btrfs wedge after 4.13.0-26 hwe kernel update

[Bill Lipa]

On Ubuntu 16.04 LTS I updated the hwe kernel from 4.10.0-42-generic x86_64 to 4.13.0-26 to get the Meltdown patch. On this box, I use systemd-nspawn to run containers using ephemeral mode on a btrfs volume (set up as a loopback device by systemd). After the kernel update, btrfs appears to be getting stuck somewhere with fairly high frequency (so far it's occurred always within an hour or two of use, on two separate machines). Anything that touches the btrfs volume will then freeze, including a btrfs scrub. The machine has to be hard rebooted and will continue to display the problem until the kernel is reverted, at which point btrfs proceeds as normal.

Attached are the sysreq 't' trace output and the hung task output from journalctl.

[Bill Lipa (masterleep)]

I was able to work around this by disabling quotas on the btrfs volume.

[Marc Pignat (swid)]

Hi Bill,

Can you please tell me if you see an unusual number of kworker threads running when you use the 4.13 kernel?

I have seen some serious problems with the 4.13.0-32-generic and I'm using btrfs for all the partitions, see also #1747597.

Best regards