Skip to contentSkip to conversation

netfilter: conntrack: remove percpu lists

RFC Superseded Florian Westphal 1 participant 16 messages Mar 3, 2022

Subsystems (2)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

v3 cover netfilter: conntrack: remove percpu lists

Do not apply, I will re-target this to next cycle.

This is a resend of the 'remove the percpu dying list' series (patches 1-7), with the reported build failures resolved.

The rest of the patches are the second part: remove the unconfirmed list. After this series, the percpu lists are gone.

Florian Westphal (15):
nfnetlink: handle already-released nl socket
netfilter: ctnetlink: make ecache event cb global again
netfilter: ecache: move to separate structure
netfilter: ecache: use dedicated list for event redelivery
netfilter: conntrack: split inner loop of list dumping to own function
netfilter: conntrack: include ecache dying list in dumps
netfilter: conntrack: remove the percpu dying list
netfilter: cttimeout: inc/dec module refcount per object, not per use
refcount
netfilter: nfnetlink_cttimeout: use rcu protection in
cttimeout_get_timeout
netfilter: cttimeout: decouple unlink and free on netns destruction
netfilter: remove nf_ct_unconfirmed_destroy helper
netfilter: extensions: introduce extension genid count
netfilter: cttimeout: decouple unlink and free on netns destruction
netfilter: conntrack: remove __nf_ct_unconfirmed_destroy
netfilter: conntrack: remove unconfirmed list

 include/net/netfilter/nf_conntrack.h         |  13 +-
 include/net/netfilter/nf_conntrack_ecache.h  |  21 +-
 include/net/netfilter/nf_conntrack_extend.h  |  31 +--
 include/net/netfilter/nf_conntrack_timeout.h |   8 -
 include/net/netns/conntrack.h                |   8 -
 net/netfilter/nf_conntrack_core.c            | 223 ++++++++-----------
 net/netfilter/nf_conntrack_ecache.c          | 172 +++++++-------
 net/netfilter/nf_conntrack_extend.c          |  32 ++-
 net/netfilter/nf_conntrack_helper.c          |   5 -
 net/netfilter/nf_conntrack_netlink.c         | 177 +++++++--------
 net/netfilter/nfnetlink.c                    |  62 ++++--
 net/netfilter/nfnetlink_cttimeout.c          |  88 +++++---
 12 files changed, 431 insertions(+), 409 deletions(-)
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

RFC v3 1/15 nfnetlink: handle already-released nl socket

At this time upper layer, e.g. nf_conntrack_event, has to make sure that its pernet exit handler runs before the nfnetlink one, otherwise we get a crash if kernel tries to send a conntrack event after the nfnetlink netns exit handler did close the socket already.

In order to move nf_conntrack_ecache to global (not pernet) netns event pointer again the nfnetlink apis need to survive attempts to send a netlink message after the socket has been destroyed in nfnetlink netns exit function.

Set the pernet socket to null in the pre_exit handler and close it in the exit_batch handler via a 'stash' pointer.

All functions now check nlsk for NULL before using it.

---
 net/netfilter/nfnetlink.c | 62 +++++++++++++++++++++++++++++----------
 1 file changed, 47 insertions(+), 15 deletions(-)
net/netfilter/nfnetlink.c +47 -15
index 7e2c8dd01408..6105dc9b8f8e 100644
@@ -46,6 +46,7 @@ static unsigned int nfnetlink_pernet_id __read_mostly;
 
 struct nfnl_net {
 	struct sock *nfnl;
+	struct sock *nfnl_stash;
 };
 
 static struct {
@@ -160,37 +161,56 @@ nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
 	return &ss->cb[cb_id];
 }
 
-int nfnetlink_has_listeners(struct net *net, unsigned int group)
+static struct sock *nfnl_pernet_sk(struct net *net)
 {
 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
 
-	return netlink_has_listeners(nfnlnet->nfnl, group);
+	return READ_ONCE(nfnlnet->nfnl);
+}
+
+int nfnetlink_has_listeners(struct net *net, unsigned int group)
+{
+	struct sock *nlsk = nfnl_pernet_sk(net);
+
+	return nlsk ? netlink_has_listeners(nlsk, group) : 0;
 }
 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
 
 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
 		   unsigned int group, int echo, gfp_t flags)
 {
-	struct nfnl_net *nfnlnet = nfnl_pernet(net);
+	struct sock *nlsk = nfnl_pernet_sk(net);
+
+	if (nlsk)
+		return nlmsg_notify(nlsk, skb, portid, group, echo, flags);
 
-	return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
+	/* nlsk already gone? This happens when .pre_exit was already called,
+	 * return 0, we can't retry.
+	 */
+	kfree_skb(skb);
+	return 0;
 }
 EXPORT_SYMBOL_GPL(nfnetlink_send);
 
 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
 {
-	struct nfnl_net *nfnlnet = nfnl_pernet(net);
+	struct sock *nlsk = nfnl_pernet_sk(net);
 
-	return netlink_set_err(nfnlnet->nfnl, portid, group, error);
+	return nlsk ? netlink_set_err(nlsk, portid, group, error) : 0;
 }
 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
 
 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
 {
-	struct nfnl_net *nfnlnet = nfnl_pernet(net);
+	struct sock *nlsk = nfnl_pernet_sk(net);
 	int err;
 
-	err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
+	if (!nlsk) {
+		kfree_skb(skb);
+		return 0;
+	}
+
+	err = nlmsg_unicast(nlsk, skb, portid);
 	if (err == -EAGAIN)
 		err = -ENOBUFS;
 
@@ -201,9 +221,12 @@ EXPORT_SYMBOL_GPL(nfnetlink_unicast);
 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
 			 __u32 group, gfp_t allocation)
 {
-	struct nfnl_net *nfnlnet = nfnl_pernet(net);
+	struct sock *nlsk = nfnl_pernet_sk(net);
 
-	netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
+	if (nlsk)
+		netlink_broadcast(nlsk, skb, portid, group, allocation);
+	else
+		kfree_skb(skb);
 }
 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
 
@@ -247,7 +270,7 @@ static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
 
 	{
 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
-		struct nfnl_net *nfnlnet = nfnl_pernet(net);
+		struct sock *nlsk = nfnl_pernet_sk(net);
 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 		struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
 		struct nlattr *attr = (void *)nlh + min_len;
@@ -255,7 +278,7 @@ static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
 		struct nfnl_info info = {
 			.net	= net,
-			.sk	= nfnlnet->nfnl,
+			.sk	= nlsk,
 			.nlh	= nlh,
 			.nfmsg	= nlmsg_data(nlh),
 			.extack	= extack,
@@ -484,14 +507,14 @@ static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
 
 		{
 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
-			struct nfnl_net *nfnlnet = nfnl_pernet(net);
+			struct sock *nlsk = nfnl_pernet_sk(net);
 			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
 			struct nlattr *attr = (void *)nlh + min_len;
 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 			int attrlen = nlh->nlmsg_len - min_len;
 			struct nfnl_info info = {
 				.net	= net,
-				.sk	= nfnlnet->nfnl,
+				.sk	= nlsk,
 				.nlh	= nlh,
 				.nfmsg	= nlmsg_data(nlh),
 				.extack	= &extack,
@@ -699,12 +722,21 @@ static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
 	list_for_each_entry(net, net_exit_list, exit_list) {
 		nfnlnet = nfnl_pernet(net);
 
-		netlink_kernel_release(nfnlnet->nfnl);
+		netlink_kernel_release(nfnlnet->nfnl_stash);
 	}
 }
 
+static void __net_exit nfnetlink_net_pre_exit(struct net *net)
+{
+	struct nfnl_net *nfnlnet = nfnl_pernet(net);
+
+	nfnlnet->nfnl_stash = nfnlnet->nfnl;
+	WRITE_ONCE(nfnlnet->nfnl, NULL);
+}
+
 static struct pernet_operations nfnetlink_net_ops = {
 	.init		= nfnetlink_net_init,
+	.pre_exit	= nfnetlink_net_pre_exit,
 	.exit_batch	= nfnetlink_net_exit_batch,
 	.id		= &nfnetlink_pernet_id,
 	.size		= sizeof(struct nfnl_net),
Signature · 1 line
2.34.1
author maintainer
#2 lore raw

RFC v3 2/15 netfilter: ctnetlink: make ecache event cb global again

In reply to Florian Westphal

This was pernet to make sure we do not trip over already closed nfnl sk.

After moving nfnl validity checks to nfnetlink core this can be global again, it only needs to be set to NULL when the module is removed.

This also avoids the need for pernet ops in ctnetlink and register mutex.

