[Bug 198216] According to man page for pthread_cond_destroy it returns EBUSY error code, but it never does

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=198216

            Bug ID: 198216
           Summary: According to man page for pthread_cond_destroy it
                    returns EBUSY error code, but it never does
           Product: Documentation
           Version: Latest
          Hardware: Any
               URL: https://www.freebsd.org/cgi/man.cgi?query=pthread_cond
                    _destroy&apropos=0&sektion=3&manpath=FreeBSD+11-curren
                    t&arch=default&format=html
                OS: Any
            Status: New
          Severity: Affects Many People
          Priority: ---
         Component: Documentation
          Assignee: freebsd-doc at FreeBSD.org
          Reporter: msiodelski at gmail.com

I am using the following FreeBSD version:

10.1-RELEASE-p6 FreeBSD 10.1-RELEASE-p6 #0 r279272: Wed Feb 25 12:27:40 CET
2015

built from the following svn branch: ^/releng/10.1

According to the man pages for the pthread_cond_destroy, it returns the
following error codes:

ERRORS
     The pthread_cond_destroy() function will fail if:

     [EINVAL]           The value specified by cond is invalid.

     [EBUSY]            The variable cond is locked by another thread.


In the course of my testing I found that this function actually never returns
EBUSY error code if the conditional variable is locked by another thread.

Briefly looking at the pthread_cond_destroy implementation in
/usr/src/lib/libthr/thread/thr_cond.c I don't see any checks for variable being
locked by another thread:

int
_pthread_cond_destroy(pthread_cond_t *cond)
{
    struct pthread_cond    *cvp;
    int            error = 0;

    if ((cvp = *cond) == THR_COND_INITIALIZER)
        error = 0;
    else if (cvp == THR_COND_DESTROYED)
        error = EINVAL;
    else {
        cvp = *cond;
        *cond = THR_COND_DESTROYED;

        /*
         * Free the memory allocated for the condition
         * variable structure:
         */
        free(cvp);
    }
    return (error);
}



So it seems that one of the: documentation or the code is wrong.

Some background about how it was tested. I am working on the Kea DHCP server
implementation (kea.isc.org). One of our unit tests is exercising the scenario
of trying to destroy the cond variable while it is locked by another thread.
The expected behavior is that the pthread_cond_destroy fails with non-success
status code, but it returns 0. However, this test uses C++ objects and
infrastructure of our implementation and it hasn't been isolated as standalone
test. If you find it useful to have such a standalone test which demonstrates
the behavior, I can write one.

For your convenience I am sending the locations of our source code where we run
the test nested in our implementation:
https://github.com/isc-projects/kea/blob/master/src/lib/util/threads/tests/condvar_unittest.cc
(see the destroyWhileWait test).

and the source code of the CondVar class under test:
https://github.com/isc-projects/kea/blob/master/src/lib/util/threads/sync.cc

Note that the destructor of the CondVar class calls the pthread_cond_destroy
and then performs an assert that the status code is 0. For this test it should
be non-zero and the process should die as expected by the test. But, it
doesn't.

Hope that helps.

-- 
You are receiving this mail because:
You are the assignee for the bug.