Skip to contentSkip to conversation

Further async re-queue tightening/fix

PATCH In review Jens Axboe 2 participants 6 messages Jul 27, 2021 – Jul 29, 2021

Subsystems (3)
lore
Share & tools
lore

Jens Axboe

author maintainer
#0 lore raw

cover Further async re-queue tightening/fix

Hi,

Two small patches to ensure we always requeue off task_work, and a sanity correction on when it is OK to do so.

Signature · 1 line
Jens Axboe
author maintainer
#1 lore raw

PATCH 1/2 io_uring: always reissue from task_work context

As a safeguard, if we're going to queue async work, do it from task_work from the original task. This ensures that we can always sanely create threads, regards of what the reissue context may be.

---
 fs/io_uring.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
fs/io_uring.c +8 -2
index a4331deb0427..6ba101cd4661 100644
@@ -2060,6 +2060,12 @@ static void io_req_task_queue(struct io_kiocb *req)
 	io_req_task_work_add(req);
 }
 
+static void io_req_task_queue_reissue(struct io_kiocb *req)
+{
+	req->io_task_work.func = io_queue_async_work;
+	io_req_task_work_add(req);
+}
+
 static inline void io_queue_next(struct io_kiocb *req)
 {
 	struct io_kiocb *nxt = io_req_find_next(req);
@@ -2248,7 +2254,7 @@ static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
 		    !(req->flags & REQ_F_DONT_REISSUE)) {
 			req->iopoll_completed = 0;
 			req_ref_get(req);
-			io_queue_async_work(req);
+			io_req_task_queue_reissue(req);
 			continue;
 		}
 
@@ -2771,7 +2777,7 @@ static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
 		req->flags &= ~REQ_F_REISSUE;
 		if (io_resubmit_prep(req)) {
 			req_ref_get(req);
-			io_queue_async_work(req);
+			io_req_task_queue_reissue(req);
 		} else {
 			int cflags = 0;
 
Signature · 1 line
2.32.0
author maintainer
#2 lore raw

PATCH 2/2 io_uring: don't block level reissue off completion path

In reply to Jens Axboe

Some setups, like SCSI, can throw spurious -EAGAIN off the softirq completion path. Normally we expect this to happen inline as part of submission, but apparently SCSI has a weird corner case where it can happen as part of normal completions.

This should be solved by having the -EAGAIN bubble back up the stack as part of submission, but previous attempts at this failed and we're not just quite there yet. Instead we currently use REQ_F_REISSUE to handle this case.

For now, catch it in io_rw_should_reissue() and prevent a reissue from a bogus path.

---
 fs/io_uring.c | 6 ++++++
 1 file changed, 6 insertions(+)
fs/io_uring.c +6 -0
index 6ba101cd4661..83f67d33bf67 100644
@@ -2447,6 +2447,12 @@ static bool io_rw_should_reissue(struct io_kiocb *req)
 	 */
 	if (percpu_ref_is_dying(&ctx->refs))
 		return false;
+	/*
+	 * Play it safe and assume not safe to re-import and reissue if we're
+	 * not in the original thread group (or in task context).
+	 */
+	if (!same_thread_group(req->task, current) || !in_task())
+		return false;
 	return true;
 }
 #else
Signature · 1 line
2.32.0
#3 lore raw

Re: [PATCH 2/2] io_uring: don't block level reissue off completion path

Am 27.07.21 um 18:58 schrieb Jens Axboe:

Quoted 28 lines

Some setups, like SCSI, can throw spurious -EAGAIN off the softirq completion path. Normally we expect this to happen inline as part of submission, but apparently SCSI has a weird corner case where it can happen as part of normal completions.

This should be solved by having the -EAGAIN bubble back up the stack as part of submission, but previous attempts at this failed and we're not just quite there yet. Instead we currently use REQ_F_REISSUE to handle this case.

For now, catch it in io_rw_should_reissue() and prevent a reissue from a bogus path.

