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: <385b7baa-8027-c4cf-948f-a1dff570befc@gmail.com>
Date: Thu, 17 Oct 2024 14:12:40 +0200
From: Jacek Anaszewski <jacek.anaszewski@...il.com>
To: anish kumar <yesanishhere@...il.com>, Mukesh Ojha <quic_mojha@...cinc.com>
Cc: Pavel Machek <pavel@....cz>, Lee Jones <lee@...nel.org>,
 linux-leds@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] leds: class: Protect brightness_show() with
 led_cdev->led_access mutex

Hi Anish and Mukesh,

On 10/16/24 18:37, anish kumar wrote:
> On Tue, Oct 15, 2024 at 10:45 PM Mukesh Ojha <quic_mojha@...cinc.com> wrote:
>>
>> On Tue, Oct 15, 2024 at 03:28:08PM -0700, anish kumar wrote:
>>> On Tue, Oct 15, 2024 at 12:28 PM Mukesh Ojha <quic_mojha@...cinc.com> wrote:
>>>>
>>>> On Tue, Oct 15, 2024 at 10:59:12AM -0700, anish kumar wrote:
>>>>> On Tue, Oct 15, 2024 at 9:26 AM Mukesh Ojha <quic_mojha@...cinc.com> wrote:
>>>>>>
>>>>>> There is NULL pointer issue observed if from Process A where hid device
>>>>>> being added which results in adding a led_cdev addition and later a
>>>>>> another call to access of led_cdev attribute from Process B can result
>>>>>> in NULL pointer issue.
>>>>>
>>>>> Which pointer is NULL? Call stack shows that dualshock4_led_get_brightness
>>>>> function could be culprit?
>>>>
>>>> in dualshock4_led_get_brightness()[1], led->dev is NULL here, as [2]
>>>> is not yet completed.
>>>>
>>>> [1]
>>>>   struct hid_device *hdev = to_hid_device(led->dev->parent);
>>>>
>>>> [2]
>>>> led_cdev->dev = device_create_with_groups(&leds_class, parent, 0,
>>>>                    led_cdev, led_cdev->groups, "%s", final_name);
>>>>
>>>>>
>>>>>>
>>>>>> Use mutex led_cdev->led_access to protect access to led->cdev and its
>>>>>> attribute inside brightness_show().
>>>>>
>>>>> I don't think it is needed here because it is just calling the led driver
>>>>> callback and updating the brightness. So, why would we need to serialize
>>>>> that using mutex? Maybe the callback needs some debugging.
>>>>> I'm curious if it is ready by the time the callback is invoked.
>>>>
>>>> Because, we should not be allowed to access led_cdev->dev as it is not
>>>> completed and since, brightness_store() has this lock brightness_show()
>>>> should also have this as we are seeing the issue without it.
>>>>
>>>> I hope, above might have answered your question.
>>>>
>>>> -Mukesh
>>>>>
>>>>>>
>>>>>>          Process A                               Process B
>>>>>>
>>>>>>   kthread+0x114
>>>>>>   worker_thread+0x244
>>>>>>   process_scheduled_works+0x248
>>>>>>   uhid_device_add_worker+0x24
>>>>>>   hid_add_device+0x120
>>>>>>   device_add+0x268
>>>>>>   bus_probe_device+0x94
>>>>>>   device_initial_probe+0x14
>>>>>>   __device_attach+0xfc
>>>>>>   bus_for_each_drv+0x10c
>>>>>>   __device_attach_driver+0x14c
>>>>>>   driver_probe_device+0x3c
>>>>>>   __driver_probe_device+0xa0
>>>>>>   really_probe+0x190
>>>>>>   hid_device_probe+0x130
>>>>>>   ps_probe+0x990
>>>>>>   ps_led_register+0x94
>>>>>>   devm_led_classdev_register_ext+0x58
>>>>>>   led_classdev_register_ext+0x1f8
>>>>>>   device_create_with_groups+0x48
>>>>>>   device_create_groups_vargs+0xc8
>>>>>>   device_add+0x244
>>>>>>   kobject_uevent+0x14
>>>>>>   kobject_uevent_env[jt]+0x224
>>>>>>   mutex_unlock[jt]+0xc4
>>>>>>   __mutex_unlock_slowpath+0xd4
>>>>>>   wake_up_q+0x70
>>>>>>   try_to_wake_up[jt]+0x48c
>>>>>>   preempt_schedule_common+0x28
>>>>>>   __schedule+0x628
>>>>>>   __switch_to+0x174
>>>>>>                                                  el0t_64_sync+0x1a8/0x1ac
>>>>>>                                                  el0t_64_sync_handler+0x68/0xbc
>>>>>>                                                  el0_svc+0x38/0x68
>>>>>>                                                  do_el0_svc+0x1c/0x28
>>>>>>                                                  el0_svc_common+0x80/0xe0
>>>>>>                                                  invoke_syscall+0x58/0x114
>>>>>>                                                  __arm64_sys_read+0x1c/0x2c
>>>>>>                                                  ksys_read+0x78/0xe8
>>>>>>                                                  vfs_read+0x1e0/0x2c8
>>>>>>                                                  kernfs_fop_read_iter+0x68/0x1b4
>>>>>>                                                  seq_read_iter+0x158/0x4ec
>>>>>>                                                  kernfs_seq_show+0x44/0x54
>>>>>>                                                  sysfs_kf_seq_show+0xb4/0x130
>>>>>>                                                  dev_attr_show+0x38/0x74
>>>>>>                                                  brightness_show+0x20/0x4c
>>>>>>                                                  dualshock4_led_get_brightness+0xc/0x74
>>>>>>
>>>>>> [ 3313.874295][ T4013] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000060
>>>>>> [ 3313.874301][ T4013] Mem abort info:
>>>>>> [ 3313.874303][ T4013]   ESR = 0x0000000096000006
>>>>>> [ 3313.874305][ T4013]   EC = 0x25: DABT (current EL), IL = 32 bits
>>>>>> [ 3313.874307][ T4013]   SET = 0, FnV = 0
>>>>>> [ 3313.874309][ T4013]   EA = 0, S1PTW = 0
>>>>>> [ 3313.874311][ T4013]   FSC = 0x06: level 2 translation fault
>>>>>> [ 3313.874313][ T4013] Data abort info:
>>>>>> [ 3313.874314][ T4013]   ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
>>>>>> [ 3313.874316][ T4013]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
>>>>>> [ 3313.874318][ T4013]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
>>>>>> [ 3313.874320][ T4013] user pgtable: 4k pages, 39-bit VAs, pgdp=00000008f2b0a000
>>>>>> ..
>>>>>>
>>>>>> [ 3313.874332][ T4013] Dumping ftrace buffer:
>>>>>> [ 3313.874334][ T4013]    (ftrace buffer empty)
>>>>>> ..
>>>>>> ..
>>>>>> [ dd3313.874639][ T4013] CPU: 6 PID: 4013 Comm: InputReader
>>>>>> [ 3313.874648][ T4013] pc : dualshock4_led_get_brightness+0xc/0x74
>>>>>> [ 3313.874653][ T4013] lr : led_update_brightness+0x38/0x60
>>>>>> [ 3313.874656][ T4013] sp : ffffffc0b910bbd0
>>>>>> ..
>>>>>> ..
>>>>>> [ 3313.874685][ T4013] Call trace:
>>>>>> [ 3313.874687][ T4013]  dualshock4_led_get_brightness+0xc/0x74
>>>>>> [ 3313.874690][ T4013]  brightness_show+0x20/0x4c
>>>>>> [ 3313.874692][ T4013]  dev_attr_show+0x38/0x74
>>>>>> [ 3313.874696][ T4013]  sysfs_kf_seq_show+0xb4/0x130
>>>>>> [ 3313.874700][ T4013]  kernfs_seq_show+0x44/0x54
>>>>>> [ 3313.874703][ T4013]  seq_read_iter+0x158/0x4ec
>>>>>> [ 3313.874705][ T4013]  kernfs_fop_read_iter+0x68/0x1b4
>>>>>> [ 3313.874708][ T4013]  vfs_read+0x1e0/0x2c8
>>>>>> [ 3313.874711][ T4013]  ksys_read+0x78/0xe8
>>>>>> [ 3313.874714][ T4013]  __arm64_sys_read+0x1c/0x2c
>>>>>> [ 3313.874718][ T4013]  invoke_syscall+0x58/0x114
>>>>>> [ 3313.874721][ T4013]  el0_svc_common+0x80/0xe0
>>>>>> [ 3313.874724][ T4013]  do_el0_svc+0x1c/0x28
>>>>>> [ 3313.874727][ T4013]  el0_svc+0x38/0x68
>>>>>> [ 3313.874730][ T4013]  el0t_64_sync_handler+0x68/0xbc
>>>>>> [ 3313.874732][ T4013]  el0t_64_sync+0x1a8/0x1ac
>>>>>>
>>>>>> Signed-off-by: Mukesh Ojha <quic_mojha@...cinc.com>
>>>>>> ---
>>>>>>   drivers/leds/led-class.c | 3 ++-
>>>>>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/leds/led-class.c b/drivers/leds/led-class.c
>>>>>> index 06b97fd49ad9..e3cb93f19c06 100644
>>>>>> --- a/drivers/leds/led-class.c
>>>>>> +++ b/drivers/leds/led-class.c
>>>>>> @@ -30,8 +30,9 @@ static ssize_t brightness_show(struct device *dev,
>>>>>>   {
>>>>>>          struct led_classdev *led_cdev = dev_get_drvdata(dev);
>>>>>>
>>>>>> -       /* no lock needed for this */
>>>
>>> just get rid of the above comment then.
>>
>> If you notice, it is already removed (-) .
>>
>>>
>>> Also, the comment below in file leds.h
>>> needs an update as originally the idea for this mutex lock was to
>>> provide quick feedback to userspace based on this commit
>>> https://github.com/torvalds/linux/commit/acd899e4f3066b6662f6047da5b795cc762093cb
>>>
>>> Basically a comment somewhere so that when a new attribute
>>> gets added, it doesn't make the same mistake of not using the mutex
>>> and run into the same issue.
>>>
>>> /* Ensures consistent access to the LED Flash Class device */
>>> struct mutex led_access;
>>
>> Thanks for accepting that it is an issue.
>> I think, comment is very obvious actually the patch you mentioned should
>> be in fixes tag as it introduced the lock but did not protect the show
>> while it does it for store.
> 
> Yes, but that patch was added for supporting flash class
> device and wasn't explicitly to take care of the scenario that you
> are trying to handle and the above comment in leds.h states the same.

