public inbox for [email protected]
 help / color / mirror / Atom feed
* [PATCH 0/2] iopoll sync fixes
@ 2021-01-04 20:36 Pavel Begunkov
  2021-01-04 20:36 ` [PATCH v3 1/2] io_uring: synchronise IOPOLL on task_submit fail Pavel Begunkov
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Pavel Begunkov @ 2021-01-04 20:36 UTC (permalink / raw)
  To: Jens Axboe, io-uring

[previously a part of "bunch of random fixes"]
since then:
- 1/2, don't set mm/files under uring_lock
- 2/2, return an overflow fast check to io_iopoll_check()

Pavel Begunkov (2):
  io_uring: synchronise IOPOLL on task_submit fail
  io_uring: patch up IOPOLL overflow_flush sync

 fs/io_uring.c | 91 +++++++++++++++++++++++++++------------------------
 1 file changed, 48 insertions(+), 43 deletions(-)

-- 
2.24.0


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v3 1/2] io_uring: synchronise IOPOLL on task_submit fail
  2021-01-04 20:36 [PATCH 0/2] iopoll sync fixes Pavel Begunkov
@ 2021-01-04 20:36 ` Pavel Begunkov
  2021-01-04 20:36 ` [PATCH v3 2/2] io_uring: patch up IOPOLL overflow_flush sync Pavel Begunkov
  2021-01-04 22:23 ` [PATCH 0/2] iopoll sync fixes Jens Axboe
  2 siblings, 0 replies; 4+ messages in thread
From: Pavel Begunkov @ 2021-01-04 20:36 UTC (permalink / raw)
  To: Jens Axboe, io-uring; +Cc: stable

io_req_task_submit() might be called for IOPOLL, do the fail path under
uring_lock to comply with IOPOLL synchronisation based solely on it.

Cc: [email protected] # 5.5+
Signed-off-by: Pavel Begunkov <[email protected]>
---
 fs/io_uring.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index ca46f314640b..5be33fd8b6bc 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2126,15 +2126,16 @@ static void io_req_task_cancel(struct callback_head *cb)
 static void __io_req_task_submit(struct io_kiocb *req)
 {
 	struct io_ring_ctx *ctx = req->ctx;
+	bool fail;
 
-	if (!__io_sq_thread_acquire_mm(ctx) &&
-	    !__io_sq_thread_acquire_files(ctx)) {
-		mutex_lock(&ctx->uring_lock);
+	fail = __io_sq_thread_acquire_mm(ctx) ||
+		__io_sq_thread_acquire_files(ctx);
+	mutex_lock(&ctx->uring_lock);
+	if (!fail)
 		__io_queue_sqe(req, NULL);
-		mutex_unlock(&ctx->uring_lock);
-	} else {
+	else
 		__io_req_task_cancel(req, -EFAULT);
-	}
+	mutex_unlock(&ctx->uring_lock);
 }
 
 static void io_req_task_submit(struct callback_head *cb)
-- 
2.24.0


^ permalink raw reply related	[flat|nested] 4+ messages in thread

* [PATCH v3 2/2] io_uring: patch up IOPOLL overflow_flush sync
  2021-01-04 20:36 [PATCH 0/2] iopoll sync fixes Pavel Begunkov
  2021-01-04 20:36 ` [PATCH v3 1/2] io_uring: synchronise IOPOLL on task_submit fail Pavel Begunkov
@ 2021-01-04 20:36 ` Pavel Begunkov
  2021-01-04 22:23 ` [PATCH 0/2] iopoll sync fixes Jens Axboe
  2 siblings, 0 replies; 4+ messages in thread
From: Pavel Begunkov @ 2021-01-04 20:36 UTC (permalink / raw)
  To: Jens Axboe, io-uring; +Cc: stable

IOPOLL skips completion locking but keeps it under uring_lock, thus
io_cqring_overflow_flush() and so io_cqring_events() need additional
locking with uring_lock in some cases for IOPOLL.

Remove __io_cqring_overflow_flush() from io_cqring_events(), introduce a
wrapper around flush doing needed synchronisation and call it by hand.

Cc: [email protected] # 5.5+
Signed-off-by: Pavel Begunkov <[email protected]>
---
 fs/io_uring.c | 78 +++++++++++++++++++++++++++------------------------
 1 file changed, 41 insertions(+), 37 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 5be33fd8b6bc..445035b24a50 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1713,9 +1713,9 @@ static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
 }
 
 /* Returns true if there are no backlogged entries after the flush */
-static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
-				     struct task_struct *tsk,
-				     struct files_struct *files)
+static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
+				       struct task_struct *tsk,
+				       struct files_struct *files)
 {
 	struct io_rings *rings = ctx->rings;
 	struct io_kiocb *req, *tmp;
@@ -1768,6 +1768,20 @@ static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
 	return all_flushed;
 }
 
+static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
+				     struct task_struct *tsk,
+				     struct files_struct *files)
+{
+	if (test_bit(0, &ctx->cq_check_overflow)) {
+		/* iopoll syncs against uring_lock, not completion_lock */
+		if (ctx->flags & IORING_SETUP_IOPOLL)
+			mutex_lock(&ctx->uring_lock);
+		__io_cqring_overflow_flush(ctx, force, tsk, files);
+		if (ctx->flags & IORING_SETUP_IOPOLL)
+			mutex_unlock(&ctx->uring_lock);
+	}
+}
+
 static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
@@ -2314,20 +2328,8 @@ static void io_double_put_req(struct io_kiocb *req)
 		io_free_req(req);
 }
 
-static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush)
+static unsigned io_cqring_events(struct io_ring_ctx *ctx)
 {
-	if (test_bit(0, &ctx->cq_check_overflow)) {
-		/*
-		 * noflush == true is from the waitqueue handler, just ensure
-		 * we wake up the task, and the next invocation will flush the
-		 * entries. We cannot safely to it from here.
-		 */
-		if (noflush)
-			return -1U;
-
-		io_cqring_overflow_flush(ctx, false, NULL, NULL);
-	}
-
 	/* See comment at the top of this file */
 	smp_rmb();
 	return __io_cqring_events(ctx);
@@ -2552,7 +2554,9 @@ static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
 		 * If we do, we can potentially be spinning for commands that
 		 * already triggered a CQE (eg in error).
 		 */
-		if (io_cqring_events(ctx, false))
+		if (test_bit(0, &ctx->cq_check_overflow))
+			__io_cqring_overflow_flush(ctx, false, NULL, NULL);
+		if (io_cqring_events(ctx))
 			break;
 
 		/*
@@ -6827,7 +6831,7 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
 
 	/* if we have a backlog and couldn't flush it all, return BUSY */
 	if (test_bit(0, &ctx->sq_check_overflow)) {
-		if (!io_cqring_overflow_flush(ctx, false, NULL, NULL))
+		if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
 			return -EBUSY;
 	}
 
@@ -7090,7 +7094,7 @@ struct io_wait_queue {
 	unsigned nr_timeouts;
 };
 
-static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
+static inline bool io_should_wake(struct io_wait_queue *iowq)
 {
 	struct io_ring_ctx *ctx = iowq->ctx;
 
@@ -7099,7 +7103,7 @@ static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
 	 * started waiting. For timeouts, we always want to return to userspace,
 	 * regardless of event count.
 	 */
-	return io_cqring_events(ctx, noflush) >= iowq->to_wait ||
+	return io_cqring_events(ctx) >= iowq->to_wait ||
 			atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
 }
 
@@ -7109,11 +7113,13 @@ static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
 	struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
 							wq);
 
-	/* use noflush == true, as we can't safely rely on locking context */
-	if (!io_should_wake(iowq, true))
-		return -1;
-
-	return autoremove_wake_function(curr, mode, wake_flags, key);
+	/*
+	 * Cannot safely flush overflowed CQEs from here, ensure we wake up
+	 * the task, and the next invocation will do it.
+	 */
+	if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
+		return autoremove_wake_function(curr, mode, wake_flags, key);
+	return -1;
 }
 
 static int io_run_task_work_sig(void)
@@ -7150,7 +7156,8 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	int ret = 0;
 
