[BusyBox] Bug#212764: 2 race conditions in init implementation

[Vladimir N. Oleynik]

Guillaume,

>>Really?!
>>WHY?
>>We blocked, not ignored SIGCHLD. If sigchld raised then signal pending
>>before unblocked. We don`t loses this signal.
> 
> 
> That's true, I was smoking crack here. 

?? Idioms? ;-)

> There is still a race here but
> for this reason, see below.

Its point not equivalent for below.
I think, must restore previous verion here.

>>>I have fixed the bug by replacing the wait() call in waitfor() with a
>>>waitpid() call and handle the error.
>>>
>>>- 2nd race condition
>>
>>And same point.
> 
> This one is not the same. You unblocked the signal before the second
> fork(). It is not a blocking/unblocking problem here. 
> 
> There is a race between the call to signal() and the signal handler
> (child_handler()). If the handler is called before signal(), the code
> will loop forever since waitpid will return -1 with errno == ECHILD.

Ah! You compared with verions have child_handler()? :-0
We remove child handler ~ 15 day ago.

> It is the same for the first race. If the handler is called before
> wait(), wait() will block.

wait MUST block!


> BTW, the SIGCHLD handler reaps all zombies, how can you expect that the
> unchecked calls to wait() and waitpid() will succeed on SMP ?

This code point must have wait(), not waitpid()

>>Glenn,
>>
>>I recomended restore previous version from CVS before realy
>>demonstrating problem.
> 
> 
> I recommend you leave it in the CVS. 

You have found problems in yours patch for this letter!

> Testing return codes is good

This code is difficult for testing for correctness.

Previous code may be have problem, I look Jean have point for demonstrate.
But your patch is bad!


--w
vodz