[linux-2.6-block.git] / net / ipv6 / anycast.c

// SPDX-License-Identifier: GPL-2.0-or-later
/*
 *	Anycast support for IPv6
 *	Linux INET6 implementation
 *
 *	Authors:
 *	David L Stevens (dlstevens@us.ibm.com)
 *
 *	based heavily on net/ipv6/mcast.c
 */

#include <linux/capability.h>
#include <linux/module.h>
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/random.h>
#include <linux/string.h>
#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/net.h>
#include <linux/in6.h>
#include <linux/netdevice.h>
#include <linux/if_arp.h>
#include <linux/route.h>
#include <linux/init.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
#include <linux/slab.h>

#include <net/net_namespace.h>
#include <net/sock.h>
#include <net/snmp.h>

#include <net/ipv6.h>
#include <net/protocol.h>
#include <net/if_inet6.h>
#include <net/ndisc.h>
#include <net/addrconf.h>
#include <net/ip6_route.h>

#include <net/checksum.h>

#define IN6_ADDR_HSIZE_SHIFT	8
#define IN6_ADDR_HSIZE		BIT(IN6_ADDR_HSIZE_SHIFT)
/*	anycast address hash table
 */
static struct hlist_head inet6_acaddr_lst[IN6_ADDR_HSIZE];
static DEFINE_SPINLOCK(acaddr_hash_lock);

static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr);

static u32 inet6_acaddr_hash(struct net *net, const struct in6_addr *addr)
{
	u32 val = ipv6_addr_hash(addr) ^ net_hash_mix(net);

	return hash_32(val, IN6_ADDR_HSIZE_SHIFT);
}

/*
 *	socket join an anycast group
 */

int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct net_device *dev = NULL;
	struct inet6_dev *idev;
	struct ipv6_ac_socklist *pac;
	struct net *net = sock_net(sk);
	int	ishost = !net->ipv6.devconf_all->forwarding;
	int	err = 0;

	ASSERT_RTNL();

	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
		return -EPERM;
	if (ipv6_addr_is_multicast(addr))
		return -EINVAL;

	if (ifindex)
		dev = __dev_get_by_index(net, ifindex);

	if (ipv6_chk_addr_and_flags(net, addr, dev, true, 0, IFA_F_TENTATIVE))
		return -EINVAL;

	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
	if (!pac)
		return -ENOMEM;
	pac->acl_next = NULL;
	pac->acl_addr = *addr;

	if (ifindex == 0) {
		struct rt6_info *rt;

		rt = rt6_lookup(net, addr, NULL, 0, NULL, 0);
		if (rt) {
			dev = rt->dst.dev;
			ip6_rt_put(rt);
		} else if (ishost) {
			err = -EADDRNOTAVAIL;
			goto error;
		} else {
			/* router, no matching interface: just pick one */
			dev = __dev_get_by_flags(net, IFF_UP,
						 IFF_UP | IFF_LOOPBACK);
		}
	}

	if (!dev) {
		err = -ENODEV;
		goto error;
	}

	idev = __in6_dev_get(dev);
	if (!idev) {
		if (ifindex)
			err = -ENODEV;
		else
			err = -EADDRNOTAVAIL;
		goto error;
	}
	/* reset ishost, now that we have a specific device */
	ishost = !idev->cnf.forwarding;

	pac->acl_ifindex = dev->ifindex;

	/* XXX
	 * 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.
	 */
	if (!ipv6_chk_prefix(addr, dev)) {
		if (ishost)
			err = -EADDRNOTAVAIL;
		if (err)
			goto error;
	}

	err = __ipv6_dev_ac_inc(idev, addr);
	if (!err) {
		pac->acl_next = np->ipv6_ac_list;
		np->ipv6_ac_list = pac;
		pac = NULL;
	}

error:
	if (pac)
		sock_kfree_s(sk, pac, sizeof(*pac));
	return err;
}

/*
 *	socket leave an anycast group
 */
int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct net_device *dev;
	struct ipv6_ac_socklist *pac, *prev_pac;
	struct net *net = sock_net(sk);

	ASSERT_RTNL();

	prev_pac = NULL;
	for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
		if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
		     ipv6_addr_equal(&pac->acl_addr, addr))
			break;
		prev_pac = pac;
	}
	if (!pac)
		return -ENOENT;
	if (prev_pac)
		prev_pac->acl_next = pac->acl_next;
	else
		np->ipv6_ac_list = pac->acl_next;

	dev = __dev_get_by_index(net, pac->acl_ifindex);
	if (dev)
		ipv6_dev_ac_dec(dev, &pac->acl_addr);

	sock_kfree_s(sk, pac, sizeof(*pac));
	return 0;
}

void ipv6_sock_ac_close(struct sock *sk)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct net_device *dev = NULL;
	struct ipv6_ac_socklist *pac;
	struct net *net = sock_net(sk);
	int	prev_index;

	if (!np->ipv6_ac_list)
		return;

	rtnl_lock();
	pac = np->ipv6_ac_list;
	np->ipv6_ac_list = NULL;

	prev_index = 0;
	while (pac) {
		struct ipv6_ac_socklist *next = pac->acl_next;

