[0/4] USB: sisusbvga: cleaning up char buffers to u8 buffer

Message ID 1593200057-245-1-git-send-email-charley.ashbringer@gmail.com
Headers show
Series
  • USB: sisusbvga: cleaning up char buffers to u8 buffer
Related show

Message

Changming June 26, 2020, 7:34 p.m.
The purpose of this patch series is to prevent
undefined behavior caused by arithemetic operations on
the content of the char buffer.This is achieved through
converting all char buffers to u8 when there is 
no need for the content to be negative value(which is the 
case for all touched buffer in this patch series).

Converting these char buffers to u8 buffers will make 
certain types of UB from producing undefined result to 
well-defined intended (modular) result, and will not 
introduce any unintended side effect.

As an example, changing the local char buf[4] in sisusb_write_mem_bulk
to u8 fixes an UB: since buf is filled with data from user space,
thus can be negative. Left-shifting this negative value produces
undefined result, this can be fixed by changing it from char 
to u8.

This however doesn't apply to ioctl interfaces functions, since the 
types for buffer of ioctl-like functions need to be 
char* instead of u8* for compatibility reason.
In these cases, the buffer declared as char* will be force cast
to u8* when being used.


Changming Liu (4):
  USB: sisusbvga: change sisusb_write_mem_bulk
  USB: sisusbvga: change the buffers of sisusb from char to u8
  USB: sisusbvga: change userbuffer for sisusb_recv_bulk_msg to u8
  USB: sisusbvga: change sisusb_read_mem_bulk

 drivers/usb/misc/sisusbvga/sisusb.c | 34 +++++++++++++++++-----------------
 drivers/usb/misc/sisusbvga/sisusb.h |  4 ++--
 2 files changed, 19 insertions(+), 19 deletions(-)

Comments

Changming July 10, 2020, 5:23 a.m. | #1
> -----Original Message-----

> From: Greg KH <gregkh@linuxfoundation.org>

> Sent: Saturday, June 27, 2020 7:28 AM

> To: Changming Liu <charley.ashbringer@gmail.com>

> Cc: linux-usb@vger.kernel.org; thomas@winischhofer.net

> Subject: Re: [PATCH 1/4] USB: sisusbvga: change the char buffer from char

to
> u8 for sisusb_write_mem_bulk and sisusb_send_bulk_msg

> 

> On Fri, Jun 26, 2020 at 03:34:14PM -0400, Changming Liu wrote:

> > This patch changes the types of char buffer declarations

> > as well as passed-in parameters to u8 for the function

> > sisusb_write_mem_bulk and sisusb_send_bulk_msg to aviod

> > any related UB.

> >

> > This patch also change the local buf[4] of sisusb_write_mem_bulk

> > to u8. This fixed an undefined behavior, since buf can be filled

> > with data from user space, thus can be negative given it's signed,

> > and its content is being left-shifted. Left-shifting a negative

> > value is undefined behavior. It's fixed by changing the buf from

> > char to u8.

> 

> In looking at this closer, it doesn't make sense to change the function

> parameters here, as everything that deals with the pointer already

> handles the change properly.

> 


Quite,  no security issue could possibly be raised without 
these unnecessary changes. 

> 

> >

> > Signed-off-by: Changming Liu <charley.ashbringer@gmail.com>

> > ---

> >  drivers/usb/misc/sisusbvga/sisusb.c | 14 +++++++-------

> >  1 file changed, 7 insertions(+), 7 deletions(-)

> >

> > diff --git a/drivers/usb/misc/sisusbvga/sisusb.c

> b/drivers/usb/misc/sisusbvga/sisusb.c

> > index fc8a5da..4aa717a 100644

> > --- a/drivers/usb/misc/sisusbvga/sisusb.c

> > +++ b/drivers/usb/misc/sisusbvga/sisusb.c

> > @@ -327,7 +327,7 @@ static int sisusb_bulkin_msg(struct sisusb_usb_data

> *sisusb,

> >   */

> >

> >  static int sisusb_send_bulk_msg(struct sisusb_usb_data *sisusb, int ep,

int
> len,

> > -		char *kernbuffer, const char __user *userbuffer, int index,

> > +		u8 *kernbuffer, const u8 __user *userbuffer, int index,

> 

> So the kernbuffer pointer might want to be changed, but in looking at

> the code, there's no difference here, it can be left alone.

> 

> The userbuffer does not need to be changed at all.

> 

> >  static int sisusb_write_mem_bulk(struct sisusb_usb_data *sisusb, u32

addr,
> > -		char *kernbuffer, int length, const char __user *userbuffer,

> > +		u8 *kernbuffer, int length, const u8 __user *userbuffer,

> 

> Same here, these do not need to be changed.


Totally agree.

> 

> >  		int index, ssize_t *bytes_written)

> >  {

> >  	struct sisusb_packet packet;

> > @@ -761,7 +761,7 @@ static int sisusb_write_mem_bulk(struct

> sisusb_usb_data *sisusb, u32 addr,

> >  	u8   swap8, fromkern = kernbuffer ? 1 : 0;

> >  	u16  swap16;

> >  	u32  swap32, flag = (length >> 28) & 1;

> > -	char buf[4];

> > +	u8 buf[4];

> 

> That is what should be changed, and in looking at the code path, I think

> that's it here.

> 

> Sorry for taking you down the wrong path, but I think you should only



It's totally fine, I took this chance to thoroughly read the code 
and learned a lot about how a typical linux driver is written : p


> change things that actually matter, and the above api changes don't

> change anything at all, right?


Yes, this is exactly what I felt when I was compiling the chances.
I really don't see necessity in the changes except the one
that has security implication.

Thanks for the feedback, these back-and-forth deepen my understanding
both of the kernel and how to submit patch.

Sorry for this late reply, I have been catching a deadline for the 
past several days :( I'll submit another patch about the change with
security implication shortly.

Best regards,
Changming