Communication Framework: Receiving
authorGoldwyn Rodrigues <rgoldwyn@suse.com>
Sat, 7 Jun 2014 06:08:29 +0000 (01:08 -0500)
committerGoldwyn Rodrigues <rgoldwyn@suse.com>
Mon, 23 Feb 2015 15:59:06 +0000 (09:59 -0600)
1. receive status

   sender                         receiver                   receiver
   ACK:CR                          ACK:CR                     ACK:CR

2. sender get EX of TOKEN
   sender get EX of MESSAGE
   sender                          receiver                   receiver
   TOKEN:EX                         ACK:CR                     ACK:CR
   MESSAGE:EX
   ACK:CR

3. sender write LVB.
   sender down-convert MESSAGE from EX to CR
   sender try to get EX of ACK
   [ wait until all receiver has *processed* the MESSAGE ]

                                     [ triggered by bast of ACK ]
                                     receiver get CR of MESSAGE
                                     receiver read LVB
                                     receiver processes the message
     [ wait finish ]
                                     receiver release ACK

   sender                         receiver                   receiver
   TOKEN:EX                       MESSAGE:CR                 MESSAGE:CR
   MESSAGE:CR
   ACK:EX

4. sender down-convert ACK from EX to CR
   sender release MESSAGE
   sender release TOKEN
  receiver upconvert to EX of MESSAGE
                                  receiver get CR of ACK
  receiver release MESSAGE

   sender                        receiver                   receiver
   ACK:CR                         ACK:CR                     ACK:CR

Signed-off-by: Lidong Zhong <lzhong@suse.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
drivers/md/md-cluster.c

index d298713..96734cd 100644 (file)
@@ -52,6 +52,23 @@ struct md_cluster_info {
        spinlock_t suspend_lock;
        struct md_thread *recovery_thread;
        unsigned long recovery_map;
+       /* communication loc resources */
+       struct dlm_lock_resource *ack_lockres;
+       struct dlm_lock_resource *message_lockres;
+       struct dlm_lock_resource *token_lockres;
+       struct md_thread *recv_thread;
+};
+
+enum msg_type {
+       METADATA_UPDATED = 0,
+       RESYNCING,
+};
+
+struct cluster_msg {
+       int type;
+       int slot;
+       sector_t low;
+       sector_t high;
 };
 
 static void sync_ast(void *arg)
@@ -283,6 +300,64 @@ static const struct dlm_lockspace_ops md_ls_ops = {
        .recover_done = recover_done,
 };
 
+/*
+ * The BAST function for the ack lock resource
+ * This function wakes up the receive thread in
+ * order to receive and process the message.
+ */
+static void ack_bast(void *arg, int mode)
+{
+       struct dlm_lock_resource *res = (struct dlm_lock_resource *)arg;
+       struct md_cluster_info *cinfo = res->mddev->cluster_info;
+
+       if (mode == DLM_LOCK_EX)
+               md_wakeup_thread(cinfo->recv_thread);
+}
+
+static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg)
+{
+       switch (msg->type) {
+       case METADATA_UPDATED:
+               pr_info("%s: %d Received message: METADATA_UPDATE from %d\n",
+                       __func__, __LINE__, msg->slot);
+               break;
+       case RESYNCING:
+               pr_info("%s: %d Received message: RESYNCING from %d\n",
+                       __func__, __LINE__, msg->slot);
+               break;
+       };
+}
+
+/*
+ * thread for receiving message
+ */
+static void recv_daemon(struct md_thread *thread)
+{
+       struct md_cluster_info *cinfo = thread->mddev->cluster_info;
+       struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres;
+       struct dlm_lock_resource *message_lockres = cinfo->message_lockres;
+       struct cluster_msg msg;
+
+       /*get CR on Message*/
+       if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) {
+               pr_err("md/raid1:failed to get CR on MESSAGE\n");
+               return;
+       }
+
+       /* read lvb and wake up thread to process this message_lockres */
+       memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg));
+       process_recvd_msg(thread->mddev, &msg);
+
+       /*release CR on ack_lockres*/
+       dlm_unlock_sync(ack_lockres);
+       /*up-convert to EX on message_lockres*/
+       dlm_lock_sync(message_lockres, DLM_LOCK_EX);
+       /*get CR on ack_lockres again*/
+       dlm_lock_sync(ack_lockres, DLM_LOCK_CR);
+       /*release CR on message_lockres*/
+       dlm_unlock_sync(message_lockres);
+}
+
 static int gather_all_resync_info(struct mddev *mddev, int total_slots)
 {
        struct md_cluster_info *cinfo = mddev->cluster_info;
@@ -368,6 +443,26 @@ static int join(struct mddev *mddev, int nodes)
                ret = -ENOMEM;
                goto err;
        }
+       /* Initiate the communication resources */
+       ret = -ENOMEM;
+       cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv");
+       if (!cinfo->recv_thread) {
+               pr_err("md-cluster: cannot allocate memory for recv_thread!\n");
+               goto err;
+       }
+       cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1);
+       if (!cinfo->message_lockres)
+               goto err;
+       cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0);
+       if (!cinfo->token_lockres)
+               goto err;
+       cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0);
+       if (!cinfo->ack_lockres)
+               goto err;
+       /* get sync CR lock on ACK. */
+       if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR))
+               pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n",
+                               ret);
 
        pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
        snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
@@ -389,6 +484,9 @@ static int join(struct mddev *mddev, int nodes)
 
        return 0;
 err:
+       lockres_free(cinfo->message_lockres);
+       lockres_free(cinfo->token_lockres);
+       lockres_free(cinfo->ack_lockres);
        lockres_free(cinfo->bitmap_lockres);
        lockres_free(cinfo->sb_lock);
        if (cinfo->lockspace)
@@ -406,6 +504,10 @@ static int leave(struct mddev *mddev)
        if (!cinfo)
                return 0;
        md_unregister_thread(&cinfo->recovery_thread);
+       md_unregister_thread(&cinfo->recv_thread);
+       lockres_free(cinfo->message_lockres);
+       lockres_free(cinfo->token_lockres);
+       lockres_free(cinfo->ack_lockres);
        lockres_free(cinfo->sb_lock);
        lockres_free(cinfo->bitmap_lockres);
        dlm_release_lockspace(cinfo->lockspace, 2);