Re: rlogind on Solaris

[Alain Magloire]

> 
> 
> Quoting Alain Magloire:
> > P.S: Did you not say you had access to Solaris box?
> > I'm trying to fix bugs to the server on solaris, they use
> > STREAM handling instead of the usual BAD ptys.  But do
> > not have a solaris box to actually test.
> <..snip..>
> > It was the rlogind, actually any r* server seems they pretty
> > much the same code.  Joachim was debugging it but I could
> > not be of any help(No Solaris machine).  IIRC the problem
> > was doing login.
> > 
> > - Maybe /bin/login was expecting some UTMPX entry.
> > - The STREAM handling that I have put in is the culprit
> >   see libinetutils/{fork,open}pty.c
> > 
> > When you contacted rlogind, the server just went away
> > or /bin/login refuse to log you.
> 
> Finally fixed :^). That was a tough one. The problem actually was
> in login_tty(): it used TIOCSCTTY on slave pty, but Solaris lacks
> this ioctl. Thus the child rlogind process was unable to set its
> controlling tty, /bin/login was receiving EOF immediately upon reading
> from its stdin and was bailing out.
> 
> Another problem was that "pckt" module was pushed onto the master
> side of a pseudo terminal by openpty(), but rlogind was trying
> to read from master side using read() call, which does not work,
> since by default read() operates in byte-stream mode. It could have
> been fixed by using I_SRDOPT ioctl on master fd, but I preferred
> to use getmsg() and provide a wrapper for systems which lack it.
> The stuff was tested on Solaris/x86 (SunOS 5.7) and GNU/Linux. 
> 
> Anyway, rlogind.c is rewritten completely. Hope I'll be able to make a
> library out of it for using with other r* servers.

That is absolute great news,  Nice work!
I'll be busy filtering my mailbox reaching ~3000(Too bad mailutils/sieve
is not operationnal 8).
I Will take a closer look later.

--
alain