From c18bdd018e8912ca73ad6c12120b7283b5038875 Mon Sep 17 00:00:00 2001 From: Sven Eckelmann Date: Sun, 31 Jan 2016 13:27:59 +0100 Subject: [PATCH 1/3] batman-adv: Only put gw_node list reference when removed The batadv_gw_node reference counter in batadv_gw_node_update can only be reduced when the list entry was actually removed. Otherwise the reference counter may reach zero when batadv_gw_node_update is called from two different contexts for the same gw_node but only one context is actually removing the entry from the list. The release function for this gw_node is not called inside the list_lock spinlock protected region because the function batadv_gw_node_update still holds a gw_node reference for the object pointer on the stack. Thus the actual release function (when required) will be called only at the end of the function. Fixes: bd3524c14bd0 ("batman-adv: remove obsolete deleted attribute for gateway node") Signed-off-by: Sven Eckelmann Signed-off-by: Marek Lindner Signed-off-by: Antonio Quartulli --- net/batman-adv/gateway_client.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/net/batman-adv/gateway_client.c b/net/batman-adv/gateway_client.c index e6c8382c79ba..ccf70bed0d0c 100644 --- a/net/batman-adv/gateway_client.c +++ b/net/batman-adv/gateway_client.c @@ -527,11 +527,12 @@ void batadv_gw_node_update(struct batadv_priv *bat_priv, * gets dereferenced. */ spin_lock_bh(&bat_priv->gw.list_lock); - hlist_del_init_rcu(&gw_node->list); + if (!hlist_unhashed(&gw_node->list)) { + hlist_del_init_rcu(&gw_node->list); + batadv_gw_node_free_ref(gw_node); + } spin_unlock_bh(&bat_priv->gw.list_lock); - batadv_gw_node_free_ref(gw_node); - curr_gw = batadv_gw_get_selected_gw_node(bat_priv); if (gw_node == curr_gw) batadv_gw_reselect(bat_priv); From 3db152093efb750bc47fd4d69355b90b18113105 Mon Sep 17 00:00:00 2001 From: Sven Eckelmann Date: Sun, 31 Jan 2016 13:28:00 +0100 Subject: [PATCH 2/3] batman-adv: Only put orig_node_vlan list reference when removed The batadv_orig_node_vlan reference counter in batadv_tt_global_size_mod can only be reduced when the list entry was actually removed. Otherwise the reference counter may reach zero when batadv_tt_global_size_mod is called from two different contexts for the same orig_node_vlan but only one context is actually removing the entry from the list. The release function for this orig_node_vlan is not called inside the vlan_list_lock spinlock protected region because the function batadv_tt_global_size_mod still holds a orig_node_vlan reference for the object pointer on the stack. Thus the actual release function (when required) will be called only at the end of the function. Fixes: 7ea7b4a14275 ("batman-adv: make the TT CRC logic VLAN specific") Signed-off-by: Sven Eckelmann Signed-off-by: Marek Lindner Signed-off-by: Antonio Quartulli --- net/batman-adv/translation-table.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/net/batman-adv/translation-table.c b/net/batman-adv/translation-table.c index cdfc85fa2743..0e80fd1461ab 100644 --- a/net/batman-adv/translation-table.c +++ b/net/batman-adv/translation-table.c @@ -303,9 +303,11 @@ static void batadv_tt_global_size_mod(struct batadv_orig_node *orig_node, if (atomic_add_return(v, &vlan->tt.num_entries) == 0) { spin_lock_bh(&orig_node->vlan_list_lock); - hlist_del_init_rcu(&vlan->list); + if (!hlist_unhashed(&vlan->list)) { + hlist_del_init_rcu(&vlan->list); + batadv_orig_node_vlan_free_ref(vlan); + } spin_unlock_bh(&orig_node->vlan_list_lock); - batadv_orig_node_vlan_free_ref(vlan); } batadv_orig_node_vlan_free_ref(vlan); From 1bc4e2b000e7fa9773d6623bc8850561ce10a4fb Mon Sep 17 00:00:00 2001 From: Andrew Lunn Date: Thu, 11 Feb 2016 22:15:57 +0100 Subject: [PATCH 3/3] batman-adv: Avoid endless loop in bat-on-bat netdevice check 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: [] ? rcu_dump_cpu_stacks+0x7e/0xd0 [] ? rcu_check_callbacks+0x3f0/0x6b0 [] ? hrtimer_run_queues+0x47/0x180 [] ? update_process_times+0x2d/0x50 [] ? tick_handle_periodic+0x1b/0x60 [] ? smp_trace_apic_timer_interrupt+0x5e/0x90 [] ? apic_timer_interrupt+0x82/0x90 [] ? __dev_get_by_index+0x37/0x40 [] ? batadv_hard_if_event+0xee/0x3a0 [batman_adv] [] ? 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 [sven@narfation.org: rewritten description, extracted fix] Signed-off-by: Sven Eckelmann Signed-off-by: Marek Lindner Signed-off-by: Antonio Quartulli --- 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 01acccc4d218..57f7107169f5 100644 --- a/net/batman-adv/hard-interface.c +++ b/net/batman-adv/hard-interface.c @@ -75,6 +75,28 @@ out: return hard_iface; } +/** + * 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 @@ -108,6 +130,9 @@ 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;