Ocfs2 devel - Jun 2010 - [PATCH] ocfs2/dlm: check dlm_state under spinlock

We should check dlm->dlm_state under dlm->spinlock though maybe in this
case it
doesn't hurt.

Signed-off-by: Wengang Wang <wen.gang.wang at oracle.com>
---
 fs/ocfs2/dlm/dlmdomain.c |   13 ++++++-------
 1 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 6b5a492..ab82add 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -796,7 +796,7 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32
len, void *data,
 	spin_lock(&dlm_domain_lock);
 	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
 	if (!dlm)
-		goto unlock_respond;
+		goto unlock_domain_respond;
 
 	/*
 	 * There is a small window where the joining node may not see the
@@ -811,7 +811,7 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32
len, void *data,
 				     "have node %u in its nodemap\n",
 				     query->node_idx, nodenum);
 				packet.code = JOIN_DISALLOW;
-				goto unlock_respond;
+				goto unlock_domain_respond;
 			}
 		}
 		nodenum++;
@@ -821,9 +821,9 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32
len, void *data,
 	 * to be put in someone's domain map.
 	 * Also, explicitly disallow joining at certain troublesome
 	 * times (ie. during recovery). */
-	if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
+	spin_lock(&dlm->spinlock);
+	if (dlm->dlm_state != DLM_CTXT_LEAVING) {
 		int bit = query->node_idx;
-		spin_lock(&dlm->spinlock);
 
 		if (dlm->dlm_state == DLM_CTXT_NEW &&
 		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
@@ -869,10 +869,9 @@ static int dlm_query_join_handler(struct o2net_msg *msg,
u32 len, void *data,
 				__dlm_set_joining_node(dlm, query->node_idx);
 			}
 		}
-
-		spin_unlock(&dlm->spinlock);
 	}
-unlock_respond:
+	spin_unlock(&dlm->spinlock);
+unlock_domain_respond:
 	spin_unlock(&dlm_domain_lock);
 
 respond:
-- 
1.6.6.1

The lock order in this code causes dead lock, not caused by your patch. 
The lock order in dlm_query_join_handler is

dlm_domain_lock
    ->dlm->spinlock

dead locks with ..

dlm_lockres_put calls dlm_lockres_release while holding dlm->spinlock 
which calls dlm_put which gets dlm_domain_lock. So the spin lock order 
here is
dlm->spinlock
    -> dlm_domain_lock


On 6/15/2010 9:08 PM, Wengang Wang wrote:
> We should check dlm->dlm_state under dlm->spinlock though maybe in
this case it
> doesn't hurt.
>
> Signed-off-by: Wengang Wang <wen.gang.wang at oracle.com>
> ---
>  fs/ocfs2/dlm/dlmdomain.c |   13 ++++++-------
>  1 files changed, 6 insertions(+), 7 deletions(-)
>
> diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
> index 6b5a492..ab82add 100644
> --- a/fs/ocfs2/dlm/dlmdomain.c
> +++ b/fs/ocfs2/dlm/dlmdomain.c
> @@ -796,7 +796,7 @@ static int dlm_query_join_handler(struct o2net_msg
*msg, u32 len, void *data,
>  	spin_lock(&dlm_domain_lock);
>  	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
>  	if (!dlm)
> -		goto unlock_respond;
> +		goto unlock_domain_respond;
>  
>  	/*
>  	 * There is a small window where the joining node may not see the
> @@ -811,7 +811,7 @@ static int dlm_query_join_handler(struct o2net_msg
*msg, u32 len, void *data,
>  				     "have node %u in its nodemap\n",
>  				     query->node_idx, nodenum);
>  				packet.code = JOIN_DISALLOW;
> -				goto unlock_respond;
> +				goto unlock_domain_respond;
>  			}
>  		}
>  		nodenum++;
> @@ -821,9 +821,9 @@ static int dlm_query_join_handler(struct o2net_msg
*msg, u32 len, void *data,
>  	 * to be put in someone's domain map.
>  	 * Also, explicitly disallow joining at certain troublesome
>  	 * times (ie. during recovery). */
> -	if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
> +	spin_lock(&dlm->spinlock);
> +	if (dlm->dlm_state != DLM_CTXT_LEAVING) {
>  		int bit = query->node_idx;
> -		spin_lock(&dlm->spinlock);
>  
>  		if (dlm->dlm_state == DLM_CTXT_NEW &&
>  		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
> @@ -869,10 +869,9 @@ static int dlm_query_join_handler(struct o2net_msg
*msg, u32 len, void *data,
>  				__dlm_set_joining_node(dlm, query->node_idx);
>  			}
>  		}
> -
> -		spin_unlock(&dlm->spinlock);
>  	}
> -unlock_respond:
> +	spin_unlock(&dlm->spinlock);
> +unlock_domain_respond:
>  	spin_unlock(&dlm_domain_lock);
>  
>  respond:
>

On 06/15/2010 09:08 PM, Wengang Wang wrote:
> We should check dlm->dlm_state under dlm->spinlock though maybe in
this case it
> doesn't hurt.
>    
NAK.

dlm->dlm_state is protected by dlm_domain_lock which is held at that time.

         /* NOTE: Next three are protected by dlm_domain_lock */
         struct kref dlm_refs;
         enum dlm_ctxt_state dlm_state;
         unsigned int num_joins;