threads/101323: fork(2) in threaded programs broken.

Daniel Eischen deischen at freebsd.org
Thu Aug 3 20:04:51 UTC 2006


On Thu, 3 Aug 2006, Poul-Henning Kamp wrote:

> In message <Pine.GSO.4.64.0608031417260.13543 at sea.ntplx.net>, Daniel Eischen wr
> ites:
>
>> No, that's not nearly enough.  This has been discussed in
>> -threads before.
>>
>> Forking from a multi-threaded program is just like an
>> asynchronous signal in an unthreaded program.  You have
>> no idea what state any of the libraries or application data
>> is in.
>
> ... Unless of course, the programmer too great care to make
> sure he did, and therefore assumes that fork() will actually
> be safe.
>
> In my case, I know the exact state of the entire process
> and I know 100% certain that there are no locks held which
> will affect the forked copy.
>
> ... except that holding all malloc's locks screws me over :-(
>
> I will agree that there is no "perfect" solution, but that is
> not what I'm after, I'm after "works in controlled circumstances.
>
> I would argue that an implementation that does:
>
> 	hold any library locks we want to handle
> 	fork
> 	if (parent)
> 		release those locks again
> 		return
> 	else
> 		unlock all locks (since they cannot possibly
> 		make sense in the child in a locked state)
> 		return

There's no easy way to hold all library locks.  They are
littered in libc and libpthread and the application doesn't
have access to them.  You would have to teach libc to
record these locks and export a function to lib<thread>
to lock and unlock these them.

> That would go a long way towards a "works if you're careful"
> implementation.

-- 
DE


More information about the freebsd-threads mailing list