2 * IP multicast routing support for mrouted 3.6/3.8
4 * (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 * Linux Consultancy and Custom Driver Development
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 * Michael Chastain : Incorrect size of copying.
14 * Alan Cox : Added the cache manager code
15 * Alan Cox : Fixed the clone/copy bug and device race.
16 * Mike McLagan : Routing by source
17 * Malcolm Beattie : Buffer handling fixes.
18 * Alexey Kuznetsov : Double buffer free and other fixes.
19 * SVR Anand : Fixed several multicast bugs and problems.
20 * Alexey Kuznetsov : Status, optimisations and more.
21 * Brad Parker : Better behaviour on mrouted upcall
23 * Carlos Picoto : PIMv1 Support
24 * Pavlin Ivanov Radoslavov: PIMv2 Registers must checksum only PIM header
25 * Relax this requirement to work with older peers.
29 #include <asm/uaccess.h>
30 #include <linux/types.h>
31 #include <linux/capability.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
35 #include <linux/kernel.h>
36 #include <linux/fcntl.h>
37 #include <linux/stat.h>
38 #include <linux/socket.h>
40 #include <linux/inet.h>
41 #include <linux/netdevice.h>
42 #include <linux/inetdevice.h>
43 #include <linux/igmp.h>
44 #include <linux/proc_fs.h>
45 #include <linux/seq_file.h>
46 #include <linux/mroute.h>
47 #include <linux/init.h>
48 #include <linux/if_ether.h>
49 #include <linux/slab.h>
50 #include <net/net_namespace.h>
52 #include <net/protocol.h>
53 #include <linux/skbuff.h>
54 #include <net/route.h>
59 #include <linux/notifier.h>
60 #include <linux/if_arp.h>
61 #include <linux/netfilter_ipv4.h>
62 #include <linux/compat.h>
63 #include <linux/export.h>
65 #include <net/checksum.h>
66 #include <net/netlink.h>
67 #include <net/fib_rules.h>
68 #include <linux/netconf.h>
70 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
71 #define CONFIG_IP_PIMSM 1
75 struct list_head list;
80 struct sock __rcu *mroute_sk;
81 struct timer_list ipmr_expire_timer;
82 struct list_head mfc_unres_queue;
83 struct list_head mfc_cache_array[MFC_LINES];
84 struct vif_device vif_table[MAXVIFS];
86 atomic_t cache_resolve_queue_len;
87 bool mroute_do_assert;
89 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
90 int mroute_reg_vif_num;
95 struct fib_rule common;
102 /* Big lock, protecting vif table, mrt cache and mroute socket state.
103 * Note that the changes are semaphored via rtnl_lock.
106 static DEFINE_RWLOCK(mrt_lock);
109 * Multicast router control variables
112 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
114 /* Special spinlock for queue of unresolved entries */
115 static DEFINE_SPINLOCK(mfc_unres_lock);
117 /* We return to original Alan's scheme. Hash table of resolved
118 * entries is changed only in process context and protected
119 * with weak lock mrt_lock. Queue of unresolved entries is protected
120 * with strong spinlock mfc_unres_lock.
122 * In this case data path is free of exclusive locks at all.
125 static struct kmem_cache *mrt_cachep __read_mostly;
127 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
128 static void ipmr_free_table(struct mr_table *mrt);
130 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
131 struct sk_buff *skb, struct mfc_cache *cache,
133 static int ipmr_cache_report(struct mr_table *mrt,
134 struct sk_buff *pkt, vifi_t vifi, int assert);
135 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
136 struct mfc_cache *c, struct rtmsg *rtm);
137 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
139 static void mroute_clean_tables(struct mr_table *mrt);
140 static void ipmr_expire_process(unsigned long arg);
142 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
143 #define ipmr_for_each_table(mrt, net) \
144 list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
146 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
148 struct mr_table *mrt;
150 ipmr_for_each_table(mrt, net) {
157 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
158 struct mr_table **mrt)
160 struct ipmr_result res;
161 struct fib_lookup_arg arg = { .result = &res, };
164 err = fib_rules_lookup(net->ipv4.mr_rules_ops,
165 flowi4_to_flowi(flp4), 0, &arg);
172 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
173 int flags, struct fib_lookup_arg *arg)
175 struct ipmr_result *res = arg->result;
176 struct mr_table *mrt;
178 switch (rule->action) {
181 case FR_ACT_UNREACHABLE:
183 case FR_ACT_PROHIBIT:
185 case FR_ACT_BLACKHOLE:
190 mrt = ipmr_get_table(rule->fr_net, rule->table);
197 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
202 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
206 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
207 struct fib_rule_hdr *frh, struct nlattr **tb)
212 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
218 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
219 struct fib_rule_hdr *frh)
227 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
228 .family = RTNL_FAMILY_IPMR,
229 .rule_size = sizeof(struct ipmr_rule),
230 .addr_size = sizeof(u32),
231 .action = ipmr_rule_action,
232 .match = ipmr_rule_match,
233 .configure = ipmr_rule_configure,
234 .compare = ipmr_rule_compare,
235 .default_pref = fib_default_rule_pref,
236 .fill = ipmr_rule_fill,
237 .nlgroup = RTNLGRP_IPV4_RULE,
238 .policy = ipmr_rule_policy,
239 .owner = THIS_MODULE,
242 static int __net_init ipmr_rules_init(struct net *net)
244 struct fib_rules_ops *ops;
245 struct mr_table *mrt;
248 ops = fib_rules_register(&ipmr_rules_ops_template, net);
252 INIT_LIST_HEAD(&net->ipv4.mr_tables);
254 mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
260 err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
264 net->ipv4.mr_rules_ops = ops;
270 fib_rules_unregister(ops);
274 static void __net_exit ipmr_rules_exit(struct net *net)
276 struct mr_table *mrt, *next;
278 list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
279 list_del(&mrt->list);
