From: Dongliang Mu <mudongliangabcd@gmail.com>

The error handling code of peak_usb_create_dev forgets to reset the
next_siblings of previous entry.

Fix this by nullifying the (dev->prev_siblings)->next_siblings in the
error handling code.

Signed-off-by: Dongliang Mu <mudongliangabcd@gmail.com>
---
 drivers/net/can/usb/peak_usb/pcan_usb_core.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/net/can/usb/peak_usb/pcan_usb_core.c b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
index b850ff8fe4bd..f858810221b6 100644
--- a/drivers/net/can/usb/peak_usb/pcan_usb_core.c
+++ b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
@@ -894,6 +894,9 @@ static int peak_usb_create_dev(const struct peak_usb_adapter *peak_usb_adapter,
 		dev->adapter->dev_free(dev);
 
 lbl_unregister_candev:
+	/* remove the dangling pointer in next_siblings */
+	if (dev->prev_siblings)
+		(dev->prev_siblings)->next_siblings = NULL;
 	unregister_candev(netdev);
 
 lbl_restore_intf_data:
-- 
2.25.1

Hi Dongliang,

On 1/20/22 16:05, Dongliang Mu wrote:
> From: Dongliang Mu <mudongliangabcd@gmail.com>
> 
> The error handling code of peak_usb_create_dev forgets to reset the
> next_siblings of previous entry.
> 
> Fix this by nullifying the (dev->prev_siblings)->next_siblings in the
> error handling code.
> 
> Signed-off-by: Dongliang Mu <mudongliangabcd@gmail.com>
> ---
>   drivers/net/can/usb/peak_usb/pcan_usb_core.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/net/can/usb/peak_usb/pcan_usb_core.c b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> index b850ff8fe4bd..f858810221b6 100644
> --- a/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> +++ b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> @@ -894,6 +894,9 @@ static int peak_usb_create_dev(const struct peak_usb_adapter *peak_usb_adapter,
>   		dev->adapter->dev_free(dev);
>   
>   lbl_unregister_candev:
> +	/* remove the dangling pointer in next_siblings */
> +	if (dev->prev_siblings)
> +		(dev->prev_siblings)->next_siblings = NULL;
>   	unregister_candev(netdev);
>   
>   lbl_restore_intf_data:


Is this pointer used somewhere? I see, that couple of
struct peak_usb_adapter::dev_free() functions use it, but
peak_usb_disconnect() sets dev->next_siblings to NULL before calling 
->dev_free().

Do you have a calltrace or oops log?




With regards,
Pavel Skripkin

On Thu, Jan 20, 2022 at 10:27 PM Pavel Skripkin <paskripkin@gmail.com> wrote:
>
> Hi Dongliang,
>
> On 1/20/22 16:05, Dongliang Mu wrote:
> > From: Dongliang Mu <mudongliangabcd@gmail.com>
> >
> > The error handling code of peak_usb_create_dev forgets to reset the
> > next_siblings of previous entry.
> >
> > Fix this by nullifying the (dev->prev_siblings)->next_siblings in the
> > error handling code.
> >
> > Signed-off-by: Dongliang Mu <mudongliangabcd@gmail.com>
> > ---
> >   drivers/net/can/usb/peak_usb/pcan_usb_core.c | 3 +++
> >   1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/net/can/usb/peak_usb/pcan_usb_core.c b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > index b850ff8fe4bd..f858810221b6 100644
> > --- a/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > +++ b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > @@ -894,6 +894,9 @@ static int peak_usb_create_dev(const struct peak_usb_adapter *peak_usb_adapter,
> >               dev->adapter->dev_free(dev);
> >
> >   lbl_unregister_candev:
> > +     /* remove the dangling pointer in next_siblings */
> > +     if (dev->prev_siblings)
> > +             (dev->prev_siblings)->next_siblings = NULL;
> >       unregister_candev(netdev);
> >
> >   lbl_restore_intf_data:
>
>
> Is this pointer used somewhere? I see, that couple of
> struct peak_usb_adapter::dev_free() functions use it, but
> peak_usb_disconnect() sets dev->next_siblings to NULL before calling
> ->dev_free().
>
> Do you have a calltrace or oops log?

Hi Pavel,

I have no calltrace or log since this dangling pointer may not be
dereferenced in the following code. But I am not sure. So the commit
title of this patch is "remove a dangling pointer in
peak_usb_create_dev".

