cgroups.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v4] memcg: Don't wait writeback completion when release memcg.
@ 2025-08-27 20:45 Julian Sun
  2025-08-27 21:25 ` Tejun Heo
  0 siblings, 1 reply; 2+ messages in thread
From: Julian Sun @ 2025-08-27 20:45 UTC (permalink / raw)
  To: cgroups, linux-mm; +Cc: jack, tj, muchun.song

Recently, we encountered the following hung task:

INFO: task kworker/4:1:1334558 blocked for more than 1720 seconds.
[Wed Jul 30 17:47:45 2025] Workqueue: cgroup_destroy css_free_rwork_fn
[Wed Jul 30 17:47:45 2025] Call Trace:
[Wed Jul 30 17:47:45 2025]  __schedule+0x934/0xe10
[Wed Jul 30 17:47:45 2025]  ? complete+0x3b/0x50
[Wed Jul 30 17:47:45 2025]  ? _cond_resched+0x15/0x30
[Wed Jul 30 17:47:45 2025]  schedule+0x40/0xb0
[Wed Jul 30 17:47:45 2025]  wb_wait_for_completion+0x52/0x80
[Wed Jul 30 17:47:45 2025]  ? finish_wait+0x80/0x80
[Wed Jul 30 17:47:45 2025]  mem_cgroup_css_free+0x22/0x1b0
[Wed Jul 30 17:47:45 2025]  css_free_rwork_fn+0x42/0x380
[Wed Jul 30 17:47:45 2025]  process_one_work+0x1a2/0x360
[Wed Jul 30 17:47:45 2025]  worker_thread+0x30/0x390
[Wed Jul 30 17:47:45 2025]  ? create_worker+0x1a0/0x1a0
[Wed Jul 30 17:47:45 2025]  kthread+0x110/0x130
[Wed Jul 30 17:47:45 2025]  ? __kthread_cancel_work+0x40/0x40
[Wed Jul 30 17:47:45 2025]  ret_from_fork+0x1f/0x30

The direct cause is that memcg spends a long time waiting for dirty page
writeback of foreign memcgs during release.

The root causes are:
    a. The wb may have multiple writeback tasks, containing millions
       of dirty pages, as shown below:

>>> for work in list_for_each_entry("struct wb_writeback_work", \
				    wb.work_list.address_of_(), "list"):
...     print(work.nr_pages, work.reason, hex(work))
...
900628  WB_REASON_FOREIGN_FLUSH 0xffff969e8d956b40
1116521 WB_REASON_FOREIGN_FLUSH 0xffff9698332a9540
1275228 WB_REASON_FOREIGN_FLUSH 0xffff969d9b444bc0
1099673 WB_REASON_FOREIGN_FLUSH 0xffff969f0954d6c0
1351522 WB_REASON_FOREIGN_FLUSH 0xffff969e76713340
2567437 WB_REASON_FOREIGN_FLUSH 0xffff9694ae208400
2954033 WB_REASON_FOREIGN_FLUSH 0xffff96a22d62cbc0
3008860 WB_REASON_FOREIGN_FLUSH 0xffff969eee8ce3c0
3337932 WB_REASON_FOREIGN_FLUSH 0xffff9695b45156c0
3348916 WB_REASON_FOREIGN_FLUSH 0xffff96a22c7a4f40
3345363 WB_REASON_FOREIGN_FLUSH 0xffff969e5d872800
3333581 WB_REASON_FOREIGN_FLUSH 0xffff969efd0f4600
3382225 WB_REASON_FOREIGN_FLUSH 0xffff969e770edcc0
3418770 WB_REASON_FOREIGN_FLUSH 0xffff96a252ceea40
3387648 WB_REASON_FOREIGN_FLUSH 0xffff96a3bda86340
3385420 WB_REASON_FOREIGN_FLUSH 0xffff969efc6eb280
3418730 WB_REASON_FOREIGN_FLUSH 0xffff96a348ab1040
3426155 WB_REASON_FOREIGN_FLUSH 0xffff969d90beac00
3397995 WB_REASON_FOREIGN_FLUSH 0xffff96a2d7288800
3293095 WB_REASON_FOREIGN_FLUSH 0xffff969dab423240
3293595 WB_REASON_FOREIGN_FLUSH 0xffff969c765ff400
3199511 WB_REASON_FOREIGN_FLUSH 0xffff969a72d5e680
3085016 WB_REASON_FOREIGN_FLUSH 0xffff969f0455e000
3035712 WB_REASON_FOREIGN_FLUSH 0xffff969d9bbf4b00

    b. The writeback might severely throttled by wbt, with a speed
       possibly less than 100kb/s, leading to a very long writeback time.

>>> wb.write_bandwidth
(unsigned long)24
>>> wb.write_bandwidth
(unsigned long)13

The wb_wait_for_completion() here is probably only used to prevent
use-after-free. Therefore, we manage 'done' separately and automatically
free it.

This allows us to remove wb_wait_for_completion() while preventing
the use-after-free issue.

Fixes: 97b27821b485 ("writeback, memcg: Implement foreign dirty flushing")
Signed-off-by: Julian Sun <sunjunchao@bytedance.com>
---
  Changes in v4:
    * Check done->cnt in memcg_cgwb_waitq_callback_fn().
    * Code cleanups as Tejun suggested.

  Changes in v3:
    * Rename cgwb_frn_wq_entry to cgwb_frn_wait.
    * Define memcg_cgwb_waitq_callback_fn() only when
      CONFIG_CGROUP_WRITEBACK is defined.
    * Embed wb_completion into struct cgwb_frn_wait.

  Change in v2:
    * Use custom waitq function to free resources


 include/linux/memcontrol.h |  7 ++++-
 mm/memcontrol.c            | 53 ++++++++++++++++++++++++++++++++------
 2 files changed, 51 insertions(+), 9 deletions(-)

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index 785173aa0739..b16bc4bd0ad0 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -157,11 +157,16 @@ struct mem_cgroup_thresholds {
  */
 #define MEMCG_CGWB_FRN_CNT	4
 
+struct cgwb_frn_wait {
+	struct wb_completion done;
+	struct wait_queue_entry wq_entry;
+};
+
 struct memcg_cgwb_frn {
 	u64 bdi_id;			/* bdi->id of the foreign inode */
 	int memcg_id;			/* memcg->css.id of foreign inode */
 	u64 at;				/* jiffies_64 at the time of dirtying */
-	struct wb_completion done;	/* tracks in-flight foreign writebacks */
+	struct cgwb_frn_wait *wait;	/* tracks in-flight foreign writebacks */
 };
 
 /*
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 8dd7fbed5a94..80257dba30f8 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -3459,7 +3459,7 @@ void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio,
 		    frn->memcg_id == wb->memcg_css->id)
 			break;
 		if (time_before64(frn->at, oldest_at) &&
-		    atomic_read(&frn->done.cnt) == 1) {
+		    atomic_read(&frn->wait->done.cnt) == 1) {
 			oldest = i;
 			oldest_at = frn->at;
 		}
@@ -3506,12 +3506,12 @@ void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
 		 * already one in flight.
 		 */
 		if (time_after64(frn->at, now - intv) &&
-		    atomic_read(&frn->done.cnt) == 1) {
+		    atomic_read(&frn->wait->done.cnt) == 1) {
 			frn->at = 0;
 			trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id);
 			cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id,
 					       WB_REASON_FOREIGN_FLUSH,
-					       &frn->done);
+					       &frn->wait->done);
 		}
 	}
 }
@@ -3702,13 +3702,29 @@ static void mem_cgroup_free(struct mem_cgroup *memcg)
 	__mem_cgroup_free(memcg);
 }
 
+#ifdef CONFIG_CGROUP_WRITEBACK
+static int memcg_cgwb_waitq_callback_fn(struct wait_queue_entry *wq_entry, unsigned int mode,
+					int flags, void *key)
+{
+	struct cgwb_frn_wait *frn_wait = container_of(wq_entry,
+						      struct cgwb_frn_wait, wq_entry);
+
+	if (!atomic_read(&frn_wait->done.cnt)) {
+		list_del(&wq_entry->entry);
+		kfree(frn_wait);
+	}
+
+	return 0;
+}
+#endif
+
 static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent)
 {
 	struct memcg_vmstats_percpu *statc;
 	struct memcg_vmstats_percpu __percpu *pstatc_pcpu;
 	struct mem_cgroup *memcg;
 	int node, cpu;
-	int __maybe_unused i;
+	int __maybe_unused i = 0;
 	long error;
 
 	memcg = kmem_cache_zalloc(memcg_cachep, GFP_KERNEL);
@@ -3763,9 +3779,17 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent)
 	INIT_LIST_HEAD(&memcg->objcg_list);
 #ifdef CONFIG_CGROUP_WRITEBACK
 	INIT_LIST_HEAD(&memcg->cgwb_list);
-	for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
-		memcg->cgwb_frn[i].done =
+	for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
+		struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i];
+
+		frn->wait = kmalloc(sizeof(struct cgwb_frn_wait), GFP_KERNEL);
+		if (!frn->wait)
+			goto fail;
+
+		frn->wait->done =
 			__WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq);
+		init_wait_func(&frn->wait->wq_entry, memcg_cgwb_waitq_callback_fn);
+	}
 #endif
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 	spin_lock_init(&memcg->deferred_split_queue.split_queue_lock);
@@ -3775,6 +3799,10 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent)
 	lru_gen_init_memcg(memcg);
 	return memcg;
 fail:
+#ifdef CONFIG_CGROUP_WRITEBACK
+	while (i--)
+		kfree(memcg->cgwb_frn[i].wait);
+#endif
 	mem_cgroup_id_remove(memcg);
 	__mem_cgroup_free(memcg);
 	return ERR_PTR(error);
@@ -3912,8 +3940,17 @@ static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
 	int __maybe_unused i;
 
 #ifdef CONFIG_CGROUP_WRITEBACK
-	for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
-		wb_wait_for_completion(&memcg->cgwb_frn[i].done);
+	for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
+		struct cgwb_frn_wait *wait = memcg->cgwb_frn[i].wait;
+
+		/*
+		 * Not necessary to wait for wb completion which might cause task hung,
+		 * only used to free resources. See memcg_cgwb_waitq_callback_fn().
+		 */
+		__add_wait_queue_entry_tail(wait->done.waitq, &wait->wq_entry);
+		if (atomic_dec_and_test(&wait->done.cnt))
+			wake_up_all(wait->done.waitq);
+	}
 #endif
 	if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
 		static_branch_dec(&memcg_sockets_enabled_key);
-- 
2.39.5


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH v4] memcg: Don't wait writeback completion when release memcg.
  2025-08-27 20:45 [PATCH v4] memcg: Don't wait writeback completion when release memcg Julian Sun
@ 2025-08-27 21:25 ` Tejun Heo
  0 siblings, 0 replies; 2+ messages in thread
From: Tejun Heo @ 2025-08-27 21:25 UTC (permalink / raw)
  To: Julian Sun; +Cc: cgroups, linux-mm, jack, muchun.song

