Skip to contentSkip to conversation

net: ipv4: fix circular-list infinite loop

PATCH In review Florian Westphal 4 participants 7 messages Jun 27, 2019 – Jul 1, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

cover net: ipv4: fix circular-list infinite loop

Tariq and Ran reported a regression caused by net-next commit 2638eb8b50cf ("net: ipv4: provide __rcu annotation for ifa_list").

This happens when net.ipv4.conf.$dev.promote_secondaries sysctl is enabled -- we can arrange for ifa->next to point at ifa, so next process that tries to walk the list loops forever.

Fix this and extend rtnetlink.sh with a small test case for this.

Florian Westphal (2):
net: ipv4: fix infinite loop on secondary addr promotion
selftests: rtnetlink: add small test case with 'promote_secondaries' enabled

 net/ipv4/devinet.c                       |    3 ++-
 tools/testing/selftests/net/rtnetlink.sh |   20 ++++++++++++++++++++
 2 files changed, 22 insertions(+), 1 deletion(-)
#1 lore raw

PATCH 1/2 net: ipv4: fix infinite loop on secondary addr promotion

secondary address promotion causes infinite loop -- it arranges for ifa->ifa_next to point back to itself.

Problem is that 'prev_prom' and 'last_prim' might point at the same entry, so 'last_sec' pointer must be obtained after prev_prom->next update.

Fixes 2638eb8b50cf ("net: ipv4: provide __rcu annotation for ifa_list")
---
 net/ipv4/devinet.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
net/ipv4/devinet.c +2 -1
index 7874303220c5..137d1892395d 100644
@@ -428,8 +428,9 @@ static void __inet_del_ifa(struct in_device *in_dev,
 		if (prev_prom) {
 			struct in_ifaddr *last_sec;
 
-			last_sec = rtnl_dereference(last_prim->ifa_next);
 			rcu_assign_pointer(prev_prom->ifa_next, next_sec);
+
+			last_sec = rtnl_dereference(last_prim->ifa_next);
 			rcu_assign_pointer(promote->ifa_next, last_sec);
 			rcu_assign_pointer(last_prim->ifa_next, promote);
 		}
Signature · 1 line
2.21.0
#2 lore raw

PATCH 2/2 selftests: rtnetlink: add small test case with 'promote_secondaries' enabled

In reply to Florian Westphal

This exercises the 'promote_secondaries' code path.

Without previous fix, this triggers infinite loop/soft lockup:
ifconfig process spinning at 100%, never to return.

---
 tools/testing/selftests/net/rtnetlink.sh | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
tools/testing/selftests/net/rtnetlink.sh +20 -0
index ed606a2e3865..505628884783 100755
@@ -269,6 +269,25 @@ kci_test_addrlft()
 	echo "PASS: preferred_lft addresses have expired"
 }
 
+kci_test_promote_secondaries()
+{
+	promote=$(sysctl -n net.ipv4.conf.$devdummy.promote_secondaries)
+
+	sysctl -q net.ipv4.conf.$devdummy.promote_secondaries=1
+
+	for i in $(seq 2 254);do
+		IP="10.23.11.$i"
+		ip -f inet addr add $IP/16 brd + dev "$devdummy"
+		ifconfig "$devdummy" $IP netmask 255.255.0.0
+	done
+
+	ip addr flush dev "$devdummy"
+
+	[ $promote -eq 0 ] && sysctl -q net.ipv4.conf.$devdummy.promote_secondaries=0
+
+	echo "PASS: promote_secondaries complete"
+}
+
 kci_test_addrlabel()
 {
 	ret=0
@@ -1161,6 +1180,7 @@ kci_test_rtnl()
 	kci_test_polrouting
 	kci_test_route_get
 	kci_test_addrlft
+	kci_test_promote_secondaries
 	kci_test_tc
 	kci_test_gre
 	kci_test_gretap
Signature · 1 line
2.21.0
maintainer
#3 lore raw
In reply to Florian Westphal

From: Florian Westphal <[email protected]>
Date: Thu, 27 Jun 2019 14:03:31 +0200

Quoted 8 lines

Tariq and Ran reported a regression caused by net-next commit 2638eb8b50cf ("net: ipv4: provide __rcu annotation for ifa_list").

This happens when net.ipv4.conf.$dev.promote_secondaries sysctl is enabled -- we can arrange for ifa->next to point at ifa, so next process that tries to walk the list loops forever.

Fix this and extend rtnetlink.sh with a small test case for this.

Series applied, thanks Florian.

#4 lore raw
Quoted 6 lines · David Miller

From: Florian Westphal <[email protected]>
Date: Thu, 27 Jun 2019 14:03:31 +0200

Quoted 3 lines

Tariq and Ran reported a regression caused by net-next commit 2638eb8b50cf ("net: ipv4: provide __rcu annotation for ifa_list").

Quoted 8 lines
Quoted 5 lines

This happens when net.ipv4.conf.$dev.promote_secondaries sysctl is enabled -- we can arrange for ifa->next to point at ifa, so next process that tries to walk the list loops forever.

Fix this and extend rtnetlink.sh with a small test case for this.

Series applied, thanks Florian.

Thanks Florian!

Ran, please test and update.

Tariq

#5 lore raw
Quoted 25 lines

-----Original Message-----
From: Tariq Toukan
Sent: Sunday, June 30, 2019 10:57
To: David Miller <[email protected]>; [email protected]
Cc: [email protected]; Ran Rozenstein <[email protected]>; Tariq Toukan <[email protected]>
Subject: Re: [PATCH net-next 0/2] net: ipv4: fix circular-list infinite loop

Quoted 12 lines · David Miller

From: Florian Westphal <[email protected]>
Date: Thu, 27 Jun 2019 14:03:31 +0200

Quoted 8 lines

Tariq and Ran reported a regression caused by net-next commit 2638eb8b50cf ("net: ipv4: provide __rcu annotation for ifa_list").

This happens when net.ipv4.conf.$dev.promote_secondaries sysctl is enabled -- we can arrange for ifa->next to point at ifa, so next process that tries to walk the list loops forever.

Fix this and extend rtnetlink.sh with a small test case for this.

Quoted 8 lines
Quoted 2 lines

Series applied, thanks Florian.

Thanks Florian!

Ran, please test and update.

Tariq

Thanks Florian.
Didn't reproduce tonight with the fixes.

Ran.

#6 lore raw
Quoted 30 lines · Ran Rozenstein
Quoted 30 lines

-----Original Message-----
From: Tariq Toukan
Sent: Sunday, June 30, 2019 10:57
To: David Miller <[email protected]>; [email protected]
Cc: [email protected]; Ran Rozenstein <[email protected]>; Tariq Toukan <[email protected]>
Subject: Re: [PATCH net-next 0/2] net: ipv4: fix circular-list infinite loop

Quoted 6 lines · David Miller

From: Florian Westphal <[email protected]>
Date: Thu, 27 Jun 2019 14:03:31 +0200

Quoted 3 lines

Tariq and Ran reported a regression caused by net-next commit 2638eb8b50cf ("net: ipv4: provide __rcu annotation for ifa_list").

Quoted 8 lines
Quoted 5 lines

This happens when net.ipv4.conf.$dev.promote_secondaries sysctl is enabled -- we can arrange for ifa->next to point at ifa, so next process that tries to walk the list loops forever.

Fix this and extend rtnetlink.sh with a small test case for this.

Series applied, thanks Florian.

Thanks Florian!

Quoted 9 lines
Quoted 3 lines

Ran, please test and update.

Tariq

Thanks Florian.
Didn't reproduce tonight with the fixes.

Ran.

Sounds good!

Thanks,
Tariq