linux-kernel - Re: [PATCH v3 1/1] PM / Freezer: Skip zombie/dead processes to

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20250707084214.GD1613200@noisy.programming.kicks-ass.net>
Date: Mon, 7 Jul 2025 10:42:14 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Zihuan Zhang <zhangzihuan@...inos.cn>
Cc: "Rafael J. Wysocki" <rafael@...nel.org>, pavel@...nel.org,
	len.brown@...el.com, linux-pm@...r.kernel.org,
	linux-kernel@...r.kernel.org, Oleg Nesterov <oleg@...hat.com>
Subject: Re: [PATCH v3 1/1] PM / Freezer: Skip zombie/dead processes to

On Mon, Jul 07, 2025 at 09:00:12AM +0800, Zihuan Zhang wrote:
> 
> 在 2025/7/4 17:21, Peter Zijlstra 写道:
> > You're missing the obvious. How about we 'fix' the PF_NOFREEZE handling
> > and help all cases that set that and not only zombies?　　　 　 　 　　 　 　 　 　 　 　 　 　 　 　　
> 
> It sounds like a good idea, but there’s a potential risk in relying solely
> on PF_NOFREEZE: it’s a mutable flag that can be set or cleared dynamically
> during runtime, even within the freeze window.

> If a task changes its PF_NOFREEZE state after passing the early check in
> try_to_freeze_task(), we might skip freezing it incorrectly, leading to
> inconsistent behavior or unexpected task escapes. This is particularly
> tricky for some kernel threads or exit paths where PF_NOFREEZE is
> manipulated as part of cleanup or teardown.

A quick browse through the code seems to suggest that for user tasks,
PF_NOFREEZE is set just like exit_state, once at death.

For kernel threads the situation is a little more complex; but typically
a kthread is spawned with PF_NOFREEZE set, and then some will clear it
again, but always before then calling a TASK_FREEZABLE wait.

The only thing I didn't fully investigate is this
{,un}lock_system_sleep() thing. But that would appear to need at least
the below fixlet.

diff --git a/kernel/power/main.c b/kernel/power/main.c
index 3d484630505a..a415e7d30a2c 100644
--- a/kernel/power/main.c
+++ b/kernel/power/main.c
@@ -52,8 +52,8 @@ void pm_restrict_gfp_mask(void)
 unsigned int lock_system_sleep(void)
 {
 	unsigned int flags = current->flags;
-	current->flags |= PF_NOFREEZE;
 	mutex_lock(&system_transition_mutex);
+	current->flags |= PF_NOFREEZE;
 	return flags;
 }
 EXPORT_SYMBOL_GPL(lock_system_sleep);


Anyway, this seems to suggest something relatively simple like this here
should do:

diff --git a/kernel/freezer.c b/kernel/freezer.c
index 8d530d0949ff..8b7cecd17564 100644
--- a/kernel/freezer.c
+++ b/kernel/freezer.c
@@ -162,20 +162,22 @@ static bool __freeze_task(struct task_struct *p)
  */
 bool freeze_task(struct task_struct *p)
 {
-	unsigned long flags;
-
-	spin_lock_irqsave(&freezer_lock, flags);
-	if (!freezing(p) || frozen(p) || __freeze_task(p)) {
-		spin_unlock_irqrestore(&freezer_lock, flags);
+	/*
+	 * User tasks get NOFREEZE in do_task_dead().
+	 */
+	if ((p->flags & (PF_NOFREEZE | PF_KTHREAD)) == PF_NOFREEZE)
 		return false;
-	}
 
-	if (!(p->flags & PF_KTHREAD))
-		fake_signal_wake_up(p);
-	else
-		wake_up_state(p, TASK_NORMAL);
+	scoped_guard (spinlock_irqsave, &freezer_lock) {
+		if (!freezing(p) || frozen(p) || __freeze_task(p))
+			return false;
+
+		if (!(p->flags & PF_KTHREAD))
+			fake_signal_wake_up(p);
+		else
+			wake_up_state(p, TASK_NORMAL);
+	}
 
-	spin_unlock_irqrestore(&freezer_lock, flags);
 	return true;
 }