zhangzhengming
2021-Apr-28 07:29 UTC
[Bridge] [PATCH] bridge: Fix possible races between assigning rx_handler_data and setting IFF_BRIDGE_PORT bit
From: Zhang Zhengming <zhangzhengming at huawei.com> There is a crash in the function br_get_link_af_size_filtered, as the port_exists(dev) is true and the rx_handler_data of dev is NULL. But the rx_handler_data of dev is correct saved in vmcore. The oops looks something like: ... pc : br_get_link_af_size_filtered+0x28/0x1c8 [bridge] ... Call trace: br_get_link_af_size_filtered+0x28/0x1c8 [bridge] if_nlmsg_size+0x180/0x1b0 rtnl_calcit.isra.12+0xf8/0x148 rtnetlink_rcv_msg+0x334/0x370 netlink_rcv_skb+0x64/0x130 rtnetlink_rcv+0x28/0x38 netlink_unicast+0x1f0/0x250 netlink_sendmsg+0x310/0x378 sock_sendmsg+0x4c/0x70 __sys_sendto+0x120/0x150 __arm64_sys_sendto+0x30/0x40 el0_svc_common+0x78/0x130 el0_svc_handler+0x38/0x78 el0_svc+0x8/0xc In br_add_if(), we found there is no guarantee that assigning rx_handler_data to dev->rx_handler_data will before setting the IFF_BRIDGE_PORT bit of priv_flags. So there is a possible data competition: CPU 0: CPU 1: (RCU read lock) (RTNL lock) rtnl_calcit() br_add_slave() if_nlmsg_size() br_add_if() br_get_link_af_size_filtered() -> netdev_rx_handler_register ... // The order is not guaranteed ... -> dev->priv_flags |= IFF_BRIDGE_PORT; // The IFF_BRIDGE_PORT bit of priv_flags has been set -> if (br_port_exists(dev)) { // The dev->rx_handler_data has NOT been assigned -> p = br_port_get_rcu(dev); .... -> rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); ... Fix this by adding memory barrier instruction to ensure the order. Signed-off-by: Zhang Zhengming <zhangzhengming at huawei.com> Reviewed-by: Zhao Lei <zhaolei69 at huawei.com> Reviewed-by: Wang Xiaogang <wangxiaogang3 at huawei.com> --- net/bridge/br_if.c | 6 ++++++ net/bridge/br_netlink.c | 1 + 2 files changed, 7 insertions(+) diff --git a/net/bridge/br_if.c b/net/bridge/br_if.c index f7d2f47..42895be 100644 --- a/net/bridge/br_if.c +++ b/net/bridge/br_if.c @@ -636,6 +636,12 @@ int br_add_if(struct net_bridge *br, struct net_device *dev, err = netdev_rx_handler_register(dev, br_get_rx_handler(dev), p); if (err) goto err4; + + /* Make sure dev->rx_handler_data is written in netdev_rx_handler_register + * before the bit IFF_BRIDGE_PORT of dev->priv_flags is set. + * coupled with smp_rmb() in br_get_link_af_size_filtered. + */ + smp_wmb(); dev->priv_flags |= IFF_BRIDGE_PORT; diff --git a/net/bridge/br_netlink.c b/net/bridge/br_netlink.c index f2b1343..ccc1fd7 100644 --- a/net/bridge/br_netlink.c +++ b/net/bridge/br_netlink.c @@ -103,6 +103,7 @@ static size_t br_get_link_af_size_filtered(const struct net_device *dev, rcu_read_lock(); if (netif_is_bridge_port(dev)) { + smp_rmb(); /* coupled with smp_wmb() in br_add_if() */ p = br_port_get_rcu(dev); vg = nbp_vlan_group_rcu(p); } else if (dev->priv_flags & IFF_EBRIDGE) { -- 2.7.4
Nikolay Aleksandrov
2021-Apr-28 08:00 UTC
[Bridge] [PATCH] bridge: Fix possible races between assigning rx_handler_data and setting IFF_BRIDGE_PORT bit
