KillFilter is unable to kill a process if the underlying file has been deleted

[Lars Kellogg-Stedman]

I was seeing the following error from neutron's dhcp-agent:

  2015-08-06 13:20:06.507 16186 DEBUG neutron.agent.linux.utils [-] Running command (rootwrap daemon): ['kill', '-9', '23224'] execute_rootwrap_daemon /usr/lib/python2.7/site-packages/neutron/agent/linux/utils.py:100
  2015-08-06 13:20:06.778 16186 ERROR neutron.agent.dhcp.agent [-] Unable to enable dhcp for ff809ade-4935-48cb-8c1e-657a9ca58ca7.
  [...]
  2015-08-06 13:20:06.778 16186 TRACE neutron.agent.dhcp.agent raise convert_to_error(kind, result)
  2015-08-06 13:20:06.778 16186 TRACE neutron.agent.dhcp.agent NoFilterMatched

That pid (23224) does refer to an active dnsmasq process:

  # ps -p 23224 -f|cat
  UID PID PPID C STIME TTY TIME CMD
  nobody 23224 1 0 09:31 ? 00:00:00 dnsmasq --no-hosts --no-resolv --strict-order --bind-interfaces ...

But the "dnsmasq" package had been updated after that process started,
so /proc/23224/exe is:

  lrwxrwxrwx. 1 root root 0 Aug 6 09:31 /proc/23224/exe -> /usr/sbin/dnsmasq;55c362c5 (deleted)

KillFilter should probably be able to deal with this situation.

[Lars Kellogg-Stedman (larsks)]

Okay, so the problem here isn't just a *deleted* file; it's one that is first *renamed*, then deleted.

[Pádraig Brady (p-draigbrady)]

Interesting. This should have been fixed with bug 1179793 back before oslo.rootwrap was split out?

[Lars Kellogg-Stedman (larsks)]

No, bug 1179793 is addressing a different problem.

In this case, the file is first *renamed* (by rpm, in this case), then deleted. The renamed file does not match the KillFilter specified values.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/210541

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/210541
Committed: https://git.openstack.org/cgit/openstack/oslo.rootwrap/commit/?id=e5f9a393ced38e2bccfd5c626cebcb036192be71
Submitter: Jenkins
Branch: master

commit e5f9a393ced38e2bccfd5c626cebcb036192be71
Author: Pádraig Brady <email address hidden>
Date: Fri Aug 7 17:12:25 2015 +0100

    Handle renamed executables with KillFilter

    A running process may refer to a renamed executable
    referenced by /proc/$pid/exe, which happens with
    updated executables on RPM based systems.
    In that case we defer to the path in /proc/$pid/cmdline
    if it exists.

    Change-Id: I113f2f8ebd56f3b05c420310c4b1e84ba6f17fcf
    Closes-Bug: 1482316

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/oslo.rootwrap 2.3.0 release.