ceph: convert mdsc->cap_dirty to a per-session list

This is a per-sb list now, but that makes it difficult to tell when
the cap is the last dirty one associated with the session. Switch
this to be a per-session list, but continue using the
mdsc->cap_dirty_lock to protect the lists.

This list is only ever walked in ceph_flush_dirty_caps, so change that
to walk the sessions array and then flush the caps for inodes on each
session's list.

If the auth cap ever changes while the inode has dirty caps, then
move the inode to the appropriate session for the new auth_cap. Also,
ensure that we never remove an auth cap while the inode is still on the
s_cap_dirty list.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

1 files changed, 65 insertions, 13 deletions

diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 53db1fcbfdd3..d5ad2434bb07 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -597,6 +597,27 @@ static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
 	}
 }
 
+/**
+ * change_auth_cap_ses - move inode to appropriate lists when auth caps change
+ * @ci: inode to be moved
+ * @session: new auth caps session
+ */
+static void change_auth_cap_ses(struct ceph_inode_info *ci,
+				struct ceph_mds_session *session)
+{
+	lockdep_assert_held(&ci->i_ceph_lock);
+
+	if (list_empty(&ci->i_dirty_item) && list_empty(&ci->i_flushing_item))
+		return;
+
+	spin_lock(&session->s_mdsc->cap_dirty_lock);
+	if (!list_empty(&ci->i_dirty_item))
+		list_move(&ci->i_dirty_item, &session->s_cap_dirty);
+	if (!list_empty(&ci->i_flushing_item))
+		list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
+	spin_unlock(&session->s_mdsc->cap_dirty_lock);
+}
+
 /*
  * Add a capability under the given MDS session.
  *
@@ -727,6 +748,9 @@ void ceph_add_cap(struct inode *inode,
 	if (flags & CEPH_CAP_FLAG_AUTH) {
 		if (!ci->i_auth_cap ||
 		    ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
+			if (ci->i_auth_cap &&
+			    ci->i_auth_cap->session != cap->session)
+				change_auth_cap_ses(ci, cap->session);
 			ci->i_auth_cap = cap;
 			cap->mds_wanted = wanted;
 		}
@@ -1123,8 +1147,10 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
 
 	/* remove from inode's cap rbtree, and clear auth cap */
 	rb_erase(&cap->ci_node, &ci->i_caps);
-	if (ci->i_auth_cap == cap)
+	if (ci->i_auth_cap == cap) {
+		WARN_ON_ONCE(!list_empty(&ci->i_dirty_item));
 		ci->i_auth_cap = NULL;
+	}
 
 	/* remove from session list */
 	spin_lock(&session->s_cap_lock);
@@ -1689,6 +1715,8 @@ int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
 	     ceph_cap_string(was | mask));
 	ci->i_dirty_caps |= mask;
 	if (was == 0) {
+		struct ceph_mds_session *session = ci->i_auth_cap->session;
+
 		WARN_ON_ONCE(ci->i_prealloc_cap_flush);
 		swap(ci->i_prealloc_cap_flush, *pcf);
 
@@ -1701,7 +1729,7 @@ int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
 		     &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
 		BUG_ON(!list_empty(&ci->i_dirty_item));
 		spin_lock(&mdsc->cap_dirty_lock);
-		list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
+		list_add(&ci->i_dirty_item, &session->s_cap_dirty);
 		spin_unlock(&mdsc->cap_dirty_lock);
 		if (ci->i_flushing_caps == 0) {
 			ihold(inode);
@@ -3749,15 +3777,9 @@ retry:
 			tcap->issue_seq = t_seq - 1;
 			tcap->issued |= issued;
 			tcap->implemented |= issued;
-			if (cap == ci->i_auth_cap)
+			if (cap == ci->i_auth_cap) {
 				ci->i_auth_cap = tcap;
-
-			if (!list_empty(&ci->i_cap_flush_list) &&
-			    ci->i_auth_cap == tcap) {
-				spin_lock(&mdsc->cap_dirty_lock);
-				list_move_tail(&ci->i_flushing_item,
-					       &tcap->session->s_cap_flushing);
-				spin_unlock(&mdsc->cap_dirty_lock);
+				change_auth_cap_ses(ci, tcap->session);
 			}
 		}
 		__ceph_remove_cap(cap, false);
@@ -4176,15 +4198,16 @@ void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
 /*
  * Flush all dirty caps to the mds
  */
-void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
+static void flush_dirty_session_caps(struct ceph_mds_session *s)
 {
+	struct ceph_mds_client *mdsc = s->s_mdsc;
 	struct ceph_inode_info *ci;
 	struct inode *inode;
 
 	dout("flush_dirty_caps\n");
 	spin_lock(&mdsc->cap_dirty_lock);
-	while (!list_empty(&mdsc->cap_dirty)) {
-		ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
+	while (!list_empty(&s->s_cap_dirty)) {
+		ci = list_first_entry(&s->s_cap_dirty, struct ceph_inode_info,
 				      i_dirty_item);
 		inode = &ci->vfs_inode;
 		ihold(inode);
@@ -4198,6 +4221,35 @@ void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
 	dout("flush_dirty_caps done\n");
 }
 
+static void iterate_sessions(struct ceph_mds_client *mdsc,
+			     void (*cb)(struct ceph_mds_session *))
+{
+	int mds;
+
+	mutex_lock(&mdsc->mutex);
+	for (mds = 0; mds < mdsc->max_sessions; ++mds) {
+		struct ceph_mds_session *s;
+
+		if (!mdsc->sessions[mds])
+			continue;
+
+		s = ceph_get_mds_session(mdsc->sessions[mds]);
+		if (!s)
+			continue;
+
+		mutex_unlock(&mdsc->mutex);
+		cb(s);
+		ceph_put_mds_session(s);
+		mutex_lock(&mdsc->mutex);
+	}
+	mutex_unlock(&mdsc->mutex);
+}
+
+void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
+{
+	iterate_sessions(mdsc, flush_dirty_session_caps);
+}
+
 void __ceph_touch_fmode(struct ceph_inode_info *ci,
 			struct ceph_mds_client *mdsc, int fmode)
 {