ceph: use list instead of rbtree to track cap flushes

We don't have requirement of searching cap flush by TID. In most cases,
we just need to know TID of the oldest cap flush. List is ideal for this
usage.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

fs/ceph/caps.c

@@ -1413,52 +1413,6 @@ int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
 	return dirty;
 }
 
-static void __add_cap_flushing_to_inode(struct ceph_inode_info *ci,
-					struct ceph_cap_flush *cf)
-{
-	struct rb_node **p = &ci->i_cap_flush_tree.rb_node;
-	struct rb_node *parent = NULL;
-	struct ceph_cap_flush *other = NULL;
-
-	while (*p) {
-		parent = *p;
-		other = rb_entry(parent, struct ceph_cap_flush, i_node);
-
-		if (cf->tid < other->tid)
-			p = &(*p)->rb_left;
-		else if (cf->tid > other->tid)
-			p = &(*p)->rb_right;
-		else
-			BUG();
-	}
-
-	rb_link_node(&cf->i_node, parent, p);
-	rb_insert_color(&cf->i_node, &ci->i_cap_flush_tree);
-}
-
-static void __add_cap_flushing_to_mdsc(struct ceph_mds_client *mdsc,
-				       struct ceph_cap_flush *cf)
-{
-	struct rb_node **p = &mdsc->cap_flush_tree.rb_node;
-	struct rb_node *parent = NULL;
-	struct ceph_cap_flush *other = NULL;
-
-	while (*p) {
-		parent = *p;
-		other = rb_entry(parent, struct ceph_cap_flush, g_node);
-
-		if (cf->tid < other->tid)
-			p = &(*p)->rb_left;
-		else if (cf->tid > other->tid)
-			p = &(*p)->rb_right;
-		else
-			BUG();
-	}
-
-	rb_link_node(&cf->g_node, parent, p);
-	rb_insert_color(&cf->g_node, &mdsc->cap_flush_tree);
-}
-
 struct ceph_cap_flush *ceph_alloc_cap_flush(void)
 {
 	return kmem_cache_alloc(ceph_cap_flush_cachep, GFP_KERNEL);
@@ -1472,10 +1426,10 @@ void ceph_free_cap_flush(struct ceph_cap_flush *cf)
 
 static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc)
 {
-	struct rb_node *n = rb_first(&mdsc->cap_flush_tree);
-	if (n) {
+	if (!list_empty(&mdsc->cap_flush_list)) {
 		struct ceph_cap_flush *cf =
-			rb_entry(n, struct ceph_cap_flush, g_node);
+			list_first_entry(&mdsc->cap_flush_list,
+					 struct ceph_cap_flush, g_list);
 		return cf->tid;
 	}
 	return 0;
@@ -1516,7 +1470,7 @@ static int __mark_caps_flushing(struct inode *inode,
 	list_del_init(&ci->i_dirty_item);
 
 	cf->tid = ++mdsc->last_cap_flush_tid;
-	__add_cap_flushing_to_mdsc(mdsc, cf);
+	list_add_tail(&cf->g_list, &mdsc->cap_flush_list);
 	*oldest_flush_tid = __get_oldest_flush_tid(mdsc);
 
 	if (list_empty(&ci->i_flushing_item)) {
@@ -1530,7 +1484,7 @@ static int __mark_caps_flushing(struct inode *inode,
 	}
 	spin_unlock(&mdsc->cap_dirty_lock);
 
-	__add_cap_flushing_to_inode(ci, cf);
+	list_add_tail(&cf->i_list, &ci->i_cap_flush_list);
 
 	*flush_tid = cf->tid;
 	return flushing;
@@ -1890,10 +1844,10 @@ retry:
 			spin_unlock(&ci->i_ceph_lock);
 		}
 	} else {
-		struct rb_node *n = rb_last(&ci->i_cap_flush_tree);
-		if (n) {
+		if (!list_empty(&ci->i_cap_flush_list)) {
 			struct ceph_cap_flush *cf =
-				rb_entry(n, struct ceph_cap_flush, i_node);
+				list_last_entry(&ci->i_cap_flush_list,
+						 struct ceph_cap_flush, i_list);
 			flush_tid = cf->tid;
 		}
 		flushing = ci->i_flushing_caps;
@@ -1913,14 +1867,13 @@ out:
 static int caps_are_flushed(struct inode *inode, u64 flush_tid)
 {
 	struct ceph_inode_info *ci = ceph_inode(inode);
-	struct ceph_cap_flush *cf;
-	struct rb_node *n;
 	int ret = 1;
 
 	spin_lock(&ci->i_ceph_lock);
-	n = rb_first(&ci->i_cap_flush_tree);
-	if (n) {
-		cf = rb_entry(n, struct ceph_cap_flush, i_node);
+	if (!list_empty(&ci->i_cap_flush_list)) {
+		struct ceph_cap_flush * cf =
+			list_first_entry(&ci->i_cap_flush_list,
+					 struct ceph_cap_flush, i_list);
 		if (cf->tid <= flush_tid)
 			ret = 0;
 	}
@@ -2083,7 +2036,6 @@ static int __kick_flushing_caps(struct ceph_mds_client *mdsc,
 	struct inode *inode = &ci->vfs_inode;
 	struct ceph_cap *cap;
 	struct ceph_cap_flush *cf;
-	struct rb_node *n;
 	int delayed = 0;
 	u64 first_tid = 0;
 	u64 oldest_flush_tid;
@@ -2092,8 +2044,11 @@ static int __kick_flushing_caps(struct ceph_mds_client *mdsc,
 	oldest_flush_tid = __get_oldest_flush_tid(mdsc);
 	spin_unlock(&mdsc->cap_dirty_lock);
 
