[PATCH 4.2.y-ckt 255/273] batman-adv: Avoid endless loop in bat-on-bat netdevice check

From: Kamal Mostafa
Date: Mon Mar 07 2016 - 18:05:51 EST


4.2.8-ckt5 -stable review patch. If anyone has any objections, please let me know.

---8<------------------------------------------------------------

From: Andrew Lunn <andrew@xxxxxxx>

commit 1bc4e2b000e7fa9773d6623bc8850561ce10a4fb upstream.

batman-adv checks in different situation if a new device is already on top
of a different batman-adv device. This is done by getting the iflink of a
device and all its parent. It assumes that this iflink is always a parent
device in an acyclic graph. But this assumption is broken by devices like
veth which are actually a pair of two devices linked to each other. The
recursive check would therefore get veth0 when calling dev_get_iflink on
veth1. And it gets veth0 when calling dev_get_iflink with veth1.

Creating a veth pair and loading batman-adv freezes parts of the system

ip link add veth0 type veth peer name veth1
modprobe batman-adv

An RCU stall will be detected on the system which cannot be fixed.

INFO: rcu_sched self-detected stall on CPU
1: (5264 ticks this GP) idle=3e9/140000000000001/0
softirq=144683/144686 fqs=5249
(t=5250 jiffies g=46 c=45 q=43)
Task dump for CPU 1:
insmod R running task 0 247 245 0x00000008
ffffffff8151f140 ffffffff8107888e ffff88000fd141c0 ffffffff8151f140
0000000000000000 ffffffff81552df0 ffffffff8107b420 0000000000000001
ffff88000e3fa700 ffffffff81540b00 ffffffff8107d667 0000000000000001
Call Trace:
<IRQ> [<ffffffff8107888e>] ? rcu_dump_cpu_stacks+0x7e/0xd0
[<ffffffff8107b420>] ? rcu_check_callbacks+0x3f0/0x6b0
[<ffffffff8107d667>] ? hrtimer_run_queues+0x47/0x180
[<ffffffff8107cf9d>] ? update_process_times+0x2d/0x50
[<ffffffff810873fb>] ? tick_handle_periodic+0x1b/0x60
[<ffffffff810290ae>] ? smp_trace_apic_timer_interrupt+0x5e/0x90
[<ffffffff813bbae2>] ? apic_timer_interrupt+0x82/0x90
<EOI> [<ffffffff812c3fd7>] ? __dev_get_by_index+0x37/0x40
[<ffffffffa0031f3e>] ? batadv_hard_if_event+0xee/0x3a0 [batman_adv]
[<ffffffff812c5801>] ? register_netdevice_notifier+0x81/0x1a0
[...]

This can be avoided by checking if two devices are each others parent and
stopping the check in this situation.

Fixes: b7eddd0b3950 ("batman-adv: prevent using any virtual device created on batman-adv as hard-interface")
Signed-off-by: Andrew Lunn <andrew@xxxxxxx>
[sven@xxxxxxxxxxxxx: rewritten description, extracted fix]
Signed-off-by: Sven Eckelmann <sven@xxxxxxxxxxxxx>
Signed-off-by: Marek Lindner <mareklindner@xxxxxxxxxxxxx>
Signed-off-by: Antonio Quartulli <a@xxxxxxxxxxx>

Signed-off-by: Kamal Mostafa <kamal@xxxxxxxxxxxxx>
---
net/batman-adv/hard-interface.c | 25 +++++++++++++++++++++++++
1 file changed, 25 insertions(+)

diff --git a/net/batman-adv/hard-interface.c b/net/batman-adv/hard-interface.c
index f4a15d2..3edf7b0 100644
--- a/net/batman-adv/hard-interface.c
+++ b/net/batman-adv/hard-interface.c
@@ -75,6 +75,28 @@ out:
}

/**
+ * batadv_mutual_parents - check if two devices are each others parent
+ * @dev1: 1st net_device
+ * @dev2: 2nd net_device
+ *
+ * veth devices come in pairs and each is the parent of the other!
+ *
+ * Return: true if the devices are each others parent, otherwise false
+ */
+static bool batadv_mutual_parents(const struct net_device *dev1,
+ const struct net_device *dev2)
+{
+ int dev1_parent_iflink = dev_get_iflink(dev1);
+ int dev2_parent_iflink = dev_get_iflink(dev2);
+
+ if (!dev1_parent_iflink || !dev2_parent_iflink)
+ return false;
+
+ return (dev1_parent_iflink == dev2->ifindex) &&
+ (dev2_parent_iflink == dev1->ifindex);
+}
+
+/**
* batadv_is_on_batman_iface - check if a device is a batman iface descendant
* @net_dev: the device to check
*
@@ -107,6 +129,9 @@ static bool batadv_is_on_batman_iface(const struct net_device *net_dev)
if (WARN(!parent_dev, "Cannot find parent device"))
return false;

+ if (batadv_mutual_parents(net_dev, parent_dev))
+ return false;
+
ret = batadv_is_on_batman_iface(parent_dev);

return ret;
--
2.7.0