On Thu, Nov 21, 2019 at 2:39 PM Nicolas Saenz Julienne
<nsaenzjulienne@xxxxxxx> wrote:
>
> Hi Marcelo,
>
> On Wed, 2019-11-20 at 15:20 -0500, Marcelo Diop-Gonzalez wrote:
> > This is a proposed fix of an issue regarding the handling of positive
> > return values from copy_to_user() in vchiq_read(). It looks like the
> > struct dump_context passed to the vchiq_dump() function keeps track
> > of the number of bytes written in the context->actual field. When
> > copy_to_user() returns a positive value, this is set to -EFAULT. The
> > problem is that this is never returned to the user, and instead the
> > function continues, adding the number of bytes that should have
> > been copied to context->actual.
> >
> > Running the following program:
> >
> > #include <stdio.h>
> > #include <stdlib.h>
> > #include <fcntl.h>
> > #include <unistd.h>
> > #include <errno.h>
> >
> > int main(int argc, char **argv) {
> > int fd = open("/dev/vchiq", 0);
> > if (fd < 0)
> > exit(-1);
> > sbrk(getpagesize());
> > char *bad_buf = sbrk(0)-100;
> > int x = read(fd, bad_buf, 2000);
> > printf("%d %d\n", x, errno);
> > puts(bad_buf);
> > }
> >
> > gives this output:
> >
> > -1 1
> > State 0: CONNECTED
> > tx_po
> >
> >
> > Remote: slots 2-32 tx_pos=578 recycle=1f
> > Slots claimed:
> >
> > Note the mangled output and incorrect errno value. Messing with the
> > constants in that toy program changes the results. Sometimes read()
> > returns with no error, sometimes it returns with a wrong error code,
> > sometimes with the right one. But it seems that it only ever returns an
> > error at all accidentally, due to the fact that the comparison between
> > context->actual and context->space in vchiq_dump() is unsigned, so that
> > that function won't do anything else if it ever sets context->actual
> > to a negative value.
> >
> > After this patchset, the above program prints this:
> >
> > -1 14
> > State 0: CONNECTED
> > tx_pos=b4a218(@165de6b4), rx_pos=ae0668(@f02b54f4)
> > Version: 8 (min 3)
> > Stats
> >
> > Help with testing would be appreciated. So far I've basically just
> > diffed the output of 'cat /dev/vchiq', run the program above with
> > a few different values, and run vchiq_test a few times.
> >
> > These were applied to the staging-next branch of the tree
> > at git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git
>
> For the whole series:
>
> Reviewed-by: Nicolas Saenz Julienne <nsaenzjulienne@xxxxxxx>
Thanks for reviewing!
-Marcelo
>
> Thanks,
> Nicolas
>
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
