svn commit: r274921 - projects/routing/sys/net

[Marko Zec]

On Mon, 24 Nov 2014 09:51:22 +0300
Gleb Smirnoff <glebius at freebsd.org> wrote:

> On Sun, Nov 23, 2014 at 05:47:12PM +0000, Alexander V. Chernikov
> wrote:
> A> Author: melifaro
> A> Date: Sun Nov 23 17:47:12 2014
> A> New Revision: 274921
> A> URL: https://svnweb.freebsd.org/changeset/base/274921
> A> 
> A> Log:
> A>   Finish r274335#2: put RT_LOCK_DESTROY() back.
> A> 
> A> Modified:
> A>   projects/routing/sys/net/route.c
> A> 
> A> Modified: projects/routing/sys/net/route.c
> A> ==============================================================================
> A> --- projects/routing/sys/net/route.c	Sun Nov 23 17:45:49
> A> 2014	(r274920) +++ projects/routing/sys/net/route.c
> A> Sun Nov 23 17:47:12 2014	(r274921) @@ -1505,6 +1505,7 @@
> A> rtrequest1_fib(int req, struct rt_addrin if (rn == NULL) {
> A>  			ifa_free(rt->rt_ifa);
> A>  			Free(rt_key(rt));
> A> +			RT_LOCK_DESTROY(rt);
> A>  			uma_zfree(V_rtzone, rt);
> A>  #ifdef FLOWTABLE
> A>  			if (rt0 != NULL)
> 
> Don't know about your branch, but in head RT_LOCK_DESTROY is done
> in V_rtzone finit method. So you are going to hit destroying of
> already destroyed mutex.
> 
> btw, I see no good reason for virtualizing the zone. (Yep, I did
> that.) :)

Having virtualized zones comes quite handy while hacking up new stuff to
track VIMAGE memory leaks, but once the bugs get cleaned up and if the
V_ stuff gets commited to head, it apparently becomes difficult to
persuade people to de-virtualize...

Marko