[1/2] y2038: linux: Provide __timerfd_gettime64 implementation

  This patch replaces auto generated wrapper (as described in
sysdeps/unix/sysv/linux/syscalls.list) for timerfd_gettime with one which
adds extra support for reading 64 bit time values on machines with
__TIMESIZE != 64.
There is no functional change for architectures already supporting 64 bit
time ABI.

This patch is conceptually identical to timer_gettime conversion already
done in sysdeps/unix/sysv/linux/timer_gettime.c.
Please refer to corresponding commit message for detailed description of
introduced functions and the testing procedure.
---
 include/time.h                            |  3 +
 sysdeps/unix/sysv/linux/Makefile          |  3 +-
 sysdeps/unix/sysv/linux/syscalls.list     |  1 -
 sysdeps/unix/sysv/linux/timerfd_gettime.c | 69 +++++++++++++++++++++++
 4 files changed, 74 insertions(+), 2 deletions(-)
 create mode 100644 sysdeps/unix/sysv/linux/timerfd_gettime.c
  

Dear Glibc Community,

I would like to share the current state of development and potentially
avoid effort duplication.

Please find the status update info regarding the effort to make glibc
supporting 64 bit time on systems with TIMESIZE != 64 (i.e. 32 bit ARM).

The most up to date setup - with "upstream first" philosophy applied
(including patch to provide support for -D_TIME_BITS=64) to glibc
can be found at [1].

Following functions to support 64 bit syscalls have been converted:

    - clock_gettime64
    - clock_settime64
    - clock_getres64
    - utimensat
    - futimens
    - clock_nanosleep_time64
    - ppoll64
    - timer_gettime
    - timer_settime
    - timerfd_gettime (WIP)
    - timerfd_settime (WIP) [2]

    - pselect64 conversion has been postponed until we advance with
      minimal supported glibc version (now it is 3.2 , we need to be >
      3.15). After 3.15 it would be far more easier to do the
      conversion.

It is possible to test this effort with QEMU and OE/Yocto. Detailed
tutorial in README file [3].

TO DO:

0. Use other 64 bit syscalls - as indicated in [4].

1. Convert in-glibc usage of internal clock functions (e.g.
    __clock_gettime()) to __clock_gettime64(), which are Y2038 safe on
    32 bit systems with TIMESIZE==32.

    Potential issue - making those functions as hidden and in the same
    time accessible by Y2038 safe systems (e.g. __clock_gettime64()).

2. The statx (and friends) conversion to Y2038 (Alistair also did some
   work there)

3. Continue replacing Y2038 unsafe functions (like gettimeofday) with
   internal, safe representation (which would use __clock_gettime64())
   as was done by Zack and Adhemerval.

4. Prepare for "big switch" to add support for -D_TIME_BITS=64 in glibc
   - I'm mostly concerned with preparing a solid test sute.


Please do not hesitate to provide input about any other ongoing work in
this area. Please also forward this mail to any potentially interested
parties.

Last, but not least, I would like to give a big thank to all community
members involved into the development and review process.


Links:

[1] -
https://github.com/lmajewski/y2038_glibc/commits/glibc_timerfd_settime_gettime-conversion-v1

[2] - posted for review:
https://patchwork.ozlabs.org/patch/1205307/

[3] - https://github.com/lmajewski/meta-y2038

[4] -
https://elixir.bootlin.com/linux/v5.3-rc5/source/arch/arm/tools/syscall.tbl#L420



Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
  

On Sat, 7 Dec 2019, Lukasz Majewski wrote:

> diff --git a/sysdeps/unix/sysv/linux/timerfd_gettime.c b/sysdeps/unix/sysv/linux/timerfd_gettime.c
> new file mode 100644
> index 0000000000..498605369b
> --- /dev/null
> +++ b/sysdeps/unix/sysv/linux/timerfd_gettime.c
> @@ -0,0 +1,69 @@
> +/* Copyright (C) 2003-2019 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +   Contributed by Ulrich Drepper <drepper@redhat.com>, 2003.

New files need to have a descriptive comment before the copyright notice, 
and no "Contributed by" line, and only have previous years in the 
copyright notice if genuinely incorporating copyrightable content from 
previous files from those years.

> +int
> +__timerfd_gettime64 (int fd, struct __itimerspec64 *value)
> +{
> +  if (fd < 0)
> +    return INLINE_SYSCALL_ERROR_RETURN_VALUE (EBADF);

Why?  In general, such checks are only needed in userspace if correct 
function semantics means not passing such a case to the kernel at all, as 
opposed to letting the kernel return an error for it.

The same comments apply to patch 2/2.
  

Hi Joseph,

> On Sat, 7 Dec 2019, Lukasz Majewski wrote:
> 
> > diff --git a/sysdeps/unix/sysv/linux/timerfd_gettime.c
> > b/sysdeps/unix/sysv/linux/timerfd_gettime.c new file mode 100644
> > index 0000000000..498605369b
> > --- /dev/null
> > +++ b/sysdeps/unix/sysv/linux/timerfd_gettime.c
> > @@ -0,0 +1,69 @@
> > +/* Copyright (C) 2003-2019 Free Software Foundation, Inc.
> > +   This file is part of the GNU C Library.
> > +   Contributed by Ulrich Drepper <drepper@redhat.com>, 2003.  
> 
> New files need to have a descriptive comment before the copyright
> notice, and no "Contributed by" line, and only have previous years in
> the copyright notice if genuinely incorporating copyrightable content
> from previous files from those years.

Thanks for the hint.

I will add the descriptive comment and treat this file as a "new" one -
with copyright description similar to newly added files.

> 
> > +int
> > +__timerfd_gettime64 (int fd, struct __itimerspec64 *value)
> > +{
> > +  if (fd < 0)
> > +    return INLINE_SYSCALL_ERROR_RETURN_VALUE (EBADF);  
       ^^^^^^^^^ - [*]

> 
> Why?  In general, such checks are only needed in userspace if correct 
> function semantics means not passing such a case to the kernel at
> all, as opposed to letting the kernel return an error for it.

I took this approach from futimens.c (from
sysdeps/unix/sysv/linux). This is how the wrong fd is handled.
According to manual [1] the EBADF shall be returned.

The manual for timerfd_gettime [2] also states that wrong fd shall
cause the EBADF return value.

And yes - you are right - the Linux kernel checks [3] if fd is valid and
returns either -EBADF or -EINVAL.
Considering the above - I will remove the check [*] from
__timerfd_gettime64 and send v2.

> 
> The same comments apply to patch 2/2.
> 

Ok.


Links:

[1] - https://linux.die.net/man/3/futimens
[2] - https://linux.die.net/man/2/timerfd_gettime
[3] - https://elixir.bootlin.com/linux/v5.4/source/fs/timerfd.c#L374



Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de