>
>
>
>
> With regards,
> Pavel Skripkin

On Fri, Jan 21, 2022 at 8:09 AM Dongliang Mu <mudongliangabcd@gmail.com> wrote:
>
> On Thu, Jan 20, 2022 at 10:27 PM Pavel Skripkin <paskripkin@gmail.com> wrote:
> >
> > Hi Dongliang,
> >
> > On 1/20/22 16:05, Dongliang Mu wrote:
> > > From: Dongliang Mu <mudongliangabcd@gmail.com>
> > >
> > > The error handling code of peak_usb_create_dev forgets to reset the
> > > next_siblings of previous entry.
> > >
> > > Fix this by nullifying the (dev->prev_siblings)->next_siblings in the
> > > error handling code.
> > >
> > > Signed-off-by: Dongliang Mu <mudongliangabcd@gmail.com>
> > > ---
> > >   drivers/net/can/usb/peak_usb/pcan_usb_core.c | 3 +++
> > >   1 file changed, 3 insertions(+)
> > >
> > > diff --git a/drivers/net/can/usb/peak_usb/pcan_usb_core.c b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > > index b850ff8fe4bd..f858810221b6 100644
> > > --- a/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > > +++ b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > > @@ -894,6 +894,9 @@ static int peak_usb_create_dev(const struct peak_usb_adapter *peak_usb_adapter,
> > >               dev->adapter->dev_free(dev);
> > >
> > >   lbl_unregister_candev:
> > > +     /* remove the dangling pointer in next_siblings */
> > > +     if (dev->prev_siblings)
> > > +             (dev->prev_siblings)->next_siblings = NULL;
> > >       unregister_candev(netdev);
> > >
> > >   lbl_restore_intf_data:
> >
> >
> > Is this pointer used somewhere? I see, that couple of
> > struct peak_usb_adapter::dev_free() functions use it, but
> > peak_usb_disconnect() sets dev->next_siblings to NULL before calling
> > ->dev_free().
> >
> > Do you have a calltrace or oops log?
>
> Hi Pavel,
>
> I have no calltrace or log since this dangling pointer may not be
> dereferenced in the following code. But I am not sure. So the commit
> title of this patch is "remove a dangling pointer in
> peak_usb_create_dev".

BTW, as you mentioned, dev->next_siblings is used in struct
peak_usb_adapter::dev_free() (i.e., pcan_usb_fd_free or
pcan_usb_pro_free), how about the following path?

peak_usb_probe
-> peak_usb_create_dev (goto adap_dev_free;)
   -> dev->adapter->dev_free()
      -> pcan_usb_fd_free or pcan_usb_pro_free (This function uses
next_siblings as condition elements)

static void pcan_usb_fd_free(struct peak_usb_device *dev)
{
        /* last device: can free shared objects now */
        if (!dev->prev_siblings && !dev->next_siblings) {
                struct pcan_usb_fd_device *pdev =
                        container_of(dev, struct pcan_usb_fd_device, dev);

                /* free commands buffer */
                kfree(pdev->cmd_buffer_addr);

                /* free usb interface object */
                kfree(pdev->usb_if);
        }
}

If next_siblings is not NULL, will it lead to the missing free of
cmd_buffer_addr and usb_if?

Please let me know if I made any mistakes.

