[LTP] [PATCH] syscalls: clock_settime: Add test around y2038 vulnerability

[Arnd Bergmann]

On Tue, May 26, 2020 at 2:10 PM Viresh Kumar <viresh.kumar@linaro.org> wrote:
>
> This adds a test around the y2038 vulnerability, it sets the system time
> to just before y2038 time (i.e. max value that can be stored in s32),
> and adds a timer to expire just after crossing it.
>
> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>

Nice!
> +static struct test_variants {
> +       int (*clock_settime)(clockid_t clk_id, void *ts);
> +       int (*timer_settime)(timer_t timerid, int flags, void *its,
> +                            void *old_its);
> +       enum tst_ts_type type;
> +       char *desc;
> +} variants[] = {
> +#if (__NR_clock_settime != __LTP__NR_INVALID_SYSCALL)
> +       { .clock_settime = sys_clock_settime, .timer_settime = sys_timer_settime, .type = TST_KERN_OLD_TIMESPEC, .desc = "syscall with old kernel spec"},
> +#endif
> +
> +#if (__NR_clock_settime64 != __LTP__NR_INVALID_SYSCALL)
> +       { .clock_settime = sys_clock_settime64, .timer_settime = sys_timer_settime64, .type = TST_KERN_TIMESPEC, .desc = "syscall time64 with kernel spec"},
> +#endif

I think the first one has to be guarded so we do not try to set
the time to just before the end with sys_clock_settime() on
32-bit machines, as kernels that don't fully support 64-bit
time_t behave in unpredictable ways when you do that and
are likely to crash.

However, we probably do want to test this on 64-bit kernels
with sys_clock_settime() anyway.

> +       tst_ts_set_sec(&ts, time);
> +       tst_ts_set_nsec(&ts, 0);
> +
> +       ret = tv->clock_settime(CLOCK_REALTIME, tst_ts_get(&ts));
> +       if (ret == -1)
> +               tst_brk(TBROK | TERRNO, "clock_settime() failed");
> +
> +       tst_its_set_time(&its, time + EXPIREDELTA, 0, 0, 0);

I suspect this is where it fails for the 32-bit time_t case, as the expiration
date for timer_settime wraps to year 1902.

     Arnd