Debian Bug report logs - #809166
ifdown stops "manual" interfaces on shutdown

Toggle useless messages

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

From: Martin Pitt <mpitt@debian.org>

To: Debian BTS Submit <submit@bugs.debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org

Subject: networking.service does not prevent ifdown with network file systems

Date: Sun, 27 Dec 2015 19:58:06 +0100

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

Package: ifupdown
Version: 0.8

Hello,

/etc/init.d/networking has some functionality to check if the root
file system or swap are on a network file system, in
check_network_file_systems() and check_network_swap(). In that case,
the interfaces are not ifdown'ed during shutdown, so that these
network file systems can be cleanly unmouted at the very end.

However, the new networking.service does not do that.  So unmounting
these network file systems will hang and fail as the interface is
already down. Root on NFS or even more so iSCSI is rather popular, so
this is quite an important regression.

Perhaps these checks can be put into ifdown itself and be done with
"ifdown -a"? Then you can still ifdown an individual interface (which
is usually deliberately done by the admin, even if it means shooting
yourself into the foot), but it would DTRT when being called by the
init.d script or the systemd unit. Alternatively this could become a
new option such as --check-netfs that the init scripts use.

Thanks,

Martin

-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #10 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: Martin Pitt <mpitt@debian.org>, 809166@bugs.debian.org

Cc: pkg-systemd-maintainers@lists.alioth.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Sun, 27 Dec 2015 20:50:20 +0100

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

On Sun, Dec 27, 2015 at 07:58:06PM +0100, Martin Pitt wrote:

> Package: ifupdown
> Version: 0.8
> 
> /etc/init.d/networking has some functionality to check if the root
> file system or swap are on a network file system, in
> check_network_file_systems() and check_network_swap(). In that case,
> the interfaces are not ifdown'ed during shutdown, so that these
> network file systems can be cleanly unmouted at the very end.
> 
> However, the new networking.service does not do that.  So unmounting
> these network file systems will hang and fail as the interface is
> already down. Root on NFS or even more so iSCSI is rather popular, so
> this is quite an important regression.

Please check if the problem still exists with version 0.8.4. Version 0.8
didn't correctly order itself with respect to network-online.target, but
0.8.4 should have fixed that.

> Perhaps these checks can be put into ifdown itself and be done with
> "ifdown -a"? Then you can still ifdown an individual interface (which
> is usually deliberately done by the admin, even if it means shooting
> yourself into the foot), but it would DTRT when being called by the
> init.d script or the systemd unit. Alternatively this could become a
> new option such as --check-netfs that the init scripts use.

No, I won't put such checks in ifdown itself, it should be fixed now.
And if you manually ifdown an interface that is necessary for a network
mount, you can easily ifup it manually to fix it.

-- 
Met vriendelijke groet / with kind regards,
      Guus Sliepen <guus@debian.org>

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

Message #15 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: 809166@bugs.debian.org, pkg-systemd-maintainers@lists.alioth.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Mon, 28 Dec 2015 08:13:18 +0100

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

Hey Guus,

Guus Sliepen [2015-12-27 20:50 +0100]:
> Please check if the problem still exists with version 0.8.4. Version 0.8
> didn't correctly order itself with respect to network-online.target, but
> 0.8.4 should have fixed that.

This should be unrelated to network-online.target. networking.service
has "Conflicts=shutdown.target" and thus will always be stopped during
shutdown/reboot, you can see

  [  OK  ] Stopped Raise network interfaces.

So with NFS or iscsi this will now mean that shutdown will be rather
broken/hanging after that point.

I don't have an iscsi installation handy right now, but I can set one
up. I filed this as I was reviewing networking.service with Michael
and we noticed that this functionality was gone, and I also got a
report that this regressed with latest systemd
(https://launchpad.net/bugs/1492546) -- that's an independent
regression though, and I'll handle that.

