Re: [PATCH v3 07/25] io_uring/kbuf: add recycling for kernel managed buffer rings

[Joanne Koong <joannelkoong@gmail.com>]

On Mon, Dec 29, 2025 at 2:20 PM Gabriel Krisman Bertazi <krisman@suse.de> wrote:
>
> Gabriel Krisman Bertazi <krisman@suse.de> writes:
>
> > Joanne Koong <joannelkoong@gmail.com> writes:
> >
> >> Add an interface for buffers to be recycled back into a kernel-managed
> >> buffer ring.
> >>
> >> This is a preparatory patch for fuse over io-uring.
> >>
> >> Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
> >> ---
> >>  include/linux/io_uring/cmd.h | 13 +++++++++++
> >>  io_uring/kbuf.c              | 42 ++++++++++++++++++++++++++++++++++++
> >>  io_uring/kbuf.h              |  3 +++
> >>  io_uring/uring_cmd.c         | 11 ++++++++++
> >>  4 files changed, 69 insertions(+)
> >>
> >> diff --git a/include/linux/io_uring/cmd.h b/include/linux/io_uring/cmd.h
> >> index 424f071f42e5..7169a2a9a744 100644
> >> --- a/include/linux/io_uring/cmd.h
> >> +++ b/include/linux/io_uring/cmd.h
> >> @@ -88,6 +88,11 @@ int io_uring_cmd_buf_ring_pin(struct io_uring_cmd *ioucmd, unsigned buf_group,
> >>                            unsigned issue_flags, struct io_buffer_list **bl);
> >>  int io_uring_cmd_buf_ring_unpin(struct io_uring_cmd *ioucmd, unsigned buf_group,
> >>                              unsigned issue_flags);
> >> +
> >> +int io_uring_cmd_kmbuffer_recycle(struct io_uring_cmd *cmd,
> >> +                              unsigned int buf_group, u64 addr,
> >> +                              unsigned int len, unsigned int bid,
> >> +                              unsigned int issue_flags);
> >>  #else
> >>  static inline int
> >>  io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
> >> @@ -143,6 +148,14 @@ static inline int io_uring_cmd_buf_ring_unpin(struct io_uring_cmd *ioucmd,
> >>  {
> >>      return -EOPNOTSUPP;
> >>  }
> >> +static inline int io_uring_cmd_kmbuffer_recycle(struct io_uring_cmd *cmd,
> >> +                                            unsigned int buf_group,
> >> +                                            u64 addr, unsigned int len,
> >> +                                            unsigned int bid,
> >> +                                            unsigned int issue_flags)
> >> +{
> >> +    return -EOPNOTSUPP;
> >> +}
> >>  #endif
> >>
> >>  static inline struct io_uring_cmd *io_uring_cmd_from_tw(struct io_tw_req tw_req)
> >> diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c
> >> index 03e05bab023a..f12d000b71c5 100644
> >> --- a/io_uring/kbuf.c
> >> +++ b/io_uring/kbuf.c
> >> @@ -101,6 +101,48 @@ void io_kbuf_drop_legacy(struct io_kiocb *req)
> >>      req->kbuf = NULL;
> >>  }
> >>
> >> +int io_kmbuf_recycle(struct io_kiocb *req, unsigned int bgid, u64 addr,
> >> +                 unsigned int len, unsigned int bid,
> >> +                 unsigned int issue_flags)
> >> +{
> >> +    struct io_ring_ctx *ctx = req->ctx;
> >> +    struct io_uring_buf_ring *br;
> >> +    struct io_uring_buf *buf;
> >> +    struct io_buffer_list *bl;
> >> +    int ret = -EINVAL;
> >> +
> >> +    if (WARN_ON_ONCE(req->flags & REQ_F_BUFFERS_COMMIT))
> >> +            return ret;
> >> +
> >> +    io_ring_submit_lock(ctx, issue_flags);
> >> +
> >> +    bl = io_buffer_get_list(ctx, bgid);
> >> +
> >> +    if (WARN_ON_ONCE(!(bl->flags & IOBL_BUF_RING)) ||
> >> +        WARN_ON_ONCE(!(bl->flags & IOBL_KERNEL_MANAGED)))
> >> +            goto done;
> >
> > Hi Joanne,
> >
> > WARN_ONs are not supposed to be reached by the user, but I think that is
> > possible here, i.e. by passing the bgid of legacy provided buffers.
>
> But now I see this is never exposed to userspace as an io_uring_cmd
> command itself, it is only used internally by other fuse operations.
> Nevertheless, it's implemented as an io_uring_cmd by
> io_uring_cmd_kmbuffer_recycle.
>
> Is it eventually going to be exposed as operations to userspace? If not,
> I'd suggest to stay out of the io_uring_cmd namespace (perhaps call
> io_kmbuf_recycle directly from fs/fuse).  Do we need to have this
> io_uring_cmd abstraction for some reason I'm missing?

Hi Gabriel,

Thanks for taking a look at the patchset.

This is not going to be exposed as an operation to userspace. Only the
kernel will be able to recycle kmbufs.

I was under the impression the io_uring_cmd_* abstraction was
preferred as the API for interfacing with io_uring from another
subsystem. In that case then, I'll get rid of the io_uring_cmd layers
for the calls then, that will make things simpler.

Thanks,
Joanne

>
> Thanks,
>
> --
> Gabriel Krisman Bertazi