Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[Pekka Savola]

On Thu, 17 Jul 2003 kuznet@ms2.inr.ac.ru wrote:
> Wait a second. What the hell is this in anycast.c? How is it possible
> to allow to any user to create reserved anycast?
> This makes them completely useless, everyone on LAN can join
> anycast service and blackhole it, which will prevent listening by real servers.
> 
> This cannot be right. I think the logic is illegally stolen
> from multicast interface: only superuser calls can create/delete anycasts.
> Non-superuser can only listen existing one.
> 
> I would block JOIN/LEAVE for non-superuser completely.

No user should be able to join anycast group, IMHO.

(Of course, that hasn't been specifed anywhere, but the implementations 
should do what they think is best -- and I certainly think this is.)

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[David S. Miller]

On Fri, 18 Jul 2003 00:52:03 +0400 (MSD)
kuznet@ms2.inr.ac.ru wrote:

> > No user should be able to join anycast group, IMHO.
> 
> OK. Done, the patch enclosed.

Pekka, please ACK Alexey's patch, I'd like to apply it.

Thanks.

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[YOSHIFUJI Hideaki / 吉藤英明]

In article <20030717145115.046fd5ee.davem@redhat.com> (at Thu, 17 Jul 2003 14:51:15 -0700), "David S. Miller" <davem@redhat.com> says:

> > > No user should be able to join anycast group, IMHO.
> > 
> > OK. Done, the patch enclosed.
> 
> Pekka, please ACK Alexey's patch, I'd like to apply it.

I'm not pekka, but It seems ok to me, too.

--yoshfuji

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[Pekka Savola]

On Fri, 18 Jul 2003 kuznet@ms2.inr.ac.ru wrote:
> > No user should be able to join anycast group, IMHO.
> 
> OK. Done, the patch enclosed.

Based on a quick glance, looks OK.  Better than the code we have now..

> Another rfc question: is random delay answering solicitations for
> anycast not required already?

I'm not sure whether this is what you're asking but..:

When responding to an NS with an NA w/ anycast address, the response
SHOULD be delayed by a random 0..MAX_ANYCAST_DELAY_TIME (1 by default)
seconds. (Override bit also SHOULD be set to 0.)