-	while (true) {
-		spin_lock(&ci->i_ceph_lock);
+	spin_lock(&ci->i_ceph_lock);
+	list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
+		if (cf->tid < first_tid)
+			continue;
+
 		cap = ci->i_auth_cap;
 		if (!(cap && cap->session == session)) {
 			pr_err("%p auth cap %p not mds%d ???\n", inode,
@@ -2102,18 +2057,6 @@ static int __kick_flushing_caps(struct ceph_mds_client *mdsc,
 			break;
 		}
 
-		for (n = rb_first(&ci->i_cap_flush_tree); n; n = rb_next(n)) {
-			cf = rb_entry(n, struct ceph_cap_flush, i_node);
-			if (cf->tid >= first_tid)
-				break;
-		}
-		if (!n) {
-			spin_unlock(&ci->i_ceph_lock);
-			break;
-		}
-
-		cf = rb_entry(n, struct ceph_cap_flush, i_node);
-
 		first_tid = cf->tid + 1;
 
 		dout("kick_flushing_caps %p cap %p tid %llu %s\n", inode,
@@ -2123,7 +2066,10 @@ static int __kick_flushing_caps(struct ceph_mds_client *mdsc,
 				      __ceph_caps_wanted(ci),
 				      cap->issued | cap->implemented,
 				      cf->caps, cf->tid, oldest_flush_tid);
+
+		spin_lock(&ci->i_ceph_lock);
 	}
+	spin_unlock(&ci->i_ceph_lock);
 	return delayed;
 }
 
@@ -2995,23 +2941,19 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
 {
 	struct ceph_inode_info *ci = ceph_inode(inode);
 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
-	struct ceph_cap_flush *cf;
-	struct rb_node *n;
+	struct ceph_cap_flush *cf, *tmp_cf;
 	LIST_HEAD(to_remove);
 	unsigned seq = le32_to_cpu(m->seq);
 	int dirty = le32_to_cpu(m->dirty);
 	int cleaned = 0;
 	int drop = 0;
 
-	n = rb_first(&ci->i_cap_flush_tree);
-	while (n) {
-		cf = rb_entry(n, struct ceph_cap_flush, i_node);
-		n = rb_next(&cf->i_node);
+	list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) {
 		if (cf->tid == flush_tid)
 			cleaned = cf->caps;
 		if (cf->tid <= flush_tid) {
-			rb_erase(&cf->i_node, &ci->i_cap_flush_tree);
-			list_add_tail(&cf->list, &to_remove);
+			list_del(&cf->i_list);
+			list_add_tail(&cf->i_list, &to_remove);
 		} else {
 			cleaned &= ~cf->caps;
 			if (!cleaned)
@@ -3033,12 +2975,12 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
 	spin_lock(&mdsc->cap_dirty_lock);
 
 	if (!list_empty(&to_remove)) {
-		list_for_each_entry(cf, &to_remove, list)
-			rb_erase(&cf->g_node, &mdsc->cap_flush_tree);
+		u64 oldest_flush_tid;
+		list_for_each_entry(cf, &to_remove, i_list)
+			list_del(&cf->g_list);
 
-		n = rb_first(&mdsc->cap_flush_tree);
-		cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL;
-		if (!cf || cf->tid > flush_tid)
+		oldest_flush_tid = __get_oldest_flush_tid(mdsc);
+		if (oldest_flush_tid == 0 || oldest_flush_tid > flush_tid)
 			wake_up_all(&mdsc->cap_flushing_wq);
 	}
 
@@ -3075,8 +3017,8 @@ out:
 
 	while (!list_empty(&to_remove)) {
 		cf = list_first_entry(&to_remove,
-				      struct ceph_cap_flush, list);
-		list_del(&cf->list);
+				      struct ceph_cap_flush, i_list);
+		list_del(&cf->i_list);
 		ceph_free_cap_flush(cf);
 	}
 	if (drop)

fs/ceph/inode.c

@@ -468,7 +468,7 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	INIT_LIST_HEAD(&ci->i_dirty_item);
 	INIT_LIST_HEAD(&ci->i_flushing_item);
 	ci->i_prealloc_cap_flush = NULL;
-	ci->i_cap_flush_tree = RB_ROOT;
+	INIT_LIST_HEAD(&ci->i_cap_flush_list);
 	init_waitqueue_head(&ci->i_cap_wq);
 	ci->i_hold_caps_min = 0;
 	ci->i_hold_caps_max = 0;

fs/ceph/mds_client.c

@@ -1148,19 +1148,17 @@ static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
 		    ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
 			invalidate = true;
 
-		while (true) {
-			struct rb_node *n = rb_first(&ci->i_cap_flush_tree);
-			if (!n)
-				break;
-			cf = rb_entry(n, struct ceph_cap_flush, i_node);
-			rb_erase(&cf->i_node, &ci->i_cap_flush_tree);
-			list_add(&cf->list, &to_remove);
+		while (!list_empty(&ci->i_cap_flush_list)) {
+			cf = list_first_entry(&ci->i_cap_flush_list,
+					      struct ceph_cap_flush, i_list);
+			list_del(&cf->i_list);
+			list_add(&cf->i_list, &to_remove);
 		}
 
 		spin_lock(&mdsc->cap_dirty_lock);
 
-		list_for_each_entry(cf, &to_remove, list)
-			rb_erase(&cf->g_node, &mdsc->cap_flush_tree);
+		list_for_each_entry(cf, &to_remove, i_list)
+			list_del(&cf->g_list);
 
 		if (!list_empty(&ci->i_dirty_item)) {
 			pr_warn_ratelimited(
@@ -1184,7 +1182,7 @@ static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
 		spin_unlock(&mdsc->cap_dirty_lock);
 
 		if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
-			list_add(&ci->i_prealloc_cap_flush->list, &to_remove);
+			list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
 			ci->i_prealloc_cap_flush = NULL;
 		}
 	}
@@ -1192,8 +1190,8 @@ static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
 	while (!list_empty(&to_remove)) {
 		struct ceph_cap_flush *cf;
 		cf = list_first_entry(&to_remove,
-				      struct ceph_cap_flush, list);
-		list_del(&cf->list);
+				      struct ceph_cap_flush, i_list);
+		list_del(&cf->i_list);
 		ceph_free_cap_flush(cf);
 	}
 
@@ -1499,17 +1497,18 @@ static int check_capsnap_flush(struct ceph_inode_info *ci,
 static int check_caps_flush(struct ceph_mds_client *mdsc,
 			    u64 want_flush_tid)
 {
-	struct rb_node *n;
-	struct ceph_cap_flush *cf;
 	int ret = 1;
 
 	spin_lock(&mdsc->cap_dirty_lock);
-	n = rb_first(&mdsc->cap_flush_tree);
-	cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL;
-	if (cf && cf->tid <= want_flush_tid) {
-		dout("check_caps_flush still flushing tid %llu <= %llu\n",
-		     cf->tid, want_flush_tid);
-		ret = 0;
+	if (!list_empty(&mdsc->cap_flush_list)) {
+		struct ceph_cap_flush *cf =
+			list_first_entry(&mdsc->cap_flush_list,
+					 struct ceph_cap_flush, g_list);
+		if (cf->tid <= want_flush_tid) {
+			dout("check_caps_flush still flushing tid "
+			     "%llu <= %llu\n", cf->tid, want_flush_tid);
+			ret = 0;
+		}
 	}
 	spin_unlock(&mdsc->cap_dirty_lock);
 	return ret;
@@ -3470,7 +3469,7 @@ int ceph_mdsc_init(struct ceph_fs_client *fsc)
 	INIT_LIST_HEAD(&mdsc->snap_flush_list);
 	spin_lock_init(&mdsc->snap_flush_lock);
 	mdsc->last_cap_flush_tid = 1;
-	mdsc->cap_flush_tree = RB_ROOT;
+	INIT_LIST_HEAD(&mdsc->cap_flush_list);
 	INIT_LIST_HEAD(&mdsc->cap_dirty);
 	INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
 	mdsc->num_cap_flushing = 0;

fs/ceph/mds_client.h

@@ -325,7 +325,7 @@ struct ceph_mds_client {
 	spinlock_t       snap_flush_lock;
 
 	u64               last_cap_flush_tid;
-	struct rb_root    cap_flush_tree;
+	struct list_head  cap_flush_list;
 	struct list_head  cap_dirty;        /* inodes with dirty caps */
 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
 	int               num_cap_flushing; /* # caps we are flushing */

fs/ceph/super.h

@@ -189,11 +189,8 @@ static inline void ceph_put_cap_snap(struct ceph_cap_snap *capsnap)
 struct ceph_cap_flush {
 	u64 tid;
 	int caps;
-	struct rb_node g_node; // global
-	union {
-		struct rb_node i_node; // inode
-		struct list_head list;
-	};
+	struct list_head g_list; // global
+	struct list_head i_list; // per inode
 };
 
 /*
@@ -310,7 +307,7 @@ struct ceph_inode_info {
 	 * overlapping, pipelined cap flushes to the mds.  we can probably
 	 * reduce the tid to 8 bits if we're concerned about inode size. */
 	struct ceph_cap_flush *i_prealloc_cap_flush;
-	struct rb_root i_cap_flush_tree;
+	struct list_head i_cap_flush_list;
 	wait_queue_head_t i_cap_wq;      /* threads waiting on a capability */
 	unsigned long i_hold_caps_min; /* jiffies */
 	unsigned long i_hold_caps_max; /* jiffies */