2007-11-07 11:49:23

by Gene Heskett

[permalink] [raw]
Subject: 2.6.24-rc2=another train wreck with amanda

Greetings;

I have this line in my /etc/modprobe.conf:
options dm-mod major=238

And I had a fsckup while building 2.6.24-rc2 cuz I thought it needed to be
based on 2.6.23.1 that has made 2.6.24-rc2 the only kernel that will boot
without a panic, killing init message.

I can survive that, but amanda (tar) went bonkers last night and tried to do a
level 0 on everything, which is about 50GB, but its virtual tape size is only
11GB.

The last time this happened that line above fixed the device mapper to a
stable address at a major of 238 which tar was happy with.

An ls -l of /dev/mapper:
[root@coyote /]# ls -l /dev/mapper
total 0
crw------- 1 root root 10, 62 Nov 6 23:40 control
brw-rw---- 1 root disk 238, 0 Nov 6 23:40 VolGroup00-LogVol00
brw-rw---- 1 root disk 238, 1 Nov 6 23:40 VolGroup00-LogVol01

So that hasn't changed, so what did? Amanda itself hasn't been changed in
several months, running the 20070727 snapshot of amanda-2.5.2p1 all this
time. Tar was updated by smart or yumex 2 or 3 days back, so it worked
correctly after the update. That leaves something in 2.6.24-rc2. I didn't
build rc1.

Other than that, rc2 seems stable. But auditd failed to start in the bootup
sequence. I think that's minor and may predate this particular kernel.
However, selinux did a relabel before it booted, could that be the cause?

--
Cheers, Gene
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
An adequate bootstrap is a contradiction in terms.


2007-11-07 18:31:24

by Andrew Morton

[permalink] [raw]
Subject: Re: 2.6.24-rc2=another train wreck with amanda

> On Wed, 07 Nov 2007 06:49:01 -0500 Gene Heskett <[email protected]> wrote:
> Greetings;
>
> I have this line in my /etc/modprobe.conf:
> options dm-mod major=238
>
> And I had a fsckup while building 2.6.24-rc2 cuz I thought it needed to be
> based on 2.6.23.1 that has made 2.6.24-rc2 the only kernel that will boot
> without a panic, killing init message.
>
> I can survive that, but amanda (tar) went bonkers last night and tried to do a
> level 0 on everything, which is about 50GB, but its virtual tape size is only
> 11GB.
>
> The last time this happened that line above fixed the device mapper to a
> stable address at a major of 238 which tar was happy with.
>
> An ls -l of /dev/mapper:
> [root@coyote /]# ls -l /dev/mapper
> total 0
> crw------- 1 root root 10, 62 Nov 6 23:40 control
> brw-rw---- 1 root disk 238, 0 Nov 6 23:40 VolGroup00-LogVol00
> brw-rw---- 1 root disk 238, 1 Nov 6 23:40 VolGroup00-LogVol01
>
> So that hasn't changed, so what did? Amanda itself hasn't been changed in
> several months, running the 20070727 snapshot of amanda-2.5.2p1 all this
> time. Tar was updated by smart or yumex 2 or 3 days back, so it worked
> correctly after the update. That leaves something in 2.6.24-rc2. I didn't
> build rc1.
>
> Other than that, rc2 seems stable. But auditd failed to start in the bootup
> sequence. I think that's minor and may predate this particular kernel.
> However, selinux did a relabel before it booted, could that be the cause?
>

An selinux relabelling could well have caused a full backup by amanda.

There's a way of forcing a relabelling so that you can confirm this, but I
forget what it is (cc's added, please).

But we don't know why the relabelling happened, do we?

2007-11-07 19:54:28

by Stephen Smalley

[permalink] [raw]
Subject: Re: 2.6.24-rc2=another train wreck with amanda

On Wed, 2007-11-07 at 10:30 -0800, Andrew Morton wrote:
> > On Wed, 07 Nov 2007 06:49:01 -0500 Gene Heskett <[email protected]> wrote:
> > Greetings;
> >
> > I have this line in my /etc/modprobe.conf:
> > options dm-mod major=238
> >
> > And I had a fsckup while building 2.6.24-rc2 cuz I thought it needed to be
> > based on 2.6.23.1 that has made 2.6.24-rc2 the only kernel that will boot
> > without a panic, killing init message.
> >
> > I can survive that, but amanda (tar) went bonkers last night and tried to do a
> > level 0 on everything, which is about 50GB, but its virtual tape size is only
> > 11GB.
> >
> > The last time this happened that line above fixed the device mapper to a
> > stable address at a major of 238 which tar was happy with.
> >
> > An ls -l of /dev/mapper:
> > [root@coyote /]# ls -l /dev/mapper
> > total 0
> > crw------- 1 root root 10, 62 Nov 6 23:40 control
> > brw-rw---- 1 root disk 238, 0 Nov 6 23:40 VolGroup00-LogVol00
> > brw-rw---- 1 root disk 238, 1 Nov 6 23:40 VolGroup00-LogVol01
> >
> > So that hasn't changed, so what did? Amanda itself hasn't been changed in
> > several months, running the 20070727 snapshot of amanda-2.5.2p1 all this
> > time. Tar was updated by smart or yumex 2 or 3 days back, so it worked
> > correctly after the update. That leaves something in 2.6.24-rc2. I didn't
> > build rc1.
> >
> > Other than that, rc2 seems stable. But auditd failed to start in the bootup
> > sequence. I think that's minor and may predate this particular kernel.
> > However, selinux did a relabel before it booted, could that be the cause?
> >
>
> An selinux relabelling could well have caused a full backup by amanda.
>
> There's a way of forcing a relabelling so that you can confirm this, but I
> forget what it is (cc's added, please).
>
> But we don't know why the relabelling happened, do we?

Gene took his question to fedora-selinux-list, which is appropriate
since the autorelabel stuff is distro-specific.

In Fedora, rc.sysinit will perform a relabel if /.autorelabel exists or
autorelabel was passed on the kernel command line. /.autorelabel can be
created by the user, by rc.sysinit itself (if you boot with SELinux
disabled, to force a relabel if you ever re-enable SELinux later), or by
utilities like system-config-{securitylevel,selinux} in response to
changing settings. Merely installing and booting a new kernel shouldn't
trigger it.

Even when triggered, a relabel shouldn't call setxattr on the file
unless its existing on-disk label doesn't match the file contexts
specification in policy. There is a force option that unconditionally
sets the label on all files, but I don't see that being used by the
autorelabel support.

--
Stephen Smalley
National Security Agency