Linux 5.16 saw some changes to the btrfs defragmentation code that introduced a regression causing extreme write I/O (and the associated single-core 100% CPU usage by a kernel thread named btrfs-cleaner). Obviously, this is defragmentation in an endless loop. This affects volumes mounted with -ossd, autodefrag in fstab, which is the default on most distributions. Check if your distribution has this option and we recommend that you remove this flag for a while if you have a Linux 5.16.x kernel!
CC: (none) => 79625490833
Source RPM: Btrfs-cleaner => linux-btrfs
Source RPM: linux-btrfs => kernel-516.*
Source RPM: kernel-516.* => kernel-5.16.*
Thank you for the report. I can do no more than assign this to the kernel team.
Assignee: bugsquad => kernel
Fixed in 5.16.5 currently building
Resolution: (none) => FIXEDStatus: NEW => RESOLVED
actually... I had the fixes already in 5.16.4-2 :)