[PATCH net-next 00/11] net: speedup netns dismantles

[PATCH net-next 00/11] net: speedup netns dismantles

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

In this series, I made network namespace deletions more scalable,
by 4x using the benchmark described in this cover letter.

- Remove bottleneck on ipv6 addrconf, by replacing a global
  hash table to a per netns one.

- rtnl mutex can be heavily contended, and is a blocker for
  the specialized kernel thread responsible for cleaning and
  freeing network namespaces.
  Rework eight (struct pernet_operations *)->exit() handlers to
  exit_batch() ones. This removes many rtnl acquisitions,
  and gives to cleanup_net() kind of a priority over rtnl
  ownership, when processing batches.

Tested on a host with 24 cpus (48 HT)

Test script:

for nr in {1..10}
do
  (for i in {1..10000}; do unshare -n /bin/bash -c "ifconfig lo up"; done) &
done
wait

for i in {1..10}
do
  sleep 1 
  echo 3 >/proc/sys/vm/drop_caches
  grep net_namespace /proc/slabinfo
done

Before this series:
We can see host struggles to clean the netns, even after there are
no new creations. Memory cost is high, because each netns consumes
a good amount of memory.

time ./unshare10.sh
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      82634  82634   3968    1    1 : tunables   24   12    8 : slabdata  82634  82634      0
net_namespace      37214  37792   3968    1    1 : tunables   24   12    8 : slabdata  37214  37792    192

real	6m57.766s
user	3m37.277s
sys	40m4.826s

After this series:
We can see the script completes much faster, the kernel thread
doing the cleanup_net() keeps up just fine.
Memory cost is not too big.

time ./unshare10.sh
net_namespace       9945   9945   4096    1    1 : tunables   24   12    8 : slabdata   9945   9945      0
net_namespace       4087   4665   4096    1    1 : tunables   24   12    8 : slabdata   4087   4665    192
net_namespace       4082   4607   4096    1    1 : tunables   24   12    8 : slabdata   4082   4607    192
net_namespace        234    761   4096    1    1 : tunables   24   12    8 : slabdata    234    761    192
net_namespace        224    751   4096    1    1 : tunables   24   12    8 : slabdata    224    751    192
net_namespace        218    745   4096    1    1 : tunables   24   12    8 : slabdata    218    745    192
net_namespace        193    667   4096    1    1 : tunables   24   12    8 : slabdata    193    667    172
net_namespace        167    609   4096    1    1 : tunables   24   12    8 : slabdata    167    609    152
net_namespace        167    609   4096    1    1 : tunables   24   12    8 : slabdata    167    609    152
net_namespace        157    609   4096    1    1 : tunables   24   12    8 : slabdata    157    609    152

real    1m43.876s
user    3m39.728s
sys 7m36.342s


Eric Dumazet (11):
  ipv6/addrconf: allocate a per netns hash table
  ipv6/addrconf: use one delayed work per netns
  ipv6/addrconf: switch to per netns inet6_addr_lst hash table
  nexthop: change nexthop_net_exit() to nexthop_net_exit_batch()
  ipv4: add fib_net_exit_batch()
  ipv6: change fib6_rules_net_exit() to batch mode
  ip6mr: introduce ip6mr_net_exit_batch()
  ipmr: introduce ipmr_net_exit_batch()
  can: gw: switch cangw_pernet_exit() to batch mode
  bonding: switch bond_net_exit() to batch mode
  net: remove default_device_exit()

 drivers/net/bonding/bond_main.c   |  27 ++++--
 drivers/net/bonding/bond_procfs.c |   1 -
 include/net/netns/ipv6.h          |   5 ++
 net/can/gw.c                      |   9 +-
 net/core/dev.c                    |  22 +++--
 net/ipv4/fib_frontend.c           |  19 +++-
 net/ipv4/ipmr.c                   |  20 +++--
 net/ipv4/nexthop.c                |  12 ++-
 net/ipv6/addrconf.c               | 139 ++++++++++++++----------------
 net/ipv6/fib6_rules.c             |  11 ++-
 net/ipv6/ip6mr.c                  |  20 +++--
 11 files changed, 172 insertions(+), 113 deletions(-)

-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 01/11] ipv6/addrconf: allocate a per netns hash table

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

Add a per netns hash table and a dedicated spinlock,
first step to get rid of the global inet6_addr_lst[] one.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/netns/ipv6.h |  4 ++++
 net/ipv6/addrconf.c      | 20 ++++++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/include/net/netns/ipv6.h b/include/net/netns/ipv6.h
index 30cdfc4e1615424b1c691b53499a1987d7fd0496..755f12001c8b2a73ad1895e73c7aebcba67c6728 100644
--- a/include/net/netns/ipv6.h
+++ b/include/net/netns/ipv6.h
@@ -92,6 +92,10 @@ struct netns_ipv6 {
 	struct sock             *tcp_sk;
 	struct sock             *igmp_sk;
 	struct sock		*mc_autojoin_sk;
+
+	struct hlist_head	*inet6_addr_lst;
+	spinlock_t		addrconf_hash_lock;
+
 #ifdef CONFIG_IPV6_MROUTE
 #ifndef CONFIG_IPV6_MROUTE_MULTIPLE_TABLES
 	struct mr_table		*mrt6;
diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index ef23e7dc538ad983a28853865dd4281f7f0ea8de..cda9e59cab4343507f670e7f59e2b72fd3cded0f 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -7111,6 +7111,13 @@ static int __net_init addrconf_init_net(struct net *net)
 	int err = -ENOMEM;
 	struct ipv6_devconf *all, *dflt;
 
+	spin_lock_init(&net->ipv6.addrconf_hash_lock);
+	net->ipv6.inet6_addr_lst = kcalloc(IN6_ADDR_HSIZE,
+					   sizeof(struct hlist_head),
+					   GFP_KERNEL);
+	if (!net->ipv6.inet6_addr_lst)
+		goto err_alloc_addr;
+
 	all = kmemdup(&ipv6_devconf, sizeof(ipv6_devconf), GFP_KERNEL);
 	if (!all)
 		goto err_alloc_all;
@@ -7172,11 +7179,15 @@ static int __net_init addrconf_init_net(struct net *net)
 err_alloc_dflt:
 	kfree(all);
 err_alloc_all:
+	kfree(net->ipv6.inet6_addr_lst);
+err_alloc_addr:
 	return err;
 }
 
 static void __net_exit addrconf_exit_net(struct net *net)
 {
+	int i;
+
 #ifdef CONFIG_SYSCTL
 	__addrconf_sysctl_unregister(net, net->ipv6.devconf_dflt,
 				     NETCONFA_IFINDEX_DEFAULT);
@@ -7187,6 +7198,15 @@ static void __net_exit addrconf_exit_net(struct net *net)
 	net->ipv6.devconf_dflt = NULL;
 	kfree(net->ipv6.devconf_all);
 	net->ipv6.devconf_all = NULL;
+
+	/*
+	 *	Check hash table, then free it.
+	 */
+	for (i = 0; i < IN6_ADDR_HSIZE; i++)
+		WARN_ON_ONCE(!hlist_empty(&net->ipv6.inet6_addr_lst[i]));
+
+	kfree(net->ipv6.inet6_addr_lst);
+	net->ipv6.inet6_addr_lst = NULL;
 }
 
 static struct pernet_operations addrconf_ops = {
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 02/11] ipv6/addrconf: use one delayed work per netns

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

Next step for using per netns inet6_addr_lst
is to have per netns work item to ultimately
call addrconf_verify_rtnl() and addrconf_verify()
with a new 'struct net*' argument.

