From: Uday Shankar <ushankar@purestorage.com>
To: Ming Lei <ming.lei@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>,
linux-block@vger.kernel.org,
Caleb Sander Mateos <csander@purestorage.com>
Subject: Re: [PATCH 2/9] ublk: properly serialize all FETCH_REQs
Date: Tue, 15 Apr 2025 19:04:13 -0600 [thread overview]
Message-ID: <Z/8CDUwcKNDRzyOU@dev-ushankar.dev.purestorage.com> (raw)
In-Reply-To: <Z/1o946/z43QETPr@dev-ushankar.dev.purestorage.com>
On Mon, Apr 14, 2025 at 01:58:47PM -0600, Uday Shankar wrote:
> On Mon, Apr 14, 2025 at 07:25:43PM +0800, Ming Lei wrote:
> > From: Uday Shankar <ushankar@purestorage.com>
> >
> > Most uring_cmds issued against ublk character devices are serialized
> > because each command affects only one queue, and there is an early check
> > which only allows a single task (the queue's ubq_daemon) to issue
> > uring_cmds against that queue. However, this mechanism does not work for
> > FETCH_REQs, since they are expected before ubq_daemon is set. Since
> > FETCH_REQs are only used at initialization and not in the fast path,
> > serialize them using the per-ublk-device mutex. This fixes a number of
> > data races that were previously possible if a badly behaved ublk server
> > decided to issue multiple FETCH_REQs against the same qid/tag
> > concurrently.
> >
> > Reviewed-by: Ming Lei <ming.lei@redhat.com>
> > Reported-by: Caleb Sander Mateos <csander@purestorage.com>
> > Signed-off-by: Uday Shankar <ushankar@purestorage.com>
>
> Thanks for picking this up. Can you use the following patch instead? It
> has two changes compared to [1]:
>
> - Factor FETCH command into its own function
> - Return -EAGAIN for non-blocking dispatch because we are taking a
> mutex.
>
> [1] https://lore.kernel.org/linux-block/20250410-ublk_task_per_io-v3-1-b811e8f4554a@purestorage.com/
>
> diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> index 2fd05c1bd30b..8efb7668ab2c 100644
> --- a/drivers/block/ublk_drv.c
> +++ b/drivers/block/ublk_drv.c
> @@ -1809,8 +1809,8 @@ static void ublk_nosrv_work(struct work_struct *work)
>
> /* device can only be started after all IOs are ready */
> static void ublk_mark_io_ready(struct ublk_device *ub, struct ublk_queue *ubq)
> + __must_hold(&ub->mutex)
> {
> - mutex_lock(&ub->mutex);
> ubq->nr_io_ready++;
> if (ublk_queue_ready(ubq)) {
> ubq->ubq_daemon = current;
> @@ -1822,7 +1822,6 @@ static void ublk_mark_io_ready(struct ublk_device *ub, struct ublk_queue *ubq)
> }
> if (ub->nr_queues_ready == ub->dev_info.nr_hw_queues)
> complete_all(&ub->completion);
> - mutex_unlock(&ub->mutex);
> }
>
> static void ublk_handle_need_get_data(struct ublk_device *ub, int q_id,
> @@ -1906,6 +1905,52 @@ static int ublk_unregister_io_buf(struct io_uring_cmd *cmd,
> return io_buffer_unregister_bvec(cmd, index, issue_flags);
> }
>
> +static int ublk_fetch(struct io_uring_cmd *cmd, struct ublk_device *ub,
> + struct ublk_queue *ubq, struct ublk_io *io,
> + const struct ublksrv_io_cmd *ub_cmd,
> + unsigned int issue_flags)
> +{
> + int ret = 0;
> +
> + if (issue_flags & IO_URING_F_NONBLOCK)
> + return -EAGAIN;
> +
> + mutex_lock(&ub->mutex);
> + /* UBLK_IO_FETCH_REQ is only allowed before queue is setup */
> + if (ublk_queue_ready(ubq)) {
> + ret = -EBUSY;
> + goto out;
> + }
> +
> + /* allow each command to be FETCHed at most once */
> + if (io->flags & UBLK_IO_FLAG_ACTIVE) {
> + ret = -EINVAL;
> + goto out;
> + }
> +
> + WARN_ON_ONCE(io->flags & UBLK_IO_FLAG_OWNED_BY_SRV);
> +
> + if (ublk_need_map_io(ubq)) {
> + /*
> + * FETCH_RQ has to provide IO buffer if NEED GET
> + * DATA is not enabled
> + */
> + if (!ub_cmd->addr && !ublk_need_get_data(ubq))
> + goto out;
> + } else if (ub_cmd->addr) {
> + /* User copy requires addr to be unset */
> + ret = -EINVAL;
> + goto out;
> + }
> +
> + ublk_fill_io_cmd(io, cmd, ub_cmd->addr);
> + ublk_mark_io_ready(ub, ubq);
> +
> +out:
> + mutex_unlock(&ub->mutex);
> + return ret;
> +}
> +
> static int __ublk_ch_uring_cmd(struct io_uring_cmd *cmd,
> unsigned int issue_flags,
> const struct ublksrv_io_cmd *ub_cmd)
> @@ -1962,34 +2007,7 @@ static int __ublk_ch_uring_cmd(struct io_uring_cmd *cmd,
> case UBLK_IO_UNREGISTER_IO_BUF:
> return ublk_unregister_io_buf(cmd, ub_cmd->addr, issue_flags);
> case UBLK_IO_FETCH_REQ:
> - /* UBLK_IO_FETCH_REQ is only allowed before queue is setup */
> - if (ublk_queue_ready(ubq)) {
> - ret = -EBUSY;
> - goto out;
> - }
> - /*
> - * The io is being handled by server, so COMMIT_RQ is expected
> - * instead of FETCH_REQ
> - */
> - if (io->flags & UBLK_IO_FLAG_OWNED_BY_SRV)
> - goto out;
> -
> - if (ublk_need_map_io(ubq)) {
> - /*
> - * FETCH_RQ has to provide IO buffer if NEED GET
> - * DATA is not enabled
> - */
> - if (!ub_cmd->addr && !ublk_need_get_data(ubq))
> - goto out;
> - } else if (ub_cmd->addr) {
> - /* User copy requires addr to be unset */
> - ret = -EINVAL;
> - goto out;
> - }
> -
> - ublk_fill_io_cmd(io, cmd, ub_cmd->addr);
> - ublk_mark_io_ready(ub, ubq);
> - break;
> + return ublk_fetch(cmd, ub, ubq, io, ub_cmd, issue_flags);
One more bug here, this skips the
ublk_prep_cancel(cmd, issue_flags, ubq, tag);
return -EIOCBQUEUED;
that is after the switch statement.
> case UBLK_IO_COMMIT_AND_FETCH_REQ:
> req = blk_mq_tag_to_rq(ub->tag_set.tags[ub_cmd->q_id], tag);
>
>
next prev parent reply other threads:[~2025-04-16 1:04 UTC|newest]
Thread overview: 32+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-04-14 11:25 [PATCH 0/9] ublk: simplify & improve IO canceling Ming Lei
2025-04-14 11:25 ` [PATCH 1/9] ublk: don't try to stop disk if ->ub_disk is NULL Ming Lei
2025-04-14 19:44 ` Uday Shankar
2025-04-15 1:32 ` Ming Lei
2025-04-14 11:25 ` [PATCH 2/9] ublk: properly serialize all FETCH_REQs Ming Lei
2025-04-14 19:58 ` Uday Shankar
2025-04-14 20:39 ` Jens Axboe
2025-04-14 20:52 ` Uday Shankar
2025-04-14 21:00 ` Jens Axboe
2025-04-15 1:40 ` Ming Lei
2025-04-16 1:13 ` Ming Lei
2025-04-16 1:17 ` Jens Axboe
2025-04-16 2:04 ` Ming Lei
2025-04-16 1:04 ` Uday Shankar [this message]
2025-04-14 11:25 ` [PATCH 3/9] ublk: add ublk_force_abort_dev() Ming Lei
2025-04-14 20:06 ` Uday Shankar
2025-04-14 11:25 ` [PATCH 4/9] ublk: rely on ->canceling for dealing with ublk_nosrv_dev_should_queue_io Ming Lei
2025-04-14 20:15 ` Uday Shankar
2025-04-15 1:48 ` Ming Lei
2025-04-14 11:25 ` [PATCH 5/9] ublk: move device reset into ublk_ch_release() Ming Lei
2025-04-14 20:29 ` Uday Shankar
2025-04-15 1:50 ` Ming Lei
2025-04-14 11:25 ` [PATCH 6/9] ublk: improve detection and handling of ublk server exit Ming Lei
2025-04-14 20:36 ` Uday Shankar
2025-04-15 1:54 ` Ming Lei
2025-04-14 11:25 ` [PATCH 7/9] ublk: remove __ublk_quiesce_dev() Ming Lei
2025-04-14 20:37 ` Uday Shankar
2025-04-14 11:25 ` [PATCH 8/9] ublk: simplify aborting ublk request Ming Lei
2025-04-14 20:42 ` Uday Shankar
2025-04-14 11:25 ` [PATCH 9/9] selftests: ublk: add generic_06 for covering fault inject Ming Lei
2025-04-14 20:44 ` Uday Shankar
2025-04-15 1:57 ` 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 \
--in-reply-to=Z/8CDUwcKNDRzyOU@dev-ushankar.dev.purestorage.com \
--to=ushankar@purestorage.com \
--cc=axboe@kernel.dk \
--cc=csander@purestorage.com \
--cc=linux-block@vger.kernel.org \
--cc=ming.lei@redhat.com \
/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;
as well as URLs for NNTP newsgroup(s).