280 ipmr_free_table(mrt);
282 fib_rules_unregister(net->ipv4.mr_rules_ops);
285 #define ipmr_for_each_table(mrt, net) \
286 for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
288 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
290 return net->ipv4.mrt;
293 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
294 struct mr_table **mrt)
296 *mrt = net->ipv4.mrt;
300 static int __net_init ipmr_rules_init(struct net *net)
302 net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
303 return net->ipv4.mrt ? 0 : -ENOMEM;
306 static void __net_exit ipmr_rules_exit(struct net *net)
308 ipmr_free_table(net->ipv4.mrt);
312 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
314 struct mr_table *mrt;
317 mrt = ipmr_get_table(net, id);
321 mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
324 write_pnet(&mrt->net, net);
327 /* Forwarding cache */
328 for (i = 0; i < MFC_LINES; i++)
329 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
331 INIT_LIST_HEAD(&mrt->mfc_unres_queue);
333 setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
336 #ifdef CONFIG_IP_PIMSM
337 mrt->mroute_reg_vif_num = -1;
339 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
340 list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
345 static void ipmr_free_table(struct mr_table *mrt)
347 del_timer_sync(&mrt->ipmr_expire_timer);
348 mroute_clean_tables(mrt);
352 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
354 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
356 struct net *net = dev_net(dev);
360 dev = __dev_get_by_name(net, "tunl0");
362 const struct net_device_ops *ops = dev->netdev_ops;
364 struct ip_tunnel_parm p;
366 memset(&p, 0, sizeof(p));
367 p.iph.daddr = v->vifc_rmt_addr.s_addr;
368 p.iph.saddr = v->vifc_lcl_addr.s_addr;
371 p.iph.protocol = IPPROTO_IPIP;
372 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
373 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
375 if (ops->ndo_do_ioctl) {
376 mm_segment_t oldfs = get_fs();
379 ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
386 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
388 struct net_device *dev;
390 dev = __dev_get_by_name(net, "tunl0");
393 const struct net_device_ops *ops = dev->netdev_ops;
396 struct ip_tunnel_parm p;
397 struct in_device *in_dev;
399 memset(&p, 0, sizeof(p));
400 p.iph.daddr = v->vifc_rmt_addr.s_addr;
401 p.iph.saddr = v->vifc_lcl_addr.s_addr;
404 p.iph.protocol = IPPROTO_IPIP;
405 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
406 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
408 if (ops->ndo_do_ioctl) {
409 mm_segment_t oldfs = get_fs();
412 err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
420 (dev = __dev_get_by_name(net, p.name)) != NULL) {
421 dev->flags |= IFF_MULTICAST;
423 in_dev = __in_dev_get_rtnl(dev);
427 ipv4_devconf_setall(in_dev);
428 IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
438 /* allow the register to be completed before unregistering. */
442 unregister_netdevice(dev);
446 #ifdef CONFIG_IP_PIMSM
448 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
450 struct net *net = dev_net(dev);
451 struct mr_table *mrt;
452 struct flowi4 fl4 = {
453 .flowi4_oif = dev->ifindex,
454 .flowi4_iif = skb->skb_iif,
455 .flowi4_mark = skb->mark,
459 err = ipmr_fib_lookup(net, &fl4, &mrt);
465 read_lock(&mrt_lock);
466 dev->stats.tx_bytes += skb->len;
467 dev->stats.tx_packets++;
468 ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
469 read_unlock(&mrt_lock);
474 static const struct net_device_ops reg_vif_netdev_ops = {
475 .ndo_start_xmit = reg_vif_xmit,
478 static void reg_vif_setup(struct net_device *dev)
480 dev->type = ARPHRD_PIMREG;
481 dev->mtu = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
482 dev->flags = IFF_NOARP;
483 dev->netdev_ops = ®_vif_netdev_ops,
484 dev->destructor = free_netdev;
485 dev->features |= NETIF_F_NETNS_LOCAL;
488 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
490 struct net_device *dev;
491 struct in_device *in_dev;
494 if (mrt->id == RT_TABLE_DEFAULT)
495 sprintf(name, "pimreg");
497 sprintf(name, "pimreg%u", mrt->id);
499 dev = alloc_netdev(0, name, reg_vif_setup);
504 dev_net_set(dev, net);
506 if (register_netdevice(dev)) {
513 in_dev = __in_dev_get_rcu(dev);
519 ipv4_devconf_setall(in_dev);
520 IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
531 /* allow the register to be completed before unregistering. */
535 unregister_netdevice(dev);
541 * vif_delete - Delete a VIF entry
542 * @notify: Set to 1, if the caller is a notifier_call
545 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
546 struct list_head *head)
548 struct vif_device *v;
549 struct net_device *dev;
550 struct in_device *in_dev;
552 if (vifi < 0 || vifi >= mrt->maxvif)
553 return -EADDRNOTAVAIL;
555 v = &mrt->vif_table[vifi];
557 write_lock_bh(&mrt_lock);
562 write_unlock_bh(&mrt_lock);
563 return -EADDRNOTAVAIL;
566 #ifdef CONFIG_IP_PIMSM
567 if (vifi == mrt->mroute_reg_vif_num)
568 mrt->mroute_reg_vif_num = -1;
571 if (vifi + 1 == mrt->maxvif) {
574 for (tmp = vifi - 1; tmp >= 0; tmp--) {
575 if (VIF_EXISTS(mrt, tmp))
581 write_unlock_bh(&mrt_lock);
583 dev_set_allmulti(dev, -1);
585 in_dev = __in_dev_get_rtnl(dev);
587 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
588 inet_netconf_notify_devconf(dev_net(dev),
589 NETCONFA_MC_FORWARDING,
590 dev->ifindex, &in_dev->cnf);
591 ip_rt_multicast_event(in_dev);
594 if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
595 unregister_netdevice_queue(dev, head);
601 static void ipmr_cache_free_rcu(struct rcu_head *head)
603 struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
605 kmem_cache_free(mrt_cachep, c);
608 static inline void ipmr_cache_free(struct mfc_cache *c)
610 call_rcu(&c->rcu, ipmr_cache_free_rcu);
613 /* Destroy an unresolved cache entry, killing queued skbs
614 * and reporting error to netlink readers.
