From: Jiayuan Chen <jiayuan.chen@linux.dev>
To: linux-rt-devel@lists.linux.dev
Cc: Jiayuan Chen <jiayuan.chen@linux.dev>,
Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
Clark Williams <clrkwllms@kernel.org>,
Steven Rostedt <rostedt@goodmis.org>,
linux-kernel@vger.kernel.org
Subject: [PATCH v1] irq_work: Fix use-after-free in irq_work_single on PREEMPT_RT
Date: Wed, 25 Mar 2026 11:05:04 +0800 [thread overview]
Message-ID: <20260325030508.321405-1-jiayuan.chen@linux.dev> (raw)
On PREEMPT_RT, non-HARD irq_work runs in a per-CPU kthread, so
irq_work_sync() uses rcuwait (sleeping) to wait for BUSY==0.
After irq_work_single() clears BUSY via atomic_cmpxchg(), an
irq_work_sync() caller on another CPU that enters *after* BUSY is
cleared can observe BUSY==0 immediately (without sleeping), return,
and free the work. Meanwhile irq_work_single() still dereferences
@work for irq_work_is_hard() and rcuwait_wake_up(), causing a
use-after-free.
Note: if a sync waiter is actually sleeping, @work is still alive
(it can't be freed until the waiter returns), so there is no UAF in
that case. The UAF only occurs when sync checks BUSY==0 without
going through schedule().
Fix this by extracting and pinning the irq_work_sync waiter's
task_struct (if any) while BUSY is still set and @work is guaranteed
alive. After clearing BUSY, wake the pinned task directly without
touching @work.
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
---
kernel/irq_work.c | 29 ++++++++++++++++++++++++++---
1 file changed, 26 insertions(+), 3 deletions(-)
diff --git a/kernel/irq_work.c b/kernel/irq_work.c
index 73f7e1fd4ab4..b10b75d1cc09 100644
--- a/kernel/irq_work.c
+++ b/kernel/irq_work.c
@@ -200,6 +200,7 @@ bool irq_work_needs_cpu(void)
void irq_work_single(void *arg)
{
+ struct task_struct *waiter = NULL;
struct irq_work *work = arg;
int flags;
@@ -221,15 +222,37 @@ void irq_work_single(void *arg)
work->func(work);
lockdep_irq_work_exit(flags);
+ /*
+ * Extract and pin the irq_work_sync() waiter before clearing
+ * BUSY. Once BUSY is cleared, @work may be freed immediately
+ * by a sync caller that observes BUSY==0 without sleeping, so
+ * @work must not be dereferenced after the cmpxchg below.
+ */
+ if ((IS_ENABLED(CONFIG_PREEMPT_RT) && !irq_work_is_hard(work)) ||
+ !arch_irq_work_has_interrupt()) {
+ rcu_read_lock();
+ waiter = rcu_dereference(work->irqwait.task);
+ if (waiter)
+ get_task_struct(waiter);
+ rcu_read_unlock();
+ }
+
/*
* Clear the BUSY bit, if set, and return to the free state if no-one
* else claimed it meanwhile.
*/
(void)atomic_cmpxchg(&work->node.a_flags, flags, flags & ~IRQ_WORK_BUSY);
- if ((IS_ENABLED(CONFIG_PREEMPT_RT) && !irq_work_is_hard(work)) ||
- !arch_irq_work_has_interrupt())
- rcuwait_wake_up(&work->irqwait);
+ /*
+ * @work must not be dereferenced past this point. Wake the
+ * pinned waiter if one was sleeping; if none was sleeping,
+ * either irq_work_sync() has not been called or it will
+ * observe BUSY==0 on its own.
+ */
+ if (waiter) {
+ wake_up_process(waiter);
+ put_task_struct(waiter);
+ }
}
static void irq_work_run_list(struct llist_head *list)
--
2.43.0
next reply other threads:[~2026-03-25 3:05 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-03-25 3:05 Jiayuan Chen [this message]
2026-03-25 15:13 ` [PATCH v1] irq_work: Fix use-after-free in irq_work_single on PREEMPT_RT Steven Rostedt
2026-03-25 15:38 ` Sebastian Andrzej Siewior
2026-03-25 15:53 ` Steven Rostedt
2026-03-25 15:55 ` Sebastian Andrzej Siewior
2026-03-25 16:34 ` Jiayuan Chen
2026-03-25 17:05 ` Sebastian Andrzej Siewior
2026-03-25 17:44 ` Steven Rostedt
2026-03-25 17:51 ` Sebastian Andrzej Siewior
2026-03-25 17:55 ` Steven Rostedt
2026-03-25 17:59 ` Sebastian Andrzej Siewior
2026-03-26 2:27 ` Jiayuan Chen
2026-03-26 8:11 ` Sebastian Andrzej Siewior
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 \
--in-reply-to=20260325030508.321405-1-jiayuan.chen@linux.dev \
--to=jiayuan.chen@linux.dev \
--cc=bigeasy@linutronix.de \
--cc=clrkwllms@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rt-devel@lists.linux.dev \
--cc=rostedt@goodmis.org \
/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