Everything is still using the global inet6_addr_lst[] table.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/netns/ipv6.h |  1 +
 net/ipv6/addrconf.c      | 44 ++++++++++++++++++++++------------------
 2 files changed, 25 insertions(+), 20 deletions(-)

diff --git a/include/net/netns/ipv6.h b/include/net/netns/ipv6.h
index 755f12001c8b2a73ad1895e73c7aebcba67c6728..d145f196668240bf5c3e509255f3f9b06f0e91bc 100644
--- a/include/net/netns/ipv6.h
+++ b/include/net/netns/ipv6.h
@@ -95,6 +95,7 @@ struct netns_ipv6 {
 
 	struct hlist_head	*inet6_addr_lst;
 	spinlock_t		addrconf_hash_lock;
+	struct delayed_work	addr_chk_work;
 
 #ifdef CONFIG_IPV6_MROUTE
 #ifndef CONFIG_IPV6_MROUTE_MULTIPLE_TABLES
diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index cda9e59cab4343507f670e7f59e2b72fd3cded0f..dab291cd39ba20c8dad29854297e8699c067b1e7 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -152,12 +152,10 @@ static int ipv6_generate_stable_address(struct in6_addr *addr,
 static struct hlist_head inet6_addr_lst[IN6_ADDR_HSIZE];
 static DEFINE_SPINLOCK(addrconf_hash_lock);
 
-static void addrconf_verify(void);
-static void addrconf_verify_rtnl(void);
-static void addrconf_verify_work(struct work_struct *);
+static void addrconf_verify(struct net *net);
+static void addrconf_verify_rtnl(struct net *net);
 
 static struct workqueue_struct *addrconf_wq;
-static DECLARE_DELAYED_WORK(addr_chk_work, addrconf_verify_work);
 
 static void addrconf_join_anycast(struct inet6_ifaddr *ifp);
 static void addrconf_leave_anycast(struct inet6_ifaddr *ifp);
@@ -2675,7 +2673,7 @@ int addrconf_prefix_rcv_add_addr(struct net *net, struct net_device *dev,
 				 create, now);
 
 		in6_ifa_put(ifp);
-		addrconf_verify();
+		addrconf_verify(net);
 	}
 
 	return 0;
@@ -2987,7 +2985,7 @@ static int inet6_addr_add(struct net *net, int ifindex,
 			manage_tempaddrs(idev, ifp, cfg->valid_lft,
 					 cfg->preferred_lft, true, jiffies);
 		in6_ifa_put(ifp);
-		addrconf_verify_rtnl();
+		addrconf_verify_rtnl(net);
 		return 0;
 	} else if (cfg->ifa_flags & IFA_F_MCAUTOJOIN) {
 		ipv6_mc_config(net->ipv6.mc_autojoin_sk, false,
@@ -3027,7 +3025,7 @@ static int inet6_addr_del(struct net *net, int ifindex, u32 ifa_flags,
 				manage_tempaddrs(idev, ifp, 0, 0, false,
 						 jiffies);
 			ipv6_del_addr(ifp);
-			addrconf_verify_rtnl();
+			addrconf_verify_rtnl(net);
 			if (ipv6_addr_is_multicast(pfx)) {
 				ipv6_mc_config(net->ipv6.mc_autojoin_sk,
 					       false, pfx, dev->ifindex);
@@ -4246,7 +4244,7 @@ static void addrconf_dad_completed(struct inet6_ifaddr *ifp, bool bump_id,
 	 * before this temporary address becomes deprecated.
 	 */
 	if (ifp->flags & IFA_F_TEMPORARY)
-		addrconf_verify_rtnl();
+		addrconf_verify_rtnl(dev_net(dev));
 }
 
 static void addrconf_dad_run(struct inet6_dev *idev, bool restart)
@@ -4484,7 +4482,7 @@ int ipv6_chk_rpl_srh_loop(struct net *net, const struct in6_addr *segs,
  *	Periodic address status verification
  */
 
-static void addrconf_verify_rtnl(void)
+static void addrconf_verify_rtnl(struct net *net)
 {
 	unsigned long now, next, next_sec, next_sched;
 	struct inet6_ifaddr *ifp;
@@ -4496,7 +4494,7 @@ static void addrconf_verify_rtnl(void)
 	now = jiffies;
 	next = round_jiffies_up(now + ADDR_CHECK_FREQUENCY);
 
-	cancel_delayed_work(&addr_chk_work);
+	cancel_delayed_work(&net->ipv6.addr_chk_work);
 
 	for (i = 0; i < IN6_ADDR_HSIZE; i++) {
 restart:
@@ -4599,20 +4597,23 @@ static void addrconf_verify_rtnl(void)
 
 	pr_debug("now = %lu, schedule = %lu, rounded schedule = %lu => %lu\n",
 		 now, next, next_sec, next_sched);
-	mod_delayed_work(addrconf_wq, &addr_chk_work, next_sched - now);
+	mod_delayed_work(addrconf_wq, &net->ipv6.addr_chk_work, next_sched - now);
 	rcu_read_unlock_bh();
 }
 
 static void addrconf_verify_work(struct work_struct *w)
 {
+	struct net *net = container_of(to_delayed_work(w), struct net,
+				       ipv6.addr_chk_work);
+
 	rtnl_lock();
-	addrconf_verify_rtnl();
+	addrconf_verify_rtnl(net);
 	rtnl_unlock();
 }
 
-static void addrconf_verify(void)
+static void addrconf_verify(struct net *net)
 {
-	mod_delayed_work(addrconf_wq, &addr_chk_work, 0);
+	mod_delayed_work(addrconf_wq, &net->ipv6.addr_chk_work, 0);
 }
 
 static struct in6_addr *extract_addr(struct nlattr *addr, struct nlattr *local,
@@ -4708,7 +4709,8 @@ static int modify_prefix_route(struct inet6_ifaddr *ifp,
 	return 0;
 }
 
-static int inet6_addr_modify(struct inet6_ifaddr *ifp, struct ifa6_config *cfg)
+static int inet6_addr_modify(struct net *net, struct inet6_ifaddr *ifp,
+			     struct ifa6_config *cfg)
 {
 	u32 flags;
 	clock_t expires;
@@ -4822,7 +4824,7 @@ static int inet6_addr_modify(struct inet6_ifaddr *ifp, struct ifa6_config *cfg)
 				 jiffies);
 	}
 
-	addrconf_verify_rtnl();
+	addrconf_verify_rtnl(net);
 
 	return 0;
 }
@@ -4909,7 +4911,7 @@ inet6_rtm_newaddr(struct sk_buff *skb, struct nlmsghdr *nlh,
 	    !(nlh->nlmsg_flags & NLM_F_REPLACE))
 		err = -EEXIST;
 	else
-		err = inet6_addr_modify(ifa, &cfg);
+		err = inet6_addr_modify(net, ifa, &cfg);
 
 	in6_ifa_put(ifa);
 
@@ -5794,7 +5796,7 @@ static int inet6_set_iftoken(struct inet6_dev *idev, struct in6_addr *token,
 
 	write_unlock_bh(&idev->lock);
 	inet6_ifinfo_notify(RTM_NEWLINK, idev);
-	addrconf_verify_rtnl();
+	addrconf_verify_rtnl(dev_net(dev));
 	return 0;
 }
 
@@ -7112,6 +7114,7 @@ static int __net_init addrconf_init_net(struct net *net)
 	struct ipv6_devconf *all, *dflt;
 
 	spin_lock_init(&net->ipv6.addrconf_hash_lock);
+	INIT_DEFERRABLE_WORK(&net->ipv6.addr_chk_work, addrconf_verify_work);
 	net->ipv6.inet6_addr_lst = kcalloc(IN6_ADDR_HSIZE,
 					   sizeof(struct hlist_head),
 					   GFP_KERNEL);
@@ -7199,6 +7202,7 @@ static void __net_exit addrconf_exit_net(struct net *net)
 	kfree(net->ipv6.devconf_all);
 	net->ipv6.devconf_all = NULL;
 
+	cancel_delayed_work(&net->ipv6.addr_chk_work);
 	/*
 	 *	Check hash table, then free it.
 	 */
@@ -7281,7 +7285,7 @@ int __init addrconf_init(void)
 
 	register_netdevice_notifier(&ipv6_dev_notf);
 
-	addrconf_verify();
+	addrconf_verify(&init_net);
 
 	rtnl_af_register(&inet6_ops);
 
@@ -7364,7 +7368,7 @@ void addrconf_cleanup(void)
 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
 		WARN_ON(!hlist_empty(&inet6_addr_lst[i]));
 	spin_unlock_bh(&addrconf_hash_lock);
-	cancel_delayed_work(&addr_chk_work);
+
 	rtnl_unlock();
 
 	destroy_workqueue(addrconf_wq);
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 03/11] ipv6/addrconf: switch to per netns inet6_addr_lst hash table

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

IPv6 does not scale very well with the number of IPv6 addresses.
It uses a global (shared by all netns) hash table with 256 buckets.

Some functions like addrconf_verify_rtnl() and addrconf_ifdown()
have to iterate all addresses in the hash table.

I have seen addrconf_verify_rtnl() holding the cpu for 10ms or more.

Switch to the per netns hashtable (and spinlock) added
in prior patches.

This considerably speeds up netns dismantle times on hosts
with thousands of netns.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv6/addrconf.c | 77 ++++++++++++++-------------------------------
 1 file changed, 23 insertions(+), 54 deletions(-)

diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index dab291cd39ba20c8dad29854297e8699c067b1e7..4f402bc38f056e08f3761e63a7bc7a51e54e9384 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -146,11 +146,6 @@ static int ipv6_generate_stable_address(struct in6_addr *addr,
 
 #define IN6_ADDR_HSIZE_SHIFT	8
 #define IN6_ADDR_HSIZE		(1 << IN6_ADDR_HSIZE_SHIFT)
-/*
- *	Configured unicast address hash table
- */
-static struct hlist_head inet6_addr_lst[IN6_ADDR_HSIZE];
-static DEFINE_SPINLOCK(addrconf_hash_lock);
 
 static void addrconf_verify(struct net *net);
 static void addrconf_verify_rtnl(struct net *net);
@@ -1009,9 +1004,7 @@ static bool ipv6_chk_same_addr(struct net *net, const struct in6_addr *addr,
 {
 	struct inet6_ifaddr *ifp;
 
-	hlist_for_each_entry(ifp, &inet6_addr_lst[hash], addr_lst) {
-		if (!net_eq(dev_net(ifp->idev->dev), net))
-			continue;
+	hlist_for_each_entry(ifp, &net->ipv6.inet6_addr_lst[hash], addr_lst) {
 		if (ipv6_addr_equal(&ifp->addr, addr)) {
 			if (!dev || ifp->idev->dev == dev)
 				return true;
@@ -1022,20 +1015,21 @@ static bool ipv6_chk_same_addr(struct net *net, const struct in6_addr *addr,
 
 static int ipv6_add_addr_hash(struct net_device *dev, struct inet6_ifaddr *ifa)
 {
-	unsigned int hash = inet6_addr_hash(dev_net(dev), &ifa->addr);
+	struct net *net = dev_net(dev);
+	unsigned int hash = inet6_addr_hash(net, &ifa->addr);
 	int err = 0;
 
-	spin_lock(&addrconf_hash_lock);
+	spin_lock(&net->ipv6.addrconf_hash_lock);
 
 	/* Ignore adding duplicate addresses on an interface */
-	if (ipv6_chk_same_addr(dev_net(dev), &ifa->addr, dev, hash)) {
+	if (ipv6_chk_same_addr(net, &ifa->addr, dev, hash)) {
 		netdev_dbg(dev, "ipv6_add_addr: already assigned\n");
 		err = -EEXIST;
 	} else {
-		hlist_add_head_rcu(&ifa->addr_lst, &inet6_addr_lst[hash]);
+		hlist_add_head_rcu(&ifa->addr_lst, &net->ipv6.inet6_addr_lst[hash]);
 	}
 
-	spin_unlock(&addrconf_hash_lock);
+	spin_unlock(&net->ipv6.addrconf_hash_lock);
 
 	return err;
 }
@@ -1259,9 +1253,10 @@ cleanup_prefix_route(struct inet6_ifaddr *ifp, unsigned long expires,
 
 static void ipv6_del_addr(struct inet6_ifaddr *ifp)
 {
-	int state;
 	enum cleanup_prefix_rt_t action = CLEANUP_PREFIX_RT_NOP;
+	struct net *net = dev_net(ifp->idev->dev);
 	unsigned long expires;
+	int state;
 
 	ASSERT_RTNL();
 
@@ -1273,9 +1268,9 @@ static void ipv6_del_addr(struct inet6_ifaddr *ifp)
 	if (state == INET6_IFADDR_STATE_DEAD)
 		goto out;
 
-	spin_lock_bh(&addrconf_hash_lock);
+	spin_lock_bh(&net->ipv6.addrconf_hash_lock);
 	hlist_del_init_rcu(&ifp->addr_lst);
-	spin_unlock_bh(&addrconf_hash_lock);
+	spin_unlock_bh(&net->ipv6.addrconf_hash_lock);
 
 	write_lock_bh(&ifp->idev->lock);
 
@@ -1918,10 +1913,8 @@ __ipv6_chk_addr_and_flags(struct net *net, const struct in6_addr *addr,
 	if (skip_dev_check)
 		dev = NULL;
 
-	hlist_for_each_entry_rcu(ifp, &inet6_addr_lst[hash], addr_lst) {
+	hlist_for_each_entry_rcu(ifp, &net->ipv6.inet6_addr_lst[hash], addr_lst) {
 		ndev = ifp->idev->dev;
-		if (!net_eq(dev_net(ndev), net))
-			continue;
 
 		if (l3mdev_master_dev_rcu(ndev) != l3mdev)
 			continue;
@@ -2025,9 +2018,7 @@ struct inet6_ifaddr *ipv6_get_ifaddr(struct net *net, const struct in6_addr *add
 	struct inet6_ifaddr *ifp, *result = NULL;
 
 	rcu_read_lock();
-	hlist_for_each_entry_rcu(ifp, &inet6_addr_lst[hash], addr_lst) {
-		if (!net_eq(dev_net(ifp->idev->dev), net))
-			continue;
+	hlist_for_each_entry_rcu(ifp, &net->ipv6.inet6_addr_lst[hash], addr_lst) {
 		if (ipv6_addr_equal(&ifp->addr, addr)) {
 			if (!dev || ifp->idev->dev == dev ||
 			    !(ifp->scope&(IFA_LINK|IFA_HOST) || strict)) {
@@ -2094,7 +2085,7 @@ static int addrconf_dad_end(struct inet6_ifaddr *ifp)
 void addrconf_dad_failure(struct sk_buff *skb, struct inet6_ifaddr *ifp)
 {
 	struct inet6_dev *idev = ifp->idev;
-	struct net *net = dev_net(ifp->idev->dev);
+	struct net *net = dev_net(idev->dev);
 
 	if (addrconf_dad_end(ifp)) {
 		in6_ifa_put(ifp);
@@ -3770,9 +3761,9 @@ static int addrconf_ifdown(struct net_device *dev, bool unregister)
 
 	/* Step 2: clear hash table */
 	for (i = 0; i < IN6_ADDR_HSIZE; i++) {
-		struct hlist_head *h = &inet6_addr_lst[i];
+		struct hlist_head *h = &net->ipv6.inet6_addr_lst[i];
 
-		spin_lock_bh(&addrconf_hash_lock);
+		spin_lock_bh(&net->ipv6.addrconf_hash_lock);
 restart:
 		hlist_for_each_entry_rcu(ifa, h, addr_lst) {
 			if (ifa->idev == idev) {
@@ -3788,7 +3779,7 @@ static int addrconf_ifdown(struct net_device *dev, bool unregister)
 				}
 			}
 		}
-		spin_unlock_bh(&addrconf_hash_lock);
+		spin_unlock_bh(&net->ipv6.addrconf_hash_lock);
 	}
 
 	write_lock_bh(&idev->lock);
@@ -4286,10 +4277,8 @@ static struct inet6_ifaddr *if6_get_first(struct seq_file *seq, loff_t pos)
 	}
 
 	for (; state->bucket < IN6_ADDR_HSIZE; ++state->bucket) {
-		hlist_for_each_entry_rcu(ifa, &inet6_addr_lst[state->bucket],
+		hlist_for_each_entry_rcu(ifa, &net->ipv6.inet6_addr_lst[state->bucket],
 					 addr_lst) {
-			if (!net_eq(dev_net(ifa->idev->dev), net))
-				continue;
 			/* sync with offset */
 			if (p < state->offset) {
 				p++;
@@ -4312,8 +4301,6 @@ static struct inet6_ifaddr *if6_get_next(struct seq_file *seq,
 	struct net *net = seq_file_net(seq);
 
 	hlist_for_each_entry_continue_rcu(ifa, addr_lst) {
-		if (!net_eq(dev_net(ifa->idev->dev), net))
-			continue;
 		state->offset++;
 		return ifa;
 	}
@@ -4321,9 +4308,7 @@ static struct inet6_ifaddr *if6_get_next(struct seq_file *seq,
 	state->offset = 0;
 	while (++state->bucket < IN6_ADDR_HSIZE) {
 		hlist_for_each_entry_rcu(ifa,
-				     &inet6_addr_lst[state->bucket], addr_lst) {
-			if (!net_eq(dev_net(ifa->idev->dev), net))
-				continue;
+				     &net->ipv6.inet6_addr_lst[state->bucket], addr_lst) {
 			return ifa;
 		}
 	}
@@ -4411,9 +4396,7 @@ int ipv6_chk_home_addr(struct net *net, const struct in6_addr *addr)
 	int ret = 0;
 
 	rcu_read_lock();
-	hlist_for_each_entry_rcu(ifp, &inet6_addr_lst[hash], addr_lst) {
-		if (!net_eq(dev_net(ifp->idev->dev), net))
-			continue;
+	hlist_for_each_entry_rcu(ifp, &net->ipv6.inet6_addr_lst[hash], addr_lst) {
 		if (ipv6_addr_equal(&ifp->addr, addr) &&
 		    (ifp->flags & IFA_F_HOMEADDRESS)) {
 			ret = 1;
@@ -4451,9 +4434,7 @@ int ipv6_chk_rpl_srh_loop(struct net *net, const struct in6_addr *segs,
 		hash = inet6_addr_hash(net, addr);
 
 		hash_found = false;
-		hlist_for_each_entry_rcu(ifp, &inet6_addr_lst[hash], addr_lst) {
-			if (!net_eq(dev_net(ifp->idev->dev), net))
-				continue;
+		hlist_for_each_entry_rcu(ifp, &net->ipv6.inet6_addr_lst[hash], addr_lst) {
 
 			if (ipv6_addr_equal(&ifp->addr, addr)) {
 				hash_found = true;
@@ -4498,7 +4479,7 @@ static void addrconf_verify_rtnl(struct net *net)
 
 	for (i = 0; i < IN6_ADDR_HSIZE; i++) {
 restart:
-		hlist_for_each_entry_rcu_bh(ifp, &inet6_addr_lst[i], addr_lst) {
+		hlist_for_each_entry_rcu_bh(ifp, &net->ipv6.inet6_addr_lst[i], addr_lst) {
 			unsigned long age;
 
