[PATCH net-next 0/4] batch calls to fib_flush and arp_ifdown

STALE3868d

Revision v1 of 3 in this series.

8 messages, 2 authors, 2016-01-12 · open the first message on its own page

[PATCH net-next 0/4] batch calls to fib_flush and arp_ifdown

From: Salam Noureddine <hidden>
Date: 2016-01-04 23:18:03

fib_flush walks the whole fib in a net_namespace and is called for
each net_device being closed or unregistered. This can be very expensive
when dealing with 100k or more routes in the fib and removal of a lot
of interfaces. These four patches deal with this issue by calling fib_flush
just once for each net namespace and introduce a new function arp_ifdown_all
that does a similar optimization for the neighbour table.

Salam Noureddine (4):
  net: add event_list to struct net and provide utility functions
  net: dev: add batching to net_device notifiers
  net: core: introduce neigh_ifdown_all for all down interfaces
  net: fib: avoid calling fib_flush for each device when doing batch
    close     and unregister

 include/linux/netdevice.h   |  2 ++
 include/net/arp.h           |  1 +
 include/net/neighbour.h     |  1 +
 include/net/net_namespace.h | 22 ++++++++++++++++++++++
 include/net/netns/ipv4.h    |  1 +
 net/core/dev.c              | 39 ++++++++++++++++++++++++++++++++++++---
 net/core/neighbour.c        | 38 +++++++++++++++++++++++++++++++-------
 net/ipv4/arp.c              |  4 ++++
 net/ipv4/fib_frontend.c     | 16 ++++++++++++++--
 9 files changed, 112 insertions(+), 12 deletions(-)

-- 
1.8.1.4

[PATCH net-next 2/4] net: dev: add batching to net_device notifiers

From: Salam Noureddine <hidden>
Date: 2016-01-04 23:18:02

This can be used to optimize bringing down and unregsitering
net_devices by running certain cleanup operations only on the
net namespace instead of on each net_device.

Signed-off-by: Salam Noureddine <redacted>
---
 include/linux/netdevice.h |  2 ++
 net/core/dev.c            | 39 ++++++++++++++++++++++++++++++++++++---
 2 files changed, 38 insertions(+), 3 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c20b814..1b12269 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h

@@ -2183,6 +2183,8 @@ struct netdev_lag_lower_state_info {
 #define NETDEV_BONDING_INFO	0x0019
 #define NETDEV_PRECHANGEUPPER	0x001A
 #define NETDEV_CHANGELOWERSTATE	0x001B
+#define NETDEV_UNREGISTER_BATCH	0x001C
+#define NETDEV_DOWN_BATCH	0x001D
 
 int register_netdevice_notifier(struct notifier_block *nb);
 int unregister_netdevice_notifier(struct notifier_block *nb);

diff --git a/net/core/dev.c b/net/core/dev.c
index 914b4a2..77410a3 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c

@@ -1439,11 +1439,16 @@ static int __dev_close(struct net_device *dev)
 int dev_close_many(struct list_head *head, bool unlink)
 {
 	struct net_device *dev, *tmp;
+	struct net *net, *net_tmp;
+	LIST_HEAD(net_head);
 
 	/* Remove the devices that don't need to be closed */
-	list_for_each_entry_safe(dev, tmp, head, close_list)
+	list_for_each_entry_safe(dev, tmp, head, close_list) {
 		if (!(dev->flags & IFF_UP))
 			list_del_init(&dev->close_list);
+		else
+			net_add_event_list(&net_head, dev_net(dev));
+	}
 
 	__dev_close_many(head);

@@ -1454,6 +1459,11 @@ int dev_close_many(struct list_head *head, bool unlink)
 			list_del_init(&dev->close_list);
 	}
 
+	list_for_each_entry_safe(net, net_tmp, &net_head, event_list) {
+		call_netdevice_notifiers(NETDEV_DOWN_BATCH, net->loopback_dev);
+		net_del_event_list(net);
+	}
+
 	return 0;
 }
 EXPORT_SYMBOL(dev_close_many);

@@ -1572,8 +1582,12 @@ rollback:
 				call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
 							dev);
 				call_netdevice_notifier(nb, NETDEV_DOWN, dev);
+				call_netdevice_notifier(nb, NETDEV_DOWN_BATCH,
+							dev);
 			}
 			call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
+			call_netdevice_notifier(nb, NETDEV_UNREGISTER_BATCH,
+						dev);
 		}
 	}

