[DLM] fix requestqueue race

Red Hat BZ 211914

There's a race between dlm_recoverd (1) enabling locking and (2) clearing
out the requestqueue, and dlm_recvd (1) checking if locking is enabled and
(2) adding a message to the requestqueue.  An order of recoverd(1),
recvd(1), recvd(2), recoverd(2) will result in a message being left on the
requestqueue.  The fix is to have dlm_recvd check if dlm_recoverd has
enabled locking after taking the mutex for the requestqueue and if it has
processing the message instead of queueing it.

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

3 files changed, 29 insertions, 9 deletions

diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c
index 3f2befa..6088a16 100644
--- a/fs/dlm/lock.c
+++ b/fs/dlm/lock.c
@@ -3028,10 +3028,17 @@ int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery)
 
 	while (1) {
 		if (dlm_locking_stopped(ls)) {
-			if (!recovery)
-				dlm_add_requestqueue(ls, nodeid, hd);
-			error = -EINTR;
-			goto out;
+			if (recovery) {
+				error = -EINTR;
+				goto out;
+			}
+			error = dlm_add_requestqueue(ls, nodeid, hd);
+			if (error == -EAGAIN)
+				continue;
+			else {
+				error = -EINTR;
+				goto out;
+			}
 		}
 
 		if (lock_recovery_try(ls))
diff --git a/fs/dlm/requestqueue.c b/fs/dlm/requestqueue.c
index 7b2b089..0226d2a 100644
--- a/fs/dlm/requestqueue.c
+++ b/fs/dlm/requestqueue.c
@@ -30,26 +30,39 @@ struct rq_entry {
  * lockspace is enabled on some while still suspended on others.
  */
 
-void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd)
+int dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd)
 {
 	struct rq_entry *e;
 	int length = hd->h_length;
+	int rv = 0;
 
 	if (dlm_is_removed(ls, nodeid))
-		return;
+		return 0;
 
 	e = kmalloc(sizeof(struct rq_entry) + length, GFP_KERNEL);
 	if (!e) {
 		log_print("dlm_add_requestqueue: out of memory\n");
-		return;
+		return 0;
 	}
 
 	e->nodeid = nodeid;
 	memcpy(e->request, hd, length);
 
+	/* We need to check dlm_locking_stopped() after taking the mutex to
+	   avoid a race where dlm_recoverd enables locking and runs
+	   process_requestqueue between our earlier dlm_locking_stopped check
+	   and this addition to the requestqueue. */
+
 	mutex_lock(&ls->ls_requestqueue_mutex);
-	list_add_tail(&e->list, &ls->ls_requestqueue);
+	if (dlm_locking_stopped(ls))
+		list_add_tail(&e->list, &ls->ls_requestqueue);
+	else {
+		log_debug(ls, "dlm_add_requestqueue skip from %d", nodeid);
+		kfree(e);
+		rv = -EAGAIN;
+	}
 	mutex_unlock(&ls->ls_requestqueue_mutex);
+	return rv;
 }
 
 int dlm_process_requestqueue(struct dlm_ls *ls)
diff --git a/fs/dlm/requestqueue.h b/fs/dlm/requestqueue.h
index 349f0d2..6a53ea0 100644
--- a/fs/dlm/requestqueue.h
+++ b/fs/dlm/requestqueue.h
@@ -13,7 +13,7 @@
 #ifndef __REQUESTQUEUE_DOT_H__
 #define __REQUESTQUEUE_DOT_H__
 
-void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd);
+int dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd);
 int dlm_process_requestqueue(struct dlm_ls *ls);
 void dlm_wait_requestqueue(struct dlm_ls *ls);
 void dlm_purge_requestqueue(struct dlm_ls *ls);