[v5] wifi: brcmfmac: Fix use-after-free bug in brcmf_cfg80211_detach

Message ID 20231106141704.866455-1-zyytlz.wz@163.com
State New
Headers
Series [v5] wifi: brcmfmac: Fix use-after-free bug in brcmf_cfg80211_detach |

Commit Message

Zheng Wang Nov. 6, 2023, 2:17 p.m. UTC
  This is the candidate patch of CVE-2023-47233 :
https://nvd.nist.gov/vuln/detail/CVE-2023-47233

In brcm80211 driver,it starts with the following invoking chain
to start init a timeout worker:

->brcmf_usb_probe
  ->brcmf_usb_probe_cb
    ->brcmf_attach
      ->brcmf_bus_started
        ->brcmf_cfg80211_attach
          ->wl_init_priv
            ->brcmf_init_escan
              ->INIT_WORK(&cfg->escan_timeout_work,
		  brcmf_cfg80211_escan_timeout_worker);

If we disconnect the USB by hotplug, it will call
brcmf_usb_disconnect to make cleanup. The invoking chain is :

brcmf_usb_disconnect
  ->brcmf_usb_disconnect_cb
    ->brcmf_detach
      ->brcmf_cfg80211_detach
        ->kfree(cfg);

While the timeout woker may still be running. This will cause
a use-after-free bug on cfg in brcmf_cfg80211_escan_timeout_worker.

Fix it by deleting the timer and canceling the worker in
brcmf_cfg80211_detach.

Fixes: e756af5b30b0 ("brcmfmac: add e-scan support.")
Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
Cc: stable@vger.kernel.org
---
v5:
- replace del_timer_sync with timer_shutdown_sync suggested by
Arend and Takashi
v4:
- rename the subject and add CVE number as Ping-Ke Shih suggested
v3:
- rename the subject as Johannes suggested
v2:
- fix the error of kernel test bot reported
---
 drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 2 ++
 1 file changed, 2 insertions(+)
  

Comments

Kalle Valo Nov. 6, 2023, 2:41 p.m. UTC | #1
Zheng Wang <zyytlz.wz@163.com> writes:

> This is the candidate patch of CVE-2023-47233 :
> https://nvd.nist.gov/vuln/detail/CVE-2023-47233
>
> In brcm80211 driver,it starts with the following invoking chain
> to start init a timeout worker:
>
> ->brcmf_usb_probe
>   ->brcmf_usb_probe_cb
>     ->brcmf_attach
>       ->brcmf_bus_started
>         ->brcmf_cfg80211_attach
>           ->wl_init_priv
>             ->brcmf_init_escan
>               ->INIT_WORK(&cfg->escan_timeout_work,
> 		  brcmf_cfg80211_escan_timeout_worker);
>
> If we disconnect the USB by hotplug, it will call
> brcmf_usb_disconnect to make cleanup. The invoking chain is :
>
> brcmf_usb_disconnect
>   ->brcmf_usb_disconnect_cb
>     ->brcmf_detach
>       ->brcmf_cfg80211_detach
>         ->kfree(cfg);
>
> While the timeout woker may still be running. This will cause
> a use-after-free bug on cfg in brcmf_cfg80211_escan_timeout_worker.
>
> Fix it by deleting the timer and canceling the worker in
> brcmf_cfg80211_detach.
>
> Fixes: e756af5b30b0 ("brcmfmac: add e-scan support.")
> Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
> Cc: stable@vger.kernel.org
> ---
> v5:
> - replace del_timer_sync with timer_shutdown_sync suggested by
> Arend and Takashi
> v4:
> - rename the subject and add CVE number as Ping-Ke Shih suggested
> v3:
> - rename the subject as Johannes suggested
> v2:
> - fix the error of kernel test bot reported
> ---
>  drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> index 667462369a32..a8723a61c9e4 100644
> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> @@ -8431,6 +8431,8 @@ void brcmf_cfg80211_detach(struct brcmf_cfg80211_info *cfg)
>  	if (!cfg)
>  		return;
>  
> +	timer_shutdown_sync(&cfg->escan_timeout);
> +	cancel_work_sync(&cfg->escan_timeout_work);
>  	brcmf_pno_detach(cfg);
>  	brcmf_btcoex_detach(cfg);
>  	wiphy_unregister(cfg->wiphy);

