[PVE-User] Disk data corruption triggered with fstrim on qcow2/Gluster spotted

Igor Podlesny pve-user at poige.ru
Wed Mar 13 08:35:51 CET 2019


* Proxmox 5
* HN kernel version Linux 4.15.18-11-pve #1 SMP PVE 4.15.18-33
* pve-qemu-kvm 2.12.1-2

I didn't have time to figure it all out down to the narrowest possible
setup yet, but since data integrity is at stake I'm warning Proxmox
users and developers as soon as possible.

=== Setup ===

I have a VM with Ubuntu 16.04 that uses Btrfs formatted partition.
This VM's disk is virtio-scsi based and resides on Glusterfs using qcow2 format.

=== How I trigger the issue ===

Just before updating VM's kernel to 4.15.0-46-generic, I ran

    fstrim /

After update completion I've noticed dmesg entries about detected csum
mismatches. Having run Btrfs scrub I found out that only newly
installed package files had been corrupted.

=== How it doesn't break ===

Changing VM disk image type to VMDK or RAW makes same sequence of
steps harmless.
Hence I doubt that Glusterfs has been related, but since I didn't have
enough time to check it, I'm reporting the issue exactly as found.

-- 
End of message. Next message?


More information about the pve-user mailing list