[PATCH v2 0/5] dma-buf backed NVMe passthrough I/O

[PATCH v2 0/5] dma-buf backed NVMe passthrough I/O

[Anuj Gupta]

This series adds support for dma-buf registered buffers in NVMe
uring_cmd passthrough.

It depends on this fix:
https://lore.kernel.org/io-uring/20261022132027.2293931-1-anuj1072538@gmail.com/

Patches 1-2 add import flags to the fixed buffer import helpers, so
uring_cmd users have to opt in to dma-buf. Patch 3 adds the io_uring
side, patch 4 lets the block layer map dma-buf iterators for
passthrough, and patch 5 enables it for the NVMe namespace character
device.

Tested with this liburing test:
https://github.com/anuj7781/liburing/tree/rw-dmabuf-pt

Changes since v1:
- split into io_uring, block and nvme patches (Christoph)
- pass import flags to io_import_reg_buf() and
  io_uring_cmd_import_fixed(), and make callers opt in to dma-buf
  (Christoph, Pavel)
- add io_uring_cmd_hw_done(), so nvme no longer handles the dma-buf
  map (Pavel)
- warn if task_work is queued with a dma-buf map attached (Pavel)
- make the -EREMOTEIO conversion explicit and comment it (Christoph)
- refuse dma-bufs larger than 4GiB on the character device

v1: https://lore.kernel.org/linux-nvme/20261016103453.1129150-1-anuj1072538@gmail.com/

Anuj Gupta (5):
  io_uring/rsrc: pass import flags to io_import_reg_buf()
  io_uring/uring_cmd: add import flags to io_uring_cmd_import_fixed()
  io_uring/uring_cmd: support dma-buf registered buffers
  block: map dma-buf iterators in blk_rq_map_user_iov()
  nvme: support dma-buf registered buffers for passthrough

 block/blk-map.c              | 15 ++++++++++++++-
 drivers/nvme/host/core.c     | 16 ++++++++++++++++
 drivers/nvme/host/ioctl.c    |  5 +++--
 drivers/scsi/scsi_bsg.c      |  2 +-
 fs/fuse/dev_uring.c          |  2 +-
 include/linux/io_uring/cmd.h | 14 ++++++++++++--
 io_uring/net.c               |  6 +++---
 io_uring/rsrc.c              |  6 +++---
 io_uring/rsrc.h              | 13 +++----------
 io_uring/rw.c                |  5 +++--
 io_uring/uring_cmd.c         | 31 +++++++++++++++++++++++++++++--
 11 files changed, 88 insertions(+), 27 deletions(-)


base-commit: fd13f31ac62abac4df2166bb0a6fbff53b288e0c
prerequisite-patch-id: 6ad52e1ffc603df71c6855926838f2df402c3ed1
-- 
2.43.0

[PATCH v2 1/5] io_uring/rsrc: pass import flags to io_import_reg_buf()

[Anuj Gupta]

From: Anuj Gupta <anuj20.g@samsung.com>

Give io_import_reg_buf() the import flags argument and drop the
__io_import_reg_buf() variant, which only existed to pass them.

No functional change.

Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
---
 io_uring/net.c       |  6 +++---
 io_uring/rsrc.c      |  6 +++---
 io_uring/rsrc.h      | 13 +++----------
 io_uring/rw.c        |  5 +++--
 io_uring/uring_cmd.c |  2 +-
 5 files changed, 13 insertions(+), 19 deletions(-)

diff --git a/io_uring/net.c b/io_uring/net.c
index 050ed274170a..63089744d43e 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -644,7 +644,7 @@ int io_send(struct io_kiocb *req, unsigned int issue_flags)
 	if (req->flags & REQ_F_IMPORT_BUFFER) {
 		ret = io_import_reg_buf(req, &kmsg->msg.msg_iter,
 					(u64)(uintptr_t)sr->buf, sr->len,
-					ITER_SOURCE, issue_flags);
+					ITER_SOURCE, issue_flags, 0);
 		if (unlikely(ret))
 			return ret;
 		req->flags &= ~REQ_F_IMPORT_BUFFER;
@@ -1216,7 +1216,7 @@ int io_recv(struct io_kiocb *req, unsigned int issue_flags)
 	if (req->flags & REQ_F_IMPORT_BUFFER) {
 		ret = io_import_reg_buf(req, &kmsg->msg.msg_iter,
 					(u64)(uintptr_t)sr->buf, sr->len,
-					ITER_DEST, issue_flags);
+					ITER_DEST, issue_flags, 0);
 		if (unlikely(ret)) {
 			kmsg->msg.msg_inq = -1;
 			sel.buf_list = NULL;
@@ -1509,7 +1509,7 @@ static int io_send_zc_import(struct io_kiocb *req,
 	if (!(sr->flags & IORING_SEND_VECTORIZED)) {
 		ret = io_import_reg_buf(notif, &kmsg->msg.msg_iter,
 					(u64)(uintptr_t)sr->buf, sr->len,
-					ITER_SOURCE, issue_flags);
+					ITER_SOURCE, issue_flags, 0);
 	} else {
 		unsigned uvec_segs = kmsg->msg.msg_iter.nr_segs;
 
diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
index d64228b0e19e..1abfbb86dae8 100644
--- a/io_uring/rsrc.c
+++ b/io_uring/rsrc.c
@@ -1365,9 +1365,9 @@ inline struct io_rsrc_node *io_find_buf_node(struct io_kiocb *req,
 	return NULL;
 }
 
-int __io_import_reg_buf(struct io_kiocb *req, struct iov_iter *iter,
-			u64 buf_addr, size_t len, int ddir,
-			unsigned issue_flags, unsigned import_flags)
+int io_import_reg_buf(struct io_kiocb *req, struct iov_iter *iter,
+		      u64 buf_addr, size_t len, int ddir,
+		      unsigned issue_flags, unsigned import_flags)
 {
 	struct io_rsrc_node *node;
 
diff --git a/io_uring/rsrc.h b/io_uring/rsrc.h
index 418d236152ff..6226c7862d4e 100644
--- a/io_uring/rsrc.h
+++ b/io_uring/rsrc.h
@@ -66,21 +66,14 @@ int io_rsrc_data_alloc(struct io_rsrc_data *data, unsigned nr);
 
 struct io_rsrc_node *io_find_buf_node(struct io_kiocb *req,
 				      unsigned issue_flags);
-int __io_import_reg_buf(struct io_kiocb *req, struct iov_iter *iter,
-			u64 buf_addr, size_t len, int ddir,
-			unsigned issue_flags, unsigned import_flags);
+int io_import_reg_buf(struct io_kiocb *req, struct iov_iter *iter,
+		      u64 buf_addr, size_t len, int ddir,
+		      unsigned issue_flags, unsigned import_flags);
 int __io_import_reg_vec(int ddir, struct iov_iter *iter,
 			struct io_kiocb *req, struct iou_vec *vec,
 			unsigned nr_iovs, unsigned issue_flags,
 			unsigned import_flags);
 
-static inline int io_import_reg_buf(struct io_kiocb *req, struct iov_iter *iter,
-				    u64 buf_addr, size_t len, int ddir,
-				    unsigned issue_flags)
-{
-	return __io_import_reg_buf(req, iter, buf_addr, len, ddir, issue_flags, 0);
-}
-
 static inline int io_import_reg_vec(int ddir, struct iov_iter *iter,
 				    struct io_kiocb *req, struct iou_vec *vec,
 				    unsigned nr_iovs, unsigned issue_flags)
diff --git a/io_uring/rw.c b/io_uring/rw.c
index d33b925a848e..71782a23e16b 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -360,8 +360,9 @@ static int io_init_rw_fixed(struct io_kiocb *req, unsigned int issue_flags,
 	int ret;
 
 	if (!io->bytes_done) {
-		ret = __io_import_reg_buf(req, &io->iter, rw->addr, rw->len, ddir,
-					  issue_flags, IO_REGBUF_IMPORT_ALLOW_DMABUF);
+		ret = io_import_reg_buf(req, &io->iter, rw->addr, rw->len, ddir,
+					issue_flags,
+					IO_REGBUF_IMPORT_ALLOW_DMABUF);
 		if (ret)
 			return ret;
 		iov_iter_save_state(&io->iter, &io->iter_state);
diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
index 73b606899833..45c9c845f5d0 100644
--- a/io_uring/uring_cmd.c
+++ b/io_uring/uring_cmd.c
@@ -279,7 +279,7 @@ int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
 	if (WARN_ON_ONCE(!(ioucmd->flags & IORING_URING_CMD_FIXED)))
 		return -EINVAL;
 
-	return io_import_reg_buf(req, iter, ubuf, len, rw, issue_flags);
+	return io_import_reg_buf(req, iter, ubuf, len, rw, issue_flags, 0);
 }
 EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed);
 
-- 
2.43.0

[PATCH v2 2/5] io_uring/uring_cmd: add import flags to io_uring_cmd_import_fixed()

[Anuj Gupta]

From: Anuj Gupta <anuj20.g@samsung.com>

Add an import flags argument to io_uring_cmd_import_fixed(), so callers
can opt in to buffer types they support. No flags are defined yet, and
all callers pass 0.

No functional change.

Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
---
 drivers/nvme/host/ioctl.c    | 2 +-
 drivers/scsi/scsi_bsg.c      | 2 +-
 fs/fuse/dev_uring.c          | 2 +-
 include/linux/io_uring/cmd.h | 5 +++--
 io_uring/uring_cmd.c         | 5 ++++-
 5 files changed, 10 insertions(+), 6 deletions(-)

diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
index 3aa5d2f2dfbb..612d3009dd29 100644
--- a/drivers/nvme/host/ioctl.c
+++ b/drivers/nvme/host/ioctl.c
@@ -533,7 +533,7 @@ static int nvme_uring_cmd_io(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 					ddir, &iter, issue_flags);
 		else
 			ret = io_uring_cmd_import_fixed(d.addr, d.data_len,
-					ddir, &iter, ioucmd, issue_flags);
+					ddir, &iter, ioucmd, issue_flags, 0);
 		if (ret < 0)
 			return ret;
 
diff --git a/drivers/scsi/scsi_bsg.c b/drivers/scsi/scsi_bsg.c
index 5eec248a77a6..e53b417c27b9 100644
--- a/drivers/scsi/scsi_bsg.c
+++ b/drivers/scsi/scsi_bsg.c
@@ -87,7 +87,7 @@ static int scsi_bsg_map_user_buffer(struct request *req,
 	if (ioucmd->flags & IORING_URING_CMD_FIXED) {
 		ret = io_uring_cmd_import_fixed(buf_addr, buf_len,
 						is_write ? WRITE : READ,
-						&iter, ioucmd, issue_flags);
+						&iter, ioucmd, issue_flags, 0);
 		if (ret < 0)
 			return ret;
 		ret = blk_rq_map_user_iov(req->q, req, NULL, &iter, gfp_mask);
diff --git a/fs/fuse/dev_uring.c b/fs/fuse/dev_uring.c
index c6dd420c4034..5cc206abccc0 100644
--- a/fs/fuse/dev_uring.c
+++ b/fs/fuse/dev_uring.c
@@ -727,7 +727,7 @@ static int fuse_uring_import_payload(struct fuse_ring_ent *ent, int dir,
 
 	if (bufpool_registered(ent->queue))
 		err = io_uring_cmd_import_fixed((u64)(uintptr_t)base, len, dir,
-						iter, ent->cmd, issue_flags);
+						iter, ent->cmd, issue_flags, 0);
 	else
 		err = import_ubuf(dir, base, len, iter);
 
diff --git a/include/linux/io_uring/cmd.h b/include/linux/io_uring/cmd.h
index e18d4af62e1a..23fb66a1e2c1 100644
--- a/include/linux/io_uring/cmd.h
+++ b/include/linux/io_uring/cmd.h
@@ -64,7 +64,8 @@ static inline void io_uring_cmd_set_res32(struct io_uring_cmd *cmd, s32 ret, u64
 int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
 			      struct iov_iter *iter,
 			      struct io_uring_cmd *ioucmd,
-			      unsigned int issue_flags);
+			      unsigned int issue_flags,
+			      unsigned int import_flags);
 int io_uring_cmd_import_fixed_vec(struct io_uring_cmd *ioucmd,
 				  const struct iovec __user *uvec,
 				  size_t uvec_segs,
@@ -124,7 +125,7 @@ int io_buffer_unregister(struct io_uring_cmd *cmd, unsigned int index,
 static inline int
 io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
 			  struct iov_iter *iter, struct io_uring_cmd *ioucmd,
-			  unsigned int issue_flags)
+			  unsigned int issue_flags, unsigned int import_flags)
 {
 	return -EOPNOTSUPP;
 }
diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
index 45c9c845f5d0..c2bc9dd242ec 100644
--- a/io_uring/uring_cmd.c
+++ b/io_uring/uring_cmd.c
@@ -272,12 +272,15 @@ int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
 int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
 			      struct iov_iter *iter,
 			      struct io_uring_cmd *ioucmd,
-			      unsigned int issue_flags)
+			      unsigned int issue_flags,
+			      unsigned int import_flags)
 {
 	struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
 
 	if (WARN_ON_ONCE(!(ioucmd->flags & IORING_URING_CMD_FIXED)))
 		return -EINVAL;
+	if (WARN_ON_ONCE(import_flags))
+		return -EINVAL;
 
 	return io_import_reg_buf(req, iter, ubuf, len, rw, issue_flags, 0);
 }
-- 
2.43.0

[PATCH v2 3/5] io_uring/uring_cmd: support dma-buf registered buffers

[Anuj Gupta]

From: Anuj Gupta <anuj20.g@samsung.com>

Let callers of io_uring_cmd_import_fixed() opt in to dma-buf registered
buffers with IO_URING_CMD_IMPORT_DMABUF, and attach the current map on
import.

The map has to be dropped at hardware completion rather than from
task_work. Add io_uring_cmd_hw_done() for drivers to call from end_io,
which drops the map and queues task_work, and warn if task_work is
queued with a map still attached.

Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
---
 include/linux/io_uring/cmd.h |  9 +++++++++
 io_uring/uring_cmd.c         | 28 ++++++++++++++++++++++++++--
 2 files changed, 35 insertions(+), 2 deletions(-)

diff --git a/include/linux/io_uring/cmd.h b/include/linux/io_uring/cmd.h
index 23fb66a1e2c1..d4e17c2f59fd 100644
--- a/include/linux/io_uring/cmd.h
+++ b/include/linux/io_uring/cmd.h
@@ -12,6 +12,9 @@
 /* io_uring_cmd is being issued again */
 #define IORING_URING_CMD_REISSUE	(1U << 31)
 
+/* io_uring_cmd_import_fixed() flags */
+#define IO_URING_CMD_IMPORT_DMABUF	(1U << 0)
+
 struct io_uring_cmd {
 	struct file	*file;
 	const struct io_uring_sqe *sqe;
@@ -86,6 +89,8 @@ void __io_uring_cmd_done(struct io_uring_cmd *, unsigned issue_flags);
 void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd,
 			    io_req_tw_func_t task_work_cb,
 			    unsigned flags);
+void io_uring_cmd_hw_done(struct io_uring_cmd *ioucmd,
+			  io_req_tw_func_t task_work_cb);
 
 /*
  * Note: the caller should never hard code @issue_flags and only use the
@@ -145,6 +150,10 @@ static inline void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd,
 			    io_req_tw_func_t task_work_cb, unsigned flags)
 {
 }
+static inline void io_uring_cmd_hw_done(struct io_uring_cmd *ioucmd,
+					io_req_tw_func_t task_work_cb)
+{
+}
 static inline void io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
 		unsigned int issue_flags)
 {
diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
index c2bc9dd242ec..c89f154190b1 100644
--- a/io_uring/uring_cmd.c
+++ b/io_uring/uring_cmd.c
@@ -10,6 +10,7 @@
 
 #include "io_uring.h"
 #include "alloc_cache.h"
+#include "dma-buf.h"
 #include "rsrc.h"
 #include "kbuf.h"
 #include "uring_cmd.h"
@@ -28,6 +29,7 @@ static void io_req_uring_cleanup(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
 	struct io_async_cmd *ac = req->async_data;
 
