mgmt/glusterd: Fix replace brick to pass src-brick and dst-brick port nos properly.

Fix also contains patch from shishir ng for maintaining replace-brick
states in glusterd.
Signed-off-by: Pavan Vilas Sondur <pavan@gluster.com>
Signed-off-by: Vijay Bellur <vijay@dev.gluster.com>

BUG: 1235 (Bug for all pump/migrate commits)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1235

8 files changed, 486 insertions, 182 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-handler.c b/xlators/mgmt/glusterd/src/glusterd-handler.c
index 7e5cddac8..70734074f 100644
--- a/xlators/mgmt/glusterd/src/glusterd-handler.c
+++ b/xlators/mgmt/glusterd/src/glusterd-handler.c
@@ -1890,11 +1890,12 @@ out:
 
 int
 glusterd_op_stage_send_resp (rpcsvc_request_t   *req,
-                             int32_t op, int32_t status, char *op_errstr)
+                             int32_t op, int32_t status,
+                             char *op_errstr, dict_t *rsp_dict)
 {
 
-        gd1_mgmt_stage_op_rsp           rsp = {{0},};
-        int                             ret = -1;
+        gd1_mgmt_stage_op_rsp           rsp      = {{0},};
+        int                             ret      = -1;
 
         GF_ASSERT (req);
         rsp.op_ret = status;
@@ -1905,6 +1906,15 @@ glusterd_op_stage_send_resp (rpcsvc_request_t   *req,
         else
                 rsp.op_errstr = "";
 
+        ret = dict_allocate_and_serialize (rsp_dict,
+                                           &rsp.dict.dict_val,
+                                           (size_t *)&rsp.dict.dict_len);
+        if (ret < 0) {
+                gf_log ("", GF_LOG_DEBUG,
+                        "failed to get serialized length of dict");
+                return ret;
+        }
+
         ret = glusterd_submit_reply (req, &rsp, NULL, 0, NULL,
                                      gd_xdr_serialize_mgmt_stage_op_rsp);
 
diff --git a/xlators/mgmt/glusterd/src/glusterd-op-sm.c b/xlators/mgmt/glusterd/src/glusterd-op-sm.c
index d9d324b2e..34caacd5e 100644
--- a/xlators/mgmt/glusterd/src/glusterd-op-sm.c
+++ b/xlators/mgmt/glusterd/src/glusterd-op-sm.c
@@ -828,7 +828,8 @@ out:
 }
 
 static int
-glusterd_op_stage_replace_brick (gd1_mgmt_stage_op_req *req, char **op_errstr)
+glusterd_op_stage_replace_brick (gd1_mgmt_stage_op_req *req, char **op_errstr,
+                                 dict_t *rsp_dict)
 {
         int                                      ret           = 0;
         dict_t                                  *dict          = NULL;
@@ -844,6 +845,7 @@ glusterd_op_stage_replace_brick (gd1_mgmt_stage_op_req *req, char **op_errstr)
         char                                    *dup_dstbrick  = NULL;
         glusterd_peerinfo_t                     *peerinfo = NULL;
         struct stat                             st_buf = {0,};
+        glusterd_brickinfo_t                    *dst_brickinfo = NULL;
 
         GF_ASSERT (req);
 
@@ -917,6 +919,56 @@ glusterd_op_stage_replace_brick (gd1_mgmt_stage_op_req *req, char **op_errstr)
                 goto out;
         }
 
+        switch (replace_op) {
+        case GF_REPLACE_OP_START:
+                if (glusterd_is_rb_started (volinfo)) {
+                        gf_log ("", GF_LOG_ERROR, "Replace brick is already "
+                                "started for volume ");
+                        ret = -1;
+                        goto out;
+                }
+                break;
+        case GF_REPLACE_OP_PAUSE:
+                if (glusterd_is_rb_paused (volinfo)) {
+                        gf_log ("", GF_LOG_ERROR, "Replace brick is already"
+                                " paused for volume ");
+                        ret = -1;
+                        goto out;
+                } else if (!glusterd_is_rb_started(volinfo)) {
+                        gf_log ("", GF_LOG_ERROR, "Replace brick is not"
+                                " started for volume ");
+                        ret = -1;
+                        goto out;
+                }
+                break;
+
+        case GF_REPLACE_OP_ABORT:
+                if ((!glusterd_is_rb_paused (volinfo)) &&
+                     (!glusterd_is_rb_started (volinfo))) {
+                        gf_log ("", GF_LOG_ERROR, "Replace brick is not "
+                                " started or paused for volume ");
+                        ret = -1;
+                        goto out;
+                }
+                break;
+
+        case GF_REPLACE_OP_COMMIT:
+                if (!glusterd_is_rb_started (volinfo)) {
+                        gf_log ("", GF_LOG_ERROR, "Replace brick is not "
+                                "started for volume ");
+                        ret = -1;
+                        goto out;
+                }
+                break;
+
+        case GF_REPLACE_OP_COMMIT_FORCE: break;
+        case GF_REPLACE_OP_STATUS:
+                break;
+        default:
+                ret = -1;
+                goto out;
+        }
+
         ret = glusterd_volume_brickinfo_get_by_brick (src_brick, volinfo,
                                       &src_brickinfo);
         if (ret) {
@@ -929,6 +981,16 @@ glusterd_op_stage_replace_brick (gd1_mgmt_stage_op_req *req, char **op_errstr)
         if (!glusterd_is_local_addr (src_brickinfo->hostname)) {
                 gf_log ("", GF_LOG_DEBUG,
                         "I AM THE SOURCE HOST");
+                if (src_brickinfo->port) {
+                        ret = dict_set_int32 (rsp_dict, "src-brick-port",
+                                              src_brickinfo->port);
+                        if (ret) {
+                                gf_log ("", GF_LOG_DEBUG,
+                                        "Could not set src-brick-port=%d",
+                                        src_brickinfo->port);
+                        }
+                }
+
         }
 
