Bug#816227: ping: socket: Address family not supported by protocol (raw socket required by specified options).

Jamie Heilman

unread,

Feb 28, 2016, 5:10:03 PM2/28/16

to

Package: iputils-ping
Version: 3:20150815-1
Severity: grave

root@cucamonga:~# ping 127.0.0.1
ping: socket: Address family not supported by protocol (raw socket required by specified options).

I actually can't ping anything at all, it all fails with the above
error message.

I wouldn't be stunned to learn this is because I have no ipv6
connectivity and no ipv6 support in my kernel, but I don't think
that makes this behavior anymore OK.

--
Jamie Heilman http://audible.transient.net/~jamie/

Jamie Heilman

unread,

Feb 28, 2016, 5:50:03 PM2/28/16

to

Noah Meyerhans wrote:
> I cannot reproduce this on a host with no ipv6 connectivity. Does
> 'ping -4 127.0.0.1' work any differently?

Yeah, that works.

> By "no ipv6 support", do you mean you're running a custom kernel with a
> different configuration than provided by Debian?

Correct.

Florent Rougon

unread,

Feb 29, 2016, 3:10:03 AM2/29/16

to

Hello,

On Sun, 28 Feb 2016 22:38:05 +0000 Jamie Heilman <ja...@audible.transient.net> wrote:

> Noah Meyerhans wrote:
> > I cannot reproduce this on a host with no ipv6 connectivity. Does
> > 'ping -4 127.0.0.1' work any differently?
>
> Yeah, that works.

Same here. 'ping -4 <ipv4_address>' works fine but 'ping <ipv4_address>'
doesn't (I tried with 127.0.0.1, with the address of a local Ethernet
interface, with the address of a host I can ssh to on the local
network...).

> > By "no ipv6 support", do you mean you're running a custom kernel with a
> > different configuration than provided by Debian?

For me, the kernel is that from linux-image-4.4.0-1-amd64 version
4.4.2-3 (recently updated from unstable), rebuilt with one tiny patch
that is not network-related in any way, and which I have been using
since June 2015 without any problem: it is just reverting upstream Linux
kernel commit 79346d620e9de87912de73337f6df8b7f9a46888 ("HID: input:
force generic axis to be mapped to their user space axis"; I
unfortunately have to apply this patch to every kernel release since
then, because nobody bothered to even answer bug #785606 despite my
'git bisect'ing it).

Apart from that, I use:

GRUB_CMDLINE_LINUX="init=/bin/systemd ipv6.disable=1"

in /etc/default/grub, which effectively disables IPv6, AFAICT.

My aptitude.log shows:

[UPGRADE] iputils-ping:amd64 3:20121221-5+b2 -> 3:20150815-1

on Fri, Feb 26 2016 20:48:43 +0100, and I only noticed the problem
today.

Thanks for considering!

--
Florent

Noah Meyerhans

unread,

Feb 29, 2016, 4:30:03 PM2/29/16

to

Control: tags -1 + upstream pending fixed-upstream

On Mon, Feb 29, 2016 at 09:02:02AM +0100, Florent Rougon wrote:
> Apart from that, I use:
>
> GRUB_CMDLINE_LINUX="init=/bin/systemd ipv6.disable=1"

Confirmed that this breaks ping when run without an explicit address
family. This is actually already fixed upstream, but not in a tagged
snapshot. It'll be fixed in the next upload, by cherry-picking that
commit or syncing a new snapshot.

Thanks
noah

signature.asc

Florent Rougon

unread,

Mar 1, 2016, 5:10:03 AM3/1/16

to

Hello,

Noah Meyerhans <no...@debian.org> wrote:

> Confirmed that this breaks ping when run without an explicit address
> family. This is actually already fixed upstream, but not in a tagged
> snapshot. It'll be fixed in the next upload, by cherry-picking that
> commit or syncing a new snapshot.

Good news, thank you. :-)

--
Florent