Re: [PATCH v2] io_uring: introduce add/post_event_and_complete function

[Pavel Begunkov <asml.silence@gmail.com>]

On 11/25/2019 12:04 PM, Bob Liu wrote:
> * Only complie-tested right now. *
> There are many duplicated code doing add/post event, set REQ_F_FAIL_LINK and
> then put req. Put them into common funcs io_cqring_event_posted_and_complete()
> and io_cqring_add_event_and_complete().

There are some comments below. This one looks much better, but I think,
it's still not self-expressing enough. And it looks like
io_cqring_add_event_and_complete() is doing 2 orthogonal things, mostly
because we anyway need to do 2+ puts() for a single request.

How about to try extract something like io_fail_event(req, err_code)
first, which fails it unconditionally? As you noticed, in the most cases
setting REQ_F_FAIL_LINK is followed by io_cqring_add_event().

It may also be interesting to set REQ_F_FAIL_LINK for all request types,
not only for linked (i.e. REQ_F_FAIL). This may (or may not...) prove to
be clearer.

> 
> Signed-off-by: Bob Liu <bob.liu@oracle.com>
> ---
> Changes since v1:
> - Using FAIL_LINK_* enums.
> 
> ---
>  fs/io_uring.c | 131 +++++++++++++++++++++++++++++-----------------------------
>  1 file changed, 65 insertions(+), 66 deletions(-)
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 9f9c2d4..d91864e 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -363,6 +363,12 @@ struct io_kiocb {
>  	struct io_wq_work	work;
>  };
>  
> +enum set_f_fail_link {
> +	FAIL_LINK_NONE,
> +	FAIL_LINK_ON_NEGATIVE,
> +	FAIL_LINK_ON_NOTEQUAL_RES,
> +	FAIL_LINK_ALWAYS,
> +};
>  #define IO_PLUG_THRESHOLD		2
>  #define IO_IOPOLL_BATCH			8
>  
> @@ -1253,34 +1259,66 @@ static void kiocb_end_write(struct io_kiocb *req)
>  	file_end_write(req->file);
>  }
>  
> -static void io_complete_rw_common(struct kiocb *kiocb, long res)
> +void set_f_fail_link(struct io_kiocb *req, long ret, unsigned int fail_link)
> +{

if (!(req->flags & REQ_F_LINK))
	return;

Less bulky and is good for fast-path. And this single check
_should_ be inlined. Could you check assembly for that?

> +	if (fail_link == FAIL_LINK_ALWAYS) {
> +		if (req->flags & REQ_F_LINK)
> +			req->flags |= REQ_F_FAIL_LINK;
> +	} else if (fail_link == FAIL_LINK_ON_NEGATIVE) {
> +		if (ret < 0 && (req->flags & REQ_F_LINK))
> +			req->flags |= REQ_F_FAIL_LINK;
> +	} else if (fail_link == FAIL_LINK_ON_NOTEQUAL_RES) {
> +		if ((ret != req->result) && (req->flags & REQ_F_LINK))
> +			req->flags |= REQ_F_FAIL_LINK;
> +	}
> +}
> +
> +static void io_cqring_add_event_and_complete(struct io_kiocb *req, long ret,
> +		unsigned int fail_link, struct io_kiocb **nxt)

io_cqring_add_event_and_put() maybe?
That's what it really do.

> +{
> +	set_f_fail_link(req, ret, fail_link);
> +	io_cqring_add_event(req, ret);
> +
> +	if (nxt)
> +		io_put_req_find_next(req, nxt);

Just removed this pattern with possible nxt==NULL.
It always tends getting nested multiplying such checks.

