Suspected bug in mq_timedreceive

Suspected bug in mq_timedreceive

[sundeep.kokkonda]

Hello,

 

The mq_timedreceive code in
glibc-2.33/sysdeps/unix/sysv/linux/mq_timedreceive.c does this:

 

ssize_t

_mq_timedreceive_time64 (mqd_t mqdes, char *_restrict msg_ptr, size_t
msg_len,

                          unsigned int *__restrict msg_prio,

                          const struct _timespec64 *_restrict abs_timeout)

{

#ifndef __NR_mq_timedreceive_time64

#define __NR_mq_timedreceive_time64 __NR_mq_timedreceive

#endif

  int ret = SYSCALL_CANCEL (mq_timedreceive_time64, mqdes, msg_ptr, msg_len,

                            msg_prio, abs_timeout);

#ifndef __ASSUME_TIME64_SYSCALLS

  if (ret == 0 || errno != ENOSYS)

    return ret;

 

  struct timespec ts32;

  if (abs_timeout != NULL)

    {

      if (! in_time_t_range (abs_timeout->tv_sec))

       

 

{           __set_errno (EOVERFLOW);           return -1;         }

      ts32 = valid_timespec64_to_timespec (*abs_timeout);

    }

 

  ret = SYSCALL_CANCEL (mq_timedreceive, mqdes, msg_ptr, msg_len, msg_prio,

                        abs_timeout != NULL ? &ts32 : NULL);

#endif

 

  return ret;

}

 

This test is wrong:

 

  if (ret == 0 || errno != ENOSYS)

    return ret;

 

That test would be correct if mq_timedreceive returned 0 on success, as is
the case for mq_timedsend. However, mq_timedreceive actually returns the
number of bytes read, so the correct test is:

 

  If (ret >= 0 || errno != ENOSYS)

     return ret;

 

 

@community: Can you comment on this, whether it is a bug or implemented
intentionally for any specific reason? I checked the latest glibc-2.35
sources as well, the check for ret==0 is still the same.

 

 

 

Thanks,

Sundeep K.

Re: Suspected bug in mq_timedreceive

[Adhemerval Zanella]

> On 30 Jun 2022, at 03:37, sundeep.kokkonda--- via Libc-help <libc-help@sourceware.org> wrote:
> 
> Hello,
> 
> 
> 
> The mq_timedreceive code in
> glibc-2.33/sysdeps/unix/sysv/linux/mq_timedreceive.c does this:
> 
> 
> 
> ssize_t
> 
> _mq_timedreceive_time64 (mqd_t mqdes, char *_restrict msg_ptr, size_t
> msg_len,
> 
>                          unsigned int *__restrict msg_prio,
> 
>                          const struct _timespec64 *_restrict abs_timeout)
> 
> {
> 
> #ifndef __NR_mq_timedreceive_time64
> 
> #define __NR_mq_timedreceive_time64 __NR_mq_timedreceive
> 
> #endif
> 
>  int ret = SYSCALL_CANCEL (mq_timedreceive_time64, mqdes, msg_ptr, msg_len,
> 
>                            msg_prio, abs_timeout);
> 
> #ifndef __ASSUME_TIME64_SYSCALLS
> 
>  if (ret == 0 || errno != ENOSYS)
> 
>    return ret;
> 
> 
> 
>  struct timespec ts32;
> 
>  if (abs_timeout != NULL)
> 
>    {
> 
>      if (! in_time_t_range (abs_timeout->tv_sec))
> 
> 
> 
> 
> 
> {           __set_errno (EOVERFLOW);           return -1;         }
> 
>      ts32 = valid_timespec64_to_timespec (*abs_timeout);
> 
>    }
> 
> 
> 
>  ret = SYSCALL_CANCEL (mq_timedreceive, mqdes, msg_ptr, msg_len, msg_prio,
> 
>                        abs_timeout != NULL ? &ts32 : NULL);
> 
> #endif
> 
> 
> 
>  return ret;
> 
> }
> 
> 
> 
> This test is wrong:
> 
> 
> 
>  if (ret == 0 || errno != ENOSYS)
> 
>    return ret;
> 
> 
> 
> That test would be correct if mq_timedreceive returned 0 on success, as is
> the case for mq_timedsend. However, mq_timedreceive actually returns the
> number of bytes read, so the correct test is:
> 
> 
> 
>  If (ret >= 0 || errno != ENOSYS)
> 
>     return ret;

Yeah, this is incorrect.  I will open a bug an fix it on all affected releases.
At least for 2.36 the time64 code path is only used if the target date can not
be fit in 32-bit time_t.