openzfs/zfs

MD RAID headers can cause Linux to steal disks from ZFS

Open

#634 创建于 2012年4月2日

在 GitHub 查看
 (30 评论) (0 反应) (0 负责人)C (9,908 star) (1,703 fork)batch import
Bot: Not StaleStatus: Understoodgood first issue

描述

I received a report from a veteran Gentoo user that MD RAID had stolen two disks from a raidz2 vdev. It seems that ZFS leaves a hole for a bootloader that an MD RAID header can occupy, causing this issue. This occurred on partitions that were 0xBF, not 0xFD.

I have found myself recommending people do mdadm --zero-superblock ... before creating a ZFS pool to avoid this situation. It would be a good idea to modify the code to zero this at vdev creation to preclude any possibility that an old MD RAID header will cause the kernel module to steal the disks.

I am not sure if I have time to do this right now, but I figured that I would file a report so that people are aware of the issue.

贡献者指南

技术栈
c
领域
backendinfrastructure
议题类型
bug
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
4
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
over 1 week
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
stale
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
C programmingLinux kernel conceptsZFS architecturemdadm utility
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
10
研究方向
Investigate the vdev creation code in the ZFS source tree (likely in module/zfs/vdev.c or similar). The issue suggests zeroing the MD RAID superblock header during vdev creation to prevent Linux from stealing disks. Review the 30 comments on the issue for any pointers or proposed implementations. Look for existing tests or documentation related to disk initialization.