Has anyone tested this on a real device? As v1 didn't even compile I am
very cautious:

https://patchwork.kernel.org/project/linux-wireless/patch/20231104054709.716585-1-zyytlz.wz@163.com/
  
Zheng Hacker Nov. 6, 2023, 2:44 p.m. UTC | #2
Thanks! I didn't test it for I don't have a device. Very appreciated
if anyone could help with that.

Kalle Valo <kvalo@kernel.org> 于2023年11月6日周一 22:41写道:
>
> Zheng Wang <zyytlz.wz@163.com> writes:
>
> > This is the candidate patch of CVE-2023-47233 :
> > https://nvd.nist.gov/vuln/detail/CVE-2023-47233
> >
> > In brcm80211 driver,it starts with the following invoking chain
> > to start init a timeout worker:
> >
> > ->brcmf_usb_probe
> >   ->brcmf_usb_probe_cb
> >     ->brcmf_attach
> >       ->brcmf_bus_started
> >         ->brcmf_cfg80211_attach
> >           ->wl_init_priv
> >             ->brcmf_init_escan
> >               ->INIT_WORK(&cfg->escan_timeout_work,
> >                 brcmf_cfg80211_escan_timeout_worker);
> >
> > If we disconnect the USB by hotplug, it will call
> > brcmf_usb_disconnect to make cleanup. The invoking chain is :
> >
> > brcmf_usb_disconnect
> >   ->brcmf_usb_disconnect_cb
> >     ->brcmf_detach
> >       ->brcmf_cfg80211_detach
> >         ->kfree(cfg);
> >
> > While the timeout woker may still be running. This will cause
> > a use-after-free bug on cfg in brcmf_cfg80211_escan_timeout_worker.
> >
> > Fix it by deleting the timer and canceling the worker in
> > brcmf_cfg80211_detach.
> >
> > Fixes: e756af5b30b0 ("brcmfmac: add e-scan support.")
> > Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
> > Cc: stable@vger.kernel.org
> > ---
> > v5:
> > - replace del_timer_sync with timer_shutdown_sync suggested by
> > Arend and Takashi
> > v4:
> > - rename the subject and add CVE number as Ping-Ke Shih suggested
> > v3:
> > - rename the subject as Johannes suggested
> > v2:
> > - fix the error of kernel test bot reported
> > ---
> >  drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> > index 667462369a32..a8723a61c9e4 100644
> > --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> > +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> > @@ -8431,6 +8431,8 @@ void brcmf_cfg80211_detach(struct brcmf_cfg80211_info *cfg)
> >       if (!cfg)
> >               return;
> >
> > +     timer_shutdown_sync(&cfg->escan_timeout);
> > +     cancel_work_sync(&cfg->escan_timeout_work);
> >       brcmf_pno_detach(cfg);
> >       brcmf_btcoex_detach(cfg);
> >       wiphy_unregister(cfg->wiphy);
>
> Has anyone tested this on a real device? As v1 didn't even compile I am
> very cautious:
>
> https://patchwork.kernel.org/project/linux-wireless/patch/20231104054709.716585-1-zyytlz.wz@163.com/
>
> --
> https://patchwork.kernel.org/project/linux-wireless/list/
>
> https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
  
Arend van Spriel Nov. 6, 2023, 3:48 p.m. UTC | #3
On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:

> Thanks! I didn't test it for I don't have a device. Very appreciated
> if anyone could help with that.

I would volunteer, but it made me dig deep and not sure if there is a 
problem to solve here.

brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() -> 
brcmf_notify_escan_complete() which does delete the timer.

What am I missing here?

