glusterd/afr: store afr pending xattrs as a volume option

Backport of http://review.gluster.org/#/c/12738/

Problem:
When AFR xlator initialises, it uses the name of the client xlators
below it for storing the pending changelogs (xattrs). This can be
problem when some other xlator is loaded in between AFR and the client.
Though that is a trivial 'traverse-graph-till-the-client-and-use-the-name'
fix in AFR's init(), there are other issues like when there's no client
xlator at all when, say, AFR is moved to the server side.

Fix:
The client xlator names are currenly unique and stored as
brickinfo->brick_ids. So persist these ids as comma separated values in
AFR's volume_options and use them as xattr values during init().

Change-Id: Ie761ffeb3373a4c4d85ad05c84a768c4188aa90d
BUG: 1291985
Signed-off-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-on: http://review.gluster.org/12977
Tested-by: NetBSD Build System <jenkins@build.gluster.org>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>

4 files changed, 122 insertions, 16 deletions

diff --git a/tests/bugs/glusterfs/bug-853690.t b/tests/bugs/glusterfs/bug-853690.t
index 59facfcddb0..7880b64488f 100755
--- a/tests/bugs/glusterfs/bug-853690.t
+++ b/tests/bugs/glusterfs/bug-853690.t
@@ -53,6 +53,7 @@ end-volume
 
 volume test-replicate-0
     type cluster/replicate
+    option afr-pending-xattr test-locks-0,test-locks-1
     option background-self-heal-count 0
     subvolumes test-locks-0 test-locks-1
 end-volume
diff --git a/tests/bugs/glusterfs/bug-892730.t b/tests/bugs/glusterfs/bug-892730.t
index a76961134c5..1fa0ff3bfb4 100755
--- a/tests/bugs/glusterfs/bug-892730.t
+++ b/tests/bugs/glusterfs/bug-892730.t
@@ -53,6 +53,7 @@ end-volume
 
 volume test-replicate-0
     type cluster/replicate
+    option afr-pending-xattr test-locks-0,test-locks-1
     option background-self-heal-count 0
     subvolumes test-locks-0 test-locks-1
 end-volume
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index 5ef920a13d1..4c072b6c69f 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -248,6 +248,8 @@ init (xlator_t *this)
         int            read_subvol_index = -1;
         xlator_t      *fav_child   = NULL;
         char          *qtype       = NULL;
+        char          *xattrs_list = NULL;
+        char          *ptr         = NULL;
 
         if (!this->children) {
                 gf_msg (this->name, GF_LOG_ERROR, 0,
@@ -401,6 +403,7 @@ init (xlator_t *this)
                 goto out;
         }
 
+        GF_OPTION_INIT ("afr-pending-xattr", xattrs_list, str, out);
         priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
                                        child_count,
                                        gf_afr_mt_char);
@@ -408,20 +411,25 @@ init (xlator_t *this)
                 ret = -ENOMEM;
                 goto out;
         }
-
-        trav = this->children;
-        i = 0;
-        while (i < child_count) {
-                priv->children[i] = trav->xlator;
-
+        ptr = gf_strdup (xattrs_list);
+        if (!ptr) {
+                ret = -ENOMEM;
+                goto out;
+        }
+        for (i = 0, ptr = strtok (ptr, ","); ptr; ptr = strtok (NULL, ",")) {
                 ret = gf_asprintf (&priv->pending_key[i], "%s.%s",
-                                   AFR_XATTR_PREFIX,
-                                   trav->xlator->name);
-                if (-1 == ret) {
+                                   AFR_XATTR_PREFIX, ptr);
+                if (ret  == -1) {
                         ret = -ENOMEM;
                         goto out;
                 }
+                i++;
+        }
 
+        trav = this->children;
+        i = 0;
+        while (i < child_count) {
+                priv->children[i] = trav->xlator;
                 trav = trav->next;
                 i++;
         }
@@ -457,6 +465,7 @@ init (xlator_t *this)
 
         ret = 0;
 out:
+        GF_FREE (ptr);
         return ret;
 }
 
@@ -780,6 +789,11 @@ struct volume_options options[] = {
 	  .type = GF_OPTION_TYPE_STR,
 	  .default_value = AFR_DIRTY_DEFAULT,
 	},
+	{ .key = {"afr-pending-xattr"},
+	  .type = GF_OPTION_TYPE_STR,
+          .description = "Comma seperated list of xattrs that are used to  "
+                         "capture information on pending heals."
+	},
 	{ .key = {"metadata-splitbrain-forced-heal"},
 	  .type = GF_OPTION_TYPE_BOOL,
 	  .default_value = "off",
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index 988c6053d41..29e5fe9b3f9 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -3286,6 +3286,80 @@ out:
 }
 
 static int
+set_afr_pending_xattrs_option (volgen_graph_t *graph,
+                              glusterd_volinfo_t *volinfo,
+                              int clusters)
+{
+        xlator_t *xlator            = NULL;
+        xlator_t **afr_xlators_list = NULL;
+        xlator_t  *this             = NULL;
+        glusterd_conf_t *conf       = NULL;
+        glusterd_brickinfo_t *brick = NULL;
+        char *ptr                   = NULL;
+        int i                       = 0;
+        int index                   = -1;
+        int ret                     = 0;
+        char *afr_xattrs_list       = NULL;
+        int list_size               = -1;
+
+        this = THIS;
+        GF_VALIDATE_OR_GOTO ("glusterd", this, out);
+        conf = this->private;
+        GF_VALIDATE_OR_GOTO (this->name, conf, out);
+
+        if (conf->op_version < GD_OP_VERSION_3_7_7)
+                return ret;
+
+        /* (brick_id x rep.count) + (rep.count-1 commas) + NULL*/
+        list_size = (1024 * volinfo->replica_count) +
+                    (volinfo->replica_count - 1) + 1;
+        afr_xattrs_list = GF_CALLOC (1, list_size, gf_common_mt_char);
+        if (!afr_xattrs_list)
+                goto out;
+
+        ptr = afr_xattrs_list;
+        afr_xlators_list = GF_CALLOC (clusters, sizeof (xlator_t *),
+                                      gf_common_mt_xlator_t);
+        if (!afr_xlators_list)
+                goto out;
+
+        xlator = first_of (graph);
+
+        for (i = 0, index = clusters - 1; i < clusters; i++) {
+                afr_xlators_list[index--] = xlator;
+                xlator = xlator->next;
+        }
+
+        i = 1;
+        index = 0;
+
+        cds_list_for_each_entry (brick, &volinfo->bricks, brick_list) {
+                if (index == clusters)
+                        break;
+                strncat (ptr, brick->brick_id, strlen(brick->brick_id));
+                if (i == volinfo->replica_count) {
+                        ret = xlator_set_option (afr_xlators_list[index++],
+                                                 "afr-pending-xattr",
+                                                 afr_xattrs_list);
+                        if (ret)
+                                return ret;
+                        memset (afr_xattrs_list, 0, list_size);
+                        ptr = afr_xattrs_list;
+                        i = 1;
+                        continue;
+                }
+                ptr[strlen(brick->brick_id)] = ',';
+                ptr += strlen (brick->brick_id) + 1;
+                i++;
+        }
+
+out:
+        GF_FREE (afr_xattrs_list);
+        GF_FREE (afr_xlators_list);
+        return ret;
+}
+
+static int
 volgen_graph_build_afr_clusters (volgen_graph_t *graph,
                                  glusterd_volinfo_t *volinfo)
 {
@@ -3296,7 +3370,7 @@ volgen_graph_build_afr_clusters (volgen_graph_t *graph,
                                                 "%s-replicate-%d"};
         xlator_t        *afr                 = NULL;
         char            option[32]           = {0};
-        int             start_count = 0;
+        int             start_count          = 0;
 
         if (volinfo->tier_info.cold_type == GF_CLUSTER_TYPE_REPLICATE)
                 start_count = volinfo->tier_info.cold_brick_count /
@@ -3321,6 +3395,11 @@ volgen_graph_build_afr_clusters (volgen_graph_t *graph,
         if (clusters < 0)
                 goto out;
 
+        ret = set_afr_pending_xattrs_option (graph, volinfo, clusters);
+        if (ret) {
+                clusters = -1;
+                goto out;
+        }
         if (!volinfo->arbiter_count)
                 goto out;
 
@@ -3343,8 +3422,6 @@ volume_volgen_graph_build_clusters (volgen_graph_t *graph,
                                     glusterd_volinfo_t *volinfo,
                                     gf_boolean_t is_quotad)
 {
-        char                    *replicate_args[]   = {"cluster/replicate",
-                                                       "%s-replicate-%d"};
         char                    *tier_args[]        = {"cluster/tier",
                                                        "%s-tier-%d"};
         char                    *stripe_args[]      = {"cluster/stripe",
@@ -3536,7 +3613,6 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
         int                ret                     = -1;
         xlator_t          *root                    = NULL;
         xlator_t          *xl, *hxl, *cxl;
-        glusterd_brickinfo_t    *brick             = NULL;
         char              *rule                    = NULL;
         int                st_brick_count          = 0;
         int                st_replica_count        = 0;
@@ -3546,6 +3622,8 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
         int                dist_count              = 0;
         int                start_count             = 0;
         char              *decommissioned_children = NULL;
+        glusterd_volinfo_t *dup_volinfo            = NULL;
+        gf_boolean_t       is_hot_tier             = _gf_false;
 
         st_brick_count     = volinfo->brick_count;
         st_replica_count   = volinfo->replica_count;
@@ -3560,8 +3638,13 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
         volinfo->redundancy_count = volinfo->tier_info.cold_redundancy_count;
         volinfo->type           = volinfo->tier_info.cold_type;
         volinfo->tier_info.cur_tier_hot = 0;
+        ret = glusterd_create_sub_tier_volinfo (volinfo, &dup_volinfo,
+                                                is_hot_tier, volinfo->volname);
+        if (ret)
+                goto out;
 
-        ret = volume_volgen_graph_build_clusters (graph, volinfo, is_quotad);
+        ret = volume_volgen_graph_build_clusters (graph, dup_volinfo,
+                                                  is_quotad);
         if (ret)
                 goto out;
         cxl = first_of(graph);
@@ -3588,12 +3671,17 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
                          volinfo->brick_count,
                          volinfo->replica_count,
                          start_count);
-                if (ret != -1)
-                        volgen_link_bricks_from_list_tail (graph,  volinfo,
+                if (ret != -1) {
+                        ret = set_afr_pending_xattrs_option (graph, volinfo,
+                                                             ret);
+                        if (ret)
+                                goto out;
+                         volgen_link_bricks_from_list_tail (graph,  volinfo,
                                                            "cluster/distribute",
                                                            "%s-hot-dht",
                                                            dist_count,
                                                            dist_count);
+                }
         } else {
                 ret = volgen_link_bricks_from_list_head (graph,  volinfo,
                                                  "cluster/distribute",
@@ -3653,6 +3741,8 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
         volinfo->dist_leaf_count = st_dist_leaf_count;
         volinfo->tier_info.cur_tier_hot = 0;
 
+        if (dup_volinfo)
+                glusterd_volinfo_delete (dup_volinfo);
         GF_FREE (rule);
         return ret;
 }