617 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
619 struct net *net = read_pnet(&mrt->net);
623 atomic_dec(&mrt->cache_resolve_queue_len);
625 while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
626 if (ip_hdr(skb)->version == 0) {
627 struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
628 nlh->nlmsg_type = NLMSG_ERROR;
629 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
630 skb_trim(skb, nlh->nlmsg_len);
632 e->error = -ETIMEDOUT;
633 memset(&e->msg, 0, sizeof(e->msg));
635 rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
645 /* Timer process for the unresolved queue. */
647 static void ipmr_expire_process(unsigned long arg)
649 struct mr_table *mrt = (struct mr_table *)arg;
651 unsigned long expires;
652 struct mfc_cache *c, *next;
654 if (!spin_trylock(&mfc_unres_lock)) {
655 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
659 if (list_empty(&mrt->mfc_unres_queue))
665 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
666 if (time_after(c->mfc_un.unres.expires, now)) {
667 unsigned long interval = c->mfc_un.unres.expires - now;
668 if (interval < expires)
674 mroute_netlink_event(mrt, c, RTM_DELROUTE);
675 ipmr_destroy_unres(mrt, c);
678 if (!list_empty(&mrt->mfc_unres_queue))
679 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
682 spin_unlock(&mfc_unres_lock);
685 /* Fill oifs list. It is called under write locked mrt_lock. */
687 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
692 cache->mfc_un.res.minvif = MAXVIFS;
693 cache->mfc_un.res.maxvif = 0;
694 memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
696 for (vifi = 0; vifi < mrt->maxvif; vifi++) {
697 if (VIF_EXISTS(mrt, vifi) &&
698 ttls[vifi] && ttls[vifi] < 255) {
699 cache->mfc_un.res.ttls[vifi] = ttls[vifi];
700 if (cache->mfc_un.res.minvif > vifi)
701 cache->mfc_un.res.minvif = vifi;
702 if (cache->mfc_un.res.maxvif <= vifi)
703 cache->mfc_un.res.maxvif = vifi + 1;
708 static int vif_add(struct net *net, struct mr_table *mrt,
709 struct vifctl *vifc, int mrtsock)
711 int vifi = vifc->vifc_vifi;
712 struct vif_device *v = &mrt->vif_table[vifi];
713 struct net_device *dev;
714 struct in_device *in_dev;
718 if (VIF_EXISTS(mrt, vifi))
721 switch (vifc->vifc_flags) {
722 #ifdef CONFIG_IP_PIMSM
725 * Special Purpose VIF in PIM
726 * All the packets will be sent to the daemon
728 if (mrt->mroute_reg_vif_num >= 0)
730 dev = ipmr_reg_vif(net, mrt);
733 err = dev_set_allmulti(dev, 1);
735 unregister_netdevice(dev);
742 dev = ipmr_new_tunnel(net, vifc);
745 err = dev_set_allmulti(dev, 1);
747 ipmr_del_tunnel(dev, vifc);
753 case VIFF_USE_IFINDEX:
755 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
756 dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
757 if (dev && __in_dev_get_rtnl(dev) == NULL) {
759 return -EADDRNOTAVAIL;
762 dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
765 return -EADDRNOTAVAIL;
766 err = dev_set_allmulti(dev, 1);
776 in_dev = __in_dev_get_rtnl(dev);
779 return -EADDRNOTAVAIL;
781 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
782 inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
784 ip_rt_multicast_event(in_dev);
786 /* Fill in the VIF structures */
788 v->rate_limit = vifc->vifc_rate_limit;
789 v->local = vifc->vifc_lcl_addr.s_addr;
790 v->remote = vifc->vifc_rmt_addr.s_addr;
791 v->flags = vifc->vifc_flags;
793 v->flags |= VIFF_STATIC;
794 v->threshold = vifc->vifc_threshold;
799 v->link = dev->ifindex;
800 if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
801 v->link = dev->iflink;
803 /* And finish update writing critical data */
804 write_lock_bh(&mrt_lock);
806 #ifdef CONFIG_IP_PIMSM
807 if (v->flags & VIFF_REGISTER)
808 mrt->mroute_reg_vif_num = vifi;
810 if (vifi+1 > mrt->maxvif)
811 mrt->maxvif = vifi+1;
812 write_unlock_bh(&mrt_lock);
816 /* called with rcu_read_lock() */
817 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
821 int line = MFC_HASH(mcastgrp, origin);
824 list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
825 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
832 * Allocate a multicast cache entry
834 static struct mfc_cache *ipmr_cache_alloc(void)
836 struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
839 c->mfc_un.res.minvif = MAXVIFS;
843 static struct mfc_cache *ipmr_cache_alloc_unres(void)
845 struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
848 skb_queue_head_init(&c->mfc_un.unres.unresolved);
849 c->mfc_un.unres.expires = jiffies + 10*HZ;
855 * A cache entry has gone into a resolved state from queued
858 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
859 struct mfc_cache *uc, struct mfc_cache *c)
864 /* Play the pending entries through our router */
866 while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
867 if (ip_hdr(skb)->version == 0) {
868 struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
870 if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
871 nlh->nlmsg_len = skb_tail_pointer(skb) -
874 nlh->nlmsg_type = NLMSG_ERROR;
875 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
876 skb_trim(skb, nlh->nlmsg_len);
878 e->error = -EMSGSIZE;
879 memset(&e->msg, 0, sizeof(e->msg));
882 rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
884 ip_mr_forward(net, mrt, skb, c, 0);
890 * Bounce a cache query up to mrouted. We could use netlink for this but mrouted
891 * expects the following bizarre scheme.
893 * Called under mrt_lock.
896 static int ipmr_cache_report(struct mr_table *mrt,
897 struct sk_buff *pkt, vifi_t vifi, int assert)
900 const int ihl = ip_hdrlen(pkt);
901 struct igmphdr *igmp;
903 struct sock *mroute_sk;
906 #ifdef CONFIG_IP_PIMSM
907 if (assert == IGMPMSG_WHOLEPKT)
908 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
911 skb = alloc_skb(128, GFP_ATOMIC);
916 #ifdef CONFIG_IP_PIMSM
917 if (assert == IGMPMSG_WHOLEPKT) {
918 /* Ugly, but we have no choice with this interface.
919 * Duplicate old header, fix ihl, length etc.
920 * And all this only to mangle msg->im_msgtype and
921 * to set msg->im_mbz to "mbz" :-)
923 skb_push(skb, sizeof(struct iphdr));
924 skb_reset_network_header(skb);
925 skb_reset_transport_header(skb);
926 msg = (struct igmpmsg *)skb_network_header(skb);
927 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
928 msg->im_msgtype = IGMPMSG_WHOLEPKT;
930 msg->im_vif = mrt->mroute_reg_vif_num;
931 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
932 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
933 sizeof(struct iphdr));
938 /* Copy the IP header */
940 skb->network_header = skb->tail;
942 skb_copy_to_linear_data(skb, pkt->data, ihl);
943 ip_hdr(skb)->protocol = 0; /* Flag to the kernel this is a route add */
944 msg = (struct igmpmsg *)skb_network_header(skb);
946 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
950 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
952 msg->im_msgtype = assert;
954 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
955 skb->transport_header = skb->network_header;
959 mroute_sk = rcu_dereference(mrt->mroute_sk);
960 if (mroute_sk == NULL) {
966 /* Deliver to mrouted */
968 ret = sock_queue_rcv_skb(mroute_sk, skb);
971 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
979 * Queue a packet for resolution. It gets locked cache entry!
983 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
988 const struct iphdr *iph = ip_hdr(skb);
990 spin_lock_bh(&mfc_unres_lock);
991 list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
992 if (c->mfc_mcastgrp == iph->daddr &&
993 c->mfc_origin == iph->saddr) {
1000 /* Create a new entry if allowable */
1002 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1003 (c = ipmr_cache_alloc_unres()) == NULL) {
1004 spin_unlock_bh(&mfc_unres_lock);
1010 /* Fill in the new cache entry */
1013 c->mfc_origin = iph->saddr;
1014 c->mfc_mcastgrp = iph->daddr;
1016 /* Reflect first query at mrouted. */
1018 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1020 /* If the report failed throw the cache entry
1023 spin_unlock_bh(&mfc_unres_lock);
1030 atomic_inc(&mrt->cache_resolve_queue_len);
1031 list_add(&c->list, &mrt->mfc_unres_queue);
1032 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1034 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1035 mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1038 /* See if we can append the packet */
1040 if (c->mfc_un.unres.unresolved.qlen > 3) {
1044 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1048 spin_unlock_bh(&mfc_unres_lock);
1053 * MFC cache manipulation by user space mroute daemon
1056 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1059 struct mfc_cache *c, *next;
1061 line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1063 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1064 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1065 c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1066 list_del_rcu(&c->list);
1067 mroute_netlink_event(mrt, c, RTM_DELROUTE);
1075 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1076 struct mfcctl *mfc, int mrtsock)
1080 struct mfc_cache *uc, *c;
1082 if (mfc->mfcc_parent >= MAXVIFS)
1085 line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1087 list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1088 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1089 c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1096 write_lock_bh(&mrt_lock);
1097 c->mfc_parent = mfc->mfcc_parent;
1098 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1100 c->mfc_flags |= MFC_STATIC;
1101 write_unlock_bh(&mrt_lock);
1102 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1106 if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1109 c = ipmr_cache_alloc();
1113 c->mfc_origin = mfc->mfcc_origin.s_addr;
1114 c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1115 c->mfc_parent = mfc->mfcc_parent;
1116 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1118 c->mfc_flags |= MFC_STATIC;
1120 list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1123 * Check to see if we resolved a queued list. If so we
1124 * need to send on the frames and tidy up.
1127 spin_lock_bh(&mfc_unres_lock);
1128 list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1129 if (uc->mfc_origin == c->mfc_origin &&
1130 uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1131 list_del(&uc->list);
1132 atomic_dec(&mrt->cache_resolve_queue_len);
1137 if (list_empty(&mrt->mfc_unres_queue))
1138 del_timer(&mrt->ipmr_expire_timer);
1139 spin_unlock_bh(&mfc_unres_lock);
1142 ipmr_cache_resolve(net, mrt, uc, c);
1143 ipmr_cache_free(uc);
1145 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1150 * Close the multicast socket, and clear the vif tables etc
1153 static void mroute_clean_tables(struct mr_table *mrt)
1157 struct mfc_cache *c, *next;
1159 /* Shut down all active vif entries */
1161 for (i = 0; i < mrt->maxvif; i++) {
1162 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1163 vif_delete(mrt, i, 0, &list);
1165 unregister_netdevice_many(&list);
1167 /* Wipe the cache */
1169 for (i = 0; i < MFC_LINES; i++) {
1170 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1171 if (c->mfc_flags & MFC_STATIC)
1173 list_del_rcu(&c->list);
1174 mroute_netlink_event(mrt, c, RTM_DELROUTE);
1179 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1180 spin_lock_bh(&mfc_unres_lock);
1181 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1183 mroute_netlink_event(mrt, c, RTM_DELROUTE);
1184 ipmr_destroy_unres(mrt, c);
1186 spin_unlock_bh(&mfc_unres_lock);
1190 /* called from ip_ra_control(), before an RCU grace period,
1191 * we dont need to call synchronize_rcu() here
1193 static void mrtsock_destruct(struct sock *sk)
1195 struct net *net = sock_net(sk);
1196 struct mr_table *mrt;
1199 ipmr_for_each_table(mrt, net) {
1200 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1201 IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1202 inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1203 NETCONFA_IFINDEX_ALL,
1204 net->ipv4.devconf_all);
1205 RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1206 mroute_clean_tables(mrt);
1213 * Socket options and virtual interface manipulation. The whole
1214 * virtual interface system is a complete heap, but unfortunately
1215 * that's how BSD mrouted happens to think. Maybe one day with a proper
1216 * MOSPF/PIM router set up we can clean this up.
1219 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1224 struct net *net = sock_net(sk);
1225 struct mr_table *mrt;
1227 if (sk->sk_type != SOCK_RAW ||
1228 inet_sk(sk)->inet_num != IPPROTO_IGMP)
1231 mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1235 if (optname != MRT_INIT) {
1236 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1237 !ns_capable(net->user_ns, CAP_NET_ADMIN))
1243 if (optlen != sizeof(int))
1247 if (rtnl_dereference(mrt->mroute_sk)) {
1252 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1254 rcu_assign_pointer(mrt->mroute_sk, sk);
1255 IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1256 inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1257 NETCONFA_IFINDEX_ALL,
1258 net->ipv4.devconf_all);
1263 if (sk != rcu_access_pointer(mrt->mroute_sk))
1265 return ip_ra_control(sk, 0, NULL);
1268 if (optlen != sizeof(vif))
1270 if (copy_from_user(&vif, optval, sizeof(vif)))
1272 if (vif.vifc_vifi >= MAXVIFS)
1275 if (optname == MRT_ADD_VIF) {
1276 ret = vif_add(net, mrt, &vif,
1277 sk == rtnl_dereference(mrt->mroute_sk));
1279 ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1285 * Manipulate the forwarding caches. These live
1286 * in a sort of kernel/user symbiosis.
1290 if (optlen != sizeof(mfc))
1292 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1295 if (optname == MRT_DEL_MFC)
1296 ret = ipmr_mfc_delete(mrt, &mfc);
1298 ret = ipmr_mfc_add(net, mrt, &mfc,
1299 sk == rtnl_dereference(mrt->mroute_sk));
1303 * Control PIM assert.
1308 if (optlen != sizeof(v))
1310 if (get_user(v, (int __user *)optval))
1312 mrt->mroute_do_assert = v;
1315 #ifdef CONFIG_IP_PIMSM
1320 if (optlen != sizeof(v))
1322 if (get_user(v, (int __user *)optval))
1328 if (v != mrt->mroute_do_pim) {
1329 mrt->mroute_do_pim = v;
1330 mrt->mroute_do_assert = v;
1336 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1341 if (optlen != sizeof(u32))
1343 if (get_user(v, (u32 __user *)optval))
1346 /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
1347 if (v != RT_TABLE_DEFAULT && v >= 1000000000)
1352 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1355 if (!ipmr_new_table(net, v))
1358 raw_sk(sk)->ipmr_table = v;
1365 * Spurious command, or MRT_VERSION which you cannot
1369 return -ENOPROTOOPT;
1374 * Getsock opt support for the multicast routing system.
1377 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1381 struct net *net = sock_net(sk);
1382 struct mr_table *mrt;
1384 if (sk->sk_type != SOCK_RAW ||
1385 inet_sk(sk)->inet_num != IPPROTO_IGMP)
1388 mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1392 if (optname != MRT_VERSION &&
1393 #ifdef CONFIG_IP_PIMSM
1394 optname != MRT_PIM &&
1396 optname != MRT_ASSERT)
1397 return -ENOPROTOOPT;
1399 if (get_user(olr, optlen))
1402 olr = min_t(unsigned int, olr, sizeof(int));
1406 if (put_user(olr, optlen))
1408 if (optname == MRT_VERSION)
1410 #ifdef CONFIG_IP_PIMSM
1411 else if (optname == MRT_PIM)
1412 val = mrt->mroute_do_pim;
1415 val = mrt->mroute_do_assert;
1416 if (copy_to_user(optval, &val, olr))
1422 * The IP multicast ioctl support routines.
