public inbox for [email protected]
 help / color / mirror / Atom feed
From: Ming Lei <[email protected]>
To: Jens Axboe <[email protected]>
Cc: [email protected], [email protected],
	Gabriel Krisman Bertazi <[email protected]>,
	[email protected]
Subject: Re: [PATCH V2] io_uring: cancelable uring_cmd
Date: Fri, 22 Sep 2023 23:24:01 +0800	[thread overview]
Message-ID: <ZQ2xkRRO6boDo6LD@fedora> (raw)
In-Reply-To: <[email protected]>

On Fri, Sep 22, 2023 at 09:02:44AM -0600, Jens Axboe wrote:
> On 9/22/23 8:28 AM, Ming Lei wrote:
> > uring_cmd may never complete, such as ublk, in which uring cmd isn't
> > completed until one new block request is coming from ublk block device.
> > 
> > Add cancelable uring_cmd to provide mechanism to driver for cancelling
> > pending commands in its own way.
> > 
> > Add API of io_uring_cmd_mark_cancelable() for driver to mark one command as
> > cancelable, then io_uring will cancel this command in
> > io_uring_cancel_generic(). ->uring_cmd() callback is reused for canceling
> > command in driver's way, then driver gets notified with the cancelling
> > from io_uring.
> > 
> > Add API of io_uring_cmd_get_task() to help driver cancel handler
> > deal with the canceling.
> 
> This looks better, a few comments:
> 
> > diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> > index 8e61f8b7c2ce..29a7a7e71f57 100644
> > --- a/include/uapi/linux/io_uring.h
> > +++ b/include/uapi/linux/io_uring.h
> > @@ -249,10 +249,13 @@ enum io_uring_op {
> >   * sqe->uring_cmd_flags
> >   * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
> >   *				along with setting sqe->buf_index.
> > + * IORING_URING_CANCELABLE	not for userspace
> >   * IORING_URING_CMD_POLLED	driver use only
> >   */
> > -#define IORING_URING_CMD_FIXED	(1U << 0)
> > -#define IORING_URING_CMD_POLLED	(1U << 31)
> > +#define IORING_URING_CMD_FIXED		(1U << 0)
> > +/* set by driver, and handled by io_uring to cancel this cmd */
> > +#define IORING_URING_CMD_CANCELABLE	(1U << 30)
> > +#define IORING_URING_CMD_POLLED		(1U << 31)
> 
> If IORING_URING_CANCELABLE isn't UAPI, why stuff it in here? Should we
> have a split where we retain the upper 8 bits for internal use, or
> something like that?

Yeah, it is for internal use, same with IORING_URING_CMD_POLLED.

I think we can retain upper 8 bits for internal use, and move
the two definitions into kernel header.

> 
> > diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
> > index 783ed0fff71b..a3135fd47a4e 100644
> > --- a/io_uring/io_uring.c
> > +++ b/io_uring/io_uring.c
> > @@ -3256,6 +3256,40 @@ static __cold bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
> >  	return ret;
> >  }
> >  
> > +static bool io_uring_try_cancel_uring_cmd(struct io_ring_ctx *ctx,
> > +		struct task_struct *task, bool cancel_all)
> > +{
> > +	struct hlist_node *tmp;
> > +	struct io_kiocb *req;
> > +	bool ret = false;
> > +
> > +	mutex_lock(&ctx->uring_lock);
> > +	hlist_for_each_entry_safe(req, tmp, &ctx->cancelable_uring_cmd,
> > +			hash_node) {
> > +		struct io_uring_cmd *cmd = io_kiocb_to_cmd(req,
> > +				struct io_uring_cmd);
> > +		struct file *file = req->file;
> > +
> > +		if (WARN_ON_ONCE(!file->f_op->uring_cmd))
> > +			continue;
> > +
> > +		if (!cancel_all && req->task != task)
> > +			continue;
> > +
> > +		if (cmd->flags & IORING_URING_CMD_CANCELABLE) {
> > +			/* ->sqe isn't available if no async data */
> > +			if (!req_has_async_data(req))
> > +				cmd->sqe = NULL;
> > +			file->f_op->uring_cmd(cmd, IO_URING_F_CANCEL);
> > +			ret = true;
> > +		}
> > +	}
> > +	io_submit_flush_completions(ctx);
> > +	mutex_unlock(&ctx->uring_lock);
> > +
> > +	return ret;
> > +}
> 
> I think it'd be saner to drop uring_lock here, and then:
> 
> > @@ -3307,6 +3341,7 @@ static __cold bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
> >  	ret |= io_kill_timeouts(ctx, task, cancel_all);
> >  	if (task)
> >  		ret |= io_run_task_work() > 0;
> > +	ret |= io_uring_try_cancel_uring_cmd(ctx, task, cancel_all);
> >  	return ret;
> >  }
> 
> move this hunk into the uring_lock section. Also ensure that we do run
> task_work for cancelation, should the uring_cmd side require that
> (either now or eventually).

