cli/glusterd: add/remove brick fixes for arbiter volumes

1.Provide a command to convert replica 2 volumes to arbiter volumes.
Existing self-heal logic will automatically heal the file hierarchy into
the arbiter brick, the progress of which can be monitored using the
heal info command.

Syntax: gluster volume add-brick <VOLNAME> replica 3 arbiter 1
<HOST:arbiter-brick-path>

2. Add checks when removing bricks from arbiter volumes:
- When converting from arbiter to replica 2 volume, allow only arbiter
  brick to be removed.
- When converting from arbiter to plain distribute volume, allow only if
  arbiter is one of the bricks that is removed.

3. Some clean-up:
- Use GD_MSG_DICT_GET_SUCCESS instead of GD_MSG_DICT_GET_FAILED to
log messages that are not failures.
- Remove unused variable `brick_list`
- Move 'brickinfo->group' related functions to glusted-utils.

Change-Id: Ic87b8c7e4d7d3ab03f93e7b9f372b314d80947ce
BUG: 1318289
Signed-off-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-on: http://review.gluster.org/14126
Smoke: Gluster Build System <jenkins@build.gluster.com>
CentOS-regression: Gluster Build System <jenkins@build.gluster.com>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
Reviewed-by: Atin Mukherjee <amukherj@redhat.com>

7 files changed, 204 insertions, 66 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
index 5b2f559b546..a90114ab2b3 100644
--- a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
+++ b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
@@ -219,8 +219,8 @@ out:
 
 static int
 gd_addbr_validate_replica_count (glusterd_volinfo_t *volinfo, int replica_count,
-                                 int total_bricks, int *type, char *err_str,
-                                 int err_len)
+                                 int arbiter_count, int total_bricks, int *type,
+                                 char *err_str, int err_len)
 {
         int ret = -1;
 
@@ -283,6 +283,14 @@ gd_addbr_validate_replica_count (glusterd_volinfo_t *volinfo, int replica_count,
                         goto out;
                 }
                 if (replica_count == volinfo->replica_count) {
+                        if (arbiter_count && !volinfo->arbiter_count) {
+                                snprintf (err_str, err_len,
+                                          "Cannot convert replica 3 volume "
+                                          "to arbiter volume.");
+                                gf_msg (THIS->name, GF_LOG_ERROR, EINVAL,
+                                        GD_MSG_INVALID_ENTRY, "%s", err_str);
+                                goto out;
+                        }
                         if (!(total_bricks % volinfo->dist_leaf_count)) {
                                 ret = 1;
                                 goto out;
@@ -413,6 +421,7 @@ __glusterd_handle_add_brick (rpcsvc_request_t *req)
         xlator_t                        *this = NULL;
         int                             total_bricks = 0;
         int32_t                         replica_count = 0;
+        int32_t                         arbiter_count = 0;
         int32_t                         stripe_count = 0;
         int                             type = 0;
         glusterd_conf_t                 *conf = NULL;
@@ -486,14 +495,21 @@ __glusterd_handle_add_brick (rpcsvc_request_t *req)
         ret = dict_get_int32 (dict, "replica-count", &replica_count);
         if (!ret) {
                 gf_msg (this->name, GF_LOG_INFO, errno,
-                        GD_MSG_DICT_GET_FAILED, "replica-count is %d",
+                        GD_MSG_DICT_GET_SUCCESS, "replica-count is %d",
                         replica_count);
         }
 
+        ret = dict_get_int32 (dict, "arbiter-count", &arbiter_count);
+        if (!ret) {
+                gf_msg (this->name, GF_LOG_INFO, errno,
+                        GD_MSG_DICT_GET_SUCCESS, "arbiter-count is %d",
+                        arbiter_count);
+        }
+
         ret = dict_get_int32 (dict, "stripe-count", &stripe_count);
         if (!ret) {
                 gf_msg (this->name, GF_LOG_INFO, errno,
-                        GD_MSG_DICT_GET_FAILED, "stripe-count is %d",
+                        GD_MSG_DICT_GET_SUCCESS, "stripe-count is %d",
                         stripe_count);
         }
 
@@ -602,7 +618,7 @@ __glusterd_handle_add_brick (rpcsvc_request_t *req)
         }
 
         ret = gd_addbr_validate_replica_count (volinfo, replica_count,
-                                               total_bricks,
+                                               arbiter_count, total_bricks,
                                                &type, err_str,
                                                sizeof (err_str));
         if (ret == -1) {
@@ -791,6 +807,71 @@ glusterd_set_detach_bricks(dict_t *dict, glusterd_volinfo_t *volinfo)
         return hot_brick_num;
 }
 