Regards,
Arend

>
> Kalle Valo <kvalo@kernel.org> 于2023年11月6日周一 22:41写道:
>>
>> Zheng Wang <zyytlz.wz@163.com> writes:
>>
>>> This is the candidate patch of CVE-2023-47233 :
>>> https://nvd.nist.gov/vuln/detail/CVE-2023-47233
>>>
>>> In brcm80211 driver,it starts with the following invoking chain
>>> to start init a timeout worker:
>>>
>>> ->brcmf_usb_probe
>>> ->brcmf_usb_probe_cb
>>> ->brcmf_attach
>>> ->brcmf_bus_started
>>>  ->brcmf_cfg80211_attach
>>>    ->wl_init_priv
>>>      ->brcmf_init_escan
>>>        ->INIT_WORK(&cfg->escan_timeout_work,
>>>          brcmf_cfg80211_escan_timeout_worker);
>>>
>>> If we disconnect the USB by hotplug, it will call
>>> brcmf_usb_disconnect to make cleanup. The invoking chain is :
>>>
>>> brcmf_usb_disconnect
>>> ->brcmf_usb_disconnect_cb
>>> ->brcmf_detach
>>> ->brcmf_cfg80211_detach
>>>  ->kfree(cfg);
>>>
>>> While the timeout woker may still be running. This will cause
>>> a use-after-free bug on cfg in brcmf_cfg80211_escan_timeout_worker.
>>>
>>> Fix it by deleting the timer and canceling the worker in
>>> brcmf_cfg80211_detach.
>>>
>>> Fixes: e756af5b30b0 ("brcmfmac: add e-scan support.")
>>> Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
>>> Cc: stable@vger.kernel.org
>>> ---
>>> v5:
>>> - replace del_timer_sync with timer_shutdown_sync suggested by
>>> Arend and Takashi
>>> v4:
>>> - rename the subject and add CVE number as Ping-Ke Shih suggested
>>> v3:
>>> - rename the subject as Johannes suggested
>>> v2:
>>> - fix the error of kernel test bot reported
>>> ---
>>> drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 2 ++
>>> 1 file changed, 2 insertions(+)
>>>
>>> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c 
>>> b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
>>> index 667462369a32..a8723a61c9e4 100644
>>> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
>>> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
>>> @@ -8431,6 +8431,8 @@ void brcmf_cfg80211_detach(struct brcmf_cfg80211_info 
>>> *cfg)
>>> if (!cfg)
>>>        return;
>>>
>>> +     timer_shutdown_sync(&cfg->escan_timeout);
>>> +     cancel_work_sync(&cfg->escan_timeout_work);
>>> brcmf_pno_detach(cfg);
>>> brcmf_btcoex_detach(cfg);
>>> wiphy_unregister(cfg->wiphy);
>>
>> Has anyone tested this on a real device? As v1 didn't even compile I am
>> very cautious:
>>
>> https://patchwork.kernel.org/project/linux-wireless/patch/20231104054709.716585-1-zyytlz.wz@163.com/
>>
>> --
>> https://patchwork.kernel.org/project/linux-wireless/list/
>>
>> https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
  
Zheng Hacker Nov. 8, 2023, 3:03 a.m. UTC | #4
Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
>
> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>
> > Thanks! I didn't test it for I don't have a device. Very appreciated
> > if anyone could help with that.
>
> I would volunteer, but it made me dig deep and not sure if there is a
> problem to solve here.
>
> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
> brcmf_notify_escan_complete() which does delete the timer.
>
> What am I missing here?

Thanks four your detailed review. I did see the code and not sure if
brcmf_notify_escan_complete
would be triggered for sure. So in the first version I want to delete
the pending timer ahead of time.
As I'm not very familiar with the logic here. I'm still not sure if we
should delete the timer_shutdown_sync.
Looking forward to your reply :)

```cpp
if (cfg->int_escan_map || cfg->scan_request) {
    escan->escan_state = WL_ESCAN_STATE_IDLE;
    brcmf_notify_escan_complete(cfg, escan->ifp, true, true);
  }
```

Best regards,
Zheng

>
> Regards,
> Arend
>
> >
> > Kalle Valo <kvalo@kernel.org> 于2023年11月6日周一 22:41写道:
> >>
> >> Zheng Wang <zyytlz.wz@163.com> writes:
> >>
> >>> This is the candidate patch of CVE-2023-47233 :
> >>> https://nvd.nist.gov/vuln/detail/CVE-2023-47233
> >>>
> >>> In brcm80211 driver,it starts with the following invoking chain
> >>> to start init a timeout worker:
> >>>
> >>> ->brcmf_usb_probe
> >>> ->brcmf_usb_probe_cb
> >>> ->brcmf_attach
> >>> ->brcmf_bus_started
> >>>  ->brcmf_cfg80211_attach
> >>>    ->wl_init_priv
> >>>      ->brcmf_init_escan
> >>>        ->INIT_WORK(&cfg->escan_timeout_work,
> >>>          brcmf_cfg80211_escan_timeout_worker);
> >>>
> >>> If we disconnect the USB by hotplug, it will call
> >>> brcmf_usb_disconnect to make cleanup. The invoking chain is :
> >>>
> >>> brcmf_usb_disconnect
> >>> ->brcmf_usb_disconnect_cb
> >>> ->brcmf_detach
> >>> ->brcmf_cfg80211_detach
> >>>  ->kfree(cfg);
> >>>
> >>> While the timeout woker may still be running. This will cause
> >>> a use-after-free bug on cfg in brcmf_cfg80211_escan_timeout_worker.
> >>>
> >>> Fix it by deleting the timer and canceling the worker in
> >>> brcmf_cfg80211_detach.
> >>>
> >>> Fixes: e756af5b30b0 ("brcmfmac: add e-scan support.")
> >>> Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
> >>> Cc: stable@vger.kernel.org
> >>> ---
> >>> v5:
> >>> - replace del_timer_sync with timer_shutdown_sync suggested by
> >>> Arend and Takashi
> >>> v4:
> >>> - rename the subject and add CVE number as Ping-Ke Shih suggested
> >>> v3:
> >>> - rename the subject as Johannes suggested
> >>> v2:
> >>> - fix the error of kernel test bot reported
> >>> ---
> >>> drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 2 ++
> >>> 1 file changed, 2 insertions(+)
> >>>
> >>> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> >>> b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> >>> index 667462369a32..a8723a61c9e4 100644
> >>> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> >>> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> >>> @@ -8431,6 +8431,8 @@ void brcmf_cfg80211_detach(struct brcmf_cfg80211_info
> >>> *cfg)
> >>> if (!cfg)
> >>>        return;
> >>>
> >>> +     timer_shutdown_sync(&cfg->escan_timeout);
> >>> +     cancel_work_sync(&cfg->escan_timeout_work);
> >>> brcmf_pno_detach(cfg);
> >>> brcmf_btcoex_detach(cfg);
> >>> wiphy_unregister(cfg->wiphy);
> >>
> >> Has anyone tested this on a real device? As v1 didn't even compile I am
> >> very cautious:
> >>
> >> https://patchwork.kernel.org/project/linux-wireless/patch/20231104054709.716585-1-zyytlz.wz@163.com/
> >>
> >> --
> >> https://patchwork.kernel.org/project/linux-wireless/list/
> >>
> >> https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
>
>
>
  
Arend van Spriel Nov. 13, 2023, 9:18 a.m. UTC | #5
On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com> 
wrote:

> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
>>
>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>
>>> Thanks! I didn't test it for I don't have a device. Very appreciated
>>> if anyone could help with that.
>>
>> I would volunteer, but it made me dig deep and not sure if there is a
>> problem to solve here.
>>
>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
>> brcmf_notify_escan_complete() which does delete the timer.
>>
>> What am I missing here?
>
> Thanks four your detailed review. I did see the code and not sure if
> brcmf_notify_escan_complete
> would be triggered for sure. So in the first version I want to delete
> the pending timer ahead of time.

Why requesting a CVE when you are not sure? Seems a bit hasty to put it 
mildly.

> As I'm not very familiar with the logic here. I'm still not sure if we
> should delete the timer_shutdown_sync.
> Looking forward to your reply :)

Reading the kerneldoc of timer_shutdown_sync() has the advantage that 
the timer can not be rearmed by another thread. However, that will only 
happen when a new scan is initiated in firmware, but the bus is already 
down so that can not happen. The only improvement (no bug fix!) I see 
here is to replace timer handling code in brcmf_notify_escan_complete():

-	if (timer_pending(&cfg->escan_timeout))
-		del_timer_sync(&cfg->escan_timeout);
+	timer_delete_sync(&cfg->escan_timeout);

Regards,
Arend
  
Kalle Valo Nov. 13, 2023, 12:11 p.m. UTC | #6
Arend van Spriel <arend.vanspriel@broadcom.com> writes:

> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
> wrote:
>
>> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
>>>
>>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>>
>>>> Thanks! I didn't test it for I don't have a device. Very appreciated
>>>> if anyone could help with that.
>>>
>>> I would volunteer, but it made me dig deep and not sure if there is a
>>> problem to solve here.
>>>
>>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
>>> brcmf_notify_escan_complete() which does delete the timer.
>>>
>>> What am I missing here?
>>
>> Thanks four your detailed review. I did see the code and not sure if
>> brcmf_notify_escan_complete
>> would be triggered for sure. So in the first version I want to delete
>> the pending timer ahead of time.
>
> Why requesting a CVE when you are not sure? Seems a bit hasty to put
> it mildly.

TBH I don't take CVE entries seriously anymore. I don't know what has
happened there.
  
Zheng Hacker Nov. 15, 2023, 3 p.m. UTC | #7
Arend van Spriel <arend.vanspriel@broadcom.com> 于2023年11月13日周一 17:18写道:
>
> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
> wrote:
>
> > Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
> >>
> >> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
> >>
> >>> Thanks! I didn't test it for I don't have a device. Very appreciated
> >>> if anyone could help with that.
> >>
> >> I would volunteer, but it made me dig deep and not sure if there is a
> >> problem to solve here.
> >>
> >> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
> >> brcmf_notify_escan_complete() which does delete the timer.
> >>
> >> What am I missing here?
> >
> > Thanks four your detailed review. I did see the code and not sure if
> > brcmf_notify_escan_complete
> > would be triggered for sure. So in the first version I want to delete
> > the pending timer ahead of time.
>
> Why requesting a CVE when you are not sure? Seems a bit hasty to put it
> mildly.

I'm sure the issue exists because there's only cancler of timer but not woker.
As there's similar CVEs before like : https://github.com/V4bel/CVE-2022-41218,
I submit it as soon as I found it.

>
> > As I'm not very familiar with the logic here. I'm still not sure if we
> > should delete the timer_shutdown_sync.
> > Looking forward to your reply :)
>
> Reading the kerneldoc of timer_shutdown_sync() has the advantage that
> the timer can not be rearmed by another thread. However, that will only
> happen when a new scan is initiated in firmware, but the bus is already
> down so that can not happen. The only improvement (no bug fix!) I see
> here is to replace timer handling code in brcmf_notify_escan_complete():
>
> -       if (timer_pending(&cfg->escan_timeout))
> -               del_timer_sync(&cfg->escan_timeout);
> +       timer_delete_sync(&cfg->escan_timeout);
>

Very thanks for your reviews and suggestions! I thinks it's a good
idea. I'll make
another patch sooner or later.

Best regards,
Zheng

> Regards,
> Arend
  
Arend van Spriel Nov. 16, 2023, 6:20 p.m. UTC | #8
On November 15, 2023 4:00:46 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:

> Arend van Spriel <arend.vanspriel@broadcom.com> 于2023年11月13日周一 17:18写道:
>>
>> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
>> wrote:
>>
>>> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
>>>>
>>>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>>>
>>>>> Thanks! I didn't test it for I don't have a device. Very appreciated
>>>>> if anyone could help with that.
>>>>
>>>> I would volunteer, but it made me dig deep and not sure if there is a
>>>> problem to solve here.
>>>>
>>>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
>>>> brcmf_notify_escan_complete() which does delete the timer.
>>>>
>>>> What am I missing here?
>>>
>>> Thanks four your detailed review. I did see the code and not sure if
>>> brcmf_notify_escan_complete
>>> would be triggered for sure. So in the first version I want to delete
>>> the pending timer ahead of time.
>>
>> Why requesting a CVE when you are not sure? Seems a bit hasty to put it
>> mildly.
>
> I'm sure the issue exists because there's only cancler of timer but not woker.
> As there's similar CVEs before like : https://github.com/V4bel/CVE-2022-41218,
> I submit it as soon as I found it.

Ah, yes. The cancel_work_sync() can also be done in 
brcmf_notify_escan_complete().

Regards,
Arend
  
Takashi Iwai Nov. 16, 2023, 6:25 p.m. UTC | #9
On Thu, 16 Nov 2023 19:20:06 +0100,
Arend Van Spriel wrote:
> 
> On November 15, 2023 4:00:46 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
> 
> > Arend van Spriel <arend.vanspriel@broadcom.com> 于2023年11月13日周一 17:18写道:
> >> 
> >> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
> >> wrote:
> >> 
> >>> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
> >>>> 
> >>>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
> >>>> 
> >>>>> Thanks! I didn't test it for I don't have a device. Very appreciated
> >>>>> if anyone could help with that.
> >>>> 
> >>>> I would volunteer, but it made me dig deep and not sure if there is a
> >>>> problem to solve here.
> >>>> 
> >>>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
> >>>> brcmf_notify_escan_complete() which does delete the timer.
> >>>> 
> >>>> What am I missing here?
> >>> 
> >>> Thanks four your detailed review. I did see the code and not sure if
> >>> brcmf_notify_escan_complete
> >>> would be triggered for sure. So in the first version I want to delete
> >>> the pending timer ahead of time.
> >> 
> >> Why requesting a CVE when you are not sure? Seems a bit hasty to put it
> >> mildly.
> > 
> > I'm sure the issue exists because there's only cancler of timer but not woker.
> > As there's similar CVEs before like : https://github.com/V4bel/CVE-2022-41218,
> > I submit it as soon as I found it.
> 
> Ah, yes. The cancel_work_sync() can also be done in
> brcmf_notify_escan_complete().

AFAIUC, brcmf_notify_scan_complete() is called from the work itself,
too, hence you can't issue cancel_work_sync() there (unless you make
it conditional).


Takashi
  
Arend van Spriel Nov. 16, 2023, 7:02 p.m. UTC | #10
On November 16, 2023 7:25:16 PM Takashi Iwai <tiwai@suse.de> wrote:

> On Thu, 16 Nov 2023 19:20:06 +0100,
> Arend Van Spriel wrote:
>>
>> On November 15, 2023 4:00:46 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>
>>> Arend van Spriel <arend.vanspriel@broadcom.com> 于2023年11月13日周一 17:18写道:
>>>>
>>>> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
>>>> wrote:
>>>>
>>>>> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
>>>>>>
>>>>>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>>>>>
>>>>>>> Thanks! I didn't test it for I don't have a device. Very appreciated
>>>>>>> if anyone could help with that.
>>>>>>
>>>>>> I would volunteer, but it made me dig deep and not sure if there is a
>>>>>> problem to solve here.
>>>>>>
>>>>>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
>>>>>> brcmf_notify_escan_complete() which does delete the timer.
>>>>>>
>>>>>> What am I missing here?
>>>>>
>>>>> Thanks four your detailed review. I did see the code and not sure if
>>>>> brcmf_notify_escan_complete
>>>>> would be triggered for sure. So in the first version I want to delete
>>>>> the pending timer ahead of time.
>>>>
>>>> Why requesting a CVE when you are not sure? Seems a bit hasty to put it
>>>> mildly.
>>>
>>> I'm sure the issue exists because there's only cancler of timer but not woker.
>>> As there's similar CVEs before like : https://github.com/V4bel/CVE-2022-41218,
>>> I submit it as soon as I found it.
>>
>> Ah, yes. The cancel_work_sync() can also be done in
>> brcmf_notify_escan_complete().
>
> AFAIUC, brcmf_notify_scan_complete() is called from the work itself,
> too, hence you can't issue cancel_work_sync() there (unless you make
> it conditional).