         dup_dstbrick = gf_strdup (dst_brick);
@@ -955,6 +1017,19 @@ glusterd_op_stage_replace_brick (gd1_mgmt_stage_op_req *req, char **op_errstr)
                 goto out;
         }
 
+        ret = glusterd_brickinfo_from_brick (dst_brick, &dst_brickinfo);
+        if ((volinfo->rb_status ==GF_RB_STATUS_NONE) &&
+            (replace_op == GF_REPLACE_OP_START)) {
+                volinfo->src_brick = src_brickinfo;
+                volinfo->dst_brick = dst_brickinfo;
+        }
+
+        if (glusterd_rb_check_bricks (volinfo, src_brickinfo, dst_brickinfo)) {
+                gf_log ("", GF_LOG_ERROR, "replace brick: incorrect source or"
+                       "  destination bricks specified");
+                ret = -1;
+                goto out;
+       }
         if (!glusterd_is_local_addr (host)) {
                 ret = stat (path, &st_buf);
                 if (ret == -1) {
@@ -2669,6 +2744,114 @@ out:
         return ret;
 }
 
+/* Set src-brick's port number to be used in the maintainance mount
+ * after all commit acks are received.
+ */
+static int
+rb_update_srcbrick_port (glusterd_brickinfo_t *src_brickinfo, dict_t *rsp_dict,
+                         dict_t *req_dict, int32_t replace_op)
+{
+        xlator_t *this            = NULL;
+        dict_t   *ctx             = NULL;
+        int       ret             = 0;
+        int       dict_ret        = 0;
+        int       src_port        = 0;
+
+        this = THIS;
+
+        ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+        if (ctx) {
+                dict_ret = dict_get_int32 (req_dict, "src-brick-port", &src_port);
+                if (src_port)
+                        src_brickinfo->port = src_port;
+        }
+
+        if (!glusterd_is_local_addr (src_brickinfo->hostname)) {
+                gf_log ("", GF_LOG_NORMAL,
+                        "adding src-brick port no");
+
+                src_brickinfo->port = pmap_registry_search (this,
+                                      src_brickinfo->path, GF_PMAP_PORT_BRICKSERVER);
+                if (!src_brickinfo->port &&
+                    replace_op != GF_REPLACE_OP_COMMIT_FORCE ) {
+                        gf_log ("", GF_LOG_ERROR,
+                                "Src brick port not available");
+                        ret = -1;
+                        goto out;
+                }
+
+                if (rsp_dict) {
+                        ret = dict_set_int32 (rsp_dict, "src-brick-port", src_brickinfo->port);
+                        if (ret) {
+                                gf_log ("", GF_LOG_DEBUG,
+                                        "Could not set src-brick port no");
+                                goto out;
+                        }
+                }
+
+                ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+                if (ctx) {
+                        ret = dict_set_int32 (ctx, "src-brick-port", src_brickinfo->port);
+                        if (ret) {
+                                gf_log ("", GF_LOG_DEBUG,
+                                        "Could not set src-brick port no");
+                                goto out;
+                        }
+                }
+
+        }
+
+out:
+        return ret;
+
+}
+
+static int
+rb_update_dstbrick_port (glusterd_brickinfo_t *dst_brickinfo, dict_t *rsp_dict,
+                         dict_t *req_dict, int32_t replace_op)
+{
+        dict_t *ctx           = NULL;
+        int     ret           = 0;
+        int     dict_ret      = 0;
+        int     dst_port      = 0;
+
+        ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+        if (ctx) {
+                dict_ret = dict_get_int32 (req_dict, "dst-brick-port", &dst_port);
+                if (dst_port)
+                        dst_brickinfo->port = dst_port;
+
+        }
+
+        if (!glusterd_is_local_addr (dst_brickinfo->hostname)) {
+                gf_log ("", GF_LOG_NORMAL,
+                        "adding dst-brick port no");
+
+                if (rsp_dict) {
+                        ret = dict_set_int32 (rsp_dict, "dst-brick-port",
+                                              dst_brickinfo->port);
+                        if (ret) {
+                                gf_log ("", GF_LOG_DEBUG,
+                                        "Could not set dst-brick port no in rsp dict");
+                                goto out;
+                        }
+                }
+
+                ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+                if (ctx) {
+                        ret = dict_set_int32 (ctx, "dst-brick-port",
+                                              dst_brickinfo->port);
+                        if (ret) {
+                                gf_log ("", GF_LOG_DEBUG,
+                                        "Could not set dst-brick port no");
+                                goto out;
+                        }
+                }
+        }
+out:
+        return ret;
+}
+
 static int
 glusterd_op_replace_brick (gd1_mgmt_stage_op_req *req, dict_t *rsp_dict)
 {
@@ -2758,43 +2941,18 @@ glusterd_op_replace_brick (gd1_mgmt_stage_op_req *req, dict_t *rsp_dict)
                 gf_log ("", GF_LOG_DEBUG, "Unable to resolve dst-brickinfo");
                 goto out;
         }
-        /* Set src-brick's port number to be used in the maintainance mount
-         * after all commit acks are received.
-         */
-        if (!glusterd_is_local_addr (src_brickinfo->hostname)) {
-                gf_log ("", GF_LOG_NORMAL,
-                        "adding src-brick port no");
 
-                src_brickinfo->port = pmap_registry_search (this,
-                                      src_brickinfo->path, GF_PMAP_PORT_BRICKSERVER);
-                if (!src_brickinfo->port &&
-                    replace_op != GF_REPLACE_OP_COMMIT_FORCE ) {
-                        gf_log ("", GF_LOG_ERROR,
-                                "Src brick port not available");
-                        ret = -1;
-                        goto out;
-                }
-
-                if (rsp_dict) {
-                        ret = dict_set_int32 (rsp_dict, "src-brick-port", src_brickinfo->port);
-                        if (ret) {
-                                gf_log ("", GF_LOG_DEBUG,
-                                        "Could not set src-brick port no");
-                                goto out;
-                        }
-                } else {
-                        ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
-                        GF_ASSERT (ctx);
-
-                        ret = dict_set_int32 (ctx, "src-brick-port", src_brickinfo->port);
-                        if (ret) {
-                                gf_log ("", GF_LOG_DEBUG,
-                                        "Could not set src-brick port no");
-                                goto out;
-                        }
-                }
+        ret = rb_update_srcbrick_port (src_brickinfo, rsp_dict,
+                                       dict, replace_op);
+        if (ret)
+                goto out;
 
-        }
+	if ((GF_REPLACE_OP_START != replace_op)) {
+                ret = rb_update_dstbrick_port (dst_brickinfo, rsp_dict,
+                                               dict, replace_op);
+                if (ret)
+                        goto out;
+	}
 
         switch (replace_op) {
         case GF_REPLACE_OP_START:
@@ -2802,40 +2960,19 @@ glusterd_op_replace_brick (gd1_mgmt_stage_op_req *req, dict_t *rsp_dict)
                 if (!glusterd_is_local_addr (dst_brickinfo->hostname)) {
                         gf_log ("", GF_LOG_NORMAL,
                                 "I AM THE DESTINATION HOST");
-                        ret = rb_spawn_destination_brick (volinfo, dst_brickinfo);
-                        if (ret) {
-                                gf_log ("", GF_LOG_DEBUG,
-                                        "Failed to spawn destination brick");
-                                goto out;
-                        }
-
-                        if (!glusterd_is_local_addr (dst_brickinfo->hostname)) {
-                                gf_log ("", GF_LOG_NORMAL,
-                                        "adding dst-brick port no");
-
-                                if (rsp_dict) {
-                                        ret = dict_set_int32 (rsp_dict, "dst-brick-port",
-                                                              dst_brickinfo->port);
-                                        if (ret) {
-                                                gf_log ("", GF_LOG_DEBUG,
-                                                        "Could not set dst-brick port no");
-                                                goto out;
-                                        }
-                                } else {
-                                        ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
-                                        GF_ASSERT (ctx);
-
-                                        ret = dict_set_int32 (ctx, "dst-brick-port",
-                                                              dst_brickinfo->port);
-                                        if (ret) {
-                                                gf_log ("", GF_LOG_DEBUG,
-                                                        "Could not set dst-brick port no");
-                                                goto out;
-                                        }
+                        if (!glusterd_is_rb_paused (volinfo)) {
+                                ret = rb_spawn_destination_brick (volinfo, dst_brickinfo);
+                                if (ret) {
+                                        gf_log ("", GF_LOG_DEBUG,
+                                                "Failed to spawn destination brick");
+                                        goto out;
                                 }
-
+                        } else {
+                                gf_log ("", GF_LOG_ERROR, "Replace brick is already "
+                                        "started=> no need to restart dst brick ");
                         }
-                }
+		}
+
 
 		if (!glusterd_is_local_addr (src_brickinfo->hostname)) {
 		        ret = rb_src_brick_restart (volinfo, src_brickinfo,
@@ -2846,13 +2983,24 @@ glusterd_op_replace_brick (gd1_mgmt_stage_op_req *req, dict_t *rsp_dict)
 			        goto out;
 			}
 		}
+
+		if (!glusterd_is_local_addr (dst_brickinfo->hostname)) {
+			gf_log ("", GF_LOG_NORMAL,
+				"adding dst-brick port no");
+
+                        ret = rb_update_dstbrick_port (dst_brickinfo, rsp_dict,
+                                                       dict, replace_op);
+                        if (ret)
+                                goto out;
+                }
+
+                glusterd_set_rb_status (volinfo, GF_RB_STATUS_STARTED);
 		break;
 	}
 
         case GF_REPLACE_OP_COMMIT:
         case GF_REPLACE_OP_COMMIT_FORCE:
         {
-
                 ret = dict_set_int32 (volinfo->dict, "enable-pump", 0);
                 gf_log ("", GF_LOG_DEBUG,
                         "Received commit - will be adding dst brick and "
@@ -2876,65 +3024,67 @@ glusterd_op_replace_brick (gd1_mgmt_stage_op_req *req, dict_t *rsp_dict)
                         goto out;
                 }
 
-                ret = glusterd_op_perform_replace_brick (volinfo, src_brick,
-                                                         dst_brick);
-                if (ret) {
-                        gf_log ("", GF_LOG_CRITICAL, "Unable to add "
-                                "dst-brick: %s to volume: %s",
-                                dst_brick, volinfo->volname);
-                        goto out;
-                }
 
-                volinfo->version++;
-                volinfo->defrag_status = 0;
+		ret = glusterd_op_perform_replace_brick (volinfo, src_brick,
+							 dst_brick);
+		if (ret) {
+			gf_log ("", GF_LOG_CRITICAL, "Unable to add "
+				"dst-brick: %s to volume: %s",
+				dst_brick, volinfo->volname);
+			goto out;
+		}
 
-                ret = glusterd_store_update_volume (volinfo);
+		volinfo->version++;
+		volinfo->defrag_status = 0;
 
-                if (ret)
-                        goto out;
+		ret = glusterd_store_update_volume (volinfo);
 
-                ret = glusterd_volume_compute_cksum (volinfo);
-                if (ret)
-                        goto out;
+		if (ret)
+			goto out;
 
-                ret = glusterd_check_generate_start_nfs (volinfo);
+		ret = glusterd_volume_compute_cksum (volinfo);
+		if (ret)
+			goto out;
 
-                if (ret) {
-                        gf_log ("", GF_LOG_CRITICAL, "Failed to generate "
-                                " nfs volume file");
-                }
+		ret = glusterd_check_generate_start_nfs (volinfo);
 
-                ret = glusterd_fetchspec_notify (THIS);
+		if (ret) {
+			gf_log ("", GF_LOG_CRITICAL, "Failed to generate "
+				" nfs volume file");
+		}
 
+		ret = glusterd_fetchspec_notify (THIS);
+                glusterd_set_rb_status (volinfo, GF_RB_STATUS_NONE);
+                volinfo->src_brick = volinfo->dst_brick = NULL;
         }
-                break;
+        break;
 
         case GF_REPLACE_OP_PAUSE:
         {
                 gf_log ("", GF_LOG_DEBUG,
                         "Recieved pause - doing nothing");
-                ret = rb_do_operation_pause (volinfo, src_brickinfo, dst_brickinfo);
-                if (!glusterd_is_local_addr (dst_brickinfo->hostname)) {
-                        gf_log ("", GF_LOG_NORMAL,
-                                "I AM THE DESTINATION HOST");
-                        ret = rb_kill_destination_brick (volinfo, dst_brickinfo);
+                ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+                if (ctx) {
+                        ret = rb_do_operation_pause (volinfo, src_brickinfo, dst_brickinfo);
                         if (ret) {
-                                gf_log ("", GF_LOG_DEBUG,
-                                        "Failed to kill destination brick");
+                                gf_log ("", GF_LOG_ERROR,
+                                        "Pause operation failed");
                                 goto out;
                         }
                 }
+
+                glusterd_set_rb_status (volinfo, GF_RB_STATUS_PAUSED);
         }
                 break;
 
         case GF_REPLACE_OP_ABORT:
         {
+
                 ret = dict_set_int32 (volinfo->dict, "enable-pump", 0);
 		if (ret) {
 			gf_log ("", GF_LOG_CRITICAL, "Unable to disable pump");
 		}
 
-                ret = rb_do_operation_abort (volinfo, src_brickinfo, dst_brickinfo);
                 if (!glusterd_is_local_addr (dst_brickinfo->hostname)) {
                         gf_log ("", GF_LOG_NORMAL,
                                 "I AM THE DESTINATION HOST");
@@ -2946,15 +3096,32 @@ glusterd_op_replace_brick (gd1_mgmt_stage_op_req *req, dict_t *rsp_dict)
                         }
                 }
 