		if (pac->acl_ifindex != prev_index) {
			dev = __dev_get_by_index(net, pac->acl_ifindex);
			prev_index = pac->acl_ifindex;
		}
		if (dev)
			ipv6_dev_ac_dec(dev, &pac->acl_addr);
		sock_kfree_s(sk, pac, sizeof(*pac));
		pac = next;
	}
	rtnl_unlock();
}

static void ipv6_add_acaddr_hash(struct net *net, struct ifacaddr6 *aca)
{
	unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr);

	spin_lock(&acaddr_hash_lock);
	hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]);
	spin_unlock(&acaddr_hash_lock);
}

static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca)
{
	spin_lock(&acaddr_hash_lock);
	hlist_del_init_rcu(&aca->aca_addr_lst);
	spin_unlock(&acaddr_hash_lock);
}

static void aca_get(struct ifacaddr6 *aca)
{
	refcount_inc(&aca->aca_refcnt);
}

static void aca_free_rcu(struct rcu_head *h)
{
	struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu);

	fib6_info_release(aca->aca_rt);
	kfree(aca);
}

static void aca_put(struct ifacaddr6 *ac)
{
	if (refcount_dec_and_test(&ac->aca_refcnt)) {
		call_rcu(&ac->rcu, aca_free_rcu);
	}
}

static struct ifacaddr6 *aca_alloc(struct fib6_info *f6i,
				   const struct in6_addr *addr)
{
	struct ifacaddr6 *aca;

	aca = kzalloc(sizeof(*aca), GFP_ATOMIC);
	if (!aca)
		return NULL;

	aca->aca_addr = *addr;
	fib6_info_hold(f6i);
	aca->aca_rt = f6i;
	INIT_HLIST_NODE(&aca->aca_addr_lst);
	aca->aca_users = 1;
	/* aca_tstamp should be updated upon changes */
	aca->aca_cstamp = aca->aca_tstamp = jiffies;
	refcount_set(&aca->aca_refcnt, 1);

	return aca;
}

/*
 *	device anycast group inc (add if not found)
 */
int __ipv6_dev_ac_inc(struct inet6_dev *idev, const struct in6_addr *addr)
{
	struct ifacaddr6 *aca;
	struct fib6_info *f6i;
	struct net *net;
	int err;

	ASSERT_RTNL();

	write_lock_bh(&idev->lock);
	if (idev->dead) {
		err = -ENODEV;
		goto out;
	}

	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
		if (ipv6_addr_equal(&aca->aca_addr, addr)) {
			aca->aca_users++;
			err = 0;
			goto out;
		}
	}

	net = dev_net(idev->dev);
	f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC);
	if (IS_ERR(f6i)) {
		err = PTR_ERR(f6i);
		goto out;
	}
	aca = aca_alloc(f6i, addr);
	if (!aca) {
		fib6_info_release(f6i);
		err = -ENOMEM;
		goto out;
	}

	aca->aca_next = idev->ac_list;
	idev->ac_list = aca;

	/* Hold this for addrconf_join_solict() below before we unlock,
	 * it is already exposed via idev->ac_list.
	 */
	aca_get(aca);
	write_unlock_bh(&idev->lock);

	ipv6_add_acaddr_hash(net, aca);

	ip6_ins_rt(net, f6i);

	addrconf_join_solict(idev->dev, &aca->aca_addr);

	aca_put(aca);
	return 0;
out:
	write_unlock_bh(&idev->lock);
	return err;
}

/*
 *	device anycast group decrement
 */
int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
{
	struct ifacaddr6 *aca, *prev_aca;

	ASSERT_RTNL();

	write_lock_bh(&idev->lock);
	prev_aca = NULL;
	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
		if (ipv6_addr_equal(&aca->aca_addr, addr))
			break;
		prev_aca = aca;
	}
	if (!aca) {
		write_unlock_bh(&idev->lock);
		return -ENOENT;
	}
	if (--aca->aca_users > 0) {
		write_unlock_bh(&idev->lock);
		return 0;
	}
	if (prev_aca)
		prev_aca->aca_next = aca->aca_next;
	else
		idev->ac_list = aca->aca_next;
	write_unlock_bh(&idev->lock);
	ipv6_del_acaddr_hash(aca);
	addrconf_leave_solict(idev, &aca->aca_addr);

	ip6_del_rt(dev_net(idev->dev), aca->aca_rt);

	aca_put(aca);
	return 0;
}

/* called with rtnl_lock() */
static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
{
	struct inet6_dev *idev = __in6_dev_get(dev);

	if (!idev)
		return -ENODEV;
	return __ipv6_dev_ac_dec(idev, addr);
}

void ipv6_ac_destroy_dev(struct inet6_dev *idev)
{
	struct ifacaddr6 *aca;

	write_lock_bh(&idev->lock);
	while ((aca = idev->ac_list) != NULL) {
		idev->ac_list = aca->aca_next;
		write_unlock_bh(&idev->lock);

		ipv6_del_acaddr_hash(aca);

		addrconf_leave_solict(idev, &aca->aca_addr);

		ip6_del_rt(dev_net(idev->dev), aca->aca_rt);

		aca_put(aca);

		write_lock_bh(&idev->lock);
	}
	write_unlock_bh(&idev->lock);
}

/*
 *	check if the interface has this anycast address
 *	called with rcu_read_lock()
 */
static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
{
	struct inet6_dev *idev;
	struct ifacaddr6 *aca;

	idev = __in6_dev_get(dev);
	if (idev) {
		read_lock_bh(&idev->lock);
		for (aca = idev->ac_list; aca; aca = aca->aca_next)
			if (ipv6_addr_equal(&aca->aca_addr, addr))
				break;
		read_unlock_bh(&idev->lock);
		return aca != NULL;
	}
	return false;
}

