Merge pull request #3336 from ceph/wip-fs-reset

mon: implement `fs reset`

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

qa/workunits/cephtool/test.sh

@@ -715,6 +715,11 @@ function test_mon_mds()
   metadata_poolnum=$(ceph osd dump | grep "pool.* 'fs_metadata" | awk '{print $2;}')
 
   fail_all_mds
+
+  # Check that 'fs reset' runs
+  ceph fs reset cephfs --yes-i-really-mean-it
+
+  # Clean up to enable subsequent newfs tests
   ceph fs rm cephfs --yes-i-really-mean-it
 
   set +e

src/mon/MDSMonitor.cc

@@ -1212,6 +1212,53 @@ int MDSMonitor::management_command(
     pending_mdsmap.created = ceph_clock_now(g_ceph_context);
 
     return 0;
+  } else if (prefix == "fs reset") {
+    string fs_name;
+    cmd_getval(g_ceph_context, cmdmap, "fs_name", fs_name);
+    if (!pending_mdsmap.get_enabled() || fs_name != pending_mdsmap.fs_name) {
+        ss << "filesystem '" << fs_name << "' does not exist";
+        // Unlike fs rm, we consider this case an error
+        return -ENOENT;
+    }
+
+    // Check that no MDS daemons are active
+    if (!pending_mdsmap.up.empty()) {
+      ss << "all MDS daemons must be inactive before resetting filesystem: set the cluster_down flag"
+            " and use `ceph mds fail` to make this so";
+      return -EINVAL;
+    }
+
+    // Check for confirmation flag
+    string sure;
+    cmd_getval(g_ceph_context, cmdmap, "sure", sure);
+    if (sure != "--yes-i-really-mean-it") {
+      ss << "this is a potentially destructive operation, only for use by experts in disaster recovery.  "
+        "Add --yes-i-really-mean-it if you are sure you wish to continue.";
+      return -EPERM;
+    }
+
+    MDSMap newmap;
+
+    // Populate rank 0 as existing (so don't go into CREATING)
+    // but failed (so that next available MDS is assigned the rank)
+    newmap.in.insert(mds_rank_t(0));
+    newmap.failed.insert(mds_rank_t(0));
+
+    // Carry forward what makes sense
+    newmap.data_pools = mdsmap.data_pools;
+    newmap.metadata_pool = mdsmap.metadata_pool;
+    newmap.cas_pool = mdsmap.cas_pool;
+    newmap.fs_name = mdsmap.fs_name;
+    newmap.created = ceph_clock_now(g_ceph_context);
+    newmap.epoch = mdsmap.epoch + 1;
+    newmap.inc = mdsmap.inc;
+    newmap.enabled = mdsmap.enabled;
+    newmap.inline_data_enabled = mdsmap.inline_data_enabled;
+
+    // Persist the new MDSMap
+    pending_mdsmap = newmap;
+    return 0;
+
   } else {
     return -ENOSYS;
   }

src/mon/MonCommands.h

@@ -307,6 +307,11 @@ COMMAND("fs rm " \
 	"name=sure,type=CephChoices,strings=--yes-i-really-mean-it,req=false", \
 	"disable the named filesystem", \
 	"fs", "rw", "cli,rest")
+COMMAND("fs reset " \
+	"name=fs_name,type=CephString " \
+	"name=sure,type=CephChoices,strings=--yes-i-really-mean-it,req=false", \
+	"disaster recovery only: reset to a single-MDS map", \
+	"fs", "rw", "cli,rest")
 COMMAND("fs ls ", \
 	"list filesystems", \
 	"fs", "r", "cli,rest")