[GIT PULL] io_uring dma-buf support for 7.4

[GIT PULL] io_uring dma-buf support for 7.4

[Jens Axboe]

Hi Linus,

This adds native dma-buf support to the I/O path, exposed through
io_uring. It's been in development for a while and should be ready
to go now. Since it depends on both the io_uring and block trees,
I've deferred sending it out until those were merged. This pull
request contains:

- dma-buf file I/O infrastructure: a dma_buf_io_ctx created through a
  new ->init_dma_buf_io_ctx file operation, and a dma_buf_io_map holding
  the device DMA mapping. The two are split so a mapping can be
  invalidated and re-established underneath a live context, with the map
  counting in-flight requests so it can be torn down safely.

- A new iov_iter type for dmabuf maps, and a dma map backed bio type for
  the block layer, with raw block device support on top.

- nvme-pci backend implementing dma-buf backed requests, including SGL
  support.

- io_uring dmabuf backed registered buffers, built on a reworked buffer
  registration structure with an extended update path, regbuf import
  flags, an uncloneable flag for buffer types that can't be cloned, and
  cloning support for registered dma-bufs.

Note that this conflicts with block/fops.c in your tree, as the dma-buf
rejection in the buffered write fallback landed next to the NOWAIT and
i_rwsem changes from the block pull. Fixup is trivial, it's just adding
the iov_iter_is_dmabuf_map() check in the changed
blkdev_buffered_write().

Please pull!


The following changes since commit e889b0e4cafddb3d68c3595dc01b7128fd35fbef:

  io_uring/io-wq: don't call try_to_wake_up() twice in io_wq_worker_wake() (2026-10-12 10:19:52 -0600)

are available in the Git repository at:

  https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git tags/for-7.4/io_uring-dmabuf-20261023

for you to fetch changes up to 45b716456dd99488a0cef58482120dc0474aa696:

  io_uring/dma-buf: don't fail fixed buffers without dma-buf support (2026-10-22 09:27:00 -0600)

----------------------------------------------------------------
for-7.4/io_uring-dmabuf-20261023

----------------------------------------------------------------
Anuj Gupta (3):
      nvme-pci: rename nvme_pci_sgl_set_data to nvme_pci_dma_iter_set_sgl
      nvme-pci: add SGL support for the dmabuf path
      io_uring/dma-buf: don't fail fixed buffers without dma-buf support

Keith Busch (2):
      block: reject dma-bufs larger than 4GiB
      nvme-pci: detach dma-bufs only on removal

Pavel Begunkov (15):
      dma-buf: introduce initial file I/O infrastructure
      iov_iter: add iterator type for dmabuf maps
      block: always adjust bi_offset on bio_advance_iter
      block: introduce dma map backed bio type
      block: add dma-buf support for raw bdev
      nvme-pci: implement dma-buf backed requests
      io_uring/rsrc: introduce buf registration structure
      io_uring/rsrc: extend buffer update
      io_uring/rsrc: add uncloneable regbuf flag
      io_uring/rsrc: add regbuf import flags
      io_uring/rsrc: add dmabuf backed registered buffers
      block: fix indention for dma-buf checks
      dma-buf-io: fix typos in kernel docs
      dma-buf-io: document dma_buf_io_init_map
      io_uring/dma-buf: allow cloning registered dma-bufs

 block/bio.c                    |  15 +-
 block/blk-merge.c              |  52 +++++++
 block/blk-mq.c                 |   6 +
 block/fops.c                   |  29 +++-
 drivers/dma-buf/Makefile       |   2 +-
 drivers/dma-buf/dma-buf-io.c   | 241 ++++++++++++++++++++++++++++++++
 drivers/md/dm-io-rewind.c      |   6 +-
 drivers/nvme/host/Makefile     |   1 +
 drivers/nvme/host/core.c       |  12 ++
 drivers/nvme/host/nvme.h       |   2 +
 drivers/nvme/host/pci-dmabuf.c | 169 ++++++++++++++++++++++
 drivers/nvme/host/pci-dmabuf.h |  75 ++++++++++
 drivers/nvme/host/pci.c        | 247 ++++++++++++++++++++++++++++++++-
 include/linux/bio.h            |  21 +--
 include/linux/blk-mq.h         |   7 +
 include/linux/blk_types.h      |  14 +-
 include/linux/blkdev.h         |   2 +
 include/linux/bvec.h           |   3 +-
 include/linux/dma-buf-io.h     | 126 +++++++++++++++++
 include/linux/fs.h             |   2 +
 include/linux/io_uring_types.h |   5 +
 include/linux/iov_iter.h       |   2 +
 include/linux/uio.h            |  11 ++
 include/uapi/linux/io_uring.h  |  31 ++++-
 io_uring/Makefile              |   1 +
 io_uring/dma-buf.c             |  71 ++++++++++
 io_uring/dma-buf.h             |  76 ++++++++++
 io_uring/rsrc.c                | 178 ++++++++++++++++++------
 io_uring/rsrc.h                |  31 ++++-
 io_uring/rw.c                  |  33 +++--
 lib/iov_iter.c                 |  29 +++-
 31 files changed, 1414 insertions(+), 86 deletions(-)
 create mode 100644 drivers/dma-buf/dma-buf-io.c
 create mode 100644 drivers/nvme/host/pci-dmabuf.c
 create mode 100644 drivers/nvme/host/pci-dmabuf.h
 create mode 100644 include/linux/dma-buf-io.h
 create mode 100644 io_uring/dma-buf.c
 create mode 100644 io_uring/dma-buf.h

-- 
Jens Axboe

[GIT PULL] io_uring dma-buf support for 7.4

[Jens Axboe]

Hi Linus,

This adds native dma-buf support to the I/O path, exposed through
io_uring. It's been in development for a while and should be ready
to go now. Since it depends on both the io_uring and block trees,
I've deferred sending it out until those were merged. This pull
request contains:

- dma-buf file I/O infrastructure: a dma_buf_io_ctx created through a
  new ->init_dma_buf_io_ctx file operation, and a dma_buf_io_map holding
  the device DMA mapping. The two are split so a mapping can be
  invalidated and re-established underneath a live context, with the map
  counting in-flight requests so it can be torn down safely.

- A new iov_iter type for dmabuf maps, and a dma map backed bio type for
  the block layer, with raw block device support on top.

- nvme-pci backend implementing dma-buf backed requests, including SGL
  support.

- io_uring dmabuf backed registered buffers, built on a reworked buffer
  registration structure with an extended update path, regbuf import
  flags, an uncloneable flag for buffer types that can't be cloned, and
  cloning support for registered dma-bufs.

Note that this conflicts with block/fops.c in your tree, as the dma-buf
rejection in the buffered write fallback landed next to the NOWAIT and
i_rwsem changes from the block pull. Fixup is trivial, it's just adding
the iov_iter_is_dmabuf_map() check in the changed
blkdev_buffered_write().

Please pull!


The following changes since commit e889b0e4cafddb3d68c3595dc01b7128fd35fbef:

  io_uring/io-wq: don't call try_to_wake_up() twice in io_wq_worker_wake() (2026-10-12 10:19:52 -0600)

are available in the Git repository at:

  https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git tags/for-7.4/io_uring-dmabuf-20261023

for you to fetch changes up to 45b716456dd99488a0cef58482120dc0474aa696:

  io_uring/dma-buf: don't fail fixed buffers without dma-buf support (2026-10-22 09:27:00 -0600)

----------------------------------------------------------------
for-7.4/io_uring-dmabuf-20261023

----------------------------------------------------------------
Anuj Gupta (3):
      nvme-pci: rename nvme_pci_sgl_set_data to nvme_pci_dma_iter_set_sgl
      nvme-pci: add SGL support for the dmabuf path
      io_uring/dma-buf: don't fail fixed buffers without dma-buf support

Keith Busch (2):
      block: reject dma-bufs larger than 4GiB
      nvme-pci: detach dma-bufs only on removal

Pavel Begunkov (15):
      dma-buf: introduce initial file I/O infrastructure
      iov_iter: add iterator type for dmabuf maps
      block: always adjust bi_offset on bio_advance_iter
      block: introduce dma map backed bio type
      block: add dma-buf support for raw bdev
      nvme-pci: implement dma-buf backed requests
      io_uring/rsrc: introduce buf registration structure
      io_uring/rsrc: extend buffer update
      io_uring/rsrc: add uncloneable regbuf flag
      io_uring/rsrc: add regbuf import flags
      io_uring/rsrc: add dmabuf backed registered buffers
      block: fix indention for dma-buf checks
      dma-buf-io: fix typos in kernel docs
      dma-buf-io: document dma_buf_io_init_map
      io_uring/dma-buf: allow cloning registered dma-bufs

