[karmic] udev requires new kernel, breaks on EC2

Bug #397187 reported by Andreas Hasenack
12
This bug affects 2 people
Affects Status Importance Assigned to Milestone
Ubuntu on EC2
Confirmed
Undecided
Unassigned
udev (Ubuntu)
Won't Fix
Undecided
Unassigned

Bug Description

Binary package hint: udev

At some point, udev started using signalfd(2), which requires a new kernel. IIRC it was implemented in 2.6.25.

While this is not a problem for karmic itself, it is a problem for EC2, which runs an older kernel. If one tries to upgrade the current jaunty image to karmic, udev won't start anymore:

root@ip-10-244-15-193:~# udevd
error getting signalfd

strace shows:
8706 SYS_327(0xffffffff, 0xbfc5238c, 0x8, 0, 0x2f4ff4) = -1 ENOSYS (Function not implemented)
8706 write(2, "error getting signalfd\n", 23) = 23

The same happens when booting Alestic's karmic image (ami-19a34270). Out of the box it still works, because it has udev 142, but dist-upgrade it and udev breaks too.

Revision history for this message
Eric Hammond (esh) wrote :

Running the latest udev (144) on a Karmic EC2 instance causes the network to not come back up after a reboot because of this problem (Andreas found this out, but I confirmed and thought it was also worth mentioning this severe effect in this bug). I'm not sure this is really a udev bug; the right answer might be to build a newer kernel for EC2, though I can't judge the effort required to "fix" either side.

Changed in ubuntu-on-ec2:
status: New → Confirmed
Revision history for this message
Scott James Remnant (Canonical) (canonical-scott) wrote :

Upstart will also soon start requiring signalfd()

The EC2 kernel should be updated

Changed in udev (Ubuntu):
status: New → Won't Fix
Chuck Short (zulcss)
Changed in ubuntu-on-ec2:
status: Confirmed → Won't Fix
Revision history for this message
Eric Hammond (esh) wrote :

By "won't fix" I assume this means that it will not be possible to upgrade an instance running an Intrepid/Jaunty AMI to Karmic and expect it to keep running on the older Intrepid/Jaunty kernel. This makes sense given the issues with the kernel and the inability to upgrade a kernel on a running instance in the EC2 environment.

I assume that "won't fix" does not mean that the plan is to not build a Karmic kernel.

Revision history for this message
Scott Moser (smoser) wrote : Re: [Bug 397187] Re: [karmic] udev requires new kernel, breaks on EC2

You are correct that there will not be a change to karmic's udev to
support running on top of a kernel without signalfd support. Thus, karmic
images will not run with older kernel images.

> I assume that "won't fix" does not mean that the plan is to not build a
> Karmic kernel.

Also correct. We're working on getting both Jaunty and Karmic kernel
images.

Revision history for this message
Scott Moser (smoser) wrote :

I'm re-opening this bug. We'll leave it open until there is a solution provided.

The solution is likely to be karmic kernels on amazon.

Changed in ubuntu-on-ec2:
status: Won't Fix → Confirmed
Revision history for this message
Eric Hammond (esh) wrote :

The fix for this will be LP#418130 (building a Karmic kernel), so I think it's appropriate to call this a duplicate.

Revision history for this message
C.J. Collier (cjcollier) wrote :

This bug is not EC2-specific. This effects all Xen HVM virtual machines.

Revision history for this message
C.J. Collier (cjcollier) wrote :

Let me clarify: all Xen HVM virtual machines which depend on the paravirt network driver (only available in the 2.6.18 kernel, AFAICT)

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.