Hi Takashi,

You are obviously right. Let's wait and see what v6 will look like ;-)

Regards,
Arend
  
Zheng Hacker Nov. 17, 2023, 2:31 a.m. UTC | #11
Yes, that makes this issue hard to fix. I was wondering why it binds the
worker with the timer rather than using just one of them.

Takashi Iwai <tiwai@suse.de> 于2023年11月17日周五 02:25写道:
>
> On Thu, 16 Nov 2023 19:20:06 +0100,
> Arend Van Spriel wrote:
> >
> > On November 15, 2023 4:00:46 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
> >
> > > Arend van Spriel <arend.vanspriel@broadcom.com> 于2023年11月13日周一 17:18写道:
> > >>
> > >> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
> > >> wrote:
> > >>
> > >>> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
> > >>>>
> > >>>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
> > >>>>
> > >>>>> Thanks! I didn't test it for I don't have a device. Very appreciated
> > >>>>> if anyone could help with that.
> > >>>>
> > >>>> I would volunteer, but it made me dig deep and not sure if there is a
> > >>>> problem to solve here.
> > >>>>
> > >>>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
> > >>>> brcmf_notify_escan_complete() which does delete the timer.
> > >>>>
> > >>>> What am I missing here?
> > >>>
> > >>> Thanks four your detailed review. I did see the code and not sure if
> > >>> brcmf_notify_escan_complete
> > >>> would be triggered for sure. So in the first version I want to delete
> > >>> the pending timer ahead of time.
> > >>
> > >> Why requesting a CVE when you are not sure? Seems a bit hasty to put it
> > >> mildly.
> > >
> > > I'm sure the issue exists because there's only cancler of timer but not woker.
> > > As there's similar CVEs before like : https://github.com/V4bel/CVE-2022-41218,
> > > I submit it as soon as I found it.
> >
> > Ah, yes. The cancel_work_sync() can also be done in
> > brcmf_notify_escan_complete().
>
> AFAIUC, brcmf_notify_scan_complete() is called from the work itself,
> too, hence you can't issue cancel_work_sync() there (unless you make
> it conditional).
>
>
> Takashi
  
Arend van Spriel Nov. 17, 2023, 6:24 a.m. UTC | #12
On November 17, 2023 3:31:40 AM Zheng Hacker <hackerzheng666@gmail.com> wrote:

> Yes, that makes this issue hard to fix. I was wondering why it binds the
> worker with the timer rather than using just one of them.

No top posting please!

The timer context is softirq and worker is thread context. The ability to 
sleep is the big difference between the two or at least the reason for 
using them here.

Regards,
Arend

>
> Takashi Iwai <tiwai@suse.de> 于2023年11月17日周五 02:25写道:
>>
>> On Thu, 16 Nov 2023 19:20:06 +0100,
>> Arend Van Spriel wrote:
>>>
>>> On November 15, 2023 4:00:46 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>>
>>>> Arend van Spriel <arend.vanspriel@broadcom.com> 于2023年11月13日周一 17:18写道:
>>>>>
>>>>> On November 8, 2023 4:03:26 AM Zheng Hacker <hackerzheng666@gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Arend Van Spriel <arend.vanspriel@broadcom.com> 于2023年11月6日周一 23:48写道:
>>>>>>>
>>>>>>> On November 6, 2023 3:44:53 PM Zheng Hacker <hackerzheng666@gmail.com> wrote:
>>>>>>>
>>>>>>>> Thanks! I didn't test it for I don't have a device. Very appreciated
>>>>>>>> if anyone could help with that.
>>>>>>>
>>>>>>> I would volunteer, but it made me dig deep and not sure if there is a
>>>>>>> problem to solve here.
>>>>>>>
>>>>>>> brcmf_cfg80211_detach() calls wl_deinit_priv() -> brcmf_abort_scanning() ->
>>>>>>> brcmf_notify_escan_complete() which does delete the timer.
>>>>>>>
>>>>>>> What am I missing here?
>>>>>>
>>>>>> Thanks four your detailed review. I did see the code and not sure if
>>>>>> brcmf_notify_escan_complete
>>>>>> would be triggered for sure. So in the first version I want to delete
>>>>>> the pending timer ahead of time.
>>>>>
>>>>> Why requesting a CVE when you are not sure? Seems a bit hasty to put it
>>>>> mildly.
>>>>
>>>> I'm sure the issue exists because there's only cancler of timer but not woker.
>>>> As there's similar CVEs before like : https://github.com/V4bel/CVE-2022-41218,
>>>> I submit it as soon as I found it.
>>>
>>> Ah, yes. The cancel_work_sync() can also be done in
>>> brcmf_notify_escan_complete().
>>
>> AFAIUC, brcmf_notify_scan_complete() is called from the work itself,
>> too, hence you can't issue cancel_work_sync() there (unless you make
>> it conditional).
>>
>>
>> Takashi
  
