[PATCH] a multiplication overflow in drivers/pps/pps.c

[PATCH] a multiplication overflow in drivers/pps/pps.c

[xqx12]

there is an overflow in the following code :

ticks = fdata.timeout.sec * HZ;

while ticks is a signed 64-bit, but the result of fdata.timeout.sec *
HZ will be converted be 32-bit first. So ticks will be a wrong value
after multiplication overflow.

Reported-by: Qixue Xiao <xiaoqixue_1@163.com>
Suggested-by: Yongjian Xu <xuyongjiande@gmail.com>
Suggested-by: Yu Chen <chyyuu@gmail.com>
Signed-off-by: Qixue Xiao <xiaoqixue_1@163.com>
---
 drivers/pps/pps.c |    2 +-
 gentags.sh        |    4 +++
 memory_leak.txt   |   88 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 93 insertions(+), 1 deletion(-)
 create mode 100755 gentags.sh
 create mode 100644 memory_leak.txt

diff --git a/drivers/pps/pps.c b/drivers/pps/pps.c
index 2f07cd6..44ddd22 100644
--- a/drivers/pps/pps.c
+++ b/drivers/pps/pps.c
@@ -164,7 +164,7 @@ static long pps_cdev_ioctl(struct file *file,
 			dev_dbg(pps->dev, "timeout %lld.%09d\n",
 					(long long) fdata.timeout.sec,
 					fdata.timeout.nsec);
-			ticks = fdata.timeout.sec * HZ;
+			ticks = (s64)(fdata.timeout.sec) * HZ;
 			ticks += fdata.timeout.nsec / (NSEC_PER_SEC / HZ);
 
 			if (ticks != 0) {
-- 
1.7.9.5

Re: [PATCH] a multiplication overflow in drivers/pps/pps.c

[Andrew Morton]

On Wed, 18 Dec 2013 15:36:11 +0800 xqx12 <s2exqx@gmail.com> wrote:

> there is an overflow in the following code :
> 
> ticks = fdata.timeout.sec * HZ;
> 
> while ticks is a signed 64-bit, but the result of fdata.timeout.sec *
> HZ will be converted be 32-bit first. So ticks will be a wrong value
> after multiplication overflow.

When fixing a bug, please always include a full description of the
user-visible effects of that bug.  This helps others decide which
kernel version(s) should be fixed.

> ...
>
> --- a/drivers/pps/pps.c
> +++ b/drivers/pps/pps.c
> @@ -164,7 +164,7 @@ static long pps_cdev_ioctl(struct file *file,
>  			dev_dbg(pps->dev, "timeout %lld.%09d\n",
>  					(long long) fdata.timeout.sec,
>  					fdata.timeout.nsec);
> -			ticks = fdata.timeout.sec * HZ;
> +			ticks = (s64)(fdata.timeout.sec) * HZ;
>  			ticks += fdata.timeout.nsec / (NSEC_PER_SEC / HZ);
>  
>  			if (ticks != 0) {

pps_fdata.timeout.sec has type __s64, so the patch should be a no-op?

Did you really observe a runtime problem from this?  If so, I suspect
your compiler is busted.