From: Caleb Sander Mateos <csander@purestorage.com>
To: Jens Axboe <axboe@kernel.dk>, Keith Busch <kbusch@kernel.org>,
Christoph Hellwig <hch@lst.de>, Sagi Grimberg <sagi@grimberg.me>
Cc: io-uring@vger.kernel.org, linux-nvme@lists.infradead.org,
linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
Caleb Sander Mateos <csander@purestorage.com>
Subject: [PATCH 6/6] nvme/ioctl: support fixed buffer for metadata
Date: Wed, 9 Sep 2026 16:28:36 -0600 [thread overview]
Message-ID: <20260909222836.2475352-7-csander@purestorage.com> (raw)
In-Reply-To: <20260909222836.2475352-1-csander@purestorage.com>
io_uring NVMe passthrough supports using a "fixed" (registered) buffer
for data, but not metadata. On high-IOPS workloads, the pinning and
unpinning overhead for the metadata pages is significant and could be
avoided if fixed metadata buffers were supported.
Define a NVME_URING_CMD_FIXED_METADATA bit for the nvme_uring_cmd's
flags field (which is currently required to be 0) to indicate that the
metadata buffer belongs to a fixed buffer registered with the io_uring.
The metadata fixed buffer index is specified in a metadata_buf_index
field replacing the existing rsvd1 in nvme_uring_cmd.
If NVME_URING_CMD_FIXED_METADATA is set, the metadata iov_iter is
obtained from io_uring_cmd_import_fixed_metadata() instead of
iov_iter_ubuf().
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
---
drivers/nvme/host/ioctl.c | 21 ++++++++++++++++-----
include/uapi/linux/nvme_ioctl.h | 5 ++++-
2 files changed, 20 insertions(+), 6 deletions(-)
diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
index 8e722184ecf1..4da8cf92c2e0 100644
--- a/drivers/nvme/host/ioctl.c
+++ b/drivers/nvme/host/ioctl.c
@@ -488,21 +488,22 @@ static int nvme_uring_cmd_io(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
struct iov_iter iter, meta_iter;
struct iov_iter *map_iter = NULL, *map_meta_iter = NULL;
struct request *req;
blk_opf_t rq_flags = 0;
blk_mq_req_flags_t blk_flags = 0;
+ u8 flags = READ_ONCE(cmd->flags);
u32 metadata_len, data_len;
u64 metadata, addr;
u32 timeout_ms;
int ddir;
int ret;
- c.common.opcode = READ_ONCE(cmd->opcode);
- c.common.flags = READ_ONCE(cmd->flags);
- if (c.common.flags)
+ if (flags & ~NVME_URING_CMD_FIXED_METADATA)
return -EINVAL;
+ c.common.opcode = READ_ONCE(cmd->opcode);
+ c.common.flags = 0;
c.common.command_id = 0;
c.common.nsid = cpu_to_le32(cmd->nsid);
if (!nvme_validate_passthru_nsid(ctrl, ns, le32_to_cpu(c.common.nsid)))
return -EINVAL;
@@ -539,12 +540,22 @@ static int nvme_uring_cmd_io(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
return ret;
map_iter = &iter;
}
if (data_len && metadata && metadata_len) {
- iov_iter_ubuf(&meta_iter, ddir, nvme_to_user_ptr(metadata),
- metadata_len);
+ if (flags & NVME_URING_CMD_FIXED_METADATA) {
+ u16 buf_index = READ_ONCE(cmd->metadata_buf_index);
+
+ ret = io_uring_cmd_import_fixed_metadata(
+ ioucmd, buf_index, metadata, metadata_len, ddir,
+ &meta_iter, issue_flags);
+ if (ret < 0)
+ return ret;
+ } else {
+ iov_iter_ubuf(&meta_iter, ddir, nvme_to_user_ptr(metadata),
+ metadata_len);
+ }
map_meta_iter = &meta_iter;
}
if (issue_flags & IO_URING_F_NONBLOCK) {
rq_flags |= REQ_NOWAIT;
diff --git a/include/uapi/linux/nvme_ioctl.h b/include/uapi/linux/nvme_ioctl.h
index 2f76cba67166..93973f636b48 100644
--- a/include/uapi/linux/nvme_ioctl.h
+++ b/include/uapi/linux/nvme_ioctl.h
@@ -68,15 +68,18 @@ struct nvme_passthru_cmd64 {
__u32 timeout_ms;
__u32 rsvd2;
__u64 result;
};
+/* struct nvme_uring_cmd flags field bits */
+#define NVME_URING_CMD_FIXED_METADATA (1U << 0)
+
/* same as struct nvme_passthru_cmd64, minus the 8b result field */
struct nvme_uring_cmd {
__u8 opcode;
__u8 flags;
- __u16 rsvd1;
+ __u16 metadata_buf_index;
__u32 nsid;
__u32 cdw2;
__u32 cdw3;
__u64 metadata;
__u64 addr;
--
2.55.0
prev parent reply other threads:[~2026-09-09 22:29 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-09-09 22:28 [PATCH 0/6] io_uring/nvme: support fixed buffer for metadata Caleb Sander Mateos
2026-09-09 22:28 ` [PATCH 1/6] bio-integrity: remove dead bio_integrity_copy_user() error path Caleb Sander Mateos
2026-09-09 22:28 ` [PATCH 2/6] nvme/ioctl: remove struct nvme_uring_data Caleb Sander Mateos
2026-09-09 22:28 ` [PATCH 3/6] blk-integrity: pass iov_iter to blk_rq_integrity_map_user() Caleb Sander Mateos
2026-09-09 22:28 ` [PATCH 4/6] nvme/ioctl: pass iov_iter to nvme_map_user_request() Caleb Sander Mateos
2026-09-09 22:28 ` [PATCH 5/6] io_uring/cmd: support fixed buffer for metadata Caleb Sander Mateos
2026-09-09 22:28 ` Caleb Sander Mateos [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260909222836.2475352-7-csander@purestorage.com \
--to=csander@purestorage.com \
--cc=axboe@kernel.dk \
--cc=hch@lst.de \
--cc=io-uring@vger.kernel.org \
--cc=kbusch@kernel.org \
--cc=linux-block@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-nvme@lists.infradead.org \
--cc=sagi@grimberg.me \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox