Ocfs2 devel - Dec 2008 - [PATCH] ocfs2/dlm: fix lockres mastery race, v2

dlm_get_lock_resource() is supposed to return a lock resource with a proper
master. If multiple
concurrent threads attempt to lookup the lockres for the same lockid, one or
more threads are likely
to return a lockres without a proper master, if the lock mastery is underway.

This patch makes the threads wait in dlm_get_lock_resource() while the mastery
is underway, ensuring
all threads return the lockres with a proper master.

This issue is known limited to users using the flock() syscall. For all other fs
operations, dlmglue
should ensure that only one thread is performing dlm operations for a given
lockid at any one time.

Thank Sunil for his review and comments.

Signed-off-by: Coly Li <coyli at suse.de>
Cc: Sunil Mushran <sunil.mushran at oracle.com>
Cc: Mark Fasheh <mfasheh at suse.com>
Cc: Jeff Mahoney <jeffm at suse.com>
---
 fs/ocfs2/dlm/dlmmaster.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 44f87ca..dd1e754 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -742,6 +742,17 @@ lookup:
 			goto lookup;
 		}

+		/* wait for lock resource is being mastered by another thread */
+		spin_lock(&tmpres->spinlock);
+		if (tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
+			__dlm_wait_on_lockres_flags(tmpres, DLM_LOCK_RES_IN_PROGRESS);
+			spin_unlock(&tmpres->spinlock);
+			dlm_lockres_put(tmpres);
+			tmpres = NULL;
+			goto lookup;
+		}
+		spin_unlock(&tmpres->spinlock);
+
 		mlog(0, "found in hash!\n");
 		if (res)
 			dlm_lockres_put(res);
-- 
Coly Li
SuSE PRC Labs

nak

The reason is listed in yesterday's email.

Coly Li wrote:
> dlm_get_lock_resource() is supposed to return a lock resource with a proper
master. If multiple
> concurrent threads attempt to lookup the lockres for the same lockid, one
or more threads are likely
> to return a lockres without a proper master, if the lock mastery is
underway.
>
> This patch makes the threads wait in dlm_get_lock_resource() while the
mastery is underway, ensuring
> all threads return the lockres with a proper master.
>
> This issue is known limited to users using the flock() syscall. For all
other fs operations, dlmglue
> should ensure that only one thread is performing dlm operations for a given
lockid at any one time.
>
> Thank Sunil for his review and comments.
>
> Signed-off-by: Coly Li <coyli at suse.de>
> Cc: Sunil Mushran <sunil.mushran at oracle.com>
> Cc: Mark Fasheh <mfasheh at suse.com>
> Cc: Jeff Mahoney <jeffm at suse.com>
> ---
>  fs/ocfs2/dlm/dlmmaster.c |   11 +++++++++++
>  1 files changed, 11 insertions(+), 0 deletions(-)
>
> diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
> index 44f87ca..dd1e754 100644
> --- a/fs/ocfs2/dlm/dlmmaster.c
> +++ b/fs/ocfs2/dlm/dlmmaster.c
> @@ -742,6 +742,17 @@ lookup:
>  			goto lookup;
>  		}
>
> +		/* wait for lock resource is being mastered by another thread */
> +		spin_lock(&tmpres->spinlock);
> +		if (tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
> +			__dlm_wait_on_lockres_flags(tmpres, DLM_LOCK_RES_IN_PROGRESS);
> +			spin_unlock(&tmpres->spinlock);
> +			dlm_lockres_put(tmpres);
> +			tmpres = NULL;
> +			goto lookup;
> +		}
> +		spin_unlock(&tmpres->spinlock);
> +
>  		mlog(0, "found in hash!\n");
>  		if (res)
>  			dlm_lockres_put(res);
>