netdev - Re: [Patch net] ipv4: fib: check forwarding before checking send

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.LFD.2.11.1404091027460.1435@ja.home.ssi.bg>
Date:	Wed, 9 Apr 2014 11:03:31 +0300 (EEST)
From:	Julian Anastasov <ja@....bg>
To:	Cong Wang <xiyou.wangcong@...il.com>
cc:	David Miller <davem@...emloft.net>,
	Linux Kernel Network Developers <netdev@...r.kernel.org>,
	"Eric W. Biederman" <ebiederm@...ssion.com>,
	Cong Wang <cwang@...pensource.com>
Subject: Re: [Patch net] ipv4: fib: check forwarding before checking
 send_redirects


	Hello,

On Tue, 8 Apr 2014, Cong Wang wrote:

> >         Yes, the code will fail if input route is
> > tried on loopback device. My first thought was that
> > you forgot to 'ip route flush cache' after clearing
> > rp_filter.
> >
> 
> route cache is removed after 3.6, so I don't need to do it?

	route cache is now FIB cache :)

	My 'ip' tool still does

open("/proc/sys/net/ipv4/route/flush", O_WRONLY) = 4
write(4, "-1", 2) = 2
close(4)          = 0

	Which invalidates the FIB cache:
ipv4_sysctl_rtcache_flush -> rt_cache_flush, fnhe_genid_bump

	The change is noticed via rt_cache_valid and
rt_is_expired.

> >         What kind of packet is that and why the skb lost its
> > output route and now it is trying for input route?
> >
> 
> Yeah, I just noticed that for normal loopback packets its ->dst
> are forced to be kept so that it will not need to lookup route
> again on rx. However, in our case, we redirect the packets
> from veth0 to lo when src_addr == dst_addr, its ->dst is dropped
> after coming out of the network namespace, therefore goes into
> the slow path to validate the reverse path and lookup the route.

	I'm still not sure that it is valid to process
input route on loopback device. There is also a case
of broadcasts/multicasts that should preserve their local output
route while being looped in ip_mc_output(). In this
case device is not loopback. I need more time to check
the code again...

> I think David's point is still valid in this case, __fib_validate_source()
> should not reject the packets. So hmm... the only reason I can find is
> the following code:
> 
>         if (fib_lookup(net, &fl4, &res) == 0) {
>                 if (res.type == RTN_UNICAST)
>                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
>         }

	I think fib_validate_source is called with
oif=LOOPBACK_IFINDEX and dev/idev=lo for skb without route,
that is the problem. In this case fib_validate_source
thinks this is forwarding from lo to lo. So, there is
a reason your patch fixes the problem, alternative
is to add oif == LOOPBACK_IFINDEX check:

	dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev) ||
	oif == LOOPBACK_IFINDEX

> where its nh_scope == RT_SCOPE_HOST?

	The idea is to avoid calling __fib_validate_source,
right? No need to look into __fib_validate_source. Also,
I'm not sure how legal is to receive packets on lo without
skb->dst...

Regards

--
Julian Anastasov <ja@....bg>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html