linux-kernel - Re: [PATCH v2 3/3] io_uring/uring_cmd: avoid double indirect call in task work dispatch

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CADUfDZp21icTKrWHcgRTfmsxtdab85b6R75wAYXW2dA+dzXmoA@mail.gmail.com>
Date: Thu, 23 Oct 2025 20:49:40 -0700
From: Caleb Sander Mateos <csander@...estorage.com>
To: Ming Lei <ming.lei@...hat.com>
Cc: Jens Axboe <axboe@...nel.dk>, Miklos Szeredi <miklos@...redi.hu>, Keith Busch <kbusch@...nel.org>, 
	Christoph Hellwig <hch@....de>, Sagi Grimberg <sagi@...mberg.me>, Chris Mason <clm@...com>, 
	David Sterba <dsterba@...e.com>, io-uring@...r.kernel.org, linux-fsdevel@...r.kernel.org, 
	linux-block@...r.kernel.org, linux-nvme@...ts.infradead.org, 
	linux-btrfs@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 3/3] io_uring/uring_cmd: avoid double indirect call in
 task work dispatch

On Thu, Oct 23, 2025 at 8:42 PM Ming Lei <ming.lei@...hat.com> wrote:
>
> On Thu, Oct 23, 2025 at 02:18:30PM -0600, Caleb Sander Mateos wrote:
> > io_uring task work dispatch makes an indirect call to struct io_kiocb's
> > io_task_work.func field to allow running arbitrary task work functions.
> > In the uring_cmd case, this calls io_uring_cmd_work(), which immediately
> > makes another indirect call to struct io_uring_cmd's task_work_cb field.
> > Define the uring_cmd task work callbacks as functions whose signatures
> > match io_req_tw_func_t. Define a IO_URING_CMD_TASK_WORK_ISSUE_FLAGS
> > constant in io_uring/cmd.h to avoid manufacturing issue_flags in the
> > uring_cmd task work callbacks. Now uring_cmd task work dispatch makes a
> > single indirect call to the uring_cmd implementation's callback. This
> > also allows removing the task_work_cb field from struct io_uring_cmd,
> > freeing up some additional storage space.
>
> The idea looks good.
>
> >
> > Signed-off-by: Caleb Sander Mateos <csander@...estorage.com>
> > ---
> >  block/ioctl.c                |  4 +++-
> >  drivers/block/ublk_drv.c     | 15 +++++++++------
> >  drivers/nvme/host/ioctl.c    |  5 +++--
> >  fs/btrfs/ioctl.c             |  4 +++-
> >  fs/fuse/dev_uring.c          |  5 +++--
> >  include/linux/io_uring/cmd.h | 16 +++++++---------
> >  io_uring/uring_cmd.c         | 13 ++-----------
> >  7 files changed, 30 insertions(+), 32 deletions(-)
> >
> > diff --git a/block/ioctl.c b/block/ioctl.c
> > index d7489a56b33c..5c10d48fab27 100644
> > --- a/block/ioctl.c
> > +++ b/block/ioctl.c
> > @@ -767,13 +767,15 @@ long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
> >  struct blk_iou_cmd {
> >       int res;
> >       bool nowait;
> >  };
> >
> > -static void blk_cmd_complete(struct io_uring_cmd *cmd, unsigned int issue_flags)
> > +static void blk_cmd_complete(struct io_kiocb *req, io_tw_token_t tw)
> >  {
> > +     struct io_uring_cmd *cmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
> >       struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
> > +     unsigned int issue_flags = IO_URING_CMD_TASK_WORK_ISSUE_FLAGS;
>
> Now `io_kiocb` is exposed to driver, it could be perfect if 'io_uring_cmd'
> is kept in kernel API interface, IMO.

You mean change the io_req_tw_func_t signature to pass struct
io_uring_cmd * instead of struct io_kiocb *? I don't think that would
make sense because task work is a more general concept, not just for
uring_cmd. I agree it's a bit ugly exposing struct io_kiocb * outside
of the io_uring core, but I don't see a way to encapsulate it without
other downsides (the additional indirect call or the gross macro from
v1). Treating it as an opaque pointer type seems like the least bad
option...

>
> ...
>
> > diff --git a/include/linux/io_uring/cmd.h b/include/linux/io_uring/cmd.h
> > index b84b97c21b43..3efad93404f9 100644
> > --- a/include/linux/io_uring/cmd.h
> > +++ b/include/linux/io_uring/cmd.h
> > @@ -9,18 +9,13 @@
> >  /* only top 8 bits of sqe->uring_cmd_flags for kernel internal use */
> >  #define IORING_URING_CMD_CANCELABLE  (1U << 30)
> >  /* io_uring_cmd is being issued again */
> >  #define IORING_URING_CMD_REISSUE     (1U << 31)
> >
> > -typedef void (*io_uring_cmd_tw_t)(struct io_uring_cmd *cmd,
> > -                               unsigned issue_flags);
> > -
> >  struct io_uring_cmd {
> >       struct file     *file;
> >       const struct io_uring_sqe *sqe;
> > -     /* callback to defer completions to task context */
> > -     io_uring_cmd_tw_t task_work_cb;
> >       u32             cmd_op;
> >       u32             flags;
> >       u8              pdu[32]; /* available inline for free use */
>
> pdu[40]

I considered that, but wondered if we might want to reuse the 8 bytes
for something internal to uring_cmd rather than providing it to the
driver's uring_cmd implementation. If we increase pdu and a driver
starts using more than 32 bytes, it will be difficult to claw back. It
seems reasonable to reserve half the space for the io_uring/uring_cmd
layer and half for the driver.

Best,
Caleb

>
>
>
> Thanks,
> Ming
>