net: bridge: br_vlan_get_pvid_rcu() should dereference the VLAN group under RCU
authorVladimir Oltean <vladimir.oltean@nxp.com>
Mon, 21 Sep 2020 22:07:09 +0000 (01:07 +0300)
committerDavid S. Miller <davem@davemloft.net>
Tue, 22 Sep 2020 00:37:44 +0000 (17:37 -0700)
commit99f62a746066fa436aa15d4606a538569540db08
treeb7fdc47fff971eaa6b97174ee47df797f24fdc2e
parent47cec3f68c6ca33fdfe381934c6c581cd716df96
net: bridge: br_vlan_get_pvid_rcu() should dereference the VLAN group under RCU

When calling the RCU brother of br_vlan_get_pvid(), lockdep warns:

=============================
WARNING: suspicious RCU usage
5.9.0-rc3-01631-g13c17acb8e38-dirty #814 Not tainted
-----------------------------
net/bridge/br_private.h:1054 suspicious rcu_dereference_protected() usage!

Call trace:
 lockdep_rcu_suspicious+0xd4/0xf8
 __br_vlan_get_pvid+0xc0/0x100
 br_vlan_get_pvid_rcu+0x78/0x108

The warning is because br_vlan_get_pvid_rcu() calls nbp_vlan_group()
which calls rtnl_dereference() instead of rcu_dereference(). In turn,
rtnl_dereference() calls rcu_dereference_protected() which assumes
operation under an RCU write-side critical section, which obviously is
not the case here. So, when the incorrect primitive is used to access
the RCU-protected VLAN group pointer, READ_ONCE() is not used, which may
cause various unexpected problems.

I'm sad to say that br_vlan_get_pvid() and br_vlan_get_pvid_rcu() cannot
share the same implementation. So fix the bug by splitting the 2
functions, and making br_vlan_get_pvid_rcu() retrieve the VLAN groups
under proper locking annotations.

Fixes: 7582f5b70f9a ("bridge: add br_vlan_get_pvid_rcu()")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/bridge/br_vlan.c