linux-kernel - [PATCH 00/13] leds: Fix devm vs. non-devm ordering

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]

Message-ID: <1667983694-15040-1-git-send-email-wangyufen@huawei.com>
Date:   Wed, 9 Nov 2022 16:48:01 +0800
From:   Wang Yufen <wangyufen@...wei.com>
To:     <linux-leds@...r.kernel.org>, <linux-kernel@...r.kernel.org>
CC:     <pavel@....cz>, Wang Yufen <wangyufen@...wei.com>
Subject: [PATCH 00/13] leds: Fix devm vs. non-devm ordering

Fault-inject tests reports this issue:

DEBUG_LOCKS_WARN_ON(lock->magic != lock)
WARNING: CPU: 2 PID: 52 at kernel/locking/mutex.c:582 __mutex_lock+0x1366/0x15b0
Call Trace:
 <TASK>
   cr0014114_set_sync+0x2d/0x80 [leds_cr0014114 dbd1de3fefae3e163bcc08f4eeaa6d1b243203a9]
   set_brightness_delayed+0xc2/0x140
   process_one_work+0x651/0xc30
   worker_thread+0x30b/0x820
   kthread+0x1a0/0x1e0
   ret_from_fork+0x1f/0x30
 </TASK>

The issue occurs in the following scenarios:

.probe()
  mutex_init()
  led->brightness_set_blocking = callback  <-- e.g. cr0014114_set_sync()
  devm_led_classdev_register_ext()
  <-- dr->node.release = devm_led_classdev_release()
...

.remove()
  mutex_destroy(lock)         <-- lock destroy

worker_thread()
  set_brightness_work
    set_brightness_delayed()
      __led_set_brightness_blocking()
        led_cdev->brightness_set_blocking()
        <-- callback, e.g. cr0014114_set_sync()
          mutex_lock(lock)              <-- lock is used after destroy

.release()
  devm_led_classdev_release()
    led_classdev_unregister()
      <-- flush set_brightness_work

When non-devm resources are allocated they mustn't be followed by devm
allocations, otherwise it will break the tear down ordering and might
lead to crashes or other bugs during ->remove() stage. Fix this by
wrapping mutex_destroy() call with devm_add_action_or_reset().

Wang Yufen (13):
  leds: cr0014114: Fix devm vs. non-devm ordering
  leds: el15203000: Fix devm vs. non-devm ordering
  leds: lm3532: Fix devm vs. non-devm ordering
  leds: lm3692x: Fix devm vs. non-devm ordering
  leds: lm3697: Fix devm vs. non-devm ordering
  leds: lp50xx: Fix devm vs. non-devm ordering
  leds: lp8860: Fix devm vs. non-devm ordering
  leds: mlxreg: Fix devm vs. non-devm ordering
  leds: mt6323: Fix devm vs. non-devm ordering
  leds: powernv: Fix devm vs. non-devm ordering
  leds: sc27xx: Fix devm vs. non-devm ordering
  leds: spi-byte: Fix devm vs. non-devm ordering
  leds: rt8515: Fix devm vs. non-devm ordering

 drivers/leds/flash/leds-rt8515.c | 11 +++++++++--
 drivers/leds/leds-cr0014114.c    | 11 ++++++++++-
 drivers/leds/leds-el15203000.c   | 18 +++++++++++-------
 drivers/leds/leds-lm3532.c       | 10 ++++++++++
 drivers/leds/leds-lm3692x.c      | 11 ++++++++++-
 drivers/leds/leds-lm3697.c       | 13 ++++++++++---
 drivers/leds/leds-lp50xx.c       | 12 ++++++++++--
 drivers/leds/leds-lp8860.c       | 11 +++++++++--
 drivers/leds/leds-mlxreg.c       | 20 ++++++++++----------
 drivers/leds/leds-mt6323.c       | 11 +++++++++--
 drivers/leds/leds-powernv.c      | 12 +++++++++---
 drivers/leds/leds-sc27xx-bltc.c  | 27 +++++++++++----------------
 drivers/leds/leds-spi-byte.c     | 21 ++++++++++-----------
 13 files changed, 128 insertions(+), 60 deletions(-)

-- 
1.8.3.1