> +	else
> +		io_put_req(req);
> +}
> +
> +static void io_cqring_ev_posted_and_complete(struct io_kiocb *req, long ret,
> +		unsigned int fail_link, struct io_kiocb **nxt)
> +{
> +	io_cqring_ev_posted(req->ctx);
> +	set_f_fail_link(req, ret, fail_link);
> +
> +	if (nxt)
> +		io_put_req_find_next(req, nxt);
> +	else
> +		io_put_req(req);
> +}
> +
> +static void io_complete_rw_common(struct kiocb *kiocb, long res,
> +		struct io_kiocb **nxt)
>  {
>  	struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw);
>  
>  	if (kiocb->ki_flags & IOCB_WRITE)
>  		kiocb_end_write(req);
>  
> -	if ((req->flags & REQ_F_LINK) && res != req->result)
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_cqring_add_event(req, res);
> +	io_cqring_add_event_and_complete(req, res,
> +			FAIL_LINK_ON_NOTEQUAL_RES, nxt);
>  }
>  
>  static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
>  {
> -	struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw);
> -
> -	io_complete_rw_common(kiocb, res);
> -	io_put_req(req);
> +	io_complete_rw_common(kiocb, res, NULL);
>  }
>  
>  static struct io_kiocb *__io_complete_rw(struct kiocb *kiocb, long res)
>  {
> -	struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw);
>  	struct io_kiocb *nxt = NULL;
>  
> -	io_complete_rw_common(kiocb, res);
> -	io_put_req_find_next(req, &nxt);
> -
> +	io_complete_rw_common(kiocb, res, &nxt);
>  	return nxt;
>  }
>  
> @@ -1831,10 +1869,7 @@ static int io_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>  				end > 0 ? end : LLONG_MAX,
>  				fsync_flags & IORING_FSYNC_DATASYNC);
>  
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_cqring_add_event(req, ret);
> -	io_put_req_find_next(req, nxt);
> +	io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, nxt);
>  	return 0;
>  }
>  
> @@ -1878,10 +1913,7 @@ static int io_sync_file_range(struct io_kiocb *req,
>  
>  	ret = sync_file_range(req->rw.ki_filp, sqe_off, sqe_len, flags);
>  
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_cqring_add_event(req, ret);
> -	io_put_req_find_next(req, nxt);
> +	io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, nxt);
>  	return 0;
>  }
>  
> @@ -1916,10 +1948,7 @@ static int io_send_recvmsg(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>  			return ret;
>  	}
>  
> -	io_cqring_add_event(req, ret);
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_put_req_find_next(req, nxt);
> +	io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, nxt);
>  	return 0;
>  }
>  #endif
> @@ -1972,10 +2001,7 @@ static int io_accept(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>  	}
>  	if (ret == -ERESTARTSYS)
>  		ret = -EINTR;
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_cqring_add_event(req, ret);
> -	io_put_req_find_next(req, nxt);
> +	io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, nxt);
>  	return 0;
>  #else
>  	return -EOPNOTSUPP;
> @@ -2005,10 +2031,8 @@ static int io_connect(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>  		return -EAGAIN;
>  	if (ret == -ERESTARTSYS)
>  		ret = -EINTR;
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_cqring_add_event(req, ret);
> -	io_put_req_find_next(req, nxt);
> +
> +	io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, nxt);
>  	return 0;
>  #else
>  	return -EOPNOTSUPP;
> @@ -2091,10 +2115,7 @@ static int io_poll_remove(struct io_kiocb *req, const struct io_uring_sqe *sqe)
>  	ret = io_poll_cancel(ctx, READ_ONCE(sqe->addr));
>  	spin_unlock_irq(&ctx->completion_lock);
>  
> -	io_cqring_add_event(req, ret);
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_put_req(req);
> +	io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, NULL);
>  	return 0;
>  }
>  
> @@ -2148,11 +2169,7 @@ static void io_poll_complete_work(struct io_wq_work **workptr)
>  	io_poll_complete(req, mask, ret);
>  	spin_unlock_irq(&ctx->completion_lock);
>  
> -	io_cqring_ev_posted(ctx);
> -
> -	if (ret < 0 && req->flags & REQ_F_LINK)
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_put_req_find_next(req, &nxt);
> +	io_cqring_ev_posted_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, &nxt);
>  	if (nxt)
>  		*workptr = &nxt->work;
>  }
> @@ -2338,10 +2355,7 @@ static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
>  	io_commit_cqring(ctx);
>  	spin_unlock_irqrestore(&ctx->completion_lock, flags);
>  
> -	io_cqring_ev_posted(ctx);
> -	if (req->flags & REQ_F_LINK)
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_put_req(req);
> +	io_cqring_ev_posted_and_complete(req, 0, FAIL_LINK_ALWAYS, NULL);
>  	return HRTIMER_NORESTART;
>  }
>  
> @@ -2396,10 +2410,7 @@ static int io_timeout_remove(struct io_kiocb *req,
>  	io_cqring_fill_event(req, ret);
>  	io_commit_cqring(ctx);
>  	spin_unlock_irq(&ctx->completion_lock);
> -	io_cqring_ev_posted(ctx);
> -	if (ret < 0 && req->flags & REQ_F_LINK)
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_put_req(req);
> +	io_cqring_ev_posted_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, NULL);
>  	return 0;
>  }
>  
> @@ -2560,11 +2571,7 @@ static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
>  	io_cqring_fill_event(req, ret);
>  	io_commit_cqring(ctx);
>  	spin_unlock_irqrestore(&ctx->completion_lock, flags);
> -	io_cqring_ev_posted(ctx);
> -
> -	if (ret < 0 && (req->flags & REQ_F_LINK))
> -		req->flags |= REQ_F_FAIL_LINK;
> -	io_put_req_find_next(req, nxt);
> +	io_cqring_ev_posted_and_complete(req, ret, FAIL_LINK_ON_NEGATIVE, nxt);
>  }
>  
>  static int io_async_cancel(struct io_kiocb *req, const struct io_uring_sqe *sqe,
> @@ -2738,12 +2745,8 @@ static void io_wq_submit_work(struct io_wq_work **workptr)
>  	/* drop submission reference */
>  	io_put_req(req);
>  
> -	if (ret) {
> -		if (req->flags & REQ_F_LINK)
> -			req->flags |= REQ_F_FAIL_LINK;
> -		io_cqring_add_event(req, ret);
> -		io_put_req(req);
> -	}
> +	if (ret)
> +		io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ALWAYS, NULL);
>  
>  	/* if a dependent link is ready, pass it back */
>  	if (!ret && nxt) {
> @@ -2981,12 +2984,8 @@ static void __io_queue_sqe(struct io_kiocb *req)
>  	}
>  
>  	/* and drop final reference, if we failed */
> -	if (ret) {
> -		io_cqring_add_event(req, ret);
> -		if (req->flags & REQ_F_LINK)
> -			req->flags |= REQ_F_FAIL_LINK;
> -		io_put_req(req);
> -	}
> +	if (ret)
> +		io_cqring_add_event_and_complete(req, ret, FAIL_LINK_ALWAYS, NULL);
>  }
>  
>  static void io_queue_sqe(struct io_kiocb *req)
> 

-- 
Pavel Begunkov