cleaned up mds map requesting

src/TODO

@@ -24,6 +24,11 @@ kernel client
   - clean up ceph_send_fault
   - more testing
   - callbacks for 'remote reset' (once 'caps' branch is merged in)
+
+ - TCP_CLOSE case in callback.. should it set CLOSED bit?
+ - ceph_queue_read .. check READING bit?
+ - 
+
 - mds client
   - pin inodes with caps (igrab/iput?)
   - mechanism to close out old caps

src/kernel/mds_client.c

@@ -162,6 +162,18 @@ static struct ceph_msg *create_session_msg(__u32 op, __u64 seq)
 	return msg;
 }
 
+static void wait_for_new_map(struct ceph_mds_client *mdsc)
+{
+	dout(30, "wait_for_new_map enter\n");
+	if (mdsc->last_requested_map < mdsc->mdsmap->m_epoch)
+		ceph_monc_request_mdsmap(&mdsc->client->monc, mdsc->mdsmap->m_epoch);
+
+	spin_unlock(&mdsc->lock);
+	wait_for_completion(&mdsc->map_waiters);
+	spin_lock(&mdsc->lock);
+	dout(30, "wait_for_new_map exit\n");
+}
+
 static int open_session(struct ceph_mds_client *mdsc, struct ceph_mds_session *session, int mds)
 {
 	struct ceph_msg *msg;
@@ -171,16 +183,11 @@ static int open_session(struct ceph_mds_client *mdsc, struct ceph_mds_session *s
 	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);
-		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);
-
+		wait_for_new_map(mdsc);
 		mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
 		if (mstate < CEPH_MDS_STATE_ACTIVE) {
-			dout(30, "open_session still not active...\n");
+			dout(30, "open_session mds%d now %d, still not active...\n",
+			     mds, mstate);
 			return -EAGAIN;  /* hrm, try again? */
 		}
 	} 
@@ -258,15 +265,7 @@ bad:
 }
 
 
-static void wait_for_new_map(struct ceph_mds_client *mdsc)
-{
-	dout(30, "wait_for_new_map enter\n");
-	if (mdsc->last_requested_map < mdsc->mdsmap->m_epoch)
-		ceph_monc_request_mdsmap(&mdsc->client->monc, mdsc->mdsmap->m_epoch);
 
-	wait_for_completion(&mdsc->map_waiters);
-	dout(30, "wait_for_new_map exit\n");
-}
 
 /* exported functions */
 
@@ -362,11 +361,8 @@ int ceph_mdsc_do_request(struct ceph_mds_client *mdsc, struct ceph_msg *msg,
 retry:
 	mds = choose_mds(mdsc, req);
 	if (mds < 0) {
-		/* wait for new mdsmap */
-		spin_unlock(&mdsc->lock);
 		dout(30, "do_request waiting for new mdsmap\n");
 		wait_for_new_map(mdsc);
-		spin_lock(&mdsc->lock);
 		goto retry;
 	}
 	dout(30, "do_request chose mds%d\n", mds);
@@ -900,6 +896,7 @@ void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
 		dout(2, "got first mdsmap %u\n", newmap->m_epoch);
 		mdsc->mdsmap = newmap;
 	}
+	ceph_monc_got_mdsmap(&mdsc->client->monc, newmap->m_epoch);  /* stop asking */
 	spin_unlock(&mdsc->lock);
 	complete(&mdsc->map_waiters);
 	return;

src/kernel/mon_client.c

@@ -82,36 +82,39 @@ int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
 		return -ENOMEM;
 	spin_lock_init(&monc->lock);
 	INIT_RADIX_TREE(&monc->statfs_request_tree, GFP_KERNEL);
+	monc->last_tid = 0;
+	monc->want_mdsmap = 0;
 	return 0;
 }
 
-
-static int request_mdsmap(struct ceph_mon_client *monc)
+int ceph_monc_request_mdsmap(struct ceph_mon_client *monc, __u32 have)
 {
 	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);
+	dout(5, "ceph_monc_request_mdsmap from mon%d have %u\n", mon, have);
+	monc->want_mdsmap = have;
+	msg = ceph_msg_new(CEPH_MSG_MDS_GETMAP, sizeof(__u32), 0, 0, 0);
 	if (IS_ERR(msg))
 		return PTR_ERR(msg);
-	*(__le32*)msg->front.iov_base = cpu_to_le32(monc->want_mdsmap_epoch_gt);
+	*(__le32*)msg->front.iov_base = cpu_to_le32(have);
 	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)
+int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, __u32 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;
+	if (have > monc->want_mdsmap) {
+		monc->want_mdsmap = 0;
+		dout(5, "ceph_monc_got_mdsmap have %u > wanted %u\n", 
+		     have, monc->want_mdsmap);
+		return 0;
+	} else {
+		dout(5, "ceph_monc_got_mdsmap have %u <= wanted %u *****\n", 
+		     have, monc->want_mdsmap);
+		return -EAGAIN;
 	}
-	monc->want_mdsmap_epoch_gt = have;
-	request_mdsmap(monc);
 }
 
 

src/kernel/mon_client.h

@@ -31,7 +31,7 @@ struct ceph_mon_client {
 	struct radix_tree_root statfs_request_tree;  /* statfs requests */
 	u64 last_tid;
 
-	u32 want_mdsmap_epoch_gt;
+	u32 want_mdsmap;  /* protected by caller's lock */
 };
 
 extern struct ceph_monmap *ceph_monmap_decode(void *p, void *end);
@@ -39,7 +39,10 @@ extern int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *
 
 extern int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl);
 
-extern void ceph_monc_request_mdsmap(struct ceph_mon_client *monc, __u64 have);
+extern int ceph_monc_request_mdsmap(struct ceph_mon_client *monc, __u32 have);
+extern int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, __u32 have);
+
+
 extern void ceph_monc_request_osdmap(struct ceph_mon_client *monc, __u64 have);
 extern void ceph_monc_request_umount(struct ceph_mon_client *monc);
 extern void ceph_monc_report_failure(struct ceph_mon_client *monc, struct ceph_entity_inst *who);