[1716] in Kerberos_V5_Development
Re: Solaris UDP connect() lossage
daemon@ATHENA.MIT.EDU (Marc Horowitz)
Fri Sep 6 13:04:49 1996
To: Tom Yu <tlyu@MIT.EDU>
Cc: krbcore@MIT.EDU
In-Reply-To: Your message of "Fri, 06 Sep 1996 05:56:53 EDT."
<9609060956.AA22861@tesla-coil.MIT.EDU>
Date: Fri, 06 Sep 1996 13:04:41 EDT
From: Marc Horowitz <marc@MIT.EDU>
In message <9609060956.AA22861@tesla-coil.MIT.EDU>, Tom Yu <tlyu@MIT.EDU> writes:
>> Under Solaris, if you connect() a UDP socket, then you don't get the
>> BSD behaviour of having send() and recv() return ECONNREFUSED when
>> called after a send() ends up sending to a nonlistening port. Do we
>> want to work around this or not?
I think that as a free software group, there's only so much effort we
can put into working around braindamage in various vendor operating
systems. I'd rather spend the time working on some feature or fixing
some bug that users of the other n-1 operating systems will find
helpful.
>> I first noticed it in login.krb5 when it's trying to convert the
>> tickets to krb4 tickets and was hanging because krb524d wasn't runnign
>> on fmult.mit.edu.
Do you just want to put in a timeout? This seems reasonable, as it is
pretty straightforward, and will help all platforms when the host is
down. I wouldn't go out of my way to figure out a way to get solaris
not to pause for n seconds if the daemon is down, though.
Marc
