[1/2] y2038: linux: Provide __mq_timedsend_time64 implementation
Commit Message
This patch provides new __mq_timedsend_time64 explicit 64 bit function for
sending messages with absolute timeout.
Moreover, a 32 bit version - __mq_timedsend has been refactored to internally
use __mq_timedsend_time64.
The __mq_timedsend is now supposed to be used on systems still supporting 32
bit time (__TIMESIZE != 64) - hence the necessary conversion to 64 bit struct
__timespec64 from struct timespec.
The new __mq_timerdsend_time64 syscall available from Linux 5.1+ has been used,
when applicable.
As this wrapper function is also used internally in the glibc, to e.g. provide
mq_send implementation, an explicit check for abs_timeout being NULL has been
added due to conversions between struct timespec and struct __timespec64 .
Before this change the Linux kernel handled this NULL pointer.
Build tests:
- ./src/scripts/build-many-glibcs.py glibcs
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
Linux kernel, headers and minimal kernel version for glibc build test matrix:
- Linux v5.1 (with mq_timedsend_time64) and glibc built with v5.1 as a
minimal kernel version (--enable-kernel="5.1.0")
The __ASSUME_TIME64_SYSCALLS flag defined.
- Linux v5.1 and default minimal kernel version
The __ASSUME_TIME64_SYSCALLS not defined, but kernel supports
mq_timedsend_time64 syscall.
- Linux v4.19 (no mq_timedsend_time64 support) with default minimal kernel
version for contemporary glibc (3.2.0)
This kernel doesn't support mq_timedsend_time64 syscall, so the fallback to
mq_timedsend is tested.
Above tests were performed with Y2038 redirection applied as well as without
(so the __TIMESIZE != 64 execution path is checked as well).
---
include/mqueue.h | 9 +++++
sysdeps/unix/sysv/linux/mq_timedsend.c | 49 ++++++++++++++++++++++++--
2 files changed, 55 insertions(+), 3 deletions(-)
Comments
On Feb 11 2020, Lukasz Majewski wrote:
> diff --git a/include/mqueue.h b/include/mqueue.h
> index 3c66f1711e..3de890905c 100644
> --- a/include/mqueue.h
> +++ b/include/mqueue.h
> @@ -10,4 +10,13 @@ extern __typeof (mq_timedreceive) __mq_timedreceive __nonnull ((2, 5));
> hidden_proto (__mq_timedreceive)
> hidden_proto (mq_setattr)
> # endif
> +#if __TIMESIZE == 64
> +# define __mq_timedsend_time64 __mq_timedsend
> +#else
> +#include <include/time.h>
I think struct __timespec64 should be moved to a separate file, so that
we do not need to pull in the whole <time.h> namespace here.
Andreas.
On Tue, 11 Feb 2020, Lukasz Majewski wrote:
> +int
> +__mq_timedsend (mqd_t mqdes, const char *msg_ptr, size_t msg_len,
> + unsigned int msg_prio, const struct timespec *abs_timeout)
> +{
> + struct __timespec64 ts64;
> + if (abs_timeout)
> + ts64 = valid_timespec_to_timespec64 (*abs_timeout);
> +
> + return __mq_timedsend_time64 (mqdes, msg_ptr, msg_len, msg_prio,
> + abs_timeout ? &ts64 : NULL);
> }
The indentation seems wrong here. Likewise in patch 2.
Hi Joseph,
> On Tue, 11 Feb 2020, Lukasz Majewski wrote:
>
> > +int
> > +__mq_timedsend (mqd_t mqdes, const char *msg_ptr, size_t msg_len,
> > + unsigned int msg_prio, const struct timespec
> > *abs_timeout) +{
> > + struct __timespec64 ts64;
> > + if (abs_timeout)
> > + ts64 = valid_timespec_to_timespec64 (*abs_timeout);
> > +
> > + return __mq_timedsend_time64 (mqdes, msg_ptr, msg_len,
> > msg_prio,
> > + abs_timeout ? &ts64 : NULL);
> > }
>
> The indentation seems wrong here. Likewise in patch 2.
>
Thanks for spotting it. I will fix it in v2.
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
Hi Andreas,
> On Feb 11 2020, Lukasz Majewski wrote:
>
> > diff --git a/include/mqueue.h b/include/mqueue.h
> > index 3c66f1711e..3de890905c 100644
> > --- a/include/mqueue.h
> > +++ b/include/mqueue.h
> > @@ -10,4 +10,13 @@ extern __typeof (mq_timedreceive)
> > __mq_timedreceive __nonnull ((2, 5)); hidden_proto
> > (__mq_timedreceive) hidden_proto (mq_setattr)
> > # endif
> > +#if __TIMESIZE == 64
> > +# define __mq_timedsend_time64 __mq_timedsend
> > +#else
> > +#include <include/time.h>
>
> I think struct __timespec64 should be moved to a separate file, so
> that we do not need to pull in the whole <time.h> namespace here.
