Re: [PATCH 17/23] y2038: time: avoid timespec usage in settimeofday()

On Fri, Nov 15, 2019 at 11:27 AM Rasmus Villemoes
[off-list ref] wrote:

On 15/11/2019 08.58, Arnd Bergmann wrote:

quoted

On Fri, Nov 15, 2019 at 12:01 AM Abel Vesa [off-list ref] wrote:

quoted

--- a/kernel/time/time.c
+++ b/kernel/time/time.c

@@ -207,7 +207,7 @@ SYSCALL_DEFINE2(settimeofday, struct

__kernel_old_timeval __user *, tv,
                    get_user(new_ts.tv_nsec, &tv->tv_usec))
                        return -EFAULT;

-               if (tv->tv_usec > USEC_PER_SEC)
+               if (new_ts->tv_usec > USEC_PER_SEC)
                        return -EINVAL;

Hopefully not :)

No, I misquoted from a fix that I had temporarily applied, not the
version in linux-next.

quoted

                new_ts.tv_nsec *= NSEC_PER_USEC;

So the actual patch in next-20191115 does

-               if (copy_from_user(&user_tv, tv, sizeof(*tv)))
+               if (get_user(new_ts.tv_sec, &tv->tv_sec) ||
+                   get_user(new_ts.tv_nsec, &tv->tv_usec))
                        return -EFAULT;

-               if (!timeval_valid(&user_tv))
+               if (new_ts.tv_nsec > USEC_PER_SEC)
                        return -EINVAL;

-               new_ts.tv_sec = user_tv.tv_sec;
-               new_ts.tv_nsec = user_tv.tv_usec * NSEC_PER_USEC;
+               new_ts.tv_nsec *= NSEC_PER_USEC;

But removing the "user value is < 0" check, relying on the timespec
value being rejected later, is wrong

You are right of course, so many ways to get this one line wrong...
Pushed more more update to the branch now.

Thanks for the careful review!

        Arnd