+static int
+glusterd_remove_brick_validate_arbiters (glusterd_volinfo_t *volinfo,
+                                         int32_t count, int32_t replica_count,
+                                         glusterd_brickinfo_t **brickinfo_list,
+                                         char *err_str, size_t err_len)
+{
+        int i = 0;
+        int ret = 0;
+        glusterd_brickinfo_t *brickinfo = NULL;
+        glusterd_brickinfo_t *last = NULL;
+        char *arbiter_array = NULL;
+
+        if ((volinfo->type != GF_CLUSTER_TYPE_REPLICATE) &&
+            (volinfo->type != GF_CLUSTER_TYPE_STRIPE_REPLICATE))
+                goto out;
+
+        if (!replica_count || !volinfo->arbiter_count)
+                goto out;
+
+        if (replica_count == 2) {
+                /* If it is an arbiter to replica 2 conversion, only permit
+                *  removal of the arbiter brick.*/
+                for (i = 0; i < count; i++) {
+                        brickinfo = brickinfo_list[i];
+                        last = get_last_brick_of_brick_group (volinfo,
+                                                              brickinfo);
+                        if (last != brickinfo) {
+                                snprintf (err_str, err_len, "Remove arbiter "
+                                          "brick(s) only when converting from "
+                                           "arbiter to replica 2 subvolume.");
+                                ret = -1;
+                                goto out;
+                        }
+                }
+        } else if (replica_count == 1) {
+                /* If it is an arbiter to plain distribute conversion, in every
+                 * replica subvol, the arbiter has to be one of the bricks that
+                 * are removed. */
+                arbiter_array = GF_CALLOC (volinfo->subvol_count,
+                                           sizeof (*arbiter_array),
+                                           gf_common_mt_char);
+                if (!arbiter_array)
+                        return -1;
+                for (i = 0; i < count; i++) {
+                        brickinfo = brickinfo_list[i];
+                        last = get_last_brick_of_brick_group (volinfo,
+                                                              brickinfo);
+                        if (last == brickinfo)
+                                arbiter_array[brickinfo->group] = 1;
+                }
+                for (i = 0; i < volinfo->subvol_count; i++)
+                        if (!arbiter_array[i]) {
+                                snprintf (err_str, err_len, "Removed bricks "
+                                          "must contain arbiter when converting"
+                                           " to plain distrubute.");
+                                ret = -1;
+                                break;
+                        }
+                GF_FREE (arbiter_array);
+        }
+
+out:
+        return ret;
+}
+
 int
 __glusterd_handle_remove_brick (rpcsvc_request_t *req)
 {
@@ -800,10 +881,10 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
         int32_t                   count            = 0;
         char                     *brick            = NULL;
         char                      key[256]         = {0,};
-        char                     *brick_list       = NULL;
         int                       i                = 1;
         glusterd_volinfo_t       *volinfo          = NULL;
         glusterd_brickinfo_t     *brickinfo        = NULL;
+        glusterd_brickinfo_t     **brickinfo_list  = NULL;
         int                      *subvols          = NULL;
         char                      err_str[2048]    = {0};
         gf_cli_rsp                rsp              = {0,};
@@ -998,16 +1079,6 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
                 }
         }
 
-        brick_list = GF_MALLOC (120000 * sizeof(*brick_list),gf_common_mt_char);
-
-        if (!brick_list) {
-                ret = -1;
-                goto out;
-        }
-
-
-        strcpy (brick_list, " ");
-
         /* subvol match is not required for tiered volume*/
         if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
              (volinfo->type != GF_CLUSTER_TYPE_TIER) &&
@@ -1020,6 +1091,13 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
         if (volinfo->type == GF_CLUSTER_TYPE_TIER)
                 count = glusterd_set_detach_bricks(dict, volinfo);
 
+        brickinfo_list = GF_CALLOC (count, sizeof (*brickinfo_list),
+                                    gf_common_mt_pointer);
+        if (!brickinfo_list) {
+                ret = -1;
+                goto out;
+        }
+
         while ( i <= count) {
                 snprintf (key, sizeof (key), "brick%d", i);
                 ret = dict_get_str (dict, key, &brick);
@@ -1044,8 +1122,7 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
                                 GD_MSG_BRICK_NOT_FOUND, "%s", err_str);
                         goto out;
                 }
-                strcat(brick_list, brick);
-                strcat(brick_list, " ");
+                brickinfo_list[i-1] = brickinfo;
 
                 i++;
                 if ((volinfo->type == GF_CLUSTER_TYPE_NONE) ||
@@ -1072,6 +1149,14 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
                         goto out;
         }
 