On Thu, Aug 28, 2025 at 04:45:57AM +0800, Julian Sun wrote:
> Recently, we encountered the following hung task:
> 
> INFO: task kworker/4:1:1334558 blocked for more than 1720 seconds.
> [Wed Jul 30 17:47:45 2025] Workqueue: cgroup_destroy css_free_rwork_fn
> [Wed Jul 30 17:47:45 2025] Call Trace:
> [Wed Jul 30 17:47:45 2025]  __schedule+0x934/0xe10
> [Wed Jul 30 17:47:45 2025]  ? complete+0x3b/0x50
> [Wed Jul 30 17:47:45 2025]  ? _cond_resched+0x15/0x30
> [Wed Jul 30 17:47:45 2025]  schedule+0x40/0xb0
> [Wed Jul 30 17:47:45 2025]  wb_wait_for_completion+0x52/0x80
> [Wed Jul 30 17:47:45 2025]  ? finish_wait+0x80/0x80
> [Wed Jul 30 17:47:45 2025]  mem_cgroup_css_free+0x22/0x1b0
> [Wed Jul 30 17:47:45 2025]  css_free_rwork_fn+0x42/0x380
> [Wed Jul 30 17:47:45 2025]  process_one_work+0x1a2/0x360
> [Wed Jul 30 17:47:45 2025]  worker_thread+0x30/0x390
> [Wed Jul 30 17:47:45 2025]  ? create_worker+0x1a0/0x1a0
> [Wed Jul 30 17:47:45 2025]  kthread+0x110/0x130
> [Wed Jul 30 17:47:45 2025]  ? __kthread_cancel_work+0x40/0x40
> [Wed Jul 30 17:47:45 2025]  ret_from_fork+0x1f/0x30
> 
> The direct cause is that memcg spends a long time waiting for dirty page
> writeback of foreign memcgs during release.
> 
> The root causes are:
>     a. The wb may have multiple writeback tasks, containing millions
>        of dirty pages, as shown below:
> 
> >>> for work in list_for_each_entry("struct wb_writeback_work", \
> 				    wb.work_list.address_of_(), "list"):
> ...     print(work.nr_pages, work.reason, hex(work))
> ...
> 900628  WB_REASON_FOREIGN_FLUSH 0xffff969e8d956b40
> 1116521 WB_REASON_FOREIGN_FLUSH 0xffff9698332a9540
> 1275228 WB_REASON_FOREIGN_FLUSH 0xffff969d9b444bc0
> 1099673 WB_REASON_FOREIGN_FLUSH 0xffff969f0954d6c0
> 1351522 WB_REASON_FOREIGN_FLUSH 0xffff969e76713340
> 2567437 WB_REASON_FOREIGN_FLUSH 0xffff9694ae208400
> 2954033 WB_REASON_FOREIGN_FLUSH 0xffff96a22d62cbc0
> 3008860 WB_REASON_FOREIGN_FLUSH 0xffff969eee8ce3c0
> 3337932 WB_REASON_FOREIGN_FLUSH 0xffff9695b45156c0
> 3348916 WB_REASON_FOREIGN_FLUSH 0xffff96a22c7a4f40
> 3345363 WB_REASON_FOREIGN_FLUSH 0xffff969e5d872800
> 3333581 WB_REASON_FOREIGN_FLUSH 0xffff969efd0f4600
> 3382225 WB_REASON_FOREIGN_FLUSH 0xffff969e770edcc0
> 3418770 WB_REASON_FOREIGN_FLUSH 0xffff96a252ceea40
> 3387648 WB_REASON_FOREIGN_FLUSH 0xffff96a3bda86340
> 3385420 WB_REASON_FOREIGN_FLUSH 0xffff969efc6eb280
> 3418730 WB_REASON_FOREIGN_FLUSH 0xffff96a348ab1040
> 3426155 WB_REASON_FOREIGN_FLUSH 0xffff969d90beac00
> 3397995 WB_REASON_FOREIGN_FLUSH 0xffff96a2d7288800
> 3293095 WB_REASON_FOREIGN_FLUSH 0xffff969dab423240
> 3293595 WB_REASON_FOREIGN_FLUSH 0xffff969c765ff400
> 3199511 WB_REASON_FOREIGN_FLUSH 0xffff969a72d5e680
> 3085016 WB_REASON_FOREIGN_FLUSH 0xffff969f0455e000
> 3035712 WB_REASON_FOREIGN_FLUSH 0xffff969d9bbf4b00
> 
>     b. The writeback might severely throttled by wbt, with a speed
>        possibly less than 100kb/s, leading to a very long writeback time.
> 
> >>> wb.write_bandwidth
> (unsigned long)24
> >>> wb.write_bandwidth
> (unsigned long)13
> 
> The wb_wait_for_completion() here is probably only used to prevent
> use-after-free. Therefore, we manage 'done' separately and automatically
> free it.
> 
> This allows us to remove wb_wait_for_completion() while preventing
> the use-after-free issue.
> 
> Fixes: 97b27821b485 ("writeback, memcg: Implement foreign dirty flushing")
> Signed-off-by: Julian Sun <sunjunchao@bytedance.com>

Acked-by: Tejun Heo <tj@kernel.org>

Thanks.

-- 
tejun

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2025-08-27 21:25 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-08-27 20:45 [PATCH v4] memcg: Don't wait writeback completion when release memcg Julian Sun
2025-08-27 21:25 ` Tejun Heo

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).