>
Just to clarify - the time.h here is from ./include glibc sources
directory, so it is "internal" for glibc.
In other words - I would like to know if your comment was regarding
polluting the glibc internal namespace?
> Andreas.
>
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 Feb 12 2020, Lukasz Majewski wrote:
> Just to clarify - the time.h here is from ./include glibc sources
> directory, so it is "internal" for glibc.
But it is a wrapper around <time.h>.
> In other words - I would like to know if your comment was regarding
> polluting the glibc internal namespace?
It's not about polluting, but since the #include is conditional, we have
inconsistent environment between configurations, which isn't nice.
By moving struct __timespec64 to <include/bits/struct/timespec.h> we can
avoid that inconsistency.
Andreas.
Hi Andreas,
> On Feb 12 2020, Lukasz Majewski wrote:
>
> > Just to clarify - the time.h here is from ./include glibc sources
> > directory, so it is "internal" for glibc.
>
> But it is a wrapper around <time.h>.
Yes, indeed - it includes <time/time.h>
>
> > In other words - I would like to know if your comment was regarding
> > polluting the glibc internal namespace?
>
> It's not about polluting, but since the #include is conditional, we
> have inconsistent environment between configurations, which isn't
> nice.
Ach... I see.
>
> By moving struct __timespec64 to <include/bits/struct/timespec.h> we
> can avoid that inconsistency.
Maybe <include/bits/types/struct___timespec64.h> would be a better
place/name?
In that way we would follow other "internal" structs definitions - i.e.
<include/bits/types/struct_timespec.h>
(which only has #include·<time/bits/types/struct_timespec.h>)
>
> Andreas.
>
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
* Lukasz Majewski:
>> By moving struct __timespec64 to <include/bits/struct/timespec.h> we
>> can avoid that inconsistency.
>
> Maybe <include/bits/types/struct___timespec64.h> would be a better
> place/name?
bits/ headers should be installed headers. If it is just an internal
type, it should not go into a bits/ header, but a header under
include/ or maybe sysdeps/generic/. The latter is useful if sysdeps
overrides are expected.
On Feb 12 2020, Lukasz Majewski wrote:
>> By moving struct __timespec64 to <include/bits/struct/timespec.h> we
>> can avoid that inconsistency.
>
> Maybe <include/bits/types/struct___timespec64.h> would be a better
> place/name?
By adding it to <include/bits/struct/timespec.h> it would be included
automatically.
Andreas.
Hi Florian,
> * Lukasz Majewski:
>
> >> By moving struct __timespec64 to <include/bits/struct/timespec.h>
> >> we can avoid that inconsistency.
> >
> > Maybe <include/bits/types/struct___timespec64.h> would be a better
> > place/name?
>
> bits/ headers should be installed headers. If it is just an internal
> type, it should not go into a bits/ header, but a header under
> include/ or maybe sysdeps/generic/. The latter is useful if sysdeps
> overrides are expected.
the idea would be to place it under ./include/bits/ in the glibc source
directory, so it would be an "internal" header.
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 Wed, 12 Feb 2020 11:44:20 +0100
Andreas Schwab <schwab@suse.de> wrote:
> On Feb 12 2020, Lukasz Majewski wrote:
>
> >> By moving struct __timespec64 to <include/bits/struct/timespec.h>
> >> we can avoid that inconsistency.
> >
> > Maybe <include/bits/types/struct___timespec64.h> would be a better
> > place/name?
>
> By adding it to <include/bits/struct/timespec.h> it would be included
> automatically.
>
Then it would be a great optimization (as __timespec64 is included in
several places)
(And I do guess that this - auto include - is related to the directory
search patch in Makefiles?)
> Andreas.
>
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
* Lukasz Majewski:
> Hi Florian,
>
>> * Lukasz Majewski:
>>
>> >> By moving struct __timespec64 to <include/bits/struct/timespec.h>
>> >> we can avoid that inconsistency.
>> >
>> > Maybe <include/bits/types/struct___timespec64.h> would be a better
>> > place/name?
>>
>> bits/ headers should be installed headers. If it is just an internal
>> type, it should not go into a bits/ header, but a header under
>> include/ or maybe sysdeps/generic/. The latter is useful if sysdeps
>> overrides are expected.