So maybe the effects with this are bearable, I didn't yet investigate
it on an actually affected system. So please feel free to close if you
are convinced that this check isn't necessary any more, I mostly filed
this as a reminder.

Thanks,

Martin
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #20 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: Martin Pitt <mpitt@debian.org>, 809166@bugs.debian.org

Cc: pkg-systemd-maintainers@lists.alioth.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Mon, 28 Dec 2015 11:08:46 +0100

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

On Mon, Dec 28, 2015 at 08:13:18AM +0100, Martin Pitt wrote:

> > Please check if the problem still exists with version 0.8.4. Version 0.8
> > didn't correctly order itself with respect to network-online.target, but
> > 0.8.4 should have fixed that.
> 
> This should be unrelated to network-online.target. networking.service
> has "Conflicts=shutdown.target" and thus will always be stopped during
> shutdown/reboot, you can see
> 
>   [  OK  ] Stopped Raise network interfaces.
> 
> So with NFS or iscsi this will now mean that shutdown will be rather
> broken/hanging after that point.

Hm, but shouldn't systemd (with networking.service from ifupdown 0.8.4)
order the unmounting of the NFS/iSCSI partitions before stoppping the
network interfaces?

> So maybe the effects with this are bearable, I didn't yet investigate
> it on an actually affected system. So please feel free to close if you
> are convinced that this check isn't necessary any more, I mostly filed
> this as a reminder.

That's fine, I'll keep it open until it's confirmed that the issue is
fixed. But I would like the solution to be that systemd, given correct
information in the .service files, correctly orders everything, rather
than hacking ifupdown itself to work around this problem.

-- 
Met vriendelijke groet / with kind regards,
      Guus Sliepen <guus@debian.org>

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

Message #25 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: 809166@bugs.debian.org, pkg-systemd-maintainers@lists.alioth.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 30 Dec 2015 13:26:51 +0100

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

Hello Guus,

Guus Sliepen [2015-12-28 11:08 +0100]:
> Hm, but shouldn't systemd (with networking.service from ifupdown 0.8.4)
> order the unmounting of the NFS/iSCSI partitions before stoppping the
> network interfaces?

This works for /home and other "sub"mounts, and there systemd should
indeed figure out a correct ordering. (And it would be a systemd bug,
not an ifupdown bug if that isn't the case).

But not for the root partition -- that happens as the very last thing
(hardcoded in systemd-shutdown(8) after after all units got stopped.
That's what I meant with "unit ordering is irrelevant for this
problem".

I. e. if the root partition is on a network file system, we must never
tear down the network interfaces. But systemd does not know that
"networking.service" does that (it's just another unit which declares
that it wants to be stopped on shutdown and has an ExecStop=). In that
regard it is no different from sysvinit or upstart, which is why
/etc/init.d/networking carries this check.

Thanks,

Martin

-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #30 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: Martin Pitt <mpitt@debian.org>, Guus Sliepen <guus@debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 30 Dec 2015 19:41:17 +0100

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

Am 30.12.2015 um 13:26 schrieb Martin Pitt:
> Hello Guus,
> 
> Guus Sliepen [2015-12-28 11:08 +0100]:
>> Hm, but shouldn't systemd (with networking.service from ifupdown 0.8.4)
>> order the unmounting of the NFS/iSCSI partitions before stoppping the
>> network interfaces?
> 
> This works for /home and other "sub"mounts, and there systemd should
> indeed figure out a correct ordering. (And it would be a systemd bug,
> not an ifupdown bug if that isn't the case).
> 
> But not for the root partition -- that happens as the very last thing
> (hardcoded in systemd-shutdown(8) after after all units got stopped.
> That's what I meant with "unit ordering is irrelevant for this
> problem".
> 
> I. e. if the root partition is on a network file system, we must never
> tear down the network interfaces. But systemd does not know that

Can you actually use ifupdown for that?
If your / is on a network file system you most likely need to setup your
network via the initramfs, or not?
Why should ifupdown tear down the network on shutdown then?


