[PATCH] HID: wacom: fix shift OOB in kfifo allocation for zero pktlen

[PATCH] HID: wacom: fix shift OOB in kfifo allocation for zero pktlen

[Qasim Ijaz]

During wacom_parse_and_register() the code calls wacom_devm_kfifo_alloc 
to allocate a fifo. During this operation it passes kfifo_alloc a 
fifo_size of 0. Kfifo attempts to round the size passed to it to the 
next power of 2 via roundup_pow_of_two (queue-type data structures
do this to maintain efficiency of operations). 

However during this phase a problem arises when the roundup_pow_of_two() 
function utilises a shift exponent of fls_long(n-1), where n is the 
fifo_size. Since n is 0 in this case and n is also an unsigned long, 
doing n-1 causes unsigned integer wrap-around to occur making the 
fifo_size 4294967295. So the code effectively does fls_long(4294967295) 
which results in 64. Returning back to roundup_pow_of_two(), the code 
utilises a shift exponent of 64. When a shift exponent of 64 is used 
on a 64-bit type such as 1UL it results in a shift-out-of-bounds.

The root cause of the issue seems to stem from insufficient validation 
of wacom_compute_pktlen(), since in this case the fifo_size comes 
from wacom_wac->features.pktlen. During wacom_parse_and_register() 
the wacom_compute_pktlen() function sets the pktlen as 0.

To fix this, we should handle cases where wacom_compute_pktlen() 
results in 0.

Reported-by: syzbot <syzbot+d5204cbbdd921f1f7cad@syzkaller.appspotmail.com>
Closes: https://syzkaller.appspot.com/bug?extid=d5204cbbdd921f1f7cad
Tested-by: Qasim Ijaz <qasdev00@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Qasim Ijaz <qasdev00@gmail.com>
---
 drivers/hid/wacom_sys.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/hid/wacom_sys.c b/drivers/hid/wacom_sys.c
index 97393a3083ca..9b2f3dbca467 100644
--- a/drivers/hid/wacom_sys.c
+++ b/drivers/hid/wacom_sys.c
@@ -2361,6 +2361,8 @@ static int wacom_parse_and_register(struct wacom *wacom, bool wireless)
 	unsigned int connect_mask = HID_CONNECT_HIDRAW;
 
 	features->pktlen = wacom_compute_pktlen(hdev);
+	if (!features->pktlen)
+		return -ENODEV;
 
 	if (!devres_open_group(&hdev->dev, wacom, GFP_KERNEL))
 		return -ENOMEM;
-- 
2.39.5

Re: [PATCH] HID: wacom: fix shift OOB in kfifo allocation for zero pktlen

[Gerecke, Jason]

On Tue, April 1, 2025 at 5:59 AM Qasim Ijaz <qasdev00@gmail.com> wrote:
> During wacom_parse_and_register() the code calls wacom_devm_kfifo_alloc 
> to allocate a fifo. During this operation it passes kfifo_alloc a 
> fifo_size of 0. Kfifo attempts to round the size passed to it to the 
> next power of 2 via roundup_pow_of_two (queue-type data structures
> do this to maintain efficiency of operations). 
> 
> However during this phase a problem arises when the roundup_pow_of_two() 
> function utilises a shift exponent of fls_long(n-1), where n is the 
> fifo_size. Since n is 0 in this case and n is also an unsigned long, 
> doing n-1 causes unsigned integer wrap-around to occur making the 
> fifo_size 4294967295. So the code effectively does fls_long(4294967295) 
> which results in 64. Returning back to roundup_pow_of_two(), the code 
> utilises a shift exponent of 64. When a shift exponent of 64 is used 
> on a 64-bit type such as 1UL it results in a shift-out-of-bounds.
> 
> The root cause of the issue seems to stem from insufficient validation 
> of wacom_compute_pktlen(), since in this case the fifo_size comes 
> from wacom_wac->features.pktlen. During wacom_parse_and_register() 
> the wacom_compute_pktlen() function sets the pktlen as 0.
> 
> To fix this, we should handle cases where wacom_compute_pktlen() 
> results in 0.
> 
> Reported-by: syzbot <syzbot+d5204cbbdd921f1f7cad@syzkaller.appspotmail.com>
> Closes: https://syzkaller.appspot.com/bug?extid=d5204cbbdd921f1f7cad
> Tested-by: Qasim Ijaz <qasdev00@gmail.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Qasim Ijaz <qasdev00@gmail.com>
> ---
>  drivers/hid/wacom_sys.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/hid/wacom_sys.c b/drivers/hid/wacom_sys.c
> index 97393a3083ca..9b2f3dbca467 100644
> --- a/drivers/hid/wacom_sys.c
> +++ b/drivers/hid/wacom_sys.c
> @@ -2361,6 +2361,8 @@ static int wacom_parse_and_register(struct wacom *wacom, bool wireless)
>  	unsigned int connect_mask = HID_CONNECT_HIDRAW;
>  
>  	features->pktlen = wacom_compute_pktlen(hdev);
> +	if (!features->pktlen)
> +		return -ENODEV;
>  
>  	if (!devres_open_group(&hdev->dev, wacom, GFP_KERNEL))
>  		return -ENOMEM;
> -- 
> 2.39.5
> 

This looks acceptable to me. Taking a closer look at this area, I have a feeling that 'wacom_compute_pktlen' should also consider feature reports in its calculation (so that we allocate enough space for e.g. GET_FEATURE responses). I'll look into this more over the next few days.

Please add the following to the end of the commit description for traceability:

> Fixes: 5e013ad20689 ("HID: wacom: Remove static WACOM_PKGLEN_MAX limit")

With that modification in place:

Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>

--------
Jason Gerecke (she / they), Senior Linux Software Engineer
Wacom Technology Corporation
tel: 503-525-3100 ext. 3229 (direct)
http://www.wacom.com