 			/* When setting preferred_lft to a value not zero or
@@ -7233,7 +7214,7 @@ static struct rtnl_af_ops inet6_ops __read_mostly = {
 int __init addrconf_init(void)
 {
 	struct inet6_dev *idev;
-	int i, err;
+	int err;
 
 	err = ipv6_addr_label_init();
 	if (err < 0) {
@@ -7280,9 +7261,6 @@ int __init addrconf_init(void)
 
 	ip6_route_init_special_entries();
 
-	for (i = 0; i < IN6_ADDR_HSIZE; i++)
-		INIT_HLIST_HEAD(&inet6_addr_lst[i]);
-
 	register_netdevice_notifier(&ipv6_dev_notf);
 
 	addrconf_verify(&init_net);
@@ -7343,7 +7321,6 @@ int __init addrconf_init(void)
 void addrconf_cleanup(void)
 {
 	struct net_device *dev;
-	int i;
 
 	unregister_netdevice_notifier(&ipv6_dev_notf);
 	unregister_pernet_subsys(&addrconf_ops);
@@ -7361,14 +7338,6 @@ void addrconf_cleanup(void)
 	}
 	addrconf_ifdown(init_net.loopback_dev, true);
 
-	/*
-	 *	Check hash table.
-	 */
-	spin_lock_bh(&addrconf_hash_lock);
-	for (i = 0; i < IN6_ADDR_HSIZE; i++)
-		WARN_ON(!hlist_empty(&inet6_addr_lst[i]));
-	spin_unlock_bh(&addrconf_hash_lock);
-
 	rtnl_unlock();
 
 	destroy_workqueue(addrconf_wq);
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 04/11] nexthop: change nexthop_net_exit() to nexthop_net_exit_batch()

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

nexthop_net_exit() seems a good candidate for exit_batch(),
as this gives chance for cleanup_net() to progress much faster,
holding rtnl a bit longer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Ahern <dsahern@kernel.org>
---
 net/ipv4/nexthop.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c
index eeafeccebb8d44629085f504e26cb4e171c8c782..e459a391e607d34e16828bef13a5ce22ba6dde89 100644
--- a/net/ipv4/nexthop.c
+++ b/net/ipv4/nexthop.c
@@ -3733,12 +3733,16 @@ void nexthop_res_grp_activity_update(struct net *net, u32 id, u16 num_buckets,
 }
 EXPORT_SYMBOL(nexthop_res_grp_activity_update);
 
-static void __net_exit nexthop_net_exit(struct net *net)
+static void __net_exit nexthop_net_exit_batch(struct list_head *net_list)
 {
+	struct net *net;
+
 	rtnl_lock();
-	flush_all_nexthops(net);
+	list_for_each_entry(net, net_list, exit_list) {
+		flush_all_nexthops(net);
+		kfree(net->nexthop.devhash);
+	}
 	rtnl_unlock();
-	kfree(net->nexthop.devhash);
 }
 
 static int __net_init nexthop_net_init(struct net *net)
@@ -3756,7 +3760,7 @@ static int __net_init nexthop_net_init(struct net *net)
 
 static struct pernet_operations nexthop_net_ops = {
 	.init = nexthop_net_init,
-	.exit = nexthop_net_exit,
+	.exit_batch = nexthop_net_exit_batch,
 };
 
 static int __init nexthop_init(void)
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 05/11] ipv4: add fib_net_exit_batch()

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

Instead of acquiring rtnl at each fib_net_exit() invocation,
add fib_net_exit_batch() so that rtnl is acquired once.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Ahern <dsahern@kernel.org>
---
 net/ipv4/fib_frontend.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/net/ipv4/fib_frontend.c b/net/ipv4/fib_frontend.c
index 4d61ddd8a0ecfc4cc47b4802eb5a573beb84ee44..8c10f671d24db7f5751b6aed8e90a902bd1be5b4 100644
--- a/net/ipv4/fib_frontend.c
+++ b/net/ipv4/fib_frontend.c
@@ -1547,7 +1547,7 @@ static void ip_fib_net_exit(struct net *net)
 {
 	int i;
 
-	rtnl_lock();
+	ASSERT_RTNL();
 #ifdef CONFIG_IP_MULTIPLE_TABLES
 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
@@ -1572,7 +1572,7 @@ static void ip_fib_net_exit(struct net *net)
 #ifdef CONFIG_IP_MULTIPLE_TABLES
 	fib4_rules_exit(net);
 #endif
-	rtnl_unlock();
+
 	kfree(net->ipv4.fib_table_hash);
 	fib4_notifier_exit(net);
 }
@@ -1599,7 +1599,9 @@ static int __net_init fib_net_init(struct net *net)
 out_proc:
 	nl_fib_lookup_exit(net);
 out_nlfl:
+	rtnl_lock();
 	ip_fib_net_exit(net);
+	rtnl_unlock();
 	goto out;
 }
 
@@ -1607,12 +1609,23 @@ static void __net_exit fib_net_exit(struct net *net)
 {
 	fib_proc_exit(net);
 	nl_fib_lookup_exit(net);
-	ip_fib_net_exit(net);
+}
+
+static void __net_exit fib_net_exit_batch(struct list_head *net_list)
+{
+	struct net *net;
+
+	rtnl_lock();
+	list_for_each_entry(net, net_list, exit_list)
+		ip_fib_net_exit(net);
+
+	rtnl_unlock();
 }
 
 static struct pernet_operations fib_net_ops = {
 	.init = fib_net_init,
 	.exit = fib_net_exit,
+	.exit_batch = fib_net_exit_batch,
 };
 
 void __init ip_fib_init(void)
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 06/11] ipv6: change fib6_rules_net_exit() to batch mode

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

fib6_rules_net_exit() seems a good candidate for exit_batch(),
as this gives chance for cleanup_net() to progress much faster,
holding rtnl a bit longer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Ahern <dsahern@kernel.org>
---
 net/ipv6/fib6_rules.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
index ec029c86ae06828c1cfe886ad0f401318b114310..8ad4f49cbe0bb1a31645bbdd4735c69b9b52d8bb 100644
--- a/net/ipv6/fib6_rules.c
+++ b/net/ipv6/fib6_rules.c
@@ -486,16 +486,21 @@ static int __net_init fib6_rules_net_init(struct net *net)
 	goto out;
 }
 
-static void __net_exit fib6_rules_net_exit(struct net *net)
+static void __net_exit fib6_rules_net_exit_batch(struct list_head *net_list)
 {
+	struct net *net;
+
 	rtnl_lock();
-	fib_rules_unregister(net->ipv6.fib6_rules_ops);
+	list_for_each_entry(net, net_list, exit_list) {
+		fib_rules_unregister(net->ipv6.fib6_rules_ops);
+		cond_resched();
+	}
 	rtnl_unlock();
 }
 
 static struct pernet_operations fib6_rules_net_ops = {
 	.init = fib6_rules_net_init,
-	.exit = fib6_rules_net_exit,
+	.exit_batch = fib6_rules_net_exit_batch,
 };
 
 int __init fib6_rules_init(void)
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 07/11] ip6mr: introduce ip6mr_net_exit_batch()

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

Avoiding to acquire rtnl for each netns before calling
ip6mr_rules_exit() gives chance for cleanup_net()
to progress much faster, holding rtnl a bit longer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv6/ip6mr.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/net/ipv6/ip6mr.c b/net/ipv6/ip6mr.c
index fd660414d482a30c6d339bb7360bd91d8f3c6f05..881fe6b503072598384b7d935eab66087cd555b6 100644
--- a/net/ipv6/ip6mr.c
+++ b/net/ipv6/ip6mr.c
@@ -253,13 +253,12 @@ static void __net_exit ip6mr_rules_exit(struct net *net)
 {
 	struct mr_table *mrt, *next;
 
-	rtnl_lock();
+	ASSERT_RTNL();
 	list_for_each_entry_safe(mrt, next, &net->ipv6.mr6_tables, list) {
 		list_del(&mrt->list);
 		ip6mr_free_table(mrt);
 	}
 	fib_rules_unregister(net->ipv6.mr6_rules_ops);
-	rtnl_unlock();
 }
 
 static int ip6mr_rules_dump(struct net *net, struct notifier_block *nb,
@@ -316,10 +315,9 @@ static int __net_init ip6mr_rules_init(struct net *net)
 
 static void __net_exit ip6mr_rules_exit(struct net *net)
 {
-	rtnl_lock();
+	ASSERT_RTNL();
 	ip6mr_free_table(net->ipv6.mrt6);
 	net->ipv6.mrt6 = NULL;
-	rtnl_unlock();
 }
 
 static int ip6mr_rules_dump(struct net *net, struct notifier_block *nb,
@@ -1323,7 +1321,9 @@ static int __net_init ip6mr_net_init(struct net *net)
 proc_cache_fail:
 	remove_proc_entry("ip6_mr_vif", net->proc_net);
 proc_vif_fail:
+	rtnl_lock();
 	ip6mr_rules_exit(net);
+	rtnl_unlock();
 #endif
 ip6mr_rules_fail:
 	ip6mr_notifier_exit(net);
@@ -1336,13 +1336,23 @@ static void __net_exit ip6mr_net_exit(struct net *net)
 	remove_proc_entry("ip6_mr_cache", net->proc_net);
 	remove_proc_entry("ip6_mr_vif", net->proc_net);
 #endif
-	ip6mr_rules_exit(net);
 	ip6mr_notifier_exit(net);
 }
 
+static void __net_exit ip6mr_net_exit_batch(struct list_head *net_list)
+{
+	struct net *net;
+
+	rtnl_lock();
+	list_for_each_entry(net, net_list, exit_list)
+		ip6mr_rules_exit(net);
+	rtnl_unlock();
+}
+
 static struct pernet_operations ip6mr_net_ops = {
 	.init = ip6mr_net_init,
 	.exit = ip6mr_net_exit,
+	.exit_batch = ip6mr_net_exit_batch,
 };
 
 int __init ip6_mr_init(void)
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 08/11] ipmr: introduce ipmr_net_exit_batch()

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

Avoiding to acquire rtnl for each netns before calling
ipmr_rules_exit() gives chance for cleanup_net()
to progress much faster, holding rtnl a bit longer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Ahern <dsahern@kernel.org>
---
 net/ipv4/ipmr.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/net/ipv4/ipmr.c b/net/ipv4/ipmr.c
index 07274619b9ea11837501f8fe812d616d20573ee0..37bc9ad0df9933290ce91ae02321dee339294163 100644
--- a/net/ipv4/ipmr.c
+++ b/net/ipv4/ipmr.c
@@ -266,13 +266,12 @@ static void __net_exit ipmr_rules_exit(struct net *net)
 {
 	struct mr_table *mrt, *next;
 
-	rtnl_lock();
+	ASSERT_RNTL();
 	list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 		list_del(&mrt->list);
 		ipmr_free_table(mrt);
 	}
 	fib_rules_unregister(net->ipv4.mr_rules_ops);
-	rtnl_unlock();
 }
 
 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
@@ -328,10 +327,9 @@ static int __net_init ipmr_rules_init(struct net *net)
 
 static void __net_exit ipmr_rules_exit(struct net *net)
 {
-	rtnl_lock();
+	ASSERT_RTNL();
 	ipmr_free_table(net->ipv4.mrt);
 	net->ipv4.mrt = NULL;
-	rtnl_unlock();
 }
 
 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
@@ -3075,7 +3073,9 @@ static int __net_init ipmr_net_init(struct net *net)
 proc_cache_fail:
 	remove_proc_entry("ip_mr_vif", net->proc_net);
 proc_vif_fail:
+	rtnl_lock();
 	ipmr_rules_exit(net);
+	rtnl_unlock();
 #endif
 ipmr_rules_fail:
 	ipmr_notifier_exit(net);
@@ -3090,12 +3090,22 @@ static void __net_exit ipmr_net_exit(struct net *net)
 	remove_proc_entry("ip_mr_vif", net->proc_net);
 #endif
 	ipmr_notifier_exit(net);
-	ipmr_rules_exit(net);
+}
+
+static void __net_exit ipmr_net_exit_batch(struct list_head *net_list)
+{
+	struct net *net;
+
+	rtnl_lock();
+	list_for_each_entry(net, net_list, exit_list)
+		ipmr_rules_exit(net);
+	rtnl_unlock();
 }
 
 static struct pernet_operations ipmr_net_ops = {
 	.init = ipmr_net_init,
 	.exit = ipmr_net_exit,
+	.exit_batch = ipmr_net_exit_batch,
 };
 
 int __init ip_mr_init(void)
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 09/11] can: gw: switch cangw_pernet_exit() to batch mode

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

Avoiding to acquire rtnl for each netns before calling
cgw_remove_all_jobs() gives chance for cleanup_net()
to progress much faster, holding rtnl a bit longer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Oliver Hartkopp <socketcan@hartkopp.net>
Cc: Marc Kleine-Budde <mkl@pengutronix.de>
---
 net/can/gw.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/net/can/gw.c b/net/can/gw.c
index d8861e862f157aec36c417b71eb7e8f59bd064b9..24221352e059be9fb9aca3819be6a7ac4cdef144 100644
--- a/net/can/gw.c
+++ b/net/can/gw.c
@@ -1239,16 +1239,19 @@ static int __net_init cangw_pernet_init(struct net *net)
 	return 0;
 }
 
-static void __net_exit cangw_pernet_exit(struct net *net)
+static void __net_exit cangw_pernet_exit_batch(struct list_head *net_list)
 {
+	struct net *net;
+
 	rtnl_lock();
-	cgw_remove_all_jobs(net);
+	list_for_each_entry(net, net_list, exit_list)
+		cgw_remove_all_jobs(net);
 	rtnl_unlock();
 }
 
 static struct pernet_operations cangw_pernet_ops = {
 	.init = cangw_pernet_init,
-	.exit = cangw_pernet_exit,
+	.exit_batch = cangw_pernet_exit_batch,
 };
 
 static __init int cgw_module_init(void)
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 10/11] bonding: switch bond_net_exit() to batch mode

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

cleanup_net() is competing with other rtnl users.

Batching bond_net_exit() factorizes all rtnl acquistions
to a single one, giving chance for cleanup_net()
to progress much faster, holding rtnl a bit longer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jay Vosburgh <j.vosburgh@gmail.com>
Cc: Veaceslav Falico <vfalico@gmail.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
---
 drivers/net/bonding/bond_main.c   | 27 +++++++++++++++++++--------
 drivers/net/bonding/bond_procfs.c |  1 -
 2 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 238b56d77c369d9595d55bc681c2191c49dd2905..617c2bf8c5a7f71ece82a20dbd3a9740b928ef6a 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -6048,27 +6048,38 @@ static int __net_init bond_net_init(struct net *net)
 	return 0;
 }
 
-static void __net_exit bond_net_exit(struct net *net)
+static void __net_exit bond_net_exit_batch(struct list_head *net_list)
 {
-	struct bond_net *bn = net_generic(net, bond_net_id);
-	struct bonding *bond, *tmp_bond;
+	struct bond_net *bn;
+	struct net *net;
 	LIST_HEAD(list);
 
-	bond_destroy_sysfs(bn);
+	list_for_each_entry(net, net_list, exit_list) {
+		bn = net_generic(net, bond_net_id);
+		bond_destroy_sysfs(bn);
+	}
 
 	/* Kill off any bonds created after unregistering bond rtnl ops */
 	rtnl_lock();
-	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
-		unregister_netdevice_queue(bond->dev, &list);
+	list_for_each_entry(net, net_list, exit_list) {
+		struct bonding *bond, *tmp_bond;
+
+		bn = net_generic(net, bond_net_id);
+		list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
+			unregister_netdevice_queue(bond->dev, &list);
+	}
 	unregister_netdevice_many(&list);
 	rtnl_unlock();
 
-	bond_destroy_proc_dir(bn);
+	list_for_each_entry(net, net_list, exit_list) {
+		bn = net_generic(net, bond_net_id);
+		bond_destroy_proc_dir(bn);
+	}
 }
 
 static struct pernet_operations bond_net_ops = {
 	.init = bond_net_init,
-	.exit = bond_net_exit,
+	.exit_batch = bond_net_exit_batch,
 	.id   = &bond_net_id,
 	.size = sizeof(struct bond_net),
 };
diff --git a/drivers/net/bonding/bond_procfs.c b/drivers/net/bonding/bond_procfs.c
index 46b150e6289ef4607c8ddbcd2b833ff4dd64cc9b..cfe37be42be4e0edb218c45127a378b53e487df8 100644
--- a/drivers/net/bonding/bond_procfs.c
+++ b/drivers/net/bonding/bond_procfs.c
@@ -307,7 +307,6 @@ void __net_init bond_create_proc_dir(struct bond_net *bn)
 }
 
 /* Destroy the bonding directory under /proc/net, if empty.
- * Caller must hold rtnl_lock.
  */
 void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
 {
-- 
2.35.0.263.gb82422642f-goog

[PATCH net-next 11/11] net: remove default_device_exit()

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

For some reason default_device_ops kept two exit method:

1) default_device_exit() is called for each netns being dismantled in
a cleanup_net() round. This acquires rtnl for each invocation.

