[PATCH 0/4] MIB: add struct net to UDP accounting macros

[PATCH 0/4] MIB: add struct net to UDP accounting macros

[Pavel Emelyanov]

This is the first small set of MIB statistics netnsization. The easiest
case is UDP stats, so I started with this one. If this set is accepted,
I will go on step-by-step with adding struct net to all the other stats' 
accounting macros, then SNMP_XXX ones and finish with a set than will put 
the stats on the struct net and fix appropriate proc files.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Denis V. Lunev <den@openvz.org>

[PATCH net-next-2.6 1/4] MIB: add struct net to UDP_INC_STATS_USER

[Pavel Emelyanov]

Nothing special - all the places already have a struct sock
at hands, so use the sock_net() net.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Denis V. Lunev <den@openvz.org>

---
 include/net/udp.h |    2 +-
 net/ipv4/udp.c    |   11 +++++++----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/include/net/udp.h b/include/net/udp.h
index 7a86848..1331909 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -158,7 +158,7 @@ DECLARE_SNMP_STAT(struct udp_mib, udplite_stats_in6);
 /*
  * 	SNMP statistics for UDP and UDP-Lite
  */
-#define UDP_INC_STATS_USER(field, is_udplite)			       do {   \
+#define UDP_INC_STATS_USER(net, field, is_udplite)	      do { (void)net; \
 	if (is_udplite) SNMP_INC_STATS_USER(udplite_statistics, field);       \
 	else		SNMP_INC_STATS_USER(udp_statistics, field);  }  while(0)
 #define UDP_INC_STATS_BH(field, is_udplite) 			       do  {  \
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 3bbf6fb..c97da53 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -526,7 +526,8 @@ out:
 	up->len = 0;
 	up->pending = 0;
 	if (!err)
-		UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
+		UDP_INC_STATS_USER(sock_net(sk),
+				UDP_MIB_OUTDATAGRAMS, is_udplite);
 	return err;
 }
 
@@ -725,7 +726,8 @@ out:
 	 * seems like overkill.
 	 */
 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
-		UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
+		UDP_INC_STATS_USER(sock_net(sk),
+				UDP_MIB_SNDBUFERRORS, is_udplite);
 	}
 	return err;
 
@@ -888,7 +890,8 @@ try_again:
 		goto out_free;
 
 	if (!peeked)
-		UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
+		UDP_INC_STATS_USER(sock_net(sk),
+				UDP_MIB_INDATAGRAMS, is_udplite);
 
 	sock_recv_timestamp(msg, sk, skb);
 
@@ -917,7 +920,7 @@ out:
 csum_copy_err:
 	lock_sock(sk);
 	if (!skb_kill_datagram(sk, skb, flags))
-		UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
+		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 	release_sock(sk);
 
 	if (noblock)
-- 
1.5.3.4

[PATCH 2/4] MIB: add struct net to UDP_INC_STATS_BH

[Pavel Emelyanov]

Two special cases here - one is rxrpc - I put init_net there
explicitly, since we haven't touched this part yet. The second
place is in __udp4_lib_rcv - we already have a struct net there,
but I have to move its initialization above to make it ready
at the "drop" label.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Denis V. Lunev <den@openvz.org>

---
 include/net/udp.h    |    6 +++---
 net/ipv4/udp.c       |   18 ++++++++++--------
 net/rxrpc/ar-input.c |    5 +++--
 3 files changed, 16 insertions(+), 13 deletions(-)

diff --git a/include/net/udp.h b/include/net/udp.h
index 1331909..7b18cfb 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -161,7 +161,7 @@ DECLARE_SNMP_STAT(struct udp_mib, udplite_stats_in6);
 #define UDP_INC_STATS_USER(net, field, is_udplite)	      do { (void)net; \
 	if (is_udplite) SNMP_INC_STATS_USER(udplite_statistics, field);       \
 	else		SNMP_INC_STATS_USER(udp_statistics, field);  }  while(0)
-#define UDP_INC_STATS_BH(field, is_udplite) 			       do  {  \
+#define UDP_INC_STATS_BH(net, field, is_udplite) 	      do { (void)net; \
 	if (is_udplite) SNMP_INC_STATS_BH(udplite_statistics, field);         \
 	else		SNMP_INC_STATS_BH(udp_statistics, field);    }  while(0)
 