 	do {
-		if (io_cqring_events(ctx, false) >= min_events)
+		io_cqring_overflow_flush(ctx, false, NULL, NULL);
+		if (io_cqring_events(ctx) >= min_events)
 			return 0;
 		if (!io_run_task_work())
 			break;
@@ -7178,6 +7185,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
 	trace_io_uring_cqring_wait(ctx, min_events);
 	do {
+		io_cqring_overflow_flush(ctx, false, NULL, NULL);
 		prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
 						TASK_INTERRUPTIBLE);
 		/* make sure we run task_work before checking for signals */
@@ -7186,8 +7194,10 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 			continue;
 		else if (ret < 0)
 			break;
-		if (io_should_wake(&iowq, false))
+		if (io_should_wake(&iowq))
 			break;
+		if (test_bit(0, &ctx->cq_check_overflow))
+			continue;
 		if (uts) {
 			timeout = schedule_timeout(timeout);
 			if (timeout == 0) {
@@ -8625,7 +8635,8 @@ static __poll_t io_uring_poll(struct file *file, poll_table *wait)
 	smp_rmb();
 	if (!io_sqring_full(ctx))
 		mask |= EPOLLOUT | EPOLLWRNORM;
-	if (io_cqring_events(ctx, false))
+	io_cqring_overflow_flush(ctx, false, NULL, NULL);
+	if (io_cqring_events(ctx))
 		mask |= EPOLLIN | EPOLLRDNORM;
 
 	return mask;
@@ -8683,7 +8694,7 @@ static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
 	/* if force is set, the ring is going away. always drop after that */
 	ctx->cq_overflow_flushed = 1;
 	if (ctx->rings)
-		io_cqring_overflow_flush(ctx, true, NULL, NULL);
+		__io_cqring_overflow_flush(ctx, true, NULL, NULL);
 	mutex_unlock(&ctx->uring_lock);
 
 	io_kill_timeouts(ctx, NULL, NULL);
@@ -8857,9 +8868,7 @@ static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
 	}
 
 	io_cancel_defer_files(ctx, task, files);
-	io_ring_submit_lock(ctx, (ctx->flags & IORING_SETUP_IOPOLL));
 	io_cqring_overflow_flush(ctx, true, task, files);
-	io_ring_submit_unlock(ctx, (ctx->flags & IORING_SETUP_IOPOLL));
 
 	if (!files)
 		__io_uring_cancel_task_requests(ctx, task);
@@ -9195,13 +9204,8 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 	 */
 	ret = 0;
 	if (ctx->flags & IORING_SETUP_SQPOLL) {
-		if (!list_empty_careful(&ctx->cq_overflow_list)) {
-			bool needs_lock = ctx->flags & IORING_SETUP_IOPOLL;
+		io_cqring_overflow_flush(ctx, false, NULL, NULL);
 
-			io_ring_submit_lock(ctx, needs_lock);
-			io_cqring_overflow_flush(ctx, false, NULL, NULL);
-			io_ring_submit_unlock(ctx, needs_lock);
-		}
 		if (flags & IORING_ENTER_SQ_WAKEUP)
 			wake_up(&ctx->sq_data->wait);
 		if (flags & IORING_ENTER_SQ_WAIT)
-- 
2.24.0


^ permalink raw reply related	[flat|nested] 4+ messages in thread

* Re: [PATCH 0/2] iopoll sync fixes
  2021-01-04 20:36 [PATCH 0/2] iopoll sync fixes Pavel Begunkov
  2021-01-04 20:36 ` [PATCH v3 1/2] io_uring: synchronise IOPOLL on task_submit fail Pavel Begunkov
  2021-01-04 20:36 ` [PATCH v3 2/2] io_uring: patch up IOPOLL overflow_flush sync Pavel Begunkov
@ 2021-01-04 22:23 ` Jens Axboe
  2 siblings, 0 replies; 4+ messages in thread
From: Jens Axboe @ 2021-01-04 22:23 UTC (permalink / raw)
  To: Pavel Begunkov, io-uring

On 1/4/21 1:36 PM, Pavel Begunkov wrote:
> [previously a part of "bunch of random fixes"]
> since then:
> - 1/2, don't set mm/files under uring_lock
> - 2/2, return an overflow fast check to io_iopoll_check()
> 
> Pavel Begunkov (2):
>   io_uring: synchronise IOPOLL on task_submit fail
>   io_uring: patch up IOPOLL overflow_flush sync
> 
>  fs/io_uring.c | 91 +++++++++++++++++++++++++++------------------------
>  1 file changed, 48 insertions(+), 43 deletions(-)

Applied, thanks.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2021-01-04 23:34 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2021-01-04 20:36 [PATCH 0/2] iopoll sync fixes Pavel Begunkov
2021-01-04 20:36 ` [PATCH v3 1/2] io_uring: synchronise IOPOLL on task_submit fail Pavel Begunkov
2021-01-04 20:36 ` [PATCH v3 2/2] io_uring: patch up IOPOLL overflow_flush sync Pavel Begunkov
2021-01-04 22:23 ` [PATCH 0/2] iopoll sync fixes Jens Axboe

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox