Skip to contentSkip to conversation

netfilter: nfnetlink: re-enable conntrack expectation events

PATCH In review Florian Westphal 2 participants 2 messages Aug 5, 2022 – Aug 9, 2022

Subsystems (2)
lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

PATCH 1/1 netfilter: nfnetlink: re-enable conntrack expectation events

To avoid allocation of the conntrack extension area when possible, the default behaviour was changed to only allocate the event extension if a userspace program is subscribed to a notification group.

Problem is that while 'conntrack -E' does enable the event allocation behind the scenes, 'conntrack -E expect' does not: no expectation events are delivered unless user sets "net.netfilter.nf_conntrack_events" back to 1 (always on).

Fix the autodetection to also consider EXP type group.

We need to track the 6 event groups (3+3, new/update/destroy for events and for expectations each) independently, else we'd disable events again if an expectation group becomes empty while there is still an active event group.

Fixes 2794cdb0b97b ("netfilter: nfnetlink: allow to detect if ctnetlink listeners exist")
---
 include/net/netns/conntrack.h |  2 +-
 net/netfilter/nfnetlink.c     | 83 ++++++++++++++++++++++++++++++-----
 2 files changed, 72 insertions(+), 13 deletions(-)
include/net/netns/conntrack.h +1 -1
index 0677cd3de034..c396a3862e80 100644
@@ -95,7 +95,7 @@ struct nf_ip_net {
 
 struct netns_ct {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-	bool ctnetlink_has_listener;
+	u8 ctnetlink_has_listener;
 	bool ecache_dwork_pending;
 #endif
 	u8			sysctl_log_invalid; /* Log invalid packets */
net/netfilter/nfnetlink.c +71 -12

111 lines not on this page · raw message

Signature · 1 line
2.35.1
#1 lore raw

Hi Florian,

a few questions, comments.

Quoted 29 lines · Florian Westphal

To avoid allocation of the conntrack extension area when possible, the default behaviour was changed to only allocate the event extension if a userspace program is subscribed to a notification group.

Problem is that while 'conntrack -E' does enable the event allocation behind the scenes, 'conntrack -E expect' does not: no expectation events are delivered unless user sets "net.netfilter.nf_conntrack_events" back to 1 (always on).

Fix the autodetection to also consider EXP type group.

We need to track the 6 event groups (3+3, new/update/destroy for events and for expectations each) independently, else we'd disable events again if an expectation group becomes empty while there is still an active event group.

Fixes 2794cdb0b97b ("netfilter: nfnetlink: allow to detect if ctnetlink listeners exist")
---
 include/net/netns/conntrack.h |  2 +-
 net/netfilter/nfnetlink.c     | 83 ++++++++++++++++++++++++++++++-----
 2 files changed, 72 insertions(+), 13 deletions(-)
include/net/netns/conntrack.h +0 -0
index 0677cd3de034..c396a3862e80 100644

include/net/netns/conntrack.h struct nf_ip_net { · line 98

Quoted 5 lines
include/net/netns/conntrack.h +1 -1
@@ -95,7 +95,7 @@ struct nf_ip_net {
 
 struct netns_ct {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
-	bool ctnetlink_has_listener;
+	u8 ctnetlink_has_listener;

so now ctnetlink_has_listener is a bitmask, right?

Quoted 20 lines
include/net/netns/conntrack.h +0 -0
 	bool ecache_dwork_pending;
 #endif
 	u8			sysctl_log_invalid; /* Log invalid packets */
net/netfilter/nfnetlink.c +4 -0
index c24b1240908f..6c268f1c201a 100644
@@ -44,6 +44,10 @@ MODULE_DESCRIPTION("Netfilter messages via netlink socket");
 
 static unsigned int nfnetlink_pernet_id __read_mostly;
 
+#ifdef CONFIG_NF_CONNTRACK_EVENTS
+static DEFINE_SPINLOCK(nfnl_grp_active_lock);
+#endif
+
 struct nfnl_net {
 	struct sock *nfnl;
 };
@@ -654,6 +658,44 @@ static void nfnetlink_rcv(struct sk_buff *skb)
 		netlink_rcv_skb(skb, nfnetlink_rcv_msg);

net/netfilter/nfnetlink.c static void nfnetlink_rcv(struct sk_buff *skb) · line 668

Quoted 10 lines
net/netfilter/nfnetlink.c +8 -0
 }
 
+static void nfnetlink_bind_event(struct net *net, unsigned int group)
+{
+#ifdef CONFIG_NF_CONNTRACK_EVENTS
+	int type, group_bit;
+	u8 v;
+
+	/* ctnetlink_has_listener is u8, all NFNLGRP_CONNTRACK_* groups
+	 * coming from userspace are < 8.

Nit. is ctnetlink_has_listener relevant in the comment? It reads a bit confusing.

net/netfilter/nfnetlink.c static void nfnetlink_rcv(struct sk_buff *skb) · line 670

Quoted 2 lines
net/netfilter/nfnetlink.c +2 -0
+	 */
+	if (group >= 8)

Nit.

enum nfnetlink_groups has __NFNLGRP_MAX, which is 9.

This skips NFNLGRP_NFTRACE which is the last one in the array, not a big deal since the switch (type) results in return; but better fix this for the future?

Quoted 6 lines
net/netfilter/nfnetlink.c +6 -0
+		return;
+
+	type = nfnl_group2type[group];
+
+	switch (type) {
+	case NFNL_SUBSYS_CTNETLINK:

net/netfilter/nfnetlink.c static void nfnetlink_rcv(struct sk_buff *skb) · line 686

Quoted 10 lines
net/netfilter/nfnetlink.c +10 -0
+		break;
+	case NFNL_SUBSYS_CTNETLINK_EXP:
+		break;
+	default:
+		return;
+	}
+
+	group_bit = (1 << group);
+
+	spin_lock(&nfnl_grp_active_lock);

suggestion: maybe mutex instead?

Quoted 46 lines
net/netfilter/nfnetlink.c +18 -12
+	v = READ_ONCE(net->ct.ctnetlink_has_listener);
+	if ((v & group_bit) == 0) {
+		v |= group_bit;
+
+		/* read concurrently without nfnl_grp_active_lock held. */
+		WRITE_ONCE(net->ct.ctnetlink_has_listener, v);
+	}
+
+	spin_unlock(&nfnl_grp_active_lock);
+#endif
+}
+
 static int nfnetlink_bind(struct net *net, int group)
 {
 	const struct nfnetlink_subsystem *ss;
@@ -670,28 +712,45 @@ static int nfnetlink_bind(struct net *net, int group)
 	if (!ss)
 		request_module_nowait("nfnetlink-subsys-%d", type);
 
-#ifdef CONFIG_NF_CONNTRACK_EVENTS
-	if (type == NFNL_SUBSYS_CTNETLINK) {
-		nfnl_lock(NFNL_SUBSYS_CTNETLINK);
-		WRITE_ONCE(net->ct.ctnetlink_has_listener, true);
-		nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
-	}
-#endif
+	nfnetlink_bind_event(net, group);
 	return 0;
 }
 
 static void nfnetlink_unbind(struct net *net, int group)
 {
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
+	int type, group_bit;
+
 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
 		return;
 
-	if (nfnl_group2type[group] == NFNL_SUBSYS_CTNETLINK) {
-		nfnl_lock(NFNL_SUBSYS_CTNETLINK);
-		if (!nfnetlink_has_listeners(net, group))
-			WRITE_ONCE(net->ct.ctnetlink_has_listener, false);
-		nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
+	type = nfnl_group2type[group];
+
+	switch (type) {

net/netfilter/nfnetlink.c static int nfnetlink_bind(struct net *net, int group) · line 739

Quoted 10 lines
net/netfilter/nfnetlink.c +10 -0
+	case NFNL_SUBSYS_CTNETLINK:
+		break;
+	case NFNL_SUBSYS_CTNETLINK_EXP:
+		break;
+	default:
+		return;
+	}
+
+	/* ctnetlink_has_listener is u8 */
+	if (group >= 8)

Same here?

Quoted 20 lines
net/netfilter/nfnetlink.c +13 -0
+		return;
+
+	group_bit = (1 << group);
+
+	spin_lock(&nfnl_grp_active_lock);
+	if (!nfnetlink_has_listeners(net, group)) {
+		u8 v = READ_ONCE(net->ct.ctnetlink_has_listener);
+
+		v &= ~group_bit;
+
+		/* read concurrently without nfnl_grp_active_lock held. */
+		WRITE_ONCE(net->ct.ctnetlink_has_listener, v);
 	}
+	spin_unlock(&nfnl_grp_active_lock);
 #endif
 }
 
Signature · 1 line
2.35.1