+	io_detach_dmabuf_map(req);
 	if (issue_flags & IO_URING_F_UNLOCKED)
 		return;
 
@@ -132,12 +134,26 @@ void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd,
 
 	if (WARN_ON_ONCE(req->flags & REQ_F_APOLL_MULTISHOT))
 		return;
+	/* the dma-buf map must be dropped first, see io_uring_cmd_hw_done() */
+	WARN_ON_ONCE(req->dmabuf_map);
 
 	req->io_task_work.func = task_work_cb;
 	__io_req_task_work_add(req, flags);
 }
 EXPORT_SYMBOL_GPL(__io_uring_cmd_do_in_task);
 
+/*
+ * Queue completion in task context after dropping references that must be
+ * released at hardware completion rather than from task_work.
+ */
+void io_uring_cmd_hw_done(struct io_uring_cmd *ioucmd,
+			  io_req_tw_func_t task_work_cb)
+{
+	io_detach_dmabuf_map(cmd_to_io_kiocb(ioucmd));
+	__io_uring_cmd_do_in_task(ioucmd, task_work_cb, IOU_F_TWQ_LAZY_WAKE);
+}
+EXPORT_SYMBOL_GPL(io_uring_cmd_hw_done);
+
 /*
  * Called by consumers of io_uring_cmd, if they originally returned
  * -EIOCBQUEUED upon receiving the command.
@@ -257,6 +273,7 @@ int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
 
 	ret = file->f_op->uring_cmd(ioucmd, issue_flags);
 	if (ret == -EAGAIN) {
+		io_detach_dmabuf_map(req);
 		ioucmd->flags |= IORING_URING_CMD_REISSUE;
 		return ret;
 	}
@@ -276,13 +293,20 @@ int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
 			      unsigned int import_flags)
 {
 	struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
+	unsigned int flags = 0;
+	int ret;
 
 	if (WARN_ON_ONCE(!(ioucmd->flags & IORING_URING_CMD_FIXED)))
 		return -EINVAL;
-	if (WARN_ON_ONCE(import_flags))
+	if (WARN_ON_ONCE(import_flags & ~IO_URING_CMD_IMPORT_DMABUF))
 		return -EINVAL;
+	if (import_flags & IO_URING_CMD_IMPORT_DMABUF)
+		flags |= IO_REGBUF_IMPORT_ALLOW_DMABUF;
 
-	return io_import_reg_buf(req, iter, ubuf, len, rw, issue_flags, 0);
+	ret = io_import_reg_buf(req, iter, ubuf, len, rw, issue_flags, flags);
+	if (ret)
+		return ret;
+	return io_attach_dmabuf_map(req, iter, issue_flags);
 }
 EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed);
 
-- 
2.43.0

[PATCH v2 4/5] block: map dma-buf iterators in blk_rq_map_user_iov()

[Anuj Gupta]

From: Anuj Gupta <anuj20.g@samsung.com>

A dma-buf iterator can't fall back to a bounce copy, so map it directly
with blk_rq_map_user_bvec(), which already handles it through
bio_iov_iter_set(). Fail with -EINVAL where a copy would otherwise be
needed, and update the blk_rq_map_user_bvec() comment to match.

Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
---
 block/blk-map.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/block/blk-map.c b/block/blk-map.c
index 9cb9605d1f62..d1e096fdb03d 100644
--- a/block/blk-map.c
+++ b/block/blk-map.c
@@ -459,7 +459,7 @@ int blk_rq_append_bio(struct request *rq, struct bio *bio)
 }
 EXPORT_SYMBOL(blk_rq_append_bio);
 
-/* Prepare bio for passthrough IO given ITER_BVEC iter */
+/* Prepare a bio for passthrough I/O from a directly mapped iterator. */
 static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter)
 {
 	unsigned int max_bytes = rq->q->limits.max_hw_sectors << SECTOR_SHIFT;
@@ -506,6 +506,19 @@ int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
 	struct iov_iter i;
 	int ret = -EINVAL;
 
+	if (iov_iter_is_dmabuf_map(iter)) {
+		if (map_data || (iov_iter_alignment(iter) & align))
+			return -EINVAL;
+		ret = blk_rq_map_user_bvec(rq, iter);
+		/*
+		 * -EREMOTEIO asks the caller to fall back to a bounce copy,
+		 * which a dma-buf can't do.
+		 */
+		if (ret == -EREMOTEIO)
+			ret = -EINVAL;
+		return ret;
+	}
+
 	if (map_data)
 		copy = true;
 	else if (iov_iter_alignment(iter) & align)
-- 
2.43.0

Re: [PATCH v2 4/5] block: map dma-buf iterators in blk_rq_map_user_iov()

[Christoph Hellwig]

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v2 5/5] nvme: support dma-buf registered buffers for passthrough

[Anuj Gupta]

From: Anuj Gupta <anuj20.g@samsung.com>

Allow registering a dma-buf against an NVMe namespace character device
and using it as the data buffer of fixed uring_cmd passthrough I/O.
Complete through io_uring_cmd_hw_done(), so the dma-buf map is released
at hardware completion.

Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
---
 drivers/nvme/host/core.c  | 16 ++++++++++++++++
 drivers/nvme/host/ioctl.c |  5 +++--
 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index d81462e864a4..6157ff839929 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -8,6 +8,7 @@
 #include <linux/blkdev.h>
 #include <linux/blk-mq.h>
 #include <linux/blk-integrity.h>
+#include <linux/dma-buf-io.h>
 #include <linux/compat.h>
 #include <linux/delay.h>
 #include <linux/errno.h>
@@ -3982,6 +3983,20 @@ static int nvme_ns_chr_release(struct inode *inode, struct file *file)
 	return 0;
 }
 
+static int nvme_ns_chr_init_dma_buf_io_ctx(struct file *file,
+					   struct dma_buf_io_ctx *ctx)
+{
+	struct nvme_ns *ns =
+		container_of(file_inode(file)->i_cdev, struct nvme_ns, cdev);
+	struct nvme_ctrl *ctrl = ns->ctrl;
+
+	if (ctx->dmabuf->size > UINT_MAX)
+		return -EINVAL;
+	if (!ctrl->ops->init_dma_buf_io_ctx)
+		return -EINVAL;
+	return ctrl->ops->init_dma_buf_io_ctx(ctrl, ctx);
+}
+
 static const struct file_operations nvme_ns_chr_fops = {
 	.owner		= THIS_MODULE,
 	.open		= nvme_ns_chr_open,
@@ -3990,6 +4005,7 @@ static const struct file_operations nvme_ns_chr_fops = {
 	.compat_ioctl	= compat_ptr_ioctl,
 	.uring_cmd	= nvme_ns_chr_uring_cmd,
 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
+	.init_dma_buf_io_ctx = nvme_ns_chr_init_dma_buf_io_ctx,
 };
 
 static void nvme_add_ns_cdev(struct nvme_ns *ns)
diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
index 612d3009dd29..645446acc10a 100644
--- a/drivers/nvme/host/ioctl.c
+++ b/drivers/nvme/host/ioctl.c
@@ -472,7 +472,7 @@ static enum rq_end_io_ret nvme_uring_cmd_end_io(struct request *req,
 			blk_rq_unmap_user(pdu->bio);
 		__io_uring_cmd_done(ioucmd, 0);
 	} else {
-		io_uring_cmd_do_in_task_lazy(ioucmd, nvme_uring_task_cb);
+		io_uring_cmd_hw_done(ioucmd, nvme_uring_task_cb);
 	}
 	return RQ_END_IO_FREE;
 }
@@ -533,7 +533,8 @@ static int nvme_uring_cmd_io(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 					ddir, &iter, issue_flags);
 		else
 			ret = io_uring_cmd_import_fixed(d.addr, d.data_len,
-					ddir, &iter, ioucmd, issue_flags, 0);
+					ddir, &iter, ioucmd, issue_flags,
+					IO_URING_CMD_IMPORT_DMABUF);
 		if (ret < 0)
 			return ret;
 
-- 
2.43.0

Re: [PATCH v2 5/5] nvme: support dma-buf registered buffers for passthrough

[Christoph Hellwig]

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>