Dear list,
The old Fedora, heavily patched OpenBSD nc package was just obsoleted by the nmap ncat implementation, available as the nmap-ncat subpackage. Those are mostly compatible and this change shouldn't cause much headache but please check your netcat dependant scripts or apps.
Related bug: #226187
Thanks to Michal Hlavinka for this.
-P
On Thu, Jul 19, 2012 at 12:12:16PM +0200, Petr Šabata wrote:
Dear list,
The old Fedora, heavily patched OpenBSD nc package was just obsoleted by the nmap ncat implementation, available as the nmap-ncat subpackage. Those are mostly compatible and this change shouldn't cause much headache but please check your netcat dependant scripts or apps.
Arrrgggh no, please no, no, no :-(
The crux of the problem is the word 'mostly' here. The NMap 'ncat' program doesn't support connecting to UNIX domain sockets. This functionality is *critical* for libvirt to connect to remote servers via our SSH tunnel facility. You've now broken the ability to connect to rawhide virt hosts :-(
$ virsh -c qemu+ssh://root@avocado.XXXXX.com/system error: End of file while reading data: 2012-07-19 10:49:00.415+0000: 10202: info : libvirt version: 0.9.13, package: 1.fc17 (Unknown, 2012-07-02-12:05:10, t500wlan.home.berrange.com) nc: invalid option -- 'U' Ncat: Try `--help' or man(1) ncat for more information, usage options and help. QUITTING.: Input/output error error: failed to connect to the hypervisor
Libvirt needs to be able to run the following command
# nc -U /var/run/libvirt/libvirt-sock
Regards, Daniel
2012/7/19 Daniel P. Berrange berrange@redhat.com:
Libvirt needs to be able to run the following command
# nc -U /var/run/libvirt/libvirt-sock
It could use socat:
# socat stdio /var/run/libvirt/libvirt-sock
- Thomas
On Thu, Jul 19, 2012 at 01:27:14PM +0200, Thomas Moschny wrote:
2012/7/19 Daniel P. Berrange berrange@redhat.com:
Libvirt needs to be able to run the following command
# nc -U /var/run/libvirt/libvirt-sock
It could use socat:
This isn't something that can simply be fixed in libvirt in rawhide because you need to consider cross-version compatibility. Every existing deployed version of libvirt in every distro will be unable to connect to rawhide, even if we changed libvirt in rawhide to use socat.
Regards, Daniel
On Thu, Jul 19, 2012 at 12:30:35PM +0100, Daniel P. Berrange wrote:
On Thu, Jul 19, 2012 at 01:27:14PM +0200, Thomas Moschny wrote:
2012/7/19 Daniel P. Berrange berrange@redhat.com:
Libvirt needs to be able to run the following command
# nc -U /var/run/libvirt/libvirt-sock
It could use socat:
This isn't something that can simply be fixed in libvirt in rawhide because you need to consider cross-version compatibility. Every existing deployed version of libvirt in every distro will be unable to connect to rawhide, even if we changed libvirt in rawhide to use socat.
Regards, Daniel
I believe using a tool that works the same on every system, such as the suggested socat, would be the right thing to do in the future libvirt releases.
About the current breakage -- I'd prefer adding UNIX sockets support to ncat over reviving nc. I'll see what can be done there.
-P
On Thu, Jul 19, 2012 at 12:30:35PM +0100, Daniel P. Berrange wrote:
On Thu, Jul 19, 2012 at 01:27:14PM +0200, Thomas Moschny wrote:
2012/7/19 Daniel P. Berrange berrange@redhat.com:
Libvirt needs to be able to run the following command
# nc -U /var/run/libvirt/libvirt-sock
It could use socat:
This isn't something that can simply be fixed in libvirt in rawhide because you need to consider cross-version compatibility. Every existing deployed version of libvirt in every distro will be unable to connect to rawhide, even if we changed libvirt in rawhide to use socat.
To clarify what Dan means here (which I found a bit confusing at first :-)
When a libvirt client connects to a remote machine, the libvirt client sends the "nc -U ..." command over ssh. So every libvirt on any distro that might connect to a Rawhide libvirtd must be changed to send a "socat ..." command instead.
[Having said that, socat does look like a better alternative to the various incompatible netcats, in the long run.]
Rich.