+        ret = glusterd_remove_brick_validate_arbiters (volinfo, count,
+                                                       replica_count,
+                                                       brickinfo_list,
+                                                       err_str,
+                                                       sizeof (err_str));
+        if (ret)
+                goto out;
+
         ret = glusterd_op_begin_synctask (req, GD_OP_REMOVE_BRICK, dict);
 
 out:
@@ -1092,8 +1177,8 @@ out:
 
         }
 
-        if (brick_list)
-                GF_FREE (brick_list);
+        if (brickinfo_list)
+                GF_FREE (brickinfo_list);
         subvol_matcher_destroy (subvols);
         free (cli_req.dict.dict_val); //its malloced by xdr
 
@@ -1224,6 +1309,7 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
         int32_t                       ret            = -1;
         int32_t                       stripe_count   = 0;
         int32_t                       replica_count  = 0;
+        int32_t                       arbiter_count  = 0;
         int32_t                       type           = 0;
         glusterd_brickinfo_t         *brickinfo      = NULL;
         glusterd_gsync_status_temp_t  param          = {0, };
@@ -1256,18 +1342,23 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
                 ret = dict_get_int32 (dict, "stripe-count", &stripe_count);
                 if (!ret)
                         gf_msg (THIS->name, GF_LOG_INFO, errno,
-                                GD_MSG_DICT_GET_FAILED,
+                                GD_MSG_DICT_GET_SUCCESS,
                                 "stripe-count is set %d", stripe_count);
 
                 ret = dict_get_int32 (dict, "replica-count", &replica_count);
                 if (!ret)
                         gf_msg (THIS->name, GF_LOG_INFO, errno,
-                                GD_MSG_DICT_GET_FAILED,
+                                GD_MSG_DICT_GET_SUCCESS,
                                 "replica-count is set %d", replica_count);
+                ret = dict_get_int32 (dict, "arbiter-count", &arbiter_count);
+                if (!ret)
+                        gf_msg (THIS->name, GF_LOG_INFO, errno,
+                                GD_MSG_DICT_GET_SUCCESS,
+                                "arbiter-count is set %d", arbiter_count);
                 ret = dict_get_int32 (dict, "type", &type);
                 if (!ret)
                         gf_msg (THIS->name, GF_LOG_INFO, errno,
-                                GD_MSG_DICT_GET_FAILED,
+                                GD_MSG_DICT_GET_SUCCESS,
                                 "type is set %d, need to change it", type);
         }
 
@@ -1328,6 +1419,9 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
         if (replica_count) {
                 volinfo->replica_count = replica_count;
         }
+        if (arbiter_count) {
+                volinfo->arbiter_count = arbiter_count;
+        }
         if (stripe_count) {
                 volinfo->stripe_count = stripe_count;
         }
@@ -1529,6 +1623,7 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
         char                                    *volname = NULL;
         int                                     count = 0;
         int                                     replica_count = 0;
+        int                                     arbiter_count = 0;
         int                                     i = 0;
         int32_t                                 local_brick_count = 0;
         char                                    *bricks    = NULL;
@@ -1578,6 +1673,12 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
                         "Unable to get replica count");
         }
 
+        ret = dict_get_int32 (dict, "arbiter-count", &arbiter_count);
+        if (ret) {
+                gf_msg_debug (THIS->name, 0,
+                        "No arbiter count present in the dict");
+        }
+
         if (replica_count > 0) {
                 ret = op_version_check (this, GD_OP_VER_PERSISTENT_AFR_XATTRS,
                                         msg, sizeof(msg));
@@ -1589,10 +1690,10 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
                 }
         }
 
-        /* Do not allow add-brick for stopped volumes when replica-count
-         * is being increased.
-         */
         if (glusterd_is_volume_replicate (volinfo)) {
+                /* Do not allow add-brick for stopped volumes when replica-count
+                 * is being increased.
+                 */
                 if (conf->op_version >= GD_OP_VERSION_3_7_10 &&
                     !dict_get (dict, "attach-tier") &&
                     replica_count &&
@@ -1606,6 +1707,20 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
                         *op_errstr = gf_strdup (msg);
                         goto out;
                 }
+                /* op-version check for replica 2 to arbiter conversion. If we
+                * dont have this check, an older peer added as arbiter brick
+                * will not have the  arbiter xlator in its volfile. */
+                if ((conf->op_version < GD_OP_VERSION_3_8_0) &&
+                    (arbiter_count == 1) && (replica_count == 3)) {
+                        ret = -1;
+                        snprintf (msg, sizeof (msg), "Cluster op-version must "
+                                  "be >= 30800 to add arbiter brick to a "
+                                  "replica 2 volume.");
+                        gf_msg (THIS->name, GF_LOG_ERROR, 0,
+                                GD_MSG_BRICK_ADD_FAIL, "%s", msg);
+                        *op_errstr = gf_strdup (msg);
+                        goto out;
+                }
         }
 
         if (conf->op_version > GD_OP_VERSION_3_7_5 &&
@@ -2689,6 +2804,10 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                         volinfo->replica_count, replica_count,
                         volinfo->volname);
                 volinfo->replica_count = replica_count;
+                /* A reduction in replica count implies an arbiter volume
+                 * earlier is now no longer one. */
+                if (volinfo->arbiter_count)
+                        volinfo->arbiter_count = 0;
                 volinfo->sub_count = replica_count;
                 volinfo->dist_leaf_count = glusterd_get_dist_leaf_count (volinfo);
 
diff --git a/xlators/mgmt/glusterd/src/glusterd-messages.h b/xlators/mgmt/glusterd/src/glusterd-messages.h
index 9fa28611f19..ba40b8f7628 100644
--- a/xlators/mgmt/glusterd/src/glusterd-messages.h
+++ b/xlators/mgmt/glusterd/src/glusterd-messages.h
@@ -41,7 +41,7 @@
 
 #define GLUSTERD_COMP_BASE      GLFS_MSGID_GLUSTERD
 
-#define GLFS_NUM_MESSAGES       577
+#define GLFS_NUM_MESSAGES       578
 
 #define GLFS_MSGID_END          (GLUSTERD_COMP_BASE + GLFS_NUM_MESSAGES + 1)
 /* Messaged with message IDs */
@@ -4647,7 +4647,6 @@
  *             on its own and try to restart the brick with a new port
  * @recommendedaction  Ensure the new port is not blocked by firewall
  */
-
 #define GD_MSG_RETRY_WITH_NEW_PORT                   (GLUSTERD_COMP_BASE + 575)
 
 /*!
@@ -4666,6 +4665,14 @@
  */
 #define GD_MSG_SLAVE_VOL_PARSE_FAIL               (GLUSTERD_COMP_BASE + 577)
 
+/*!
+ * @messageid
+ * @diagnosis
+ * @recommendedaction
+ *
+ */
+#define GD_MSG_DICT_GET_SUCCESS                   (GLUSTERD_COMP_BASE + 578)
+
 /*------------*/
 #define glfs_msg_end_x GLFS_MSGID_END, "Invalid: End of messages"
 #endif /* !_GLUSTERD_MESSAGES_H_ */
diff --git a/xlators/mgmt/glusterd/src/glusterd-store.c b/xlators/mgmt/glusterd/src/glusterd-store.c
index 113902d2e6a..aa2be17bd9a 100644
--- a/xlators/mgmt/glusterd/src/glusterd-store.c
+++ b/xlators/mgmt/glusterd/src/glusterd-store.c
@@ -2444,6 +2444,7 @@ glusterd_store_retrieve_bricks (glusterd_volinfo_t *volinfo)
                 brick_count++;
         }
 
