starting dnsmasq should be protected by semaphore

[Mandar Vaze]

Scenario :
========

while enable_dhcp() method in nova/network/quantum/manager.py is protected using @utils.synchronized decorator, update_dhcp() is NOT.

Thus multiple update_dhcp and a enable_dhcp method could be executed on same time. This could cause problem when starting dnsmasq

Following steps are done in both enable_dhcp() and update_dhcp()
1) update_dhcp_hostfile_with_texts
2) kill_dhcp (only in update_dhcp - Not done in enable_dhcp)
3) restart_dhcp

Steps 1,2,3 should be executed atomically.

Proposed Solution :
================

Create new private method inside update_dhcp:

@utils.synchronized('quantum-update-dhcp')
def _safe_restart_dhcp(self, dev, hosts,network_ref)
    self.driver.update_dhcp_hostfile_with_text(dev, hosts)
    # Restart dnsmasq
    self.driver.kill_dhcp(dev)
    self.driver.restart_dhcp(context, dev, network_ref)

Now Invoke the above method from update_dhcp() instead of three separate steps.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/6481

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/6481
Committed: http://github.com/openstack/nova/commit/72d0299d438b9bb25fba3fe4a2f5ade8544055ae
Submitter: Jenkins
Branch: master

commit 72d0299d438b9bb25fba3fe4a2f5ade8544055ae
Author: Mandar Vaze <email address hidden>
Date: Thu Apr 12 05:45:24 2012 -0700

    Introduced _atomic_restart_dhcp() Fixes Bug 977875

    Introduced a new method which executes update_dhcp_text_file
    and restart_dhcp as one atomic unit - to avoid race between these
    steps.

    Change-Id: I6fda9fbfcaf59fa6a4859b361574d98e16331882

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/essex
Review: https://review.openstack.org/7268