/*
 *	check if given interface (or any, if dev==0) has this anycast address
 */
bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
			 const struct in6_addr *addr)
{
	struct net_device *nh_dev;
	struct ifacaddr6 *aca;
	bool found = false;

	rcu_read_lock();
	if (dev)
		found = ipv6_chk_acast_dev(dev, addr);
	else {
		unsigned int hash = inet6_acaddr_hash(net, addr);

		hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash],
					 aca_addr_lst) {
			nh_dev = fib6_info_nh_dev(aca->aca_rt);
			if (!nh_dev || !net_eq(dev_net(nh_dev), net))
				continue;
			if (ipv6_addr_equal(&aca->aca_addr, addr)) {
				found = true;
				break;
			}
		}
	}
	rcu_read_unlock();
	return found;
}

/*	check if this anycast address is link-local on given interface or
 *	is global
 */
bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev,
			     const struct in6_addr *addr)
{
	return ipv6_chk_acast_addr(net,
				   (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
				    dev : NULL),
				   addr);
}

#ifdef CONFIG_PROC_FS
struct ac6_iter_state {
	struct seq_net_private p;
	struct net_device *dev;
	struct inet6_dev *idev;
};

#define ac6_seq_private(seq)	((struct ac6_iter_state *)(seq)->private)

static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
{
	struct ifacaddr6 *im = NULL;
	struct ac6_iter_state *state = ac6_seq_private(seq);
	struct net *net = seq_file_net(seq);

	state->idev = NULL;
	for_each_netdev_rcu(net, state->dev) {
		struct inet6_dev *idev;
		idev = __in6_dev_get(state->dev);
		if (!idev)
			continue;
		read_lock_bh(&idev->lock);
		im = idev->ac_list;
		if (im) {
			state->idev = idev;
			break;
		}
		read_unlock_bh(&idev->lock);
	}
	return im;
}

static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
{
	struct ac6_iter_state *state = ac6_seq_private(seq);

	im = im->aca_next;
	while (!im) {
		if (likely(state->idev != NULL))
			read_unlock_bh(&state->idev->lock);

		state->dev = next_net_device_rcu(state->dev);
		if (!state->dev) {
			state->idev = NULL;
			break;
		}
		state->idev = __in6_dev_get(state->dev);
		if (!state->idev)
			continue;
		read_lock_bh(&state->idev->lock);
		im = state->idev->ac_list;
	}
	return im;
}

static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
{
	struct ifacaddr6 *im = ac6_get_first(seq);
	if (im)
		while (pos && (im = ac6_get_next(seq, im)) != NULL)
			--pos;
	return pos ? NULL : im;
}

static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
	__acquires(RCU)
{
	rcu_read_lock();
	return ac6_get_idx(seq, *pos);
}

static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
	struct ifacaddr6 *im = ac6_get_next(seq, v);

	++*pos;
	return im;
}

static void ac6_seq_stop(struct seq_file *seq, void *v)
	__releases(RCU)
{
	struct ac6_iter_state *state = ac6_seq_private(seq);

	if (likely(state->idev != NULL)) {
		read_unlock_bh(&state->idev->lock);
		state->idev = NULL;
	}
	rcu_read_unlock();
}

static int ac6_seq_show(struct seq_file *seq, void *v)
{
	struct ifacaddr6 *im = (struct ifacaddr6 *)v;
	struct ac6_iter_state *state = ac6_seq_private(seq);

	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
		   state->dev->ifindex, state->dev->name,
		   &im->aca_addr, im->aca_users);
	return 0;
}

static const struct seq_operations ac6_seq_ops = {
	.start	=	ac6_seq_start,
	.next	=	ac6_seq_next,
	.stop	=	ac6_seq_stop,
	.show	=	ac6_seq_show,
};

int __net_init ac6_proc_init(struct net *net)
{
	if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops,
			sizeof(struct ac6_iter_state)))
		return -ENOMEM;

	return 0;
}

void ac6_proc_exit(struct net *net)
{
	remove_proc_entry("anycast6", net->proc_net);
}
#endif

