Merge pull request ceph#55374 from aclamk/wip-aclamk-bs-multi-label

os/bluestore: Multiple bdev labels on main block device

Author: aclamk

src/common/options/global.yaml.in

@@ -4411,6 +4411,40 @@ options:
   flags:
   - create
   with_legacy: true
+- name: bluestore_bdev_label_multi
+  type: bool
+  level: advanced
+  desc: Keep multiple copies of block device label.
+  long_desc: Having multiple labels is only useful in error conditions.
+    The label located at offset 0 has been known to be sometimes overwritten by unknown causes,
+    but without it OSD cannot run.
+  default: true
+  flags:
+  - create
+  with_legacy: false
+- name: bluestore_bdev_label_require_all
+  type: bool
+  level: advanced
+  desc: Require all copies to match.
+  long_desc: Under normal conditions, all copies should be the same.
+    Clearing this flag allows to run OSD if at least one of labels
+    could be properly read.
+  default: true
+  see_also:
+  - bluestore_bdev_label_multi
+  flags:
+  - runtime
+  with_legacy: false
+- name: bluestore_bdev_label_multi_upgrade
+  type: bool
+  level: advanced
+  desc: Let repair upgrade to multi label.
+  long_desc: By default single label is preserved.
+    Setting this variable before running fsck-repair upgrades single label into multi label.
+  default: false
+  flags:
+  - startup
+  with_legacy: false
 # whether preallocate space if block/db_path/wal_path is file rather that block device.
 - name: bluestore_block_preallocate_file
   type: bool
@@ -5333,12 +5367,6 @@ options:
   level: dev
   default: false
   with_legacy: true
-- name: bluestore_debug_prefill
-  type: float
-  level: dev
-  desc: simulate fragmentation
-  default: 0
-  with_legacy: true
 - name: bluestore_debug_prefragment_max
   type: size
   level: dev

src/os/bluestore/BlueFS.cc

@@ -490,7 +490,7 @@ int BlueFS::add_block_device(unsigned id, const string& path, bool trim,
       break;
     case BDEV_DB:
     case BDEV_NEWDB:
-      reserved = DB_SUPER_RESERVED;
+      reserved = SUPER_RESERVED;
       break;
     case BDEV_SLOW:
       reserved = 0;
@@ -546,6 +546,13 @@ uint64_t BlueFS::get_block_device_size(unsigned id) const
   return 0;
 }
 
+BlockDevice* BlueFS::get_block_device(unsigned id) const
+{
+  if (id < bdev.size() && bdev[id])
+    return bdev[id];
+  return nullptr;
+}
+
 void BlueFS::handle_discard(unsigned id, interval_set<uint64_t>& to_release)
 {
   dout(10) << __func__ << " bdev " << id << dendl;

src/os/bluestore/BlueFS.h

@@ -743,6 +743,7 @@ class BlueFS {
 		       bluefs_shared_alloc_context_t* _shared_alloc = nullptr);
   bool bdev_support_label(unsigned id);
   uint64_t get_block_device_size(unsigned bdev) const;
+  BlockDevice* get_block_device(unsigned bdev) const;
 
   // handler for discard event
   void handle_discard(unsigned dev, interval_set<uint64_t>& to_release);