[PATCH] [BZ #19402] Clear list of acquired robust mutexes in the child process after forking.

[Florian Weimer]

On 01/13/2017 04:37 PM, Torvald Riegel wrote:
> On Fri, 2017-01-13 at 14:11 +0100, Florian Weimer wrote:
>> On 12/23/2016 10:31 PM, Torvald Riegel wrote:
>>>> Can we add a test case for this?
>>> What do you have in mind?  Checking that the list is reset to zero after
>>> fork?  Do we need a test for that if we have documented the need to do
>>> that in the code?
>>
>> What about the attached patch?
>
> That looks fine generally, with the following exceptions / comments:
>
> According to the feedback so far we got for
> http://austingroupbugs.net/view.php?id=1112 the trylock-based test is
> actually UB.  There should be a comment making that clear (ie, that we
> rely on something we know about our current implementation to test
> this).

Yeah, I'll move the check to the parent process and add another bit for 
error-checking mutexes (and make the test conditional on that).

> If this was a multi-threaded test program, it would be invalid because
> non-async-signal-safe functions are executed in the child before exec()
> is called.  In a single-threaded program, it's not quite clear to me
> what POSIX really wants.  There's indication that only AS-safe functions
> should be called, but IIRC that's not explicitly stated.
> Florian says we still need to support this as an extension.  If so, this
> should be made clear in the test (using a suitable comment).

I'll add a comment, yes.

> What about putting the x* wrappers as static inline functions just in
> the header?  Then, we wouldn't need to add all those trivial C files.

We did that before and it caused problems because it restricted what 
feature macros we could assume in the implementation of those inline 
functions.

I verified your patch with this test, so I think it is okay to check it 
in.  I'll commit my updated patch afterwards.

Thanks,
Florian