Lee Jones Dec. 12, 2023, 1:48 p.m. UTC | #13
On Mon, 06 Nov 2023, Zheng Wang wrote:

> This is the candidate patch of CVE-2023-47233 :
> https://nvd.nist.gov/vuln/detail/CVE-2023-47233
> 
> In brcm80211 driver,it starts with the following invoking chain
> to start init a timeout worker:
> 
> ->brcmf_usb_probe
>   ->brcmf_usb_probe_cb
>     ->brcmf_attach
>       ->brcmf_bus_started
>         ->brcmf_cfg80211_attach
>           ->wl_init_priv
>             ->brcmf_init_escan
>               ->INIT_WORK(&cfg->escan_timeout_work,
> 		  brcmf_cfg80211_escan_timeout_worker);
> 
> If we disconnect the USB by hotplug, it will call
> brcmf_usb_disconnect to make cleanup. The invoking chain is :
> 
> brcmf_usb_disconnect
>   ->brcmf_usb_disconnect_cb
>     ->brcmf_detach
>       ->brcmf_cfg80211_detach
>         ->kfree(cfg);
> 
> While the timeout woker may still be running. This will cause
> a use-after-free bug on cfg in brcmf_cfg80211_escan_timeout_worker.
> 
> Fix it by deleting the timer and canceling the worker in
> brcmf_cfg80211_detach.
> 
> Fixes: e756af5b30b0 ("brcmfmac: add e-scan support.")
> Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
> Cc: stable@vger.kernel.org
> ---
> v5:
> - replace del_timer_sync with timer_shutdown_sync suggested by
> Arend and Takashi
> v4:
> - rename the subject and add CVE number as Ping-Ke Shih suggested
> v3:
> - rename the subject as Johannes suggested
> v2:
> - fix the error of kernel test bot reported
> ---
>  drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> index 667462369a32..a8723a61c9e4 100644
> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> @@ -8431,6 +8431,8 @@ void brcmf_cfg80211_detach(struct brcmf_cfg80211_info *cfg)
>  	if (!cfg)
>  		return;
>  
> +	timer_shutdown_sync(&cfg->escan_timeout);
> +	cancel_work_sync(&cfg->escan_timeout_work);
>  	brcmf_pno_detach(cfg);
>  	brcmf_btcoex_detach(cfg);
>  	wiphy_unregister(cfg->wiphy);

Has there been any progress on this please?

Are we expecting a v6 to this?
  

Patch

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
index 667462369a32..a8723a61c9e4 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
@@ -8431,6 +8431,8 @@  void brcmf_cfg80211_detach(struct brcmf_cfg80211_info *cfg)
 	if (!cfg)
 		return;
 
+	timer_shutdown_sync(&cfg->escan_timeout);
+	cancel_work_sync(&cfg->escan_timeout_work);
 	brcmf_pno_detach(cfg);
 	brcmf_btcoex_detach(cfg);
 	wiphy_unregister(cfg->wiphy);