Re: [RFC PATCH 6/7] ublk: register buffer to specified io_uring & buf index via UBLK_F_AUTO_BUF_REG

[Caleb Sander Mateos <csander@purestorage.com>]

On Wed, Apr 30, 2025 at 8:45 AM Ming Lei <ming.lei@redhat.com> wrote:
>
> On Mon, Apr 28, 2025 at 05:52:28PM -0700, Caleb Sander Mateos wrote:
> > On Mon, Apr 28, 2025 at 2:45 AM Ming Lei <ming.lei@redhat.com> wrote:
> > >
> > > Add UBLK_F_AUTO_BUF_REG for supporting to register buffer automatically
> > > to specified io_uring context and buffer index.
> > >
> > > Signed-off-by: Ming Lei <ming.lei@redhat.com>
> > > ---
> > >  drivers/block/ublk_drv.c      | 56 ++++++++++++++++++++++++++++-------
> > >  include/uapi/linux/ublk_cmd.h | 38 ++++++++++++++++++++++++
> > >  2 files changed, 84 insertions(+), 10 deletions(-)
> > >
> > > diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> > > index 1fd20e481a60..e82618442749 100644
> > > --- a/drivers/block/ublk_drv.c
> > > +++ b/drivers/block/ublk_drv.c
> > > @@ -66,7 +66,8 @@
> > >                 | UBLK_F_USER_COPY \
> > >                 | UBLK_F_ZONED \
> > >                 | UBLK_F_USER_RECOVERY_FAIL_IO \
> > > -               | UBLK_F_UPDATE_SIZE)
> > > +               | UBLK_F_UPDATE_SIZE \
> > > +               | UBLK_F_AUTO_BUF_REG)
> > >
> > >  #define UBLK_F_ALL_RECOVERY_FLAGS (UBLK_F_USER_RECOVERY \
> > >                 | UBLK_F_USER_RECOVERY_REISSUE \
> > > @@ -146,7 +147,10 @@ struct ublk_uring_cmd_pdu {
> > >
> > >  struct ublk_io {
> > >         /* userspace buffer address from io cmd */
> > > -       __u64   addr;
> > > +       union {
> > > +               __u64   addr;
> > > +               struct ublk_auto_buf_reg buf;
> >
> > Maybe add a comment justifying why these fields can overlap? From my
> > understanding, buf is valid iff UBLK_F_AUTO_BUF_REG is set on the
> > ublk_queue and addr is valid iff neither UBLK_F_USER_COPY,
> > UBLK_F_SUPPORT_ZERO_COPY, nor UBLK_F_AUTO_BUF_REG is set.
>
> ->addr is for storing the userspace buffer, which is only used in
> non-zc cases(zc, auto_buf_reg) or user copy case.

Right, could you add a comment to that effect? I think using
overlapping fields is subtle and has the potential to break in the
future if the usage of the fields changes. Documenting the assumptions
clearly would go a long way.