@@ -1614,8 +1628,12 @@ int unregister_netdevice_notifier(struct notifier_block *nb)
 				call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
 							dev);
 				call_netdevice_notifier(nb, NETDEV_DOWN, dev);
+				call_netdevice_notifier(nb, NETDEV_DOWN_BATCH,
+							dev);
 			}
 			call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
+			call_netdevice_notifier(nb, NETDEV_UNREGISTER_BATCH,
+						dev);
 		}
 	}
 unlock:

@@ -6187,10 +6205,12 @@ void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
 
 	if (changes & IFF_UP) {
-		if (dev->flags & IFF_UP)
+		if (dev->flags & IFF_UP) {
 			call_netdevice_notifiers(NETDEV_UP, dev);
-		else
+		} else {
 			call_netdevice_notifiers(NETDEV_DOWN, dev);
+			call_netdevice_notifiers(NETDEV_DOWN_BATCH, dev);
+		}
 	}
 
 	if (dev->flags & IFF_UP &&

@@ -6427,7 +6447,9 @@ static void net_set_todo(struct net_device *dev)
 static void rollback_registered_many(struct list_head *head)
 {
 	struct net_device *dev, *tmp;
+	struct net *net, *net_tmp;
 	LIST_HEAD(close_head);
+	LIST_HEAD(net_head);
 
 	BUG_ON(dev_boot_phase);
 	ASSERT_RTNL();

@@ -6504,6 +6526,15 @@ static void rollback_registered_many(struct list_head *head)
 #endif
 	}
 
+	list_for_each_entry(dev, head, unreg_list) {
+		net_add_event_list(&net_head, dev_net(dev));
+	}
+	list_for_each_entry_safe(net, net_tmp, &net_head, event_list) {
+		call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH,
+					 net->loopback_dev);
+		net_del_event_list(net);
+	}
+
 	synchronize_net();
 
 	list_for_each_entry(dev, head, unreg_list)

@@ -7065,6 +7096,7 @@ static void netdev_wait_allrefs(struct net_device *dev)
 
 			/* Rebroadcast unregister notification */
 			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
+			call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
 
 			__rtnl_unlock();
 			rcu_barrier();

@@ -7581,6 +7613,7 @@ int dev_change_net_namespace(struct net_device *dev, struct net *net, const char
 	   the device is just moving and can keep their slaves up.
 	*/
 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
+	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
 	rcu_barrier();
 	call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
 	rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);

-- 
1.8.1.4

[PATCH net-next 4/4] net: fib: avoid calling fib_flush for each device when doing batch close and unregister

From: Salam Noureddine <hidden>
Date: 2016-01-04 23:18:03

Call fib_flush at the end when closing or unregistering multiple
devices. This can save walking the fib many times and greatly
reduce rtnl_lock hold time when unregistering many devices with
a fib having hundreds of thousands of routes.

Signed-off-by: Salam Noureddine <redacted>
---
 include/net/netns/ipv4.h |  1 +
 net/ipv4/fib_frontend.c  | 16 ++++++++++++++--
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index d75be32..d59a078 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h

@@ -111,5 +111,6 @@ struct netns_ipv4 {
 #endif
 #endif
 	atomic_t	rt_genid;
+	bool		needs_fib_flush;
 };
 #endif

diff --git a/net/ipv4/fib_frontend.c b/net/ipv4/fib_frontend.c
index 4734475..808426e 100644
--- a/net/ipv4/fib_frontend.c
+++ b/net/ipv4/fib_frontend.c

@@ -1161,11 +1161,22 @@ static int fib_netdev_event(struct notifier_block *this, unsigned long event, vo
 	unsigned int flags;
 
 	if (event == NETDEV_UNREGISTER) {
-		fib_disable_ip(dev, event, true);
+		if (fib_sync_down_dev(dev, event, true))
+			net->ipv4.needs_fib_flush = true;
 		rt_flush_dev(dev);
 		return NOTIFY_DONE;
 	}
 
