linux-kernel - [PATCH] kthread: Fix kthread_mod_delayed_work vs kthread_cancel_delayed_work

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-next>] [day] [month] [year] [list]

Date:   Thu, 13 May 2021 14:54:57 +0800
From:   Martin Liu <liumartin@...gle.com>
To:     Nathan Chancellor <nathan@...nel.org>,
        Nick Desaulniers <ndesaulniers@...gle.com>,
        Petr Mladek <pmladek@...e.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Tejun Heo <tj@...nel.org>
Cc:     minchan@...gle.com, www@...gle.com, davidchao@...gle.com,
        jenhaochen@...gle.com, Martin Liu <liumartin@...gle.com>,
        linux-kernel@...r.kernel.org, clang-built-linux@...glegroups.com
Subject: [PATCH] kthread: Fix kthread_mod_delayed_work vs kthread_cancel_delayed_work_sync
 race

We encountered a system hang issue while doing the tests. The callstack
is as following

	schedule+0x80/0x100
	schedule_timeout+0x48/0x138
	wait_for_common+0xa4/0x134
	wait_for_completion+0x1c/0x2c
	kthread_flush_work+0x114/0x1cc
	kthread_cancel_work_sync.llvm.16514401384283632983+0xe8/0x144
	kthread_cancel_delayed_work_sync+0x18/0x2c
	xxxx_pm_notify+0xb0/0xd8
	blocking_notifier_call_chain_robust+0x80/0x194
	pm_notifier_call_chain_robust+0x28/0x4c
	suspend_prepare+0x40/0x260
	enter_state+0x80/0x3f4
	pm_suspend+0x60/0xdc
	state_store+0x108/0x144
	kobj_attr_store+0x38/0x88
	sysfs_kf_write+0x64/0xc0
	kernfs_fop_write_iter+0x108/0x1d0
	vfs_write+0x2f4/0x368
	ksys_write+0x7c/0xec

When we started investigating, we found race between
kthread_mod_delayed_work vs kthread_cancel_delayed_work_sync. The race's
result could be simply reproduced as a kthread_mod_delayed_work with
a following kthread_flush_work call.

Thing is we release kthread_mod_delayed_work kspin_lock in
__kthread_cancel_work so it opens a race window for
kthread_cancel_delayed_work_sync to change the canceling count used to
prevent dwork from being requeued before calling kthread_flush_work.
However, we don't check the canceling count after returning from
__kthread_cancel_work and then insert the dwork to the worker. It
results the following kthread_flush_work inserts flush work to dwork's
tail which is at worker's dealyed_work_list. Therefore, flush work will
never get moved to the worker's work_list to be executed. Finally,
kthread_cancel_delayed_work_sync will NOT be able to get completed and
wait forever. The code sequence diagram is as following

Thread A                Thread B
kthread_mod_delayed_work
  spin_lock
   __kthread_cancel_work
    canceling = 1
    spin_unlock
                        kthread_cancel_delayed_work_sync
                          spin_lock
                            kthread_cancel_work
                          canceling = 2
                          spin_unlock
    del_timer_sync
    spin_lock
    canceling = 1 // canceling count gets update in ThreadB before
  queue_delayed_work // dwork is put into the woker’s dealyed_work_list
                        without checking the canceling count
 spin_unlock
                          kthread_flush_work
                            spin_lock
                            Insert flush work // at the tail of the
			                         dwork which is at
						 the worker’s
						 dealyed_work_list
                            spin_unlock
                            wait_for_completion // Thread B stuck here as
			                           flush work will never
						   get executed

The canceling count could change in __kthread_cancel_work as the spinlock
get released and regained in between, let's check the count again before
we queue the delayed work to avoid the race.

Fixes: 37be45d49dec2 ("kthread: allow to cancel kthread work")
Tested-by: David Chao <davidchao@...gle.com>
Signed-off-by: Martin Liu <liumartin@...gle.com>
---
 kernel/kthread.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/kernel/kthread.c b/kernel/kthread.c
index fe3f2a40d61e..064eae335c1f 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -1181,6 +1181,19 @@ bool kthread_mod_delayed_work(struct kthread_worker *worker,
 		goto out;
 
 	ret = __kthread_cancel_work(work, true, &flags);
+
+	/*
+	 * Canceling could run in parallel from kthread_cancel_delayed_work_sync
+	 * and change work's canceling count as the spinlock is released and regain
+	 * in __kthread_cancel_work so we need to check the count again. Otherwise,
+	 * we might incorrectly queue the dwork and further cause
+	 * cancel_delayed_work_sync thread waiting for flush dwork endlessly.
+	 */
+	if (work->canceling) {
+		ret = false;
+		goto out;
+	}
+
 fast_queue:
 	__kthread_queue_delayed_work(worker, dwork, delay);
 out:
-- 
2.31.1.607.g51e8a6a459-goog