2) default_device_exit_batch() is called once with the list of all netns
int the batch, allowing for a single rtnl invocation.

Get rid of the .exit() method to handle the logic from
default_device_exit_batch(), to decrease the number of rtnl acquisition
to one.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/core/dev.c | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index f662c6a7d7b49b836a05efc74aeffc7fc9e4e147..e39c2897f6475dfa77c478603cfced76ba0b9078 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -10848,14 +10848,14 @@ static struct pernet_operations __net_initdata netdev_net_ops = {
 	.exit = netdev_exit,
 };
 
-static void __net_exit default_device_exit(struct net *net)
+static void __net_exit default_device_exit_net(struct net *net)
 {
 	struct net_device *dev, *aux;
 	/*
 	 * Push all migratable network devices back to the
 	 * initial network namespace
 	 */
-	rtnl_lock();
+	ASSERT_RTNL();
 	for_each_netdev_safe(net, dev, aux) {
 		int err;
 		char fb_name[IFNAMSIZ];
@@ -10879,22 +10879,22 @@ static void __net_exit default_device_exit(struct net *net)
 			BUG();
 		}
 	}
-	rtnl_unlock();
 }
 
 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
 {
-	/* Return with the rtnl_lock held when there are no network
+	/* Return (with the rtnl_lock held) when there are no network
 	 * devices unregistering in any network namespace in net_list.
 	 */
-	struct net *net;
-	bool unregistering;
 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
+	bool unregistering;
+	struct net *net;
 
+	ASSERT_RTNL();
 	add_wait_queue(&netdev_unregistering_wq, &wait);
 	for (;;) {
 		unregistering = false;
-		rtnl_lock();
+
 		list_for_each_entry(net, net_list, exit_list) {
 			if (net->dev_unreg_count > 0) {
 				unregistering = true;
@@ -10906,6 +10906,7 @@ static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
 		__rtnl_unlock();
 
 		wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
+		rtnl_lock();
 	}
 	remove_wait_queue(&netdev_unregistering_wq, &wait);
 }
@@ -10921,6 +10922,11 @@ static void __net_exit default_device_exit_batch(struct list_head *net_list)
 	struct net *net;
 	LIST_HEAD(dev_kill_list);
 
+	rtnl_lock();
+	list_for_each_entry(net, net_list, exit_list) {
+		default_device_exit_net(net);
+		cond_resched();
+	}
 	/* To prevent network device cleanup code from dereferencing
 	 * loopback devices or network devices that have been freed
 	 * wait here for all pending unregistrations to complete,
@@ -10933,6 +10939,7 @@ static void __net_exit default_device_exit_batch(struct list_head *net_list)
 	 * default_device_exit_batch.
 	 */
 	rtnl_lock_unregistering(net_list);
+
 	list_for_each_entry(net, net_list, exit_list) {
 		for_each_netdev_reverse(net, dev) {
 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
@@ -10946,7 +10953,6 @@ static void __net_exit default_device_exit_batch(struct list_head *net_list)
 }
 
 static struct pernet_operations __net_initdata default_device_ops = {
-	.exit = default_device_exit,
 	.exit_batch = default_device_exit_batch,
 };
 
-- 
2.35.0.263.gb82422642f-goog

Re: [PATCH net-next 09/11] can: gw: switch cangw_pernet_exit() to batch mode

[Oliver Hartkopp]

On 07.02.22 18:17, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> cleanup_net() is competing with other rtnl users.
> 
> Avoiding to acquire rtnl for each netns before calling
> cgw_remove_all_jobs() gives chance for cleanup_net()
> to progress much faster, holding rtnl a bit longer.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: Oliver Hartkopp <socketcan@hartkopp.net>
> Cc: Marc Kleine-Budde <mkl@pengutronix.de>
> ---
>   net/can/gw.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/net/can/gw.c b/net/can/gw.c
> index d8861e862f157aec36c417b71eb7e8f59bd064b9..24221352e059be9fb9aca3819be6a7ac4cdef144 100644
> --- a/net/can/gw.c
> +++ b/net/can/gw.c
> @@ -1239,16 +1239,19 @@ static int __net_init cangw_pernet_init(struct net *net)
>   	return 0;
>   }
>   
> -static void __net_exit cangw_pernet_exit(struct net *net)
> +static void __net_exit cangw_pernet_exit_batch(struct list_head *net_list)
>   {
> +	struct net *net;
> +
>   	rtnl_lock();
> -	cgw_remove_all_jobs(net);
> +	list_for_each_entry(net, net_list, exit_list)
> +		cgw_remove_all_jobs(net);

Instead of removing the jobs for ONE net namespace it seems you are 
remove removing the jobs for ALL net namespaces?

Looks wrong to me.

Best regards,
Oliver


>   	rtnl_unlock();
>   }
>   
>   static struct pernet_operations cangw_pernet_ops = {
>   	.init = cangw_pernet_init,
> -	.exit = cangw_pernet_exit,
> +	.exit_batch = cangw_pernet_exit_batch,
>   };
>   
>   static __init int cgw_module_init(void)

Re: [PATCH net-next 09/11] can: gw: switch cangw_pernet_exit() to batch mode

[Eric Dumazet]

On Mon, Feb 7, 2022 at 9:41 AM Oliver Hartkopp <socketcan@hartkopp.net> wrote:
>
>
>
> On 07.02.22 18:17, Eric Dumazet wrote:
> > From: Eric Dumazet <edumazet@google.com>
> >
> > cleanup_net() is competing with other rtnl users.
> >
> > Avoiding to acquire rtnl for each netns before calling
> > cgw_remove_all_jobs() gives chance for cleanup_net()
> > to progress much faster, holding rtnl a bit longer.
> >
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
> > Cc: Oliver Hartkopp <socketcan@hartkopp.net>
> > Cc: Marc Kleine-Budde <mkl@pengutronix.de>
> > ---
> >   net/can/gw.c | 9 ++++++---
> >   1 file changed, 6 insertions(+), 3 deletions(-)
> >
> > diff --git a/net/can/gw.c b/net/can/gw.c
> > index d8861e862f157aec36c417b71eb7e8f59bd064b9..24221352e059be9fb9aca3819be6a7ac4cdef144 100644
> > --- a/net/can/gw.c
> > +++ b/net/can/gw.c
> > @@ -1239,16 +1239,19 @@ static int __net_init cangw_pernet_init(struct net *net)
> >       return 0;
> >   }
> >
> > -static void __net_exit cangw_pernet_exit(struct net *net)
> > +static void __net_exit cangw_pernet_exit_batch(struct list_head *net_list)
> >   {
> > +     struct net *net;
> > +
> >       rtnl_lock();
> > -     cgw_remove_all_jobs(net);
> > +     list_for_each_entry(net, net_list, exit_list)
> > +             cgw_remove_all_jobs(net);
>
> Instead of removing the jobs for ONE net namespace it seems you are
> remove removing the jobs for ALL net namespaces?
>
> Looks wrong to me.

I see nothing wrong in my patch.

I think you have to look more closely at ops_exit_list() in
net/core/net_namespace.c


BTW, the sychronize_rcu() call in cgw_remove_all_jobs is definitely
bad, you should absolutely replace it by call_rcu() or kfree_rcu()

>
> Best regards,
> Oliver
>
>
> >       rtnl_unlock();
> >   }
> >
> >   static struct pernet_operations cangw_pernet_ops = {
> >       .init = cangw_pernet_init,
> > -     .exit = cangw_pernet_exit,
> > +     .exit_batch = cangw_pernet_exit_batch,
> >   };
> >
> >   static __init int cgw_module_init(void)

Re: [PATCH net-next 09/11] can: gw: switch cangw_pernet_exit() to batch mode

[Oliver Hartkopp]

On 07.02.22 18:54, Eric Dumazet wrote:
> On Mon, Feb 7, 2022 at 9:41 AM Oliver Hartkopp <socketcan@hartkopp.net> wrote:
(..)
>>> -static void __net_exit cangw_pernet_exit(struct net *net)
>>> +static void __net_exit cangw_pernet_exit_batch(struct list_head *net_list)
>>>    {
>>> +     struct net *net;
>>> +
>>>        rtnl_lock();
>>> -     cgw_remove_all_jobs(net);
>>> +     list_for_each_entry(net, net_list, exit_list)
>>> +             cgw_remove_all_jobs(net);
>>
>> Instead of removing the jobs for ONE net namespace it seems you are
>> remove removing the jobs for ALL net namespaces?
>>
>> Looks wrong to me.
> 
> I see nothing wrong in my patch.
> 
> I think you have to look more closely at ops_exit_list() in
> net/core/net_namespace.c

Ok, thanks. Your patch just moved the list_for_each_entry() to gw.c.
So there is no functional difference.