/*	Init / cleanup code
 */
int __init ipv6_anycast_init(void)
{
	int i;

	for (i = 0; i < IN6_ADDR_HSIZE; i++)
		INIT_HLIST_HEAD(&inet6_acaddr_lst[i]);
	return 0;
}

void ipv6_anycast_cleanup(void)
{
	int i;

	spin_lock(&acaddr_hash_lock);
	for (i = 0; i < IN6_ADDR_HSIZE; i++)
		WARN_ON(!hlist_empty(&inet6_acaddr_lst[i]));
	spin_unlock(&acaddr_hash_lock);
}
Commit	Line	Data
2874c5fd	1	// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4 LT	2	/*
1da177e4 LT	3	* Anycast support for IPv6
1ab1457c	4	* Linux INET6 implementation
1da177e4 LT	5	*
	6	* Authors:
	7	* David L Stevens (dlstevens@us.ibm.com)
	8	*
	9	* based heavily on net/ipv6/mcast.c
1da177e4 LT	10	*/
1da177e4 LT	11
4fc268d2	12	#include <linux/capability.h>
1da177e4 LT	13	#include <linux/module.h>
	14	#include <linux/errno.h>
	15	#include <linux/types.h>
	16	#include <linux/random.h>
	17	#include <linux/string.h>
	18	#include <linux/socket.h>
	19	#include <linux/sockios.h>
1da177e4 LT	20	#include <linux/net.h>
	21	#include <linux/in6.h>
	22	#include <linux/netdevice.h>
	23	#include <linux/if_arp.h>
	24	#include <linux/route.h>
	25	#include <linux/init.h>
	26	#include <linux/proc_fs.h>
	27	#include <linux/seq_file.h>
5a0e3ad6	28	#include <linux/slab.h>
1da177e4	29
457c4cbc	30	#include <net/net_namespace.h>
1da177e4 LT	31	#include <net/sock.h>
	32	#include <net/snmp.h>
	33
	34	#include <net/ipv6.h>
	35	#include <net/protocol.h>
	36	#include <net/if_inet6.h>
	37	#include <net/ndisc.h>
	38	#include <net/addrconf.h>
	39	#include <net/ip6_route.h>
	40
	41	#include <net/checksum.h>
	42
2384d025 JB	43	#define IN6_ADDR_HSIZE_SHIFT 8
	44	#define IN6_ADDR_HSIZE BIT(IN6_ADDR_HSIZE_SHIFT)
	45	/* anycast address hash table
	46	*/
	47	static struct hlist_head inet6_acaddr_lst[IN6_ADDR_HSIZE];
	48	static DEFINE_SPINLOCK(acaddr_hash_lock);
	49
b71d1d42	50	static int ipv6_dev_ac_dec(struct net_device dev, const struct in6_addr addr);
1da177e4	51
2384d025 JB	52	static u32 inet6_acaddr_hash(struct net net, const struct in6_addr addr)
	53	{
	54	u32 val = ipv6_addr_hash(addr) ^ net_hash_mix(net);
	55
	56	return hash_32(val, IN6_ADDR_HSIZE_SHIFT);
	57	}
	58
1da177e4 LT	59	/*
	60	* socket join an anycast group
	61	*/
	62
b71d1d42	63	int ipv6_sock_ac_join(struct sock sk, int ifindex, const struct in6_addr addr)
1da177e4 LT	64	{
	65	struct ipv6_pinfo *np = inet6_sk(sk);
	66	struct net_device *dev = NULL;
	67	struct inet6_dev *idev;
	68	struct ipv6_ac_socklist *pac;
6ab57e7e	69	struct net *net = sock_net(sk);
53b7997f	70	int ishost = !net->ipv6.devconf_all->forwarding;
1da177e4 LT	71	int err = 0;
1da177e4 LT	72
c4a6853d MRL	73	ASSERT_RTNL();
c4a6853d MRL	74
af31f412	75	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1da177e4 LT	76	return -EPERM;
	77	if (ipv6_addr_is_multicast(addr))
	78	return -EINVAL;
232378e8 DA	79
	80	if (ifindex)
	81	dev = __dev_get_by_index(net, ifindex);
	82
	83	if (ipv6_chk_addr_and_flags(net, addr, dev, true, 0, IFA_F_TENTATIVE))
1da177e4 LT	84	return -EINVAL;
	85
	86	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
63159f29	87	if (!pac)
1da177e4 LT	88	return -ENOMEM;
1da177e4 LT	89	pac->acl_next = NULL;
4e3fd7a0	90	pac->acl_addr = *addr;
1da177e4 LT	91
	92	if (ifindex == 0) {
	93	struct rt6_info *rt;
	94
b75cc8f9	95	rt = rt6_lookup(net, addr, NULL, 0, NULL, 0);
1da177e4	96	if (rt) {
d1918542	97	dev = rt->dst.dev;
94e187c0	98	ip6_rt_put(rt);
1da177e4 LT	99	} else if (ishost) {
1da177e4 LT	100	err = -EADDRNOTAVAIL;
bb69ae04	101	goto error;
1da177e4 LT	102	} else {
1da177e4 LT	103	/* router, no matching interface: just pick one */
6c555490 WC	104	dev = __dev_get_by_flags(net, IFF_UP,
6c555490 WC	105	IFF_UP \| IFF_LOOPBACK);
