在Bcache上启动OSD报unable to read osd superblock错误

环境信息

环境	具体信息
架构	LoongArch
处理器	Loongson-3C5000
内核版本	4.19
操作系统版本	lns8
Ceph版本	Nautilus 14.2.22
Ceph Cluster	单机最小集群，一个Monitor，两个OSD，一个Manager
PAGESIZE	16384

[root@ceph01 ~]# getconf PAGESIZE
16384

问题描述

使用Bcache加速块设备，在上述环境中创建Bcache，并在Bcache上创建OSD。但是systemctl restart ceph-osd@0.service时失败，/var/log/ceph/ceph-osd.0.log日志如下：

2023-10-13 05:26:42.705 fff37c0030 -1 bluestore(/var/lib/ceph/osd/ceph-0) _verify_csum bad crc32c/0x1000 checksum at blob offset 0x0, got 0x246e0328, expected 0x6d5d9709, device location [0x2000~1000], logical extent 0x0~1000, object #-1:7b3f43c4:::osd_superblock:0#
2023-10-13 05:26:42.705 fff37c0030 -1 bluestore(/var/lib/ceph/osd/ceph-0) _verify_csum bad crc32c/0x1000 checksum at blob offset 0x0, got 0x246e0328, expected 0x6d5d9709, device location [0x2000~1000], logical extent 0x0~1000, object #-1:7b3f43c4:::osd_superblock:0#
2023-10-13 05:26:42.705 fff37c0030 -1 bluestore(/var/lib/ceph/osd/ceph-0) _verify_csum bad crc32c/0x1000 checksum at blob offset 0x0, got 0x246e0328, expected 0x6d5d9709, device location [0x2000~1000], logical extent 0x0~1000, object #-1:7b3f43c4:::osd_superblock:0#
2023-10-13 05:26:42.705 fff37c0030 -1 bluestore(/var/lib/ceph/osd/ceph-0) _verify_csum bad crc32c/0x1000 checksum at blob offset 0x0, got 0x246e0328, expected 0x6d5d9709, device location [0x2000~1000], logical extent 0x0~1000, object #-1:7b3f43c4:::osd_superblock:0#
2023-10-13 05:26:42.705 fff37c0030 -1 osd.0 0 OSD::init() : unable to read osd superblock
2023-10-13 05:26:42.705 fff37c0030  1 bluestore(/var/lib/ceph/osd/ceph-0) umount
2023-10-13 05:26:42.705 fff37c0030  4 rocksdb: [db/db_impl.cc:390] Shutdown: canceling all background work
2023-10-13 05:26:42.705 fff37c0030  4 rocksdb: [db/db_impl.cc:563] Shutdown complete
2023-10-13 05:26:42.709 fff37c0030  1 bluefs umount
2023-10-13 05:26:42.709 fff37c0030  1 bdev(0xaac6157500 /var/lib/ceph/osd/ceph-0/block.wal) close
2023-10-13 05:26:42.989 fff37c0030  1 bdev(0xaac6157880 /var/lib/ceph/osd/ceph-0/block.db) close
2023-10-13 05:26:43.273 fff37c0030  1 bdev(0xaac6157c00 /var/lib/ceph/osd/ceph-0/block) close
2023-10-13 05:26:43.509 fff37c0030  1 freelist shutdown
2023-10-13 05:26:43.509 fff37c0030  1 bdev(0xaac6156000 /var/lib/ceph/osd/ceph-0/block) close
2023-10-13 05:26:43.709 fff37c0030 -1  ** ERROR: osd init failed: (22) Invalid argument

可以看到OSD::init() : unable to read osd superblock，在OSD初始化时，无法读取OSD superblock。

解决方法

有两种解决办法：

1. 将内核参数——PAGESIZE修改为4K。在鲲鹏BoostKit分布式存储使能套件文档中提供了将内核参数——PAGESIZE修改为4K的方法。
   • https://www.hikunpeng.com/document/detail/zh/kunpengsdss/appAccelFeatures/globalCache/kunpengglobalcache_05_0040.html
2. （推荐）在loongarch平台16K页大小情况下，OSD采用direct write写superblock到地址8K-12K，采用buffer write写设备标签到地址0-4K，对buffer write操作系统会按页对齐刷盘，superblock和设备标签刚好在同一个页上，刷盘导致superblock被覆盖，无法读出正确的数据。将写设备标签改成direct write修复此问题。
   • https://gitee.com/src-openeuler/ceph/blob/master/0007-bluestore-use-direct-write-for-bdevlabel.patch
   • 在最新版本的Ceph源码中也修复了此问题：https://github.com/ceph/ceph/blob/main/src/os/bluestore/BlueStore.cc#L6480

参考

• https://gitee.com/src-openeuler/ceph/issues/I54Q01
• https://gitee.com/src-openeuler/ceph/pulls/121/