 block/bio.c                    |  15 +-
 block/blk-merge.c              |  52 +++++++
 block/blk-mq.c                 |   6 +
 block/fops.c                   |  29 +++-
 drivers/dma-buf/Makefile       |   2 +-
 drivers/dma-buf/dma-buf-io.c   | 241 ++++++++++++++++++++++++++++++++
 drivers/md/dm-io-rewind.c      |   6 +-
 drivers/nvme/host/Makefile     |   1 +
 drivers/nvme/host/core.c       |  12 ++
 drivers/nvme/host/nvme.h       |   2 +
 drivers/nvme/host/pci-dmabuf.c | 169 ++++++++++++++++++++++
 drivers/nvme/host/pci-dmabuf.h |  75 ++++++++++
 drivers/nvme/host/pci.c        | 247 ++++++++++++++++++++++++++++++++-
 include/linux/bio.h            |  21 +--
 include/linux/blk-mq.h         |   7 +
 include/linux/blk_types.h      |  14 +-
 include/linux/blkdev.h         |   2 +
 include/linux/bvec.h           |   3 +-
 include/linux/dma-buf-io.h     | 126 +++++++++++++++++
 include/linux/fs.h             |   2 +
 include/linux/io_uring_types.h |   5 +
 include/linux/iov_iter.h       |   2 +
 include/linux/uio.h            |  11 ++
 include/uapi/linux/io_uring.h  |  31 ++++-
 io_uring/Makefile              |   1 +
 io_uring/dma-buf.c             |  71 ++++++++++
 io_uring/dma-buf.h             |  76 ++++++++++
 io_uring/rsrc.c                | 178 ++++++++++++++++++------
 io_uring/rsrc.h                |  31 ++++-
 io_uring/rw.c                  |  33 +++--
 lib/iov_iter.c                 |  29 +++-
 31 files changed, 1414 insertions(+), 86 deletions(-)
 create mode 100644 drivers/dma-buf/dma-buf-io.c
 create mode 100644 drivers/nvme/host/pci-dmabuf.c
 create mode 100644 drivers/nvme/host/pci-dmabuf.h
 create mode 100644 include/linux/dma-buf-io.h
 create mode 100644 io_uring/dma-buf.c
 create mode 100644 io_uring/dma-buf.h

-- 
Jens Axboe

Re: [GIT PULL] io_uring dma-buf support for 7.4

[Linus Torvalds]

On Sun, 25 Oct 2026 at 03:30, Jens Axboe <axboe@kernel.dk> wrote:
>
> This adds native dma-buf support to the I/O path

So while doing the trivial resolution, and looking at the code, I note
that blkdev_buffered_write() grew a

    inode_lock_shared
    ..
    inode_unlock_shared

in it. But looking at the two call sites (in blkdev_write_iter()),
_one_ of them already has that inode_lock_shared() on what seems to be
the exact same inode (bdev_file_inode(iocb->ki_filp))

These all went through you in the original block pull:

  1b1be9ea4e34 block: take the inode lock for buffered write fallback
  2fec4fd97cc3 block: take i_rwsem for the direct I/O write fallback

am I missing something? That nested locking looks not just
unnecessary, it looks like deadlock city. I'm surprised lockdep hasn't
complained, but I'm assuming hardly anybody actually uses this path.

                 Linus

Re: [GIT PULL] io_uring dma-buf support for 7.4

[Jens Axboe]

On 10/27/26 11:45 AM, Linus Torvalds wrote:
> On Sun, 25 Oct 2026 at 03:30, Jens Axboe <axboe@kernel.dk> wrote:
>>
>> This adds native dma-buf support to the I/O path
> 
> So while doing the trivial resolution, and looking at the code, I note
> that blkdev_buffered_write() grew a
> 
>     inode_lock_shared
>     ..
>     inode_unlock_shared
> 
> in it. But looking at the two call sites (in blkdev_write_iter()),
> _one_ of them already has that inode_lock_shared() on what seems to be
> the exact same inode (bdev_file_inode(iocb->ki_filp))
> 
> These all went through you in the original block pull:
> 
>   1b1be9ea4e34 block: take the inode lock for buffered write fallback
>   2fec4fd97cc3 block: take i_rwsem for the direct I/O write fallback
> 
> am I missing something? That nested locking looks not just
> unnecessary, it looks like deadlock city. I'm surprised lockdep hasn't
> complained, but I'm assuming hardly anybody actually uses this path.

Someone did actually report it recently, after it had been in
for-next for a long time. There is a fix queued up for it:

https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git/commit/?h=block-7.4&id=21b88cc8bafceee6c6c82503e3b2cc6f1d0c78bf

since this morning! It'll go out later this week, pre -rc1.

-- 
Jens Axboe

Re: [GIT PULL] io_uring dma-buf support for 7.4

[pr-tracker-bot]

The pull request you sent on Sun, 25 Oct 2026 07:29:55 -0600:

> https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git tags/for-7.4/io_uring-dmabuf-20261023

has been merged into torvalds/linux.git:
https://git.kernel.org/torvalds/c/ed02f4f7ad9b5dabae2118e642cbbafe7dc9bb62

Thank you!

-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/prtracker.html