1da177e4	106	}
232378e8	107	}
1da177e4	108
63159f29	109	if (!dev) {
1da177e4	110	err = -ENODEV;
bb69ae04	111	goto error;
1da177e4 LT	112	}
1da177e4 LT	113
bb69ae04	114	idev = __in6_dev_get(dev);
1da177e4 LT	115	if (!idev) {
	116	if (ifindex)
	117	err = -ENODEV;
	118	else
	119	err = -EADDRNOTAVAIL;
bb69ae04	120	goto error;
1da177e4 LT	121	}
	122	/* reset ishost, now that we have a specific device */
	123	ishost = !idev->cnf.forwarding;
1da177e4 LT	124
	125	pac->acl_ifindex = dev->ifindex;
	126
	127	/* XXX
	128	* For hosts, allow link-local or matching prefix anycasts.
	129	* This obviates the need for propagating anycast routes while
	130	* still allowing some non-router anycast participation.
	131	*/
52eeeb84	132	if (!ipv6_chk_prefix(addr, dev)) {
1da177e4 LT	133	if (ishost)
	134	err = -EADDRNOTAVAIL;
	135	if (err)
bb69ae04	136	goto error;
1da177e4 LT	137	}
1da177e4 LT	138
013b4d90	139	err = __ipv6_dev_ac_inc(idev, addr);
bb69ae04	140	if (!err) {
bb69ae04 ED	141	pac->acl_next = np->ipv6_ac_list;
bb69ae04 ED	142	np->ipv6_ac_list = pac;
bb69ae04 ED	143	pac = NULL;
bb69ae04 ED	144	}
1da177e4	145
bb69ae04	146	error:
bb69ae04 ED	147	if (pac)
bb69ae04 ED	148	sock_kfree_s(sk, pac, sizeof(*pac));
1da177e4 LT	149	return err;
	150	}
	151
	152	/*
	153	* socket leave an anycast group
	154	*/
b71d1d42	155	int ipv6_sock_ac_drop(struct sock sk, int ifindex, const struct in6_addr addr)
1da177e4 LT	156	{
	157	struct ipv6_pinfo *np = inet6_sk(sk);
	158	struct net_device *dev;
	159	struct ipv6_ac_socklist pac, prev_pac;
6ab57e7e	160	struct net *net = sock_net(sk);
1da177e4	161
c4a6853d MRL	162	ASSERT_RTNL();
c4a6853d MRL	163
1da177e4 LT	164	prev_pac = NULL;
	165	for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
	166	if ((ifindex == 0 \|\| pac->acl_ifindex == ifindex) &&
	167	ipv6_addr_equal(&pac->acl_addr, addr))
	168	break;
	169	prev_pac = pac;
	170	}
c4a6853d	171	if (!pac)
1da177e4	172	return -ENOENT;
1da177e4 LT	173	if (prev_pac)
	174	prev_pac->acl_next = pac->acl_next;
	175	else
	176	np->ipv6_ac_list = pac->acl_next;
	177
6c555490	178	dev = __dev_get_by_index(net, pac->acl_ifindex);
bb69ae04	179	if (dev)
1da177e4	180	ipv6_dev_ac_dec(dev, &pac->acl_addr);
bb69ae04	181
1da177e4 LT	182	sock_kfree_s(sk, pac, sizeof(*pac));
	183	return 0;
	184	}
	185
	186	void ipv6_sock_ac_close(struct sock *sk)
	187	{
	188	struct ipv6_pinfo *np = inet6_sk(sk);
	189	struct net_device *dev = NULL;
	190	struct ipv6_ac_socklist *pac;
6ab57e7e	191	struct net *net = sock_net(sk);
1da177e4 LT	192	int prev_index;
1da177e4 LT	193
0e1efe9d ED	194	if (!np->ipv6_ac_list)
	195	return;
	196
b03a9c04	197	rtnl_lock();
1da177e4 LT	198	pac = np->ipv6_ac_list;
1da177e4 LT	199	np->ipv6_ac_list = NULL;
1da177e4 LT	200
	201	prev_index = 0;
	202	while (pac) {
	203	struct ipv6_ac_socklist *next = pac->acl_next;
	204
	205	if (pac->acl_ifindex != prev_index) {
6c555490	206	dev = __dev_get_by_index(net, pac->acl_ifindex);
1da177e4 LT	207	prev_index = pac->acl_ifindex;
	208	}
	209	if (dev)
	210	ipv6_dev_ac_dec(dev, &pac->acl_addr);
	211	sock_kfree_s(sk, pac, sizeof(*pac));
	212	pac = next;
	213	}
a9ed4a29	214	rtnl_unlock();
1da177e4 LT	215	}
1da177e4 LT	216
2384d025 JB	217	static void ipv6_add_acaddr_hash(struct net net, struct ifacaddr6 aca)
	218	{
	219	unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr);
	220
	221	spin_lock(&acaddr_hash_lock);
	222	hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]);
	223	spin_unlock(&acaddr_hash_lock);
	224	}
	225
	226	static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca)
	227	{
	228	spin_lock(&acaddr_hash_lock);
	229	hlist_del_init_rcu(&aca->aca_addr_lst);
	230	spin_unlock(&acaddr_hash_lock);