+	if (event == NETDEV_UNREGISTER_BATCH || event == NETDEV_DOWN_BATCH) {
+		if (net->ipv4.needs_fib_flush) {
+			fib_flush(net);
+			net->ipv4.needs_fib_flush = false;
+		}
+		rt_cache_flush(net);
+		arp_ifdown_all();
+		return NOTIFY_DONE;
+	}
+
 	in_dev = __in_dev_get_rtnl(dev);
 	if (!in_dev)
 		return NOTIFY_DONE;

@@ -1182,7 +1193,8 @@ static int fib_netdev_event(struct notifier_block *this, unsigned long event, vo
 		rt_cache_flush(net);
 		break;
 	case NETDEV_DOWN:
-		fib_disable_ip(dev, event, false);
+		if (fib_sync_down_dev(dev, event, false))
+			net->ipv4.needs_fib_flush = true;
 		break;
 	case NETDEV_CHANGE:
 		flags = dev_get_flags(dev);

-- 
1.8.1.4

[PATCH net-next 3/4] net: core: introduce neigh_ifdown_all for all down interfaces

From: Salam Noureddine <hidden>
Date: 2016-01-04 23:18:03

This cleans up neighbour entries for all interfaces in the down
state, avoiding walking the whole neighbour table for each interface
being brought down.

Signed-off-by: Salam Noureddine <redacted>
---
 include/net/arp.h       |  1 +
 include/net/neighbour.h |  1 +
 net/core/neighbour.c    | 38 +++++++++++++++++++++++++++++++-------
 net/ipv4/arp.c          |  4 ++++
 4 files changed, 37 insertions(+), 7 deletions(-)

diff --git a/include/net/arp.h b/include/net/arp.h
index 5e0f891..0efee66 100644
--- a/include/net/arp.h
+++ b/include/net/arp.h

@@ -43,6 +43,7 @@ void arp_send(int type, int ptype, __be32 dest_ip,
 	      const unsigned char *src_hw, const unsigned char *th);
 int arp_mc_map(__be32 addr, u8 *haddr, struct net_device *dev, int dir);
 void arp_ifdown(struct net_device *dev);
+void arp_ifdown_all(void);
 
 struct sk_buff *arp_create(int type, int ptype, __be32 dest_ip,
 			   struct net_device *dev, __be32 src_ip,

diff --git a/include/net/neighbour.h b/include/net/neighbour.h
index 8b68384..8785d7b 100644
--- a/include/net/neighbour.h
+++ b/include/net/neighbour.h

@@ -318,6 +318,7 @@ int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new, u32 flags);
 void __neigh_set_probe_once(struct neighbour *neigh);
 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev);
 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
+int neigh_ifdown_all(struct neigh_table *tbl);
 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb);
 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb);
 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb);

diff --git a/net/core/neighbour.c b/net/core/neighbour.c
index f18ae91..bfbd97a 100644
--- a/net/core/neighbour.c
+++ b/net/core/neighbour.c

@@ -54,7 +54,8 @@ do {						\
 static void neigh_timer_handler(unsigned long arg);
 static void __neigh_notify(struct neighbour *n, int type, int flags);
 static void neigh_update_notify(struct neighbour *neigh);
-static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
+static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
+			 bool all_down);
 
 #ifdef CONFIG_PROC_FS
 static const struct file_operations neigh_stat_seq_fops;

@@ -192,7 +193,8 @@ static void pneigh_queue_purge(struct sk_buff_head *list)
 	}
 }
 
-static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
+static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
+			    bool all_down)
 {
 	int i;
 	struct neigh_hash_table *nht;

@@ -210,6 +212,12 @@ static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 				np = &n->next;
 				continue;
 			}
+			if (!dev && n->dev && all_down) {
+				if (n->dev->flags & IFF_UP) {
+					np = &n->next;
+					continue;
+				}
+			}
 			rcu_assign_pointer(*np,
 				   rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));

@@ -245,7 +253,7 @@ static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 {
 	write_lock_bh(&tbl->lock);
-	neigh_flush_dev(tbl, dev);
+	neigh_flush_dev(tbl, dev, false);
 	write_unlock_bh(&tbl->lock);
 }
 EXPORT_SYMBOL(neigh_changeaddr);