On 28/04/2021 10:29, zhangzhengming wrote:> From: Zhang Zhengming <zhangzhengming at huawei.com> > > There is a crash in the function br_get_link_af_size_filtered, > as the port_exists(dev) is true and the rx_handler_data of dev is NULL. > But the rx_handler_data of dev is correct saved in vmcore. > > The oops looks something like: > ... > pc : br_get_link_af_size_filtered+0x28/0x1c8 [bridge] > ... > Call trace: > br_get_link_af_size_filtered+0x28/0x1c8 [bridge] > if_nlmsg_size+0x180/0x1b0 > rtnl_calcit.isra.12+0xf8/0x148 > rtnetlink_rcv_msg+0x334/0x370 > netlink_rcv_skb+0x64/0x130 > rtnetlink_rcv+0x28/0x38 > netlink_unicast+0x1f0/0x250 > netlink_sendmsg+0x310/0x378 > sock_sendmsg+0x4c/0x70 > __sys_sendto+0x120/0x150 > __arm64_sys_sendto+0x30/0x40 > el0_svc_common+0x78/0x130 > el0_svc_handler+0x38/0x78 > el0_svc+0x8/0xc > > In br_add_if(), we found there is no guarantee that > assigning rx_handler_data to dev->rx_handler_data > will before setting the IFF_BRIDGE_PORT bit of priv_flags. > So there is a possible data competition: > > CPU 0: CPU 1: > (RCU read lock) (RTNL lock) > rtnl_calcit() br_add_slave() > if_nlmsg_size() br_add_if() > > -> netdev_rx_handler_register > ... > // The order is not guaranteed > ... -> dev->priv_flags |= IFF_BRIDGE_PORT; > // The IFF_BRIDGE_PORT bit of priv_flags has been set > -> if (br_port_exists(dev)) { > // The dev->rx_handler_data has NOT been assigned > -> p = br_port_get_rcu(dev); > .... > -> rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); > ... > > Fix this by adding memory barrier instruction to ensure the order. > > Signed-off-by: Zhang Zhengming <zhangzhengming at huawei.com> > Reviewed-by: Zhao Lei <zhaolei69 at huawei.com> > Reviewed-by: Wang Xiaogang <wangxiaogang3 at huawei.com> > --- > net/bridge/br_if.c | 6 ++++++ > net/bridge/br_netlink.c | 1 + > 2 files changed, 7 insertions(+) >No need for memory barriers, just use br_port_get_check_rcu() in br_get_link_af_size_filtered() and check the return value. It will be much simpler and shorter. Thanks> diff --git a/net/bridge/br_if.c b/net/bridge/br_if.c > index f7d2f47..42895be 100644 > --- a/net/bridge/br_if.c > +++ b/net/bridge/br_if.c > @@ -636,6 +636,12 @@ int br_add_if(struct net_bridge *br, struct net_device *dev, > err = netdev_rx_handler_register(dev, br_get_rx_handler(dev), p); > if (err) > goto err4; > + > + /* Make sure dev->rx_handler_data is written in netdev_rx_handler_register > + * before the bit IFF_BRIDGE_PORT of dev->priv_flags is set. > + * coupled with smp_rmb() in br_get_link_af_size_filtered. > + */ > + smp_wmb(); > > dev->priv_flags |= IFF_BRIDGE_PORT; > > diff --git a/net/bridge/br_netlink.c b/net/bridge/br_netlink.c > index f2b1343..ccc1fd7 100644 > --- a/net/bridge/br_netlink.c > +++ b/net/bridge/br_netlink.c > @@ -103,6 +103,7 @@ static size_t br_get_link_af_size_filtered(const struct net_device *dev, > > rcu_read_lock(); > if (netif_is_bridge_port(dev)) { > + smp_rmb(); /* coupled with smp_wmb() in br_add_if() */ > p = br_port_get_rcu(dev); > vg = nbp_vlan_group_rcu(p); > } else if (dev->priv_flags & IFF_EBRIDGE) { >