	231	}
	232
83aa29ee WC	233	static void aca_get(struct ifacaddr6 *aca)
83aa29ee WC	234	{
affa78bc	235	refcount_inc(&aca->aca_refcnt);
83aa29ee WC	236	}
83aa29ee WC	237
2384d025 JB	238	static void aca_free_rcu(struct rcu_head *h)
	239	{
	240	struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu);
	241
	242	fib6_info_release(aca->aca_rt);
	243	kfree(aca);
	244	}
	245
1da177e4 LT	246	static void aca_put(struct ifacaddr6 *ac)
1da177e4 LT	247	{
affa78bc	248	if (refcount_dec_and_test(&ac->aca_refcnt)) {
2384d025	249	call_rcu(&ac->rcu, aca_free_rcu);
1da177e4 LT	250	}
	251	}
	252
93c2fb25	253	static struct ifacaddr6 aca_alloc(struct fib6_info f6i,
83aa29ee WC	254	const struct in6_addr *addr)
83aa29ee WC	255	{
83aa29ee WC	256	struct ifacaddr6 *aca;
	257
	258	aca = kzalloc(sizeof(*aca), GFP_ATOMIC);
63159f29	259	if (!aca)
83aa29ee WC	260	return NULL;
	261
	262	aca->aca_addr = *addr;
93c2fb25 DA	263	fib6_info_hold(f6i);
93c2fb25 DA	264	aca->aca_rt = f6i;
2384d025	265	INIT_HLIST_NODE(&aca->aca_addr_lst);
83aa29ee WC	266	aca->aca_users = 1;
	267	/* aca_tstamp should be updated upon changes */
	268	aca->aca_cstamp = aca->aca_tstamp = jiffies;
affa78bc	269	refcount_set(&aca->aca_refcnt, 1);
83aa29ee WC	270
	271	return aca;
	272	}
	273
1da177e4 LT	274	/*
	275	* device anycast group inc (add if not found)
	276	*/
013b4d90	277	int __ipv6_dev_ac_inc(struct inet6_dev idev, const struct in6_addr addr)
1da177e4 LT	278	{
1da177e4 LT	279	struct ifacaddr6 *aca;
360a9887	280	struct fib6_info *f6i;
afb1d4b5	281	struct net *net;
1da177e4 LT	282	int err;
1da177e4 LT	283
a9ed4a29 SD	284	ASSERT_RTNL();
a9ed4a29 SD	285
1da177e4 LT	286	write_lock_bh(&idev->lock);
	287	if (idev->dead) {
	288	err = -ENODEV;
	289	goto out;
	290	}
	291
	292	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
	293	if (ipv6_addr_equal(&aca->aca_addr, addr)) {
	294	aca->aca_users++;
	295	err = 0;
	296	goto out;
	297	}
	298	}
	299
afb1d4b5	300	net = dev_net(idev->dev);
360a9887 DA	301	f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC);
	302	if (IS_ERR(f6i)) {
	303	err = PTR_ERR(f6i);
1da177e4 LT	304	goto out;
1da177e4 LT	305	}
360a9887	306	aca = aca_alloc(f6i, addr);
63159f29	307	if (!aca) {
360a9887	308	fib6_info_release(f6i);
83aa29ee WC	309	err = -ENOMEM;
	310	goto out;
	311	}
1da177e4 LT	312
	313	aca->aca_next = idev->ac_list;
	314	idev->ac_list = aca;
83aa29ee WC	315
	316	/* Hold this for addrconf_join_solict() below before we unlock,
	317	* it is already exposed via idev->ac_list.
	318	*/
	319	aca_get(aca);
1da177e4 LT	320	write_unlock_bh(&idev->lock);
1da177e4 LT	321
2384d025 JB	322	ipv6_add_acaddr_hash(net, aca);
2384d025 JB	323
360a9887	324	ip6_ins_rt(net, f6i);
1da177e4	325
013b4d90	326	addrconf_join_solict(idev->dev, &aca->aca_addr);
1da177e4 LT	327
	328	aca_put(aca);
	329	return 0;
	330	out:
	331	write_unlock_bh(&idev->lock);
1da177e4 LT	332	return err;
	333	}
	334
	335	/*
	336	* device anycast group decrement
	337	*/
b71d1d42	338	int __ipv6_dev_ac_dec(struct inet6_dev idev, const struct in6_addr addr)
1da177e4 LT	339	{
	340	struct ifacaddr6 aca, prev_aca;
	341
a9ed4a29 SD	342	ASSERT_RTNL();
a9ed4a29 SD	343
1da177e4 LT	344	write_lock_bh(&idev->lock);
	345	prev_aca = NULL;
	346	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
	347	if (ipv6_addr_equal(&aca->aca_addr, addr))
	348	break;
	349	prev_aca = aca;
	350	}
	351	if (!aca) {
	352	write_unlock_bh(&idev->lock);
	353	return -ENOENT;
	354	}
	355	if (--aca->aca_users > 0) {
	356	write_unlock_bh(&idev->lock);
	357	return 0;
	358	}
	359	if (prev_aca)
	360	prev_aca->aca_next = aca->aca_next;
	361	else
	362	idev->ac_list = aca->aca_next;
