Hi, Recently we found a weird failure by running xfstests generic/563 on xfs or ext4, when the test device is a general disk partition. If you don't want to try xfstests, you can see below script[1]. If we test on a disk partition (e.g. /dev/sdb1 or /dev/nvme0n1p1), the cgroup's io.stat always can't output the correct minor device number of the disk partition. LVM or loop devices don't have this issue. Is there any 'blk-cgroup' expert would like to help to answer this question? Is this expected or it's a bug will be fixed later? It can be reproduced on latest upstream kernel. Thanks, Zorro [1] # cat blkcgroup-test.sh #/bin/bash umount /dev/nvme0n1p1 2>/dev/null mkfs.xfs -f /dev/nvme0n1p1 >/dev/null 2>&1 mount /dev/nvme0n1p1 /mnt/scratch major=$((0x`stat -L -c "%t" /dev/nvme0n1p1`)) minor=$((0x`stat -L -c "%T" /dev/nvme0n1p1`)) echo "$major:$minor" echo "+io" > /sys/fs/cgroup/cgroup.subtree_control echo $$ > /sys/fs/cgroup/cgroup.procs rmdir /sys/fs/cgroup/mytest-cg xfs_io -fc "pwrite 0 8m" /mnt/scratch/file >/dev/null 2>&1 umount /dev/nvme0n1p1 && mount /dev/nvme0n1p1 /mnt/scratch stat /mnt/scratch/file >/dev/null mkdir /sys/fs/cgroup/mytest-cg echo $$ > /sys/fs/cgroup/mytest-cg/cgroup.procs xfs_io -c "pread 0 8m" -c "pwrite 0 8m" -c fsync /mnt/scratch/file >/dev/null 2>&1 mkdir -p /sys/fs/cgroup echo $$ > /sys/fs/cgroup/cgroup.procs xfs_io -c fsync /mnt/scratch/file cat /sys/fs/cgroup/mytest-cg/io.stat [2] 259:1 259:0 rbytes=8388608 wbytes=8389632 rios=67 wios=12 dbytes=0 dios=0
