| Message ID | 20220419041742.4117026-1-keescook@chromium.org (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | USB: serial: Fix heap overflow in WHITEHEAT_GET_DTR_RTS | expand |
On Mon, Apr 18, 2022 at 09:17:42PM -0700, Kees Cook wrote: > This looks like it's harmless, as both the source and the destinations are > currently the same allocation size (4 bytes) and don't use their padding, > but if anything were to ever be added after the "mcr" member in "struct > whiteheat_private", it would be overwritten. The structs both have a > single u8 "mcr" member, but are 4 bytes in padded size. The memcpy() > destination was explicitly targeting the u8 member (size 1) with the > length of the whole structure (size 4), triggering the memcpy buffer > overflow warning: Ehh... No. The size of a structure with a single u8 is 1, not 4. There's nothing wrong with the current code even if the use of memcpy for this is a bit odd. > In file included from include/linux/string.h:253, > from include/linux/bitmap.h:11, > from include/linux/cpumask.h:12, > from include/linux/smp.h:13, > from include/linux/lockdep.h:14, > from include/linux/spinlock.h:62, > from include/linux/mmzone.h:8, > from include/linux/gfp.h:6, > from include/linux/slab.h:15, > from drivers/usb/serial/whiteheat.c:17: > In function 'fortify_memcpy_chk', > inlined from 'firm_send_command' at drivers/usb/serial/whiteheat.c:587:4: > include/linux/fortify-string.h:328:25: warning: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Wattribute-warning] > 328 | __write_overflow_field(p_size_field, size); > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ So something is confused here. > Expand the memcpy() to the entire structure, though perhaps the correct > solution is to mark all the USB command structures as "__packed". Again no, why would you potentially overwrite the whole structure just to update a single field? This is just wrong. And the only structure that needs a __packed which doesn't have it already is the unused struct whiteheat_dump. > Reported-by: kernel test robot <lkp@intel.com> > Link: https://lore.kernel.org/lkml/202204142318.vDqjjSFn-lkp@intel.com > Cc: Johan Hovold <johan@kernel.org> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> > Cc: linux-usb@vger.kernel.org > Cc: stable@vger.kernel.org > Signed-off-by: Kees Cook <keescook@chromium.org> > --- > drivers/usb/serial/whiteheat.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/usb/serial/whiteheat.c b/drivers/usb/serial/whiteheat.c > index da65d14c9ed5..6e00498843fb 100644 > --- a/drivers/usb/serial/whiteheat.c > +++ b/drivers/usb/serial/whiteheat.c > @@ -584,7 +584,7 @@ static int firm_send_command(struct usb_serial_port *port, __u8 command, > switch (command) { > case WHITEHEAT_GET_DTR_RTS: > info = usb_get_serial_port_data(port); > - memcpy(&info->mcr, command_info->result_buffer, > + memcpy(info, command_info->result_buffer, > sizeof(struct whiteheat_dr_info)); > break; > } Johan
On Wed, Apr 20, 2022 at 10:14 AM Johan Hovold <johan@kernel.org> wrote: > On Mon, Apr 18, 2022 at 09:17:42PM -0700, Kees Cook wrote: > > This looks like it's harmless, as both the source and the destinations are > > currently the same allocation size (4 bytes) and don't use their padding, > > but if anything were to ever be added after the "mcr" member in "struct > > whiteheat_private", it would be overwritten. The structs both have a > > single u8 "mcr" member, but are 4 bytes in padded size. The memcpy() > > destination was explicitly targeting the u8 member (size 1) with the > > length of the whole structure (size 4), triggering the memcpy buffer > > overflow warning: > > Ehh... No. The size of a structure with a single u8 is 1, not 4. There's > nothing wrong with the current code even if the use of memcpy for this > is a bit odd. > > > In file included from include/linux/string.h:253, > > from include/linux/bitmap.h:11, > > from include/linux/cpumask.h:12, > > from include/linux/smp.h:13, > > from include/linux/lockdep.h:14, > > from include/linux/spinlock.h:62, > > from include/linux/mmzone.h:8, > > from include/linux/gfp.h:6, > > from include/linux/slab.h:15, > > from drivers/usb/serial/whiteheat.c:17: > > In function 'fortify_memcpy_chk', > > inlined from 'firm_send_command' at drivers/usb/serial/whiteheat.c:587:4: > > include/linux/fortify-string.h:328:25: warning: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Wattribute-warning] > > 328 | __write_overflow_field(p_size_field, size); > > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > So something is confused here. So something's going wrong in fortify_memcpy_chk()? It looks like it is called with constant "size" equal to 1, and the condition "p_size_field < size" (with an unsigned comparison) is either true (meaning p_size_field would have to be 0) or not known at compile time? The original report says it happened when compiling with CONFIG_CC_OPTIMIZE_FOR_SIZE=y, maybe that matters?
