Bug 515

Summary:

BindAddress and -b not working

Product:

Portable OpenSSH

Reporter:

Craig <murple>

Component:

ssh

Assignee:

OpenSSH Bugzilla mailing list <openssh-bugs>

Status:

CLOSED FIXED

Severity:

normal

Priority:

Version:

-current

Hardware:

UltraSPARC

OS:

Solaris

Attachments:

Description	Flags
Comment out AI_PASSIVE from sshconnect.c	none
Try to fix AI_PASSIVE support	none
Fixed patch	none

Description Craig 2003-03-21 15:01:25 AEDT

I'm trying to install the OpenSSH3.4p1 on an UltraSparc running SunOS 5.6. The
program compiled and runs fine with one problem... The system has two IP
addresses, one on a secure network and one facing the outside world. I need to
have the ssh client use one of these addresses for its outbound connections.
With the old SSH 1.2.x I was able to use the SourceAddress directive in
ssh_config to do this. Using OpenSSH3.4p1, neither the BindAddress directive in
ssh_config nor the -b commandline switch seem to function on Solaris.

A search of Bugzilla found that this was a known problem on Solaris with an
older version of OpenSSH about one year ago, but there was no mention of a fix
or workaround. It is still not working for me. This is something I need
functional for work pretty quickly. Please advise if there is a known fix.

Comment 1 Darren Tucker 2003-03-23 20:59:59 AEDT

Confirmed with 3.5p1 on Solaris 2.6.  Does not occur on Solaris 8 (ie -b works 
as expected).

I note that Solaris 8 has a real getaddrinfo whereas 2.6 does not and uses the 
compatibility one from openbsd-compat, so there may be something funny 
going on there.

Comment 2 Darren Tucker 2003-03-24 01:07:25 AEDT

Created attachment 253 [details]
Comment out AI_PASSIVE from sshconnect.c

It looks like getaddrinfo() in openbsd-compat doesn't do the right thing when
AI_PASSIVE is set.  It will always return a null address even when an address
is specified (either via ssh -b or sshd's ListenAddress).

The patch fixes the ssh -b thing (should AI_PASSIVE be set on a socket that's
not going to be listening?) but fake-getaddrinfo seems to need some work for
the sshd ListenAddress case too.

Comment 3 Damien Miller 2003-03-24 10:20:29 AEDT

Created attachment 254 [details]
Try to fix AI_PASSIVE support

This (untested) patch may help to fix fake-getaddrinfo's AI_PASSIVE support

Comment 4 Darren Tucker 2003-03-24 12:05:34 AEDT

Don't have a copy of the CVS tree on my Solaris 2.6 machine.  Applied patch to 
3.5p1 (minor reject of "u_long addr;", easily fixed).

Apart from a missing semicolon at the end of "addr = htonl(0x00000000)" this 
works for ssh -b and sshd -o ListenAddress.

Comment 5 Damien Miller 2003-03-24 13:34:50 AEDT

Created attachment 255 [details]
Fixed patch

Here is a corrected patch. It will be applied to -current. Users of 3.6p1 (and
earlier) will want to apply this if using BindAddress.

Comment 6 Damien Miller 2003-03-24 13:36:19 AEDT

Applied to -current

Comment 7 Damien Miller 2004-04-14 12:24:18 AEST

Mass change of RESOLVED bugs to CLOSED