Debian Bug report logs - #569018
openssh-client: recreate mux control socket

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: "Andrew O. Shadoura" <bugzilla@tut.by>

To: submit@bugs.debian.org

Subject: openssh-client: recreate mux control socket

Date: Tue, 9 Feb 2010 14:45:41 +0200

[Message part 1 (text/plain, inline)]

Package: openssh-client
Version: 1:5.3p1-1

Hello.

This patch enables re-creation of the control socket for multiplexing. Without 
it, if ssh dies, it leaves socket in place, and newly started ssh cannot bind 
to it.

-- 
WBR, Andrew

[mux.c.patch (text/x-patch, inline)]

--- openssh-5.3p1.orig/mux.c	2010-02-09 14:35:32.000000000 +0200
+++ openssh-5.3p1/mux.c	2010-02-09 11:03:25.000000000 +0200
@@ -143,16 +143,23 @@
 
 	old_umask = umask(0177);
 	if (bind(muxserver_sock, (struct sockaddr *)&addr, addr_len) == -1) {
-		muxserver_sock = -1;
 		if (errno == EINVAL || errno == EADDRINUSE) {
 			error("ControlSocket %s already exists, "
-			    "disabling multiplexing", options.control_path);
-			close(muxserver_sock);
-			muxserver_sock = -1;
-			xfree(options.control_path);
-			options.control_path = NULL;
-			options.control_master = SSHCTL_MASTER_NO;
-			return;
+			    "recreating socket", options.control_path);
+			unlink(options.control_path);
+			if (bind(muxserver_sock, (struct sockaddr *)&addr, addr_len) == -1) {
+				if (errno == EINVAL || errno == EADDRINUSE) {
+					error("ControlSocket %s already exists, "
+							"disabling multiplexing", options.control_path);
+					close(muxserver_sock);
+					muxserver_sock = -1;
+					xfree(options.control_path);
+					options.control_path = NULL;
+					options.control_master = SSHCTL_MASTER_NO;
+					return;
+				} else
+					fatal("%s bind(): %s", __func__, strerror(errno));
+			}
 		} else
 			fatal("%s bind(): %s", __func__, strerror(errno));
 	}

[signature.asc (application/pgp-signature, inline)]

Message #14 received at 569018@bugs.debian.org (full text, mbox, reply):

From: Colin Watson <cjwatson@debian.org>

To: "Andrew O. Shadoura" <bugzilla@tut.by>, 569018@bugs.debian.org

Subject: Re: Bug#569018: openssh-client: recreate mux control socket

Date: Tue, 9 Feb 2010 16:22:45 +0000

On Tue, Feb 09, 2010 at 02:45:41PM +0200, Andrew O. Shadoura wrote:
> This patch enables re-creation of the control socket for multiplexing. Without 
> it, if ssh dies, it leaves socket in place, and newly started ssh cannot bind 
> to it.

Thanks for your patch.  This has certainly long been an irritating
property of multiplexing.

It seems to me that this patch makes it a bit too easy to accidentally
trash an existing master just by setting ControlMaster to "yes", though.
Wouldn't it be better to do something like pinging the master to find
out if it's dead before recreating the socket, and disable multiplexing
as before if it's still alive?  That way there'd be no question about
whether it's appropriate to recreate the socket.  You could do the same
kind of thing as 'ssh -O check' does.

(Also, it would be best for this to go upstream to bugzilla.mindrot.org.
I can proxy for you if you like, but sometimes it's both better and
quicker for upstream to be able to discuss the patch directly with the
author rather than via an intermediary.)

Regards,

-- 
Colin Watson                                       [cjwatson@debian.org]

Message #19 received at 569018@bugs.debian.org (full text, mbox, reply):

From: "Andrew O. Shadoura" <bugzilla@tut.by>

To: Colin Watson <cjwatson@debian.org>

Cc: 569018@bugs.debian.org

Subject: Re: Bug#569018: openssh-client: recreate mux control socket

Date: Tue, 9 Feb 2010 18:34:50 +0200

[Message part 1 (text/plain, inline)]

On Tuesday 09 February 2010 18:22:45 Colin Watson wrote:
> On Tue, Feb 09, 2010 at 02:45:41PM +0200, Andrew O. Shadoura wrote:
> > This patch enables re-creation of the control socket for multiplexing.
> > Without it, if ssh dies, it leaves socket in place, and newly started ssh
> > cannot bind to it.

> Thanks for your patch.  This has certainly long been an irritating
> property of multiplexing.

> It seems to me that this patch makes it a bit too easy to accidentally
> trash an existing master just by setting ControlMaster to "yes", though.
> Wouldn't it be better to do something like pinging the master to find
> out if it's dead before recreating the socket, and disable multiplexing
> as before if it's still alive?  That way there'd be no question about
> whether it's appropriate to recreate the socket.  You could do the same
> kind of thing as 'ssh -O check' does.

Doesn't it try to connect to the socket first? I didn't dig into source deeply 
enough, but at least it says on stderr that "connection refused".

> (Also, it would be best for this to go upstream to bugzilla.mindrot.org.
> I can proxy for you if you like, but sometimes it's both better and
> quicker for upstream to be able to discuss the patch directly with the
> author rather than via an intermediary.)

Ok, will try.

-- 
WBR, Andrew

[signature.asc (application/pgp-signature, inline)]

Message #24 received at 569018@bugs.debian.org (full text, mbox, reply):

From: "Andrew O. Shadoura" <bugzilla@tut.by>

To: Colin Watson <cjwatson@debian.org>

Cc: 569018@bugs.debian.org

Subject: Re: Bug#569018: openssh-client: recreate mux control socket

Date: Mon, 29 Mar 2010 18:06:23 +0300

[Message part 1 (text/plain, inline)]

Hello Colin.

On Tuesday 09 February 2010 18:22:45 Colin Watson wrote:
> (Also, it would be best for this to go upstream to bugzilla.mindrot.org.
> I can proxy for you if you like, but sometimes it's both better and
> quicker for upstream to be able to discuss the patch directly with the
> author rather than via an intermediary.)

I've posted a bug report there, still got no reply from them. Should I ping 
someone individually there? :)

-- 
WBR, Andrew

[signature.asc (application/pgp-signature, inline)]

Message #31 received at 569018@bugs.debian.org (full text, mbox, reply):

From: "Andrew O. Shadoura" <bugzilla@tut.by>

To: 569018@bugs.debian.org

Cc: control@bugs.debian.org

Subject: openssh-client: recreate mux control socket

Date: Fri, 24 Sep 2010 17:32:06 +0300

[Message part 1 (text/plain, inline)]

tags 569018 + fixed-upstream
--

Hello.

The bug is fixed by recent upstream commit:

--- Comment #1 from Damien Miller <djm@mindrot.org> 2010-09-24 22:20:14 EST
OpenSSH -current now atomically establishes a listening mux master
socket at the specified location, so a client encountering a mux master
socket that refuses connection can safely delete it and process to
establish its own. This change will be released in OpenSSH 5.7, due in
a couple of months.
---

Here is it:
http://www.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/mux.c.diff?r1=1.21;r2=1.22;f=h

-- 
WBR, Andrew

[signature.asc (application/pgp-signature, inline)]

Send a report that this bug log contains spam.