OpenVZ Forum


Home » Mailing lists » Devel » [PATCH net-2.6.26 0/6][NETNS][SOCK]: Make "prot inuse" counters work per-net.
[PATCH net-2.6.26 3/6][NETNS][SOCK]: Swap unhash and net change for icmp and tcp service sockets. [message #28734 is a reply to message #28731] Thu, 27 March 2008 08:16 Go to previous messageGo to previous message
Pavel Emelianov is currently offline  Pavel Emelianov
Messages: 1149
Registered: September 2006
Senior Member
This is tricky. 

The problem is that these sockets are created in init_net not to pin
the new one and then are moved to it with sk_change_net. _After_ this
the socket is unhashed.

The problem is that unhash will decrement the inuse counter for the
socket in a new net, but no in the one it was incremented (init_net).
This will lead to RAW inuse counter increment each time a new net is
created and will be negative for the new net.

Swap unhashing with net changing so that RAW inuse is decremented in
the proper namespace.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>

---
 net/ipv4/icmp.c  |    2 +-
 net/ipv6/icmp.c  |    2 +-
 net/ipv6/mcast.c |    2 +-
 net/ipv6/ndisc.c |    2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/icmp.c b/net/ipv4/icmp.c
index 3697e05..74eb856 100644
--- a/net/ipv4/icmp.c
+++ b/net/ipv4/icmp.c
@@ -1160,7 +1160,6 @@ int __net_init icmp_sk_init(struct net *net)
 			goto fail;
 
 		net->ipv4.icmp_sk[i] = sk = sock->sk;
-		sk_change_net(sk, net);
 
 		sk->sk_allocation = GFP_ATOMIC;
 
@@ -1179,6 +1178,7 @@ int __net_init icmp_sk_init(struct net *net)
 		 * packets.
 		 */
 		sk->sk_prot->unhash(sk);
+		sk_change_net(sk, net);
 	}
 
 	/* Control parameters for ECHO replies. */
diff --git a/net/ipv6/icmp.c b/net/ipv6/icmp.c
index 63309d1..fec5d55 100644
--- a/net/ipv6/icmp.c
+++ b/net/ipv6/icmp.c
@@ -820,7 +820,6 @@ static int __net_init icmpv6_sk_init(struct net *net)
 		}
 
 		net->ipv6.icmp_sk[i] = sk = sock->sk;
-		sk_change_net(sk, net);
 
 		sk->sk_allocation = GFP_ATOMIC;
 		/*
@@ -839,6 +838,7 @@ static int __net_init icmpv6_sk_init(struct net *net)
 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
 
 		sk->sk_prot->unhash(sk);
+		sk_change_net(sk, net);
 	}
 	return 0;
 
diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
index d810cff..8fbfffc 100644
--- a/net/ipv6/mcast.c
+++ b/net/ipv6/mcast.c
@@ -2686,9 +2686,9 @@ static int igmp6_net_init(struct net *net)
 	}
 
 	net->ipv6.igmp_sk = sk = sock->sk;
-	sk_change_net(sk, net);
 	sk->sk_allocation = GFP_ATOMIC;
 	sk->sk_prot->unhash(sk);
+	sk_change_net(sk, net);
 
 	np = inet6_sk(sk);
 	np->hop_limit = 1;
diff --git a/net/ipv6/ndisc.c b/net/ipv6/ndisc.c
index b4d8e33..a13ec57 100644
--- a/net/ipv6/ndisc.c
+++ b/net/ipv6/ndisc.c
@@ -1731,7 +1731,6 @@ static int ndisc_net_init(struct net *net)
 	}
 
 	net->ipv6.ndisc_sk = sk = sock->sk;
-	sk_change_net(sk, net);
 
 	np = inet6_sk(sk);
 	sk->sk_allocation = GFP_ATOMIC;
@@ -1739,6 +1738,7 @@ static int ndisc_net_init(struct net *net)
 	/* Do not loopback ndisc messages */
 	np->mc_loop = 0;
 	sk->sk_prot->unhash(sk);
+	sk_change_net(sk, net);
 
 	return 0;
 }
-- 
1.5.3.4
 
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Previous Topic: [PATCH 2/2 net-2.6.26] [NETNS]: Do no include NET related headers if CONFIG_NET is not set.
Next Topic: [RFC][-mm] [1/2] Simple stats for cpu resource controller
Goto Forum:
  


Current Time: Tue Aug 26 22:06:08 GMT 2025

Total time taken to generate the page: 0.08861 seconds