OK, io_run_task_work() has run task_work already.

> 
> > diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
> > index 537795fddc87..d6b200a0be33 100644
> > --- a/io_uring/uring_cmd.c
> > +++ b/io_uring/uring_cmd.c
> > @@ -13,6 +13,52 @@
> >  #include "rsrc.h"
> >  #include "uring_cmd.h"
> >  
> > +static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd,
> > +		unsigned int issue_flags)
> > +{
> > +	if (cmd->flags & IORING_URING_CMD_CANCELABLE) {
> > +		struct io_kiocb *req = cmd_to_io_kiocb(cmd);
> > +		struct io_ring_ctx *ctx = req->ctx;
> > +
> > +		io_ring_submit_lock(ctx, issue_flags);
> > +		cmd->flags &= ~IORING_URING_CMD_CANCELABLE;
> > +		hlist_del(&req->hash_node);
> > +		io_ring_submit_unlock(ctx, issue_flags);
> > +	}
> > +}
> 
> static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd,
> 		unsigned int issue_flags)
> {
> 	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
> 	struct io_ring_ctx *ctx = req->ctx;
> 
> 	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE))
> 		return;
> 
> 	io_ring_submit_lock(ctx, issue_flags);
> 	cmd->flags &= ~IORING_URING_CMD_CANCELABLE;
> 	hlist_del(&req->hash_node);
> 	io_ring_submit_unlock(ctx, issue_flags);
> }
> 
> is cleaner imho. Minor nit.
> 
> > +
> > +/*
> > + * Mark this command as concelable, then io_uring_try_cancel_uring_cmd()
> > + * will try to cancel this issued command by sending ->uring_cmd() with
> > + * issue_flags of IO_URING_F_CANCEL.
> > + *
> > + * The command is guaranteed to not be done when calling ->uring_cmd()
> > + * with IO_URING_F_CANCEL, but it is driver's responsibility to deal
> > + * with race between io_uring canceling and normal completion.
> > + */
> > +int io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
> > +		unsigned int issue_flags)
> > +{
> > +	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
> > +	struct io_ring_ctx *ctx = req->ctx;
> > +
> > +	io_ring_submit_lock(ctx, issue_flags);
> > +	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
> > +		cmd->flags |= IORING_URING_CMD_CANCELABLE;
> > +		hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
> > +	}
> > +	io_ring_submit_unlock(ctx, issue_flags);
> > +
> > +	return 0;
> > +}
> 
> A bit inconsistent here in terms of the locking. I'm assuming the
> marking happens within issue, in which case it should be fine to do:
> 
> int io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
> 		unsigned int issue_flags)
> {
> 	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
> 	struct io_ring_ctx *ctx = req->ctx;
> 
> 	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
> 		cmd->flags |= IORING_URING_CMD_CANCELABLE;
> 		io_ring_submit_lock(ctx, issue_flags);
> 		hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
> 		io_ring_submit_unlock(ctx, issue_flags);
> 	}

OK, mask & clear bit can be moved out of lock in both two functions.


Thanks,
Ming


      reply	other threads:[~2023-09-22 15:25 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-09-22 14:28 [PATCH V2] io_uring: cancelable uring_cmd Ming Lei
2023-09-22 15:02 ` Jens Axboe
2023-09-22 15:24   ` Ming Lei [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=ZQ2xkRRO6boDo6LD@fedora \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox