Merge tag 'dlm-6.18' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm

Pull dlm updates from David Teigland:
 "This adds a dlm_release_lockspace() flag to request that node-failure
  recovery be performed for the node leaving the lockspace.

  The implementation of this flag requires coordination with userland
  clustering components. It's been requested for use by GFS2"

* tag 'dlm-6.18' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm:
  dlm: check for undefined release_option values
  dlm: handle release_option as unsigned
  dlm: move to rinfo for all middle conversion cases
  dlm: handle invalid lockspace member remove
  dlm: add new flag DLM_RELEASE_RECOVER for dlm_lockspace_release
  dlm: add new configfs entry release_recover for lockspace members
  dlm: add new RELEASE_RECOVER uevent attribute for release_lockspace
  dlm: use defines for force values in dlm_release_lockspace
  dlm: check for defined force value in dlm_lockspace_release

11 files changed, 153 insertions, 39 deletions

diff --git a/drivers/md/md-cluster.c b/drivers/md/md-cluster.c
index 5497eaee96e7..6e9a0045f0ff 100644
--- a/drivers/md/md-cluster.c
+++ b/drivers/md/md-cluster.c
@@ -979,7 +979,7 @@ err:
 	lockres_free(cinfo->resync_lockres);
 	lockres_free(cinfo->bitmap_lockres);
 	if (cinfo->lockspace)
-		dlm_release_lockspace(cinfo->lockspace, 2);
+		dlm_release_lockspace(cinfo->lockspace, DLM_RELEASE_NORMAL);
 	mddev->cluster_info = NULL;
 	kfree(cinfo);
 	return ret;
@@ -1042,7 +1042,7 @@ static int leave(struct mddev *mddev)
 	lockres_free(cinfo->resync_lockres);
 	lockres_free(cinfo->bitmap_lockres);
 	unlock_all_bitmaps(mddev);
-	dlm_release_lockspace(cinfo->lockspace, 2);
+	dlm_release_lockspace(cinfo->lockspace, DLM_RELEASE_NORMAL);
 	kfree(cinfo);
 	return 0;
 }
diff --git a/fs/dlm/config.c b/fs/dlm/config.c
index a23fd524a6ee..a0d75b5c83c6 100644
--- a/fs/dlm/config.c
+++ b/fs/dlm/config.c
@@ -26,6 +26,7 @@
 /*
  * /config/dlm/<cluster>/spaces/<space>/nodes/<node>/nodeid (refers to <node>)
  * /config/dlm/<cluster>/spaces/<space>/nodes/<node>/weight
+ * /config/dlm/<cluster>/spaces/<space>/nodes/<node>/release_recover
  * /config/dlm/<cluster>/comms/<comm>/nodeid (refers to <comm>)
  * /config/dlm/<cluster>/comms/<comm>/local
  * /config/dlm/<cluster>/comms/<comm>/addr      (write only)
@@ -267,6 +268,7 @@ enum {
 enum {
 	NODE_ATTR_NODEID = 0,
 	NODE_ATTR_WEIGHT,
+	NODE_ATTR_RELEASE_RECOVER,
 };
 
 struct dlm_clusters {
@@ -280,6 +282,8 @@ struct dlm_spaces {
 struct dlm_space {
 	struct config_group group;
 	struct list_head members;
+	struct list_head members_gone;
+	int members_gone_count;
 	struct mutex members_lock;
 	int members_count;
 	struct dlm_nodes *nds;
@@ -310,6 +314,14 @@ struct dlm_node {
 	int weight;
 	int new;
 	int comm_seq; /* copy of cm->seq when nd->nodeid is set */
+	unsigned int release_recover;
+};
+
+struct dlm_member_gone {
+	int nodeid;
+	unsigned int release_recover;
+
+	struct list_head list; /* space->members_gone */
 };
 
 static struct configfs_group_operations clusters_ops = {
@@ -480,6 +492,7 @@ static struct config_group *make_space(struct config_group *g, const char *name)
 	configfs_add_default_group(&nds->ns_group, &sp->group);
 
 	INIT_LIST_HEAD(&sp->members);
+	INIT_LIST_HEAD(&sp->members_gone);
 	mutex_init(&sp->members_lock);
 	sp->members_count = 0;
 	sp->nds = nds;
@@ -587,10 +600,20 @@ static void drop_node(struct config_group *g, struct config_item *i)
 {
 	struct dlm_space *sp = config_item_to_space(g->cg_item.ci_parent);
 	struct dlm_node *nd = config_item_to_node(i);
+	struct dlm_member_gone *mb_gone;
+
+	mb_gone = kzalloc(sizeof(*mb_gone), GFP_KERNEL);
+	if (!mb_gone)
+		return;
 
 	mutex_lock(&sp->members_lock);
 	list_del(&nd->list);
 	sp->members_count--;
+
+	mb_gone->nodeid = nd->nodeid;
+	mb_gone->release_recover = nd->release_recover;
+	list_add(&mb_gone->list, &sp->members_gone);
+	sp->members_gone_count++;
 	mutex_unlock(&sp->members_lock);
 
 	config_item_put(i);
@@ -815,12 +838,34 @@ static ssize_t node_weight_store(struct config_item *item, const char *buf,
 	return len;
 }
 
+static ssize_t node_release_recover_show(struct config_item *item, char *buf)
+{
+	struct dlm_node *n = config_item_to_node(item);
+
+	return sprintf(buf, "%u\n", n->release_recover);
+}
+
+static ssize_t node_release_recover_store(struct config_item *item,
+					  const char *buf, size_t len)
+{
+	struct dlm_node *n = config_item_to_node(item);
+	int rc;
+
+	rc = kstrtouint(buf, 0, &n->release_recover);
+	if (rc)
+		return rc;
+
+	return len;
+}
+
 CONFIGFS_ATTR(node_, nodeid);
 CONFIGFS_ATTR(node_, weight);
+CONFIGFS_ATTR(node_, release_recover);
 
 static struct configfs_attribute *node_attrs[] = {
 	[NODE_ATTR_NODEID] = &node_attr_nodeid,
 	[NODE_ATTR_WEIGHT] = &node_attr_weight,
+	[NODE_ATTR_RELEASE_RECOVER] = &node_attr_release_recover,
 	NULL,
 };
 
@@ -882,9 +927,10 @@ static void put_comm(struct dlm_comm *cm)
 int dlm_config_nodes(char *lsname, struct dlm_config_node **nodes_out,
 		     int *count_out)
 {
+	struct dlm_member_gone *mb_gone, *mb_safe;
+	struct dlm_config_node *nodes, *node;
 	struct dlm_space *sp;
 	struct dlm_node *nd;
-	struct dlm_config_node *nodes, *node;
 	int rv, count;
 
 	sp = get_space(lsname);
@@ -898,7 +944,7 @@ int dlm_config_nodes(char *lsname, struct dlm_config_node **nodes_out,
 		goto out;
 	}
 
-	count = sp->members_count;
+	count = sp->members_count + sp->members_gone_count;
 
 	nodes = kcalloc(count, sizeof(struct dlm_config_node), GFP_NOFS);
 	if (!nodes) {
@@ -917,6 +963,20 @@ int dlm_config_nodes(char *lsname, struct dlm_config_node **nodes_out,
 		nd->new = 0;
 	}
 
+	/* we delay the remove on nodes until here as configfs does
+	 * not support addtional attributes for rmdir().
+	 */
+	list_for_each_entry_safe(mb_gone, mb_safe, &sp->members_gone, list) {
+		node->nodeid = mb_gone->nodeid;
+		node->release_recover = mb_gone->release_recover;
+		node->gone = true;
+		node++;
+
+		list_del(&mb_gone->list);
+		sp->members_gone_count--;
+		kfree(mb_gone);
+	}
+
 	*count_out = count;
 	*nodes_out = nodes;
 	rv = 0;
diff --git a/fs/dlm/config.h b/fs/dlm/config.h
index 13a3d0b26194..4ebd45f75276 100644
--- a/fs/dlm/config.h
+++ b/fs/dlm/config.h
@@ -17,8 +17,10 @@
 struct dlm_config_node {
 	int nodeid;
 	int weight;
+	bool gone;
 	int new;
 	uint32_t comm_seq;
+	unsigned int release_recover;
 };
 
 extern const struct rhashtable_params dlm_rhash_rsb_params;
diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c
index 6dd3a524cd35..be938fdf17d9 100644
--- a/fs/dlm/lock.c
+++ b/fs/dlm/lock.c
@@ -5576,7 +5576,7 @@ static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
 
 	if (rl->rl_status == DLM_LKSTS_CONVERT && middle_conversion(lkb)) {
 		/* We may need to adjust grmode depending on other granted locks. */
-		log_limit(ls, "%s %x middle convert gr %d rq %d remote %d %x",
+		log_rinfo(ls, "%s %x middle convert gr %d rq %d remote %d %x",
 			  __func__, lkb->lkb_id, lkb->lkb_grmode,
 			  lkb->lkb_rqmode, lkb->lkb_nodeid, lkb->lkb_remid);
 		rsb_set_flag(r, RSB_RECOVER_CONVERT);
diff --git a/fs/dlm/lockspace.c b/fs/dlm/lockspace.c
index 1929327ffbe1..ddaa76558706 100644
--- a/fs/dlm/lockspace.c
+++ b/fs/dlm/lockspace.c
@@ -186,12 +186,17 @@ static struct kobj_type dlm_ktype = {
 
 static struct kset *dlm_kset;
 
-static int do_uevent(struct dlm_ls *ls, int in)
+static int do_uevent(struct dlm_ls *ls, int in, unsigned int release_recover)
 {
-	if (in)
+	char message[512] = {};
+	char *envp[] = { message, NULL };
+
+	if (in) {
 		kobject_uevent(&ls->ls_kobj, KOBJ_ONLINE);
-	else
-		kobject_uevent(&ls->ls_kobj, KOBJ_OFFLINE);
+	} else {
+		snprintf(message, 511, "RELEASE_RECOVER=%u", release_recover);
+		kobject_uevent_env(&ls->ls_kobj, KOBJ_OFFLINE, envp);
+	}
 
 	log_rinfo(ls, "%s the lockspace group...", in ? "joining" : "leaving");
 
@@ -575,7 +580,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	   current lockspace members are (via configfs) and then tells the
 	   lockspace to start running (via sysfs) in dlm_ls_start(). */
 
-	error = do_uevent(ls, 1);
+	error = do_uevent(ls, 1, 0);
 	if (error < 0)
 		goto out_recoverd;
 
@@ -592,7 +597,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	return 0;
 
  out_members:
-	do_uevent(ls, 0);
+	do_uevent(ls, 0, 0);
 	dlm_clear_members(ls);
 	kfree(ls->ls_node_array);
  out_recoverd:
@@ -671,19 +676,20 @@ int dlm_new_user_lockspace(const char *name, const char *cluster,
    This is because there may be LKBs queued as ASTs that have been unlinked
    from their RSBs and are pending deletion once the AST has been delivered */
 
-static int lockspace_busy(struct dlm_ls *ls, int force)
+static int lockspace_busy(struct dlm_ls *ls, unsigned int release_option)
 {
 	struct dlm_lkb *lkb;
 	unsigned long id;
 	int rv = 0;
 
 	read_lock_bh(&ls->ls_lkbxa_lock);
-	if (force == 0) {
+	if (release_option == DLM_RELEASE_NO_LOCKS) {
 		xa_for_each(&ls->ls_lkbxa, id, lkb) {
 			rv = 1;
 			break;
 		}
-	} else if (force == 1) {
+	} else if (release_option == DLM_RELEASE_UNUSED) {
+		/* TODO: handle this UNUSED option as NO_LOCKS in later patch */
 		xa_for_each(&ls->ls_lkbxa, id, lkb) {
 			if (lkb->lkb_nodeid == 0 &&
 			    lkb->lkb_grmode != DLM_LOCK_IV) {
@@ -698,11 +704,11 @@ static int lockspace_busy(struct dlm_ls *ls, int force)
 	return rv;
 }
 
-static int release_lockspace(struct dlm_ls *ls, int force)
+static int release_lockspace(struct dlm_ls *ls, unsigned int release_option)
 {
 	int busy, rv;
 
-	busy = lockspace_busy(ls, force);
+	busy = lockspace_busy(ls, release_option);
 
 	spin_lock_bh(&lslist_lock);
 	if (ls->ls_create_count == 1) {
@@ -730,8 +736,9 @@ static int release_lockspace(struct dlm_ls *ls, int force)
 
 	dlm_device_deregister(ls);
 
-	if (force < 3 && dlm_user_daemon_available())
-		do_uevent(ls, 0);
+	if (release_option != DLM_RELEASE_NO_EVENT &&
+	    dlm_user_daemon_available())
+		do_uevent(ls, 0, (release_option == DLM_RELEASE_RECOVER));
 
 	dlm_recoverd_stop(ls);
 
@@ -782,25 +789,24 @@ static int release_lockspace(struct dlm_ls *ls, int force)
  * lockspace must continue to function as usual, participating in recoveries,
  * until this returns.
  *
- * Force has 4 possible values:
- * 0 - don't destroy lockspace if it has any LKBs
- * 1 - destroy lockspace if it has remote LKBs but not if it has local LKBs
- * 2 - destroy lockspace regardless of LKBs
- * 3 - destroy lockspace as part of a forced shutdown
+ * See DLM_RELEASE defines for release_option values and their meaning.
  */
 
