Merge pull request #44221 from BenoitKnecht/ceph-mon-device-metrics

mon: Abort device health when device not found

Reviewed-by: Yaarit Hatuka <yaarit@redhat.com>

src/mon/Monitor.cc

@@ -421,6 +421,11 @@ int Monitor::do_admin_command(
     cmd_getval(cmdmap, "devid", want_devid);
 
     string devname = store->get_devname();
+    if (devname.empty()) {
+      err << "could not determine device name for " << store->get_path();
+      r = -ENOENT;
+      goto abort;
+    }
     set<string> devnames;
     get_raw_devices(devname, &devnames);
     json_spirit::mObject json_map;

src/mon/MonitorDBStore.h

@@ -57,6 +57,10 @@ class MonitorDBStore
     return devname;
   }
 
+  std::string get_path() {
+    return path;
+  }
+
   std::shared_ptr<PriorityCache::PriCache> get_priority_cache() const {
     return db->get_priority_cache();
   }

systemd/ceph-mon@.service.in

@@ -20,7 +20,10 @@ LockPersonality=true
 MemoryDenyWriteExecute=true
 # Need NewPrivileges via `sudo smartctl`
 NoNewPrivileges=false
-PrivateDevices=yes
+# We need access to block devices to check the health of the disk backing the
+# monitor DB store. It can be set to `true` if you're not interested in that
+# feature.
+PrivateDevices=false
 PrivateTmp=true
 ProtectControlGroups=true
 ProtectHome=true