On 10.05.2021 00:14 Martin Raiber wrote:

I get this (rare) issue where btrfs reports an IO error in run_delayed_refs or finish_ordered_io with no underlying device errors being reported. This is with 5.10.26 but with a few patches like the pcpu ENOMEM fix or work-arounds for btrfs ENOSPC issues:

[1885197.101981] systemd-sysv-generator[2324776]: SysV service '/etc/init.d/exim4' lacks a native systemd unit file. Automatically generating a unit file for compatibility. Please update package to include a native systemd unit file, in order to make it more safe and robust.
[2260628.156893] BTRFS: error (device dm-0) in btrfs_finish_ordered_io:2736: errno=-5 IO failure
[2260628.156980] BTRFS info (device dm-0): forced readonly

This issue occured on two different machines now (on one twice). Both with ECC RAM. One bare metal (where dm-0 is on a NVMe) and one in a VM (where dm-0 is a ceph volume).

Just got it again (5.10.43). So I guess the question is how can I trace where this error comes from... The error message points at btrfs_csum_file_blocks but nothing beyond that. Grep for EIO and put a WARN_ON at each location?