get_registers() copies whatever memory is written by the
usb_control_msg() call even if the underlying urb call ends up failing.
If get_registers() fails, or ends up reading 0 bytes, meaningless and
junk register values would end up being copied over (and eventually read
by the driver), and since most of the callers of get_registers() don't
check the return values of get_registers() either, this would go unnoticed.
It might be a better idea to try and mirror the PCI master abort
termination and set memory to 0xFFs instead in such cases.
Fixes: https://syzkaller.appspot.com/bug?extid=abbc768b560c84d92fd3
Reported-by: [email protected]
Tested-by: [email protected]
Signed-off-by: Anant Thazhemadam <[email protected]>
---
drivers/net/usb/rtl8150.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/net/usb/rtl8150.c b/drivers/net/usb/rtl8150.c
index 733f120c852b..04fca7bfcbcb 100644
--- a/drivers/net/usb/rtl8150.c
+++ b/drivers/net/usb/rtl8150.c
@@ -162,8 +162,13 @@ static int get_registers(rtl8150_t * dev, u16 indx, u16 size, void *data)
ret = usb_control_msg(dev->udev, usb_rcvctrlpipe(dev->udev, 0),
RTL8150_REQ_GET_REGS, RTL8150_REQT_READ,
indx, 0, buf, size, 500);
- if (ret > 0 && ret <= size)
+
+ if (ret < 0)
+ memset(data, 0xff, size);
+
+ else
memcpy(data, buf, ret);
+
kfree(buf);
return ret;
}
@@ -276,7 +281,7 @@ static int write_mii_word(rtl8150_t * dev, u8 phy, __u8 indx, u16 reg)
static inline void set_ethernet_addr(rtl8150_t * dev)
{
- u8 node_id[6];
+ u8 node_id[6] = {0};
get_registers(dev, IDR, sizeof(node_id), node_id);
memcpy(dev->netdev->dev_addr, node_id, sizeof(node_id));
--
2.25.1
On Wed, Sep 16, 2020 at 10:35:40AM +0530, Anant Thazhemadam wrote:
> get_registers() copies whatever memory is written by the
> usb_control_msg() call even if the underlying urb call ends up failing.
>
> If get_registers() fails, or ends up reading 0 bytes, meaningless and
> junk register values would end up being copied over (and eventually read
> by the driver), and since most of the callers of get_registers() don't
> check the return values of get_registers() either, this would go unnoticed.
>
> It might be a better idea to try and mirror the PCI master abort
> termination and set memory to 0xFFs instead in such cases.
It would be better to use this new api call instead of
usb_control_msg():
https://lore.kernel.org/r/[email protected]
How about porting this patch to run on top of that series instead? That
should make this logic much simpler.
>
> Fixes: https://syzkaller.appspot.com/bug?extid=abbc768b560c84d92fd3
> Reported-by: [email protected]
> Tested-by: [email protected]
> Signed-off-by: Anant Thazhemadam <[email protected]>
> ---
> drivers/net/usb/rtl8150.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/usb/rtl8150.c b/drivers/net/usb/rtl8150.c
> index 733f120c852b..04fca7bfcbcb 100644
> --- a/drivers/net/usb/rtl8150.c
> +++ b/drivers/net/usb/rtl8150.c
> @@ -162,8 +162,13 @@ static int get_registers(rtl8150_t * dev, u16 indx, u16 size, void *data)
> ret = usb_control_msg(dev->udev, usb_rcvctrlpipe(dev->udev, 0),
> RTL8150_REQ_GET_REGS, RTL8150_REQT_READ,
> indx, 0, buf, size, 500);
> - if (ret > 0 && ret <= size)
> +
> + if (ret < 0)
> + memset(data, 0xff, size);
> +
> + else
> memcpy(data, buf, ret);
> +
> kfree(buf);
> return ret;
> }
> @@ -276,7 +281,7 @@ static int write_mii_word(rtl8150_t * dev, u8 phy, __u8 indx, u16 reg)
>
> static inline void set_ethernet_addr(rtl8150_t * dev)
> {
> - u8 node_id[6];
> + u8 node_id[6] = {0};
This should not be needed to be done.
thanks,
greg k-h
On 20-09-16 08:22:27, Greg KH wrote:
> On Wed, Sep 16, 2020 at 10:35:40AM +0530, Anant Thazhemadam wrote:
> > get_registers() copies whatever memory is written by the
> > usb_control_msg() call even if the underlying urb call ends up failing.
> >
> > If get_registers() fails, or ends up reading 0 bytes, meaningless and
> > junk register values would end up being copied over (and eventually read
> > by the driver), and since most of the callers of get_registers() don't
> > check the return values of get_registers() either, this would go unnoticed.
> >
> > It might be a better idea to try and mirror the PCI master abort
> > termination and set memory to 0xFFs instead in such cases.
>
> It would be better to use this new api call instead of
> usb_control_msg():
> https://lore.kernel.org/r/[email protected]
Heh, wasn't aware of the new api.
> How about porting this patch to run on top of that series instead? That
> should make this logic much simpler.
I'll need to check if in this case 'size' is the right amount of bytes expected
and not an upper limit. Then i'll convert it to the new api.
cheers,
Petko
> > Fixes: https://syzkaller.appspot.com/bug?extid=abbc768b560c84d92fd3
> > Reported-by: [email protected]
> > Tested-by: [email protected]
> > Signed-off-by: Anant Thazhemadam <[email protected]>
> > ---
> > drivers/net/usb/rtl8150.c | 9 +++++++--
> > 1 file changed, 7 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/usb/rtl8150.c b/drivers/net/usb/rtl8150.c
> > index 733f120c852b..04fca7bfcbcb 100644
> > --- a/drivers/net/usb/rtl8150.c
> > +++ b/drivers/net/usb/rtl8150.c
> > @@ -162,8 +162,13 @@ static int get_registers(rtl8150_t * dev, u16 indx, u16 size, void *data)
> > ret = usb_control_msg(dev->udev, usb_rcvctrlpipe(dev->udev, 0),
> > RTL8150_REQ_GET_REGS, RTL8150_REQT_READ,
> > indx, 0, buf, size, 500);
> > - if (ret > 0 && ret <= size)
> > +
> > + if (ret < 0)
> > + memset(data, 0xff, size);
> > +
> > + else
> > memcpy(data, buf, ret);
> > +
> > kfree(buf);
> > return ret;
> > }
> > @@ -276,7 +281,7 @@ static int write_mii_word(rtl8150_t * dev, u8 phy, __u8 indx, u16 reg)
> >
> > static inline void set_ethernet_addr(rtl8150_t * dev)
> > {
> > - u8 node_id[6];
> > + u8 node_id[6] = {0};
>
> This should not be needed to be done.
>
> thanks,
>
> greg k-h
>
On 20-09-16 10:35:40, Anant Thazhemadam wrote:
> get_registers() copies whatever memory is written by the
> usb_control_msg() call even if the underlying urb call ends up failing.
Not true, memcpy() is only called if "ret" is positive.
> If get_registers() fails, or ends up reading 0 bytes, meaningless and junk
> register values would end up being copied over (and eventually read by the
> driver), and since most of the callers of get_registers() don't check the
> return values of get_registers() either, this would go unnoticed.
usb_control_msg() returns negative on error (look up usb_internal_control_msg()
to see for yourself) so it does not go unnoticed. If for some reason it return
zero, nothing is copied. Also, if usb transfer fail no register values are
being copied anywhere.
Your patch also allows for memcpy() to be called with 'size' either zero or
greater than the allocated buffer size. Please, look at the code carefully.
> It might be a better idea to try and mirror the PCI master abort
> termination and set memory to 0xFFs instead in such cases.
I wasn't aware drivers are now responsible for filling up the memory with
anything. Does not sound like a good idea to me.
> Fixes: https://syzkaller.appspot.com/bug?extid=abbc768b560c84d92fd3
> Reported-by: [email protected]
> Tested-by: [email protected]
> Signed-off-by: Anant Thazhemadam <[email protected]>
Well, NACK from me.
cheers,
Petko
> ---
> drivers/net/usb/rtl8150.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/usb/rtl8150.c b/drivers/net/usb/rtl8150.c
> index 733f120c852b..04fca7bfcbcb 100644
> --- a/drivers/net/usb/rtl8150.c
> +++ b/drivers/net/usb/rtl8150.c
> @@ -162,8 +162,13 @@ static int get_registers(rtl8150_t * dev, u16 indx, u16 size, void *data)
> ret = usb_control_msg(dev->udev, usb_rcvctrlpipe(dev->udev, 0),
> RTL8150_REQ_GET_REGS, RTL8150_REQT_READ,
> indx, 0, buf, size, 500);
> - if (ret > 0 && ret <= size)
> +
> + if (ret < 0)
> + memset(data, 0xff, size);
> +
> + else
> memcpy(data, buf, ret);
> +
> kfree(buf);
> return ret;
> }
> @@ -276,7 +281,7 @@ static int write_mii_word(rtl8150_t * dev, u8 phy, __u8 indx, u16 reg)
>
> static inline void set_ethernet_addr(rtl8150_t * dev)
> {
> - u8 node_id[6];
> + u8 node_id[6] = {0};
>
> get_registers(dev, IDR, sizeof(node_id), node_id);
> memcpy(dev->netdev->dev_addr, node_id, sizeof(node_id));
> --
> 2.25.1
>
>
On 16/09/20 11:49 am, Petko Manolov wrote:
> On 20-09-16 10:35:40, Anant Thazhemadam wrote:
>> get_registers() copies whatever memory is written by the
>> usb_control_msg() call even if the underlying urb call ends up failing.
> Not true, memcpy() is only called if "ret" is positive.
Right. I'm really sorry I fumbled and messed up the commit message
there. Thank you for pointing that out.
>> If get_registers() fails, or ends up reading 0 bytes, meaningless and junk
>> register values would end up being copied over (and eventually read by the
>> driver), and since most of the callers of get_registers() don't check the
>> return values of get_registers() either, this would go unnoticed.
> usb_control_msg() returns negative on error (look up usb_internal_control_msg()
> to see for yourself) so it does not go unnoticed.
When I said "this would go unnoticed", I meant get_register() failing would
go unnoticed, not that usb_control_msg() failing would go unnoticed.
I agree that get_registers() notices usb_control_msg() failing, and
appropriately returns the return value from usb_control_msg().
But there are many instances where get_registers() is called but the return
value of get_registers() is not checked, to see if it failed or not; hence, "this
would go unnoticed".
> If for some reason it return zero, nothing is copied. Also, if usb transfer fail
> no register values are being copied anywhere.
True.
Now consider set_ethernet_addr(), and suppose get_register() fails when
invoked from inside set_ethernet_addr().
As you said, no value is copied back, which means no value is copied back
into node_id, which leaves node_id uninitialized. This node_id (still
uninitialized) is then blindly copied into dev->netdev->dev_addr; which
is less than ideal and could also quickly prove to become an issue, right?
> Your patch also allows for memcpy() to be called with 'size' either zero or
> greater than the allocated buffer size. Please, look at the code carefully.
Oh. I apologize for this. This can be reverted relatively easily.
>> It might be a better idea to try and mirror the PCI master abort
>> termination and set memory to 0xFFs instead in such cases.
> I wasn't aware drivers are now responsible for filling up the memory with
> anything. Does not sound like a good idea to me.
Since we copy the correct register values when get_register() doesn't fail,
I thought it might be a slightly better alternative to fill node_id with 0xFFs,
instead of leaving it go uninitialized in case get_registers() fails.
Also, what are the odds that a successful get_register() call would see
0xFFs being copied?
If that's very real scenario, then I admit this doesn't work at all.
The only other alternative approach I can think of that can handle the
issue I highlighted above, is to introduce checking for get_registers()'s
return values nearly everywhere it gets called.
Would that be a more preferable and welcome approach?
Thank you for your time.
Thanks,
Anant
On 16/09/20 11:52 am, Greg KH wrote:
> On Wed, Sep 16, 2020 at 10:35:40AM +0530, Anant Thazhemadam wrote:
>> get_registers() copies whatever memory is written by the
>> usb_control_msg() call even if the underlying urb call ends up failing.
>>
>> If get_registers() fails, or ends up reading 0 bytes, meaningless and
>> junk register values would end up being copied over (and eventually read
>> by the driver), and since most of the callers of get_registers() don't
>> check the return values of get_registers() either, this would go unnoticed.
>>
>> It might be a better idea to try and mirror the PCI master abort
>> termination and set memory to 0xFFs instead in such cases.
> It would be better to use this new api call instead of
> usb_control_msg():
> https://lore.kernel.org/r/[email protected]
>
> How about porting this patch to run on top of that series instead? That
> should make this logic much simpler.
This looks viable to me. I'll be sure to try this out.
>> Fixes: https://syzkaller.appspot.com/bug?extid=abbc768b560c84d92fd3
>> Reported-by: [email protected]
>> Tested-by: [email protected]
>> Signed-off-by: Anant Thazhemadam <[email protected]>
>> ---
>> drivers/net/usb/rtl8150.c | 9 +++++++--
>> 1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/net/usb/rtl8150.c b/drivers/net/usb/rtl8150.c
>> index 733f120c852b..04fca7bfcbcb 100644
>> --- a/drivers/net/usb/rtl8150.c
>> +++ b/drivers/net/usb/rtl8150.c
>> @@ -162,8 +162,13 @@ static int get_registers(rtl8150_t * dev, u16 indx, u16 size, void *data)
>> ret = usb_control_msg(dev->udev, usb_rcvctrlpipe(dev->udev, 0),
>> RTL8150_REQ_GET_REGS, RTL8150_REQT_READ,
>> indx, 0, buf, size, 500);
>> - if (ret > 0 && ret <= size)
>> +
>> + if (ret < 0)
>> + memset(data, 0xff, size);
>> +
>> + else
>> memcpy(data, buf, ret);
>> +
>> kfree(buf);
>> return ret;
>> }
>> @@ -276,7 +281,7 @@ static int write_mii_word(rtl8150_t * dev, u8 phy, __u8 indx, u16 reg)
>>
>> static inline void set_ethernet_addr(rtl8150_t * dev)
>> {
>> - u8 node_id[6];
>> + u8 node_id[6] = {0};
> This should not be needed to be done.
Noted.
Thank you for your time.
Thanks,
Anant