2 * Linux IPv6 multicast routing support for BSD pim6sd
3 * Based on net/ipv4/ipmr.c.
5 * (c) 2004 Mickael Hoerdt, <hoerdt@clarinet.u-strasbg.fr>
6 * LSIIT Laboratory, Strasbourg, France
7 * (c) 2004 Jean-Philippe Andriot, <jean-philippe.andriot@6WIND.com>
9 * Copyright (C)2007,2008 USAGI/WIDE Project
10 * YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
19 #include <asm/system.h>
20 #include <asm/uaccess.h>
21 #include <linux/types.h>
22 #include <linux/sched.h>
23 #include <linux/errno.h>
24 #include <linux/timer.h>
26 #include <linux/kernel.h>
27 #include <linux/fcntl.h>
28 #include <linux/stat.h>
29 #include <linux/socket.h>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/inetdevice.h>
33 #include <linux/proc_fs.h>
34 #include <linux/seq_file.h>
35 #include <linux/init.h>
36 #include <net/protocol.h>
37 #include <linux/skbuff.h>
40 #include <linux/notifier.h>
41 #include <linux/if_arp.h>
42 #include <net/checksum.h>
43 #include <net/netlink.h>
46 #include <net/ip6_route.h>
47 #include <linux/mroute6.h>
48 #include <linux/pim.h>
49 #include <net/addrconf.h>
50 #include <linux/netfilter_ipv6.h>
52 struct sock *mroute6_socket;
55 /* Big lock, protecting vif table, mrt cache and mroute socket state.
56 Note that the changes are semaphored via rtnl_lock.
59 static DEFINE_RWLOCK(mrt_lock);
62 * Multicast router control variables
65 static struct mif_device vif6_table[MAXMIFS]; /* Devices */
68 #define MIF_EXISTS(idx) (vif6_table[idx].dev != NULL)
70 static int mroute_do_assert; /* Set in PIM assert */
71 #ifdef CONFIG_IPV6_PIMSM_V2
72 static int mroute_do_pim;
74 #define mroute_do_pim 0
77 static struct mfc6_cache *mfc6_cache_array[MFC6_LINES]; /* Forwarding cache */
79 static struct mfc6_cache *mfc_unres_queue; /* Queue of unresolved entries */
80 static atomic_t cache_resolve_queue_len; /* Size of unresolved */
82 /* Special spinlock for queue of unresolved entries */
83 static DEFINE_SPINLOCK(mfc_unres_lock);
85 /* We return to original Alan's scheme. Hash table of resolved
86 entries is changed only in process context and protected
87 with weak lock mrt_lock. Queue of unresolved entries is protected
88 with strong spinlock mfc_unres_lock.
90 In this case data path is free of exclusive locks at all.
93 static struct kmem_cache *mrt_cachep __read_mostly;
95 static int ip6_mr_forward(struct sk_buff *skb, struct mfc6_cache *cache);
96 static int ip6mr_cache_report(struct sk_buff *pkt, mifi_t mifi, int assert);
97 static int ip6mr_fill_mroute(struct sk_buff *skb, struct mfc6_cache *c, struct rtmsg *rtm);
99 #ifdef CONFIG_IPV6_PIMSM_V2
100 static struct inet6_protocol pim6_protocol;
103 static struct timer_list ipmr_expire_timer;
106 #ifdef CONFIG_PROC_FS
108 struct ipmr_mfc_iter {
109 struct mfc6_cache **cache;
114 static struct mfc6_cache *ipmr_mfc_seq_idx(struct ipmr_mfc_iter *it, loff_t pos)
116 struct mfc6_cache *mfc;
118 it->cache = mfc6_cache_array;
119 read_lock(&mrt_lock);
120 for (it->ct = 0; it->ct < ARRAY_SIZE(mfc6_cache_array); it->ct++)
121 for (mfc = mfc6_cache_array[it->ct]; mfc; mfc = mfc->next)
124 read_unlock(&mrt_lock);
126 it->cache = &mfc_unres_queue;
127 spin_lock_bh(&mfc_unres_lock);
128 for (mfc = mfc_unres_queue; mfc; mfc = mfc->next)
131 spin_unlock_bh(&mfc_unres_lock);
141 * The /proc interfaces to multicast routing /proc/ip6_mr_cache /proc/ip6_mr_vif
144 struct ipmr_vif_iter {
148 static struct mif_device *ip6mr_vif_seq_idx(struct ipmr_vif_iter *iter,
151 for (iter->ct = 0; iter->ct < maxvif; ++iter->ct) {
152 if (!MIF_EXISTS(iter->ct))
155 return &vif6_table[iter->ct];
160 static void *ip6mr_vif_seq_start(struct seq_file *seq, loff_t *pos)
163 read_lock(&mrt_lock);
