PERFORCE change 107903 for review

[John Baldwin]

On Saturday 14 October 2006 12:04, Roman Divacky wrote:
> http://perforce.freebsd.org/chv.cgi?CH=107903
> 
> Change 107903 by rdivacky at rdivacky_witten on 2006/10/14 16:04:47
> 
> 	A bunch of fixes that makes this not panic when killpg() is called.
> 
> Affected files ...
> 
> .. //depot/projects/linuxolator/src/sys/compat/linux/linux_emul.c#10 edit
> 
> Differences ...
> 
> ==== //depot/projects/linuxolator/src/sys/compat/linux/linux_emul.c#10 
(text+ko) ====
> 
> @@ -212,8 +212,12 @@
>  	q = LIST_FIRST(&p->p_children);
>  	for (; q != NULL; q = nq) {
>  	   	nq = LIST_NEXT(q, p_sibling);
> -		if (__predict_true(q->p_sysent != &elf_linux_sysvec))
> -   		   	break;
> +	   	PROC_LOCK(q);
> +	   	if (q->p_flag & P_WEXIT)
> +		   	continue;
> +	   	PROC_UNLOCK(q);
> +		if (__predict_false(q->p_sysent != &elf_linux_sysvec))
> +   		   	continue;
>     	   	em = em_find(q, EMUL_UNLOCKED);
>  		KASSERT(em != NULL, ("linux_reparent: emuldata not found: %i\n", 
q->p_pid));
>  		if (em->pdeath_signal != 0) {

Holding the proc lock doesn't buy you anything here.  Probably you should hold 
an slock of the proctree_lock while you walk the list, and that should be 
good enough to test P_WEXIT.  However, even if you didn't hold it, grabbing 
the lock just to do a read doesn't buy you anything as far as closing race 
conditions.

-- 
John Baldwin