> # This is a BitKeeper generated patch for the following project:
> # Project Name: Linux kernel tree
> # This patch format is intended for GNU patch command version 2.5 or higher.
> # This patch includes the following deltas:
> #	           ChangeSet	1.1469  -> 1.1470 
> #	  net/ipv6/anycast.c	1.5     -> 1.6    
> #	include/net/ip6_route.h	1.10    -> 1.11   
> #	     net/ipv6/icmp.c	1.36    -> 1.37   
> #	 net/ipv6/tcp_ipv6.c	1.64    -> 1.65   
> #	    net/ipv6/ndisc.c	1.52    -> 1.53   
> #	    net/ipv6/route.c	1.50    -> 1.51   
> #	  include/net/ipv6.h	1.22    -> 1.23   
> #	 net/ipv6/addrconf.c	1.58    -> 1.59   
> #
> # The following is the BitKeeper ChangeSet Log
> # --------------------------------------------
> # 03/07/18	kuznet@oops.inr.ac.ru	1.1470
> # IPv6: sanitize anycast address support
> # --------------------------------------------
> #
> diff -Nru a/include/net/ip6_route.h b/include/net/ip6_route.h
> --- a/include/net/ip6_route.h	Fri Jul 18 00:49:43 2003
> +++ b/include/net/ip6_route.h	Fri Jul 18 00:49:43 2003
> @@ -45,7 +45,8 @@
>  					   void *rtattr);
>  
>  extern int			ip6_rt_addr_add(struct in6_addr *addr,
> -						struct net_device *dev);
> +						struct net_device *dev,
> +						int anycast);
>  
>  extern int			ip6_rt_addr_del(struct in6_addr *addr,
>  						struct net_device *dev);
> @@ -116,6 +117,13 @@
>  	np->daddr_cache = daddr;
>  	np->dst_cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
>  	write_unlock(&sk->sk_dst_lock);
> +}
> +
> +static inline int ipv6_unicast_destination(struct sk_buff *skb)
> +{
> +	struct rt6_info *rt = (struct rt6_info *) skb->dst;
> +
> +	return rt->rt6i_flags & RTF_LOCAL;
>  }
>  
>  #endif
> diff -Nru a/include/net/ipv6.h b/include/net/ipv6.h
> --- a/include/net/ipv6.h	Fri Jul 18 00:49:43 2003
> +++ b/include/net/ipv6.h	Fri Jul 18 00:49:43 2003
> @@ -51,7 +51,7 @@
>  /*
>   *	Addr type
>   *	
> - *	type	-	unicast | multicast | anycast
> + *	type	-	unicast | multicast
>   *	scope	-	local	| site	    | global
>   *	v4	-	compat
>   *	v4mapped
> @@ -63,7 +63,6 @@
>  
>  #define IPV6_ADDR_UNICAST      	0x0001U	
>  #define IPV6_ADDR_MULTICAST    	0x0002U	
> -#define IPV6_ADDR_ANYCAST	0x0004U
>  
>  #define IPV6_ADDR_LOOPBACK	0x0010U
>  #define IPV6_ADDR_LINKLOCAL	0x0020U
> diff -Nru a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
> --- a/net/ipv6/addrconf.c	Fri Jul 18 00:49:43 2003
> +++ b/net/ipv6/addrconf.c	Fri Jul 18 00:49:43 2003
> @@ -209,15 +209,8 @@
>  		};
>  		return type;
>  	}
> -	/* check for reserved anycast addresses */
> -	
> -	if ((st & htonl(0xE0000000)) &&
> -	    ((addr->s6_addr32[2] == htonl(0xFDFFFFFF) &&
> -	    (addr->s6_addr32[3] | htonl(0x7F)) == (u32)~0) ||
> -	    (addr->s6_addr32[2] == 0 && addr->s6_addr32[3] == 0)))
> -		type = IPV6_ADDR_ANYCAST;
> -	else
> -		type = IPV6_ADDR_UNICAST;
> +
> +	type = IPV6_ADDR_UNICAST;
>  
>  	/* Consider all addresses with the first three bits different of
>  	   000 and 111 as finished.
> @@ -2552,7 +2545,7 @@
>  
>  	switch (event) {
>  	case RTM_NEWADDR:
> -		ip6_rt_addr_add(&ifp->addr, ifp->idev->dev);
> +		ip6_rt_addr_add(&ifp->addr, ifp->idev->dev, 0);
>  		break;
>  	case RTM_DELADDR:
>  		addrconf_leave_solict(ifp->idev->dev, &ifp->addr);
> diff -Nru a/net/ipv6/anycast.c b/net/ipv6/anycast.c
> --- a/net/ipv6/anycast.c	Fri Jul 18 00:49:43 2003
> +++ b/net/ipv6/anycast.c	Fri Jul 18 00:49:43 2003
> @@ -96,7 +96,6 @@
>  	return onlink;
>  }
>  
> -
>  /*
>   *	socket join an anycast group
>   */
> @@ -110,8 +109,12 @@
>  	int	ishost = !ipv6_devconf.forwarding;
>  	int	err = 0;
>  
> +	if (!capable(CAP_NET_ADMIN))
> +		return -EPERM;
>  	if (ipv6_addr_type(addr) & IPV6_ADDR_MULTICAST)
>  		return -EINVAL;
> +	if (ipv6_chk_addr(addr, NULL))
> +		return -EINVAL;
>  
>  	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
>  	if (pac == NULL)
> @@ -161,21 +164,12 @@
>  	 * For hosts, allow link-local or matching prefix anycasts.
>  	 * This obviates the need for propagating anycast routes while
>  	 * still allowing some non-router anycast participation.
> -	 *
> -	 * allow anyone to join anycasts that don't require a special route
> -	 * and can't be spoofs of unicast addresses (reserved anycast only)
>  	 */
>  	if (!ip6_onlink(addr, dev)) {
>  		if (ishost)
>  			err = -EADDRNOTAVAIL;
> -		else if (!capable(CAP_NET_ADMIN))
> -			err = -EPERM;
>  		if (err)
>  			goto out_dev_put;
> -	} else if (!(ipv6_addr_type(addr) & IPV6_ADDR_ANYCAST) &&
> -		   !capable(CAP_NET_ADMIN)) {
> -		err = -EPERM;
> -		goto out_dev_put;
>  	}
>  
>  	err = ipv6_dev_ac_inc(dev, addr);
> @@ -266,6 +260,13 @@
>  		dev_put(dev);
>  }
>  
> +#if 0
> +/* The function is not used, which is funny. Apparently, author
> + * supposed to use it to filter out datagrams inside udp/raw but forgot.
> + *
> + * It is OK, anycasts are not special comparing to delivery to unicasts.
> + */
> +
>  int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex)
>  {
>  	struct ipv6_ac_socklist *pac;
> @@ -286,6 +287,8 @@
>  	return found;
>  }
>  
> +#endif
> +
>  static void aca_put(struct ifacaddr6 *ac)
>  {
>  	if (atomic_dec_and_test(&ac->aca_refcnt)) {
> @@ -347,7 +350,7 @@
>  	idev->ac_list = aca;
>  	write_unlock_bh(&idev->lock);
>  
> -	ip6_rt_addr_add(&aca->aca_addr, dev);
> +	ip6_rt_addr_add(&aca->aca_addr, dev, 1);
>  
>  	addrconf_join_solict(dev, &aca->aca_addr);
>  
> diff -Nru a/net/ipv6/icmp.c b/net/ipv6/icmp.c
> --- a/net/ipv6/icmp.c	Fri Jul 18 00:49:43 2003
> +++ b/net/ipv6/icmp.c	Fri Jul 18 00:49:43 2003
> @@ -415,8 +415,7 @@
>  
>  	saddr = &skb->nh.ipv6h->daddr;
>  
> -	if (ipv6_addr_type(saddr) & IPV6_ADDR_MULTICAST ||
> -	    ipv6_chk_acast_addr(0, saddr)) 
> +	if (!ipv6_unicast_destination(skb))
>  		saddr = NULL;
>  
>  	memcpy(&tmp_hdr, icmph, sizeof(tmp_hdr));
> diff -Nru a/net/ipv6/ndisc.c b/net/ipv6/ndisc.c
> --- a/net/ipv6/ndisc.c	Fri Jul 18 00:49:43 2003
> +++ b/net/ipv6/ndisc.c	Fri Jul 18 00:49:43 2003
> @@ -785,8 +785,7 @@
>  			ipv6_addr_all_nodes(&maddr);
>  			ndisc_send_na(dev, NULL, &maddr, &ifp->addr, 
>  				      ifp->idev->cnf.forwarding, 0, 
> -				      ipv6_addr_type(&ifp->addr)&IPV6_ADDR_ANYCAST ? 0 : 1, 
> -				      1);
> +				      1, 1);
>  			in6_ifa_put(ifp);
>  			return;
>  		}
> @@ -809,8 +808,7 @@
>  			if (neigh || !dev->hard_header) {
>  				ndisc_send_na(dev, neigh, saddr, &ifp->addr, 
>  					      ifp->idev->cnf.forwarding, 1, 
> -					      ipv6_addr_type(&ifp->addr)&IPV6_ADDR_ANYCAST ? 0 : 1, 
> -					      1);
> +					      1, 1);
>  				if (neigh)
>  					neigh_release(neigh);
>  			}
> diff -Nru a/net/ipv6/route.c b/net/ipv6/route.c
> --- a/net/ipv6/route.c	Fri Jul 18 00:49:43 2003
> +++ b/net/ipv6/route.c	Fri Jul 18 00:49:43 2003
> @@ -1256,7 +1256,7 @@
>   *	Add address
>   */
>  
> -int ip6_rt_addr_add(struct in6_addr *addr, struct net_device *dev)
> +int ip6_rt_addr_add(struct in6_addr *addr, struct net_device *dev, int anycast)
>  {
>  	struct rt6_info *rt = ip6_dst_alloc();
>  
> @@ -1275,6 +1275,8 @@
>  	rt->u.dst.obsolete = -1;
>  
>  	rt->rt6i_flags = RTF_UP | RTF_NONEXTHOP;
> +	if (!anycast)
> +		rt->rt6i_flags |= RTF_LOCAL;
>  	rt->rt6i_nexthop = ndisc_get_neigh(rt->rt6i_dev, &rt->rt6i_gateway);
>  	if (rt->rt6i_nexthop == NULL) {
>  		dst_free((struct dst_entry *) rt);
> diff -Nru a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
> --- a/net/ipv6/tcp_ipv6.c	Fri Jul 18 00:49:43 2003
> +++ b/net/ipv6/tcp_ipv6.c	Fri Jul 18 00:49:43 2003
> @@ -971,7 +971,7 @@
>  	if (th->rst)
>  		return;
>  
> -	if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr))
> +	if (!ipv6_unicast_destination(skb))
>  		return; 
>  
>  	/*
> @@ -1175,8 +1175,7 @@
>  	if (skb->protocol == htons(ETH_P_IP))
>  		return tcp_v4_conn_request(sk, skb);
>  
> -	/* FIXME: do the same check for anycast */
> -	if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr))
> +	if (!ipv6_unicast_destination(skb))
>  		goto drop; 
>  
>  	/*
> 

