Re: iolint race condition

[Andrew J. Schorr]

Hi Miguel,

On Mon, Dec 06, 2021 at 09:37:26PM -0500, Miguel Pineiro Jr. wrote:
> I haven't done any testing, but after looking at the code, I believe
> the problem is here:
> 
>       print "hello" | "cat"
>       print "/bin/cat \"$@\"" > "cat"
>       print close("cat")
>       print close("cat")
> 
> Redirections are maintained in a most-recently-used list (io.c ::
> redirect_string). Therefore the first close() corresponds to the
> regular file of the second print statement. This creates a race
> between when the child process "cat" prints "hello" and the
> printing of the first close statement's return value.
> 
> If I'm correct, the fix is to reverse the order of the redirections
> so that nothing can print between the time the child is forked
> and the time it's waited on:
> 
>       print "/bin/cat \"$@\"" > "cat"
>       print "hello" | "cat"
>       print close("cat")
>       print close("cat")

I guess you're saying that with this change, the results should
consistently be:

hello
0
0

I don't love that it's implementation-dependent, but it may be the most
expedient way of solving the problem.

Arnold -- OK to apply the attached patch?

> P. S. Nice work refactoring efwrite. Great idea.

Thanks -- a happy moment when I was able to escape being a prisoner of
the existing code.

Regards,
Andy

iolint.patch
Description: Text document