	363	write_unlock_bh(&idev->lock);
2384d025	364	ipv6_del_acaddr_hash(aca);
1da177e4 LT	365	addrconf_leave_solict(idev, &aca->aca_addr);
1da177e4 LT	366
afb1d4b5	367	ip6_del_rt(dev_net(idev->dev), aca->aca_rt);
1da177e4 LT	368
	369	aca_put(aca);
	370	return 0;
	371	}
	372
6c555490	373	/* called with rtnl_lock() */
b71d1d42	374	static int ipv6_dev_ac_dec(struct net_device dev, const struct in6_addr addr)
1da177e4	375	{
bb69ae04 ED	376	struct inet6_dev *idev = __in6_dev_get(dev);
bb69ae04 ED	377
63159f29	378	if (!idev)
1da177e4	379	return -ENODEV;
bb69ae04	380	return __ipv6_dev_ac_dec(idev, addr);
1da177e4	381	}
1ab1457c	382
381f4dca SD	383	void ipv6_ac_destroy_dev(struct inet6_dev *idev)
	384	{
	385	struct ifacaddr6 *aca;
	386
	387	write_lock_bh(&idev->lock);
	388	while ((aca = idev->ac_list) != NULL) {
	389	idev->ac_list = aca->aca_next;
	390	write_unlock_bh(&idev->lock);
	391
2384d025 JB	392	ipv6_del_acaddr_hash(aca);
2384d025 JB	393
381f4dca SD	394	addrconf_leave_solict(idev, &aca->aca_addr);
381f4dca SD	395
afb1d4b5	396	ip6_del_rt(dev_net(idev->dev), aca->aca_rt);
381f4dca SD	397
	398	aca_put(aca);
	399
	400	write_lock_bh(&idev->lock);
	401	}
	402	write_unlock_bh(&idev->lock);
	403	}
	404
1da177e4 LT	405	/*
1da177e4 LT	406	* check if the interface has this anycast address
bb69ae04	407	* called with rcu_read_lock()
1da177e4	408	*/
a50feda5	409	static bool ipv6_chk_acast_dev(struct net_device dev, const struct in6_addr addr)
1da177e4 LT	410	{
	411	struct inet6_dev *idev;
	412	struct ifacaddr6 *aca;
	413
bb69ae04	414	idev = __in6_dev_get(dev);
1da177e4 LT	415	if (idev) {
	416	read_lock_bh(&idev->lock);
	417	for (aca = idev->ac_list; aca; aca = aca->aca_next)
	418	if (ipv6_addr_equal(&aca->aca_addr, addr))
	419	break;
	420	read_unlock_bh(&idev->lock);
cfcabdcc	421	return aca != NULL;
1da177e4	422	}
a50feda5	423	return false;
1da177e4 LT	424	}
	425
	426	/*
	427	* check if given interface (or any, if dev==0) has this anycast address
	428	*/
a50feda5 ED	429	bool ipv6_chk_acast_addr(struct net net, struct net_device dev,
a50feda5 ED	430	const struct in6_addr *addr)
1da177e4	431	{
2384d025 JB	432	struct net_device *nh_dev;
2384d025 JB	433	struct ifacaddr6 *aca;
a50feda5	434	bool found = false;
7562f876	435
c6d14c84	436	rcu_read_lock();
bb69ae04 ED	437	if (dev)
bb69ae04 ED	438	found = ipv6_chk_acast_dev(dev, addr);
1c51dc9a LR	439	else {
	440	unsigned int hash = inet6_acaddr_hash(net, addr);
	441
2384d025 JB	442	hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash],
	443	aca_addr_lst) {
	444	nh_dev = fib6_info_nh_dev(aca->aca_rt);
	445	if (!nh_dev \|\| !net_eq(dev_net(nh_dev), net))
	446	continue;
	447	if (ipv6_addr_equal(&aca->aca_addr, addr)) {
a50feda5	448	found = true;
bb69ae04 ED	449	break;
bb69ae04 ED	450	}
2384d025	451	}
1c51dc9a	452	}
c6d14c84	453	rcu_read_unlock();
7562f876	454	return found;
1da177e4 LT	455	}
1da177e4 LT	456
7c90cc2d FLB	457	/* check if this anycast address is link-local on given interface or
	458	* is global
	459	*/
	460	bool ipv6_chk_acast_addr_src(struct net net, struct net_device dev,
	461	const struct in6_addr *addr)
	462	{
	463	return ipv6_chk_acast_addr(net,
	464	(ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
	465	dev : NULL),
	466	addr);
	467	}
1da177e4 LT	468
	469	#ifdef CONFIG_PROC_FS
	470	struct ac6_iter_state {
6ab57e7e	471	struct seq_net_private p;
1da177e4 LT	472	struct net_device *dev;
	473	struct inet6_dev *idev;
	474	};
	475
	476	#define ac6_seq_private(seq) ((struct ac6_iter_state *)(seq)->private)
	477
	478	static inline struct ifacaddr6 ac6_get_first(struct seq_file seq)
	479	{
	480	struct ifacaddr6 *im = NULL;
	481	struct ac6_iter_state *state = ac6_seq_private(seq);
