Do not attempt to kill already-dead dnsmasq

Check that the dnsmasq process is running (and actually looks like a
dnsmasq process) before attempting to kill it. Fixes bug 1010275.

Change-Id: Ib49209e1624dfb30470adbe13d7fc045ec1fdf83

1 files changed, 8 insertions, 1 deletions

diff --git a/nova/network/linux_net.py b/nova/network/linux_net.py
index 811395f9c..484320261 100755
--- a/nova/network/linux_net.py
+++ b/nova/network/linux_net.py
@@ -660,7 +660,14 @@ def update_dhcp_hostfile_with_text(dev, hosts_text):
 def kill_dhcp(dev):
     pid = _dnsmasq_pid_for(dev)
     if pid:
-        _execute('kill', '-9', pid, run_as_root=True)
+        # Check that the process exists and looks like a dnsmasq process
+        conffile = _dhcp_file(dev, 'conf')
+        out, _err = _execute('cat', '/proc/%d/cmdline' % pid,
+                             check_exit_code=False)
+        if conffile.split('/')[-1] in out:
+            _execute('kill', '-9', pid, run_as_root=True)
+        else:
+            LOG.debug(_('Pid %d is stale, skip killing dnsmasq'), pid)
 
 
 # NOTE(ja): Sending a HUP only reloads the hostfile, so any