Hello. BUG: RAID6 recovery broken by commit 4f4fd7c5798bbdd5a03a60f6269cf1177fbd11ef (Linux 5.1.3+) Replacing a failed disk of a MD RAID6 array causes file system corruption and data loss on kernels containing commit 4f4fd7c5798bbdd5a03a60f6269cf1177fbd11ef. Affected kernels: 5.1.3, 5.1.4 possibly others. Unaffected kernels: 5.1.2 OS: Debian stretch amd64 Steps to reproduce the BUG: 1. Create a new 4-disk RAID6 array, create a file system and mount it: mdadm /dev/md0 --create -l 6 -n 4 /dev/sd[bcde] mkfs.ext4 /dev/md0 mount /dev/md0 /mnt 2. Store some data (a few GB should be fine) on the RAID6 arrays file system: cp -r whatever /mnt 3. Fail a disk of the RAID6 array and remove it from the array: mdadm /dev/md0 --fail /dev/sdd mdadm /dev/md0 --remove /dev/sdd 4. Drop caches: echo "3" > /proc/sys/vm/drop_caches 5. Compare data copied to the RAID6 array in step 2 with its source: diff -r whatever /mnt/whatever There should be no differences and no file system errors. 6. Add a new empty disk to the RAID6 array: mdadm /dev/md0 --add /dev/sdf 7. RAID6 recovery should start now, wait for the RAID6 recovery to finish. 8. Drop caches again: echo "3" > /proc/sys/vm/drop_caches 9. Compare data copied to the RAID6 array in step 2 with its source again: diff -r whatever /mnt/whatever diff now reports a lot of differences and the kernel log gets filled with file system errors. For example: EXT4-fs warning (device md0): ext4_dirent_csum_verify:355: inode #918549: comm diff: No space for directory leaf checksum. Please run e2fsck -D. Reverting commit 4f4fd7c5798bbdd5a03a60f6269cf1177fbd11ef from kernel 5.1.4 resolves the issues described above. Kind regards Thorsten -- ___ | | / E-Mail: linux@xxxxxxxxxxxxxxxxx |horsten |/\nabe WWW: http://linux.thorsten-knabe.de
