time: Avoid undefined behaviour in timespec64_add_safe()
authorVegard Nossum <vegard.nossum@oracle.com>
Fri, 12 Aug 2016 18:14:09 +0000 (20:14 +0200)
committerJohn Stultz <john.stultz@linaro.org>
Wed, 31 Aug 2016 21:43:35 +0000 (14:43 -0700)
commit469e857f374640f6164913835ce30d0736b40a60
tree40e99c581617200e46b1c53cbac32058a1b0de3e
parent0bf43f15db857e83daf4134aa062c8b157a80ee0
time: Avoid undefined behaviour in timespec64_add_safe()

I ran into this:

    ================================================================================
    UBSAN: Undefined behaviour in kernel/time/time.c:783:2
    signed integer overflow:
    5273 + 9223372036854771711 cannot be represented in type 'long int'
    CPU: 0 PID: 17363 Comm: trinity-c0 Not tainted 4.8.0-rc1+ #88
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.3-0-ge2fc41e-prebuilt.qemu-project.org
    04/01/2014
     0000000000000000 ffff88011457f8f0 ffffffff82344f50 0000000041b58ab3
     ffffffff84f98080 ffffffff82344ea4 ffff88011457f918 ffff88011457f8c8
     ffff88011457f8e0 7fffffffffffefff ffff88011457f6d8 dffffc0000000000
    Call Trace:
     [<ffffffff82344f50>] dump_stack+0xac/0xfc
     [<ffffffff82344ea4>] ? _atomic_dec_and_lock+0xc4/0xc4
     [<ffffffff8242f4c8>] ubsan_epilogue+0xd/0x8a
     [<ffffffff8242fc04>] handle_overflow+0x202/0x23d
     [<ffffffff8242fa02>] ? val_to_string.constprop.6+0x11e/0x11e
     [<ffffffff823c7837>] ? debug_smp_processor_id+0x17/0x20
     [<ffffffff8131b581>] ? __sigqueue_free.part.13+0x51/0x70
     [<ffffffff8146d4e0>] ? rcu_is_watching+0x110/0x110
     [<ffffffff8242fc4d>] __ubsan_handle_add_overflow+0xe/0x10
     [<ffffffff81476ef8>] timespec64_add_safe+0x298/0x340
     [<ffffffff81476c60>] ? timespec_add_safe+0x330/0x330
     [<ffffffff812f7990>] ? wait_noreap_copyout+0x1d0/0x1d0
     [<ffffffff8184bf18>] poll_select_set_timeout+0xf8/0x170
     [<ffffffff8184be20>] ? poll_schedule_timeout+0x2b0/0x2b0
     [<ffffffff813aa9bb>] ? __might_sleep+0x5b/0x260
     [<ffffffff833c8a87>] __sys_recvmmsg+0x107/0x790
     [<ffffffff833c8980>] ? SyS_recvmsg+0x20/0x20
     [<ffffffff81486378>] ? hrtimer_start_range_ns+0x3b8/0x1380
     [<ffffffff845f8bfb>] ? _raw_spin_unlock_irqrestore+0x3b/0x60
     [<ffffffff8148bcea>] ? do_setitimer+0x39a/0x8e0
     [<ffffffff813aa9bb>] ? __might_sleep+0x5b/0x260
     [<ffffffff833c9110>] ? __sys_recvmmsg+0x790/0x790
     [<ffffffff833c91e9>] SyS_recvmmsg+0xd9/0x160
     [<ffffffff833c9110>] ? __sys_recvmmsg+0x790/0x790
     [<ffffffff823c7853>] ? __this_cpu_preempt_check+0x13/0x20
     [<ffffffff8162f680>] ? __context_tracking_exit.part.3+0x30/0x1b0
     [<ffffffff833c9110>] ? __sys_recvmmsg+0x790/0x790
     [<ffffffff81007bd3>] do_syscall_64+0x1b3/0x4b0
     [<ffffffff845f936a>] entry_SYSCALL64_slow_path+0x25/0x25
    ================================================================================

Line 783 is this:

783         set_normalized_timespec64(&res, lhs.tv_sec + rhs.tv_sec,
784                         lhs.tv_nsec + rhs.tv_nsec);

In other words, since lhs.tv_sec and rhs.tv_sec are both time64_t, this
is a signed addition which will cause undefined behaviour on overflow.

Note that this is not currently a huge concern since the kernel should be
built with -fno-strict-overflow by default, but could be a problem in the
future, a problem with older compilers, or other compilers than gcc.

The easiest way to avoid the overflow is to cast one of the arguments to
unsigned (so the addition will be done using unsigned arithmetic).

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
include/linux/time64.h
kernel/time/time.c