@@ -253,8 +261,8 @@ EXPORT_SYMBOL(neigh_changeaddr);
 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 {
 	write_lock_bh(&tbl->lock);
-	neigh_flush_dev(tbl, dev);
-	pneigh_ifdown(tbl, dev);
+	neigh_flush_dev(tbl, dev, false);
+	pneigh_ifdown(tbl, dev, false);
 	write_unlock_bh(&tbl->lock);
 
 	del_timer_sync(&tbl->proxy_timer);

@@ -263,6 +271,19 @@ int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 }
 EXPORT_SYMBOL(neigh_ifdown);
 
+int neigh_ifdown_all(struct neigh_table *tbl)
+{
+	write_lock_bh(&tbl->lock);
+	neigh_flush_dev(tbl, NULL, true);
+	pneigh_ifdown(tbl, NULL, true);
+	write_unlock_bh(&tbl->lock);
+
+	del_timer_sync(&tbl->proxy_timer);
+	pneigh_queue_purge(&tbl->proxy_queue);
+	return 0;
+}
+EXPORT_SYMBOL(neigh_ifdown_all);
+
 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
 {
 	struct neighbour *n = NULL;

@@ -645,7 +666,8 @@ int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 	return -ENOENT;
 }
 
-static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
+static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
+			 bool all_down)
 {
 	struct pneigh_entry *n, **np;
 	u32 h;

@@ -653,7 +675,9 @@ static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 	for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 		np = &tbl->phash_buckets[h];
 		while ((n = *np) != NULL) {
-			if (!dev || n->dev == dev) {
+			if ((!dev && !all_down) || (all_down && n->dev &&
+						    !(n->dev->flags & IFF_UP)) ||
+			    n->dev == dev) {
 				*np = n->next;
 				if (tbl->pdestructor)
 					tbl->pdestructor(n);

diff --git a/net/ipv4/arp.c b/net/ipv4/arp.c
index 59b3e0e..1328244 100644
--- a/net/ipv4/arp.c
+++ b/net/ipv4/arp.c

@@ -1219,6 +1219,10 @@ void arp_ifdown(struct net_device *dev)
 	neigh_ifdown(&arp_tbl, dev);
 }
 
+void arp_ifdown_all(void)
+{
+	neigh_ifdown_all(&arp_tbl);
+}
 
 /*
  *	Called once on startup.

-- 
1.8.1.4

[PATCH net-next 1/4] net: add event_list to struct net and provide utility functions

From: Salam Noureddine <hidden>
Date: 2016-01-04 23:18:03

Signed-off-by: Salam Noureddine <redacted>
---
 include/net/net_namespace.h | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h
index 4089abc..4cf47de 100644
--- a/include/net/net_namespace.h
+++ b/include/net/net_namespace.h

@@ -58,6 +58,7 @@ struct net {
 	struct list_head	list;		/* list of network namespaces */
 	struct list_head	cleanup_list;	/* namespaces on death row */
 	struct list_head	exit_list;	/* Use only net_mutex */
+	struct list_head	event_list;     /* net_device notifier list */
 
 	struct user_namespace   *user_ns;	/* Owning user namespace */
 	spinlock_t		nsid_lock;

@@ -380,4 +381,25 @@ static inline void fnhe_genid_bump(struct net *net)
 	atomic_inc(&net->fnhe_genid);
 }
 
+#ifdef CONFIG_NET_NS
+static inline void net_add_event_list(struct list_head *head, struct net *net)
+{
+	if (!list_empty(&net->event_list))
+		list_add_tail(&net->event_list, head);
+}
+
+static inline void net_del_event_list(struct net *net)
+{
+	list_del_init(&net->event_list);
+}
+#else
+static inline void net_add_event_list(struct list_head *head, struct net *net)
+{
+}
+
+static inline void net_del_event_list(struct net *net)
+{
+}
+#endif
+
 #endif /* __NET_NET_NAMESPACE_H */

-- 
1.8.1.4

Re: [PATCH net-next 0/4] batch calls to fib_flush and arp_ifdown

From: Eric W. Biederman <hidden>
Date: 2016-01-05 00:44:07

Salam Noureddine [off-list ref] writes:

fib_flush walks the whole fib in a net_namespace and is called for
each net_device being closed or unregistered. This can be very expensive
when dealing with 100k or more routes in the fib and removal of a lot
of interfaces. These four patches deal with this issue by calling fib_flush
just once for each net namespace and introduce a new function arp_ifdown_all
that does a similar optimization for the neighbour table.

