[0/5,V5] vhost-scsi: IO error fixups

Message

Mike Christie Nov. 10, 2020, 5:33 a.m. UTC

The following patches were made over Michael's vhost branch:

https://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git/log/?h=vhost

They fixe a couple issues where the guest is getting IO errors:
1. The admin may set num_queues and cmd_per_lun/virtqueue_size
in a way that allows the guest to send more IO that vhost-scsi
was coded for.

2. A race where vhost-scsi completes the cmd to the guest before
it's fully completed in the target/vhost-scsi layers so if the
guest sends a new IO too quicky we fail it.

3. If the real/physical storage (lio backend device) hits a transient
error then vhost-scsi would fail commands instead of relying on
the lower levels error recovery to do what the admin had set up.

V5:
- Bring back V1 but move flush patches to a separate clean up set
- Modify test in vhost_vq_is_setup

V4:
- really really fix compile errors
- dropped threading patches so we can figure that out separately.

V3:
- fix compile errors
- fix possible crash where cmd could be freed while adding it to
completion list
- fix issue where we added the worker thread to the blk cgroup but
the blk IO was submitted by a driver workqueue.

V2:
- fix use before set cpu var errors
- drop vhost_vq_is_setup
- include patches to do a worker thread per scsi IO vq

Comments

Stefan Hajnoczi Nov. 12, 2020, 5 p.m. UTC | #1

On Mon, Nov 09, 2020 at 11:33:18PM -0600, Mike Christie wrote:
> The following patches were made over Michael's vhost branch:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git/log/?h=vhost
> 
> They fixe a couple issues where the guest is getting IO errors:
> 1. The admin may set num_queues and cmd_per_lun/virtqueue_size
> in a way that allows the guest to send more IO that vhost-scsi
> was coded for.
> 
> 2. A race where vhost-scsi completes the cmd to the guest before
> it's fully completed in the target/vhost-scsi layers so if the
> guest sends a new IO too quicky we fail it.
> 
> 3. If the real/physical storage (lio backend device) hits a transient
> error then vhost-scsi would fail commands instead of relying on
> the lower levels error recovery to do what the admin had set up.
> 
> V5:
> - Bring back V1 but move flush patches to a separate clean up set
> - Modify test in vhost_vq_is_setup
> 
> V4:
> - really really fix compile errors
> - dropped threading patches so we can figure that out separately.
> 
> V3:
> - fix compile errors
> - fix possible crash where cmd could be freed while adding it to
> completion list
> - fix issue where we added the worker thread to the blk cgroup but
> the blk IO was submitted by a driver workqueue.
> 
> V2:
> - fix use before set cpu var errors
> - drop vhost_vq_is_setup
> - include patches to do a worker thread per scsi IO vq
> 
> 

Acked-by: Stefan Hajnoczi <[e-mail]>