[DLM] fix leaking user locks

User NOQUEUE lock requests to a remote node that failed with -EAGAIN were
never being removed from a process's list of locks.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c
index 82e1ac251843ef09c6253a9a0ad873d30b3ee0a3..227443218167226763cd6e28dc67bf763a9a2e3e 100644 (file)
--- a/fs/dlm/lock.c
+++ b/fs/dlm/lock.c
@@ -526,6 +526,7 @@ static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
        lkb->lkb_nodeid = -1;
        lkb->lkb_grmode = DLM_LOCK_IV;
        kref_init(&lkb->lkb_ref);
+       INIT_LIST_HEAD(&lkb->lkb_ownqueue);
 
        get_random_bytes(&bucket, sizeof(bucket));
        bucket &= (ls->ls_lkbtbl_size - 1);
@@ -3705,7 +3706,7 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
                goto out_put;
 
        spin_lock(&ua->proc->locks_spin);
-       list_del(&lkb->lkb_ownqueue);
+       list_del_init(&lkb->lkb_ownqueue);
        spin_unlock(&ua->proc->locks_spin);
 
        /* this removes the reference for the proc->locks list added by
@@ -3749,7 +3750,7 @@ int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
        /* this lkb was removed from the WAITING queue */
        if (lkb->lkb_grmode == DLM_LOCK_IV) {
                spin_lock(&ua->proc->locks_spin);
-               list_del(&lkb->lkb_ownqueue);
+               list_del_init(&lkb->lkb_ownqueue);
                spin_unlock(&ua->proc->locks_spin);
                unhold_lkb(lkb);
        }
@@ -3817,7 +3818,7 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
                        unhold_lkb(lkb);
                }
 
-               list_del(&lkb->lkb_ownqueue);
+               list_del_init(&lkb->lkb_ownqueue);
 
                if (lkb->lkb_exflags & DLM_LKF_PERSISTENT) {
                        lkb->lkb_flags |= DLM_IFL_ORPHAN;

diff --git a/fs/dlm/user.c b/fs/dlm/user.c
index 1f05960a916f387e3fa835f3424a8c269ca71178..fd19caf9af973b888750b0ca0dc7080c232974ab 100644 (file)
--- a/fs/dlm/user.c
+++ b/fs/dlm/user.c
@@ -133,6 +133,7 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
        struct dlm_ls *ls;
        struct dlm_user_args *ua;
        struct dlm_user_proc *proc;
+       int remove_ownqueue = 0;
 
        /* dlm_clear_proc_locks() sets ORPHAN/DEAD flag on each
           lkb before dealing with it.  We need to check this
@@ -171,6 +172,14 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
                wake_up_interruptible(&proc->wait);
        }
 
+       /* noqueue requests that fail may need to be removed from the
+          proc's locks list, there should be a better way of detecting
+          this situation than checking all these things... */
+          
+       if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV &&
+           ua->lksb.sb_status == -EAGAIN && !list_empty(&lkb->lkb_ownqueue))
+               remove_ownqueue = 1;
+
        /* We want to copy the lvb to userspace when the completion
           ast is read if the status is 0, the lock has an lvb and
           lvb_ops says we should.  We could probably have set_lvb_lock()
@@ -185,6 +194,13 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
                ua->update_user_lvb = 0;
 
        spin_unlock(&proc->asts_spin);
+
+       if (remove_ownqueue) {
+               spin_lock(&ua->proc->locks_spin);
+               list_del_init(&lkb->lkb_ownqueue);
+               spin_unlock(&ua->proc->locks_spin);
+               dlm_put_lkb(lkb);
+       }
  out:
        mutex_unlock(&ls->ls_clear_proc_locks);
 }