[PATCH] Ensure that dhcp client is exited

[Mikhail Efremov <mikhefr gmail com>]

Hello!

Since NM does not stop dhcp client when it exits, there may persist an
old instance of dhcp client during NM start again and it is not a child
of the current process. So we need to wait until it is actually exited,
otherwise a race condition may occurs, as I noticed in case dhcpcd
4.x.x (the pidfile was removed by exiting old dhcpcd instance).

-- 
WBR, Mikhail Efremov

>From 45e528745f7ec9e6cd4683ffc4bf1e5ee40fc394 Mon Sep 17 00:00:00 2001
From: Mikhail Efremov <sem altlinux org>
Date: Thu, 11 Nov 2010 20:07:32 +0300
Subject: [PATCH] dhcp: Ensure that dhcp client is exited.

Do not start another dhcp client until existing one
really exited.
---
 src/dhcp-manager/nm-dhcp-client.c |   12 ++++++++----
 1 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/src/dhcp-manager/nm-dhcp-client.c b/src/dhcp-manager/nm-dhcp-client.c
index 0c7f3d4..3a8b194 100644
--- a/src/dhcp-manager/nm-dhcp-client.c
+++ b/src/dhcp-manager/nm-dhcp-client.c
@@ -156,11 +156,15 @@ stop_process (GPid pid, const char *iface)
 
 		if (ret == -1) {
 			/* Child already exited */
-			if (errno == ECHILD)
+			if (errno == ECHILD) {
+				/* Was it really our child and it exited? */
+				if (kill (pid, 0) < 0 && errno == ESRCH)
+					break;
+			} else {
+				/* Took too long; shoot it in the head */
+				i = 0;
 				break;
-			/* Took too long; shoot it in the head */
-			i = 0;
-			break;
+			}
 		}
 		g_usleep (G_USEC_PER_SEC / 5);
 	}
-- 
1.7.3.2