5.0.0-32.34 causes system instability with cephfs

[Stefan Fleischmann]

On Ubuntu 18.04

linux-image-5.0.0-32-generic 5.0.0-32.34~18.04.2

According to https://<email address hidden>/msg00940.html the same happens on Disco with this kernel version.

Processes that read data from CephFS get stuck, the system becomes unstable. dmesg contains messages like this:

[ 1510.703090] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info
[ 1510.711590] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info
[ 1510.720093] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info
[ 1510.728592] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info
[ 1510.737086] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info
[ 1510.745587] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info
[ 1510.754080] cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info

According to the discussion here: https://github.com/coreos/bugs/issues/2616
this is caused by a faulty patch. If that information is correct the problematic
patch should be https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/disco/commit/?id=100a8eb40c492f2525cdae434c50d53ec7f5cc23

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Brandon Sanabria (sana4sho)]

Seeing the same in our farm.

journalctl -k -p err -b --no-full --no-pager

-- Logs begin at Sat 2019-11-02 16:23:20 PDT, end at Sun 2019-11-03 01:57:55 PDT. --

Nov 02 20:36:59 10.173.164.1-7e6e6109 kernel: cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info

Nov 02 20:36:59 10.173.164.1-7e6e6109 kernel: cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info

Nov 02 20:37:07 10.173.164.1-7e6e6109 kernel: cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info

Nov 02 20:37:07 10.173.164.1-7e6e6109 kernel: cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info

[Stefan Fleischmann (sfleischmann)]

It looks like this has been fixed here:
https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/disco/commit/?id=3b931173c97b0d73f80ea55b72bb2966a246167f

and here:
https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/bionic/commit/?h=hwe&id=3b931173c97b0d73f80ea55b72bb2966a246167f

so supposedly already with 5.0.0-33

I've tried the latest hwe on bionic (5.0.0-36), and do not see the problem there anymore.