>
> >
> > > +       };
> > >         unsigned int flags;
> > >         int res;
> > >
> > > @@ -626,7 +630,7 @@ static inline bool ublk_support_zero_copy(const struct ublk_queue *ubq)
> > >
> > >  static inline bool ublk_support_auto_buf_reg(const struct ublk_queue *ubq)
> > >  {
> > > -       return false;
> > > +       return ubq->flags & UBLK_F_AUTO_BUF_REG;
> > >  }
> > >
> > >  static inline bool ublk_support_user_copy(const struct ublk_queue *ubq)
> > > @@ -1177,6 +1181,16 @@ static inline void __ublk_abort_rq(struct ublk_queue *ubq,
> > >                 blk_mq_end_request(rq, BLK_STS_IOERR);
> > >  }
> > >
> > > +
> > > +static inline void ublk_init_auto_buf_reg(const struct ublk_io *io,
> > > +                                         struct io_buf_data *data)
> > > +{
> > > +       data->index = io->buf.index;
> > > +       data->ring_fd = io->buf.ring_fd;
> > > +       data->has_fd = true;
> > > +       data->registered_fd = io->buf.flags & UBLK_AUTO_BUF_REGISTERED_RING;
> > > +}
> > > +
> > >  static bool ublk_auto_buf_reg(struct ublk_queue *ubq, struct request *req,
> > >                               struct ublk_io *io, unsigned int issue_flags)
> > >  {
> > > @@ -1187,6 +1201,9 @@ static bool ublk_auto_buf_reg(struct ublk_queue *ubq, struct request *req,
> > >         };
> > >         int ret;
> > >
> > > +       if (ublk_support_auto_buf_reg(ubq))
> >
> > This check seems redundant with the check in the caller? Same comment
> > about ublk_auto_buf_unreg(). That would allow you to avoid adding the
> > ubq argument to ublk_auto_buf_unreg().
>
> Yeah, actually I removed one feature which just registers buffer to
> the uring command context, then forget to update the check.
>
> >
> > > +               ublk_init_auto_buf_reg(io, &data);
> > > +
> > >         /* one extra reference is dropped by ublk_io_release */
> > >         ublk_init_req_ref(ubq, req, 2);
> > >         ret = io_buffer_register_bvec(io->cmd, &data, issue_flags);
> > > @@ -2045,7 +2062,7 @@ static int ublk_fetch(struct io_uring_cmd *cmd, struct ublk_queue *ubq,
> > >                  */
> > >                 if (!buf_addr && !ublk_need_get_data(ubq))
> > >                         goto out;
> > > -       } else if (buf_addr) {
> > > +       } else if (buf_addr && !ublk_support_auto_buf_reg(ubq)) {
> > >                 /* User copy requires addr to be unset */
> > >                 ret = -EINVAL;
> > >                 goto out;
> > > @@ -2058,13 +2075,17 @@ static int ublk_fetch(struct io_uring_cmd *cmd, struct ublk_queue *ubq,
> > >         return ret;
> > >  }
> > >
> > > -static void ublk_auto_buf_unreg(struct ublk_io *io, struct io_uring_cmd *cmd,
> > > +static void ublk_auto_buf_unreg(const struct ublk_queue *ubq,
> > > +                               struct ublk_io *io, struct io_uring_cmd *cmd,
> > >                                 struct request *req, unsigned int issue_flags)
> > >  {
> > >         struct io_buf_data data = {
> > >                 .index = req->tag,
> > >         };
> > >
> > > +       if (ublk_support_auto_buf_reg(ubq))
> > > +               ublk_init_auto_buf_reg(io, &data);
> > > +
> > >         WARN_ON_ONCE(io_buffer_unregister_bvec(cmd, &data, issue_flags));
> > >         io->flags &= ~UBLK_IO_FLAG_AUTO_BUF_REG;
> > >  }
> > > @@ -2088,7 +2109,8 @@ static int ublk_commit_and_fetch(const struct ublk_queue *ubq,
> > >                 if (!ub_cmd->addr && (!ublk_need_get_data(ubq) ||
> > >                                         req_op(req) == REQ_OP_READ))
> > >                         return -EINVAL;
> > > -       } else if (req_op(req) != REQ_OP_ZONE_APPEND && ub_cmd->addr) {
> > > +       } else if ((req_op(req) != REQ_OP_ZONE_APPEND &&
> > > +                               !ublk_support_auto_buf_reg(ubq)) && ub_cmd->addr) {
> > >                 /*
> > >                  * User copy requires addr to be unset when command is
> > >                  * not zone append
> > > @@ -2097,7 +2119,7 @@ static int ublk_commit_and_fetch(const struct ublk_queue *ubq,
> > >         }
> > >
> > >         if (io->flags & UBLK_IO_FLAG_AUTO_BUF_REG)
> > > -               ublk_auto_buf_unreg(io, cmd, req, issue_flags);
> > > +               ublk_auto_buf_unreg(ubq, io, cmd, req, issue_flags);
> > >
> > >         ublk_fill_io_cmd(io, cmd, ub_cmd->addr);
> > >
> > > @@ -2788,6 +2810,11 @@ static int ublk_ctrl_add_dev(const struct ublksrv_ctrl_cmd *header)
> > >         else if (!(info.flags & UBLK_F_UNPRIVILEGED_DEV))
> > >                 return -EPERM;
> > >
> > > +       /* F_AUTO_BUF_REG and F_SUPPORT_ZERO_COPY can't co-exist */
> > > +       if ((info.flags & UBLK_F_AUTO_BUF_REG) &&
> > > +                       (info.flags & UBLK_F_SUPPORT_ZERO_COPY))
> > > +               return -EINVAL;
> > > +