On 07/19/2012 05:46 PM, Richard W.M. Jones wrote:
When a libvirt client connects to a remote machine, the libvirt client sends the "nc -U ..." command over ssh. So every libvirt on any distro that might connect to a Rawhide libvirtd must be changed to send a "socat ..." command instead.
Alternatively, perhaps a compatibility script called 'nc' could execute the appropriate socat command?
On Fri, Jul 20, 2012 at 11:07:30AM -0400, Przemek Klosowski wrote:
On 07/19/2012 05:46 PM, Richard W.M. Jones wrote:
When a libvirt client connects to a remote machine, the libvirt client sends the "nc -U ..." command over ssh. So every libvirt on any distro that might connect to a Rawhide libvirtd must be changed to send a "socat ..." command instead.
Alternatively, perhaps a compatibility script called 'nc' could execute the appropriate socat command?
That sounds like an acceptable temporary solution until ncat supports UNIX sockets too.
-P
Hey,
On Mon, Jul 23, 2012 at 04:03:00PM +0200, Petr Šabata wrote:
On Fri, Jul 20, 2012 at 11:07:30AM -0400, Przemek Klosowski wrote:
On 07/19/2012 05:46 PM, Richard W.M. Jones wrote:
When a libvirt client connects to a remote machine, the libvirt client sends the "nc -U ..." command over ssh. So every libvirt on any distro that might connect to a Rawhide libvirtd must be changed to send a "socat ..." command instead.
Alternatively, perhaps a compatibility script called 'nc' could execute the appropriate socat command?
That sounds like an acceptable temporary solution until ncat supports UNIX sockets too.
For what it's worth, I think this change broke GNOME Boxes build: http://kojipkgs.fedoraproject.org//work/tasks/36/4270036/root.log
DEBUG util.py:257: Error: Package: libvirt-client-0.9.13-1.fc18.i686 (build) DEBUG util.py:257: Requires: nc
and I indeed get: $ repoquery --releasever=rawhide --whatprovides nc No package provides nc
libvirt.spec has:
%package client # So remote clients can access libvirt over SSH tunnel # (client invokes 'nc' against the UNIX socket on the server) Requires: nc
Christophe
On Mon, Jul 23, 2012 at 06:00:46PM +0200, Christophe Fergeau wrote:
Hey,
On Mon, Jul 23, 2012 at 04:03:00PM +0200, Petr Šabata wrote:
On Fri, Jul 20, 2012 at 11:07:30AM -0400, Przemek Klosowski wrote:
On 07/19/2012 05:46 PM, Richard W.M. Jones wrote:
When a libvirt client connects to a remote machine, the libvirt client sends the "nc -U ..." command over ssh. So every libvirt on any distro that might connect to a Rawhide libvirtd must be changed to send a "socat ..." command instead.
Alternatively, perhaps a compatibility script called 'nc' could execute the appropriate socat command?
That sounds like an acceptable temporary solution until ncat supports UNIX sockets too.
For what it's worth, I think this change broke GNOME Boxes build: http://kojipkgs.fedoraproject.org//work/tasks/36/4270036/root.log
DEBUG util.py:257: Error: Package: libvirt-client-0.9.13-1.fc18.i686 (build) DEBUG util.py:257: Requires: nc
and I indeed get: $ repoquery --releasever=rawhide --whatprovides nc No package provides nc
libvirt.spec has:
%package client # So remote clients can access libvirt over SSH tunnel # (client invokes 'nc' against the UNIX socket on the server) Requires: nc
Christophe
That was just a temporary issue (no nmap-6.01-2+ builds in the repository while nc was dead). gnome-boxes builds fine now (since July 20).
-P
On 07/19/2012 05:12 AM, Petr Šabata wrote:
The old Fedora, heavily patched OpenBSD nc package was just obsoleted by the nmap ncat implementation, available as the nmap-ncat subpackage. Those are mostly compatible and this change shouldn't cause much headache but please check your netcat dependant scripts or apps.
This appears to break freenx-server.
https://bugzilla.redhat.com/show_bug.cgi?id=891109
-
Christophe Fergeau -
Daniel P. Berrange -
Ian Pilcher -
Petr Šabata -
Przemek Klosowski -
Richard W.M. Jones -
Thomas Moschny