RE: [squid-users] NTLM - invalid callback data error

tis 2003-01-14 klockan 09.41 skrev Jay Turner:

> One thought we had here as to what is happening is that [in
> AuthenticateNTLMHandleReply()] the callback data check [cbdataValid()] is
> looking at the callback memory area and seeing that it is not valid and
> exiting gracefully citing the invalid callback data error. This as Henrik
> said is correct and the way such an error should be handled.

As both Robert and me have said: This is most certainly due to aborted
requests while the NTLM authentication is being negotiated between the
browser and Squid.

1. Browser contacts Squid and starts NTLM negotiation.
2. While the ntlm helper is processing a authentication step the browser
aborts, making the callback data invalid when the helper returns the
result.

If you do not like seeing the message logged, change the debug level of
the message to 2.

> I guess my question is why is the callback data not valid? Is it due to the
> fact that we are requesting small objects at such a fast rate and then not
> allowing them to complete downloading (the equivalent of aborting the
> request?) that the memory area is released by squid, however the NTLM helper
> does not know this and still continues to attempt to access it only to find
> that the memory area has been freed already?

This is by design of Squid. See the programmers guide where the use of
cbdata is fairly well explained.

> I was able to load test this a little and have the error occur a number of
> times in a row (roughly 30) within a minute, fortunately this didn't crash
> this server that I have been testing on. The previous crashes I reported may
> have been due to the fact that my test bed was a low spec'd Celeron 500 with
> 256Mb RAM as opposed to this PIII 700 with 768Mb RAM.

It should not crash Squid. If you can get this to crash Squid then
please get a stacktrace and register a bug report.

Regards
Henrik
Received on Tue Jan 14 2003 - 02:58:03 MST