> > >         /* forbid nonsense combinations of recovery flags */
> > >         switch (info.flags & UBLK_F_ALL_RECOVERY_FLAGS) {
> > >         case 0:
> > > @@ -2817,8 +2844,11 @@ static int ublk_ctrl_add_dev(const struct ublksrv_ctrl_cmd *header)
> > >                  * For USER_COPY, we depends on userspace to fill request
> > >                  * buffer by pwrite() to ublk char device, which can't be
> > >                  * used for unprivileged device
> > > +                *
> > > +                * Same with zero copy or auto buffer register.
> > >                  */
> > > -               if (info.flags & (UBLK_F_USER_COPY | UBLK_F_SUPPORT_ZERO_COPY))
> > > +               if (info.flags & (UBLK_F_USER_COPY | UBLK_F_SUPPORT_ZERO_COPY |
> > > +                                       UBLK_F_AUTO_BUF_REG))
> > >                         return -EINVAL;
> > >         }
> > >
> > > @@ -2876,17 +2906,22 @@ static int ublk_ctrl_add_dev(const struct ublksrv_ctrl_cmd *header)
> > >                 UBLK_F_URING_CMD_COMP_IN_TASK;
> > >
> > >         /* GET_DATA isn't needed any more with USER_COPY or ZERO COPY */
> > > -       if (ub->dev_info.flags & (UBLK_F_USER_COPY | UBLK_F_SUPPORT_ZERO_COPY))
> > > +       if (ub->dev_info.flags & (UBLK_F_USER_COPY | UBLK_F_SUPPORT_ZERO_COPY |
> > > +                               UBLK_F_AUTO_BUF_REG))
> > >                 ub->dev_info.flags &= ~UBLK_F_NEED_GET_DATA;
> > >
> > >         /*
> > >          * Zoned storage support requires reuse `ublksrv_io_cmd->addr` for
> > >          * returning write_append_lba, which is only allowed in case of
> > >          * user copy or zero copy
> > > +        *
> > > +        * UBLK_F_AUTO_BUF_REG can't be enabled for zoned because it need
> > > +        * the space for getting ring_fd and buffer index.
> > >          */
> > >         if (ublk_dev_is_zoned(ub) &&
> > >             (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) || !(ub->dev_info.flags &
> > > -            (UBLK_F_USER_COPY | UBLK_F_SUPPORT_ZERO_COPY)))) {
> > > +            (UBLK_F_USER_COPY | UBLK_F_SUPPORT_ZERO_COPY)) ||
> > > +            (ub->dev_info.flags & UBLK_F_AUTO_BUF_REG))) {
> > >                 ret = -EINVAL;
> > >                 goto out_free_dev_number;
> > >         }
> > > @@ -3403,6 +3438,7 @@ static int __init ublk_init(void)
> > >
> > >         BUILD_BUG_ON((u64)UBLKSRV_IO_BUF_OFFSET +
> > >                         UBLKSRV_IO_BUF_TOTAL_SIZE < UBLKSRV_IO_BUF_OFFSET);
> > > +       BUILD_BUG_ON(sizeof(struct ublk_auto_buf_reg) != sizeof(__u64));
> > >
> > >         init_waitqueue_head(&ublk_idr_wq);
> > >
> > > diff --git a/include/uapi/linux/ublk_cmd.h b/include/uapi/linux/ublk_cmd.h
> > > index be5c6c6b16e0..3d7c8c69cf06 100644
> > > --- a/include/uapi/linux/ublk_cmd.h
> > > +++ b/include/uapi/linux/ublk_cmd.h
> > > @@ -219,6 +219,30 @@
> > >   */
> > >  #define UBLK_F_UPDATE_SIZE              (1ULL << 10)
> > >
> > > +/*
> > > + * request buffer is registered automatically to ublk server specified
> > > + * io_uring context before delivering this io command to ublk server,
> > > + * meantime it is un-registered automatically when completing this io
> > > + * command.
> > > + *
> > > + * For using this feature:
> > > + *
> > > + * - ublk server has to create sparse buffer table
> > > + *
> > > + * - pass io_ring context FD from `ublksrv_io_cmd.buf.ring_fd`, and the FD
> > > + *   can be registered io_ring FD if `UBLK_AUTO_BUF_REGISTERED_RING` is set
> > > + *   in `ublksrv_io_cmd.flags`, or plain FD
> > > + *
> > > + * - pass buffer index from `ublksrv_io_cmd.buf.index`
> > > + *
> > > + * This way avoids extra cost from two uring_cmd, but also simplifies backend
> > > + * implementation, such as, the dependency on IO_REGISTER_IO_BUF and
> > > + * IO_UNREGISTER_IO_BUF becomes not necessary.
> > > + *
> > > + * This feature isn't available for UBLK_F_ZONED
> > > + */
> > > +#define UBLK_F_AUTO_BUF_REG    (1ULL << 11)
> > > +
> > >  /* device state */
> > >  #define UBLK_S_DEV_DEAD        0
> > >  #define UBLK_S_DEV_LIVE        1
> > > @@ -339,6 +363,14 @@ static inline __u32 ublksrv_get_flags(const struct ublksrv_io_desc *iod)
> > >         return iod->op_flags >> 8;
> > >  }
> > >
> > > +struct ublk_auto_buf_reg {
> > > +       __s32  ring_fd;
> > > +       __u16  index;
> > > +#define UBLK_AUTO_BUF_REGISTERED_RING            (1 << 0)
> > > +       __u8   flags;
> >
> > The flag could potentially be stored in ublk_io's flags field instead
> > to avoid taking up this byte.
>
> `ublk_auto_buf_reg` takes the exact ->addr space in both 'struct ublk_io'
> and 'struct ublksrv_io_cmd', this way won't take extra byte, but keep code simple
> and reuse for dealing with auto_buf_reg from both 'struct ublk_io' and
> 'struct ublksrv_io_cmd'.

Sure, either way is fine by me.

Best,
Caleb