[PATCH] io_uring: support CLOCK_BOOTTIME for timeouts

[Jens Axboe <axboe@xxxxxxxxx>]

Certain use cases want to use CLOCK_BOOTTIME rather than CLOCK_MONOTONIC,
as it doesn't stop updating over suspend. Apart from that, they should
behave the same.

Add an IORING_TIMEOUT_BOOTTIME flag that allows timeouts and linked
timeouts to use CLOCK_BOOTTIME instead.

Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>

---

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 0f827fbe8e6c..39c8631e4d10 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -508,6 +508,7 @@ struct io_timeout_data {
 	struct hrtimer			timer;
 	struct timespec64		ts;
 	enum hrtimer_mode		mode;
+	u32				flags;
 };
 
 struct io_accept {
@@ -5725,7 +5726,10 @@ static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
 	req->timeout.off = 0; /* noseq */
 	data = req->async_data;
 	list_add_tail(&req->timeout.list, &ctx->timeout_list);
-	hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
+	if (data->flags & IORING_TIMEOUT_BOOTTIME)
+		hrtimer_init(&data->timer, CLOCK_BOOTTIME, mode);
+	else
+		hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
 	data->timer.function = io_timeout_fn;
 	hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
 	return 0;
@@ -5807,7 +5811,7 @@ static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 	if (off && is_timeout_link)
 		return -EINVAL;
 	flags = READ_ONCE(sqe->timeout_flags);
-	if (flags & ~IORING_TIMEOUT_ABS)
+	if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_BOOTTIME))
 		return -EINVAL;
 
 	req->timeout.off = off;
@@ -5819,12 +5823,16 @@ static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 
 	data = req->async_data;
 	data->req = req;
+	data->flags = flags;
 
 	if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
 		return -EFAULT;
 
 	data->mode = io_translate_timeout_mode(flags);
-	hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
+	if (flags & IORING_TIMEOUT_BOOTTIME)
+		hrtimer_init(&data->timer, CLOCK_BOOTTIME, data->mode);
+	else
+		hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
 
 	if (is_timeout_link) {
 		struct io_submit_link *link = &req->ctx->submit_state.link;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index bb6845e14629..18a4ffd2bbb3 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -151,6 +151,7 @@ enum {
  */
 #define IORING_TIMEOUT_ABS	(1U << 0)
 #define IORING_TIMEOUT_UPDATE	(1U << 1)
+#define IORING_TIMEOUT_BOOTTIME	(1U << 2)
 
 /*
  * sqe->splice_flags

-- 
Jens Axboe