From: Arnd Bergmann <[email protected]>
A previous patch addressed the fortified memcpy warning for most
builds, but I still see this one with gcc-9:
In file included from include/linux/string.h:254,
from drivers/hid/hid-hyperv.c:8:
In function 'fortify_memcpy_chk',
inlined from 'mousevsc_on_receive' at drivers/hid/hid-hyperv.c:272:3:
include/linux/fortify-string.h:583:4: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror=attribute-warning]
583 | __write_overflow_field(p_size_field, size);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
My guess is that the WARN_ON() itself is what confuses gcc, so it no
longer sees that there is a correct range check. Rework the code in a
way that helps readability and avoids the warning.
Fixes: 542f25a944715 ("HID: hyperv: Replace one-element array with flexible-array member")
Signed-off-by: Arnd Bergmann <[email protected]>
---
drivers/hid/hid-hyperv.c | 10 ++++------
1 file changed, 4 insertions(+), 6 deletions(-)
diff --git a/drivers/hid/hid-hyperv.c b/drivers/hid/hid-hyperv.c
index 49d4a26895e76..f33485d83d24f 100644
--- a/drivers/hid/hid-hyperv.c
+++ b/drivers/hid/hid-hyperv.c
@@ -258,19 +258,17 @@ static void mousevsc_on_receive(struct hv_device *device,
switch (hid_msg_hdr->type) {
case SYNTH_HID_PROTOCOL_RESPONSE:
+ len = struct_size(pipe_msg, data, pipe_msg->size);
+
/*
* While it will be impossible for us to protect against
* malicious/buggy hypervisor/host, add a check here to
* ensure we don't corrupt memory.
*/
- if (struct_size(pipe_msg, data, pipe_msg->size)
- > sizeof(struct mousevsc_prt_msg)) {
- WARN_ON(1);
+ if (WARN_ON(len > sizeof(struct mousevsc_prt_msg)))
break;
- }
- memcpy(&input_dev->protocol_resp, pipe_msg,
- struct_size(pipe_msg, data, pipe_msg->size));
+ memcpy(&input_dev->protocol_resp, pipe_msg, len);
complete(&input_dev->wait_event);
break;
--
2.39.2
From: Arnd Bergmann <[email protected]> Sent: Wednesday, July 5, 2023 7:02 AM
>
> A previous patch addressed the fortified memcpy warning for most
> builds, but I still see this one with gcc-9:
>
> In file included from include/linux/string.h:254,
> from drivers/hid/hid-hyperv.c:8:
> In function 'fortify_memcpy_chk',
> inlined from 'mousevsc_on_receive' at drivers/hid/hid-hyperv.c:272:3:
> include/linux/fortify-string.h:583:4: error: call to '__write_overflow_field' declared with
> attribute warning: detected write beyond size of field (1st parameter); maybe use
> struct_group()? [-Werror=attribute-warning]
> 583 | __write_overflow_field(p_size_field, size);
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> My guess is that the WARN_ON() itself is what confuses gcc, so it no
> longer sees that there is a correct range check. Rework the code in a
> way that helps readability and avoids the warning.
>
> Fixes: 542f25a944715 ("HID: hyperv: Replace one-element array with flexible-array member")
> Signed-off-by: Arnd Bergmann <[email protected]>
> ---
> drivers/hid/hid-hyperv.c | 10 ++++------
> 1 file changed, 4 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/hid/hid-hyperv.c b/drivers/hid/hid-hyperv.c
> index 49d4a26895e76..f33485d83d24f 100644
> --- a/drivers/hid/hid-hyperv.c
> +++ b/drivers/hid/hid-hyperv.c
> @@ -258,19 +258,17 @@ static void mousevsc_on_receive(struct hv_device *device,
>
> switch (hid_msg_hdr->type) {
> case SYNTH_HID_PROTOCOL_RESPONSE:
> + len = struct_size(pipe_msg, data, pipe_msg->size);
> +
> /*
> * While it will be impossible for us to protect against
> * malicious/buggy hypervisor/host, add a check here to
> * ensure we don't corrupt memory.
> */
> - if (struct_size(pipe_msg, data, pipe_msg->size)
> - > sizeof(struct mousevsc_prt_msg)) {
> - WARN_ON(1);
> + if (WARN_ON(len > sizeof(struct mousevsc_prt_msg)))
> break;
> - }
>
> - memcpy(&input_dev->protocol_resp, pipe_msg,
> - struct_size(pipe_msg, data, pipe_msg->size));
> + memcpy(&input_dev->protocol_resp, pipe_msg, len);
> complete(&input_dev->wait_event);
> break;
>
> --
> 2.39.2
Reviewed-by: Michael Kelley <[email protected]>
On Wed, 05 Jul 2023 16:02:24 +0200, Arnd Bergmann wrote:
> A previous patch addressed the fortified memcpy warning for most
> builds, but I still see this one with gcc-9:
>
> In file included from include/linux/string.h:254,
> from drivers/hid/hid-hyperv.c:8:
> In function 'fortify_memcpy_chk',
> inlined from 'mousevsc_on_receive' at drivers/hid/hid-hyperv.c:272:3:
> include/linux/fortify-string.h:583:4: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror=attribute-warning]
> 583 | __write_overflow_field(p_size_field, size);
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> [...]
Applied to https://git.kernel.org/pub/scm/linux/kernel/git/hid/hid.git (for-6.5/upstream-fixes), thanks!
[1/1] HID: hyperv: avoid struct memcpy overrun warning
https://git.kernel.org/hid/hid/c/5f151364b1da
Cheers,
--
Benjamin Tissoires <[email protected]>
On Wed, Jul 05, 2023 at 04:02:24PM +0200, Arnd Bergmann wrote:
> From: Arnd Bergmann <[email protected]>
>
> A previous patch addressed the fortified memcpy warning for most
> builds, but I still see this one with gcc-9:
JFYI: as of today I have run Linux Next with `make W=1 allmodconfig` on x86_64
and it seems there are still tons of similar issues which break the build.
--
With Best Regards,
Andy Shevchenko
On Mon, Jul 17, 2023, at 11:36, Andy Shevchenko wrote:
> On Wed, Jul 05, 2023 at 04:02:24PM +0200, Arnd Bergmann wrote:
>> From: Arnd Bergmann <[email protected]>
>>
>> A previous patch addressed the fortified memcpy warning for most
>> builds, but I still see this one with gcc-9:
>
> JFYI: as of today I have run Linux Next with `make W=1 allmodconfig` on x86_64
> and it seems there are still tons of similar issues which break the build.
It's a bit more complex:
- yes, there are lots of warnings for memcpy() read overflow when you
build allmodconfig kernels with W=1. I have patches for all of these and
plan to submit them over time.
- This particular one is a memcpy /write/ overflow, i.e. the
destination type overflows with the length according to gcc's
understanding of the structures. This warning is enabled even
without W=1, though it may only show up in certain configurations
or compiler versions.
Arnd