@@ -176,12 +176,12 @@ DECLARE_SNMP_STAT(struct udp_mib, udplite_stats_in6);
 #define UDPX_INC_STATS_BH(sk, field) \
 	do { \
 		if ((sk)->sk_family == AF_INET) \
-			UDP_INC_STATS_BH(field, 0); \
+			UDP_INC_STATS_BH(sock_net(sk), field, 0); \
 		else \
 			UDP6_INC_STATS_BH(field, 0); \
 	} while (0);
 #else
-#define UDPX_INC_STATS_BH(sk, field) UDP_INC_STATS_BH(field, 0)
+#define UDPX_INC_STATS_BH(sk, field) UDP_INC_STATS_BH(sock_net(sk), field, 0)
 #endif
 
 /* /proc */
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index c97da53..7187121 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -991,7 +991,8 @@ int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
 
 			ret = (*up->encap_rcv)(sk, skb);
 			if (ret <= 0) {
-				UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS,
+				UDP_INC_STATS_BH(sock_net(sk),
+						 UDP_MIB_INDATAGRAMS,
 						 is_udplite);
 				return -ret;
 			}
@@ -1044,7 +1045,8 @@ int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
 		/* Note that an ENOMEM error is charged twice */
 		if (rc == -ENOMEM) {
-			UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
+			UDP_INC_STATS_BH(sock_net(sk),
+					UDP_MIB_RCVBUFERRORS, is_udplite);
 			atomic_inc(&sk->sk_drops);
 		}
 		goto drop;
@@ -1053,7 +1055,7 @@ int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
 	return 0;
 
 drop:
-	UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
+	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 	kfree_skb(skb);
 	return -1;
 }
@@ -1161,7 +1163,7 @@ int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
 	struct rtable *rt = (struct rtable*)skb->dst;
 	__be32 saddr = ip_hdr(skb)->saddr;
 	__be32 daddr = ip_hdr(skb)->daddr;
-	struct net *net;
+	struct net *net = dev_net(skb->dev);
 
 	/*
 	 *  Validate the packet.
@@ -1183,7 +1185,6 @@ int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
 	if (udp4_csum_init(skb, uh, proto))
 		goto csum_error;
 
-	net = dev_net(skb->dev);
 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
 		return __udp4_lib_mcast_deliver(net, skb, uh,
 				saddr, daddr, udptable);
@@ -1217,7 +1218,7 @@ int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
 	if (udp_lib_checksum_complete(skb))
 		goto csum_error;
 
-	UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
+	UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 
 	/*
@@ -1251,7 +1252,7 @@ csum_error:
 		       ntohs(uh->dest),
 		       ulen);
 drop:
-	UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
+	UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
 	kfree_skb(skb);
 	return 0;
 }
@@ -1458,7 +1459,8 @@ unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
 		spin_lock_bh(&rcvq->lock);
 		while ((skb = skb_peek(rcvq)) != NULL &&
 		       udp_lib_checksum_complete(skb)) {
-			UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
+			UDP_INC_STATS_BH(sock_net(sk),
+					UDP_MIB_INERRORS, is_lite);
 			__skb_unlink(skb, rcvq);
 			kfree_skb(skb);
 		}
diff --git a/net/rxrpc/ar-input.c b/net/rxrpc/ar-input.c
index f8a699e..f98c802 100644
--- a/net/rxrpc/ar-input.c
+++ b/net/rxrpc/ar-input.c
@@ -21,6 +21,7 @@
 #include <net/af_rxrpc.h>
 #include <net/ip.h>
 #include <net/udp.h>
+#include <net/net_namespace.h>
 #include "ar-internal.h"
 
 unsigned long rxrpc_ack_timeout = 1;
@@ -708,12 +709,12 @@ void rxrpc_data_ready(struct sock *sk, int count)
 	if (skb_checksum_complete(skb)) {
 		rxrpc_free_skb(skb);
 		rxrpc_put_local(local);
-		UDP_INC_STATS_BH(UDP_MIB_INERRORS, 0);
+		UDP_INC_STATS_BH(&init_net, UDP_MIB_INERRORS, 0);
 		_leave(" [CSUM failed]");
 		return;
 	}
 
-	UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, 0);
+	UDP_INC_STATS_BH(&init_net, UDP_MIB_INDATAGRAMS, 0);
 
 	/* the socket buffer we have is owned by UDP, with UDP's data all over
 	 * it, but we really want our own */
-- 
1.5.3.4

[PATCH net-next 3/4] MIB: add struct net to UDP6_INC_STATS_USER

[Pavel Emelyanov]

As simple as the patch #1 in this set.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Denis V. Lunev <den@openvz.org>