6ab57e7e	482	struct net *net = seq_file_net(seq);
1da177e4	483
7562f876	484	state->idev = NULL;
ce81b76a	485	for_each_netdev_rcu(net, state->dev) {
1da177e4	486	struct inet6_dev *idev;
ce81b76a	487	idev = __in6_dev_get(state->dev);
1da177e4 LT	488	if (!idev)
	489	continue;
	490	read_lock_bh(&idev->lock);
	491	im = idev->ac_list;
	492	if (im) {
	493	state->idev = idev;
	494	break;
	495	}
	496	read_unlock_bh(&idev->lock);
	497	}
	498	return im;
	499	}
	500
	501	static struct ifacaddr6 ac6_get_next(struct seq_file seq, struct ifacaddr6 *im)
	502	{
	503	struct ac6_iter_state *state = ac6_seq_private(seq);
	504
	505	im = im->aca_next;
	506	while (!im) {
ce81b76a	507	if (likely(state->idev != NULL))
1da177e4	508	read_unlock_bh(&state->idev->lock);
ce81b76a ED	509
ce81b76a ED	510	state->dev = next_net_device_rcu(state->dev);
1da177e4 LT	511	if (!state->dev) {
	512	state->idev = NULL;
	513	break;
	514	}
ce81b76a	515	state->idev = __in6_dev_get(state->dev);
1da177e4 LT	516	if (!state->idev)
	517	continue;
	518	read_lock_bh(&state->idev->lock);
	519	im = state->idev->ac_list;
	520	}
	521	return im;
	522	}
	523
	524	static struct ifacaddr6 ac6_get_idx(struct seq_file seq, loff_t pos)
	525	{
	526	struct ifacaddr6 *im = ac6_get_first(seq);
	527	if (im)
	528	while (pos && (im = ac6_get_next(seq, im)) != NULL)
	529	--pos;
	530	return pos ? NULL : im;
	531	}
	532
	533	static void ac6_seq_start(struct seq_file seq, loff_t *pos)
ce81b76a	534	__acquires(RCU)
1da177e4	535	{
ce81b76a	536	rcu_read_lock();
1da177e4 LT	537	return ac6_get_idx(seq, *pos);
	538	}
	539
	540	static void ac6_seq_next(struct seq_file seq, void v, loff_t pos)
	541	{
ce81b76a ED	542	struct ifacaddr6 *im = ac6_get_next(seq, v);
ce81b76a ED	543
1da177e4 LT	544	++*pos;
	545	return im;
	546	}
	547
	548	static void ac6_seq_stop(struct seq_file seq, void v)
ce81b76a	549	__releases(RCU)
1da177e4 LT	550	{
1da177e4 LT	551	struct ac6_iter_state *state = ac6_seq_private(seq);
ce81b76a	552
1da177e4 LT	553	if (likely(state->idev != NULL)) {
1da177e4 LT	554	read_unlock_bh(&state->idev->lock);
ce81b76a	555	state->idev = NULL;
1da177e4	556	}
ce81b76a	557	rcu_read_unlock();
1da177e4 LT	558	}
	559
	560	static int ac6_seq_show(struct seq_file seq, void v)
	561	{
	562	struct ifacaddr6 im = (struct ifacaddr6 )v;
	563	struct ac6_iter_state *state = ac6_seq_private(seq);
	564
4b7a4274	565	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
1da177e4	566	state->dev->ifindex, state->dev->name,
b071195d	567	&im->aca_addr, im->aca_users);
1da177e4 LT	568	return 0;
	569	}
	570
56b3d975	571	static const struct seq_operations ac6_seq_ops = {
1da177e4 LT	572	.start = ac6_seq_start,
	573	.next = ac6_seq_next,
	574	.stop = ac6_seq_stop,
	575	.show = ac6_seq_show,
	576	};
	577
2c8c1e72	578	int __net_init ac6_proc_init(struct net *net)
1da177e4	579	{
c3506372 CH	580	if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops,
c3506372 CH	581	sizeof(struct ac6_iter_state)))
1da177e4 LT	582	return -ENOMEM;
	583
	584	return 0;
	585	}
	586
6ab57e7e	587	void ac6_proc_exit(struct net *net)
1da177e4	588	{
ece31ffd	589	remove_proc_entry("anycast6", net->proc_net);
1da177e4	590	}
6915ed86	591	#endif
2384d025 JB	592
	593	/* Init / cleanup code
	594	*/
	595	int __init ipv6_anycast_init(void)
	596	{
	597	int i;
	598
	599	for (i = 0; i < IN6_ADDR_HSIZE; i++)
	600	INIT_HLIST_HEAD(&inet6_acaddr_lst[i]);
	601	return 0;
	602	}
	603
	604	void ipv6_anycast_cleanup(void)
	605	{
	606	int i;
	607
	608	spin_lock(&acaddr_hash_lock);
	609	for (i = 0; i < IN6_ADDR_HSIZE; i++)
	610	WARN_ON(!hlist_empty(&inet6_acaddr_lst[i]));
	611	spin_unlock(&acaddr_hash_lock);
	612	}