+                ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+                if (ctx) {
+                        ret = rb_do_operation_abort (volinfo, src_brickinfo, dst_brickinfo);
+                        if (ret) {
+                                gf_log ("", GF_LOG_ERROR,
+                                        "Abort operation failed");
+                                goto out;
+                        }
+                }
+                glusterd_set_rb_status (volinfo, GF_RB_STATUS_NONE);
+                volinfo->src_brick = volinfo->dst_brick = NULL;
         }
-                break;
+        break;
 
         case GF_REPLACE_OP_STATUS:
         {
                 gf_log ("", GF_LOG_DEBUG,
                         "received status - doing nothing");
-                ret = rb_do_operation_status (volinfo, src_brickinfo,
-                                              dst_brickinfo);
+                ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+                if (ctx) {
+                        ret = rb_do_operation_status (volinfo, src_brickinfo,
+                                                      dst_brickinfo);
+                        if (ret)
+                                goto out;
+                }
+
         }
                 break;
 
@@ -4015,21 +4182,6 @@ out:
         return ret;
 }
 
-static gf_boolean_t
-rb_check_brick_signin (glusterd_brickinfo_t *brickinfo)
-{
-        gf_boolean_t value;
-
-        value = brickinfo->signed_in;
-
-        if (value == _gf_true) {
-                gf_log ("", GF_LOG_DEBUG,
-                        "Brick has signed in. Continuing...");
-        }
-
-        return value;
-}
-
 void
 glusterd_do_replace_brick (void *data)
 {
@@ -4041,7 +4193,6 @@ glusterd_do_replace_brick (void *data)
         char                   *src_brick = NULL;
         char                   *dst_brick = NULL;
         char                   *volname   = NULL;
-        gf_boolean_t            brick_signin = _gf_false;
         glusterd_brickinfo_t   *src_brickinfo = NULL;
         glusterd_brickinfo_t   *dst_brickinfo = NULL;
 	glusterd_conf_t	       *priv = NULL;
@@ -4128,24 +4279,23 @@ glusterd_do_replace_brick (void *data)
         ret = dict_get_int32 (dict, "dst-brick-port", &dst_port);
         if (ret) {
                 gf_log ("", GF_LOG_ERROR, "Unable to get dst-brick port");
-                goto out;
         }
 
         dst_brickinfo->port = dst_port;
         src_brickinfo->port = src_port;
 
-        brick_signin = rb_check_brick_signin (src_brickinfo);
-        if (brick_signin == _gf_false) {
-                gf_log ("", GF_LOG_DEBUG,
-                        "Marking replace brick to fail due to brick "
-                        "not having signed-in in 10secs");
-                ret = -1;
-                goto out;
-        }
-
         switch (op) {
         case GF_REPLACE_OP_START:
+                if (!dst_port) {
+                        ret = -1;
+                        goto out;
+                }
+
                 ret = rb_do_operation_start (volinfo, src_brickinfo, dst_brickinfo);
+                if (ret) {
+                        glusterd_set_rb_status (volinfo, GF_RB_STATUS_NONE);
+                        goto out;
+                }
                 break;
         case GF_REPLACE_OP_PAUSE:
         case GF_REPLACE_OP_ABORT:
@@ -4541,6 +4691,7 @@ glusterd_op_ac_stage_op (glusterd_op_sm_event_t *event, void *ctx)
         gd1_mgmt_stage_op_req   *req = NULL;
         glusterd_op_stage_ctx_t *stage_ctx = NULL;
         int32_t                 status = 0;
+        dict_t                  *rsp_dict  = NULL;
         char                    *op_errstr = NULL;
 
         GF_ASSERT (ctx);
@@ -4549,19 +4700,32 @@ glusterd_op_ac_stage_op (glusterd_op_sm_event_t *event, void *ctx)
 
         req = &stage_ctx->stage_req;
 
-        status = glusterd_op_stage_validate (req, &op_errstr);
+
+        rsp_dict = dict_new ();
+        if (!rsp_dict) {
+                gf_log ("", GF_LOG_DEBUG,
+                        "Out of memory");
+                return -1;
+        }
+
+        status = glusterd_op_stage_validate (req, &op_errstr,
+                                             rsp_dict);
 
         if (status) {
                 gf_log ("", GF_LOG_ERROR, "Validate failed: %d", status);
         }
 
-        ret = glusterd_op_stage_send_resp (stage_ctx->req, req->op, status, op_errstr);
+        ret = glusterd_op_stage_send_resp (stage_ctx->req, req->op,
+                                           status, op_errstr, rsp_dict);
 
         if (op_errstr && (strcmp (op_errstr, "")))
                 GF_FREE (op_errstr);
 
         gf_log ("", GF_LOG_DEBUG, "Returning with %d", ret);
 
+        if (rsp_dict)
+                dict_unref (rsp_dict);
+
         return ret;
 }
 
