Re: ATTN REDHAT (was Re: dns problems with ssh,telnet,ftp on a local network)

[Jan Carlson <janc kubwa com>]

On Sun, Jan 13, 2002 at 05:13:43PM +0100, Alfredo Ferrari wrote:
> Hi,
> 
> thanks for the help, but I already have 
> 
>     order hosts,bind
> 
> inside /etc/host.conf. My setup is a "pure" RH7.2 with all updates,
> however I have the feeling I hit a really weird bug in glibc, since I
> found a thread in libc-alpha that explains exactly what the problem is, 
> that is a bad behaviour of getaddrinfo which tries to get an Ipv6 
> address as well even when a host is listed inside /etc/hosts
> (see 
> http://sources.redhat.com/ml/libc-alpha/2001-11/msg00125.html
> 
> and following...)
> 
> Now the question for RedHat: since this is really a tough problem, making 
> in practice /etc/hosts completely useless and rendering simple local 
> networks almost impossible to setup, are you going to provide some
> patched stuff and/or good suggestions how to circumvent the problem?
> 
> I found the answers of Ulrich Drepper on this issue particularly
> disappointing. Maybe he does not fully realize how it can badly break

I can appreciate what you are saying, but don't be disappointed - just
file a concise bug report in Bugzilla if you haven't already, or add
more info to an existing one.  Good bug reports in Bugzilla always get
action.

It's too easy for people to ignore bug reports by email.

BTW: I get the same results as you but only when
there is no error message that cuts short the request like

	no route to host
	network unreachable
	port is closed
	rejected

But if the DNS server CAN be reached, or there is a prompt error
message that it CANT, then the resolver quickly returns data 
from **/etc/hosts**.

(I never saw this problem before because I run a local nameserver
that covers all my private machines.)


> everything (ssh,ftp,telnet etc) on a two-three machine isolated network
> (typical home-like situation) when an external connection is not 
> available (case two of
> http://sources.redhat.com/ml/libc-alpha/2001-11/msg00153.html
> )
> 
>                    Thanks
>                 ALfredo Ferrari
> 
> 
> On Sun, 13 Jan 2002, Jan Carlson wrote:
> 
> > On Sun, Jan 13, 2002 at 03:53:23PM +0100, Alfredo Ferrari wrote:
> > > Hi
> > > 
> > > I have two computers at home which talk over a twisted ethernet cable.
> > > Till RH7.1, listing both machines inside their respective /etc/hosts
> > > files was enough to make ssh,telnet,ftp work between them.
> > > 
> > > After upgrading to RH7.2 they do not longer work (I do not even reach the 
> > > login prompt). With strace I can see that they try to reach the 
> > > nameservers listed in /etc/resolv.conf which are unreachable most of the 
> > > time (they are those of my ppp connection). Both /etc/hosts.conf and 
> > > /etc/nsswitch.conf have the proper setting to use /etc/hosts in first p
> > > lace, and ping and nfs mounts work flawlessly using the names as listed 
> > > (and resolved) in /etc/hosts.
> > > Furthermore strac'ing ssh shows that /etc/hosts is parsed, but apparently 
> > > the resolver goes on trying also the nameservers listed in 
> > > /etc/resolv.conf.
> > > 
> > > ssh,telnet,ftp'ing directly to the ip number works ok. It seems that these 
> > > services are trying to make a further identification after parsing 
> > > /etc/hosts which was not the case in the past. If I substitute 
> > > /etc/resolv.conf with an empty one everything works ok, but of course 
> > > I can no longer resolve names not in /etc/hosts when connected to the 
> > > external world.
> > 
> > If you want the kernel to use a nameserver only if matching info is
> > not in /etc/hosts, then keep the standard RH7.2 /etc/host.conf:
> > 
> > 	order hosts,bind
> > 
> > To contact foo.fun.com, you need the full "foo.fun.com" in /etc/hosts.
> > 
> > If this doesn't solve it, are you running anything besides the current
> > Red Hat RH7.2 rpms?  Are there any tarballs installed?
> > 
> > 

-- 
Jan Carlson                                 janc at kubwa dot com