're getting KTC_NOCM.  A quick
persusal of the source shows that this is probably caused by
CreateFileEx failing on \\machine-afs\all\_._IOCTL_AFS_._.  I expect
that if I were to stick a GetLastError in there, I'd see the same win32
error code.

This problem persists until the service is restarted.  As such, it's
particularly bad, as it makes the machine worthless until someone
reboots it.  At the peak of the problem, we had about 1 in 5 not
working.

I've actually hacked around the error by making our login program
restart the service when it detects the error state, but this solution
is suboptimal.

Unfortunately, the error is very hard to debug as I cannot reproduce it
on demand.  There is nothing in the event logs around the time when the
CM "goes south".  It just stops working.  Over the next few days after I
get time, I'll be adding the loopback adapter to some of the computers
to see if that makes it go away.  Also, I've turned on TraceOption
although I don't know if I'll be able to make sense of the results. 
Failing that, I may try to remotely debug an instance of the service to
see if I can figure out what's going wrong.

Has anyone run into this problem and (even better) solved it?  Anyone
know what in particular I should look for when using TraceOption?

Thanks,
Ben Creech
NCSU ITECS