From: Pavel Begunkov <[email protected]>
To: Caleb Sander Mateos <[email protected]>,
Jens Axboe <[email protected]>, Ming Lei <[email protected]>,
Keith Busch <[email protected]>, Christoph Hellwig <[email protected]>,
Sagi Grimberg <[email protected]>
Cc: Xinyu Zhang <[email protected]>,
[email protected], [email protected],
[email protected]
Subject: Re: [PATCH 3/3] io_uring/uring_cmd: import fixed buffer before going async
Date: Fri, 21 Mar 2025 20:35:24 +0000 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
On 3/21/25 18:48, Caleb Sander Mateos wrote:
> For uring_cmd operations with fixed buffers, the fixed buffer lookup
> happens in io_uring_cmd_import_fixed(), called from the ->uring_cmd()
> implementation. A ->uring_cmd() implementation could return -EAGAIN on
> the initial issue for any reason before io_uring_cmd_import_fixed().
> For example, nvme_uring_cmd_io() calls nvme_alloc_user_request() first,
> which can return -EAGAIN if all tags in the tag set are in use.
That's up to command when it resolves the buffer, you can just
move the call to io_import_reg_buf() earlier in nvme cmd code
and not working it around at the io_uring side.
In general, it's a step back, it just got cleaned up from the
mess where node resolution and buffer imports were separate
steps and duplicate by every single request type that used it.
> This ordering difference is observable when using
> UBLK_U_IO_{,UN}REGISTER_IO_BUF SQEs to modify the fixed buffer table.
> If the uring_cmd is followed by a UBLK_U_IO_UNREGISTER_IO_BUF operation
> that unregisters the fixed buffer, the uring_cmd going async will cause
> the fixed buffer lookup to fail because it happens after the unregister.
>
> Move the fixed buffer lookup out of io_uring_cmd_import_fixed() and
> instead perform it in io_uring_cmd() before calling ->uring_cmd().
> io_uring_cmd_import_fixed() now only initializes an iov_iter from the
> existing fixed buffer node. This division of responsibilities makes
> sense as the fixed buffer lookup is an io_uring implementation detail
> and independent of the ->uring_cmd() implementation. It also cuts down
> on the need to pass around the io_uring issue_flags.
>
> Signed-off-by: Caleb Sander Mateos <[email protected]>
> Fixes: 27cb27b6d5ea ("io_uring: add support for kernel registered bvecs")
> ---
> drivers/nvme/host/ioctl.c | 10 ++++------
> include/linux/io_uring/cmd.h | 6 ++----
> io_uring/rsrc.c | 6 ++++++
> io_uring/rsrc.h | 2 ++
> io_uring/uring_cmd.c | 10 +++++++---
> 5 files changed, 21 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
> index fe9fb80c6a14..3fad74563b9e 100644
> --- a/drivers/nvme/host/ioctl.c
> +++ b/drivers/nvme/host/ioctl.c
> @@ -112,12 +112,11 @@ static struct request *nvme_alloc_user_request(struct request_queue *q,
--
Pavel Begunkov
next prev parent reply other threads:[~2025-03-21 20:34 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-03-21 18:48 [PATCH 0/3] Consistently look up fixed buffers before going async Caleb Sander Mateos
2025-03-21 18:48 ` [PATCH 1/3] io_uring/net: only import send_zc buffer once Caleb Sander Mateos
2025-03-21 20:38 ` Pavel Begunkov
2025-03-21 20:44 ` Caleb Sander Mateos
2025-03-21 18:48 ` [PATCH 2/3] io_uring/net: import send_zc fixed buffer before going async Caleb Sander Mateos
2025-03-21 18:48 ` [PATCH 3/3] io_uring/uring_cmd: import " Caleb Sander Mateos
2025-03-21 20:35 ` Pavel Begunkov [this message]
2025-03-21 21:38 ` Caleb Sander Mateos
2025-03-22 12:18 ` Pavel Begunkov
2025-03-21 19:53 ` [PATCH 0/3] Consistently look up fixed buffers " Jens Axboe
2025-03-21 20:24 ` Pavel Begunkov
2025-03-21 21:24 ` Caleb Sander Mateos
2025-03-22 12:33 ` Pavel Begunkov
2025-03-22 7:42 ` Ming Lei
2025-03-22 7:33 ` Ming Lei
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 \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
/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