---
 include/net/udp.h |    2 +-
 net/ipv6/udp.c    |   11 +++++++----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/include/net/udp.h b/include/net/udp.h
index 7b18cfb..bb5b9ec 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -168,7 +168,7 @@ DECLARE_SNMP_STAT(struct udp_mib, udplite_stats_in6);
 #define UDP6_INC_STATS_BH(field, is_udplite) 			      do  {  \
 	if (is_udplite) SNMP_INC_STATS_BH(udplite_stats_in6, field);         \
 	else		SNMP_INC_STATS_BH(udp_stats_in6, field);    } while(0)
-#define UDP6_INC_STATS_USER(field, is_udplite)			       do {    \
+#define UDP6_INC_STATS_USER(net, field, is_udplite)	    do { (void)net;    \
 	if (is_udplite) SNMP_INC_STATS_USER(udplite_stats_in6, field);         \
 	else		SNMP_INC_STATS_USER(udp_stats_in6, field);    } while(0)
 
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index f91e1df..833f715 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -166,7 +166,8 @@ try_again:
 		goto out_free;
 
 	if (!peeked)
-		UDP6_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
+		UDP6_INC_STATS_USER(sock_net(sk),
+				UDP_MIB_INDATAGRAMS, is_udplite);
 
 	sock_recv_timestamp(msg, sk, skb);
 
@@ -213,7 +214,7 @@ out:
 csum_copy_err:
 	lock_sock(sk);
 	if (!skb_kill_datagram(sk, skb, flags))
-		UDP6_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
+		UDP6_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 	release_sock(sk);
 
 	if (flags & MSG_DONTWAIT)
@@ -591,7 +592,8 @@ out:
 	up->len = 0;
 	up->pending = 0;
 	if (!err)
-		UDP6_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
+		UDP6_INC_STATS_USER(sock_net(sk),
+				UDP_MIB_OUTDATAGRAMS, is_udplite);
 	return err;
 }
 
@@ -873,7 +875,8 @@ out:
 	 * seems like overkill.
 	 */
 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
-		UDP6_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
+		UDP6_INC_STATS_USER(sock_net(sk),
+				UDP_MIB_SNDBUFERRORS, is_udplite);
 	}
 	return err;
 
-- 
1.5.3.4

[PATCH 4/4] MIB: add struct net to UDP6_INC_STATS_BH

[Pavel Emelyanov]

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Denis V. Lunev <den@openvz.org>


---
 include/net/udp.h |    4 ++--
 net/ipv6/udp.c    |   13 +++++++------
 2 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/include/net/udp.h b/include/net/udp.h
index bb5b9ec..3e55159 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -165,7 +165,7 @@ DECLARE_SNMP_STAT(struct udp_mib, udplite_stats_in6);
 	if (is_udplite) SNMP_INC_STATS_BH(udplite_statistics, field);         \
 	else		SNMP_INC_STATS_BH(udp_statistics, field);    }  while(0)
 
-#define UDP6_INC_STATS_BH(field, is_udplite) 			      do  {  \
+#define UDP6_INC_STATS_BH(net, field, is_udplite) 	    do { (void)net;  \
 	if (is_udplite) SNMP_INC_STATS_BH(udplite_stats_in6, field);         \
 	else		SNMP_INC_STATS_BH(udp_stats_in6, field);    } while(0)
 #define UDP6_INC_STATS_USER(net, field, is_udplite)	    do { (void)net;    \
@@ -178,7 +178,7 @@ DECLARE_SNMP_STAT(struct udp_mib, udplite_stats_in6);
 		if ((sk)->sk_family == AF_INET) \
 			UDP_INC_STATS_BH(sock_net(sk), field, 0); \
 		else \
-			UDP6_INC_STATS_BH(field, 0); \
+			UDP6_INC_STATS_BH(sock_net(sk), field, 0); \
 	} while (0);
 #else
 #define UDPX_INC_STATS_BH(sk, field) UDP_INC_STATS_BH(sock_net(sk), field, 0)
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 833f715..d1477b3 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -299,7 +299,8 @@ int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
 		/* Note that an ENOMEM error is charged twice */
 		if (rc == -ENOMEM) {
-			UDP6_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
+			UDP6_INC_STATS_BH(sock_net(sk),
+					UDP_MIB_RCVBUFERRORS, is_udplite);
 			atomic_inc(&sk->sk_drops);
 		}
 		goto drop;
@@ -307,7 +308,7 @@ int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
 
 	return 0;
 drop:
-	UDP6_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
+	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 	kfree_skb(skb);
 	return -1;
 }
@@ -439,7 +440,7 @@ int __udp6_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
 	struct net_device *dev = skb->dev;
 	struct in6_addr *saddr, *daddr;
 	u32 ulen = 0;
-	struct net *net;
+	struct net *net = dev_net(skb->dev);
 
 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
 		goto short_packet;
@@ -474,7 +475,6 @@ int __udp6_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
 	if (udp6_csum_init(skb, uh, proto))
 		goto discard;
 
-	net = dev_net(skb->dev);
 	/*
 	 *	Multicast receive code
 	 */
@@ -497,7 +497,8 @@ int __udp6_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
 
 		if (udp_lib_checksum_complete(skb))
 			goto discard;
-		UDP6_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
+		UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
+				proto == IPPROTO_UDPLITE);
 
 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
 
@@ -522,7 +523,7 @@ short_packet:
 		       ulen, skb->len);
 
 discard:
-	UDP6_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
+	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
 	kfree_skb(skb);
 	return 0;
 }
-- 
1.5.3.4

Re: [PATCH 0/4] MIB: add struct net to UDP accounting macros

[David Miller]

From: Pavel Emelyanov <xemul@openvz.org>
Date: Fri, 04 Jul 2008 15:58:44 +0400

> This is the first small set of MIB statistics netnsization. The easiest
> case is UDP stats, so I started with this one. If this set is accepted,
> I will go on step-by-step with adding struct net to all the other stats' 
> accounting macros, then SNMP_XXX ones and finish with a set than will put 
> the stats on the struct net and fix appropriate proc files.
> 
> Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
> Acked-by: Denis V. Lunev <den@openvz.org>

Applied, thanks Pavel.

Are we going to provide some way for an administrator to fetch
stats from the perspective of all namespaces?  I know the idea
is seperation with this stuff, but admins are going to want
something like that.

Re: [PATCH 0/4] MIB: add struct net to UDP accounting macros

[Pavel Emelyanov]

David Miller wrote:
> From: Pavel Emelyanov <xemul@openvz.org>
> Date: Fri, 04 Jul 2008 15:58:44 +0400
> 
>> This is the first small set of MIB statistics netnsization. The easiest
>> case is UDP stats, so I started with this one. If this set is accepted,
>> I will go on step-by-step with adding struct net to all the other stats' 
>> accounting macros, then SNMP_XXX ones and finish with a set than will put 
>> the stats on the struct net and fix appropriate proc files.
>>
>> Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
>> Acked-by: Denis V. Lunev <den@openvz.org>
> 
> Applied, thanks Pavel.
> 
> Are we going to provide some way for an administrator to fetch
> stats from the perspective of all namespaces?  I know the idea
> is seperation with this stuff, but admins are going to want
> something like that.

Well, if we want to get the stats for each namespace separately, then
this ability is already present. Since this statistics is shown via the
/proc/net files and the /proc/net itself is seen via the /proc/<pid>/net,
then we can walk the init-s of all the containers in the system and dump
this info.

The problem that is to be solved with this approach is how to get these
init-s :) But since finding any namespace by some task living in it is a
common practice now (netdev moving, sys_hijack) this one will be solved.

BTW, are there some plans about implementing some netlink-based fetcher
of this statistics? If so, then I think it's worth making this engine
namespaces aware from the very beginning.

Thanks,
Pavel

Re: [PATCH 0/4] MIB: add struct net to UDP accounting macros

[David Miller]

From: Pavel Emelyanov <xemul@openvz.org>
Date: Mon, 07 Jul 2008 12:13:27 +0400

> BTW, are there some plans about implementing some netlink-based fetcher
> of this statistics? If so, then I think it's worth making this engine
> namespaces aware from the very beginning.

I'm not aware of any, or whether that would make any sense.

Thomas?

Re: [PATCH 0/4] MIB: add struct net to UDP accounting macros

[Thomas Graf]

* David Miller <davem@davemloft.net> 2008-07-07 01:47
> From: Pavel Emelyanov <xemul@openvz.org>
> Date: Mon, 07 Jul 2008 12:13:27 +0400
> 
> > BTW, are there some plans about implementing some netlink-based fetcher
> > of this statistics? If so, then I think it's worth making this engine
> > namespaces aware from the very beginning.
> 
> I'm not aware of any, or whether that would make any sense.

There have been plans but none was brave enough so far.

Re: [PATCH 0/4] MIB: add struct net to UDP accounting macros

[Daniel Lezcano]

Pavel Emelyanov wrote:
> David Miller wrote:
>> From: Pavel Emelyanov <xemul@openvz.org>
>> Date: Fri, 04 Jul 2008 15:58:44 +0400
>>
>>> This is the first small set of MIB statistics netnsization. The easiest
>>> case is UDP stats, so I started with this one. If this set is accepted,
>>> I will go on step-by-step with adding struct net to all the other stats' 
>>> accounting macros, then SNMP_XXX ones and finish with a set than will put 
>>> the stats on the struct net and fix appropriate proc files.
>>>
>>> Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
>>> Acked-by: Denis V. Lunev <den@openvz.org>
>> Applied, thanks Pavel.
>>
>> Are we going to provide some way for an administrator to fetch
>> stats from the perspective of all namespaces?  I know the idea
>> is seperation with this stuff, but admins are going to want
>> something like that.
> 
> Well, if we want to get the stats for each namespace separately, then
> this ability is already present. Since this statistics is shown via the
> /proc/net files and the /proc/net itself is seen via the /proc/<pid>/net,
> then we can walk the init-s of all the containers in the system and dump
> this info.
> 
> The problem that is to be solved with this approach is how to get these
> init-s :) But since finding any namespace by some task living in it is a
> common practice now (netdev moving, sys_hijack) this one will be solved.
> 
> BTW, are there some plans about implementing some netlink-based fetcher
> of this statistics? If so, then I think it's worth making this engine
> namespaces aware from the very beginning.


Shouldn't be interesting to handle the network namespaces directly with 
the iproute command ?

  * ip netns add <name>
  * ip netns del <name>
  * ip netns <oldname> set name <newname>
  * ip netns show

When a network namespace is created via clone|unshare, the name is 
automatically the pid of the creator.

In order to track the namespace, we can add an entry in /proc/net (aka 
/proc/<pid>/net) named "name" which contains the name of the namespace.
I heard Eric is thinking about a netnsfs.

 From a larger perspective, the iproute command with a new "netns" 
subcommand can be enhanced to have more features, for example the freeze 
and resume for the network.

  * ip netns set down <name>
  * ip netns set up <name>

So having the netns binded, we can plug the known subcommand (link, ip, 
...) with the netns features. For examples:

  * ip addr add 1.2.3.4/24 dev eth0 netns foo

Global namespace naming, and monitoring.

[Eric W. Biederman]

Daniel Lezcano <dlezcano@fr.ibm.com> writes:

>> Well, if we want to get the stats for each namespace separately, then
>> this ability is already present. Since this statistics is shown via the
>> /proc/net files and the /proc/net itself is seen via the /proc/<pid>/net,
>> then we can walk the init-s of all the containers in the system and dump
>> this info.
>>
>> The problem that is to be solved with this approach is how to get these
>> init-s :) But since finding any namespace by some task living in it is a
>> common practice now (netdev moving, sys_hijack) this one will be solved.

Yes.  Finding the which the a single process in each namespace to look
at (the init-s) is something we have yet to refine.

The model for multiple namespace monitoring is definitely having
filesystems mounted that we can look at to get all of the information
we care about.  /proc does a lot of this today, and with some cleanups
it should be able to display per namespace sysctls and a few other
goodies. 

...

There is one class of user that we have yet to find a good solution for.
The people who want to use isolated network stacks within a single application.
Usually because there are duplicate routes between the stacks.  In that case
indirect through processes falls down, as does being able to create a socket in
one namespace.

My latest brainstorm comes from asking how the problem would have been
solved in plan9.  The idea is to create a filesystem we can mount that
holds a reference to a netns (netnsfs).  Using a mounted filesystem
like that is a bit heavy weight, but just referring to it's the root
directory is enough to give us a name in the mount namespace.  The
auto unmount and consequent release of the network namespace when the
mount namespace goes away is attractive.

> Shouldn't be interesting to handle the network namespaces directly with the
> iproute command ?
> 
>  * ip netns add <name>
>  * ip netns del <name>
>  * ip netns <oldname> set name <newname>
>  * ip netns show

Ugh.  You have to be really careful with proposal like this to ensure that
they wind up in some namespace.  Otherwise you have created a new global
namespace, and have made nesting of containers much harder to implement.

> So having the netns binded, we can plug the known subcommand (link, ip, ...)
> with the netns features. For examples:
>
>  * ip addr add 1.2.3.4/24 dev eth0 netns foo

The only thing limiting that today is that we need someway to get a netlink
socket on the namespace in question.  Get me in a perverse mood and I will
grab the netlink socket from one of the init-s with ptrace.

Eric