[dm-devel] dm-mq and end_clone_request()

[Bart Van Assche]

On 08/08/16 06:44, Johannes Thumshirn wrote:
> So __scsi_remove_device() is also checking for sdev->sdev_state == SDEV_DEL
> and returns if so. If it would have the chance to do so the goto restart would
> be hit and we'd reatart the list traverse. The if in turn just continues and
> I've seen endless loops with this pattern (check the 40998193560 ->
> 90a88d6ef88edc -> f05795d3d771f30a7bd commit chain).
>
> Might want to give the below patch a shot?
>
>
>>From fee838ebfea88b581994b3f855eab8da20b07fc9 Mon Sep 17 00:00:00 2001
> From: Johannes Thumshirn <jthumshirn at suse.de>
> Date: Mon, 8 Aug 2016 15:41:09 +0200
> Subject: [PATCH] scsi: restart full list search when re-encountering a deleted device
>
> __scsi_remove_device() already checks for reentrency with a deleted device,
> so there's no need to do it in scsi_forget_host() as well.
>
> Signed-off-by: Johannes Thumshirn <jthumshirn at suse.de>
> ---
>  drivers/scsi/scsi_scan.c | 2 --
>  1 file changed, 2 deletions(-)
>
> diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
> index e0a78f5..1c5a4d6 100644
> --- a/drivers/scsi/scsi_scan.c
> +++ b/drivers/scsi/scsi_scan.c
> @@ -1890,8 +1890,6 @@ void scsi_forget_host(struct Scsi_Host *shost)
>   restart:
>  	spin_lock_irqsave(shost->host_lock, flags);
>  	list_for_each_entry(sdev, &shost->__devices, siblings) {
> -		if (sdev->sdev_state == SDEV_DEL)
> -			continue;
>  		spin_unlock_irqrestore(shost->host_lock, flags);
>  		__scsi_remove_device(sdev);
>  		goto restart;

Hello Johannes,

Sorry but I do not agree with this patch. This patch will namely cause 
scsi_forget_host() to busy-wait until 
scsi_device_dev_release_usercontext() is called if it encounters a SCSI 
device on the host list that is in state SDEV_DEL. Additionally, I don't 
see how this patch can avoid an endless loop.

Thanks,

Bart.