cluster/afr: Implementation of quorum-reads

Provide a way of disabling reads when quorum is not met.

Change-Id: Ic4f57c2b87a0b8514600759de3a7a47e217fe3b5
BUG: 1187885
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/9543
Reviewed-by: Ravishankar N <ravishankar@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>

6 files changed, 36 insertions, 2 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index c31fcba6c3b..a28cbf4a2a9 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -3561,6 +3561,7 @@ afr_priv_dump (xlator_t *this)
         gf_proc_dump_write("read_child", "%d", priv->read_child);
         gf_proc_dump_write("favorite_child", "%d", priv->favorite_child);
         gf_proc_dump_write("wait_count", "%u", priv->wait_count);
+        gf_proc_dump_write("quorum-reads", "%d", priv->quorum_reads);
 
         return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c
index 29a926dbd97..ec67a20e624 100644
--- a/xlators/cluster/afr/src/afr-read-txn.c
+++ b/xlators/cluster/afr/src/afr-read-txn.c
@@ -195,6 +195,15 @@ afr_read_txn (call_frame_t *frame, xlator_t *this, inode_t *inode,
 	local->readfn = readfn;
 	local->inode = inode_ref (inode);
 
+        if (priv->quorum_reads &&
+            priv->quorum_count && !afr_has_quorum (priv->child_up, this)) {
+                local->op_ret = -1;
+                local->op_errno = ENOTCONN;
+                read_subvol = -1;
+                goto read;
+        }
+
+
 	local->transaction.type = type;
 	ret = afr_inode_read_subvol_type_get (inode, this, local->readable,
 					      &event_generation, type);
@@ -232,6 +241,7 @@ afr_read_txn (call_frame_t *frame, xlator_t *this, inode_t *inode,
 
 	local->read_attempted[read_subvol] = 1;
 
+read:
 	local->readfn (frame, this, read_subvol);
 
 	return 0;
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 3cb073ecfc3..afa11bba7ab 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -28,6 +28,13 @@ int
 afr_changelog_do (call_frame_t *frame, xlator_t *this, dict_t *xattr,
 		  afr_changelog_resume_t changelog_resume);
 
+static int32_t
+afr_quorum_errno (afr_private_t *priv)
+{
+        if (priv->quorum_reads)
+                return ENOTCONN;
+        return EROFS;
+}
 
 int
 __afr_txn_write_fop (call_frame_t *frame, xlator_t *this)
@@ -558,7 +565,7 @@ afr_handle_quorum (call_frame_t *frame)
         }
 
         local->op_ret = -1;
-        local->op_errno = EROFS;
+        local->op_errno = afr_quorum_errno (priv);
 }
 
 int
@@ -992,7 +999,7 @@ afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
          * quorum number of nodes.
          */
         if (priv->quorum_count && !afr_has_fop_quorum (frame)) {
-                op_errno = EROFS;
+                op_errno = afr_quorum_errno (priv);
                 goto err;
         }
 
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index bf7ba3fb0ac..f435767f5e4 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -211,6 +211,9 @@ reconfigure (xlator_t *this, dict_t *options)
         GF_OPTION_RECONF ("heal-timeout", priv->shd.timeout, options,
                           int32, out);
 
+        GF_OPTION_RECONF ("quorum-reads", priv->quorum_reads, options,
+                          bool, out);
+
         priv->did_discovery = _gf_false;
 
         ret = 0;
@@ -359,6 +362,8 @@ init (xlator_t *this)
 	GF_OPTION_INIT ("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
         GF_OPTION_INIT ("heal-timeout", priv->shd.timeout, int32, out);
 
+        GF_OPTION_INIT ("quorum-reads", priv->quorum_reads, bool, out);
+
         priv->wait_count = 1;
 
         priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
@@ -724,6 +729,12 @@ struct volume_options options[] = {
                          "this many bricks or present.  Other quorum types "
                          "will OVERWRITE this value.",
         },
+        { .key = {"quorum-reads"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "no",
+          .description = "If quorum-reads is \"true\" only allow reads if "
+                         "quorum is met when quorum is enabled.",
+        },
         { .key  = {"node-uuid"},
           .type = GF_OPTION_TYPE_STR,
           .description = "Local glusterd uuid string, used in starting "
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 8156eaa995e..dff70e89bb3 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -97,6 +97,7 @@ typedef struct _afr_private {
         gf_boolean_t      pre_op_compat;      /* on/off */
 	uint32_t          post_op_delay_secs;
         unsigned int      quorum_count;
+        gf_boolean_t      quorum_reads;
 
         char                   vol_uuid[UUID_SIZE + 1];
         int32_t                *last_event;
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
index e9473658176..e35a607cfc2 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
@@ -1655,6 +1655,10 @@ struct volopt_map_entry glusterd_volopt_map[] = {
           .op_version    = GD_OP_VERSION_3_7_0,
           .validate_fn   = validate_disperse_heal_enable_disable
         },
+        { .key           = "cluster.quorum-reads",
+          .voltype       = "cluster/replicate",
+          .op_version    = GD_OP_VERSION_3_7_0,
+        },
         { .key         = NULL
         }
 };