+        assign_brick_groups (volinfo);
         ret = gf_store_iter_destroy (tmpiter);
         if (ret)
                 goto out;
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index 89c6d2dfa18..19173a4d015 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -11418,3 +11418,40 @@ out:
         gf_msg_debug ("glusterd", 0, "Returning with ret");
         return ret;
 }
+
+void
+assign_brick_groups (glusterd_volinfo_t *volinfo)
+{
+        glusterd_brickinfo_t    *brickinfo      = NULL;
+        uint16_t                group_num       = 0;
+        int                     in_group        = 0;
+
+        list_for_each_entry (brickinfo, &volinfo->bricks, brick_list) {
+                brickinfo->group = group_num;
+                if (++in_group >= volinfo->replica_count) {
+                        in_group = 0;
+                        ++group_num;
+                }
+        }
+}
+
+glusterd_brickinfo_t*
+get_last_brick_of_brick_group (glusterd_volinfo_t *volinfo,
+                               glusterd_brickinfo_t *brickinfo)
+{
+        glusterd_brickinfo_t  *next = NULL;
+        glusterd_brickinfo_t  *last = NULL;
+        int ret = -1;
+
+        last = brickinfo;
+        for (;;) {
+                next = list_next (last, &volinfo->bricks,
+                                  glusterd_brickinfo_t, brick_list);
+                if (!next || (next->group != brickinfo->group)) {
+                        break;
+                }
+                last = next;
+        }
+
+        return last;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.h b/xlators/mgmt/glusterd/src/glusterd-utils.h
index e6436eff419..5f4bb6d251b 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.h
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.h
@@ -708,4 +708,10 @@ int
 glusterd_handle_replicate_brick_ops (glusterd_volinfo_t *volinfo,
                                      glusterd_brickinfo_t *brickinfo,
                                      glusterd_op_t op);
+void
+assign_brick_groups (glusterd_volinfo_t *volinfo);
+
+glusterd_brickinfo_t*
+get_last_brick_of_brick_group (glusterd_volinfo_t *volinfo,
+                               glusterd_brickinfo_t *brickinfo);
 #endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index 4321ddb7ddb..231117750f6 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -1547,16 +1547,8 @@ brick_graph_add_arbiter (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
         if (volinfo->arbiter_count != 1)
                 return 0;
 
-        /* Find the last brick in the same group. */
-        last = brickinfo;
-        for (;;) {
-                next = list_next (last, &volinfo->bricks,
-                                  glusterd_brickinfo_t, brick_list);
-                if (!next || (next->group != brickinfo->group)) {
-                        break;
-                }
-                last = next;
-        }
+        /* Add arbiter only if it is the last (i.e. 3rd) brick. */
+        last = get_last_brick_of_brick_group (volinfo, brickinfo);
         if (last != brickinfo)
                 return 0;
 
@@ -1623,22 +1615,6 @@ out:
         return ret;
 }
 
-void
-assign_brick_groups (glusterd_volinfo_t *volinfo)
-{
-        glusterd_brickinfo_t    *brickinfo      = NULL;
-        uint16_t                group_num       = 0;
-        int                     in_group        = 0;
-
-        list_for_each_entry (brickinfo, &volinfo->bricks, brick_list) {
-                brickinfo->group = group_num;
-                if (++in_group >= volinfo->replica_count) {
-                        in_group = 0;
-                        ++group_num;
-                }
-        }
-}
-
 static int
 brick_graph_add_changelog (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
                             dict_t *set_dict, glusterd_brickinfo_t *brickinfo)
@@ -5304,23 +5280,18 @@ get_parent_vol_tstamp_file (char *filename, glusterd_volinfo_t *volinfo)
 }
 
 void
-assign_groups (glusterd_volinfo_t *volinfo)
+assign_jbr_uuids (glusterd_volinfo_t *volinfo)
 {
         glusterd_brickinfo_t    *brickinfo      = NULL;
-        uint16_t                group_num       = 0;
         int                     in_group        = 0;
         uuid_t                  tmp_uuid;
 
         list_for_each_entry (brickinfo, &volinfo->bricks, brick_list) {
-                if (in_group == 0) {
+                if (in_group == 0)
                         gf_uuid_generate(tmp_uuid);
-                }
-                brickinfo->group = group_num;
                 gf_uuid_copy(brickinfo->jbr_uuid, tmp_uuid);
-                if (++in_group >= volinfo->replica_count) {
+                if (++in_group >= volinfo->replica_count)
                         in_group = 0;
-                        ++group_num;
-                }
         }
 }
 
@@ -5393,7 +5364,7 @@ generate_brick_volfiles (glusterd_volinfo_t *volinfo)
         }
 
         if (glusterd_volinfo_get_boolean(volinfo, "cluster.jbr") > 0) {
-                assign_groups(volinfo);
+                assign_jbr_uuids(volinfo);
         }
 
         ret = glusterd_volume_brick_for_each (volinfo, NULL,
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.h b/xlators/mgmt/glusterd/src/glusterd-volgen.h
index df6d9e5da7e..d8dd70bc33b 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.h
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.h
@@ -230,9 +230,6 @@ glusterd_check_voloption_flags (char *key, int32_t flags);
 gf_boolean_t
 glusterd_is_valid_volfpath (char *volname, char *brick);
 
-void
-assign_brick_groups (glusterd_volinfo_t *volinfo);
-
 int
 generate_brick_volfiles (glusterd_volinfo_t *volinfo);