1425 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1427 struct sioc_sg_req sr;
1428 struct sioc_vif_req vr;
1429 struct vif_device *vif;
1430 struct mfc_cache *c;
1431 struct net *net = sock_net(sk);
1432 struct mr_table *mrt;
1434 mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1440 if (copy_from_user(&vr, arg, sizeof(vr)))
1442 if (vr.vifi >= mrt->maxvif)
1444 read_lock(&mrt_lock);
1445 vif = &mrt->vif_table[vr.vifi];
1446 if (VIF_EXISTS(mrt, vr.vifi)) {
1447 vr.icount = vif->pkt_in;
1448 vr.ocount = vif->pkt_out;
1449 vr.ibytes = vif->bytes_in;
1450 vr.obytes = vif->bytes_out;
1451 read_unlock(&mrt_lock);
1453 if (copy_to_user(arg, &vr, sizeof(vr)))
1457 read_unlock(&mrt_lock);
1458 return -EADDRNOTAVAIL;
1460 if (copy_from_user(&sr, arg, sizeof(sr)))
1464 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1466 sr.pktcnt = c->mfc_un.res.pkt;
1467 sr.bytecnt = c->mfc_un.res.bytes;
1468 sr.wrong_if = c->mfc_un.res.wrong_if;
1471 if (copy_to_user(arg, &sr, sizeof(sr)))
1476 return -EADDRNOTAVAIL;
1478 return -ENOIOCTLCMD;
1482 #ifdef CONFIG_COMPAT
1483 struct compat_sioc_sg_req {
1486 compat_ulong_t pktcnt;
1487 compat_ulong_t bytecnt;
1488 compat_ulong_t wrong_if;
1491 struct compat_sioc_vif_req {
1492 vifi_t vifi; /* Which iface */
1493 compat_ulong_t icount;
1494 compat_ulong_t ocount;
1495 compat_ulong_t ibytes;
1496 compat_ulong_t obytes;
1499 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1501 struct compat_sioc_sg_req sr;
1502 struct compat_sioc_vif_req vr;
1503 struct vif_device *vif;
1504 struct mfc_cache *c;
1505 struct net *net = sock_net(sk);
1506 struct mr_table *mrt;
1508 mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1514 if (copy_from_user(&vr, arg, sizeof(vr)))
1516 if (vr.vifi >= mrt->maxvif)
1518 read_lock(&mrt_lock);
1519 vif = &mrt->vif_table[vr.vifi];
1520 if (VIF_EXISTS(mrt, vr.vifi)) {
1521 vr.icount = vif->pkt_in;
1522 vr.ocount = vif->pkt_out;
1523 vr.ibytes = vif->bytes_in;
1524 vr.obytes = vif->bytes_out;
1525 read_unlock(&mrt_lock);
1527 if (copy_to_user(arg, &vr, sizeof(vr)))
1531 read_unlock(&mrt_lock);
1532 return -EADDRNOTAVAIL;
1534 if (copy_from_user(&sr, arg, sizeof(sr)))
1538 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1540 sr.pktcnt = c->mfc_un.res.pkt;
1541 sr.bytecnt = c->mfc_un.res.bytes;
1542 sr.wrong_if = c->mfc_un.res.wrong_if;
1545 if (copy_to_user(arg, &sr, sizeof(sr)))
1550 return -EADDRNOTAVAIL;
1552 return -ENOIOCTLCMD;
1558 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1560 struct net_device *dev = ptr;
1561 struct net *net = dev_net(dev);
1562 struct mr_table *mrt;
1563 struct vif_device *v;
1566 if (event != NETDEV_UNREGISTER)
1569 ipmr_for_each_table(mrt, net) {
1570 v = &mrt->vif_table[0];
1571 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1573 vif_delete(mrt, ct, 1, NULL);
1580 static struct notifier_block ip_mr_notifier = {
1581 .notifier_call = ipmr_device_event,
1585 * Encapsulate a packet by attaching a valid IPIP header to it.
1586 * This avoids tunnel drivers and other mess and gives us the speed so
1587 * important for multicast video.
1590 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1593 const struct iphdr *old_iph = ip_hdr(skb);
1595 skb_push(skb, sizeof(struct iphdr));
1596 skb->transport_header = skb->network_header;
1597 skb_reset_network_header(skb);
1601 iph->tos = old_iph->tos;
1602 iph->ttl = old_iph->ttl;
1606 iph->protocol = IPPROTO_IPIP;
1608 iph->tot_len = htons(skb->len);
1609 ip_select_ident(iph, skb_dst(skb), NULL);
1612 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1616 static inline int ipmr_forward_finish(struct sk_buff *skb)
1618 struct ip_options *opt = &(IPCB(skb)->opt);
1620 IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1621 IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
1623 if (unlikely(opt->optlen))
1624 ip_forward_options(skb);
1626 return dst_output(skb);
1630 * Processing handlers for ipmr_forward
1633 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1634 struct sk_buff *skb, struct mfc_cache *c, int vifi)
1636 const struct iphdr *iph = ip_hdr(skb);
1637 struct vif_device *vif = &mrt->vif_table[vifi];
1638 struct net_device *dev;
1643 if (vif->dev == NULL)
1646 #ifdef CONFIG_IP_PIMSM
1647 if (vif->flags & VIFF_REGISTER) {
1649 vif->bytes_out += skb->len;
1650 vif->dev->stats.tx_bytes += skb->len;
1651 vif->dev->stats.tx_packets++;
1652 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1657 if (vif->flags & VIFF_TUNNEL) {
1658 rt = ip_route_output_ports(net, &fl4, NULL,
1659 vif->remote, vif->local,
1662 RT_TOS(iph->tos), vif->link);
1665 encap = sizeof(struct iphdr);
1667 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1670 RT_TOS(iph->tos), vif->link);
1677 if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1678 /* Do not fragment multicasts. Alas, IPv4 does not
1679 * allow to send ICMP, so that packets will disappear
1683 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1688 encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1690 if (skb_cow(skb, encap)) {
1696 vif->bytes_out += skb->len;
1699 skb_dst_set(skb, &rt->dst);
1700 ip_decrease_ttl(ip_hdr(skb));
1702 /* FIXME: forward and output firewalls used to be called here.
1703 * What do we do with netfilter? -- RR
1705 if (vif->flags & VIFF_TUNNEL) {
1706 ip_encap(skb, vif->local, vif->remote);
1707 /* FIXME: extra output firewall step used to be here. --RR */
1708 vif->dev->stats.tx_packets++;
1709 vif->dev->stats.tx_bytes += skb->len;
1712 IPCB(skb)->flags |= IPSKB_FORWARDED;
1715 * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1716 * not only before forwarding, but after forwarding on all output
1717 * interfaces. It is clear, if mrouter runs a multicasting
1718 * program, it should receive packets not depending to what interface
1719 * program is joined.
1720 * If we will not make it, the program will have to join on all
1721 * interfaces. On the other hand, multihoming host (or router, but
1722 * not mrouter) cannot join to more than one interface - it will
1723 * result in receiving multiple packets.
1725 NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1726 ipmr_forward_finish);
1733 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1737 for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1738 if (mrt->vif_table[ct].dev == dev)
1744 /* "local" means that we should preserve one skb (for local delivery) */
1746 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1747 struct sk_buff *skb, struct mfc_cache *cache,
1753 vif = cache->mfc_parent;
1754 cache->mfc_un.res.pkt++;
1755 cache->mfc_un.res.bytes += skb->len;
1758 * Wrong interface: drop packet and (maybe) send PIM assert.
1760 if (mrt->vif_table[vif].dev != skb->dev) {
1763 if (rt_is_output_route(skb_rtable(skb))) {
1764 /* It is our own packet, looped back.
1765 * Very complicated situation...
1767 * The best workaround until routing daemons will be
1768 * fixed is not to redistribute packet, if it was
1769 * send through wrong interface. It means, that
1770 * multicast applications WILL NOT work for
1771 * (S,G), which have default multicast route pointing
1772 * to wrong oif. In any case, it is not a good
1773 * idea to use multicasting applications on router.
1778 cache->mfc_un.res.wrong_if++;
1779 true_vifi = ipmr_find_vif(mrt, skb->dev);
1781 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1782 /* pimsm uses asserts, when switching from RPT to SPT,
1783 * so that we cannot check that packet arrived on an oif.
1784 * It is bad, but otherwise we would need to move pretty
1785 * large chunk of pimd to kernel. Ough... --ANK
1787 (mrt->mroute_do_pim ||
1788 cache->mfc_un.res.ttls[true_vifi] < 255) &&
1790 cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1791 cache->mfc_un.res.last_assert = jiffies;
1792 ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1797 mrt->vif_table[vif].pkt_in++;
1798 mrt->vif_table[vif].bytes_in += skb->len;
1803 for (ct = cache->mfc_un.res.maxvif - 1;
1804 ct >= cache->mfc_un.res.minvif; ct--) {
1805 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1807 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1810 ipmr_queue_xmit(net, mrt, skb2, cache,
1818 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1821 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1823 ipmr_queue_xmit(net, mrt, skb, cache, psend);
1834 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1836 struct rtable *rt = skb_rtable(skb);
1837 struct iphdr *iph = ip_hdr(skb);
1838 struct flowi4 fl4 = {
1839 .daddr = iph->daddr,
1840 .saddr = iph->saddr,
1841 .flowi4_tos = RT_TOS(iph->tos),
1842 .flowi4_oif = (rt_is_output_route(rt) ?
1843 skb->dev->ifindex : 0),
1844 .flowi4_iif = (rt_is_output_route(rt) ?
1847 .flowi4_mark = skb->mark,
1849 struct mr_table *mrt;
1852 err = ipmr_fib_lookup(net, &fl4, &mrt);
1854 return ERR_PTR(err);
1859 * Multicast packets for forwarding arrive here
1860 * Called with rcu_read_lock();
1863 int ip_mr_input(struct sk_buff *skb)
1865 struct mfc_cache *cache;
1866 struct net *net = dev_net(skb->dev);
1867 int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1868 struct mr_table *mrt;
1870 /* Packet is looped back after forward, it should not be
1871 * forwarded second time, but still can be delivered locally.
1873 if (IPCB(skb)->flags & IPSKB_FORWARDED)
1876 mrt = ipmr_rt_fib_lookup(net, skb);
1879 return PTR_ERR(mrt);
1882 if (IPCB(skb)->opt.router_alert) {
1883 if (ip_call_ra_chain(skb))
1885 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1886 /* IGMPv1 (and broken IGMPv2 implementations sort of
1887 * Cisco IOS <= 11.2(8)) do not put router alert
1888 * option to IGMP packets destined to routable
1889 * groups. It is very bad, because it means
1890 * that we can forward NO IGMP messages.
1892 struct sock *mroute_sk;
1894 mroute_sk = rcu_dereference(mrt->mroute_sk);
1897 raw_rcv(mroute_sk, skb);
1903 /* already under rcu_read_lock() */
1904 cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1907 * No usable cache entry
1909 if (cache == NULL) {
1913 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1914 ip_local_deliver(skb);
1920 read_lock(&mrt_lock);
1921 vif = ipmr_find_vif(mrt, skb->dev);
1923 int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1924 read_unlock(&mrt_lock);
1928 read_unlock(&mrt_lock);
1933 read_lock(&mrt_lock);
1934 ip_mr_forward(net, mrt, skb, cache, local);
1935 read_unlock(&mrt_lock);
1938 return ip_local_deliver(skb);
1944 return ip_local_deliver(skb);
1949 #ifdef CONFIG_IP_PIMSM
1950 /* called with rcu_read_lock() */
1951 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1952 unsigned int pimlen)
1954 struct net_device *reg_dev = NULL;
1955 struct iphdr *encap;
1957 encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1960 * a. packet is really sent to a multicast group
1961 * b. packet is not a NULL-REGISTER
1962 * c. packet is not truncated
1964 if (!ipv4_is_multicast(encap->daddr) ||
1965 encap->tot_len == 0 ||
1966 ntohs(encap->tot_len) + pimlen > skb->len)
1969 read_lock(&mrt_lock);
1970 if (mrt->mroute_reg_vif_num >= 0)
1971 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1972 read_unlock(&mrt_lock);
1974 if (reg_dev == NULL)
1977 skb->mac_header = skb->network_header;
1978 skb_pull(skb, (u8 *)encap - skb->data);
1979 skb_reset_network_header(skb);
1980 skb->protocol = htons(ETH_P_IP);
1981 skb->ip_summed = CHECKSUM_NONE;
1982 skb->pkt_type = PACKET_HOST;
1984 skb_tunnel_rx(skb, reg_dev);
1988 return NET_RX_SUCCESS;
1992 #ifdef CONFIG_IP_PIMSM_V1
1994 * Handle IGMP messages of PIMv1
1997 int pim_rcv_v1(struct sk_buff *skb)
1999 struct igmphdr *pim;
2000 struct net *net = dev_net(skb->dev);
2001 struct mr_table *mrt;
2003 if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2006 pim = igmp_hdr(skb);
2008 mrt = ipmr_rt_fib_lookup(net, skb);
2011 if (!mrt->mroute_do_pim ||
2012 pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2015 if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2023 #ifdef CONFIG_IP_PIMSM_V2
2024 static int pim_rcv(struct sk_buff *skb)
2026 struct pimreghdr *pim;
2027 struct net *net = dev_net(skb->dev);
2028 struct mr_table *mrt;
2030 if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2033 pim = (struct pimreghdr *)skb_transport_header(skb);
2034 if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2035 (pim->flags & PIM_NULL_REGISTER) ||
2036 (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2037 csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2040 mrt = ipmr_rt_fib_lookup(net, skb);
2043 if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2051 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2052 struct mfc_cache *c, struct rtmsg *rtm)
2055 struct rtnexthop *nhp;
2056 struct nlattr *mp_attr;
2057 struct rta_mfc_stats mfcs;
2059 /* If cache is unresolved, don't try to parse IIF and OIF */
2060 if (c->mfc_parent >= MAXVIFS)
2063 if (VIF_EXISTS(mrt, c->mfc_parent) &&
2064 nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2067 if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2070 for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2071 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2072 if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2073 nla_nest_cancel(skb, mp_attr);
2077 nhp->rtnh_flags = 0;
2078 nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2079 nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2080 nhp->rtnh_len = sizeof(*nhp);
2084 nla_nest_end(skb, mp_attr);
2086 mfcs.mfcs_packets = c->mfc_un.res.pkt;
2087 mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2088 mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2089 if (nla_put(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs) < 0)
2092 rtm->rtm_type = RTN_MULTICAST;
2096 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2097 __be32 saddr, __be32 daddr,
2098 struct rtmsg *rtm, int nowait)
2100 struct mfc_cache *cache;
2101 struct mr_table *mrt;
2104 mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2109 cache = ipmr_cache_find(mrt, saddr, daddr);
2111 if (cache == NULL) {
2112 struct sk_buff *skb2;
2114 struct net_device *dev;
2123 read_lock(&mrt_lock);
2125 vif = ipmr_find_vif(mrt, dev);
2127 read_unlock(&mrt_lock);
2131 skb2 = skb_clone(skb, GFP_ATOMIC);
2133 read_unlock(&mrt_lock);
2138 skb_push(skb2, sizeof(struct iphdr));
2139 skb_reset_network_header(skb2);
2141 iph->ihl = sizeof(struct iphdr) >> 2;
2145 err = ipmr_cache_unresolved(mrt, vif, skb2);
2146 read_unlock(&mrt_lock);
2151 read_lock(&mrt_lock);
2152 if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2153 cache->mfc_flags |= MFC_NOTIFY;
2154 err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2155 read_unlock(&mrt_lock);
2160 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2161 u32 portid, u32 seq, struct mfc_cache *c, int cmd)
2163 struct nlmsghdr *nlh;
2167 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), NLM_F_MULTI);
2171 rtm = nlmsg_data(nlh);
2172 rtm->rtm_family = RTNL_FAMILY_IPMR;
2173 rtm->rtm_dst_len = 32;
2174 rtm->rtm_src_len = 32;
2176 rtm->rtm_table = mrt->id;
2177 if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2178 goto nla_put_failure;
2179 rtm->rtm_type = RTN_MULTICAST;
2180 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
2181 if (c->mfc_flags & MFC_STATIC)
2182 rtm->rtm_protocol = RTPROT_STATIC;
2184 rtm->rtm_protocol = RTPROT_MROUTED;
2187 if (nla_put_be32(skb, RTA_SRC, c->mfc_origin) ||
2188 nla_put_be32(skb, RTA_DST, c->mfc_mcastgrp))
2189 goto nla_put_failure;
2190 err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2191 /* do not break the dump if cache is unresolved */
2192 if (err < 0 && err != -ENOENT)
2193 goto nla_put_failure;
2195 return nlmsg_end(skb, nlh);
2198 nlmsg_cancel(skb, nlh);
2202 static size_t mroute_msgsize(bool unresolved, int maxvif)
2205 NLMSG_ALIGN(sizeof(struct rtmsg))
2206 + nla_total_size(4) /* RTA_TABLE */
2207 + nla_total_size(4) /* RTA_SRC */
2208 + nla_total_size(4) /* RTA_DST */
2213 + nla_total_size(4) /* RTA_IIF */
2214 + nla_total_size(0) /* RTA_MULTIPATH */
2215 + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2217 + nla_total_size(sizeof(struct rta_mfc_stats))
2223 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2226 struct net *net = read_pnet(&mrt->net);
2227 struct sk_buff *skb;
2230 skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2235 err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd);
2239 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2245 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2248 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2250 struct net *net = sock_net(skb->sk);
2251 struct mr_table *mrt;
2252 struct mfc_cache *mfc;
2253 unsigned int t = 0, s_t;
2254 unsigned int h = 0, s_h;
2255 unsigned int e = 0, s_e;
2262 ipmr_for_each_table(mrt, net) {
2267 for (h = s_h; h < MFC_LINES; h++) {
2268 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2271 if (ipmr_fill_mroute(mrt, skb,
2272 NETLINK_CB(cb->skb).portid,
2274 mfc, RTM_NEWROUTE) < 0)
2281 spin_lock_bh(&mfc_unres_lock);
2282 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2285 if (ipmr_fill_mroute(mrt, skb,
2286 NETLINK_CB(cb->skb).portid,
2288 mfc, RTM_NEWROUTE) < 0) {
2289 spin_unlock_bh(&mfc_unres_lock);
2295 spin_unlock_bh(&mfc_unres_lock);
2311 #ifdef CONFIG_PROC_FS
2313 * The /proc interfaces to multicast routing :
2314 * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2316 struct ipmr_vif_iter {
2317 struct seq_net_private p;
2318 struct mr_table *mrt;
2322 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2323 struct ipmr_vif_iter *iter,
2326 struct mr_table *mrt = iter->mrt;
2328 for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2329 if (!VIF_EXISTS(mrt, iter->ct))
2332 return &mrt->vif_table[iter->ct];
2337 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2338 __acquires(mrt_lock)
2340 struct ipmr_vif_iter *iter = seq->private;
2341 struct net *net = seq_file_net(seq);
2342 struct mr_table *mrt;
2344 mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2346 return ERR_PTR(-ENOENT);
2350 read_lock(&mrt_lock);
2351 return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2355 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2357 struct ipmr_vif_iter *iter = seq->private;
2358 struct net *net = seq_file_net(seq);
2359 struct mr_table *mrt = iter->mrt;
2362 if (v == SEQ_START_TOKEN)
2363 return ipmr_vif_seq_idx(net, iter, 0);
2365 while (++iter->ct < mrt->maxvif) {
2366 if (!VIF_EXISTS(mrt, iter->ct))
2368 return &mrt->vif_table[iter->ct];
2373 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2374 __releases(mrt_lock)
2376 read_unlock(&mrt_lock);
2379 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2381 struct ipmr_vif_iter *iter = seq->private;
2382 struct mr_table *mrt = iter->mrt;
2384 if (v == SEQ_START_TOKEN) {
2386 "Interface BytesIn PktsIn BytesOut PktsOut Flags Local Remote\n");
2388 const struct vif_device *vif = v;
2389 const char *name = vif->dev ? vif->dev->name : "none";
2392 "%2Zd %-10s %8ld %7ld %8ld %7ld %05X %08X %08X\n",
2393 vif - mrt->vif_table,
2394 name, vif->bytes_in, vif->pkt_in,
2395 vif->bytes_out, vif->pkt_out,
2396 vif->flags, vif->local, vif->remote);
2401 static const struct seq_operations ipmr_vif_seq_ops = {
2402 .start = ipmr_vif_seq_start,
2403 .next = ipmr_vif_seq_next,
2404 .stop = ipmr_vif_seq_stop,
2405 .show = ipmr_vif_seq_show,
2408 static int ipmr_vif_open(struct inode *inode, struct file *file)
2410 return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2411 sizeof(struct ipmr_vif_iter));
2414 static const struct file_operations ipmr_vif_fops = {
2415 .owner = THIS_MODULE,
2416 .open = ipmr_vif_open,
2418 .llseek = seq_lseek,
2419 .release = seq_release_net,
2422 struct ipmr_mfc_iter {
2423 struct seq_net_private p;
2424 struct mr_table *mrt;
2425 struct list_head *cache;
2430 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2431 struct ipmr_mfc_iter *it, loff_t pos)
2433 struct mr_table *mrt = it->mrt;
2434 struct mfc_cache *mfc;
2437 for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2438 it->cache = &mrt->mfc_cache_array[it->ct];
2439 list_for_each_entry_rcu(mfc, it->cache, list)
2445 spin_lock_bh(&mfc_unres_lock);
2446 it->cache = &mrt->mfc_unres_queue;
2447 list_for_each_entry(mfc, it->cache, list)
2450 spin_unlock_bh(&mfc_unres_lock);
2457 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2459 struct ipmr_mfc_iter *it = seq->private;
2460 struct net *net = seq_file_net(seq);
2461 struct mr_table *mrt;
2463 mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2465 return ERR_PTR(-ENOENT);
2470 return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2474 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2476 struct mfc_cache *mfc = v;
2477 struct ipmr_mfc_iter *it = seq->private;
2478 struct net *net = seq_file_net(seq);
2479 struct mr_table *mrt = it->mrt;
2483 if (v == SEQ_START_TOKEN)
2484 return ipmr_mfc_seq_idx(net, seq->private, 0);
2486 if (mfc->list.next != it->cache)
2487 return list_entry(mfc->list.next, struct mfc_cache, list);
2489 if (it->cache == &mrt->mfc_unres_queue)
2492 BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2494 while (++it->ct < MFC_LINES) {
2495 it->cache = &mrt->mfc_cache_array[it->ct];
2496 if (list_empty(it->cache))
2498 return list_first_entry(it->cache, struct mfc_cache, list);
2501 /* exhausted cache_array, show unresolved */
2503 it->cache = &mrt->mfc_unres_queue;
2506 spin_lock_bh(&mfc_unres_lock);
2507 if (!list_empty(it->cache))
2508 return list_first_entry(it->cache, struct mfc_cache, list);
2511 spin_unlock_bh(&mfc_unres_lock);
2517 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2519 struct ipmr_mfc_iter *it = seq->private;
2520 struct mr_table *mrt = it->mrt;
2522 if (it->cache == &mrt->mfc_unres_queue)
2523 spin_unlock_bh(&mfc_unres_lock);
2524 else if (it->cache == &mrt->mfc_cache_array[it->ct])
2528 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2532 if (v == SEQ_START_TOKEN) {
2534 "Group Origin Iif Pkts Bytes Wrong Oifs\n");
2536 const struct mfc_cache *mfc = v;
2537 const struct ipmr_mfc_iter *it = seq->private;
2538 const struct mr_table *mrt = it->mrt;
2540 seq_printf(seq, "%08X %08X %-3hd",
2541 (__force u32) mfc->mfc_mcastgrp,
2542 (__force u32) mfc->mfc_origin,
2545 if (it->cache != &mrt->mfc_unres_queue) {
2546 seq_printf(seq, " %8lu %8lu %8lu",
2547 mfc->mfc_un.res.pkt,
2548 mfc->mfc_un.res.bytes,
2549 mfc->mfc_un.res.wrong_if);
2550 for (n = mfc->mfc_un.res.minvif;
2551 n < mfc->mfc_un.res.maxvif; n++) {
2552 if (VIF_EXISTS(mrt, n) &&
2553 mfc->mfc_un.res.ttls[n] < 255)
2556 n, mfc->mfc_un.res.ttls[n]);
2559 /* unresolved mfc_caches don't contain
2560 * pkt, bytes and wrong_if values
2562 seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2564 seq_putc(seq, '\n');
2569 static const struct seq_operations ipmr_mfc_seq_ops = {
2570 .start = ipmr_mfc_seq_start,
2571 .next = ipmr_mfc_seq_next,
2572 .stop = ipmr_mfc_seq_stop,
2573 .show = ipmr_mfc_seq_show,
2576 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2578 return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2579 sizeof(struct ipmr_mfc_iter));
2582 static const struct file_operations ipmr_mfc_fops = {
2583 .owner = THIS_MODULE,
2584 .open = ipmr_mfc_open,
2586 .llseek = seq_lseek,
2587 .release = seq_release_net,
2591 #ifdef CONFIG_IP_PIMSM_V2
2592 static const struct net_protocol pim_protocol = {
2600 * Setup for IP multicast routing
2602 static int __net_init ipmr_net_init(struct net *net)
2606 err = ipmr_rules_init(net);
2610 #ifdef CONFIG_PROC_FS
2612 if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2614 if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2615 goto proc_cache_fail;
2619 #ifdef CONFIG_PROC_FS
2621 proc_net_remove(net, "ip_mr_vif");
2623 ipmr_rules_exit(net);
2629 static void __net_exit ipmr_net_exit(struct net *net)
2631 #ifdef CONFIG_PROC_FS
2632 proc_net_remove(net, "ip_mr_cache");
2633 proc_net_remove(net, "ip_mr_vif");
2635 ipmr_rules_exit(net);
2638 static struct pernet_operations ipmr_net_ops = {
2639 .init = ipmr_net_init,
2640 .exit = ipmr_net_exit,
2643 int __init ip_mr_init(void)
2647 mrt_cachep = kmem_cache_create("ip_mrt_cache",
2648 sizeof(struct mfc_cache),
2649 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2654 err = register_pernet_subsys(&ipmr_net_ops);
2656 goto reg_pernet_fail;
2658 err = register_netdevice_notifier(&ip_mr_notifier);
2660 goto reg_notif_fail;
2661 #ifdef CONFIG_IP_PIMSM_V2
2662 if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2663 pr_err("%s: can't add PIM protocol\n", __func__);
2665 goto add_proto_fail;
2668 rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2669 NULL, ipmr_rtm_dumproute, NULL);
2672 #ifdef CONFIG_IP_PIMSM_V2
2674 unregister_netdevice_notifier(&ip_mr_notifier);
2677 unregister_pernet_subsys(&ipmr_net_ops);
2679 kmem_cache_destroy(mrt_cachep);