ceph: map snapid to anonymous bdev ID

ceph_getattr() return zero dev ID for head inodes and set dev ID to
snapid directly for snaphost inodes. This is not good because userspace
utilities may consider device ID of 0 as invalid, snapid may conflict
with other device's ID.

This patch introduces "snapids to anonymous bdev IDs" map. we create a
new mapping when we see a snapid for the first time. we trim unused
mapping after it is ilde for 5 minutes.

Link: http://tracker.ceph.com/issues/22353
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

1 files changed, 154 insertions, 2 deletions

diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
index dfc25ceeffed..89aa37fa0f84 100644
--- a/fs/ceph/snap.c
+++ b/fs/ceph/snap.c
@@ -3,12 +3,13 @@
 
 #include <linux/sort.h>
 #include <linux/slab.h>
-
 #include "super.h"
 #include "mds_client.h"
-
 #include <linux/ceph/decode.h>
 
+/* unused map expires after 5 minutes */
+#define CEPH_SNAPID_MAP_TIMEOUT	(5 * 60 * HZ)
+
 /*
  * Snapshots in ceph are driven in large part by cooperation from the
  * client.  In contrast to local file systems or file servers that
@@ -989,3 +990,154 @@ out:
 		up_write(&mdsc->snap_rwsem);
 	return;
 }
+
+struct ceph_snapid_map* ceph_get_snapid_map(struct ceph_mds_client *mdsc,
+					    u64 snap)
+{
+	struct ceph_snapid_map *sm, *exist;
+	struct rb_node **p, *parent;
+	int ret;
+
+	exist = NULL;
+	spin_lock(&mdsc->snapid_map_lock);
+	p = &mdsc->snapid_map_tree.rb_node;
+	while (*p) {
+		exist = rb_entry(*p, struct ceph_snapid_map, node);
+		if (snap > exist->snap) {
+			p = &(*p)->rb_left;
+		} else if (snap < exist->snap) {
+			p = &(*p)->rb_right;
+		} else {
+			if (atomic_inc_return(&exist->ref) == 1)
+				list_del_init(&exist->lru);
+			break;
+		}
+		exist = NULL;
+	}
+	spin_unlock(&mdsc->snapid_map_lock);
+	if (exist) {
+		dout("found snapid map %llx -> %x\n", exist->snap, exist->dev);
+		return exist;
+	}
+
+	sm = kmalloc(sizeof(*sm), GFP_NOFS);
+	if (!sm)
+		return NULL;
+
+	ret = get_anon_bdev(&sm->dev);
+	if (ret < 0) {
+		kfree(sm);
+		return NULL;
+	}
+
+	INIT_LIST_HEAD(&sm->lru);
+	atomic_set(&sm->ref, 1);
+	sm->snap = snap;
+
+	exist = NULL;
+	parent = NULL;
+	p = &mdsc->snapid_map_tree.rb_node;
+	spin_lock(&mdsc->snapid_map_lock);
+	while (*p) {
+		parent = *p;
+		exist = rb_entry(*p, struct ceph_snapid_map, node);
+		if (snap > exist->snap)
+			p = &(*p)->rb_left;
+		else if (snap < exist->snap)
+			p = &(*p)->rb_right;
+		else
+			break;
+		exist = NULL;
+	}
+	if (exist) {
+		if (atomic_inc_return(&exist->ref) == 1)
+			list_del_init(&exist->lru);
+	} else {
+		rb_link_node(&sm->node, parent, p);
+		rb_insert_color(&sm->node, &mdsc->snapid_map_tree);
+	}
+	spin_unlock(&mdsc->snapid_map_lock);
+	if (exist) {
+		free_anon_bdev(sm->dev);
+		kfree(sm);
+		dout("found snapid map %llx -> %x\n", exist->snap, exist->dev);
+		return exist;
+	}
+
+	dout("create snapid map %llx -> %x\n", sm->snap, sm->dev);
+	return sm;
+}
+
+void ceph_put_snapid_map(struct ceph_mds_client* mdsc,
+			 struct ceph_snapid_map *sm)
+{
+	if (!sm)
+		return;
+	if (atomic_dec_and_lock(&sm->ref, &mdsc->snapid_map_lock)) {
+		if (!RB_EMPTY_NODE(&sm->node)) {
+			sm->last_used = jiffies;
+			list_add_tail(&sm->lru, &mdsc->snapid_map_lru);
+			spin_unlock(&mdsc->snapid_map_lock);
+		} else {
+			/* already cleaned up by
+			 * ceph_cleanup_snapid_map() */
+			spin_unlock(&mdsc->snapid_map_lock);
+			kfree(sm);
+		}
+	}
+}
+
+void ceph_trim_snapid_map(struct ceph_mds_client *mdsc)
+{
+	struct ceph_snapid_map *sm;
+	unsigned long now;
+	LIST_HEAD(to_free);
+
+	spin_lock(&mdsc->snapid_map_lock);
+	now = jiffies;
+
+	while (!list_empty(&mdsc->snapid_map_lru)) {
+		sm = list_first_entry(&mdsc->snapid_map_lru,
+				      struct ceph_snapid_map, lru);
+		if (time_after(sm->last_used + CEPH_SNAPID_MAP_TIMEOUT, now))
+			break;
+
+		rb_erase(&sm->node, &mdsc->snapid_map_tree);
+		list_move(&sm->lru, &to_free);
+	}
+	spin_unlock(&mdsc->snapid_map_lock);
+
+	while (!list_empty(&to_free)) {
+		sm = list_first_entry(&to_free, struct ceph_snapid_map, lru);
+		list_del(&sm->lru);
+		dout("trim snapid map %llx -> %x\n", sm->snap, sm->dev);
+		free_anon_bdev(sm->dev);
+		kfree(sm);
+	}
+}
+
+void ceph_cleanup_snapid_map(struct ceph_mds_client *mdsc)
+{
+	struct ceph_snapid_map *sm;
+	struct rb_node *p;
+	LIST_HEAD(to_free);
+
+	spin_lock(&mdsc->snapid_map_lock);
+	while ((p = rb_first(&mdsc->snapid_map_tree))) {
+		sm = rb_entry(p, struct ceph_snapid_map, node);
+		rb_erase(p, &mdsc->snapid_map_tree);
+		RB_CLEAR_NODE(p);
+		list_move(&sm->lru, &to_free);
+	}
+	spin_unlock(&mdsc->snapid_map_lock);
+
+	while (!list_empty(&to_free)) {
+		sm = list_first_entry(&to_free, struct ceph_snapid_map, lru);
+		list_del(&sm->lru);
+		free_anon_bdev(sm->dev);
+		if (WARN_ON_ONCE(atomic_read(&sm->ref))) {
+			pr_err("snapid map %llx -> %x still in use\n",
+			       sm->snap, sm->dev);
+		}
+	}
+}