Bug#753790: systemd: process 1 should load new versions of shared objects

[Russell Coker]

On Sat, 5 Jul 2014 04:40:33 Michael Biebl wrote:
> > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=753726
> > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=753727
> > 
> > The above bugs concern the ability of library packages to request that
> > systemd use the new version on an upgrade.  I don't think it's reasonable
> > for the library updates to never be applied because there's the risk of a
> > security flaw being discovered in one of them that affects the operation
> > of systemd.
> While I agree with you in general, keep in mind that this is actually
> also a general issue. PID 1 is in no way special in that regard and this
> concerns every long running process / daemon.

Pid 1 is special in that it must always exist.

> It's not like a security update of libselinux (or any other library for
> that matter) restarts all daemons / binaries linking against said library.

I think it should.  We already have pam and libc6 restarting all daemons that 
link against them.

> Incidentally we discussed exactly that within the pkg-systemd team
> before I filed this bug. Our conclusion was, that the right answer for
> that is probably something like checkrestart which is run *after* the
> upgrade has completed.

Sounds reasonable.

> > As there is apparently a reliability issue in the library postinst calling
> > "telinit u" I think that systemd should have triggers to allow it to take
> > the new libraries when they are installed.
> 
> I'm not convinced that a package-individual trigger is the right answer
> for this (we also discussed this possibility within the team). Every
> package providing a long running system service would have to provide
> such a trigger and every library would have to call all triggers. That
> doesn't scale.

Why not?  A typical system only has a couple of dozen long running daemons and 
for most of them there aren't many libraries that they link against (except 
systemd).  Of the daemons that are long running there are only a few that may 
cause difficulty to restart, that's systemd, xen daemons, an XDM, and 
rsyslogd.

> We need a general solution for this.
> 
> What I'm convinced about though is, that restarting a daemon (or
> re-execing PID 1) midway through an upgrade is bound to fail one way or
> another.
> 
> So I still kindly ask you to apply the patches in #753726 and #753727

OK I think we should do that next time we update them.  I don't think it's 
worth doing a special update for that.

-- 
My Main Blog         http://etbe.coker.com.au/
My Documents Blog    http://doc.coker.com.au/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: This is a digitally signed message part.
URL: <http://alioth-lists.debian.net/pipermail/pkg-systemd-maintainers/attachments/20140705/923edfbc/attachment-0002.sig>