Regards,
Michael

-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

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

Message #35 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Michael Biebl <biebl@debian.org>

Cc: Guus Sliepen <guus@debian.org>, pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 16:00:16 +0100

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

Hello Guus, Michael,

sorry, took a bit to get back to this. I did some research about this
now.

Michael Biebl [2015-12-30 19:41 +0100]:
> Can you actually use ifupdown for that?
> If your / is on a network file system you most likely need to setup your
> network via the initramfs, or not?

Correct; scenarios that use remote root file systems (NFS, open-iscsi,
LTSP) ship initramfs-tools hooks with use things like "ipconfig" or
"udhcpd" to bring up the ethernet interface, they don't use ifupdown.

> Why should ifupdown tear down the network on shutdown then?

E. g. LTSP creates an ifupdown stanza like "iface eth0 inet manual"
for the remote root interface to prevent NetworkManager from
(auto-)connecting/disconnecting eth0. NM ignores interfaces that are
configured in ifupdown, thus preventing the user from  shooting
herself into the foot by accidentally/maliciously clicking on
"disconnect" in the network icon. This use case actually got broken
by the overzealous patch in #809169 which added this:

   down
+    /bin/ip link set dev %iface% down 2>/dev/null || true \
+        if (iface_is_link())

for "manual" interfaces. I. e. "ifdown eth0" on the above "manual"
interface example would actually down the interface. This isn't
documented behaviour, and breaks this (IMHO) valid use case to mark an
interface as "please leave me alone". This is what's causing the
shutdown hang with LTSP now. (See https://launchpad.net/1492546 for
details).

The attached debdiff [1] reverts this bit, sorry about that. We've had
this for ages in Ubuntu and I didn't quite know what this was about.
As long as we had the check_network_file_systems() bits in
/etc/init.d/networking this papered over this misbehaviour, but with
that gone there is no protection any more from tearing down interfaces
with remote root fs.


I also looked at open-iscsi in Ubuntu. That package has a weird patch
which hacks ifupdown's state files to think that eth0 is already up,
so that ifupdown does not try to start it again during boot. It does
that because the cloud images have a hardcoded "iface eth0 inet dhcp"
in them, and open-iscsi needs to work with those. Again the
check_network_file_systems() check prevented a shutdown hang here. But
this whole approach is just wrong, and we need to fix that on the
Ubuntu side to generate an ifupdown interface stanza on first boot on
the fly (this is also necessary for the ifnames changes, as it's not
guaranteed to be eth0 these days). This part isn't really relevant for
you, but I wanted to mention it here for the record.

Thanks,

Martin

[1] Sorry, anonscm etc. are down ATM, I don't have access to current git
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

[ifdown.no-manual-down.debdiff (text/plain, attachment)]

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

Message #40 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 16:05:09 +0100

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

Martin Pitt [2016-01-13 16:00 +0100]:
> +  * inet{,6}.defn: Don't down an interface when stopping. This got introduced
> +    in 0.8.5, but breaks use cases like LTSP which create a "manual"
> +    interfaces stanza to prevent NetworkManager or other management software
> +    to *prevent* touching the interface of the remote root file system.

Argh, today I can't talk today :-) Double-"prevent".

Can you please s/to *prevent*/from/ here when you apply this?

Thanks!

Martin
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #47 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: Martin Pitt <mpitt@debian.org>

Cc: Michael Biebl <biebl@debian.org>, pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 16:30:03 +0100

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

On Wed, Jan 13, 2016 at 04:00:16PM +0100, Martin Pitt wrote:

> > Why should ifupdown tear down the network on shutdown then?
> 
> E. g. LTSP creates an ifupdown stanza like "iface eth0 inet manual"
> for the remote root interface to prevent NetworkManager from
> (auto-)connecting/disconnecting eth0. NM ignores interfaces that are
> configured in ifupdown, thus preventing the user from  shooting
> herself into the foot by accidentally/maliciously clicking on
> "disconnect" in the network icon.