On Wed, Apr 20, 2022 at 02:33:06PM +0200, Jann Horn wrote: > On Wed, Apr 20, 2022 at 10:14 AM Johan Hovold <johan@kernel.org> wrote: > > On Mon, Apr 18, 2022 at 09:17:42PM -0700, Kees Cook wrote: > > > This looks like it's harmless, as both the source and the destinations are > > > currently the same allocation size (4 bytes) and don't use their padding, > > > but if anything were to ever be added after the "mcr" member in "struct > > > whiteheat_private", it would be overwritten. The structs both have a > > > single u8 "mcr" member, but are 4 bytes in padded size. The memcpy() > > > destination was explicitly targeting the u8 member (size 1) with the > > > length of the whole structure (size 4), triggering the memcpy buffer > > > overflow warning: > > > > Ehh... No. The size of a structure with a single u8 is 1, not 4. There's > > nothing wrong with the current code even if the use of memcpy for this > > is a bit odd. I thought that was surprising too, and suspected it was something specific to the build (as Jann also suggested). I tracked it down[1] to "-mabi=apcs-gnu", which is from CONFIG_AEABI=n. whiteheat_private { __u8 mcr; /* 0 1 */ /* size: 4, cachelines: 1, members: 1 */ /* padding: 3 */ /* last cacheline: 4 bytes */ }; > > > > > In file included from include/linux/string.h:253, > > > from include/linux/bitmap.h:11, > > > from include/linux/cpumask.h:12, > > > from include/linux/smp.h:13, > > > from include/linux/lockdep.h:14, > > > from include/linux/spinlock.h:62, > > > from include/linux/mmzone.h:8, > > > from include/linux/gfp.h:6, > > > from include/linux/slab.h:15, > > > from drivers/usb/serial/whiteheat.c:17: > > > In function 'fortify_memcpy_chk', > > > inlined from 'firm_send_command' at drivers/usb/serial/whiteheat.c:587:4: > > > include/linux/fortify-string.h:328:25: warning: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Wattribute-warning] > > > 328 | __write_overflow_field(p_size_field, size); > > > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > > > So something is confused here. > > So something's going wrong in fortify_memcpy_chk()? It looks like it > is called with constant "size" equal to 1, and the condition sizeof(info->mcr) is 1. sizeof(struct whiteheat_dr_info) (with CONFIG_AEABI=n) is 4. Given nothing actually uses "struct whiteheat_dr_info", except for the sizing (har har), I suspect the better solution is just to do: info->mcr = command_info->result_buffer[0]; -Kees [1] https://godbolt.org/z/3YvM1MYWW
[ +CC: Arnd ] On Wed, Apr 20, 2022 at 11:11:26AM -0700, Kees Cook wrote: > On Wed, Apr 20, 2022 at 02:33:06PM +0200, Jann Horn wrote: > > On Wed, Apr 20, 2022 at 10:14 AM Johan Hovold <johan@kernel.org> wrote: > > > On Mon, Apr 18, 2022 at 09:17:42PM -0700, Kees Cook wrote: > > > > This looks like it's harmless, as both the source and the destinations are > > > > currently the same allocation size (4 bytes) and don't use their padding, > > > > but if anything were to ever be added after the "mcr" member in "struct > > > > whiteheat_private", it would be overwritten. The structs both have a > > > > single u8 "mcr" member, but are 4 bytes in padded size. The memcpy() > > > > destination was explicitly targeting the u8 member (size 1) with the > > > > length of the whole structure (size 4), triggering the memcpy buffer > > > > overflow warning: > > > > > > Ehh... No. The size of a structure with a single u8 is 1, not 4. There's > > > nothing wrong with the current code even if the use of memcpy for this > > > is a bit odd. > > I thought that was surprising too, and suspected it was something > specific to the build (as Jann also suggested). I tracked it down[1] to > "-mabi=apcs-gnu", which is from CONFIG_AEABI=n. > > whiteheat_private { > __u8 mcr; /* 0 1 */ > > /* size: 4, cachelines: 1, members: 1 */ > /* padding: 3 */ > /* last cacheline: 4 bytes */ > }; I stand corrected, thanks. Do we have other ABIs that can increase the alignment of structures like this? Skimming lore reveals a few subsystems that have started depending on !OABI to not have to deal with this. Apparently the old ARM ABI is deprecated in user space since gcc-4.6: https://lore.kernel.org/all/20190304193723.657089-1-arnd@arndb.de/ Perhaps time to drop support from the kernel too? > Given nothing actually uses "struct whiteheat_dr_info", except for the > sizing (har har), I suspect the better solution is just to do: > > info->mcr = command_info->result_buffer[0]; Yeah, that works for now. Ideally, we'd cast the result buffer to struct whiteheat_dr_info and access its single field. But that's not what current code does and the above is no less confusing. Patch applied, thanks. Johan
diff --git a/drivers/usb/serial/whiteheat.c b/drivers/usb/serial/whiteheat.c index da65d14c9ed5..6e00498843fb 100644 --- a/drivers/usb/serial/whiteheat.c +++ b/drivers/usb/serial/whiteheat.c @@ -584,7 +584,7 @@ static int firm_send_command(struct usb_serial_port *port, __u8 command, switch (command) { case WHITEHEAT_GET_DTR_RTS: info = usb_get_serial_port_data(port); - memcpy(&info->mcr, command_info->result_buffer, + memcpy(info, command_info->result_buffer, sizeof(struct whiteheat_dr_info)); break; }
This looks like it's harmless, as both the source and the destinations are currently the same allocation size (4 bytes) and don't use their padding, but if anything were to ever be added after the "mcr" member in "struct whiteheat_private", it would be overwritten. The structs both have a single u8 "mcr" member, but are 4 bytes in padded size. The memcpy() destination was explicitly targeting the u8 member (size 1) with the length of the whole structure (size 4), triggering the memcpy buffer overflow warning: In file included from include/linux/string.h:253, from include/linux/bitmap.h:11, from include/linux/cpumask.h:12, from include/linux/smp.h:13, from include/linux/lockdep.h:14, from include/linux/spinlock.h:62, from include/linux/mmzone.h:8, from include/linux/gfp.h:6, from include/linux/slab.h:15, from drivers/usb/serial/whiteheat.c:17: In function 'fortify_memcpy_chk', inlined from 'firm_send_command' at drivers/usb/serial/whiteheat.c:587:4: include/linux/fortify-string.h:328:25: warning: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Wattribute-warning] 328 | __write_overflow_field(p_size_field, size); | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Expand the memcpy() to the entire structure, though perhaps the correct solution is to mark all the USB command structures as "__packed". Reported-by: kernel test robot <lkp@intel.com> Link: https://lore.kernel.org/lkml/202204142318.vDqjjSFn-lkp@intel.com Cc: Johan Hovold <johan@kernel.org> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: linux-usb@vger.kernel.org Cc: stable@vger.kernel.org Signed-off-by: Kees Cook <keescook@chromium.org> --- drivers/usb/serial/whiteheat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)