>
> the idea would be to place it under ./include/bits/ in the glibc source
> directory, so it would be an "internal" header.
Okay, that part makes sense.
On Feb 12 2020, Lukasz Majewski wrote:
> (And I do guess that this - auto include - is related to the directory
> search patch in Makefiles?)
For the same reason <time.h> finds include/time.h first during build.
Andreas.
Hi Andreas,
> On Feb 12 2020, Lukasz Majewski wrote:
>
> > (And I do guess that this - auto include - is related to the
> > directory search patch in Makefiles?)
>
> For the same reason <time.h> finds include/time.h first during build.
>
> Andreas.
>
I've added the struct __timespec64 definition to
./include/bits/struct/timespec.h
and I had build errors on ./include/time.h which uses struct
__timespec64 massively.
It requires explicit adding of #include
<include/bits/struct/timespec.h> or <bits/struct/timespec.h>
The same #include was required in ./include/mqueue.h to have proper
struct __timespec64 declaration.
It seems like the glibc build system fails to #include it automatically
to internal headers (like ./include/time.h and ./include/mqueue.h).
IMHO, it would be better to add definition of struct __timespec64 into
./include/bits/types/struct___timespec64.h (which is intended for
internal glibc use).
Such approach would follow what we do have now in ./include/time.h
with # include <bits/types/struct_timeval.h>
Or what I'm missing here?
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
@@ -10,4 +10,13 @@ extern __typeof (mq_timedreceive) __mq_timedreceive __nonnull ((2, 5));
hidden_proto (__mq_timedreceive)
hidden_proto (mq_setattr)
# endif
+#if __TIMESIZE == 64
+# define __mq_timedsend_time64 __mq_timedsend
+#else
+#include <include/time.h>
+extern int __mq_timedsend_time64 (mqd_t mqdes, const char *msg_ptr,
+ size_t msg_len, unsigned int msg_prio,
+ const struct __timespec64 *abs_timeout);
+librt_hidden_proto (__mq_timedsend_time64)
+#endif
#endif
@@ -22,12 +22,55 @@
/* Add message pointed by MSG_PTR to message queue MQDES, stop blocking
on full message queue if ABS_TIMEOUT expires. */
int
-__mq_timedsend (mqd_t mqdes, const char *msg_ptr, size_t msg_len,
- unsigned int msg_prio, const struct timespec *abs_timeout)
+__mq_timedsend_time64 (mqd_t mqdes, const char *msg_ptr, size_t msg_len,
+ unsigned int msg_prio,
+ const struct __timespec64 *abs_timeout)
{
+#ifdef __ASSUME_TIME64_SYSCALLS
+# ifndef __NR_mq_timedsend_time64
+# define __NR_mq_timedsend_time64 __NR_mq_timedsend
+# endif
+ return SYSCALL_CANCEL (mq_timedsend_time64, mqdes, msg_ptr, msg_len, msg_prio,
+ abs_timeout);
+#else
+ int ret = SYSCALL_CANCEL (mq_timedsend_time64, mqdes, msg_ptr, msg_len,
+ msg_prio, abs_timeout);
+ if (ret == 0 || errno != ENOSYS)
+ return ret;
+
+ struct timespec ts32;
+ if (abs_timeout)
+ {
+ if (! in_time_t_range (abs_timeout->tv_sec))
+ {
+ __set_errno (EOVERFLOW);
+ return -1;
+ }
+
+ ts32 = valid_timespec64_to_timespec (*abs_timeout);
+ }
+
return SYSCALL_CANCEL (mq_timedsend, mqdes, msg_ptr, msg_len, msg_prio,
- abs_timeout);
+ abs_timeout ? &ts32 : NULL);
+#endif
+}
+
+#if __TIMESIZE != 64
+librt_hidden_def (__mq_timedsend_time64)
+
+int
+__mq_timedsend (mqd_t mqdes, const char *msg_ptr, size_t msg_len,
+ unsigned int msg_prio, const struct timespec *abs_timeout)
+{
+ struct __timespec64 ts64;
+ if (abs_timeout)
+ ts64 = valid_timespec_to_timespec64 (*abs_timeout);
+
+ return __mq_timedsend_time64 (mqdes, msg_ptr, msg_len, msg_prio,
+ abs_timeout ? &ts64 : NULL);
}
+#endif
+
hidden_def (__mq_timedsend)
weak_alias (__mq_timedsend, mq_timedsend)
hidden_weak (mq_timedsend)