@@ -4630,7 +4794,8 @@ glusterd_op_sm_transition_state (glusterd_op_info_t *opinfo,
 }
 
 int32_t
-glusterd_op_stage_validate (gd1_mgmt_stage_op_req *req, char **op_errstr)
+glusterd_op_stage_validate (gd1_mgmt_stage_op_req *req, char **op_errstr,
+                            dict_t *rsp_dict)
 {
         int     ret = -1;
 
@@ -4658,7 +4823,8 @@ glusterd_op_stage_validate (gd1_mgmt_stage_op_req *req, char **op_errstr)
                         break;
 
                 case GD_OP_REPLACE_BRICK:
-                        ret = glusterd_op_stage_replace_brick (req, op_errstr);
+                        ret = glusterd_op_stage_replace_brick (req, op_errstr,
+                                                               rsp_dict);
                         break;
 
                 case GD_OP_SET_VOLUME:
diff --git a/xlators/mgmt/glusterd/src/glusterd-op-sm.h b/xlators/mgmt/glusterd/src/glusterd-op-sm.h
index 440154bd0..775cab07c 100644
--- a/xlators/mgmt/glusterd/src/glusterd-op-sm.h
+++ b/xlators/mgmt/glusterd/src/glusterd-op-sm.h
@@ -171,7 +171,8 @@ int
 glusterd_op_build_payload (glusterd_op_t op, gd1_mgmt_stage_op_req **req);
 
 int32_t
-glusterd_op_stage_validate (gd1_mgmt_stage_op_req *req, char **op_errstr);
+glusterd_op_stage_validate (gd1_mgmt_stage_op_req *req, char **op_errstr,
+                            dict_t *rsp_dict);
 
 int32_t
 glusterd_op_commit_perform (gd1_mgmt_stage_op_req *req, char **op_errstr,
diff --git a/xlators/mgmt/glusterd/src/glusterd-rebalance.c b/xlators/mgmt/glusterd/src/glusterd-rebalance.c
index f24ff85c9..934fc7030 100644
--- a/xlators/mgmt/glusterd/src/glusterd-rebalance.c
+++ b/xlators/mgmt/glusterd/src/glusterd-rebalance.c
@@ -408,6 +408,13 @@ glusterd_handle_defrag_volume (rpcsvc_request_t *req)
                         goto out;
                 }
 
+                if (glusterd_is_rb_started (volinfo) ||
+                    glusterd_is_rb_paused (volinfo)) {
+                        gf_log ("glusterd", GF_LOG_DEBUG,
+                                "Replace brick is in progress on volume %s",
+                                cli_req.volname);
+                        goto out;
+                }
                 volinfo->defrag = GF_CALLOC (1, sizeof (glusterd_defrag_info_t),
                                              gf_gld_mt_defrag_info);
                 if (!volinfo->defrag)
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index 11a19d2a2..cceecc11c 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -2337,3 +2337,53 @@ out:
         gf_log ("", GF_LOG_DEBUG, "returning %d ", ret);
         return ret;
 }
+
+inline int
+glusterd_is_rb_started(glusterd_volinfo_t *volinfo)
+{
+        gf_log ("", GF_LOG_DEBUG,
+                "is_rb_started:status=%d", volinfo->rb_status);
+        return (volinfo->rb_status == GF_RB_STATUS_STARTED);
+
+}
+
+inline int
+glusterd_is_rb_paused ( glusterd_volinfo_t *volinfo)
+{
+        gf_log ("", GF_LOG_DEBUG,
+                "is_rb_paused:status=%d", volinfo->rb_status);
+
+        return (volinfo->rb_status == GF_RB_STATUS_PAUSED);
+}
+
+inline int
+glusterd_set_rb_status (glusterd_volinfo_t *volinfo, gf_rb_status_t status)
+{
+        gf_log ("", GF_LOG_DEBUG,
+                "setting status from %d to %d",
+                volinfo->rb_status,
+                status);
+
+        volinfo->rb_status = status;
+        return 0;
+}
+
+inline int
+glusterd_rb_check_bricks (glusterd_volinfo_t *volinfo,
+                          glusterd_brickinfo_t *src, glusterd_brickinfo_t *dst)
+{
+        if (!volinfo->src_brick || !volinfo->dst_brick)
+                return -1;
+
+        if (strcmp (volinfo->src_brick->hostname, src->hostname) ||
+            strcmp (volinfo->src_brick->path, src->path)) {
+                gf_log("", GF_LOG_ERROR, "Replace brick src bricks differ");
+                return -1;
+        }
+        if (strcmp (volinfo->dst_brick->hostname, dst->hostname) ||
+            strcmp (volinfo->dst_brick->path, dst->path)) {
+                gf_log ("", GF_LOG_ERROR, "Replace brick dst bricks differ");
+                return -1;
+        }
+        return 0;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.h b/xlators/mgmt/glusterd/src/glusterd-utils.h
index 79a15c022..9db91fd58 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.h
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.h
@@ -215,4 +215,16 @@ glusterd_volume_brickinfo_get (uuid_t uuid, char *hostname, char *path,
 int
 glusterd_brickinfo_get (uuid_t uuid, char *hostname, char *path,
                        glusterd_brickinfo_t **brickinfo);
+int
+glusterd_is_rb_started (glusterd_volinfo_t *volinfo);
+
+int
+glusterd_is_rb_paused (glusterd_volinfo_t *volinfo);
+
+int
+glusterd_set_rb_status (glusterd_volinfo_t *volinfo, gf_rb_status_t status);
+
+int
+glusterd_rb_check_bricks (glusterd_volinfo_t *volinfo,
+                          glusterd_brickinfo_t *src_brick, glusterd_brickinfo_t *dst_brick);
 #endif
diff --git a/xlators/mgmt/glusterd/src/glusterd.h b/xlators/mgmt/glusterd/src/glusterd.h
index 7763962cd..9d7401f99 100644
--- a/xlators/mgmt/glusterd/src/glusterd.h
+++ b/xlators/mgmt/glusterd/src/glusterd.h
@@ -151,6 +151,13 @@ typedef enum gf_transport_type_ {
         GF_TRANSPORT_RDMA,
 } gf_transport_type;
 
+
+typedef enum gf_rb_status_ {
+        GF_RB_STATUS_NONE,
+        GF_RB_STATUS_STARTED,
+        GF_RB_STATUS_PAUSED,
+} gf_rb_status_t;
+
 struct glusterd_volinfo_ {
         char                    volname[GLUSTERD_MAX_VOLUME_NAME];
         int                     type;
@@ -169,6 +176,11 @@ struct glusterd_volinfo_ {
         uint64_t                lookedup_files;
         glusterd_defrag_info_t  *defrag;
 
+        /* Replace brick status */
+        gf_rb_status_t          rb_status;
+        glusterd_brickinfo_t    *src_brick;
+        glusterd_brickinfo_t    *dst_brick;
+
         int                     version;
         uint32_t                cksum;
         gf_transport_type       transport_type;
@@ -291,7 +303,8 @@ glusterd_op_unlock_send_resp (rpcsvc_request_t *req, int32_t status);
 
 int
 glusterd_op_stage_send_resp (rpcsvc_request_t *req,
-                             int32_t op, int32_t status, char *op_errstr);
+                             int32_t op, int32_t status,
+                             char *op_errstr, dict_t *rsp_dict);
 
 int
 glusterd_op_commmit_send_resp (rpcsvc_request_t *req,
diff --git a/xlators/mgmt/glusterd/src/glusterd3_1-mops.c b/xlators/mgmt/glusterd/src/glusterd3_1-mops.c
index cc2f84d38..b7ffd0dd6 100644
--- a/xlators/mgmt/glusterd/src/glusterd3_1-mops.c
+++ b/xlators/mgmt/glusterd/src/glusterd3_1-mops.c
@@ -496,6 +496,63 @@ out:
         return ret;
 }
 
+static int32_t
+glusterd_rb_use_rsp_dict (dict_t *rsp_dict)
+{
+        int32_t  src_port = 0;
+        int32_t  dst_port = 0;
+        int      ret      = 0;
+        dict_t  *ctx      = NULL;
+
+
+        ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
+        if (!ctx) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Operation Context is not present");
+                GF_ASSERT (0);
+        }
+
+        if (rsp_dict) {
+                ret = dict_get_int32 (rsp_dict, "src-brick-port", &src_port);
+                if (ret == 0) {
+                        gf_log ("", GF_LOG_DEBUG,
+                                "src-brick-port=%d found", src_port);
+                }
+
+                ret = dict_get_int32 (rsp_dict, "dst-brick-port", &dst_port);
+                if (ret == 0) {
+                        gf_log ("", GF_LOG_DEBUG,
+                                "dst-brick-port=%d found", dst_port);
+                }
+
+        }
+
+        if (src_port) {
+                ret = dict_set_int32 (ctx, "src-brick-port",
+                                      src_port);
+                if (ret) {
+                        gf_log ("", GF_LOG_DEBUG,
+                                "Could not set src-brick");
+                        goto out;
+                }
+        }
+
+        if (dst_port) {
+                ret = dict_set_int32 (ctx, "dst-brick-port",
+                                      dst_port);
+                if (ret) {
+                        gf_log ("", GF_LOG_DEBUG,
+                                "Could not set dst-brick");
+                        goto out;
+                }
+
+        }
+
+out:
+        return ret;
+
+}
+
 int32_t
 glusterd3_1_stage_op_cbk (struct rpc_req *req, struct iovec *iov,
                           int count, void *myframe)
@@ -506,6 +563,7 @@ glusterd3_1_stage_op_cbk (struct rpc_req *req, struct iovec *iov,
         glusterd_op_sm_event_type_t   event_type = GD_OP_EVENT_NONE;
         glusterd_peerinfo_t           *peerinfo = NULL;
         char                          str[50] = {0,};
+        dict_t                        *dict   = NULL;
 
         GF_ASSERT (req);
 
@@ -526,6 +584,24 @@ glusterd3_1_stage_op_cbk (struct rpc_req *req, struct iovec *iov,
         }
         uuid_unparse (rsp.uuid, str);
 
+        if (rsp.dict.dict_len) {
+                /* Unserialize the dictionary */
+                dict  = dict_new ();
+
+                ret = dict_unserialize (rsp.dict.dict_val,
+                                        rsp.dict.dict_len,
+                                        &dict);
+                if (ret < 0) {
+                        gf_log ("glusterd", GF_LOG_ERROR,
+                                "failed to "
+                                "unserialize rsp-buffer to dictionary");
+			event_type = GD_OP_EVENT_RCVD_RJT;
+                        goto out;
+                } else {
+                        dict->extra_stdfree = rsp.dict.dict_val;
+                }
+        }
+
         op_ret = rsp.op_ret;
 
         gf_log ("glusterd", GF_LOG_NORMAL,
@@ -552,6 +628,12 @@ glusterd3_1_stage_op_cbk (struct rpc_req *req, struct iovec *iov,
                 event_type = GD_OP_EVENT_RCVD_ACC;
         }
 
+        switch (rsp.op) {
+        case GD_OP_REPLACE_BRICK:
+                glusterd_rb_use_rsp_dict (dict);
+                break;
+        }
+
         ret = glusterd_op_sm_inject_event (event_type, NULL);
 
         if (!ret) {
@@ -583,47 +665,6 @@ out:
 
 }
 
-static int32_t
-glusterd_rb_use_rsp_dict (dict_t *rsp_dict)
-{
-        int32_t  src_port = 0;
-        int      ret      = 0;
-        dict_t  *ctx      = NULL;
-
-
-        if (rsp_dict) {
-                ret = dict_get_int32 (rsp_dict, "src-brick-port", &src_port);
-                if (ret) {
-                        gf_log ("", GF_LOG_DEBUG,
-                                "src-brick-port not present");
-                        ret = 0;
-                        goto out;
-                }
-
-                if (src_port) {
-                        ctx = glusterd_op_get_ctx (GD_OP_REPLACE_BRICK);
-                        if (!ctx) {
-                                gf_log ("", GF_LOG_ERROR,
-                                        "Operation Context is not present");
-                                ret = 0;
-                                goto out;
-                        }
-
-                        ret = dict_set_int32 (ctx, "src-brick-port",
-                                              src_port);
-                        if (ret) {
-                                gf_log ("", GF_LOG_DEBUG,
-                                        "Could not set src-brick");
-                                goto out;
-                        }
-                }
-        }
-
-out:
-        return ret;
-
-}
-
 int32_t
 glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                           int count, void *myframe)
@@ -643,6 +684,7 @@ glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                 rsp.op_ret   = -1;
                 rsp.op_errno = EINVAL;
                 rsp.op_errstr = "error";
+		event_type = GD_OP_EVENT_RCVD_RJT;
                 goto out;
         }
 
@@ -652,6 +694,7 @@ glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                 rsp.op_ret   = -1;
                 rsp.op_errno = EINVAL;
                 rsp.op_errstr = "error";
+		event_type = GD_OP_EVENT_RCVD_RJT;
                 goto out;
         }
         uuid_unparse (rsp.uuid, str);
@@ -667,6 +710,7 @@ glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                         gf_log ("glusterd", GF_LOG_ERROR,
                                 "failed to "
                                 "unserialize rsp-buffer to dictionary");
+			event_type = GD_OP_EVENT_RCVD_RJT;
                         goto out;
                 } else {
                         dict->extra_stdfree = rsp.dict.dict_val;
@@ -696,6 +740,7 @@ glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                         goto out;
                 }
         } else {
+                event_type = GD_OP_EVENT_RCVD_ACC;
                 switch (rsp.op) {
                 case GD_OP_REPLACE_BRICK:
                         ret = glusterd_rb_use_rsp_dict (dict);
@@ -710,9 +755,9 @@ glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                 default:
                         break;
                 }
-                event_type = GD_OP_EVENT_RCVD_ACC;
         }
 
+out:
         ret = glusterd_op_sm_inject_event (event_type, NULL);
 
         if (!ret) {
@@ -720,7 +765,6 @@ glusterd3_1_commit_op_cbk (struct rpc_req *req, struct iovec *iov,
                 glusterd_op_sm ();
         }
 
-out:
         if (dict)
                 dict_unref (dict);
         if (rsp.op_errstr && strcmp (rsp.op_errstr, "error"))
@@ -1158,7 +1202,8 @@ glusterd3_1_stage_op (call_frame_t *frame, xlator_t *this,
         if (ret)
                 goto out;
 
-        ret = glusterd_op_stage_validate (req, &op_errstr);
+        /* rsp_dict NULL from source */
+        ret = glusterd_op_stage_validate (req, &op_errstr, NULL);
 
         if (ret) {
                 gf_log ("", GF_LOG_ERROR, "Staging failed");