-int dlm_release_lockspace(void *lockspace, int force)
+int dlm_release_lockspace(void *lockspace, unsigned int release_option)
 {
 	struct dlm_ls *ls;
 	int error;
 
+	if (release_option > __DLM_RELEASE_MAX)
+		return -EINVAL;
+
 	ls = dlm_find_lockspace_local(lockspace);
 	if (!ls)
 		return -EINVAL;
 	dlm_put_lockspace(ls);
 
 	mutex_lock(&ls_lock);
-	error = release_lockspace(ls, force);
+	error = release_lockspace(ls, release_option);
 	if (!error)
 		ls_count--;
 	if (!ls_count)
diff --git a/fs/dlm/member.c b/fs/dlm/member.c
index b0864c93230f..c0f557a80a75 100644
--- a/fs/dlm/member.c
+++ b/fs/dlm/member.c
@@ -478,7 +478,8 @@ static void dlm_lsop_recover_prep(struct dlm_ls *ls)
 	ls->ls_ops->recover_prep(ls->ls_ops_arg);
 }
 
-static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb)
+static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb,
+				  unsigned int release_recover)
 {
 	struct dlm_slot slot;
 	uint32_t seq;
@@ -495,7 +496,7 @@ static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb)
 
 	error = dlm_comm_seq(memb->nodeid, &seq, false);
 
-	if (!error && seq == memb->comm_seq)
+	if (!release_recover && !error && seq == memb->comm_seq)
 		return;
 
 	slot.nodeid = memb->nodeid;
@@ -552,6 +553,7 @@ int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
 	struct dlm_member *memb, *safe;
 	struct dlm_config_node *node;
 	int i, error, neg = 0, low = -1;
+	unsigned int release_recover;
 
 	/* previously removed members that we've not finished removing need to
 	 * count as a negative change so the "neg" recovery steps will happen
@@ -569,11 +571,21 @@ int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
 
 	list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
 		node = find_config_node(rv, memb->nodeid);
-		if (node && !node->new)
+		if (!node) {
+			log_error(ls, "remove member %d invalid",
+				  memb->nodeid);
+			return -EFAULT;
+		}
+
+		if (!node->new && !node->gone)
 			continue;
 
-		if (!node) {
-			log_rinfo(ls, "remove member %d", memb->nodeid);
+		release_recover = 0;
+
+		if (node->gone) {
+			release_recover = node->release_recover;
+			log_rinfo(ls, "remove member %d%s", memb->nodeid,
+				  release_recover ? " (release_recover)" : "");
 		} else {
 			/* removed and re-added */
 			log_rinfo(ls, "remove member %d comm_seq %u %u",
@@ -584,13 +596,16 @@ int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
 		list_move(&memb->list, &ls->ls_nodes_gone);
 		remove_remote_member(memb->nodeid);
 		ls->ls_num_nodes--;
-		dlm_lsop_recover_slot(ls, memb);
+		dlm_lsop_recover_slot(ls, memb, release_recover);
 	}
 
 	/* add new members to ls_nodes */
 
 	for (i = 0; i < rv->nodes_count; i++) {
 		node = &rv->nodes[i];
+		if (node->gone)
+			continue;
+
 		if (dlm_is_member(ls, node->nodeid))
 			continue;
 		error = dlm_add_member(ls, node);
diff --git a/fs/dlm/recover.c b/fs/dlm/recover.c
index be4240f09abd..3ac020fb8139 100644
--- a/fs/dlm/recover.c
+++ b/fs/dlm/recover.c
@@ -842,7 +842,7 @@ static void recover_conversion(struct dlm_rsb *r)
 		 */
 		if (((lkb->lkb_grmode == DLM_LOCK_PR) && (other_grmode == DLM_LOCK_CW)) ||
 		    ((lkb->lkb_grmode == DLM_LOCK_CW) && (other_grmode == DLM_LOCK_PR))) {
-			log_limit(ls, "%s %x gr %d rq %d, remote %d %x, other_lkid %u, other gr %d, set gr=NL",
+			log_rinfo(ls, "%s %x gr %d rq %d, remote %d %x, other_lkid %u, other gr %d, set gr=NL",
 				  __func__, lkb->lkb_id, lkb->lkb_grmode,
 				  lkb->lkb_rqmode, lkb->lkb_nodeid,
 				  lkb->lkb_remid, other_lkid, other_grmode);
diff --git a/fs/dlm/user.c b/fs/dlm/user.c
index 5cb3896be826..51daf4acbe31 100644
--- a/fs/dlm/user.c
+++ b/fs/dlm/user.c
@@ -425,7 +425,7 @@ static int device_create_lockspace(struct dlm_lspace_params *params)
 	dlm_put_lockspace(ls);
 
 	if (error)
-		dlm_release_lockspace(lockspace, 0);
+		dlm_release_lockspace(lockspace, DLM_RELEASE_NO_LOCKS);
 	else
 		error = ls->ls_device.minor;
 
@@ -436,7 +436,7 @@ static int device_remove_lockspace(struct dlm_lspace_params *params)
 {
 	dlm_lockspace_t *lockspace;
 	struct dlm_ls *ls;
-	int error, force = 0;
+	int error, force = DLM_RELEASE_NO_LOCKS;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
@@ -446,7 +446,7 @@ static int device_remove_lockspace(struct dlm_lspace_params *params)
 		return -ENOENT;
 
 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
-		force = 2;
+		force = DLM_RELEASE_NORMAL;
 
 	lockspace = ls;
 	dlm_put_lockspace(ls);
diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
index ae058b1f9435..4f00af7dd256 100644
--- a/fs/gfs2/lock_dlm.c
+++ b/fs/gfs2/lock_dlm.c
@@ -1418,7 +1418,7 @@ static int gdlm_mount(struct gfs2_sbd *sdp, const char *table)
 	return 0;
 
 fail_release:
-	dlm_release_lockspace(ls->ls_dlm, 2);
+	dlm_release_lockspace(ls->ls_dlm, DLM_RELEASE_NORMAL);
 fail_free:
 	free_recover_size(ls);
 fail:
@@ -1456,7 +1456,7 @@ static void gdlm_unmount(struct gfs2_sbd *sdp)
 release:
 	down_write(&ls->ls_sem);
 	if (ls->ls_dlm) {
-		dlm_release_lockspace(ls->ls_dlm, 2);
+		dlm_release_lockspace(ls->ls_dlm, DLM_RELEASE_NORMAL);
 		ls->ls_dlm = NULL;
 	}
 	up_write(&ls->ls_sem);
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
index 0f045e45fa0c..765105f1ff8a 100644
--- a/fs/ocfs2/stack_user.c
+++ b/fs/ocfs2/stack_user.c
@@ -952,7 +952,7 @@ static const struct dlm_lockspace_ops ocfs2_ls_ops = {
 static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 {
 	version_unlock(conn);
-	dlm_release_lockspace(conn->cc_lockspace, 2);
+	dlm_release_lockspace(conn->cc_lockspace, DLM_RELEASE_NORMAL);
 	conn->cc_lockspace = NULL;
 	ocfs2_live_connection_drop(conn->cc_private);
 	conn->cc_private = NULL;
diff --git a/include/linux/dlm.h b/include/linux/dlm.h
index bacda9898f2b..7e7b45b0d097 100644
--- a/include/linux/dlm.h
+++ b/include/linux/dlm.h
@@ -88,12 +88,43 @@ int dlm_new_lockspace(const char *name, const char *cluster,
 		      int *ops_result, dlm_lockspace_t **lockspace);
 
 /*
+ * dlm_release_lockspace() release_option values:
+ *
+ * DLM_RELEASE_NO_LOCKS returns -EBUSY if any locks (lkb's)
+ *   exist in the local lockspace.
+ *
+ * DLM_RELEASE_UNUSED previous value that is no longer used.
+ *
+ * DLM_RELEASE_NORMAL releases the lockspace regardless of any
+ *   locks managed in the local lockspace.
+ *
+ * DLM_RELEASE_NO_EVENT release the lockspace regardless of any
+ *   locks managed in the local lockspace, and does not submit
+ *   a leave event to the cluster manager, so other nodes will
+ *   not be notified that the node should be removed from the
+ *   list of lockspace members.
+ *
+ * DLM_RELEASE_RECOVER like DLM_RELEASE_NORMAL, but the remaining
+ *   nodes will handle the removal of the node as if the node
+ *   had failed, e.g. the recover_slot() callback would be used.
+ */
+#define DLM_RELEASE_NO_LOCKS		0
+#define DLM_RELEASE_UNUSED		1
+#define DLM_RELEASE_NORMAL		2
+#define DLM_RELEASE_NO_EVENT		3
+#define DLM_RELEASE_RECOVER		4
+#define __DLM_RELEASE_MAX		DLM_RELEASE_RECOVER
+
+/*
  * dlm_release_lockspace
  *
  * Stop a lockspace.
+ *
+ * release_option: see DLM_RELEASE values above.
  */
 
-int dlm_release_lockspace(dlm_lockspace_t *lockspace, int force);
+int dlm_release_lockspace(dlm_lockspace_t *lockspace,
+			  unsigned int release_option);
 
 /*
  * dlm_lock