glusterd: Always copy old brick ports when importing

  Backport of ecf6243 from master.
When an updated volinfo is imported in, the brick ports from the old
volinfo should be always copied.

Earlier, this was being done only if the old volinfo was stopped and
new volinfo was started. This could lead to brick ports chaging when the
following sequence of steps happened.
- A volume is stopped
- GlusterD is stopped on a peer
- The stopped volume is started
- The stopped GlusterD is started
This sequence would lead to bricks on the peer with re-started GlusterD
to get new ports, which could break firewall rules and could prevent
client access. This sequence could be hit when enabling management
encryption in a Gluster trusted storage pool.

Change-Id: I808ad478038d12ed2b19752511bdd7aa6f663bfc
BUG: 1316391
Reviewed-originally-on: http://review.gluster.org/13578
Signed-off-by: Kaushal M <kaushal@redhat.com>
Reviewed-on: http://review.gluster.org/13664
Smoke: Gluster Build System <jenkins@build.gluster.com>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Atin Mukherjee <amukherj@redhat.com>

1 files changed, 15 insertions, 20 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index 73cdf16badf..cfee26dfe7e 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -3735,8 +3735,8 @@ glusterd_volume_disconnect_all_bricks (glusterd_volinfo_t *volinfo)
 }
 
 int32_t
-glusterd_volinfo_copy_brick_portinfo (glusterd_volinfo_t *new_volinfo,
-                                      glusterd_volinfo_t *old_volinfo)
+glusterd_volinfo_copy_brick_portinfo (glusterd_volinfo_t *old_volinfo,
+                                      glusterd_volinfo_t *new_volinfo)
 {
         char                    pidfile[PATH_MAX+1] = {0,};
         glusterd_brickinfo_t   *new_brickinfo       = NULL;
@@ -3752,9 +3752,6 @@ glusterd_volinfo_copy_brick_portinfo (glusterd_volinfo_t *new_volinfo,
         priv = this->private;
         GF_ASSERT (priv);
 
-        if (_gf_false == glusterd_is_volume_started (new_volinfo))
-                goto out;
-
         cds_list_for_each_entry (new_brickinfo, &new_volinfo->bricks,
                                  brick_list) {
                 ret = glusterd_volume_brickinfo_get (new_brickinfo->uuid,
@@ -3763,15 +3760,11 @@ glusterd_volinfo_copy_brick_portinfo (glusterd_volinfo_t *new_volinfo,
                                                      old_volinfo,
                                                      &old_brickinfo);
                 if (ret == 0) {
-                        GLUSTERD_GET_BRICK_PIDFILE (pidfile, old_volinfo,
-                                                    old_brickinfo, priv);
-                        if (gf_is_service_running (pidfile, NULL))
-                                new_brickinfo->port = old_brickinfo->port;
-
+                        new_brickinfo->port = old_brickinfo->port;
                 }
         }
-out:
         ret = 0;
+
         return ret;
 }
 
@@ -3854,21 +3847,16 @@ glusterd_delete_stale_volume (glusterd_volinfo_t *stale_volinfo,
                 }
         }
 
-        /* If stale volume is in started state, copy the port numbers of the
-         * local bricks if they exist in the valid volume information.
-         * stop stale bricks. Stale volume information is going to be deleted.
-         * Which deletes the valid brick information inside stale volinfo.
-         * We dont want brick_rpc_notify to access already deleted brickinfo.
-         * Disconnect all bricks from stale_volinfo (unconditionally), since
+        /* If stale volume is in started state, stop the stale bricks if the new
+         * volume is started else, stop all bricks.
+         * We dont want brick_rpc_notify to access already deleted brickinfo,
+         * so disconnect all bricks from stale_volinfo (unconditionally), since
          * they are being deleted subsequently.
          */
         if (glusterd_is_volume_started (stale_volinfo)) {
                 if (glusterd_is_volume_started (valid_volinfo)) {
                         (void) glusterd_volinfo_stop_stale_bricks (valid_volinfo,
                                                                    stale_volinfo);
-                        //Only valid bricks will be running now.
-                        (void) glusterd_volinfo_copy_brick_portinfo (valid_volinfo,
-                                                                     stale_volinfo);
 
                 } else {
                         (void) glusterd_stop_bricks (stale_volinfo);
@@ -3982,6 +3970,13 @@ glusterd_import_friend_volume (dict_t *peer_data, size_t count)
         if (0 == ret) {
                 (void) gd_check_and_update_rebalance_info (old_volinfo,
                                                            new_volinfo);
+
+                /* Copy brick ports from the old volinfo always. The old_volinfo
+                 * will be cleaned up and this information could be lost
+                 */
+                (void) glusterd_volinfo_copy_brick_portinfo (old_volinfo,
+                                                             new_volinfo);
+
                 (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo);
         }