> >
> >
> >
> >
> > With regards,
> > Pavel Skripkin

On Fri, Jan 21, 2022 at 11:36 AM Dongliang Mu <mudongliangabcd@gmail.com> wrote:
>
> On Fri, Jan 21, 2022 at 8:09 AM Dongliang Mu <mudongliangabcd@gmail.com> wrote:
> >
> > On Thu, Jan 20, 2022 at 10:27 PM Pavel Skripkin <paskripkin@gmail.com> wrote:
> > >
> > > Hi Dongliang,
> > >
> > > On 1/20/22 16:05, Dongliang Mu wrote:
> > > > From: Dongliang Mu <mudongliangabcd@gmail.com>
> > > >
> > > > The error handling code of peak_usb_create_dev forgets to reset the
> > > > next_siblings of previous entry.
> > > >
> > > > Fix this by nullifying the (dev->prev_siblings)->next_siblings in the
> > > > error handling code.
> > > >
> > > > Signed-off-by: Dongliang Mu <mudongliangabcd@gmail.com>
> > > > ---
> > > >   drivers/net/can/usb/peak_usb/pcan_usb_core.c | 3 +++
> > > >   1 file changed, 3 insertions(+)
> > > >
> > > > diff --git a/drivers/net/can/usb/peak_usb/pcan_usb_core.c b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > > > index b850ff8fe4bd..f858810221b6 100644
> > > > --- a/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > > > +++ b/drivers/net/can/usb/peak_usb/pcan_usb_core.c
> > > > @@ -894,6 +894,9 @@ static int peak_usb_create_dev(const struct peak_usb_adapter *peak_usb_adapter,
> > > >               dev->adapter->dev_free(dev);
> > > >
> > > >   lbl_unregister_candev:
> > > > +     /* remove the dangling pointer in next_siblings */
> > > > +     if (dev->prev_siblings)
> > > > +             (dev->prev_siblings)->next_siblings = NULL;
> > > >       unregister_candev(netdev);
> > > >
> > > >   lbl_restore_intf_data:
> > >
> > >
> > > Is this pointer used somewhere? I see, that couple of
> > > struct peak_usb_adapter::dev_free() functions use it, but
> > > peak_usb_disconnect() sets dev->next_siblings to NULL before calling
> > > ->dev_free().
> > >
> > > Do you have a calltrace or oops log?
> >
> > Hi Pavel,
> >
> > I have no calltrace or log since this dangling pointer may not be
> > dereferenced in the following code. But I am not sure. So the commit
> > title of this patch is "remove a dangling pointer in
> > peak_usb_create_dev".
>
> BTW, as you mentioned, dev->next_siblings is used in struct
> peak_usb_adapter::dev_free() (i.e., pcan_usb_fd_free or
> pcan_usb_pro_free), how about the following path?
>
> peak_usb_probe
> -> peak_usb_create_dev (goto adap_dev_free;)
>    -> dev->adapter->dev_free()
>       -> pcan_usb_fd_free or pcan_usb_pro_free (This function uses
> next_siblings as condition elements)
>
> static void pcan_usb_fd_free(struct peak_usb_device *dev)
> {
>         /* last device: can free shared objects now */
>         if (!dev->prev_siblings && !dev->next_siblings) {
>                 struct pcan_usb_fd_device *pdev =
>                         container_of(dev, struct pcan_usb_fd_device, dev);
>
>                 /* free commands buffer */
>                 kfree(pdev->cmd_buffer_addr);
>
>                 /* free usb interface object */
>                 kfree(pdev->usb_if);
>         }
> }
>
> If next_siblings is not NULL, will it lead to the missing free of
> cmd_buffer_addr and usb_if?

The answer is No. Forget my silly thought.

>
> Please let me know if I made any mistakes.
>
> > >
> > >
> > >
> > >
> > > With regards,
> > > Pavel Skripkin

Hi Dongliang,

On 1/21/22 08:58, Dongliang Mu wrote:
[...]>> BTW, as you mentioned, dev->next_siblings is used in struct
>> peak_usb_adapter::dev_free() (i.e., pcan_usb_fd_free or
>> pcan_usb_pro_free), how about the following path?
>>
>> peak_usb_probe
>> -> peak_usb_create_dev (goto adap_dev_free;)
>>    -> dev->adapter->dev_free()
>>       -> pcan_usb_fd_free or pcan_usb_pro_free (This function uses
>> next_siblings as condition elements)
>>
>> static void pcan_usb_fd_free(struct peak_usb_device *dev)
>> {
>>         /* last device: can free shared objects now */
>>         if (!dev->prev_siblings && !dev->next_siblings) {
>>                 struct pcan_usb_fd_device *pdev =
>>                         container_of(dev, struct pcan_usb_fd_device, dev);
>>
>>                 /* free commands buffer */
>>                 kfree(pdev->cmd_buffer_addr);
>>
>>                 /* free usb interface object */
>>                 kfree(pdev->usb_if);
>>         }
>> }
>>
>> If next_siblings is not NULL, will it lead to the missing free of
>> cmd_buffer_addr and usb_if?
> 
> The answer is No. Forget my silly thought.
> 

Yeah, it seems like (at least based on code), that this dangling pointer 
is not dangerous, since nothing accesses it. And next_siblings 
_guaranteed_ to be NULL, since dev->next_siblings is set NULL in 
disconnect()




With regards,
Pavel Skripkin

On Sat, Jan 22, 2022 at 3:36 AM Pavel Skripkin <paskripkin@gmail.com> wrote:
>
> Hi Dongliang,
>
> On 1/21/22 08:58, Dongliang Mu wrote:
> [...]>> BTW, as you mentioned, dev->next_siblings is used in struct
> >> peak_usb_adapter::dev_free() (i.e., pcan_usb_fd_free or
> >> pcan_usb_pro_free), how about the following path?
> >>
> >> peak_usb_probe
> >> -> peak_usb_create_dev (goto adap_dev_free;)
> >>    -> dev->adapter->dev_free()
> >>       -> pcan_usb_fd_free or pcan_usb_pro_free (This function uses
> >> next_siblings as condition elements)
> >>
> >> static void pcan_usb_fd_free(struct peak_usb_device *dev)
> >> {
> >>         /* last device: can free shared objects now */
> >>         if (!dev->prev_siblings && !dev->next_siblings) {
> >>                 struct pcan_usb_fd_device *pdev =
> >>                         container_of(dev, struct pcan_usb_fd_device, dev);
> >>
> >>                 /* free commands buffer */
> >>                 kfree(pdev->cmd_buffer_addr);
> >>
> >>                 /* free usb interface object */
> >>                 kfree(pdev->usb_if);
> >>         }
> >> }
> >>
> >> If next_siblings is not NULL, will it lead to the missing free of
> >> cmd_buffer_addr and usb_if?
> >
> > The answer is No. Forget my silly thought.
> >
>
> Yeah, it seems like (at least based on code), that this dangling pointer
> is not dangerous, since nothing accesses it. And next_siblings
> _guaranteed_ to be NULL, since dev->next_siblings is set NULL in
> disconnect()

Yes, you're right. As a security researcher, I am sensitive to such
dangling pointers.

As its nullifying site is across functions, I suggest developers
remove this dangling pointer in case that any newly added code in this
function or before the nullifying location would touch next_siblings.

If Pavel and others think it's fine, then it's time to close this patch.

>
>
>
>
> With regards,
> Pavel Skripkin

Hi Dongliang,

On 1/22/22 09:45, Dongliang Mu wrote:
[...]

>> Yeah, it seems like (at least based on code), that this dangling pointer
>> is not dangerous, since nothing accesses it. And next_siblings
>> _guaranteed_ to be NULL, since dev->next_siblings is set NULL in
>> disconnect()
> 
> Yes, you're right. As a security researcher, I am sensitive to such
> dangling pointers.
> 
> As its nullifying site is across functions, I suggest developers
> remove this dangling pointer in case that any newly added code in this
> function or before the nullifying location would touch next_siblings.
> 

Based on git blame this driver is very old (was added in 2012), so, I 
guess, nothing really new will come up.

Anyway, I am absolutely not a security person and if you think, that 
this dangling pointer can be somehow used in exploitation you should 
state it in commit message.


> If Pavel and others think it's fine, then it's time to close this patch.
> 

I don't have any big objections on the code itself. Maybe only 'if' can 
be removed to just speed up the code, but I don't see why this change is 
needed :)




With regards,
Pavel Skripkin

On Sun, Jan 23, 2022 at 9:48 PM Pavel Skripkin <paskripkin@gmail.com> wrote:
>
> Hi Dongliang,
>
> On 1/22/22 09:45, Dongliang Mu wrote:
> [...]
>
> >> Yeah, it seems like (at least based on code), that this dangling pointer
> >> is not dangerous, since nothing accesses it. And next_siblings
> >> _guaranteed_ to be NULL, since dev->next_siblings is set NULL in
> >> disconnect()
> >
> > Yes, you're right. As a security researcher, I am sensitive to such
> > dangling pointers.
> >
> > As its nullifying site is across functions, I suggest developers
> > remove this dangling pointer in case that any newly added code in this
> > function or before the nullifying location would touch next_siblings.
> >
>
> Based on git blame this driver is very old (was added in 2012), so, I
> guess, nothing really new will come up.
>
> Anyway, I am absolutely not a security person and if you think, that
> this dangling pointer can be somehow used in exploitation you should
> state it in commit message.
>
>
> > If Pavel and others think it's fine, then it's time to close this patch.
> >
>
> I don't have any big objections on the code itself. Maybe only 'if' can
> be removed to just speed up the code, but I don't see why this change is
> needed :)

OK, let's move on. Leave alone this patch.

>
>
>
>
> With regards,
> Pavel Skripkin