Ugh. That's totally wrong. Why doesn't LTSP tell NetworkManager itself
to stop messing with eth0? Instead of adding a stanza to /e/n/i, just
write this to /etc/NetworkManager/NetworkManager.conf:

[keyfile]
unmanaged-devices=interface-name=eth0

> This use case actually got broken
> by the overzealous patch in #809169 which added this:
> 
>    down
> +    /bin/ip link set dev %iface% down 2>/dev/null || true \
> +        if (iface_is_link())
> 
> for "manual" interfaces. I. e. "ifdown eth0" on the above "manual"
> interface example would actually down the interface. This isn't
> documented behaviour, and breaks this (IMHO) valid use case to mark an
> interface as "please leave me alone".

I'd say that the latter use case is invalid.

> I also looked at open-iscsi in Ubuntu. That package has a weird patch
> which hacks ifupdown's state files to think that eth0 is already up,

Oh god. That's also completely wrong. Ifupdown 0.8 and later uses
per-interface state files (because they can be locked individually), the
/run/network/ifstate file is still there for now but it might be phased
out.

> Again the check_network_file_systems() check prevented a shutdown hang
> here. But this whole approach is just wrong, and we need to fix that
> on the Ubuntu side to generate an ifupdown interface stanza on first
> boot on the fly (this is also necessary for the ifnames changes, as
> it's not guaranteed to be eth0 these days). This part isn't really
> relevant for you, but I wanted to mention it here for the record.

What I am willing to do is to add the keyword "no-auto-down" to
interface stanza's, which will prevent the interface from being brought
down using "ifdown -a".

-- 
Met vriendelijke groet / with kind regards,
      Guus Sliepen <guus@debian.org>

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

Message #52 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: Michael Biebl <biebl@debian.org>, pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 16:43:33 +0100

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

Hey Guus,

Guus Sliepen [2016-01-13 16:30 +0100]:
> Ugh. That's totally wrong. Why doesn't LTSP tell NetworkManager itself
> to stop messing with eth0? Instead of adding a stanza to /e/n/i, just
> write this to /etc/NetworkManager/NetworkManager.conf:
> 
> [keyfile]
> unmanaged-devices=interface-name=eth0

I guess it could do that. This is currently done in

  https://sources.debian.net/src/ltsp/5.5.5-1/client/share/ltsp/init-ltsp.d/50-interfaces/

and it could be equally well writing to NetworkManager.conf.

> 
> > This use case actually got broken
> > by the overzealous patch in #809169 which added this:
> > 
> >    down
> > +    /bin/ip link set dev %iface% down 2>/dev/null || true \
> > +        if (iface_is_link())
> > 
> > for "manual" interfaces. I. e. "ifdown eth0" on the above "manual"
> > interface example would actually down the interface. This isn't
> > documented behaviour, and breaks this (IMHO) valid use case to mark an
> > interface as "please leave me alone".
> 
> I'd say that the latter use case is invalid.

I do agree it's inelegant, but these LTSP versions are out there.

But independently of that: Do you actually think it is right that
"ifdown" on a manual interface without any "down" command downs it?
That isn't documented and hasn't happened until 0.8.4, and IMHO that's
overzealous and might break other setups too.

> > [open-iscsi]
> Oh god. That's also completely wrong.

Heh, amen :-/ As I said, we'll fix that.

> What I am willing to do is to add the keyword "no-auto-down" to
> interface stanza's, which will prevent the interface from being brought
> down using "ifdown -a".

In the interest of backward compatibility, it would perhaps make more
sense to add a keyword "auto-down" which will then "down" a manual
interface on ifdown? This would then be a shortcut for

  iface eth0 inet manual
     down ifconfig $IFACE down

Thanks,

Martin
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #57 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: Martin Pitt <mpitt@debian.org>

Cc: Michael Biebl <biebl@debian.org>, pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 16:57:48 +0100

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

On Wed, Jan 13, 2016 at 04:43:33PM +0100, Martin Pitt wrote:

> > [keyfile]
> > unmanaged-devices=interface-name=eth0
> 
> I guess it could do that. This is currently done in
> 
>   https://sources.debian.net/src/ltsp/5.5.5-1/client/share/ltsp/init-ltsp.d/50-interfaces/
> 
> and it could be equally well writing to NetworkManager.conf.

Great. Who shall do the honors of writing a patch/bug report?

> But independently of that: Do you actually think it is right that
> "ifdown" on a manual interface without any "down" command downs it?
> That isn't documented and hasn't happened until 0.8.4, and IMHO that's
> overzealous and might break other setups too.

Uh, didn't we discuss that in #809169? Wasn't it already the default in
Ubuntu?

Anyway, I'm not for or against it, but I was slightly convinced by the
use case of lots of bonding slave interfaces, that it would make sense
to have empty inet manual interfaces that did bring interfaces up or
down. And that still is more convincing than the use case of LTSP doing
the wrong thing.

> > What I am willing to do is to add the keyword "no-auto-down" to
> > interface stanza's, which will prevent the interface from being brought
> > down using "ifdown -a".
> 
> In the interest of backward compatibility, it would perhaps make more
> sense to add a keyword "auto-down" which will then "down" a manual
> interface on ifdown? This would then be a shortcut for
> 
>   iface eth0 inet manual
>      down ifconfig $IFACE down

But do we need backwards compatibility? If we get LTSP and open-iscsi
fixed in unstable, is there still a problem somewhere?

-- 
Met vriendelijke groet / with kind regards,
      Guus Sliepen <guus@debian.org>

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

Message #62 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: Guus Sliepen <guus@debian.org>, Martin Pitt <mpitt@debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 20:51:02 +0100

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

Am 13.01.2016 um 16:57 schrieb Guus Sliepen:
> On Wed, Jan 13, 2016 at 04:43:33PM +0100, Martin Pitt wrote:
> 
>>> [keyfile]
>>> unmanaged-devices=interface-name=eth0
>>
>> I guess it could do that. This is currently done in
>>
>>   https://sources.debian.net/src/ltsp/5.5.5-1/client/share/ltsp/init-ltsp.d/50-interfaces/
>>
>> and it could be equally well writing to NetworkManager.conf.
>

Better then mangling NetworkManager.conf would be to drop a config file
into /etc/NetworkManager/conf.d/


-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

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

Message #67 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 23:30:32 +0100

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

Hey Guus,

Guus Sliepen [2016-01-13 16:57 +0100]:
> Great. Who shall do the honors of writing a patch/bug report?

I filed https://bugs.debian.org/XXXXXX, thanks Michael for pointing
out conf.d/.

> > But independently of that: Do you actually think it is right that
> > "ifdown" on a manual interface without any "down" command downs it?
> > That isn't documented and hasn't happened until 0.8.4, and IMHO that's
> > overzealous and might break other setups too.
> 
> Uh, didn't we discuss that in #809169? Wasn't it already the default in
> Ubuntu?

It was, but as I said the check_network_file_systems() function
short-circuited this until recently, so interfaces were never really
downed on shutdown as soon as there was any remote file system. So
this (mis)design issue in ltsp and friends didn't really pop up until
now.

> Anyway, I'm not for or against it, but I was slightly convinced by the
> use case of lots of bonding slave interfaces, that it would make sense
> to have empty inet manual interfaces that did bring interfaces up or
> down. And that still is more convincing than the use case of LTSP doing
> the wrong thing.

From a clean slate POV I agree. I'm just a bit concerned about
breaking existing systems with that. I don't have a really strong
opinion, I just noticed that this change already caused regressions
with at least one package in Debian (ltsp) and two in Ubuntu (that and
open-isci).

That downing of interfaces is really only relevant for manual
operations, right? I can't think of anything bad that would happen if
a real interface is not downed during shutdown -- in fact, doing that
has no real benefit that I can see.

> But do we need backwards compatibility? If we get LTSP and open-iscsi
> fixed in unstable, is there still a problem somewhere?

There are more use cases/packages which set up a remote root fs;
initramfs-tools itself supports root-on-NFS, tcos, custom setups with
PXE boot, etc. I haven't searched for them or analyzed them all.

Thanks,

Martin
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #72 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Wed, 13 Jan 2016 23:33:12 +0100

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

Martin Pitt [2016-01-13 23:30 +0100]:
> I filed https://bugs.debian.org/XXXXXX, thanks Michael for pointing
> out conf.d/.

"Let's type this mail while I wait for the bug number" syndrome,
sorry. It's <http://bugs.debian.org/810934>.

Martin
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #77 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Alkis Georgopoulos <alkisg@gmail.com>

To: 809166@bugs.debian.org

Subject: Re: networking.service does not prevent ifdown with network file systems

Date: Fri, 15 Jan 2016 09:37:31 +0200

Hi, I'm the LTSP developer that added the "manual" stanza a few years 
ago (r1944).

I've answered in 
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=810934#10 about it, 
but I should probably add a comment here as well as this bug sees the 
issue from a different view.

The "manual" stanza isn't there just for network-manager. LTSP chroots 
don't have network-manager installed by default.
A netbooted system needs these 2 things from ifupdown:
1) the scripts at /etc/network/if-up.d to get executed for the boot 
interface,
2) while preventing the boot interface from being DOWNed at shutdown.

You mention that using "manual" is the wrong approach to accomplish 
this. Could you please advise us on how to notify ifupdown to do those 2 
things in a backwards/forward compatible way?

Message #82 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: Alkis Georgopoulos <alkisg@gmail.com>, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Fri, 15 Jan 2016 09:14:52 +0100

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

On Fri, Jan 15, 2016 at 09:37:31AM +0200, Alkis Georgopoulos wrote:

> The "manual" stanza isn't there just for network-manager. LTSP chroots don't
> have network-manager installed by default.
> A netbooted system needs these 2 things from ifupdown:
> 1) the scripts at /etc/network/if-up.d to get executed for the boot
> interface,
> 2) while preventing the boot interface from being DOWNed at shutdown.
> 
> You mention that using "manual" is the wrong approach to accomplish this.
> Could you please advise us on how to notify ifupdown to do those 2 things in
> a backwards/forward compatible way?

Which scripts in /etc/network/if-up.d are required for the system to
work?

If the network has been brought up by other means before mounting the
root filesystem, I don't see why ifupdown needs to be used at all. If it
does need to be used, can you explain in more detail what is happening
during the boot of an LTSP client?

The best way to prevent the boot interface from being downed at shutdown
is not to have it managed by ifupdown in the first place. Ifupdown will
only touch interfaces that are defined in /etc/network/interfaces.

-- 
Met vriendelijke groet / with kind regards,
      Guus Sliepen <guus@debian.org>

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

Message #87 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Alkis Georgopoulos <alkisg@gmail.com>

To: 809166@bugs.debian.org

Cc: Guus Sliepen <guus@debian.org>

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Fri, 15 Jan 2016 11:42:48 +0200

About the boot process of LTSP clients, I think it's better not to focus 
on LTSP, but on the simplest possible netbooted system that works out of 
the box in all Debian/Ubuntu versions (and many other Linux distros) 
without using a custom initramfs:
Let's say we have an "rw" NFS export for a single client on some server.
And we just pass "root=/dev/nfs ip=dhcp boot=nfs" in the client command 
line.
The *stock* initramfs (related packages: initramfs-tools, klibc) then 
brings up the network interface and mounts the NFS root file system and 
chains to it.

That NFS export (chroot) may have any package installed in it, gnome, 
kde, apache, network-manager, avahi etc.
There's no special handling of anything; everything works without 
customizing any configuration file, the only part that is needed is to 
put "manual" in $CHROOT/etc/network/intefaces.
If "manual" is not inserted there, then the packages that use ifupdown 
won't receive the appropriate events.
We can get a list of those packages from http://packages.debian.org, but 
I don't think it's important; we don't want to special-case any one of them.
For example, ntpdate needs an if-up.d event to sync the clock, ethtool 
adjusts the WOL and other NIC settings, and there's also avahi, openssh, 
resolvconf etc.

So to sum up, a netbooted system needs a way to "have all the events 
sent for the boot interface, but prohibit its IP from being changed and 
prohibit it from getting down".
In practise, the "manual" flag covered that fine so far.
But if that's ...abusing its intented purpose, and another solution is 
necessary, I would propose this:
1) initramfs-tools, dracut etc could create a file somewhere in /run to 
mark the boot interface
2) ifupdown, network-manager, wicd etc could read that file and handle 
that interface in that aforementioned special way (send events, maybe 
even show the connection speed and status in the network-manager UI, but 
don't change the IP or ifdown it).

Message #92 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: Alkis Georgopoulos <alkisg@gmail.com>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Thu, 21 Jan 2016 22:10:18 +0100

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

On Fri, Jan 15, 2016 at 09:14:52AM +0100, Guus Sliepen wrote:

> > The "manual" stanza isn't there just for network-manager. LTSP chroots don't
> > have network-manager installed by default.
> > A netbooted system needs these 2 things from ifupdown:
> > 1) the scripts at /etc/network/if-up.d to get executed for the boot
> > interface,
> > 2) while preventing the boot interface from being DOWNed at shutdown.

ifupdown version 0.8.9 will not bring the link down of manual interfaces
during shutdown (but every "down" and "post-down" keyword is still
executed).

ifupdown 0.8.9 can also mark interfaces with the "no-auto-down" keyword,
which causes those interfaces to be ignored during "ifdown -a". This is
a more explicit way to tell an interface should not be brought down
during shutdown. An example:

auto eth0
no-auto-down eth0

iface eth0 inet dhcp
	...

I hope this will help.

-- 
Met vriendelijke groet / with kind regards,
      Guus Sliepen <guus@debian.org>

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

Message #97 received at 809166-close@bugs.debian.org (full text, mbox, reply):

From: Guus Sliepen <guus@debian.org>

To: 809166-close@bugs.debian.org

Subject: Bug#809166: fixed in ifupdown 0.8.9

Date: Thu, 21 Jan 2016 22:02:54 +0000

Source: ifupdown
Source-Version: 0.8.9

We believe that the bug you reported is fixed in the latest version of
ifupdown, which is due to be installed in the Debian FTP archive.

A summary of the changes between this version and the previous one is
attached.

Thank you for reporting the bug, which will now be closed.  If you
have further comments please address them to 809166@bugs.debian.org,
and the maintainer will reopen the bug report if appropriate.

Debian distribution maintenance software
pp.
Guus Sliepen <guus@debian.org> (supplier of updated ifupdown package)

(This message was generated automatically at their request; if you
believe that there is a problem with it please contact the archive
administrators by mailing ftpmaster@ftp-master.debian.org)


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Format: 1.8
Date: Thu, 21 Jan 2016 22:02:11 +0100
Source: ifupdown
Binary: ifupdown
Architecture: source amd64
Version: 0.8.9
Distribution: unstable
Urgency: medium
Maintainer: Guus Sliepen <guus@debian.org>
Changed-By: Guus Sliepen <guus@debian.org>
Description:
 ifupdown   - high level tools to configure network interfaces
Closes: 615130 809166
Changes:
 ifupdown (0.8.9) unstable; urgency=medium
 .
   * It's network-online.target, not .service.
   * Do not run "ip link set down" on manual interfaces during "ifdown -a".
     Closes: #809166
   * Add the keywords "no-auto-down" and "no-scripts". Closes: #615130
Checksums-Sha1:
 d929c137f01e50edda15e849cf0727c0df0e6009 1536 ifupdown_0.8.9.dsc
 2e2de639fd424cbfc7ce6bd9ff0793aad7fac434 70208 ifupdown_0.8.9.tar.xz
 d72b036f32a6b875505f43a8b673050c87146949 60674 ifupdown-dbgsym_0.8.9_amd64.deb
 55e20efc2027a61a1ae25f5ff64581b9d5d278e7 73174 ifupdown_0.8.9_amd64.deb
Checksums-Sha256:
 f3d733ceaa283fdb974336d3b34a1a9ca248b65d3d811eabb70265002fd9805d 1536 ifupdown_0.8.9.dsc
 9c12da542f43c604055817a28a58b8704e1598a7dde42b18ea45b6e46d8925f6 70208 ifupdown_0.8.9.tar.xz
 963b059165eb1049e2cf54102a260c73194ae73b3b5b49687adc45f50ab53f0c 60674 ifupdown-dbgsym_0.8.9_amd64.deb
 20d19ace8bedf9651b6adfbd8be80bfbda90df152da4cc39d2e6d71c1cbdf24e 73174 ifupdown_0.8.9_amd64.deb
Files:
 db29243dcb36f0bd4a7cc460be501a92 1536 admin important ifupdown_0.8.9.dsc
 83be4dd9c5599583f13e20181c43f07f 70208 admin important ifupdown_0.8.9.tar.xz
 8cdb7e62c6cb2277526bf6474732cab8 60674 debug extra ifupdown-dbgsym_0.8.9_amd64.deb
 f67dd914f58f537a4433d7b62739cab5 73174 admin important ifupdown_0.8.9_amd64.deb

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
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=BP9T
-----END PGP SIGNATURE-----

Message #102 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Martin Pitt <mpitt@debian.org>

To: Guus Sliepen <guus@debian.org>

Cc: Alkis Georgopoulos <alkisg@gmail.com>, pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Fri, 22 Jan 2016 11:49:35 +0100

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

Hey Guus,

Guus Sliepen [2016-01-21 22:10 +0100]:
> ifupdown version 0.8.9 will not bring the link down of manual interfaces
> during shutdown (but every "down" and "post-down" keyword is still
> executed).
> 
> ifupdown 0.8.9 can also mark interfaces with the "no-auto-down" keyword,
> which causes those interfaces to be ignored during "ifdown -a".

Great, thanks Guus!

Martin
-- 
Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

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

Message #107 received at 809166@bugs.debian.org (full text, mbox, reply):

From: Alkis Georgopoulos <alkisg@gmail.com>

To: Guus Sliepen <guus@debian.org>

Cc: pkg-systemd-maintainers@lists.alioth.debian.org, 809166@bugs.debian.org

Subject: Re: Bug#809166: networking.service does not prevent ifdown with network file systems

Date: Fri, 22 Jan 2016 13:02:24 +0200

On 21/01/2016 11:10 μμ, Guus Sliepen wrote:
> On Fri, Jan 15, 2016 at 09:14:52AM +0100, Guus Sliepen wrote:
>
> ifupdown 0.8.9 can also mark interfaces with the "no-auto-down" keyword,
> which causes those interfaces to be ignored during "ifdown -a". This is
> a more explicit way to tell an interface should not be brought down
> during shutdown. An example:
>
> auto eth0
> no-auto-down eth0
>
> iface eth0 inet dhcp
> 	...
>
> I hope this will help.
>


Perfect, thanks a lot! :)

Send a report that this bug log contains spam.