mutex held in a thread which is cancelled stays busy

[Daniel Eischen]

> On Aug 7, 2019, at 6:25 AM, Erich Dollansky <freebsd.ed.lists at sumeritec.com> wrote:
> 
> Hi,
> 
> On Wed, 7 Aug 2019 06:07:25 -0400
> Daniel Eischen <deischen at freebsd.org> wrote:
> 
>>> On Aug 7, 2019, at 5:20 AM, Konstantin Belousov
>>> <kostikbel at gmail.com> wrote: 
>>>> On Wed, Aug 07, 2019 at 04:37:57PM +0800, Erich Dollansky wrote:
>>>> Hi,
>>>> 
>>>> On Wed, 7 Aug 2019 10:10:02 +0300
>>>> Konstantin Belousov <kostikbel at gmail.com> wrote:
>>>> 
>>>>>>> On Tue, Aug 06, 2019 at 08:58:30PM -0400, Daniel Eischen wrote:
>>>>>>> 
>>>>>>> On Aug 6, 2019, at 4:54 AM, Erich Dollansky
>>>>>>> <freebsd.ed.lists at sumeritec.com> wrote:
>>>>>>> 
>>>>>>> Hi,
>>>>>>> 
>>>>>>> for testing purpose, I did the following.
>>>>>>> 
>>>>>>> Start a thread, initialise a mutex in a global variable, lock
>>>>>>> the mutex and wait in that thread.
>>>>>>> 
>>>>>>> Wait in the main program until above's thread waits and cancel
>>>>>>> it.
>>>>>>> 
>>>>>>> Clean up behind the cancelled thread but leave intentional the
>>>>>>> mutex locked.
>>>>>>> 
>>>>>>> I would have expected now to get an error like 'EOWNERDEAD'
>>>>>>> doing operations with that mutex. But I get 'EBUSY' as the
>>>>>>> error.    
>>>>>> 
>>>>>> Are you initializing the mutex as a robust mutex, via
>>>>>> pthread_mutexattr_setrobust()?  Are you using _lock() or
>>>>>> _trylock()?   
>>>>> Robust mutexes only have special properties on the process
>>>>> termination. They behave same as the normal mutexes if the owning
>>>>> thread is terminated.
>>>>> 
>>>> man says:
>>>> 
>>>> [EOWNERDEAD]  The argument mutex points to a robust mutex and the
>>>> previous owning thread terminated while holding the mutex lock.  
>>> 
>>> So what ?  It describes the case when error can be returned, but it
>>> is not required to do so.  POSIX wording is the following:
>>> 
>>> If mutex is a robust mutex and the process containing the owning
>>> thread terminated while holding the mutex lock, a call to
>>> pthread_mutex_lock() shall return the error value [EOWNERDEAD]. If
>>> mutex is a robust mutex and the owning thread terminated while
>>> holding the mutex lock, a call to pthread_mutex_lock( ) may return
>>> the error value [EOWNERDEAD] even if the process in which the
>>> owning thread resides has not terminated.
>>> 
>>> Note the difference between shall and may.  We only process robust
>>> list on the process termination.  If the process is still alive,
>>> but the thread terminated, it can only happen because the process
>>> code asked for the thread termination explicitly, and then the code
>>> should be able to keep its own state.  On really fatal conditions,
>>> like unhandled signals, kernel terminates the process, not a
>>> thread.  
>> 
>> But pthread_mutex_lock() should not return EBUSY; that is only for
>> _trylock().  It seems to me _lock() should either return EOWNERDEAD
>> or EDEADLK, or it just blocks indefinitely.
>> 
>> Erich, are you getting EBUSY for pthread_mutex_lock() or is that only
>> for pthread_mutex_trylock()?
>> 
> EBUSY is only returned when I call 'pthread_mutex_trylock'. The other
> one just hangs.

In this case, I think FreeBSD is behaving correctly.  I think perhaps the only problem is that the man page isn't reflecting the POSIX wording.

--
DE