protocol/client: handle the subdir handshake properly for add-brick

There should be different way we handle handshake in case of subdir
mount for the first time, and in case of subsequent graph changes.

Change-Id: I2a7ba836433bb0a0f4a861809e2bb0d7fbc4da54
BUG: 1505323
Signed-off-by: Amar Tumballi <amarts@redhat.com>

2 files changed, 35 insertions, 6 deletions

diff --git a/tests/features/subdir-mount.t b/tests/features/subdir-mount.t
index ab7ef35815b..1742f8655e8 100644
--- a/tests/features/subdir-mount.t
+++ b/tests/features/subdir-mount.t
@@ -82,17 +82,38 @@ TEST $CLI volume stop $V0
 
 TEST $CLI volume start $V0
 
-# /subdir2 has not allowed IP
-TEST $GFS --subdir-mount /subdir2 -s $H0 --volfile-id $V0 $M1
-TEST stat $M1
-
 TEST $GFS --subdir-mount /subdir1/subdir1.1/subdir1.2 -s $H0 --volfile-id $V0 $M2
 TEST stat $M2
 
+# mount shouldn't fail even after add-brick
+TEST $CLI volume add-brick $V0 replica 2 $H0:$B0/${V0}{5,6};
+
+# Give time for client process to get notified and use the new
+# volfile after add-brick
+sleep 1
+
+# Existing mount should still be active
+mount_inode=$(stat --format "%i" "$M2")
+TEST test "$mount_inode" == "1"
+
+TEST umount $M2
+
+# because the subdir is not yet 'healed', below should fail.
+TEST $GFS --subdir-mount /subdir2 -s $H0 --volfile-id $V0 $M2
+mount_inode=$(stat --format "%i" "$M2")
+TEST test "$mount_inode" != "1"
+
+# Allow the heal to complete
+TEST stat $M0/subdir1/subdir1.1/subdir1.2/subdir1.2_file;
+TEST stat $M0/subdir2/
+
+# Now the mount should succeed
+TEST $GFS --subdir-mount /subdir2 -s $H0 --volfile-id $V0 $M1
+TEST stat $M1
+
 # umount $M1 / $M2
 TEST umount $M0
 TEST umount $M1
-TEST umount $M2
 
 
 TEST $CLI volume stop $V0;
diff --git a/xlators/protocol/client/src/client-handshake.c b/xlators/protocol/client/src/client-handshake.c
index cd71c2c34a2..3cf2b06cb03 100644
--- a/xlators/protocol/client/src/client-handshake.c
+++ b/xlators/protocol/client/src/client-handshake.c
@@ -1079,10 +1079,14 @@ client_setvolume_cbk (struct rpc_req *req, struct iovec *iov, int count, void *m
         int32_t               op_errno      = 0;
         gf_boolean_t          auth_fail     = _gf_false;
         uint32_t              lk_ver        = 0;
+        glusterfs_ctx_t      *ctx           = NULL;
 
         frame = myframe;
         this  = frame->this;
         conf  = this->private;
+        GF_VALIDATE_OR_GOTO (this->name, conf, out);
+        ctx = this->ctx;
+        GF_VALIDATE_OR_GOTO (this->name, ctx, out);
 
         if (-1 == req->rpc_status) {
                 gf_msg (frame->this->name, GF_LOG_WARNING, ENOTCONN,
@@ -1145,9 +1149,13 @@ client_setvolume_cbk (struct rpc_req *req, struct iovec *iov, int count, void *m
                         auth_fail = _gf_true;
                         op_ret = 0;
                 }
-                if ((op_errno == ENOENT) && this->ctx->cmd_args.subdir_mount) {
+                if ((op_errno == ENOENT) && this->ctx->cmd_args.subdir_mount &&
+                    (ctx->graph_id <= 1)) {
                         /* A case of subdir not being present at the moment,
                            ride on auth_fail framework to notify the error */
+                        /* Make sure this case is handled only in the new
+                           graph, so mount may fail in this case. In case
+                           of 'add-brick' etc, we need to continue retry */
                         auth_fail = _gf_true;
                         op_ret = 0;
                 }