Two things would be very valuable with this patchset.

Some numbers on how much your changes have improved the code in the case
you care about.  I suspect the improvements are not subtle so this
should not be hard.

Can you please provide a justification for event_list.  Just skimming
through it appears that event_list because a duplicate of the list of
batched network devices that are passed to dev_close_many and friends.
If the list is actually a duplicate it appears foolish to create it.

Eric

Salam Noureddine (4):
  net: add event_list to struct net and provide utility functions
  net: dev: add batching to net_device notifiers
  net: core: introduce neigh_ifdown_all for all down interfaces
  net: fib: avoid calling fib_flush for each device when doing batch
    close     and unregister

 include/linux/netdevice.h   |  2 ++
 include/net/arp.h           |  1 +
 include/net/neighbour.h     |  1 +
 include/net/net_namespace.h | 22 ++++++++++++++++++++++
 include/net/netns/ipv4.h    |  1 +
 net/core/dev.c              | 39 ++++++++++++++++++++++++++++++++++++---
 net/core/neighbour.c        | 38 +++++++++++++++++++++++++++++++-------
 net/ipv4/arp.c              |  4 ++++
 net/ipv4/fib_frontend.c     | 16 ++++++++++++++--
 9 files changed, 112 insertions(+), 12 deletions(-)

Re: [PATCH net-next 0/4] batch calls to fib_flush and arp_ifdown

From: Salam Noureddine <hidden>
Date: 2016-01-05 01:12:10

On Mon, Jan 4, 2016 at 4:35 PM, Eric W. Biederman [off-list ref] wrote:

Two things would be very valuable with this patchset.

Some numbers on how much your changes have improved the code in the case
you care about.  I suspect the improvements are not subtle so this
should not be hard.

Can you please provide a justification for event_list.  Just skimming
through it appears that event_list because a duplicate of the list of
batched network devices that are passed to dev_close_many and friends.
If the list is actually a duplicate it appears foolish to create it.

Eric

The performance test I ran tries to unregister 1000 dummy interfaces
with 512K routes in the fib.
Without the patch I could unregister 35 interfaces per second and with
the patch it jumped to 620
interfaces per second. 512K is a lot of routes but I am assuming we
would get a good improvement
even with 100K routes in the fib.

I am using event_list to put all the net namespaces in the current
net_device batch on a list and only
call the NETDEV_UNREGISTER_BATCH on those namespaces. It would be
possible to just call the
notifier for NETDEV_DOWN/UNREGISTER_BATCH for all the devices on the
list and rely on the
needs_fib_flush flag to only call fib_flush once per namespace but it
seems like a waste to me.

Salam

Re: [PATCH net-next 0/4] batch calls to fib_flush and arp_ifdown

From: Salam Noureddine <hidden>
Date: 2016-01-12 01:53:17

Any thoughts on the correctness of the patchset? Or a different way to
solve this issue leading to long rtnl_lock hold time?

Thanks,

Salam

On Mon, Jan 4, 2016 at 5:12 PM, Salam Noureddine [off-list ref] wrote:

On Mon, Jan 4, 2016 at 4:35 PM, Eric W. Biederman [off-list ref] wrote:

quoted

Two things would be very valuable with this patchset.

Some numbers on how much your changes have improved the code in the case
you care about.  I suspect the improvements are not subtle so this
should not be hard.

Can you please provide a justification for event_list.  Just skimming
through it appears that event_list because a duplicate of the list of
batched network devices that are passed to dev_close_many and friends.
If the list is actually a duplicate it appears foolish to create it.

Eric

The performance test I ran tries to unregister 1000 dummy interfaces
with 512K routes in the fib.
Without the patch I could unregister 35 interfaces per second and with
the patch it jumped to 620
interfaces per second. 512K is a lot of routes but I am assuming we
would get a good improvement
even with 100K routes in the fib.

I am using event_list to put all the net namespaces in the current
net_device batch on a list and only
call the NETDEV_UNREGISTER_BATCH on those namespaces. It would be
possible to just call the
notifier for NETDEV_DOWN/UNREGISTER_BATCH for all the devices on the
list and rely on the
needs_fib_flush flag to only call fib_flush once per namespace but it
seems like a waste to me.

Salam

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help