2019-07-08 10:33:36

by zhangxiaoxu (A)

Subject: [PATCH] time: Validate the usec before covert to nsec in do_adjtimex

When covert the usec to nsec, it will multiple 1000, it maybe
overflow and lead an undefined behavior.

For example, users may input an negative tv_usec values when
call adjtimex syscall, then multiple 1000 maybe overflow it
to a positive and legal number.

So, we should validate the usec before coverted it to nsec.

Signed-off-by: ZhangXiaoxu <[email protected]>
---
kernel/time/timekeeping.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 44b726b..e5c1d00 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -1272,9 +1272,6 @@ static int timekeeping_inject_offset(const struct timespec64 *ts)
struct timespec64 tmp;
int ret = 0;

- if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC)
- return -EINVAL;
-
raw_spin_lock_irqsave(&timekeeper_lock, flags);
write_seqcount_begin(&tk_core.seq);

@@ -2321,6 +2318,9 @@ int do_adjtimex(struct __kernel_timex *txc)

if (txc->modes & ADJ_SETOFFSET) {
struct timespec64 delta;
+
+ if (txc->time.tv_usec < 0 || txc->time.tv_usec >= USEC_PER_SEC)
+ return -EINVAL;
delta.tv_sec = txc->time.tv_sec;
delta.tv_nsec = txc->time.tv_usec;
if (!(txc->modes & ADJ_NANO))
--
2.7.4

2019-07-08 13:49:13

by Thomas Gleixner

Subject: Re: [PATCH] time: Validate the usec before covert to nsec in do_adjtimex

On Mon, 8 Jul 2019, ZhangXiaoxu wrote:

> When covert the usec to nsec, it will multiple 1000, it maybe
> overflow and lead an undefined behavior.
>
> For example, users may input an negative tv_usec values when
> call adjtimex syscall, then multiple 1000 maybe overflow it
> to a positive and legal number.
>
> So, we should validate the usec before coverted it to nsec.

That's correct, but the actuall inject function wants to keep the sanity
check,

Thanks,

tglx

2019-07-08 15:12:50

by zhangxiaoxu (A)

Subject: Re: [PATCH] time: Validate the usec before covert to nsec in do_adjtimex

?? 2019/7/8 17:24, Thomas Gleixner д??:
> On Mon, 8 Jul 2019, ZhangXiaoxu wrote:
>
>> When covert the usec to nsec, it will multiple 1000, it maybe
>> overflow and lead an undefined behavior.
>>
>> For example, users may input an negative tv_usec values when
>> call adjtimex syscall, then multiple 1000 maybe overflow it
>> to a positive and legal number.
>>
>> So, we should validate the usec before coverted it to nsec.
>
> That's correct, but the actuall inject function wants to keep the sanity
> check,
timekeeping_inject_offset is called only by timekeeping_warp_clock and do_adjtimex.
The do_adjtimex already validate it, and timekeeping_warp_clock is set tv_nsec=0.
We keep the sanity check is for some other maybe use this function?
I had send a v2 to keep the sanity check in timekeeping_inject_offset.

Thanks.
>
> Thanks,
>
> tglx
>
> .
>

2019-07-08 22:43:28

by Richard Cochran

Subject: Re: [PATCH] time: Validate the usec before covert to nsec in do_adjtimex

On Mon, Jul 08, 2019 at 03:55:04PM +0800, ZhangXiaoxu wrote:
> When covert the usec to nsec, it will multiple 1000, it maybe
> overflow and lead an undefined behavior.
>
> For example, users may input an negative tv_usec values when
> call adjtimex syscall, then multiple 1000 maybe overflow it
> to a positive and legal number.
>
> So, we should validate the usec before coverted it to nsec.
>
> Signed-off-by: ZhangXiaoxu <[email protected]>
> ---
> kernel/time/timekeeping.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
> index 44b726b..e5c1d00 100644
> --- a/kernel/time/timekeeping.c
> +++ b/kernel/time/timekeeping.c
> @@ -1272,9 +1272,6 @@ static int timekeeping_inject_offset(const struct timespec64 *ts)
> struct timespec64 tmp;
> int ret = 0;
>
> - if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC)
> - return -EINVAL;
> -
> raw_spin_lock_irqsave(&timekeeper_lock, flags);
> write_seqcount_begin(&tk_core.seq);
>
> @@ -2321,6 +2318,9 @@ int do_adjtimex(struct __kernel_timex *txc)
>
> if (txc->modes & ADJ_SETOFFSET) {
> struct timespec64 delta;
> +
> + if (txc->time.tv_usec < 0 || txc->time.tv_usec >= USEC_PER_SEC)
> + return -EINVAL;

This test is wrong. If the tv_usec field is in nanoseconds, then the
value can easily be greater than USEC_PER_SEC.

> delta.tv_sec = txc->time.tv_sec;
> delta.tv_nsec = txc->time.tv_usec;
> if (!(txc->modes & ADJ_NANO))
> --
> 2.7.4
>

Thanks,
Richard