Date: Tue, 24 Feb 2004 13:20:18 +0100 From: Andre Oppermann <andre@freebsd.org> To: Gleb Smirnoff <glebius@cell.sick.ru> Cc: freebsd-net@freebsd.org Subject: Re: rtalloc()/rtfree() problems on CURRENT Message-ID: <403B4182.F6BD7101@freebsd.org> References: <20040224080353.GA76272@cell.sick.ru> <403B2423.DABF2E48@freebsd.org> <20040224102113.GB77406@cell.sick.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
Gleb Smirnoff wrote:
>
> Dear Andre,
>
> first, thanks for your feedback.
>
> On Tue, Feb 24, 2004 at 11:14:59AM +0100, Andre Oppermann wrote:
> A> Most of the time, if you don't need a cloned route allocated, you can
> A> just use rtalloc_ign(&ro, RTF_CLONING) to just get a reference to the
> A> existing rtentry. Not allocating a new cloned route saves work and
> A> time for the rtalloc code.
>
> Thanks for this hint. So if using rtalloc_ign() I do not need to call rtfree()?
You still have to call rtfree() to decrement the refcount of the rtentry.
> Why does this situation is not clear in manpage? :)
I should update it.
> A> > if (ro.ro_rt != NULL) {
> A> > struct rtentry *rt = ro.ro_rt;
> A> >
> A> > /* here some read-only things are done with rt */
> A>
> A> Maybe you do some nasty things to rt here?
>
> Really, I don't :) Exact code is given at end of the mail.
You indeed do some nasty things.
> A> > rtfree(ro.ro_rt);
> A> > }
> A> >
> A> > But on CURRENT rtfree() causes panic. Here is backtrace:
> A> ...
> A> > Does this mean that something is broken in CURRENT or I am doing something wrong?
> A>
> A> rtalloc() and rtfree() are used quite often in the network code and
> A> otherwise work fine. The odds are high that you are doing something
> A> wrong.
>
> Here is exact code:
>
> bzero((caddr_t)&ro, sizeof(ro));
> sin = (struct sockaddr_in *)&ro.ro_dst;
> sin->sin_len = sizeof(*sin);
> sin->sin_family = AF_INET;
> sin->sin_addr = fle->r.r_dst;
> rtalloc(&ro);
> if (ro.ro_rt != NULL) {
> struct rtentry *rt = ro.ro_rt;
>
> /* This is cloned route, use its parent */
> if (ro.ro_rt->rt_flags & RTF_WASCLONED &&
> ro.ro_rt->rt_parent)
> rt = ro.ro_rt->rt_parent;
With rtalloc() you get a cloned route essentially every time. Use
rtalloc_ign(&ro, RTF_CLONING) to get the parent route directly.
> fle->o_ifx = rt->rt_ifp->if_index;
>
> if (rt->rt_flags & RTF_GATEWAY &&
> rt->rt_gateway->sa_family == AF_INET)
> fle->next_hop =
> ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr;
>
> if (rt_mask(rt))
> fle->dst_mask =
> bit_count(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
> else if (rt->rt_flags & RTF_HOST)
> /* Give up. We can't determine mask :( */
> fle->dst_mask = 32;
>
> rtfree(ro.ro_rt);
Use the macro RTFREE() instead of rtfree(), it will take care of some
locking issues.
> }
--
Andre
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?403B4182.F6BD7101>