On Fri, Sep 22, 2023 at 09:02:44AM -0600, Jens Axboe wrote:

On 9/22/23 8:28 AM, Ming Lei wrote:

quoted

uring_cmd may never complete, such as ublk, in which uring cmd isn't
completed until one new block request is coming from ublk block device.

Add cancelable uring_cmd to provide mechanism to driver for cancelling
pending commands in its own way.

Add API of io_uring_cmd_mark_cancelable() for driver to mark one command as
cancelable, then io_uring will cancel this command in
io_uring_cancel_generic(). ->uring_cmd() callback is reused for canceling
command in driver's way, then driver gets notified with the cancelling
from io_uring.

Add API of io_uring_cmd_get_task() to help driver cancel handler
deal with the canceling.

This looks better, a few comments:

quoted

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 8e61f8b7c2ce..29a7a7e71f57 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h

@@ -249,10 +249,13 @@ enum io_uring_op {
  * sqe->uring_cmd_flags
  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
  *				along with setting sqe->buf_index.
+ * IORING_URING_CANCELABLE	not for userspace
  * IORING_URING_CMD_POLLED	driver use only
  */
-#define IORING_URING_CMD_FIXED	(1U << 0)
-#define IORING_URING_CMD_POLLED	(1U << 31)
+#define IORING_URING_CMD_FIXED		(1U << 0)
+/* set by driver, and handled by io_uring to cancel this cmd */
+#define IORING_URING_CMD_CANCELABLE	(1U << 30)
+#define IORING_URING_CMD_POLLED		(1U << 31)

If IORING_URING_CANCELABLE isn't UAPI, why stuff it in here? Should we
have a split where we retain the upper 8 bits for internal use, or
something like that?

Yeah, it is for internal use, same with IORING_URING_CMD_POLLED.

I think we can retain upper 8 bits for internal use, and move
the two definitions into kernel header.

quoted

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 783ed0fff71b..a3135fd47a4e 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c

@@ -3256,6 +3256,40 @@ static __cold bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
 	return ret;
 }
 
+static bool io_uring_try_cancel_uring_cmd(struct io_ring_ctx *ctx,
+		struct task_struct *task, bool cancel_all)
+{
+	struct hlist_node *tmp;
+	struct io_kiocb *req;
+	bool ret = false;
+
+	mutex_lock(&ctx->uring_lock);
+	hlist_for_each_entry_safe(req, tmp, &ctx->cancelable_uring_cmd,
+			hash_node) {
+		struct io_uring_cmd *cmd = io_kiocb_to_cmd(req,
+				struct io_uring_cmd);
+		struct file *file = req->file;
+
+		if (WARN_ON_ONCE(!file->f_op->uring_cmd))
+			continue;
+
+		if (!cancel_all && req->task != task)
+			continue;
+
+		if (cmd->flags & IORING_URING_CMD_CANCELABLE) {
+			/* ->sqe isn't available if no async data */
+			if (!req_has_async_data(req))
+				cmd->sqe = NULL;
+			file->f_op->uring_cmd(cmd, IO_URING_F_CANCEL);
+			ret = true;
+		}
+	}
+	io_submit_flush_completions(ctx);
+	mutex_unlock(&ctx->uring_lock);
+
+	return ret;
+}

I think it'd be saner to drop uring_lock here, and then:

quoted

@@ -3307,6 +3341,7 @@ static __cold bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
 	ret |= io_kill_timeouts(ctx, task, cancel_all);
 	if (task)
 		ret |= io_run_task_work() > 0;
+	ret |= io_uring_try_cancel_uring_cmd(ctx, task, cancel_all);
 	return ret;
 }

move this hunk into the uring_lock section. Also ensure that we do run
task_work for cancelation, should the uring_cmd side require that
(either now or eventually).

OK, io_run_task_work() has run task_work already.

quoted

diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
index 537795fddc87..d6b200a0be33 100644
--- a/io_uring/uring_cmd.c
+++ b/io_uring/uring_cmd.c

@@ -13,6 +13,52 @@
 #include "rsrc.h"
 #include "uring_cmd.h"
 
+static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd,
+		unsigned int issue_flags)
+{
+	if (cmd->flags & IORING_URING_CMD_CANCELABLE) {
+		struct io_kiocb *req = cmd_to_io_kiocb(cmd);
+		struct io_ring_ctx *ctx = req->ctx;
+
+		io_ring_submit_lock(ctx, issue_flags);
+		cmd->flags &= ~IORING_URING_CMD_CANCELABLE;
+		hlist_del(&req->hash_node);
+		io_ring_submit_unlock(ctx, issue_flags);
+	}
+}

static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd,
		unsigned int issue_flags)
{
	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
	struct io_ring_ctx *ctx = req->ctx;

	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE))
		return;

	io_ring_submit_lock(ctx, issue_flags);
	cmd->flags &= ~IORING_URING_CMD_CANCELABLE;
	hlist_del(&req->hash_node);
	io_ring_submit_unlock(ctx, issue_flags);
}

is cleaner imho. Minor nit.

quoted

+
+/*
+ * Mark this command as concelable, then io_uring_try_cancel_uring_cmd()
+ * will try to cancel this issued command by sending ->uring_cmd() with
+ * issue_flags of IO_URING_F_CANCEL.
+ *
+ * The command is guaranteed to not be done when calling ->uring_cmd()
+ * with IO_URING_F_CANCEL, but it is driver's responsibility to deal
+ * with race between io_uring canceling and normal completion.
+ */
+int io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
+		unsigned int issue_flags)
+{
+	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
+	struct io_ring_ctx *ctx = req->ctx;
+
+	io_ring_submit_lock(ctx, issue_flags);
+	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
+		cmd->flags |= IORING_URING_CMD_CANCELABLE;
+		hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
+	}
+	io_ring_submit_unlock(ctx, issue_flags);
+
+	return 0;
+}

A bit inconsistent here in terms of the locking. I'm assuming the
marking happens within issue, in which case it should be fine to do:

int io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
		unsigned int issue_flags)
{
	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
	struct io_ring_ctx *ctx = req->ctx;

	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
		cmd->flags |= IORING_URING_CMD_CANCELABLE;
		io_ring_submit_lock(ctx, issue_flags);
		hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
		io_ring_submit_unlock(ctx, issue_flags);
	}

OK, mask & clear bit can be moved out of lock in both two functions.


Thanks,
Ming