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[David S. Miller]

On Fri, 18 Jul 2003 00:52:03 +0400 (MSD)
kuznet@ms2.inr.ac.ru wrote:

> Done, the patch enclosed.

Applied, thanks everyone.

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[Mika Liljeberg]

On Thu, 2003-07-17 at 12:06, kuznet@ms2.inr.ac.ru wrote:
> They cannot intersect, otherwise RTF_LOCAL thing will not work.
> 
> I deliberately blocked attempt to add a local address as anycast
> in anycast.c, see another chunk.

Ok, I missed that one.

I guess it's safe to assume that the anycast and unicast spaces will not
intersect, even though the addresses are allocated from the same range.
I was wondering how to dynamically assign anycast addresses. In theory
one could abuse the unicast address assignment mechanisms (in the
absence of anything else users might be tempted to try this). But that's
a different issue.

	MikaL

Re: Anycast usage, final diagnosis? (was: IPv6: Fix broken anycast

[kuznet]

Hello!

> I'm not sure you can just remove these. It seems possible (?) to have
> the anycast address configured on one of the interfaces as a unicast at
> the same time. I.e., one of the anycast members could own the address.

They cannot intersect, otherwise RTF_LOCAL thing will not work.

I deliberately blocked attempt to add a local address as anycast
in anycast.c, see another chunk. But even that check is not necessary:
non-superuser may listen only for reserved unicasts, which are
excluded from allowed local addresses by policy. Kernel does not need even
to worry about this.

Actually, the test in ndisc.c was bogus by another reason:
inet_addr_type() checks only for reserved anycasts and non-reserved unicasts,
which would conflict with local addresses, were not detected in any case.

Alexey