From 9296475397c72ace967bcdffed2a964a19664c24 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@newdream.net>
Date: Thu, 17 Jan 2008 11:54:53 -0800
Subject: [PATCH] clean up mds session open; request mdsmap when necessary

---
 src/kernel/mds_client.c | 40 +++++++++++++++++++++-------------------
 src/kernel/mon_client.c | 23 ++++++++++++++++++++++-
 src/kernel/mon_client.h |  2 ++
 3 files changed, 45 insertions(+), 20 deletions(-)

diff --git a/src/kernel/mds_client.c b/src/kernel/mds_client.c
index d281169cf9241..c8036e266eea1 100644
--- a/src/kernel/mds_client.c
+++ b/src/kernel/mds_client.c
@@ -167,12 +167,22 @@ static int open_session(struct ceph_mds_client *mdsc, struct ceph_mds_session *s
 	struct ceph_msg *msg;
 	int mstate;
 
-	/* connect */
+	/* mds active? */
 	mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
 	dout(10, "open_session to mds%d, state %d\n", mds, mstate);
 	if (mstate < CEPH_MDS_STATE_ACTIVE) {
 		ceph_monc_request_mdsmap(&mdsc->client->monc, mdsc->mdsmap->m_epoch);
-		return -EINPROGRESS;
+		spin_unlock(&mdsc->lock);
+		dout(30, "open_session waiting on map\n");
+		wait_for_completion(&mdsc->map_waiters);
+		dout(30, "open_session done waiting on map\n");
+		spin_lock(&mdsc->lock);
+
+		mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
+		if (mstate < CEPH_MDS_STATE_ACTIVE) {
+			dout(30, "open_session still not active...\n");
+			return -EAGAIN;  /* hrm, try again? */
+		}
 	} 
 	
 	/* send connect message */
@@ -181,6 +191,14 @@ static int open_session(struct ceph_mds_client *mdsc, struct ceph_mds_session *s
 		return PTR_ERR(msg);  /* fixme */
 	session->s_state = CEPH_MDS_SESSION_OPENING;
 	send_msg_mds(mdsc, msg, mds);
+
+	/* wait for session to open (or fail, or close) */
+	spin_unlock(&mdsc->lock);
+	dout(30, "open_session waiting on session %p\n", session);
+	wait_for_completion(&session->s_completion);
+	dout(30, "open_session done waiting on session %p, state %d\n", 
+	     session, session->s_state);
+	spin_lock(&mdsc->lock);
 	return 0;
 }
 
@@ -361,23 +379,7 @@ retry:
 	if (session->s_state == CEPH_MDS_SESSION_NEW ||
 	    session->s_state == CEPH_MDS_SESSION_CLOSING) {
 		err = open_session(mdsc, session, mds);
-		if (err == -EINPROGRESS) {
-			/* waiting for new mdsmap.  bleh, this is a little messy. */
-			spin_unlock(&mdsc->lock);
-			wait_for_completion(&mdsc->map_waiters);
-			spin_lock(&mdsc->lock);
-			goto retry;
-		}
-		BUG_ON(err);  /* implement me */
-	}
-	if (session->s_state == CEPH_MDS_SESSION_OPENING) {
-		/* wait for session to open (or fail, or close) */
-		spin_unlock(&mdsc->lock);
-		dout(30, "do_request waiting on session %p\n", session);
-		wait_for_completion(&session->s_completion);
-		dout(30, "do_request done waiting on session %p, state %d\n", 
-		     session, session->s_state);
-		spin_lock(&mdsc->lock);
+		BUG_ON(err && err != -EAGAIN);
 	}
 	if (session->s_state != CEPH_MDS_SESSION_OPEN) {
 		dout(30, "do_request session %p not open, %d\n", session, session->s_state);
diff --git a/src/kernel/mon_client.c b/src/kernel/mon_client.c
index 38aa2cd5f7041..1ef19c7ea6eef 100644
--- a/src/kernel/mon_client.c
+++ b/src/kernel/mon_client.c
@@ -86,11 +86,32 @@ int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
 }
 
 
+static int request_mdsmap(struct ceph_mon_client *monc)
+{
+	struct ceph_msg *msg;
+	int mon = pick_mon(monc, -1);
+
+	dout(10, "request_mdsmap to mon%d have %u\n", mon, monc->want_mdsmap_epoch_gt);
+	msg = ceph_msg_new(CEPH_MSG_MDS_GETMAP, sizeof(ceph_epoch_t), 0, 0, 0);
+	if (IS_ERR(msg))
+		return PTR_ERR(msg);
+	*(__le32*)msg->front.iov_base = cpu_to_le32(monc->want_mdsmap_epoch_gt);
+	msg->hdr.dst = monc->monmap->mon_inst[mon];
+	ceph_msg_send(monc->client->msgr, msg, 0);
+	return 0;
+}
+
 
 void ceph_monc_request_mdsmap(struct ceph_mon_client *monc, __u64 have)
 {
 	dout(5, "ceph_monc_request_mdsmap -- IMPLEMENT ME\n");
 	
+	if (have <= monc->want_mdsmap_epoch_gt) {
+		dout(5, " already waiting for > %u\n", monc->want_mdsmap_epoch_gt);
+		return;
+	}
+	monc->want_mdsmap_epoch_gt = have;
+	request_mdsmap(monc);
 }
 
 
@@ -144,7 +165,7 @@ int send_statfs(struct ceph_mon_client *monc, u64 tid)
 		return PTR_ERR(msg);
 	*(__le64*)msg->front.iov_base = cpu_to_le64(tid);
 	msg->hdr.dst = monc->monmap->mon_inst[mon];
-	ceph_msg_send(monc->client->msgr, msg, BASE_DELAY_INTERVAL);
+	ceph_msg_send(monc->client->msgr, msg, 0);
 	return 0;
 }
 
diff --git a/src/kernel/mon_client.h b/src/kernel/mon_client.h
index fc88b4d98e4c2..9bb74feafa49f 100644
--- a/src/kernel/mon_client.h
+++ b/src/kernel/mon_client.h
@@ -30,6 +30,8 @@ struct ceph_mon_client {
 	spinlock_t lock;
 	struct radix_tree_root statfs_request_tree;  /* statfs requests */
 	u64 last_tid;
+
+	u32 want_mdsmap_epoch_gt;
 };
 
 extern struct ceph_monmap *ceph_monmap_decode(void *p, void *end);
-- 
2.39.5