Acked-by: Oliver Hartkopp <socketcan@hartkopp.net>

> BTW, the sychronize_rcu() call in cgw_remove_all_jobs is definitely
> bad, you should absolutely replace it by call_rcu() or kfree_rcu()

Advise is welcome!

The synchronize_rcu() has been introduced in fb8696ab14ad ("can: gw: 
synchronize rcu operations before removing gw job entry") as 
can_can_gw_rcv() is called under RCU protection (NET_RX softirq).

That patch was a follow-up to d5f9023fa61e ("can: bcm: delay release of 
struct bcm_op after synchronize_rcu()") where Thadeu Lima de Souza 
Cascardo detected a race in the BCM code.

When call_rcu() is enough to make sure we do not get a race in 
can_can_gw_rcv() while receiving skbs and removing filters with 
cgw_unregister_filter() I would be happy this rcu thing being fixed up.

Best regards,
Oliver

Re: [PATCH net-next 09/11] can: gw: switch cangw_pernet_exit() to batch mode

[Eric Dumazet]

On Mon, Feb 7, 2022 at 10:40 AM Oliver Hartkopp <socketcan@hartkopp.net> wrote:
>
>
>
> On 07.02.22 18:54, Eric Dumazet wrote:
> > On Mon, Feb 7, 2022 at 9:41 AM Oliver Hartkopp <socketcan@hartkopp.net> wrote:
> (..)
> >>> -static void __net_exit cangw_pernet_exit(struct net *net)
> >>> +static void __net_exit cangw_pernet_exit_batch(struct list_head *net_list)
> >>>    {
> >>> +     struct net *net;
> >>> +
> >>>        rtnl_lock();
> >>> -     cgw_remove_all_jobs(net);
> >>> +     list_for_each_entry(net, net_list, exit_list)
> >>> +             cgw_remove_all_jobs(net);
> >>
> >> Instead of removing the jobs for ONE net namespace it seems you are
> >> remove removing the jobs for ALL net namespaces?
> >>
> >> Looks wrong to me.
> >
> > I see nothing wrong in my patch.
> >
> > I think you have to look more closely at ops_exit_list() in
> > net/core/net_namespace.c
>
> Ok, thanks. Your patch just moved the list_for_each_entry() to gw.c.
> So there is no functional difference.
>
> Acked-by: Oliver Hartkopp <socketcan@hartkopp.net>
>
> > BTW, the sychronize_rcu() call in cgw_remove_all_jobs is definitely
> > bad, you should absolutely replace it by call_rcu() or kfree_rcu()
>
> Advise is welcome!
>
> The synchronize_rcu() has been introduced in fb8696ab14ad ("can: gw:
> synchronize rcu operations before removing gw job entry") as
> can_can_gw_rcv() is called under RCU protection (NET_RX softirq).
>
> That patch was a follow-up to d5f9023fa61e ("can: bcm: delay release of
> struct bcm_op after synchronize_rcu()") where Thadeu Lima de Souza
> Cascardo detected a race in the BCM code.
>
> When call_rcu() is enough to make sure we do not get a race in
> can_can_gw_rcv() while receiving skbs and removing filters with
> cgw_unregister_filter() I would be happy this rcu thing being fixed up.
>

Can you test this straightforward patch, thanks !
diff --git a/net/can/gw.c b/net/can/gw.c
index d8861e862f157aec36c417b71eb7e8f59bd064b9..20e74fe7d0906dccc65732b8f9e7e14e2d1192c3
100644
--- a/net/can/gw.c
+++ b/net/can/gw.c
@@ -577,6 +577,13 @@ static inline void cgw_unregister_filter(struct
net *net, struct cgw_job *gwj)
                          gwj->ccgw.filter.can_mask, can_can_gw_rcv, gwj);
 }

+static void cgw_job_free_rcu(struct rcu_head *rcu_head)
+{
+       struct cgw_job *gwj = container_of(rcu_head, struct cgw_job, rcu);
+
+       kmem_cache_free(cgw_cache, gwj);
+}
+
 static int cgw_notifier(struct notifier_block *nb,
                        unsigned long msg, void *ptr)
 {
@@ -596,8 +603,7 @@ static int cgw_notifier(struct notifier_block *nb,
                        if (gwj->src.dev == dev || gwj->dst.dev == dev) {
                                hlist_del(&gwj->list);
                                cgw_unregister_filter(net, gwj);
-                               synchronize_rcu();
-                               kmem_cache_free(cgw_cache, gwj);
+                               call_rcu(&gwj->rcu, cgw_job_free_rcu);
                        }
                }
        }
@@ -1155,8 +1161,7 @@ static void cgw_remove_all_jobs(struct net *net)
        hlist_for_each_entry_safe(gwj, nx, &net->can.cgw_list, list) {
                hlist_del(&gwj->list);
                cgw_unregister_filter(net, gwj);
-               synchronize_rcu();
-               kmem_cache_free(cgw_cache, gwj);
+               call_rcu(&gwj->rcu, cgw_job_free_rcu);
        }
 }

@@ -1224,8 +1229,7 @@ static int cgw_remove_job(struct sk_buff *skb,
struct nlmsghdr *nlh,

                hlist_del(&gwj->list);
                cgw_unregister_filter(net, gwj);
-               synchronize_rcu();
-               kmem_cache_free(cgw_cache, gwj);
+               call_rcu(&gwj->rcu, cgw_job_free_rcu);
                err = 0;
                break;
        }

Re: [PATCH net-next 08/11] ipmr: introduce ipmr_net_exit_batch()

[kernel test robot]

[-- Attachment #1: Type: text/plain, Size: 2224 bytes --]

Hi Eric,

I love your patch! Yet something to improve:

[auto build test ERROR on net-next/master]

url:    https://github.com/0day-ci/linux/commits/Eric-Dumazet/net-speedup-netns-dismantles/20220208-013735
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git ff62433883b3ab753a78954ecf46e2c514f5c407
config: ia64-allmodconfig (https://download.01.org/0day-ci/archive/20220208/202202080332.UKy4LO6e-lkp(a)intel.com/config)
compiler: ia64-linux-gcc (GCC) 11.2.0
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/0day-ci/linux/commit/4a23446f3f1d4d348348210ec16e5e5372177b48
        git remote add linux-review https://github.com/0day-ci/linux
        git fetch --no-tags linux-review Eric-Dumazet/net-speedup-netns-dismantles/20220208-013735
        git checkout 4a23446f3f1d4d348348210ec16e5e5372177b48
        # save the config file to linux build tree
        mkdir build_dir
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross O=build_dir ARCH=ia64 SHELL=/bin/bash net/ipv4/

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>

All errors (new ones prefixed by >>):

   net/ipv4/ipmr.c: In function 'ipmr_rules_exit':
>> net/ipv4/ipmr.c:269:9: error: implicit declaration of function 'ASSERT_RNTL'; did you mean 'ASSERT_RTNL'? [-Werror=implicit-function-declaration]
     269 |         ASSERT_RNTL();
         |         ^~~~~~~~~~~
         |         ASSERT_RTNL
   cc1: some warnings being treated as errors


vim +269 net/ipv4/ipmr.c

   264	
   265	static void __net_exit ipmr_rules_exit(struct net *net)
   266	{
   267		struct mr_table *mrt, *next;
   268	
 > 269		ASSERT_RNTL();
   270		list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
   271			list_del(&mrt->list);
   272			ipmr_free_table(mrt);
   273		}
   274		fib_rules_unregister(net->ipv4.mr_rules_ops);
   275	}
   276	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

Re: [PATCH net-next 08/11] ipmr: introduce ipmr_net_exit_batch()

[kernel test robot]

Hi Eric,

I love your patch! Yet something to improve:

[auto build test ERROR on net-next/master]

url:    https://github.com/0day-ci/linux/commits/Eric-Dumazet/net-speedup-netns-dismantles/20220208-013735
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git ff62433883b3ab753a78954ecf46e2c514f5c407
config: x86_64-randconfig-r003-20220207 (https://download.01.org/0day-ci/archive/20220208/202202080620.vPGy8XEF-lkp@intel.com/config)
compiler: clang version 15.0.0 (https://github.com/llvm/llvm-project 0d8850ae2cae85d49bea6ae0799fa41c7202c05c)
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/0day-ci/linux/commit/4a23446f3f1d4d348348210ec16e5e5372177b48
        git remote add linux-review https://github.com/0day-ci/linux
        git fetch --no-tags linux-review Eric-Dumazet/net-speedup-netns-dismantles/20220208-013735
        git checkout 4a23446f3f1d4d348348210ec16e5e5372177b48
        # save the config file to linux build tree
        mkdir build_dir
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=x86_64 SHELL=/bin/bash net/

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>

All errors (new ones prefixed by >>):

>> net/ipv4/ipmr.c:269:2: error: implicit declaration of function 'ASSERT_RNTL' [-Werror,-Wimplicit-function-declaration]
           ASSERT_RNTL();
           ^
   1 error generated.


vim +/ASSERT_RNTL +269 net/ipv4/ipmr.c

   264	
   265	static void __net_exit ipmr_rules_exit(struct net *net)
   266	{
   267		struct mr_table *mrt, *next;
   268	
 > 269		ASSERT_RNTL();
   270		list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
   271			list_del(&mrt->list);
   272			ipmr_free_table(mrt);
   273		}
   274		fib_rules_unregister(net->ipv4.mr_rules_ops);
   275	}
   276	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org

Re: [PATCH net-next 08/11] ipmr: introduce ipmr_net_exit_batch()

[kernel test robot]

[-- Attachment #1: Type: text/plain, Size: 2160 bytes --]

Hi Eric,

I love your patch! Yet something to improve:

[auto build test ERROR on net-next/master]

url:    https://github.com/0day-ci/linux/commits/Eric-Dumazet/net-speedup-netns-dismantles/20220208-013735
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git ff62433883b3ab753a78954ecf46e2c514f5c407
config: x86_64-randconfig-r003-20220207 (https://download.01.org/0day-ci/archive/20220208/202202080620.vPGy8XEF-lkp(a)intel.com/config)
compiler: clang version 15.0.0 (https://github.com/llvm/llvm-project 0d8850ae2cae85d49bea6ae0799fa41c7202c05c)
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/0day-ci/linux/commit/4a23446f3f1d4d348348210ec16e5e5372177b48
        git remote add linux-review https://github.com/0day-ci/linux
        git fetch --no-tags linux-review Eric-Dumazet/net-speedup-netns-dismantles/20220208-013735
        git checkout 4a23446f3f1d4d348348210ec16e5e5372177b48
        # save the config file to linux build tree
        mkdir build_dir
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=x86_64 SHELL=/bin/bash net/

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>

All errors (new ones prefixed by >>):

>> net/ipv4/ipmr.c:269:2: error: implicit declaration of function 'ASSERT_RNTL' [-Werror,-Wimplicit-function-declaration]
           ASSERT_RNTL();
           ^
   1 error generated.


vim +/ASSERT_RNTL +269 net/ipv4/ipmr.c

   264	
   265	static void __net_exit ipmr_rules_exit(struct net *net)
   266	{
   267		struct mr_table *mrt, *next;
   268	
 > 269		ASSERT_RNTL();
   270		list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
   271			list_del(&mrt->list);
   272			ipmr_free_table(mrt);
   273		}
   274		fib_rules_unregister(net->ipv4.mr_rules_ops);
   275	}
   276	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

Re: [PATCH net-next 01/11] ipv6/addrconf: allocate a per netns hash table

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> Add a per netns hash table and a dedicated spinlock,
> first step to get rid of the global inet6_addr_lst[] one.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  include/net/netns/ipv6.h |  4 ++++
>  net/ipv6/addrconf.c      | 20 ++++++++++++++++++++
>  2 files changed, 24 insertions(+)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 02/11] ipv6/addrconf: use one delayed work per netns

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> Next step for using per netns inet6_addr_lst
> is to have per netns work item to ultimately
> call addrconf_verify_rtnl() and addrconf_verify()
> with a new 'struct net*' argument.
> 
> Everything is still using the global inet6_addr_lst[] table.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  include/net/netns/ipv6.h |  1 +
>  net/ipv6/addrconf.c      | 44 ++++++++++++++++++++++------------------
>  2 files changed, 25 insertions(+), 20 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 03/11] ipv6/addrconf: switch to per netns inet6_addr_lst hash table

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> IPv6 does not scale very well with the number of IPv6 addresses.
> It uses a global (shared by all netns) hash table with 256 buckets.
> 
> Some functions like addrconf_verify_rtnl() and addrconf_ifdown()
> have to iterate all addresses in the hash table.
> 
> I have seen addrconf_verify_rtnl() holding the cpu for 10ms or more.
> 
> Switch to the per netns hashtable (and spinlock) added
> in prior patches.
> 
> This considerably speeds up netns dismantle times on hosts
> with thousands of netns.

this has to speed up many aspects of ipv6 addresses, not just the teardown.

> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv6/addrconf.c | 77 ++++++++++++++-------------------------------
>  1 file changed, 23 insertions(+), 54 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 04/11] nexthop: change nexthop_net_exit() to nexthop_net_exit_batch()

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> cleanup_net() is competing with other rtnl users.
> 
> nexthop_net_exit() seems a good candidate for exit_batch(),
> as this gives chance for cleanup_net() to progress much faster,
> holding rtnl a bit longer.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: David Ahern <dsahern@kernel.org>
> ---
>  net/ipv4/nexthop.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 05/11] ipv4: add fib_net_exit_batch()

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> cleanup_net() is competing with other rtnl users.
> 
> Instead of acquiring rtnl at each fib_net_exit() invocation,
> add fib_net_exit_batch() so that rtnl is acquired once.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: David Ahern <dsahern@kernel.org>
> ---
>  net/ipv4/fib_frontend.c | 19 ++++++++++++++++---
>  1 file changed, 16 insertions(+), 3 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 06/11] ipv6: change fib6_rules_net_exit() to batch mode

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> cleanup_net() is competing with other rtnl users.
> 
> fib6_rules_net_exit() seems a good candidate for exit_batch(),
> as this gives chance for cleanup_net() to progress much faster,
> holding rtnl a bit longer.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: David Ahern <dsahern@kernel.org>
> ---
>  net/ipv6/fib6_rules.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 07/11] ip6mr: introduce ip6mr_net_exit_batch()

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> cleanup_net() is competing with other rtnl users.
> 
> Avoiding to acquire rtnl for each netns before calling
> ip6mr_rules_exit() gives chance for cleanup_net()
> to progress much faster, holding rtnl a bit longer.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv6/ip6mr.c | 20 +++++++++++++++-----
>  1 file changed, 15 insertions(+), 5 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 08/11] ipmr: introduce ipmr_net_exit_batch()

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> cleanup_net() is competing with other rtnl users.
> 
> Avoiding to acquire rtnl for each netns before calling
> ipmr_rules_exit() gives chance for cleanup_net()
> to progress much faster, holding rtnl a bit longer.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: David Ahern <dsahern@kernel.org>
> ---
>  net/ipv4/ipmr.c | 20 +++++++++++++++-----
>  1 file changed, 15 insertions(+), 5 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 11/11] net: remove default_device_exit()

[David Ahern]

On 2/7/22 9:17 AM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> For some reason default_device_ops kept two exit method:
> 
> 1) default_device_exit() is called for each netns being dismantled in
> a cleanup_net() round. This acquires rtnl for each invocation.
> 
> 2) default_device_exit_batch() is called once with the list of all netns
> int the batch, allowing for a single rtnl invocation.
> 
> Get rid of the .exit() method to handle the logic from
> default_device_exit_batch(), to decrease the number of rtnl acquisition
> to one.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/core/dev.c | 22 ++++++++++++++--------
>  1 file changed, 14 insertions(+), 8 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 11/11] net: remove default_device_exit()

[Eric Dumazet]

On Mon, Feb 7, 2022 at 8:29 PM David Ahern <dsahern@gmail.com> wrote:
>
> On 2/7/22 9:17 AM, Eric Dumazet wrote:
> > From: Eric Dumazet <edumazet@google.com>
> >
> > For some reason default_device_ops kept two exit method:
> >
> > 1) default_device_exit() is called for each netns being dismantled in
> > a cleanup_net() round. This acquires rtnl for each invocation.
> >
> > 2) default_device_exit_batch() is called once with the list of all netns
> > int the batch, allowing for a single rtnl invocation.
> >
> > Get rid of the .exit() method to handle the logic from
> > default_device_exit_batch(), to decrease the number of rtnl acquisition
> > to one.
> >
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
> > ---
> >  net/core/dev.c | 22 ++++++++++++++--------
> >  1 file changed, 14 insertions(+), 8 deletions(-)
> >
>
> Reviewed-by: David Ahern <dsahern@kernel.org>

Thanks for the review !

I am sending a v2, with your tags, plus the RTNL_ASSERT() typo fixed.