---
  fs/io_uring.c | 6 ++++++
  1 file changed, 6 insertions(+)
fs/io_uring.c +0 -0
index 6ba101cd4661..83f67d33bf67 100644
@@ -2447,6 +2447,12 @@ static bool io_rw_should_reissue(struct io_kiocb *req)
  	 */
  	if (percpu_ref_is_dying(&ctx->refs))

fs/io_uring.c static bool io_rw_should_reissue(struct io_kiocb *req) · line 2458

Quoted 10 lines
fs/io_uring.c +6 -0
  		return false;
+	/*
+	 * Play it safe and assume not safe to re-import and reissue if we're
+	 * not in the original thread group (or in task context).
+	 */
+	if (!same_thread_group(req->task, current) || !in_task())
+		return false;
  	return true;
  }
  #else

Hi,

thank you for the fix! This does indeed prevent the panic (with 5.11.22) and hang (with 5.13.3) with my problematic workload.

Best Regards,
Fabian

author maintainer
#4 lore raw

Re: [PATCH 2/2] io_uring: don't block level reissue off completion path

Quoted 42 lines · Fabian Ebner

Am 27.07.21 um 18:58 schrieb Jens Axboe:

Quoted 28 lines

Some setups, like SCSI, can throw spurious -EAGAIN off the softirq completion path. Normally we expect this to happen inline as part of submission, but apparently SCSI has a weird corner case where it can happen as part of normal completions.

This should be solved by having the -EAGAIN bubble back up the stack as part of submission, but previous attempts at this failed and we're not just quite there yet. Instead we currently use REQ_F_REISSUE to handle this case.

For now, catch it in io_rw_should_reissue() and prevent a reissue from a bogus path.

---
  fs/io_uring.c | 6 ++++++
  1 file changed, 6 insertions(+)
fs/io_uring.c +0 -0
index 6ba101cd4661..83f67d33bf67 100644
@@ -2447,6 +2447,12 @@ static bool io_rw_should_reissue(struct io_kiocb *req)
  	 */
  	if (percpu_ref_is_dying(&ctx->refs))

fs/io_uring.c static bool io_rw_should_reissue(struct io_kiocb *req) · line 2458

Quoted 10 lines
fs/io_uring.c +6 -0
  		return false;
+	/*
+	 * Play it safe and assume not safe to re-import and reissue if we're
+	 * not in the original thread group (or in task context).
+	 */
+	if (!same_thread_group(req->task, current) || !in_task())
+		return false;
  	return true;
  }
  #else

Hi,

Quoted 2 lines

thank you for the fix! This does indeed prevent the panic (with 5.11.22) and hang (with 5.13.3) with my problematic workload.

Perfect, thanks for re-testing! Can I add your Tested-by to the patch?

Signature · 1 line
Jens Axboe
#5 lore raw

Re: [PATCH 2/2] io_uring: don't block level reissue off completion path

Am 28.07.21 um 15:23 schrieb Jens Axboe:

Quoted 43 lines · Fabian Ebner
Quoted 42 lines

Am 27.07.21 um 18:58 schrieb Jens Axboe:

Quoted 28 lines

Some setups, like SCSI, can throw spurious -EAGAIN off the softirq completion path. Normally we expect this to happen inline as part of submission, but apparently SCSI has a weird corner case where it can happen as part of normal completions.

This should be solved by having the -EAGAIN bubble back up the stack as part of submission, but previous attempts at this failed and we're not just quite there yet. Instead we currently use REQ_F_REISSUE to handle this case.

For now, catch it in io_rw_should_reissue() and prevent a reissue from a bogus path.

---
   fs/io_uring.c | 6 ++++++
   1 file changed, 6 insertions(+)
fs/io_uring.c +0 -0
index 6ba101cd4661..83f67d33bf67 100644
@@ -2447,6 +2447,12 @@ static bool io_rw_should_reissue(struct io_kiocb *req)
   	 */
   	if (percpu_ref_is_dying(&ctx->refs))

fs/io_uring.c static bool io_rw_should_reissue(struct io_kiocb *req) · line 2458

Quoted 10 lines
fs/io_uring.c +6 -0
   		return false;
+	/*
+	 * Play it safe and assume not safe to re-import and reissue if we're
+	 * not in the original thread group (or in task context).
+	 */
+	if (!same_thread_group(req->task, current) || !in_task())
+		return false;
   	return true;
   }
   #else

Hi,

Quoted 5 lines
Quoted 2 lines

thank you for the fix! This does indeed prevent the panic (with 5.11.22) and hang (with 5.13.3) with my problematic workload.

Perfect, thanks for re-testing! Can I add your Tested-by to the patch?

Sure, feel free to do so.

Best Regards,
Fabian