[RFC PATCH] nvme-fc: move tagset removal to nvme_fc_delete_ctrl()

[Hannes Reinecke]

On 1/13/25 13:44, Ming Lei wrote:
> Now target is removed from nvme_fc_ctrl_free() which is the ctrl->ref
> release handler. And even admin queue is unquiesced there, this way
> is definitely wrong because the ctr->ref is grabbed when submitting
> command.
> 
> And Marco observed that nvme_fc_ctrl_free() can be called from request
> completion code path, and trigger kernel warning since request completes
> from softirq context.
> 
> Fix the issue by moveing target removal into nvme_fc_delete_ctrl(),
> which is also aligned with nvme-tcp and nvme-rdma.
> 
> Cc: Marco Patalano <mpatalan at redhat.com>
> Cc: Ewan Milne <emilne at redhat.com>
> Cc: James Smart <james.smart at broadcom.com>
> Cc: Sagi Grimberg <sagi at grimberg.me>
> Signed-off-by: Ming Lei <ming.lei at redhat.com>
> ---
>   drivers/nvme/host/fc.c | 13 +++++++------
>   1 file changed, 7 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/nvme/host/fc.c b/drivers/nvme/host/fc.c
> index b81af7919e94..b94b4e50a3df 100644
> --- a/drivers/nvme/host/fc.c
> +++ b/drivers/nvme/host/fc.c
> @@ -2389,17 +2389,11 @@ nvme_fc_ctrl_free(struct kref *ref)
>   		container_of(ref, struct nvme_fc_ctrl, ref);
>   	unsigned long flags;
>   
> -	if (ctrl->ctrl.tagset)
> -		nvme_remove_io_tag_set(&ctrl->ctrl);
> -
>   	/* remove from rport list */
>   	spin_lock_irqsave(&ctrl->rport->lock, flags);
>   	list_del(&ctrl->ctrl_list);
>   	spin_unlock_irqrestore(&ctrl->rport->lock, flags);
>   
> -	nvme_unquiesce_admin_queue(&ctrl->ctrl);
> -	nvme_remove_admin_tag_set(&ctrl->ctrl);
> -
>   	kfree(ctrl->queues);
>   
>   	put_device(ctrl->dev);
> @@ -3288,6 +3282,13 @@ nvme_fc_delete_ctrl(struct nvme_ctrl *nctrl)
>   
>   	cancel_work_sync(&ctrl->ioerr_work);
>   	cancel_delayed_work_sync(&ctrl->connect_work);
> +
> +	if (ctrl->ctrl.tagset)
> +		nvme_remove_io_tag_set(&ctrl->ctrl);
> +
> +	nvme_unquiesce_admin_queue(&ctrl->ctrl);
> +	nvme_remove_admin_tag_set(&ctrl->ctrl);
> +
>   	/*
>   	 * kill the association on the link side.  this will block
>   	 * waiting for io to terminate

Not sure that'll work. We're waiting for I/Os to complete just after
the line in the last hunk, and I guess the need a tagset to complete
properly.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                  Kernel Storage Architect
hare at suse.de                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich