[6510] in Kerberos
Re: SIGINT in the rlogind child
daemon@ATHENA.MIT.EDU (Sam Hartman)
Sun Jan 21 15:41:30 1996
To: schwartz@galapagos.cse.psu.edu (Scott Schwartz)
Cc: kerberos@MIT.EDU
From: hartmans@MIT.EDU (Sam Hartman)
Date: 21 Jan 1996 15:28:52 -0500
In-Reply-To: schwartz@galapagos.cse.psu.edu's message of 21 Jan 1996 07:07:49 GMT
>>>>> "Scott" == Scott Schwartz <schwartz@galapagos.cse.psu.edu> writes:
Scott> hartmans@MIT.EDU (Sam Hartman) writes: | Gary> SIGINT is
Scott> ignored in the client, but the ^C is sent to the | Gary>
Scott> server, where it kills the child. This kills the session,
Scott> | Gary> and the client reports "Connection Closed".
Scott> | Gary> So I got into krlogind.c and in the child , told it
Scott> to | Gary> ignore any SIGINT signals. | Gary> The
Scott> question: Any reason why this should not be done?
Scott> Because that's treating the symptoms, not the real bug.
Scott> The real bug is that krlogind, being a daemon, should not
Scott> have a controlling tty. Ditto for login.krb5. But the
Scott> kerberos code is broken in such a way that they share the
Scott> controlling tty with the one for shell that krlogind
Scott> creates for you. That's why it is getting the signals in
Scott> the first place.
The krlogind in the current source tree is fixed so that it
doesn't get a controlling terminal. I am unconvinced that login.krb5
shouldn't get a controlling terminal; while I could do what you say, I
would argue that the lerker is a daemon started on the session's
behalf (AKA leave or others) and it does want to get the SIGHUP from
the tty.
Why do you see a problem associated with the login.krb5 having
the session as a tty?
--Sam