Correct. led_access mutex was introduced to add support for preventing
any LED class device state changes originating from sysfs while
v4l2_flash wrapper owns the device.

Since the inception of LED subsystem all the locking was deemed to be
the responsibility of every single LED class driver and initially sysfs
attr callbacks didn't have any locking. After some time when LED core
started to grow it turned out that it was required to lock the LED class
initialization sequence, so as not to give the userspace an opportunity
to set LED brightness on not fully initialized device, which was
introduced in [0]. led_access mutex was already in place so it was used.
However as you noticed, it is not used consistently across all LED class
sysfs attrs callbacks.

Since brightness_show() does not acquire led_access mutex it is still
possible to call brightness_get op when LED class initialization
sequence is not yet finished.

Still, I'd propose to first narrow down the issue and figure out what
actually causes NULL pointer dereference, as it apparently
originates from dualshock4_led_get_brightness and not from LED core.

I bet that the driver is not fully initialized up to the point when
devm_led_classdev_register_ext() is called in it.

> 
> I think we should modify that comment and state clearly that
> the aforementioned mutex is also to handle access to led_cdev->dev.
> Either here in this .h or where attributes are defined, so that new attributes
> that get added doesn't suffer from the same bug.
> 
> led_trigger_set also this function also suffers from the same bug so you
> need to handle it the same way.

led_trigger_set() is already called with led_access mutex held in
led_trigger_write(), i.e. from "trigger" sysfs attr.

>>
>> Fixes: acd899e4f306 ("leds: implement sysfs interface locking mechanism")
>>
>> -Mukesh
>>>
>>>
>>>>>
>>>>>>> also you missed this.
>>>>>
>>>>>> +       mutex_lock(&led_cdev->led_access);
>>>>>>          led_update_brightness(led_cdev);
>>>>>> +       mutex_unlock(&led_cdev->led_access);
>>>>>>
>>>>>>          return sprintf(buf, "%u\n", led_cdev->brightness);
>>>>>>   }
>>>>>> --
>>>>>> 2.34.1
>>>>>>
>>>>>>
> 

[0] 
https://lore.kernel.org/linux-leds/20180523222221.27621-1-lhenriques@suse.com/

-- 
Best regards,
Jacek Anaszewski

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