From 878463ef3a26f49b9ccad9696ad130fb5aee47e7 Mon Sep 17 00:00:00 2001
From: "yite.gu" <yitegu0@gmail.com>
Date: Mon, 29 Jan 2024 18:26:20 +0800
Subject: [PATCH] client: check mds down status bofore getting mds_gid_t from
 mdsmap

Get mds_gid_t from up of MDSMap, will cause to core dump if target
mds_rank_t does not exist up map:

from: src/mds/MDSMap.h
  const auto& get_info(mds_rank_t m) const {
    return mds_info.at(up.at(m));
  }

reproduct:
1. ceph fs fail <fs_name>
2. curl -X GET "example.com:8080/api/cephfs/1/clients" ...
up.at(m) will cause to core dump.

Fixes: https://tracker.ceph.com/issues/61844
Signed-off-by: Yite Gu <yitegu0@gmail.com>
---
 qa/tasks/mgr/dashboard/test_cephfs.py          | 14 ++++++++++++++
 src/client/Client.cc                           |  5 +++++
 src/pybind/mgr/dashboard/controllers/cephfs.py |  3 ++-
 3 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/qa/tasks/mgr/dashboard/test_cephfs.py b/qa/tasks/mgr/dashboard/test_cephfs.py
index d8e00fe60d9d..7a6b912a3fb8 100644
--- a/qa/tasks/mgr/dashboard/test_cephfs.py
+++ b/qa/tasks/mgr/dashboard/test_cephfs.py
@@ -347,3 +347,17 @@ class CephfsTest(DashboardTestCase):
         self.assertEqual(stats['subdirs'], 1)
 
         self.rm_dir('/animal')
+
+    def test_cephfs_clients_get_after_mds_down(self):
+        fs_id = self.get_fs_id()
+        self._get(f"/api/cephfs/{fs_id}/clients")
+        self.assertStatus(200)
+
+        self.fs.fail()
+        params = {'suppress_client_ls_errors': 'False'}
+        self._get(f"/api/cephfs/{fs_id}/clients", params=params)
+        self.assertStatus(500)
+
+        self.fs.set_joinable()
+        self._get(f"/api/cephfs/{fs_id}/clients")
+        self.assertStatus(200)
diff --git a/src/client/Client.cc b/src/client/Client.cc
index 8966bf60adb0..6af4d24cae53 100644
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -6273,6 +6273,11 @@ int Client::resolve_mds(
   if (role_r == 0) {
     // We got a role, resolve it to a GID
     const auto& mdsmap = fsmap->get_filesystem(role.fscid).get_mds_map();
+    if (mdsmap.is_down(role.rank)) {
+      lderr(cct) << __func__ << ": targets rank: " << role.rank
+                 << " is down" << dendl;
+      return -CEPHFS_EAGAIN;
+    }
     auto& info = mdsmap.get_info(role.rank);
     ldout(cct, 10) << __func__ << ": resolved " << mds_spec << " to role '"
       << role << "' aka " << info.human_name() << dendl;
diff --git a/src/pybind/mgr/dashboard/controllers/cephfs.py b/src/pybind/mgr/dashboard/controllers/cephfs.py
index fca5ddc95ee1..84af1b9e2f2e 100644
--- a/src/pybind/mgr/dashboard/controllers/cephfs.py
+++ b/src/pybind/mgr/dashboard/controllers/cephfs.py
@@ -1,4 +1,5 @@
 # -*- coding: utf-8 -*-
+# pylint: disable=too-many-lines
 import errno
 import json
 import logging
@@ -110,7 +111,7 @@ class CephFS(RESTController):
         if flag not in ('True', 'False'):
             raise DashboardException(msg='suppress_client_ls_errors value '
                                          'needs to be either True or False '
-                                         f'but provided {flag}',
+                                         f'but provided "{flag}"',
                                      component='cephfs')
 
         fs_id = self.fs_id_to_int(fs_id)
-- 
2.47.3