[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <YmuNpbnXoOJbh07G@kroah.com>
Date: Fri, 29 Apr 2022 09:03:01 +0200
From: Greg KH <gregkh@...uxfoundation.org>
To: Duoming Zhou <duoming@....edu.cn>
Cc: linux-kernel@...r.kernel.org, kuba@...nel.org,
krzysztof.kozlowski@...aro.org, davem@...emloft.net,
edumazet@...gle.com, pabeni@...hat.com, alexander.deucher@....com,
akpm@...ux-foundation.org, broonie@...nel.org,
netdev@...r.kernel.org, linma@....edu.cn
Subject: Re: [PATCH net v5 1/2] nfc: replace improper check
device_is_registered() in netlink related functions
On Fri, Apr 29, 2022 at 09:14:32AM +0800, Duoming Zhou wrote:
> The device_is_registered() in nfc core is used to check whether
> nfc device is registered in netlink related functions such as
> nfc_fw_download(), nfc_dev_up() and so on. Although device_is_registered()
> is protected by device_lock, there is still a race condition between
> device_del() and device_is_registered(). The root cause is that
> kobject_del() in device_del() is not protected by device_lock.
>
> (cleanup task) | (netlink task)
> |
> nfc_unregister_device | nfc_fw_download
> device_del | device_lock
> ... | if (!device_is_registered)//(1)
> kobject_del//(2) | ...
> ... | device_unlock
>
> The device_is_registered() returns the value of state_in_sysfs and
> the state_in_sysfs is set to zero in kobject_del(). If we pass check in
> position (1), then set zero in position (2). As a result, the check
> in position (1) is useless.
>
> This patch uses bool variable instead of device_is_registered() to judge
> whether the nfc device is registered, which is well synchronized.
>
> Fixes: 3e256b8f8dfa ("NFC: add nfc subsystem core")
> Signed-off-by: Duoming Zhou <duoming@....edu.cn>
> ---
> Changes in v5:
> - Replace device_is_registered() to bool variable.
Reviewed-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Powered by blists - more mailing lists