libceph: use keepalive2 to verify the mon session is alive

Signed-off-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

3 files changed, 84 insertions, 13 deletions

diff --git a/net/ceph/ceph_common.c b/net/ceph/ceph_common.c
index f30329f..3f56eef 100644
--- a/net/ceph/ceph_common.c
+++ b/net/ceph/ceph_common.c
@@ -357,6 +357,7 @@ ceph_parse_options(char *options, const char *dev_name,
 	opt->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
 	opt->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT;
 	opt->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;
+	opt->monc_ping_timeout = CEPH_MONC_PING_TIMEOUT_DEFAULT;
 
 	/* get mon ip(s) */
 	/* ip1[:port1][,ip2[:port2]...] */
diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c
index 101ab62..36757d4 100644
--- a/net/ceph/messenger.c
+++ b/net/ceph/messenger.c
@@ -163,6 +163,7 @@ static struct kmem_cache	*ceph_msg_data_cache;
 static char tag_msg = CEPH_MSGR_TAG_MSG;
 static char tag_ack = CEPH_MSGR_TAG_ACK;
 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
+static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
 
 #ifdef CONFIG_LOCKDEP
 static struct lock_class_key socket_class;
@@ -1351,7 +1352,15 @@ static void prepare_write_keepalive(struct ceph_connection *con)
 {
 	dout("prepare_write_keepalive %p\n", con);
 	con_out_kvec_reset(con);
-	con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
+	if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
+		struct timespec ts = CURRENT_TIME;
+		struct ceph_timespec ceph_ts;
+		ceph_encode_timespec(&ceph_ts, &ts);
+		con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
+		con_out_kvec_add(con, sizeof(ceph_ts), &ceph_ts);
+	} else {
+		con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
+	}
 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
 }
 
@@ -1625,6 +1634,12 @@ static void prepare_read_tag(struct ceph_connection *con)
 	con->in_tag = CEPH_MSGR_TAG_READY;
 }
 
+static void prepare_read_keepalive_ack(struct ceph_connection *con)
+{
+	dout("prepare_read_keepalive_ack %p\n", con);
+	con->in_base_pos = 0;
+}
+
 /*
  * Prepare to read a message.
  */
@@ -2457,6 +2472,17 @@ static void process_message(struct ceph_connection *con)
 	mutex_lock(&con->mutex);
 }
 
+static int read_keepalive_ack(struct ceph_connection *con)
+{
+	struct ceph_timespec ceph_ts;
+	size_t size = sizeof(ceph_ts);
+	int ret = read_partial(con, size, size, &ceph_ts);
+	if (ret <= 0)
+		return ret;
+	ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
+	prepare_read_tag(con);
+	return 1;
+}
 
 /*
  * Write something to the socket.  Called in a worker thread when the
@@ -2526,6 +2552,10 @@ more_kvec:
 
 do_next:
 	if (con->state == CON_STATE_OPEN) {
+		if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
+			prepare_write_keepalive(con);
+			goto more;
+		}
 		/* is anything else pending? */
 		if (!list_empty(&con->out_queue)) {
 			prepare_write_message(con);
@@ -2535,10 +2565,6 @@ do_next:
 			prepare_write_ack(con);
 			goto more;
 		}
-		if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
-			prepare_write_keepalive(con);
-			goto more;
-		}
 	}
 
 	/* Nothing to do! */
@@ -2641,6 +2667,9 @@ more:
 		case CEPH_MSGR_TAG_ACK:
 			prepare_read_ack(con);
 			break;
+		case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
+			prepare_read_keepalive_ack(con);
+			break;
 		case CEPH_MSGR_TAG_CLOSE:
 			con_close_socket(con);
 			con->state = CON_STATE_CLOSED;
@@ -2684,6 +2713,12 @@ more:
 		process_ack(con);
 		goto more;
 	}
+	if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
+		ret = read_keepalive_ack(con);
+		if (ret <= 0)
+			goto out;
+		goto more;
+	}
 
 out:
 	dout("try_read done on %p ret %d\n", con, ret);
@@ -3101,6 +3136,20 @@ void ceph_con_keepalive(struct ceph_connection *con)
 }
 EXPORT_SYMBOL(ceph_con_keepalive);
 
+bool ceph_con_keepalive_expired(struct ceph_connection *con,
+			       unsigned long interval)
+{
+	if (interval > 0 &&
+	    (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
+		struct timespec now = CURRENT_TIME;
+		struct timespec ts;
+		jiffies_to_timespec(interval, &ts);
+		ts = timespec_add(con->last_keepalive_ack, ts);
+		return timespec_compare(&now, &ts) >= 0;
+	}
+	return false;
+}
+
 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
 {
 	struct ceph_msg_data *data;
diff --git a/net/ceph/mon_client.c b/net/ceph/mon_client.c
index 9d6ff12..edda016 100644
--- a/net/ceph/mon_client.c
+++ b/net/ceph/mon_client.c
@@ -149,6 +149,10 @@ static int __open_session(struct ceph_mon_client *monc)
 			      CEPH_ENTITY_TYPE_MON, monc->cur_mon,
 			      &monc->monmap->mon_inst[monc->cur_mon].addr);
 
+		/* send an initial keepalive to ensure our timestamp is
+		 * valid by the time we are in an OPENED state */
+		ceph_con_keepalive(&monc->con);
+
 		/* initiatiate authentication handshake */
 		ret = ceph_auth_build_hello(monc->auth,
 					    monc->m_auth->front.iov_base,
@@ -170,14 +174,19 @@ static bool __sub_expired(struct ceph_mon_client *monc)
  */
 static void __schedule_delayed(struct ceph_mon_client *monc)
 {
-	unsigned int delay;
+	struct ceph_options *opt = monc->client->options;
+	unsigned long delay;
 
-	if (monc->cur_mon < 0 || __sub_expired(monc))
+	if (monc->cur_mon < 0 || __sub_expired(monc)) {
 		delay = 10 * HZ;
-	else
+	} else {
 		delay = 20 * HZ;
-	dout("__schedule_delayed after %u\n", delay);
-	schedule_delayed_work(&monc->delayed_work, delay);
+		if (opt->monc_ping_timeout > 0)
+			delay = min(delay, opt->monc_ping_timeout / 3);
+	}
+	dout("__schedule_delayed after %lu\n", delay);
+	schedule_delayed_work(&monc->delayed_work,
+			      round_jiffies_relative(delay));
 }
 
 /*
@@ -743,11 +752,23 @@ static void delayed_work(struct work_struct *work)
 		__close_session(monc);
 		__open_session(monc);  /* continue hunting */
 	} else {
-		ceph_con_keepalive(&monc->con);
+		struct ceph_options *opt = monc->client->options;
+		int is_auth = ceph_auth_is_authenticated(monc->auth);
+		if (ceph_con_keepalive_expired(&monc->con,
+					       opt->monc_ping_timeout)) {
+			dout("monc keepalive timeout\n");
+			is_auth = 0;
+			__close_session(monc);
+			monc->hunting = true;
+			__open_session(monc);
+		}
 
-		__validate_auth(monc);
+		if (!monc->hunting) {
+			ceph_con_keepalive(&monc->con);
+			__validate_auth(monc);
+		}
 
-		if (ceph_auth_is_authenticated(monc->auth))
+		if (is_auth)
 			__send_subscribe(monc);
 	}
 	__schedule_delayed(monc);