public inbox for [email protected]
 help / color / mirror / Atom feed
From: Jens Axboe <[email protected]>
To: [email protected]
Cc: Jens Axboe <[email protected]>
Subject: [PATCH 10/33] io-wq: provide an io_wq_put_and_exit() helper
Date: Wed,  3 Mar 2021 17:26:37 -0700	[thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>

If we put the io-wq from io_uring, we really want it to exit. Provide
a helper that does that for us. Couple that with not having the manager
hold a reference to the 'wq' and the normal SQPOLL exit will tear down
the io-wq context appropriate.

On the io-wq side, our wq context is per task, so only the task itself
is manipulating ->manager and hence it's safe to check and clear without
any extra locking. We just need to ensure that the manager task stays
around, in case it exits.

Signed-off-by: Jens Axboe <[email protected]>
---
 fs/io-wq.c    | 29 +++++++++++++++++++----------
 fs/io-wq.h    |  1 +
 fs/io_uring.c |  2 +-
 3 files changed, 21 insertions(+), 11 deletions(-)

diff --git a/fs/io-wq.c b/fs/io-wq.c
index 23f431747cd2..65ae35ca8dba 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -748,7 +748,7 @@ static int io_wq_manager(void *data)
 	sprintf(buf, "iou-mgr-%d", wq->task_pid);
 	set_task_comm(current, buf);
 	current->flags |= PF_IO_WORKER;
-	wq->manager = current;
+	wq->manager = get_task_struct(current);
 
 	complete(&wq->started);
 
@@ -764,9 +764,7 @@ static int io_wq_manager(void *data)
 	/* we might not ever have created any workers */
 	if (atomic_read(&wq->worker_refs))
 		wait_for_completion(&wq->worker_done);
-	wq->manager = NULL;
 	complete(&wq->exited);
-	io_wq_put(wq);
 	do_exit(0);
 }
 
@@ -809,8 +807,6 @@ static int io_wq_fork_manager(struct io_wq *wq)
 		return 0;
 
 	reinit_completion(&wq->worker_done);
-	clear_bit(IO_WQ_BIT_EXIT, &wq->state);
-	refcount_inc(&wq->refs);
 	current->flags |= PF_IO_WORKER;
 	ret = io_wq_fork_thread(io_wq_manager, wq);
 	current->flags &= ~PF_IO_WORKER;
@@ -1082,6 +1078,16 @@ struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data)
 	return ERR_PTR(ret);
 }
 
+static void io_wq_destroy_manager(struct io_wq *wq)
+{
+	if (wq->manager) {
+		wake_up_process(wq->manager);
+		wait_for_completion(&wq->exited);
+		put_task_struct(wq->manager);
+		wq->manager = NULL;
+	}
+}
+
 static void io_wq_destroy(struct io_wq *wq)
 {
 	int node;
@@ -1089,10 +1095,7 @@ static void io_wq_destroy(struct io_wq *wq)
 	cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node);
 
 	set_bit(IO_WQ_BIT_EXIT, &wq->state);
-	if (wq->manager) {
-		wake_up_process(wq->manager);
-		wait_for_completion(&wq->exited);
-	}
+	io_wq_destroy_manager(wq);
 
 	rcu_read_lock();
 	for_each_node(node)
@@ -1110,7 +1113,6 @@ static void io_wq_destroy(struct io_wq *wq)
 	io_wq_put_hash(wq->hash);
 	kfree(wq->wqes);
 	kfree(wq);
-
 }
 
 void io_wq_put(struct io_wq *wq)
@@ -1119,6 +1121,13 @@ void io_wq_put(struct io_wq *wq)
 		io_wq_destroy(wq);
 }
 
+void io_wq_put_and_exit(struct io_wq *wq)
+{
+	set_bit(IO_WQ_BIT_EXIT, &wq->state);
+	io_wq_destroy_manager(wq);
+	io_wq_put(wq);
+}
+
 static bool io_wq_worker_affinity(struct io_worker *worker, void *data)
 {
 	struct task_struct *task = worker->task;
diff --git a/fs/io-wq.h b/fs/io-wq.h
index b6ca12b60c35..f6ef433df8a8 100644
--- a/fs/io-wq.h
+++ b/fs/io-wq.h
@@ -114,6 +114,7 @@ struct io_wq_data {
 
 struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data);
 void io_wq_put(struct io_wq *wq);
+void io_wq_put_and_exit(struct io_wq *wq);
 
 void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work);
 void io_wq_hash_work(struct io_wq_work *work, void *val);
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 904bf0fecc36..cb65e54c1b09 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -8857,7 +8857,7 @@ void __io_uring_files_cancel(struct files_struct *files)
 	if (files) {
 		io_uring_remove_task_files(tctx);
 		if (tctx->io_wq) {
-			io_wq_put(tctx->io_wq);
+			io_wq_put_and_exit(tctx->io_wq);
 			tctx->io_wq = NULL;
 		}
 	}
-- 
2.30.1


  parent reply	other threads:[~2021-03-04  1:10 UTC|newest]

Thread overview: 36+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-03-04  0:26 [PATCHSET 0/33] Fixes queued up for 5.12 Jens Axboe
2021-03-04  0:26 ` [PATCH 01/33] io-wq: wait for worker startup when forking a new one Jens Axboe
2021-03-04  0:26 ` [PATCH 02/33] io-wq: have manager wait for all workers to exit Jens Axboe
2021-03-04  0:26 ` [PATCH 03/33] io-wq: don't ask for a new worker if we're exiting Jens Axboe
2021-03-04  0:26 ` [PATCH 04/33] io-wq: rename wq->done completion to wq->started Jens Axboe
2021-03-04  0:26 ` [PATCH 05/33] io-wq: wait for manager exit on wq destroy Jens Axboe
2021-03-04  0:26 ` [PATCH 06/33] io-wq: fix double put of 'wq' in error path Jens Axboe
2021-03-04  0:26 ` [PATCH 07/33] io_uring: SQPOLL stop error handling fixes Jens Axboe
2021-03-04  0:26 ` [PATCH 08/33] io_uring: run fallback on cancellation Jens Axboe
2021-03-04  0:26 ` [PATCH 09/33] io_uring: don't use complete_all() on SQPOLL thread exit Jens Axboe
2021-03-04  0:26 ` Jens Axboe [this message]
2021-03-04  0:26 ` [PATCH 11/33] io_uring: fix race condition in task_work add and clear Jens Axboe
2021-03-04  0:26 ` [PATCH 12/33] io_uring: signal worker thread unshare Jens Axboe
2021-03-04 12:15   ` Stefan Metzmacher
2021-03-04 14:05     ` Jens Axboe
2021-03-04  0:26 ` [PATCH 13/33] io_uring: warn on not destroyed io-wq Jens Axboe
2021-03-04  0:26 ` [PATCH 14/33] io_uring: destroy io-wq on exec Jens Axboe
2021-03-04  0:26 ` [PATCH 15/33] io_uring: remove unused argument 'tsk' from io_req_caches_free() Jens Axboe
2021-03-04  0:26 ` [PATCH 16/33] io_uring: kill unnecessary REQ_F_WORK_INITIALIZED checks Jens Axboe
2021-03-04  0:26 ` [PATCH 17/33] io_uring: move cred assignment into io_issue_sqe() Jens Axboe
2021-03-04  0:26 ` [PATCH 18/33] io_uring: kill unnecessary io_run_ctx_fallback() in io_ring_exit_work() Jens Axboe
2021-03-04  0:26 ` [PATCH 19/33] io_uring: kill io_uring_flush() Jens Axboe
2021-03-04  0:26 ` [PATCH 20/33] io_uring: fix __tctx_task_work() ctx race Jens Axboe
2021-03-04  0:26 ` [PATCH 21/33] io_uring: replace cmpxchg in fallback with xchg Jens Axboe
2021-03-04  0:26 ` [PATCH 22/33] io_uring: ensure that SQPOLL thread is started for exit Jens Axboe
2021-03-04  0:26 ` [PATCH 23/33] io_uring: ignore double poll add on the same waitqueue head Jens Axboe
2021-03-04  0:26 ` [PATCH 24/33] io_uring: kill sqo_dead and sqo submission halting Jens Axboe
2021-03-04  0:26 ` [PATCH 25/33] io_uring: remove sqo_task Jens Axboe
2021-03-04  0:26 ` [PATCH 26/33] io-wq: fix error path leak of buffered write hash map Jens Axboe
2021-03-04  0:26 ` [PATCH 27/33] io_uring: fix -EAGAIN retry with IOPOLL Jens Axboe
2021-03-04  0:26 ` [PATCH 28/33] io_uring: choose right tctx->io_wq for try cancel Jens Axboe
2021-03-04  0:26 ` [PATCH 29/33] io_uring: inline io_req_clean_work() Jens Axboe
2021-03-04  0:26 ` [PATCH 30/33] io_uring: inline __io_queue_async_work() Jens Axboe
2021-03-04  0:26 ` [PATCH 31/33] io_uring: remove extra in_idle wake up Jens Axboe
2021-03-04  0:26 ` [PATCH 32/33] io_uring: ensure that threads freeze on suspend Jens Axboe
2021-03-04  0:27 ` [PATCH 33/33] io-wq: ensure all pending work is canceled on exit Jens Axboe

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 \
    [email protected] \
    [email protected] \
    [email protected] \
    /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