---
 include/net/netfilter/nf_conntrack_ecache.h | 18 ++++------
 include/net/netns/conntrack.h               |  1 -
 net/netfilter/nf_conntrack_ecache.c         | 32 ++++++------------
 net/netfilter/nf_conntrack_netlink.c        | 37 +++++----------------
 4 files changed, 25 insertions(+), 63 deletions(-)
include/net/netfilter/nf_conntrack_ecache.h +7 -11
index 6c4c490a3e34..ebd816af9c40 100644
@@ -83,9 +83,8 @@ struct nf_ct_event_notifier {
 	int (*exp_event)(unsigned int events, const struct nf_exp_event *item);
 };
 
-void nf_conntrack_register_notifier(struct net *net,
-				   const struct nf_ct_event_notifier *nb);
-void nf_conntrack_unregister_notifier(struct net *net);
+void nf_conntrack_register_notifier(const struct nf_ct_event_notifier *nb);
+void nf_conntrack_unregister_notifier(void);
 
 void nf_ct_deliver_cached_events(struct nf_conn *ct);
 int nf_conntrack_eventmask_report(unsigned int eventmask, struct nf_conn *ct,
@@ -107,14 +106,15 @@ static inline int nf_conntrack_eventmask_report(unsigned int eventmask,
 
 #endif
 
+extern const struct nf_ct_event_notifier __rcu *nf_conntrack_event_cb __read_mostly;
+
 static inline void
 nf_conntrack_event_cache(enum ip_conntrack_events event, struct nf_conn *ct)
 {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-	struct net *net = nf_ct_net(ct);
 	struct nf_conntrack_ecache *e;
 
-	if (!rcu_access_pointer(net->ct.nf_conntrack_event_cb))
+	if (!rcu_access_pointer(nf_conntrack_event_cb))
 		return;
 
 	e = nf_ct_ecache_find(ct);
@@ -130,9 +130,7 @@ nf_conntrack_event_report(enum ip_conntrack_events event, struct nf_conn *ct,
 			  u32 portid, int report)
 {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-	const struct net *net = nf_ct_net(ct);
-
-	if (!rcu_access_pointer(net->ct.nf_conntrack_event_cb))
+	if (!rcu_access_pointer(nf_conntrack_event_cb))
 		return 0;
 
 	return nf_conntrack_eventmask_report(1 << event, ct, portid, report);
@@ -145,9 +143,7 @@ static inline int
 nf_conntrack_event(enum ip_conntrack_events event, struct nf_conn *ct)
 {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-	const struct net *net = nf_ct_net(ct);
-
-	if (!rcu_access_pointer(net->ct.nf_conntrack_event_cb))
+	if (!rcu_access_pointer(nf_conntrack_event_cb))
 		return 0;
 
 	return nf_conntrack_eventmask_report(1 << event, ct, 0, 0);
include/net/netns/conntrack.h +0 -1
index 0294f3d473af..3bb62e938fa9 100644
@@ -112,7 +112,6 @@ struct netns_ct {
 
 	struct ct_pcpu __percpu *pcpu_lists;
 	struct ip_conntrack_stat __percpu *stat;
-	struct nf_ct_event_notifier __rcu *nf_conntrack_event_cb;
 	struct nf_ip_net	nf_ct_proto;
 #if defined(CONFIG_NF_CONNTRACK_LABELS)
 	unsigned int		labels_used;
net/netfilter/nf_conntrack_ecache.c +10 -22
index 07e65b4e92f8..13ba6aa82ec1 100644
@@ -27,7 +27,7 @@
 #include <net/netfilter/nf_conntrack_ecache.h>
 #include <net/netfilter/nf_conntrack_extend.h>
 
-static DEFINE_MUTEX(nf_ct_ecache_mutex);
+const struct nf_ct_event_notifier __rcu *nf_conntrack_event_cb __read_mostly;
 
 #define ECACHE_RETRY_WAIT (HZ/10)
 #define ECACHE_STACK_ALLOC (256 / sizeof(void *))
@@ -135,8 +135,7 @@ static int __nf_conntrack_eventmask_report(struct nf_conntrack_ecache *e,
 					   const u32 missed,
 					   const struct nf_ct_event *item)
 {
-	struct net *net = nf_ct_net(item->ct);
-	struct nf_ct_event_notifier *notify;
+	const struct nf_ct_event_notifier *notify;
 	u32 old, want;
 	int ret;
 
@@ -145,7 +144,7 @@ static int __nf_conntrack_eventmask_report(struct nf_conntrack_ecache *e,
 
 	rcu_read_lock();
 
-	notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
+	notify = rcu_dereference(nf_conntrack_event_cb);
 	if (!notify) {
 		rcu_read_unlock();
 		return 0;
@@ -240,12 +239,11 @@ void nf_ct_expect_event_report(enum ip_conntrack_expect_events event,
 			       u32 portid, int report)
 
 {
-	struct net *net = nf_ct_exp_net(exp);
-	struct nf_ct_event_notifier *notify;
+	const struct nf_ct_event_notifier *notify;
 	struct nf_conntrack_ecache *e;
 
 	rcu_read_lock();
-	notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
+	notify = rcu_dereference(nf_conntrack_event_cb);
 	if (!notify)
 		goto out_unlock;
 
@@ -265,26 +263,16 @@ void nf_ct_expect_event_report(enum ip_conntrack_expect_events event,
 	rcu_read_unlock();
 }
 
-void nf_conntrack_register_notifier(struct net *net,
-				    const struct nf_ct_event_notifier *new)
+void nf_conntrack_register_notifier(const struct nf_ct_event_notifier *new)
 {
-	struct nf_ct_event_notifier *notify;
-
-	mutex_lock(&nf_ct_ecache_mutex);
-	notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
-					   lockdep_is_held(&nf_ct_ecache_mutex));
-	WARN_ON_ONCE(notify);
-	rcu_assign_pointer(net->ct.nf_conntrack_event_cb, new);
-	mutex_unlock(&nf_ct_ecache_mutex);
+	WARN_ON_ONCE(rcu_access_pointer(nf_conntrack_event_cb));
+	rcu_assign_pointer(nf_conntrack_event_cb, new);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_register_notifier);
 
-void nf_conntrack_unregister_notifier(struct net *net)
+void nf_conntrack_unregister_notifier(void)
 {
-	mutex_lock(&nf_ct_ecache_mutex);
-	RCU_INIT_POINTER(net->ct.nf_conntrack_event_cb, NULL);
-	mutex_unlock(&nf_ct_ecache_mutex);
-	/* synchronize_rcu() is called after netns pre_exit */
+	RCU_INIT_POINTER(nf_conntrack_event_cb, NULL);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_notifier);
 
net/netfilter/nf_conntrack_netlink.c +8 -29
index 1ea2ad732d57..4a460565f275 100644
@@ -3772,7 +3772,7 @@ static int ctnetlink_stat_exp_cpu(struct sk_buff *skb,
 }
 
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-static struct nf_ct_event_notifier ctnl_notifier = {
+static const struct nf_ct_event_notifier ctnl_notifier = {
 	.ct_event = ctnetlink_conntrack_event,
 	.exp_event = ctnetlink_expect_event,
 };
@@ -3864,26 +3864,6 @@ MODULE_ALIAS("ip_conntrack_netlink");
 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK);
 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK_EXP);
 
-static int __net_init ctnetlink_net_init(struct net *net)
-{
-#ifdef CONFIG_NF_CONNTRACK_EVENTS
-	nf_conntrack_register_notifier(net, &ctnl_notifier);
-#endif
-	return 0;
-}
-
-static void ctnetlink_net_pre_exit(struct net *net)
-{
-#ifdef CONFIG_NF_CONNTRACK_EVENTS
-	nf_conntrack_unregister_notifier(net);
-#endif
-}
-
-static struct pernet_operations ctnetlink_net_ops = {
-	.init		= ctnetlink_net_init,
-	.pre_exit	= ctnetlink_net_pre_exit,
-};
-
 static int __init ctnetlink_init(void)
 {
 	int ret;
@@ -3902,19 +3882,16 @@ static int __init ctnetlink_init(void)
 		goto err_unreg_subsys;
 	}
 
-	ret = register_pernet_subsys(&ctnetlink_net_ops);
-	if (ret < 0) {
-		pr_err("ctnetlink_init: cannot register pernet operations\n");
-		goto err_unreg_exp_subsys;
-	}
+#ifdef CONFIG_NF_CONNTRACK_EVENTS
+	nf_conntrack_register_notifier(&ctnl_notifier);
+#endif
+
 #ifdef CONFIG_NETFILTER_NETLINK_GLUE_CT
 	/* setup interaction between nf_queue and nf_conntrack_netlink. */
 	RCU_INIT_POINTER(nfnl_ct_hook, &ctnetlink_glue_hook);
 #endif
 	return 0;
 
-err_unreg_exp_subsys:
-	nfnetlink_subsys_unregister(&ctnl_exp_subsys);
 err_unreg_subsys:
 	nfnetlink_subsys_unregister(&ctnl_subsys);
 err_out:
@@ -3923,7 +3900,9 @@ static int __init ctnetlink_init(void)
 
 static void __exit ctnetlink_exit(void)
 {
-	unregister_pernet_subsys(&ctnetlink_net_ops);
+#ifdef CONFIG_NF_CONNTRACK_EVENTS
+	nf_conntrack_unregister_notifier();
+#endif
 	nfnetlink_subsys_unregister(&ctnl_exp_subsys);
 	nfnetlink_subsys_unregister(&ctnl_subsys);
 #ifdef CONFIG_NETFILTER_NETLINK_GLUE_CT
Signature · 1 line
2.34.1
author maintainer
#3 lore raw

RFC v3 3/15 netfilter: ecache: move to separate structure

In reply to Florian Westphal

This makes it easier for a followup patch to only expose ecache related parts of nf_conntrack_net structure.

---
 include/net/netfilter/nf_conntrack.h |  8 ++++++--
 net/netfilter/nf_conntrack_ecache.c  | 19 ++++++++++---------
 2 files changed, 16 insertions(+), 11 deletions(-)
include/net/netfilter/nf_conntrack.h +6 -2
index 8731d5bcb47d..20fefe043850 100644
@@ -43,6 +43,11 @@ union nf_conntrack_expect_proto {
 	/* insert expect proto private data here */
 };
 
+struct nf_conntrack_net_ecache {
+	struct delayed_work dwork;
+	struct netns_ct *ct_net;
+};
+
 struct nf_conntrack_net {
 	/* only used when new connection is allocated: */
 	atomic_t count;
@@ -58,8 +63,7 @@ struct nf_conntrack_net {
 	struct ctl_table_header	*sysctl_header;
 #endif
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-	struct delayed_work ecache_dwork;
-	struct netns_ct *ct_net;
+	struct nf_conntrack_net_ecache ecache;
 #endif
 };
 
net/netfilter/nf_conntrack_ecache.c +10 -9
index 13ba6aa82ec1..d7a07873e605 100644
@@ -96,8 +96,8 @@ static enum retry_state ecache_work_evict_list(struct ct_pcpu *pcpu)
 
 static void ecache_work(struct work_struct *work)
 {
-	struct nf_conntrack_net *cnet = container_of(work, struct nf_conntrack_net, ecache_dwork.work);
-	struct netns_ct *ctnet = cnet->ct_net;
+	struct nf_conntrack_net *cnet = container_of(work, struct nf_conntrack_net, ecache.dwork.work);
+	struct netns_ct *ctnet = cnet->ecache.ct_net;
 	int cpu, delay = -1;
 	struct ct_pcpu *pcpu;
 
@@ -127,7 +127,7 @@ static void ecache_work(struct work_struct *work)
 
 	ctnet->ecache_dwork_pending = delay > 0;
 	if (delay >= 0)
-		schedule_delayed_work(&cnet->ecache_dwork, delay);
+		schedule_delayed_work(&cnet->ecache.dwork, delay);
 }
 
 static int __nf_conntrack_eventmask_report(struct nf_conntrack_ecache *e,
@@ -281,12 +281,12 @@ void nf_conntrack_ecache_work(struct net *net, enum nf_ct_ecache_state state)
 	struct nf_conntrack_net *cnet = nf_ct_pernet(net);
 
 	if (state == NFCT_ECACHE_DESTROY_FAIL &&
-	    !delayed_work_pending(&cnet->ecache_dwork)) {
-		schedule_delayed_work(&cnet->ecache_dwork, HZ);
+	    !delayed_work_pending(&cnet->ecache.dwork)) {
+		schedule_delayed_work(&cnet->ecache.dwork, HZ);
 		net->ct.ecache_dwork_pending = true;
 	} else if (state == NFCT_ECACHE_DESTROY_SENT) {
 		net->ct.ecache_dwork_pending = false;
-		mod_delayed_work(system_wq, &cnet->ecache_dwork, 0);
+		mod_delayed_work(system_wq, &cnet->ecache.dwork, 0);
 	}
 }
 
@@ -298,8 +298,9 @@ void nf_conntrack_ecache_pernet_init(struct net *net)
 	struct nf_conntrack_net *cnet = nf_ct_pernet(net);
 
 	net->ct.sysctl_events = nf_ct_events;
-	cnet->ct_net = &net->ct;
-	INIT_DELAYED_WORK(&cnet->ecache_dwork, ecache_work);
+
+	cnet->ecache.ct_net = &net->ct;
+	INIT_DELAYED_WORK(&cnet->ecache.dwork, ecache_work);
 
 	BUILD_BUG_ON(__IPCT_MAX >= 16);	/* e->ctmask is u16 */
 }
@@ -308,5 +309,5 @@ void nf_conntrack_ecache_pernet_fini(struct net *net)
 {
 	struct nf_conntrack_net *cnet = nf_ct_pernet(net);
 
-	cancel_delayed_work_sync(&cnet->ecache_dwork);
+	cancel_delayed_work_sync(&cnet->ecache.dwork);
 }
Signature · 1 line
2.34.1
author maintainer
#4 lore raw

RFC v3 4/15 netfilter: ecache: use dedicated list for event redelivery

In reply to Florian Westphal

This disentangles event redelivery and the percpu dying list.

Because entries are now stored on a dedicated list, all entries are in NFCT_ECACHE_DESTROY_FAIL state and all entries still have confirmed bit set -- the reference count is at least 1.

The 'struct net' back-pointer can be removed as well.

The pcpu dying list will be removed eventually, it has no functionality.

---
 include/net/netfilter/nf_conntrack.h        |   3 +-
 include/net/netfilter/nf_conntrack_ecache.h |   1 -
 net/netfilter/nf_conntrack_core.c           |  33 +++++-
 net/netfilter/nf_conntrack_ecache.c         | 118 +++++++++-----------
 4 files changed, 83 insertions(+), 72 deletions(-)
include/net/netfilter/nf_conntrack.h +2 -1
index 20fefe043850..dbbb0e206901 100644
@@ -45,7 +45,8 @@ union nf_conntrack_expect_proto {
 
 struct nf_conntrack_net_ecache {
 	struct delayed_work dwork;
-	struct netns_ct *ct_net;
+	spinlock_t dying_lock;
+	struct hlist_nulls_head dying_list;
 };
 
 struct nf_conntrack_net {
include/net/netfilter/nf_conntrack_ecache.h +0 -1
index ebd816af9c40..c63a8fc3225e 100644
@@ -14,7 +14,6 @@
 #include <net/netfilter/nf_conntrack_extend.h>
 
 enum nf_ct_ecache_state {
-	NFCT_ECACHE_UNKNOWN,		/* destroy event not sent */
 	NFCT_ECACHE_DESTROY_FAIL,	/* tried but failed to send destroy event */
 	NFCT_ECACHE_DESTROY_SENT,	/* sent destroy event after failure */
 };
net/netfilter/nf_conntrack_core.c +28 -5
index 9b7f9c966f73..7eefcfa55fc2 100644
@@ -647,15 +647,12 @@ void nf_ct_destroy(struct nf_conntrack *nfct)
 }
 EXPORT_SYMBOL(nf_ct_destroy);
 
-static void nf_ct_delete_from_lists(struct nf_conn *ct)
+static void __nf_ct_delete_from_lists(struct nf_conn *ct)
 {
 	struct net *net = nf_ct_net(ct);
 	unsigned int hash, reply_hash;
 	unsigned int sequence;
 
-	nf_ct_helper_destroy(ct);
-
-	local_bh_disable();
 	do {
 		sequence = read_seqcount_begin(&nf_conntrack_generation);
 		hash = hash_conntrack(net,
@@ -668,12 +665,33 @@ static void nf_ct_delete_from_lists(struct nf_conn *ct)
 
 	clean_from_lists(ct);
 	nf_conntrack_double_unlock(hash, reply_hash);
+}
 
+static void nf_ct_delete_from_lists(struct nf_conn *ct)
+{
+	nf_ct_helper_destroy(ct);
+	local_bh_disable();
+
+	__nf_ct_delete_from_lists(ct);
 	nf_ct_add_to_dying_list(ct);
 
 	local_bh_enable();
 }
 
+static void nf_ct_add_to_ecache_list(struct nf_conn *ct)
+{
+#ifdef CONFIG_NF_CONNTRACK_EVENTS
+	struct nf_conntrack_net *cnet = nf_ct_pernet(nf_ct_net(ct));
+
+	spin_lock(&cnet->ecache.dying_lock);
+	hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
+				 &cnet->ecache.dying_list);
+	spin_unlock(&cnet->ecache.dying_lock);
+#else
+	nf_ct_add_to_dying_list(ct);
+#endif
+}
+
 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 {
 	struct nf_conn_tstamp *tstamp;
@@ -696,7 +714,12 @@ bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 		/* destroy event was not delivered. nf_ct_put will
 		 * be done by event cache worker on redelivery.
 		 */
-		nf_ct_delete_from_lists(ct);
+		nf_ct_helper_destroy(ct);
+		local_bh_disable();
+		__nf_ct_delete_from_lists(ct);
+		nf_ct_add_to_ecache_list(ct);
+		local_bh_enable();
+
 		nf_conntrack_ecache_work(nf_ct_net(ct), NFCT_ECACHE_DESTROY_FAIL);
 		return false;
 	}
net/netfilter/nf_conntrack_ecache.c +53 -65

184 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#5 lore raw

RFC v3 5/15 netfilter: conntrack: split inner loop of list dumping to own function

In reply to Florian Westphal

This allows code re-use in the followup patch.
No functional changes intended.

---
 net/netfilter/nf_conntrack_netlink.c | 68 ++++++++++++++++++----------
 1 file changed, 43 insertions(+), 25 deletions(-)
net/netfilter/nf_conntrack_netlink.c +43 -25
index 4a460565f275..5f3d211a41e3 100644
@@ -1708,6 +1708,47 @@ static int ctnetlink_done_list(struct netlink_callback *cb)
 	return 0;
 }
 
+static int ctnetlink_dump_one_entry(struct sk_buff *skb,
+				    struct netlink_callback *cb,
+				    struct nf_conn *ct,
+				    bool dying)
+{
+	struct ctnetlink_list_dump_ctx *ctx = (void *)cb->ctx;
+	struct nfgenmsg *nfmsg = nlmsg_data(cb->nlh);
+	u8 l3proto = nfmsg->nfgen_family;
+	int res;
+
+	if (l3proto && nf_ct_l3num(ct) != l3proto)
+		return 0;
+
+	if (ctx->last) {
+		if (ct != ctx->last)
+			return 0;
+
+		ctx->last = NULL;
+	}
+
+	/* We can't dump extension info for the unconfirmed
+	 * list because unconfirmed conntracks can have
+	 * ct->ext reallocated (and thus freed).
+	 *
+	 * In the dying list case ct->ext can't be free'd
+	 * until after we drop pcpu->lock.
+	 */
+	res = ctnetlink_fill_info(skb, NETLINK_CB(cb->skb).portid,
+				  cb->nlh->nlmsg_seq,
+				  NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
+				  ct, dying, 0);
+	if (res < 0) {
+		if (!refcount_inc_not_zero(&ct->ct_general.use))
+			return 0;
+
+		ctx->last = ct;
+	}
+
+	return res;
+}
+
 static int
 ctnetlink_dump_list(struct sk_buff *skb, struct netlink_callback *cb, bool dying)
 {
@@ -1715,12 +1756,9 @@ ctnetlink_dump_list(struct sk_buff *skb, struct netlink_callback *cb, bool dying
 	struct nf_conn *ct, *last;
 	struct nf_conntrack_tuple_hash *h;
 	struct hlist_nulls_node *n;
-	struct nfgenmsg *nfmsg = nlmsg_data(cb->nlh);
-	u_int8_t l3proto = nfmsg->nfgen_family;
-	int res;
-	int cpu;
 	struct hlist_nulls_head *list;
 	struct net *net = sock_net(skb->sk);
+	int res, cpu;
 
 	if (ctx->done)
 		return 0;
@@ -1739,30 +1777,10 @@ ctnetlink_dump_list(struct sk_buff *skb, struct netlink_callback *cb, bool dying
 restart:
 		hlist_nulls_for_each_entry(h, n, list, hnnode) {
 			ct = nf_ct_tuplehash_to_ctrack(h);
-			if (l3proto && nf_ct_l3num(ct) != l3proto)
-				continue;
-			if (ctx->last) {
-				if (ct != last)
-					continue;
-				ctx->last = NULL;
-			}
 
-			/* We can't dump extension info for the unconfirmed
-			 * list because unconfirmed conntracks can have
-			 * ct->ext reallocated (and thus freed).
-			 *
-			 * In the dying list case ct->ext can't be free'd
-			 * until after we drop pcpu->lock.
-			 */
-			res = ctnetlink_fill_info(skb, NETLINK_CB(cb->skb).portid,
-						  cb->nlh->nlmsg_seq,
-						  NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
-						  ct, dying, 0);
+			res = ctnetlink_dump_one_entry(skb, cb, ct, dying);
 			if (res < 0) {
-				if (!refcount_inc_not_zero(&ct->ct_general.use))
-					continue;
 				ctx->cpu = cpu;
-				ctx->last = ct;
 				spin_unlock_bh(&pcpu->lock);
 				goto out;
 			}
Signature · 1 line
2.34.1
author maintainer
#6 lore raw

RFC v3 6/15 netfilter: conntrack: include ecache dying list in dumps

In reply to Florian Westphal

The new pernet dying list includes conntrack entries that await delivery of the 'destroy' event via ctnetlink.

The old percpu dying list will be removed soon.

---
 include/net/netfilter/nf_conntrack_ecache.h |  2 +
 net/netfilter/nf_conntrack_ecache.c         | 10 +++++
 net/netfilter/nf_conntrack_netlink.c        | 43 +++++++++++++++++++++
 3 files changed, 55 insertions(+)
include/net/netfilter/nf_conntrack_ecache.h +2 -0
index c63a8fc3225e..54051e663ff4 100644
@@ -161,6 +161,8 @@ void nf_conntrack_ecache_work(struct net *net, enum nf_ct_ecache_state state);
 void nf_conntrack_ecache_pernet_init(struct net *net);
 void nf_conntrack_ecache_pernet_fini(struct net *net);
 
+struct nf_conntrack_net_ecache *nf_conn_pernet_ecache(const struct net *net);
+
 static inline bool nf_conntrack_ecache_dwork_pending(const struct net *net)
 {
 	return net->ct.ecache_dwork_pending;
net/netfilter/nf_conntrack_ecache.c +10 -0
index be111218899d..ce3ebd420585 100644
@@ -38,6 +38,16 @@ enum retry_state {
 	STATE_DONE,
 };
 
+struct nf_conntrack_net_ecache *nf_conn_pernet_ecache(const struct net *net)
+{
+	struct nf_conntrack_net *cnet = nf_ct_pernet(net);
+
+	return &cnet->ecache;
+}
+#if IS_MODULE(CONFIG_NF_CT_NETLINK)
+EXPORT_SYMBOL_GPL(nf_conn_pernet_ecache);
+#endif
+
 static enum retry_state ecache_work_evict_list(struct nf_conntrack_net *cnet)
 {
 	unsigned long stop = jiffies + ECACHE_MAX_JIFFIES;
net/netfilter/nf_conntrack_netlink.c +43 -0

55 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#7 lore raw

RFC v3 7/15 netfilter: conntrack: remove the percpu dying list

In reply to Florian Westphal

Its no longer needed. Entries that need event redelivery are placed on the new pernet dying list.

The advantage is that there is no need to take additional spinlock on conntrack removal unless event redelivery failed or the conntrack entry was never added to the table in the first place (confirmed bit not set).

---
 include/net/netns/conntrack.h        |  1 -
 net/netfilter/nf_conntrack_core.c    | 35 +++++-----------------------
 net/netfilter/nf_conntrack_ecache.c  |  1 -
 net/netfilter/nf_conntrack_netlink.c | 23 ++++++------------
 4 files changed, 13 insertions(+), 47 deletions(-)
include/net/netns/conntrack.h +0 -1
index 3bb62e938fa9..dd1d096b2ada 100644
@@ -96,7 +96,6 @@ struct nf_ip_net {
 struct ct_pcpu {
 	spinlock_t		lock;
 	struct hlist_nulls_head unconfirmed;
-	struct hlist_nulls_head dying;
 };
 
 struct netns_ct {
net/netfilter/nf_conntrack_core.c +6 -29
index 7eefcfa55fc2..6f7471ba0744 100644
@@ -512,21 +512,6 @@ clean_from_lists(struct nf_conn *ct)
 	nf_ct_remove_expectations(ct);
 }
 
-/* must be called with local_bh_disable */
-static void nf_ct_add_to_dying_list(struct nf_conn *ct)
-{
-	struct ct_pcpu *pcpu;
-
-	/* add this conntrack to the (per cpu) dying list */
-	ct->cpu = smp_processor_id();
-	pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
-
-	spin_lock(&pcpu->lock);
-	hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
-			     &pcpu->dying);
-	spin_unlock(&pcpu->lock);
-}
-
 /* must be called with local_bh_disable */
 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 {
@@ -543,11 +528,11 @@ static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 }
 
 /* must be called with local_bh_disable */
-static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
+static void nf_ct_del_from_unconfirmed_list(struct nf_conn *ct)
 {
 	struct ct_pcpu *pcpu;
 
-	/* We overload first tuple to link into unconfirmed or dying list.*/
+	/* We overload first tuple to link into unconfirmed list.*/
 	pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 
 	spin_lock(&pcpu->lock);
@@ -635,7 +620,8 @@ void nf_ct_destroy(struct nf_conntrack *nfct)
 	 */
 	nf_ct_remove_expectations(ct);
 
-	nf_ct_del_from_dying_or_unconfirmed_list(ct);
+	if (unlikely(!nf_ct_is_confirmed(ct)))
+		nf_ct_del_from_unconfirmed_list(ct);
 
 	local_bh_enable();
 
@@ -673,7 +659,6 @@ static void nf_ct_delete_from_lists(struct nf_conn *ct)
 	local_bh_disable();
 
 	__nf_ct_delete_from_lists(ct);
-	nf_ct_add_to_dying_list(ct);
 
 	local_bh_enable();
 }
@@ -687,8 +672,6 @@ static void nf_ct_add_to_ecache_list(struct nf_conn *ct)
 	hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 				 &cnet->ecache.dying_list);
 	spin_unlock(&cnet->ecache.dying_lock);
-#else
-	nf_ct_add_to_dying_list(ct);
 #endif
 }
 
@@ -982,7 +965,6 @@ static void __nf_conntrack_insert_prepare(struct nf_conn *ct)
 	struct nf_conn_tstamp *tstamp;
 
 	refcount_inc(&ct->ct_general.use);
-	ct->status |= IPS_CONFIRMED;
 
 	/* set conntrack timestamp, if enabled. */
 	tstamp = nf_conn_tstamp_find(ct);
@@ -1011,7 +993,6 @@ static int __nf_ct_resolve_clash(struct sk_buff *skb,
 		nf_conntrack_get(&ct->ct_general);
 
 		nf_ct_acct_merge(ct, ctinfo, loser_ct);
-		nf_ct_add_to_dying_list(loser_ct);
 		nf_ct_put(loser_ct);
 		nf_ct_set(skb, ct, ctinfo);
 
@@ -1144,7 +1125,6 @@ nf_ct_resolve_clash(struct sk_buff *skb, struct nf_conntrack_tuple_hash *h,
 		return ret;
 
 drop:
-	nf_ct_add_to_dying_list(loser_ct);
 	NF_CT_STAT_INC(net, drop);
 	NF_CT_STAT_INC(net, insert_failed);
 	return NF_DROP;
@@ -1211,10 +1191,10 @@ __nf_conntrack_confirm(struct sk_buff *skb)
 	 * user context, else we insert an already 'dead' hash, blocking
 	 * further use of that particular connection -JM.
 	 */
-	nf_ct_del_from_dying_or_unconfirmed_list(ct);
+	nf_ct_del_from_unconfirmed_list(ct);
+	ct->status |= IPS_CONFIRMED;
 
 	if (unlikely(nf_ct_is_dying(ct))) {
-		nf_ct_add_to_dying_list(ct);
 		NF_CT_STAT_INC(net, insert_failed);
 		goto dying;
 	}
@@ -1238,7 +1218,6 @@ __nf_conntrack_confirm(struct sk_buff *skb)
 			goto out;
 		if (chainlen++ > max_chainlen) {
 chaintoolong:
-			nf_ct_add_to_dying_list(ct);
 			NF_CT_STAT_INC(net, chaintoolong);
 			NF_CT_STAT_INC(net, insert_failed);
 			ret = NF_DROP;
@@ -2752,7 +2731,6 @@ void nf_conntrack_init_end(void)
  * We need to use special "null" values, not used in hash table
  */
 #define UNCONFIRMED_NULLS_VAL	((1<<30)+0)
-#define DYING_NULLS_VAL		((1<<30)+1)
 
 int nf_conntrack_init_net(struct net *net)
 {
@@ -2773,7 +2751,6 @@ int nf_conntrack_init_net(struct net *net)
 
 		spin_lock_init(&pcpu->lock);
 		INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
-		INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
 	}
 
 	net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
net/netfilter/nf_conntrack_ecache.c +0 -1
index ce3ebd420585..a6ae8fd0c00e 100644
@@ -95,7 +95,6 @@ static enum retry_state ecache_work_evict_list(struct nf_conntrack_net *cnet)
 	hlist_nulls_for_each_entry_safe(h, n, &evicted_list, hnnode) {
 		struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 
-		hlist_nulls_add_fake(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 		hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 		nf_ct_put(ct);
 
net/netfilter/nf_conntrack_netlink.c +7 -16
index 3cd0d9d35b13..51e387e8da85 100644
@@ -62,7 +62,6 @@ struct ctnetlink_list_dump_ctx {
 	struct nf_conn *last;
 	unsigned int cpu;
 	bool done;
-	bool retrans_done;
 };
 
 static int ctnetlink_dump_tuples_proto(struct sk_buff *skb,
@@ -1751,13 +1750,12 @@ static int ctnetlink_dump_one_entry(struct sk_buff *skb,
 }
 
 static int
-ctnetlink_dump_list(struct sk_buff *skb, struct netlink_callback *cb, bool dying)
+ctnetlink_dump_unconfirmed(struct sk_buff *skb, struct netlink_callback *cb)
 {
 	struct ctnetlink_list_dump_ctx *ctx = (void *)cb->ctx;
 	struct nf_conn *ct, *last;
 	struct nf_conntrack_tuple_hash *h;
 	struct hlist_nulls_node *n;
-	struct hlist_nulls_head *list;
 	struct net *net = sock_net(skb->sk);
 	int res, cpu;
 
@@ -1774,12 +1772,11 @@ ctnetlink_dump_list(struct sk_buff *skb, struct netlink_callback *cb, bool dying
 
 		pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
 		spin_lock_bh(&pcpu->lock);
-		list = dying ? &pcpu->dying : &pcpu->unconfirmed;
 restart:
-		hlist_nulls_for_each_entry(h, n, list, hnnode) {
+		hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
 			ct = nf_ct_tuplehash_to_ctrack(h);
 
-			res = ctnetlink_dump_one_entry(skb, cb, ct, dying);
+			res = ctnetlink_dump_one_entry(skb, cb, ct, false);
 			if (res < 0) {
 				ctx->cpu = cpu;
 				spin_unlock_bh(&pcpu->lock);
@@ -1812,8 +1809,8 @@ ctnetlink_dump_dying(struct sk_buff *skb, struct netlink_callback *cb)
 	struct hlist_nulls_node *n;
 #endif
 
-	if (ctx->retrans_done)
-		return ctnetlink_dump_list(skb, cb, true);
+	if (ctx->done)
+		return 0;
 
 	ctx->last = NULL;
 
@@ -1842,10 +1839,10 @@ ctnetlink_dump_dying(struct sk_buff *skb, struct netlink_callback *cb)
 
 	spin_unlock_bh(&ecache_net->dying_lock);
 #endif
+	ctx->done = true;
 	nf_ct_put(last);
-	ctx->retrans_done = true;
 
-	return ctnetlink_dump_list(skb, cb, true);
+	return skb->len;
 }
 
 static int ctnetlink_get_ct_dying(struct sk_buff *skb,
@@ -1863,12 +1860,6 @@ static int ctnetlink_get_ct_dying(struct sk_buff *skb,
 	return -EOPNOTSUPP;
 }
 
-static int
-ctnetlink_dump_unconfirmed(struct sk_buff *skb, struct netlink_callback *cb)
-{
-	return ctnetlink_dump_list(skb, cb, false);
-}
-
 static int ctnetlink_get_ct_unconfirmed(struct sk_buff *skb,
 					const struct nfnl_info *info,
 					const struct nlattr * const cda[])
Signature · 1 line
2.34.1
author maintainer
#8 lore raw

RFC v3 8/15 netfilter: cttimeout: inc/dec module refcount per object, not per use refcount

In reply to Florian Westphal

There is no need to increment the module refcount again, its enough to obtain one reference per object, i.e. take a reference on object creation and put it on object destruction.

---
 net/netfilter/nfnetlink_cttimeout.c | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)
net/netfilter/nfnetlink_cttimeout.c +5 -9
index b0d8888a539b..eea486f32971 100644
@@ -158,6 +158,7 @@ static int cttimeout_new_timeout(struct sk_buff *skb,
 	timeout->timeout.l3num = l3num;
 	timeout->timeout.l4proto = l4proto;
 	refcount_set(&timeout->refcnt, 1);
+	__module_get(THIS_MODULE);
 	list_add_tail_rcu(&timeout->head, &pernet->nfct_timeout_list);
 
 	return 0;
@@ -506,13 +507,8 @@ static struct nf_ct_timeout *ctnl_timeout_find_get(struct net *net,
 		if (strncmp(timeout->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
 			continue;
 
-		if (!try_module_get(THIS_MODULE))
+		if (!refcount_inc_not_zero(&timeout->refcnt))
 			goto err;
-
-		if (!refcount_inc_not_zero(&timeout->refcnt)) {
-			module_put(THIS_MODULE);
-			goto err;
-		}
 		matching = timeout;
 		break;
 	}
@@ -525,10 +521,10 @@ static void ctnl_timeout_put(struct nf_ct_timeout *t)
 	struct ctnl_timeout *timeout =
 		container_of(t, struct ctnl_timeout, timeout);
 
-	if (refcount_dec_and_test(&timeout->refcnt))
+	if (refcount_dec_and_test(&timeout->refcnt)) {
 		kfree_rcu(timeout, rcu_head);
-
-	module_put(THIS_MODULE);
+		module_put(THIS_MODULE);
+	}
 }
 
 static const struct nfnl_callback cttimeout_cb[IPCTNL_MSG_TIMEOUT_MAX] = {
Signature · 1 line
2.34.1
author maintainer
#9 lore raw

RFC v3 9/15 netfilter: nfnetlink_cttimeout: use rcu protection in cttimeout_get_timeout

In reply to Florian Westphal

I'd like to be able to switch lifetime management of ctnl_timeout to free-on-zero-refcount.

This isn't possible at the moment because removal of the structures from the pernet list requires the nfnl mutex and release may happen from softirq.

Current solution is to prevent this by disallowing policy object removal if the refcount is > 1 (i.e., policy is still referenced from the ruleset).

Switch traversal to rcu-read-lock as a first step to reduce reliance on nfnl mutex protection: removal from softirq would require a extra list spinlock.

---
 net/netfilter/nfnetlink_cttimeout.c | 27 ++++++++++++++-------------
 1 file changed, 14 insertions(+), 13 deletions(-)
net/netfilter/nfnetlink_cttimeout.c +14 -13
index eea486f32971..c9cdc605f486 100644
@@ -253,6 +253,7 @@ static int cttimeout_get_timeout(struct sk_buff *skb,
 				 const struct nlattr * const cda[])
 {
 	struct nfct_timeout_pernet *pernet = nfct_timeout_pernet(info->net);
+	struct sk_buff *skb2;
 	int ret = -ENOENT;
 	char *name;
 	struct ctnl_timeout *cur;
@@ -268,31 +269,31 @@ static int cttimeout_get_timeout(struct sk_buff *skb,
 		return -EINVAL;
 	name = nla_data(cda[CTA_TIMEOUT_NAME]);
 
-	list_for_each_entry(cur, &pernet->nfct_timeout_list, head) {
-		struct sk_buff *skb2;
+	skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+	if (skb2 == NULL)
+		return -ENOMEM;
 
+	rcu_read_lock();
+
+	list_for_each_entry_rcu(cur, &pernet->nfct_timeout_list, head) {
 		if (strncmp(cur->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
 			continue;
 
-		skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
-		if (skb2 == NULL) {
-			ret = -ENOMEM;
-			break;
-		}
-
 		ret = ctnl_timeout_fill_info(skb2, NETLINK_CB(skb).portid,
 					     info->nlh->nlmsg_seq,
 					     NFNL_MSG_TYPE(info->nlh->nlmsg_type),
 					     IPCTNL_MSG_TIMEOUT_NEW, cur);
-		if (ret <= 0) {
-			kfree_skb(skb2);
+		if (ret <= 0)
 			break;
-		}
 
-		ret = nfnetlink_unicast(skb2, info->net, NETLINK_CB(skb).portid);
-		break;
+		rcu_read_unlock();
+
+		return nfnetlink_unicast(skb2, info->net, NETLINK_CB(skb).portid);
 	}
 
+	rcu_read_unlock();
+	kfree_skb(skb2);
+
 	return ret;
 }
 
Signature · 1 line
2.34.1
author maintainer
#10 lore raw

RFC v3 10/15 netfilter: cttimeout: decouple unlink and free on netns destruction

In reply to Florian Westphal

Make it so netns pre_exit unlinks the objects from the pernet list, so they cannot be found anymore.

netns core issues a synchronize_rcu() before calling the exit hooks so any the time the exit hooks run unconfirmed nf_conn entries have been free'd or they have been committed to the hashtable.

The exit hook still tags unconfirmed entries as dying, this can now be removed in a followup change.

---
 include/net/netfilter/nf_conntrack_timeout.h |  8 ------
 net/netfilter/nfnetlink_cttimeout.c          | 30 ++++++++++++++++++--
 2 files changed, 28 insertions(+), 10 deletions(-)
include/net/netfilter/nf_conntrack_timeout.h +0 -8
index 3ea94f6f3844..fea258983d23 100644
@@ -17,14 +17,6 @@ struct nf_ct_timeout {
 	char			data[];
 };
 
-struct ctnl_timeout {
-	struct list_head	head;
-	struct rcu_head		rcu_head;
-	refcount_t		refcnt;
-	char			name[CTNL_TIMEOUT_NAME_MAX];
-	struct nf_ct_timeout	timeout;
-};
-
 struct nf_conn_timeout {
 	struct nf_ct_timeout __rcu *timeout;
 };
net/netfilter/nfnetlink_cttimeout.c +28 -2
index c9cdc605f486..5bd660b45976 100644
@@ -33,8 +33,19 @@
 
 static unsigned int nfct_timeout_id __read_mostly;
 
+struct ctnl_timeout {
+	struct list_head	head;
+	struct rcu_head		rcu_head;
+	refcount_t		refcnt;
+	char			name[CTNL_TIMEOUT_NAME_MAX];
+	struct nf_ct_timeout	timeout;
+
+	struct list_head	free_head;
+};
+
 struct nfct_timeout_pernet {
 	struct list_head	nfct_timeout_list;
+	struct list_head	nfct_timeout_freelist;
 };
 
 MODULE_LICENSE("GPL");
@@ -575,10 +586,24 @@ static int __net_init cttimeout_net_init(struct net *net)
 	struct nfct_timeout_pernet *pernet = nfct_timeout_pernet(net);
 
 	INIT_LIST_HEAD(&pernet->nfct_timeout_list);
+	INIT_LIST_HEAD(&pernet->nfct_timeout_freelist);
 
 	return 0;
 }
 
+static void __net_exit cttimeout_net_pre_exit(struct net *net)
+{
+	struct nfct_timeout_pernet *pernet = nfct_timeout_pernet(net);
+	struct ctnl_timeout *cur, *tmp;
+
+	list_for_each_entry_safe(cur, tmp, &pernet->nfct_timeout_list, head) {
+		list_del_rcu(&cur->head);
+		list_add(&cur->free_head, &pernet->nfct_timeout_freelist);
+	}
+
+	/* core calls synchronize_rcu() after this */
+}
+
 static void __net_exit cttimeout_net_exit(struct net *net)
 {
 	struct nfct_timeout_pernet *pernet = nfct_timeout_pernet(net);
@@ -587,8 +612,8 @@ static void __net_exit cttimeout_net_exit(struct net *net)
 	nf_ct_unconfirmed_destroy(net);
 	nf_ct_untimeout(net, NULL);
 
-	list_for_each_entry_safe(cur, tmp, &pernet->nfct_timeout_list, head) {
-		list_del_rcu(&cur->head);
+	list_for_each_entry_safe(cur, tmp, &pernet->nfct_timeout_freelist, head) {
+		list_del(&cur->free_head);
 
 		if (refcount_dec_and_test(&cur->refcnt))
 			kfree_rcu(cur, rcu_head);
@@ -597,6 +622,7 @@ static void __net_exit cttimeout_net_exit(struct net *net)
 
 static struct pernet_operations cttimeout_ops = {
 	.init	= cttimeout_net_init,
+	.pre_exit = cttimeout_net_pre_exit,
 	.exit	= cttimeout_net_exit,
 	.id     = &nfct_timeout_id,
 	.size   = sizeof(struct nfct_timeout_pernet),
Signature · 1 line
2.34.1
author maintainer
#11 lore raw

RFC v3 11/15 netfilter: remove nf_ct_unconfirmed_destroy helper

In reply to Florian Westphal

This helper tags connetions not yet in the conntrack table as dying. These nf_conn entries will be dropped instead when the core attempts to insert them from the input or postrouting 'confirm' hook.

After the previous change, the entries get unlinked from the list earlier, so that by the time the actual exit hook runs, new connections no longer have a timeout policy assigned.

Its enough to walk the hashtable instead.

---
 include/net/netfilter/nf_conntrack.h |  3 ---
 net/netfilter/nf_conntrack_core.c    | 14 --------------
 net/netfilter/nfnetlink_cttimeout.c  |  4 +++-
 3 files changed, 3 insertions(+), 18 deletions(-)
include/net/netfilter/nf_conntrack.h +0 -3
index dbbb0e206901..c823f0e33dcc 100644
@@ -238,9 +238,6 @@ static inline bool nf_ct_kill(struct nf_conn *ct)
 	return nf_ct_delete(ct, 0, 0);
 }
 
-/* Set all unconfirmed conntrack as dying */
-void nf_ct_unconfirmed_destroy(struct net *);
-
 /* Iterate over all conntracks: if iter returns true, it's deleted. */
 void nf_ct_iterate_cleanup_net(struct net *net,
 			       int (*iter)(struct nf_conn *i, void *data),
net/netfilter/nf_conntrack_core.c +0 -14
index 6f7471ba0744..cbfd79dae8e7 100644
@@ -2383,20 +2383,6 @@ __nf_ct_unconfirmed_destroy(struct net *net)
 	}
 }
 
-void nf_ct_unconfirmed_destroy(struct net *net)
-{
-	struct nf_conntrack_net *cnet = nf_ct_pernet(net);
-
-	might_sleep();
-
-	if (atomic_read(&cnet->count) > 0) {
-		__nf_ct_unconfirmed_destroy(net);
-		nf_queue_nf_hook_drop(net);
-		synchronize_net();
-	}
-}
-EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
-
 void nf_ct_iterate_cleanup_net(struct net *net,
 			       int (*iter)(struct nf_conn *i, void *data),
 			       void *data, u32 portid, int report)
net/netfilter/nfnetlink_cttimeout.c +3 -1
index 5bd660b45976..a98cf956f7c7 100644
@@ -609,7 +609,9 @@ static void __net_exit cttimeout_net_exit(struct net *net)
 	struct nfct_timeout_pernet *pernet = nfct_timeout_pernet(net);
 	struct ctnl_timeout *cur, *tmp;
 
-	nf_ct_unconfirmed_destroy(net);
+	if (list_empty(&pernet->nfct_timeout_freelist))
+		return;
+
 	nf_ct_untimeout(net, NULL);
 
 	list_for_each_entry_safe(cur, tmp, &pernet->nfct_timeout_freelist, head) {
Signature · 1 line
2.34.1
author maintainer
#12 lore raw

RFC v3 12/15 netfilter: extensions: introduce extension genid count

In reply to Florian Westphal

Multiple netfilter extensions store pointers to external data in their extension area struct.

Examples:
1. Timeout policies
2. Connection tracking helpers.

No references are taken for these.

When a helper or timeout policy is removed, the conntrack table gets traversed and affected extensions are cleared.

Conntrack entries not yet in the hashtable are referenced via a special list, the unconfirmed list.

On removal of a policy or connection tracking helper, the unconfirmed list gets traversed an all entries are marked as dying, this prevents them from getting committed to the table at insertion time: core checks for dying bit, if set, the conntrack entry gets destroyed at confirm time.

The disadvantage is that each new conntrack has to be added to the percpu unconfirmed list, and each insertion needs to remove it from this list. The list is only ever needed when a policy or helper is removed -- a rare occurence.

Add a generation ID count: Instead of adding to the list and then traversing that list on policy/helper removal, increment a counter that is stored in the extension area.

For unconfirmed conntracks, the extension has the genid valid at conntrack allocation time.

Removal of a helper/policy etc. increments the counter. At confirmation time, validate that ext->genid == global_id.

If the stored number is not the same, do not allow the conntrack insertion, just like as if a confirmed-list traversal would have flagged the entry as dying.

After insertion, the genid is no longer relevant (conntrack entries are now reachable via the conntrack table iterators and is set to 0.

This allows removal of the percpu unconfirmed list.

---
 include/net/netfilter/nf_conntrack_extend.h | 31 ++++++------
 net/netfilter/nf_conntrack_core.c           | 55 +++++++++++++++++++++
 net/netfilter/nf_conntrack_extend.c         | 32 +++++++++++-
 3 files changed, 102 insertions(+), 16 deletions(-)
include/net/netfilter/nf_conntrack_extend.h +16 -15
index 96635ad2acc7..0b247248b032 100644
@@ -34,21 +34,11 @@ enum nf_ct_ext_id {
 	NF_CT_EXT_NUM,
 };
 
-#define NF_CT_EXT_HELPER_TYPE struct nf_conn_help
-#define NF_CT_EXT_NAT_TYPE struct nf_conn_nat
-#define NF_CT_EXT_SEQADJ_TYPE struct nf_conn_seqadj
-#define NF_CT_EXT_ACCT_TYPE struct nf_conn_acct
-#define NF_CT_EXT_ECACHE_TYPE struct nf_conntrack_ecache
-#define NF_CT_EXT_TSTAMP_TYPE struct nf_conn_tstamp
-#define NF_CT_EXT_TIMEOUT_TYPE struct nf_conn_timeout
-#define NF_CT_EXT_LABELS_TYPE struct nf_conn_labels
-#define NF_CT_EXT_SYNPROXY_TYPE struct nf_conn_synproxy
-#define NF_CT_EXT_ACT_CT_TYPE struct nf_conn_act_ct_ext
-
 /* Extensions: optional stuff which isn't permanently in struct. */
 struct nf_ct_ext {
 	u8 offset[NF_CT_EXT_NUM];
 	u8 len;
+	unsigned int gen_id;
 	char data[] __aligned(8);
 };
 
@@ -62,17 +52,28 @@ static inline bool nf_ct_ext_exist(const struct nf_conn *ct, u8 id)
 	return (ct->ext && __nf_ct_ext_exist(ct->ext, id));
 }
 
-static inline void *__nf_ct_ext_find(const struct nf_conn *ct, u8 id)
+void *__nf_ct_ext_find(const struct nf_ct_ext *ext, u8 id);
+
+static inline void *nf_ct_ext_find(const struct nf_conn *ct, u8 id)
 {
-	if (!nf_ct_ext_exist(ct, id))
+	struct nf_ct_ext *ext = ct->ext;
+
+	if (!ext || !__nf_ct_ext_exist(ext, id))
 		return NULL;
 
+	if (unlikely(ext->gen_id))
+		return __nf_ct_ext_find(ext, id);
+
 	return (void *)ct->ext + ct->ext->offset[id];
 }
-#define nf_ct_ext_find(ext, id)	\
-	((id##_TYPE *)__nf_ct_ext_find((ext), (id)))
 
 /* Add this type, returns pointer to data or NULL. */
 void *nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp);
 
+/* ext genid.  if ext->id != ext_genid, extensions cannot be used
+ * anymore unless conntrack has CONFIRMED bit set.
+ */
+extern atomic_t nf_conntrack_ext_genid;
+void nf_ct_ext_bump_genid(void);
+
 #endif /* _NF_CONNTRACK_EXTEND_H */
net/netfilter/nf_conntrack_core.c +55 -0

91 lines not on this page · raw message

net/netfilter/nf_conntrack_extend.c +31 -1
index 1296fda54ac6..0b513f7bf9f3 100644
@@ -27,6 +27,8 @@
 
 #define NF_CT_EXT_PREALLOC	128u /* conntrack events are on by default */
 
+atomic_t nf_conntrack_ext_genid __read_mostly = ATOMIC_INIT(1);
+
 static const u8 nf_ct_ext_type_len[NF_CT_EXT_NUM] = {
 	[NF_CT_EXT_HELPER] = sizeof(struct nf_conn_help),
 #if IS_ENABLED(CONFIG_NF_NAT)
@@ -116,8 +118,10 @@ void *nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 	if (!new)
 		return NULL;
 
-	if (!ct->ext)
+	if (!ct->ext) {
 		memset(new->offset, 0, sizeof(new->offset));
+		new->gen_id = atomic_read(&nf_conntrack_ext_genid);
+	}
 
 	new->offset[id] = newoff;
 	new->len = newlen;
@@ -127,3 +131,29 @@ void *nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 	return (void *)new + newoff;
 }
 EXPORT_SYMBOL(nf_ct_ext_add);
+
+/* Use nf_ct_ext_find wrapper. This is only useful for unconfirmed entries. */
+void *__nf_ct_ext_find(const struct nf_ct_ext *ext, u8 id)
+{
+	unsigned int gen_id = atomic_read(&nf_conntrack_ext_genid);
+	unsigned int this_id = READ_ONCE(ext->gen_id);
+
+	if (!__nf_ct_ext_exist(ext, id))
+		return NULL;
+
+	if (this_id == 0 || ext->gen_id == gen_id)
+		return (void *)ext + ext->offset[id];
+
+	return NULL;
+}
+EXPORT_SYMBOL(__nf_ct_ext_find);
+
+void nf_ct_ext_bump_genid(void)
+{
+	unsigned int value = atomic_inc_return(&nf_conntrack_ext_genid);
+
+	if (value == UINT_MAX)
+		atomic_set(&nf_conntrack_ext_genid, 1);
+
+	msleep(HZ);
+}
Signature · 1 line
2.34.1
author maintainer
#13 lore raw

RFC v3 13/15 netfilter: cttimeout: decouple unlink and free on netns destruction

In reply to Florian Westphal

Increment the extid on module removal; this makes sure that even in extreme cases any old uncofirmed entry that happened to be kept e.g. on nfnetlink_queue list will not trip over a stale timeout reference.

---
 net/netfilter/nfnetlink_cttimeout.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
net/netfilter/nfnetlink_cttimeout.c +12 -0
index a98cf956f7c7..fecebc991e6f 100644
@@ -657,12 +657,24 @@ static int __init cttimeout_init(void)
 	return ret;
 }
 
+static int untimeout(struct nf_conn *ct, void *timeout)
+{
+	struct nf_conn_timeout *timeout_ext = nf_ct_timeout_find(ct);
+
+	if (timeout_ext)
+		RCU_INIT_POINTER(timeout_ext->timeout, NULL);
+
+	return 0;
+}
+
 static void __exit cttimeout_exit(void)
 {
 	nfnetlink_subsys_unregister(&cttimeout_subsys);
 
 	unregister_pernet_subsys(&cttimeout_ops);
 	RCU_INIT_POINTER(nf_ct_timeout_hook, NULL);
+
+	nf_ct_iterate_destroy(untimeout, NULL);
 	synchronize_rcu();
 }
 
Signature · 1 line
2.34.1
author maintainer
#14 lore raw

RFC v3 14/15 netfilter: conntrack: remove __nf_ct_unconfirmed_destroy

In reply to Florian Westphal

Its not needed anymore:

A. If entry is totally new, then the rcu-protected resource must already have been removed from global visibility before call to nf_ct_iterate_destroy.

B. If entry was allocated before, but is not yet in the hash table
(uncofirmed case), genid gets incremented and synchronize_rcu() call makes sure access has completed.

C. Next attempt to peek at extension area will fail for unconfirmed
conntracks, because ext->genid != genid.

D. Conntracks in the hash are iterated as before.

---
 net/netfilter/nf_conntrack_core.c   | 46 ++++++++---------------------
 net/netfilter/nf_conntrack_helper.c |  5 ----
 net/netfilter/nfnetlink_cttimeout.c |  1 -
 3 files changed, 13 insertions(+), 39 deletions(-)
net/netfilter/nf_conntrack_core.c +13 -33
index 7c775a6c27ff..3c89f8fd2203 100644
@@ -2409,34 +2409,6 @@ static int iter_net_only(struct nf_conn *i, void *data)
 	return d->iter(i, d->data);
 }
 
-static void
-__nf_ct_unconfirmed_destroy(struct net *net)
-{
-	int cpu;
-
-	for_each_possible_cpu(cpu) {
-		struct nf_conntrack_tuple_hash *h;
-		struct hlist_nulls_node *n;
-		struct ct_pcpu *pcpu;
-
-		pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
-
-		spin_lock_bh(&pcpu->lock);
-		hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
-			struct nf_conn *ct;
-
-			ct = nf_ct_tuplehash_to_ctrack(h);
-
-			/* we cannot call iter() on unconfirmed list, the
-			 * owning cpu can reallocate ct->ext at any time.
-			 */
-			set_bit(IPS_DYING_BIT, &ct->status);
-		}
-		spin_unlock_bh(&pcpu->lock);
-		cond_resched();
-	}
-}
-
 void nf_ct_iterate_cleanup_net(struct net *net,
 			       int (*iter)(struct nf_conn *i, void *data),
 			       void *data, u32 portid, int report)
@@ -2479,26 +2451,34 @@ nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
 
 		if (atomic_read(&cnet->count) == 0)
 			continue;
-		__nf_ct_unconfirmed_destroy(net);
 		nf_queue_nf_hook_drop(net);
 	}
 	up_read(&net_rwsem);
 
 	/* Need to wait for netns cleanup worker to finish, if its
 	 * running -- it might have deleted a net namespace from
-	 * the global list, so our __nf_ct_unconfirmed_destroy() might
-	 * not have affected all namespaces.
+	 * the global list, so hook drop above might not have
+	 * affected all namespaces.
 	 */
 	net_ns_barrier();
 
-	/* a conntrack could have been unlinked from unconfirmed list
-	 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
+	/* a skb w. unconfirmed conntrack could have been reinjected just
+	 * before we called nf_queue_nf_hook_drop().
+	 *
 	 * This makes sure its inserted into conntrack table.
 	 */
 	synchronize_net();
 
 	nf_ct_ext_bump_genid();
 	nf_ct_iterate_cleanup(iter, data, 0, 0);
+
+	/* Another cpu might be in a rcu read section with
+	 * rcu protected pointer cleared in iter callback
+	 * or hidden via nf_ct_ext_bump_genid() above.
+	 *
+	 * Wait until those are done.
+	 */
+	synchronize_rcu();
 }
 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
 
net/netfilter/nf_conntrack_helper.c +0 -5
index a97ddb1497aa..c44ac2d2b4b6 100644
@@ -468,11 +468,6 @@ void nf_conntrack_helper_unregister(struct nf_conntrack_helper *me)
 
 	nf_ct_expect_iterate_destroy(expect_iter_me, NULL);
 	nf_ct_iterate_destroy(unhelp, me);
-
-	/* Maybe someone has gotten the helper already when unhelp above.
-	 * So need to wait it.
-	 */
-	synchronize_rcu();
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_helper_unregister);
 
net/netfilter/nfnetlink_cttimeout.c +0 -1
index fecebc991e6f..6128a34db210 100644
@@ -675,7 +675,6 @@ static void __exit cttimeout_exit(void)
 	RCU_INIT_POINTER(nf_ct_timeout_hook, NULL);
 
 	nf_ct_iterate_destroy(untimeout, NULL);
-	synchronize_rcu();
 }
 
 module_init(cttimeout_init);
Signature · 1 line
2.34.1
author maintainer
#15 lore raw

RFC v3 15/15 netfilter: conntrack: remove unconfirmed list

In reply to Florian Westphal

It has no function anymore and can be removed.

---
 include/net/netfilter/nf_conntrack.h |  1 -
 include/net/netns/conntrack.h        |  6 ----
 net/netfilter/nf_conntrack_core.c    | 54 ++--------------------------
 net/netfilter/nf_conntrack_netlink.c | 44 +----------------------
 4 files changed, 3 insertions(+), 102 deletions(-)
include/net/netfilter/nf_conntrack.h +0 -1
index c823f0e33dcc..096b898860d6 100644
@@ -101,7 +101,6 @@ struct nf_conn {
 	/* Have we seen traffic both ways yet? (bitset) */
 	unsigned long status;
 
-	u16		cpu;
 	u16		local_origin:1;
 	possible_net_t ct_net;
 
include/net/netns/conntrack.h +0 -6
index dd1d096b2ada..22d8d1909a3b 100644
@@ -93,11 +93,6 @@ struct nf_ip_net {
 #endif
 };
 
-struct ct_pcpu {
-	spinlock_t		lock;
-	struct hlist_nulls_head unconfirmed;
-};
-
 struct netns_ct {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
 	bool ecache_dwork_pending;
@@ -109,7 +104,6 @@ struct netns_ct {
 	u8			sysctl_tstamp;
 	u8			sysctl_checksum;
 
-	struct ct_pcpu __percpu *pcpu_lists;
 	struct ip_conntrack_stat __percpu *stat;
 	struct nf_ip_net	nf_ct_proto;
 #if defined(CONFIG_NF_CONNTRACK_LABELS)
net/netfilter/nf_conntrack_core.c +2 -52

103 lines not on this page · raw message

net/netfilter/nf_conntrack_netlink.c +1 -43

50 lines not on this page · raw message

Signature · 1 line
2.34.1