Ocfs2 devel - Jan 2019 - [PATCH v2] ocfs2: fix a panic problem caused by o2cb_ctl

In the process of creating a node, it will cause NULL pointer
dereference in kernel if o2cb_ctl failed in the interval
(mkdir, o2cb_set_node_attribute(node_num)] in function o2cb_add_node.

The node num is initialized to 0 in function o2nm_node_group_make_item,
o2nm_node_group_drop_item will mistake the node number 0 for a
valid node number when we delete the node before the node number is set
correctly. If the local node number of the current host happens to be 0,
cluster->cl_local_node will be set to O2NM_INVALID_NODE_NUM while
o2hb_thread still running. The panic stack is generated as follows:

o2hb_thread
    \-o2hb_do_disk_heartbeat
        \-o2hb_check_own_slot
            |-slot = &reg->hr_slots[o2nm_this_node()];
            //o2nm_this_node() return O2NM_INVALID_NODE_NUM

We need to check whether the node number is set when we delete the node.

Signed-off-by: Jia Guo <guojia12 at huawei.com>
---
 fs/ocfs2/cluster/nodemanager.c | 22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)

diff --git a/fs/ocfs2/cluster/nodemanager.c b/fs/ocfs2/cluster/nodemanager.c
index 0e4166c..989abad 100644
--- a/fs/ocfs2/cluster/nodemanager.c
+++ b/fs/ocfs2/cluster/nodemanager.c
@@ -620,14 +620,19 @@ static void o2nm_node_group_drop_item(struct config_group
*group,
 {
 	struct o2nm_node *node = to_o2nm_node(item);
 	struct o2nm_cluster *cluster = to_o2nm_cluster(group->cg_item.ci_parent);
+	int nd_num_set = 0;

-	o2net_disconnect_node(node);
-
-	if (cluster->cl_has_local &&
-	    (cluster->cl_local_node == node->nd_num)) {
-		cluster->cl_has_local = 0;
-		cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
-		o2net_stop_listening(node);
+	/* nd_num might be 0 if the node number hasn't been set.. */
+	if (cluster->cl_nodes[node->nd_num] == node) {
+		nd_num_set = 1;
+		o2net_disconnect_node(node);
+
+		if (cluster->cl_has_local &&
+		    (cluster->cl_local_node == node->nd_num)) {
+			cluster->cl_has_local = 0;
+			cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
+			o2net_stop_listening(node);
+		}
 	}

 	/* XXX call into net to stop this node from trading messages */
@@ -638,8 +643,7 @@ static void o2nm_node_group_drop_item(struct config_group
*group,
 	if (node->nd_ipv4_address)
 		rb_erase(&node->nd_ip_node, &cluster->cl_node_ip_tree);

-	/* nd_num might be 0 if the node number hasn't been set.. */
-	if (cluster->cl_nodes[node->nd_num] == node) {
+	if (nd_num_set) {
 		cluster->cl_nodes[node->nd_num] = NULL;
 		clear_bit(node->nd_num, cluster->cl_nodes_bitmap);
 	}

Hi,

On 19/1/26 16:19, Jia Guo wrote:
> In the process of creating a node, it will cause NULL pointer
> dereference in kernel if o2cb_ctl failed in the interval
> (mkdir, o2cb_set_node_attribute(node_num)] in function o2cb_add_node.
> 
> The node num is initialized to 0 in function o2nm_node_group_make_item,
> o2nm_node_group_drop_item will mistake the node number 0 for a
> valid node number when we delete the node before the node number is set
> correctly. If the local node number of the current host happens to be 0,
> cluster->cl_local_node will be set to O2NM_INVALID_NODE_NUM while
> o2hb_thread still running. The panic stack is generated as follows:
> 
> o2hb_thread
>     \-o2hb_do_disk_heartbeat
>         \-o2hb_check_own_slot
>             |-slot = &reg->hr_slots[o2nm_this_node()];
>             //o2nm_this_node() return O2NM_INVALID_NODE_NUM
> 
> We need to check whether the node number is set when we delete the node.
> 
> Signed-off-by: Jia Guo <guojia12 at huawei.com>
> ---
>  fs/ocfs2/cluster/nodemanager.c | 22 +++++++++++++---------
>  1 file changed, 13 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/ocfs2/cluster/nodemanager.c
b/fs/ocfs2/cluster/nodemanager.c
> index 0e4166c..989abad 100644
> --- a/fs/ocfs2/cluster/nodemanager.c
> +++ b/fs/ocfs2/cluster/nodemanager.c
> @@ -620,14 +620,19 @@ static void o2nm_node_group_drop_item(struct
config_group *group,
>  {
>  	struct o2nm_node *node = to_o2nm_node(item);
>  	struct o2nm_cluster *cluster =
to_o2nm_cluster(group->cg_item.ci_parent);
> +	int nd_num_set = 0;
Could we just add the condition for the first part, and leave the second
part no change? Then we won't introduce the local nd_num_set.

With the above comments addressed, you can add:
Reviewed-by: Joseph Qi <jiangqi903 at gmail.com>
> 
> -	o2net_disconnect_node(node);
> -
> -	if (cluster->cl_has_local &&
> -	    (cluster->cl_local_node == node->nd_num)) {
> -		cluster->cl_has_local = 0;
> -		cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
> -		o2net_stop_listening(node);
> +	/* nd_num might be 0 if the node number hasn't been set.. */
> +	if (cluster->cl_nodes[node->nd_num] == node) {
> +		nd_num_set = 1;
> +		o2net_disconnect_node(node);
> +
> +		if (cluster->cl_has_local &&
> +		    (cluster->cl_local_node == node->nd_num)) {
> +			cluster->cl_has_local = 0;
> +			cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
> +			o2net_stop_listening(node);
> +		}
>  	}
> 
>  	/* XXX call into net to stop this node from trading messages */
> @@ -638,8 +643,7 @@ static void o2nm_node_group_drop_item(struct
config_group *group,
>  	if (node->nd_ipv4_address)
>  		rb_erase(&node->nd_ip_node, &cluster->cl_node_ip_tree);
> 
> -	/* nd_num might be 0 if the node number hasn't been set.. */
> -	if (cluster->cl_nodes[node->nd_num] == node) {
> +	if (nd_num_set) {
>  		cluster->cl_nodes[node->nd_num] = NULL;
>  		clear_bit(node->nd_num, cluster->cl_nodes_bitmap);
>  	}
>