Patchwork [1/2] y2038: linux: Provide __mq_timedsend_time64 implementation

login
register
mail settings
Submitter Lukasz Majewski
Date Feb. 11, 2020, 4:52 p.m.
Message ID <20200211165256.21848-1-lukma@denx.de>
Download mbox | patch
Permalink /patch/37973/
State New
Headers show

Comments

Lukasz Majewski - Feb. 11, 2020, 4:52 p.m.
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(-)
Andreas Schwab - Feb. 11, 2020, 5:16 p.m.
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.
Joseph Myers - Feb. 12, 2020, 1:05 a.m.
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.
Lukasz Majewski - Feb. 12, 2020, 9:09 a.m.
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
Lukasz Majewski - Feb. 12, 2020, 9:18 a.m.
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
Andreas Schwab - Feb. 12, 2020, 10:19 a.m.
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.
Lukasz Majewski - Feb. 12, 2020, 10:37 a.m.
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
Florian Weimer - Feb. 12, 2020, 10:39 a.m.
* 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.
Andreas Schwab - Feb. 12, 2020, 10:44 a.m.
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.
Lukasz Majewski - Feb. 12, 2020, 11:21 a.m.
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
Lukasz Majewski - Feb. 12, 2020, 11:31 a.m.
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
Florian Weimer - Feb. 12, 2020, 11:33 a.m.
* 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.
Andreas Schwab - Feb. 12, 2020, 11:47 a.m.
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.
Lukasz Majewski - Feb. 12, 2020, 1:19 p.m.
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

Patch

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>
+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
diff --git a/sysdeps/unix/sysv/linux/mq_timedsend.c b/sysdeps/unix/sysv/linux/mq_timedsend.c
index 888ec6744a..662309c365 100644
--- a/sysdeps/unix/sysv/linux/mq_timedsend.c
+++ b/sysdeps/unix/sysv/linux/mq_timedsend.c
@@ -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)