linux-kernel - Re: [PATCH v3 0/7] Isolate time_t data types for clock/timer syscalls

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Mon, 3 Jul 2017 13:19:30 +0200
From:   Arnd Bergmann <arnd@...db.de>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     Deepa Dinamani <deepa.kernel@...il.com>,
        Al Viro <viro@...iv.linux.org.uk>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        John Stultz <john.stultz@...aro.org>,
        Nicolas Pitre <nicolas.pitre@...aro.org>,
        y2038 Mailman List <y2038@...ts.linaro.org>,
        Linux FS-devel Mailing List <linux-fsdevel@...r.kernel.org>
Subject: Re: [PATCH v3 0/7] Isolate time_t data types for clock/timer syscalls

On Mon, Jul 3, 2017 at 12:23 PM, Thomas Gleixner <tglx@...utronix.de> wrote:
> On Mon, 26 Jun 2017, Arnd Bergmann wrote:
>> On Mon, Jun 26, 2017 at 8:17 PM, Deepa Dinamani <deepa.kernel@...il.com> wrote:
>> > On Sun, Jun 25, 2017 at 7:35 PM, Al Viro <viro@...iv.linux.org.uk> wrote:
>> >> On Sat, Jun 24, 2017 at 11:45:01AM -0700, Deepa Dinamani wrote:
>> >>> The series aims at isolating data conversions of time_t based structures:
>> >>> struct timespec and struct itimerspec at user space boundaries.
>> >>> This helps to later change the underlying types to handle y2038 changes
>> >>> to these.
>> >>
>> >> Nice...  A few questions:
>> >>
>> >> * what about setitimer(2)?  Right now that's the only remaining user of
>> >> get_compat_itimerval(); similar for getitimer(2) and put_compat_itimerval().
>> >
>> > We do not plan to support these beyond y2038 on 32 bit systems.
>> > timer_settime() and timer_gettime() are considered to be replacements
>> > for these, respectively.
>> >
>> > There is also going to be a cleanup of timeval/ timespec/ time_t data
>> > types and apis after the new syscalls are ready.
>> > At that time I might choose to get rid of these itimerval apis. I'm
>> > not sure yet.
>>
>> I see that internally, alarm/getitimer/setitimer all use ktime_t, so
>> one possible solution would be to push down the use of ktime_t
>> into the callers and do both the conversion and range check in the
>> user copy function.
>
> We still can decide to not support the itimer API with the new y2038 ready
> syscalls.
>
> Actually there is no real need to do so because the itimer interfaces are
> relative and never absolute. Keeping relative time limited to 68 years from
> now should be good enough :)

I really want to have all syscalls to use 64-bit time_t for the new
API, otherwise
we get into the really silly state where even for future architectures, glibc
has to convert the 64-bit time_t coming from an application into a 32-bit
timeval to pass it to the kernel, which then converts it back to an internal
type (64-bit time_t or ktime_t).

In case of the itimer interfaces, this is really no question though, as
Deepa said, since glibc can simply implement the new version by calling
timer_create/timer_settime instead of calling the 32-bit setitimer.
timer_settime() needs to take 64-bit arguments anyway because it
can use either relative or absolute arguments.

       Arnd