Re: [PATCH 1/6] block: Add blk_completion

[Christoph Hellwig <hch@infradead.org>]

On Thu, Oct 22, 2020 at 10:22:23PM +0100, Matthew Wilcox (Oracle) wrote:
> This new data structure allows a task to wait for N things to complete.
> Usually the submitting task will handle cleanup, but if it is killed,
> the last completer will take care of it.
> 
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
>  block/blk-core.c    | 61 +++++++++++++++++++++++++++++++++++++++++++++
>  include/linux/bio.h | 11 ++++++++
>  2 files changed, 72 insertions(+)
> 
> diff --git a/block/blk-core.c b/block/blk-core.c
> index 10c08ac50697..2892246f2176 100644
> --- a/block/blk-core.c
> +++ b/block/blk-core.c
> @@ -1900,6 +1900,67 @@ void blk_io_schedule(void)
>  }
>  EXPORT_SYMBOL_GPL(blk_io_schedule);
>  
> +void blk_completion_init(struct blk_completion *cmpl, int n)
> +{
> +	spin_lock_init(&cmpl->cmpl_lock);
> +	cmpl->cmpl_count = n;
> +	cmpl->cmpl_task = current;
> +	cmpl->cmpl_status = BLK_STS_OK;
> +}
> +
> +int blk_completion_sub(struct blk_completion *cmpl, blk_status_t status, int n)

This needs documentation.  e.g. to explain what 'n' is.

> +int blk_completion_wait_killable(struct blk_completion *cmpl)
> +{
> +	int err = 0;
> +
> +	for (;;) {
> +		set_current_state(TASK_KILLABLE);
> +		spin_lock_bh(&cmpl->cmpl_lock);
> +		if (cmpl->cmpl_count == 0)
> +			break;
> +		spin_unlock_bh(&cmpl->cmpl_lock);
> +		blk_io_schedule();
> +		if (fatal_signal_pending(current)) {
> +			spin_lock_bh(&cmpl->cmpl_lock);
> +			cmpl->cmpl_task = NULL;
> +			if (cmpl->cmpl_count != 0) {
> +				spin_unlock_bh(&cmpl->cmpl_lock);
> +				cmpl = NULL;
> +			}
> +			err = -ERESTARTSYS;
> +			break;
> +		}
> +	}
> +	set_current_state(TASK_RUNNING);
> +	if (cmpl) {
> +		spin_unlock_bh(&cmpl->cmpl_lock);
> +		err = blk_status_to_errno(cmpl->cmpl_status);
> +		kfree(cmpl);
> +	}
> +
> +	return err;
> +}

What are the life time rules for cmpl?  Who frees it in the case
of a fatal signal?