164 return (*pos ? ip6mr_vif_seq_idx(seq->private, *pos - 1)
168 static void *ip6mr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
170 struct ipmr_vif_iter *iter = seq->private;
173 if (v == SEQ_START_TOKEN)
174 return ip6mr_vif_seq_idx(iter, 0);
176 while (++iter->ct < maxvif) {
177 if (!MIF_EXISTS(iter->ct))
179 return &vif6_table[iter->ct];
184 static void ip6mr_vif_seq_stop(struct seq_file *seq, void *v)
187 read_unlock(&mrt_lock);
190 static int ip6mr_vif_seq_show(struct seq_file *seq, void *v)
192 if (v == SEQ_START_TOKEN) {
194 "Interface BytesIn PktsIn BytesOut PktsOut Flags\n");
196 const struct mif_device *vif = v;
197 const char *name = vif->dev ? vif->dev->name : "none";
200 "%2Zd %-10s %8ld %7ld %8ld %7ld %05X\n",
202 name, vif->bytes_in, vif->pkt_in,
203 vif->bytes_out, vif->pkt_out,
209 static struct seq_operations ip6mr_vif_seq_ops = {
210 .start = ip6mr_vif_seq_start,
211 .next = ip6mr_vif_seq_next,
212 .stop = ip6mr_vif_seq_stop,
213 .show = ip6mr_vif_seq_show,
216 static int ip6mr_vif_open(struct inode *inode, struct file *file)
218 return seq_open_private(file, &ip6mr_vif_seq_ops,
219 sizeof(struct ipmr_vif_iter));
222 static struct file_operations ip6mr_vif_fops = {
223 .owner = THIS_MODULE,
224 .open = ip6mr_vif_open,
227 .release = seq_release,
230 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
232 return (*pos ? ipmr_mfc_seq_idx(seq->private, *pos - 1)
236 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
238 struct mfc6_cache *mfc = v;
239 struct ipmr_mfc_iter *it = seq->private;
243 if (v == SEQ_START_TOKEN)
244 return ipmr_mfc_seq_idx(seq->private, 0);
249 if (it->cache == &mfc_unres_queue)
252 BUG_ON(it->cache != mfc6_cache_array);
254 while (++it->ct < ARRAY_SIZE(mfc6_cache_array)) {
255 mfc = mfc6_cache_array[it->ct];
260 /* exhausted cache_array, show unresolved */
261 read_unlock(&mrt_lock);
262 it->cache = &mfc_unres_queue;
265 spin_lock_bh(&mfc_unres_lock);
266 mfc = mfc_unres_queue;
271 spin_unlock_bh(&mfc_unres_lock);
277 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
279 struct ipmr_mfc_iter *it = seq->private;
281 if (it->cache == &mfc_unres_queue)
282 spin_unlock_bh(&mfc_unres_lock);
283 else if (it->cache == mfc6_cache_array)
284 read_unlock(&mrt_lock);
287 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
291 if (v == SEQ_START_TOKEN) {
295 "Iif Pkts Bytes Wrong Oifs\n");
297 const struct mfc6_cache *mfc = v;
298 const struct ipmr_mfc_iter *it = seq->private;
301 NIP6_FMT " " NIP6_FMT " %-3d %8ld %8ld %8ld",
302 NIP6(mfc->mf6c_mcastgrp), NIP6(mfc->mf6c_origin),
305 mfc->mfc_un.res.bytes,
306 mfc->mfc_un.res.wrong_if);
308 if (it->cache != &mfc_unres_queue) {
309 for (n = mfc->mfc_un.res.minvif;
310 n < mfc->mfc_un.res.maxvif; n++) {
312 mfc->mfc_un.res.ttls[n] < 255)
315 n, mfc->mfc_un.res.ttls[n]);
323 static struct seq_operations ipmr_mfc_seq_ops = {
324 .start = ipmr_mfc_seq_start,
325 .next = ipmr_mfc_seq_next,
326 .stop = ipmr_mfc_seq_stop,
327 .show = ipmr_mfc_seq_show,
330 static int ipmr_mfc_open(struct inode *inode, struct file *file)
332 return seq_open_private(file, &ipmr_mfc_seq_ops,
333 sizeof(struct ipmr_mfc_iter));
336 static struct file_operations ip6mr_mfc_fops = {
337 .owner = THIS_MODULE,
338 .open = ipmr_mfc_open,
341 .release = seq_release,
345 #ifdef CONFIG_IPV6_PIMSM_V2
346 static int reg_vif_num = -1;
348 static int pim6_rcv(struct sk_buff *skb)
350 struct pimreghdr *pim;
351 struct ipv6hdr *encap;
352 struct net_device *reg_dev = NULL;
354 if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(*encap)))
357 pim = (struct pimreghdr *)skb_transport_header(skb);
358 if (pim->type != ((PIM_VERSION << 4) | PIM_REGISTER) ||
359 (pim->flags & PIM_NULL_REGISTER) ||
360 (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
361 csum_fold(skb_checksum(skb, 0, skb->len, 0))))
364 /* check if the inner packet is destined to mcast group */
365 encap = (struct ipv6hdr *)(skb_transport_header(skb) +
368 if (!ipv6_addr_is_multicast(&encap->daddr) ||
369 encap->payload_len == 0 ||
370 ntohs(encap->payload_len) + sizeof(*pim) > skb->len)
373 read_lock(&mrt_lock);
374 if (reg_vif_num >= 0)
375 reg_dev = vif6_table[reg_vif_num].dev;
378 read_unlock(&mrt_lock);
383 skb->mac_header = skb->network_header;
384 skb_pull(skb, (u8 *)encap - skb->data);
385 skb_reset_network_header(skb);
387 skb->protocol = htons(ETH_P_IP);
389 skb->pkt_type = PACKET_HOST;
390 dst_release(skb->dst);
391 reg_dev->stats.rx_bytes += skb->len;
392 reg_dev->stats.rx_packets++;
403 static struct inet6_protocol pim6_protocol = {
407 /* Service routines creating virtual interfaces: PIMREG */
409 static int reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
411 read_lock(&mrt_lock);
412 dev->stats.tx_bytes += skb->len;
413 dev->stats.tx_packets++;
414 ip6mr_cache_report(skb, reg_vif_num, MRT6MSG_WHOLEPKT);
415 read_unlock(&mrt_lock);
420 static void reg_vif_setup(struct net_device *dev)
422 dev->type = ARPHRD_PIMREG;
423 dev->mtu = 1500 - sizeof(struct ipv6hdr) - 8;
424 dev->flags = IFF_NOARP;
425 dev->hard_start_xmit = reg_vif_xmit;
426 dev->destructor = free_netdev;
429 static struct net_device *ip6mr_reg_vif(void)
431 struct net_device *dev;
433 dev = alloc_netdev(0, "pim6reg", reg_vif_setup);
437 if (register_netdevice(dev)) {
449 /* allow the register to be completed before unregistering. */
453 unregister_netdevice(dev);
462 static int mif6_delete(int vifi)
464 struct mif_device *v;
465 struct net_device *dev;
466 if (vifi < 0 || vifi >= maxvif)
467 return -EADDRNOTAVAIL;
469 v = &vif6_table[vifi];
471 write_lock_bh(&mrt_lock);
476 write_unlock_bh(&mrt_lock);
477 return -EADDRNOTAVAIL;
480 #ifdef CONFIG_IPV6_PIMSM_V2
481 if (vifi == reg_vif_num)
485 if (vifi + 1 == maxvif) {
487 for (tmp = vifi - 1; tmp >= 0; tmp--) {
494 write_unlock_bh(&mrt_lock);
496 dev_set_allmulti(dev, -1);
498 if (v->flags & MIFF_REGISTER)
499 unregister_netdevice(dev);
505 /* Destroy an unresolved cache entry, killing queued skbs
506 and reporting error to netlink readers.
509 static void ip6mr_destroy_unres(struct mfc6_cache *c)
513 atomic_dec(&cache_resolve_queue_len);
515 while((skb = skb_dequeue(&c->mfc_un.unres.unresolved)) != NULL) {
516 if (ipv6_hdr(skb)->version == 0) {
517 struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct ipv6hdr));
518 nlh->nlmsg_type = NLMSG_ERROR;
519 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
520 skb_trim(skb, nlh->nlmsg_len);
521 ((struct nlmsgerr *)NLMSG_DATA(nlh))->error = -ETIMEDOUT;
522 rtnl_unicast(skb, &init_net, NETLINK_CB(skb).pid);
527 kmem_cache_free(mrt_cachep, c);
531 /* Single timer process for all the unresolved queue. */
533 static void ipmr_do_expire_process(unsigned long dummy)
535 unsigned long now = jiffies;
536 unsigned long expires = 10 * HZ;
537 struct mfc6_cache *c, **cp;
539 cp = &mfc_unres_queue;
541 while ((c = *cp) != NULL) {
542 if (time_after(c->mfc_un.unres.expires, now)) {
544 unsigned long interval = c->mfc_un.unres.expires - now;
545 if (interval < expires)
552 ip6mr_destroy_unres(c);
555 if (atomic_read(&cache_resolve_queue_len))
556 mod_timer(&ipmr_expire_timer, jiffies + expires);
559 static void ipmr_expire_process(unsigned long dummy)
561 if (!spin_trylock(&mfc_unres_lock)) {
562 mod_timer(&ipmr_expire_timer, jiffies + 1);
566 if (atomic_read(&cache_resolve_queue_len))
567 ipmr_do_expire_process(dummy);
569 spin_unlock(&mfc_unres_lock);
572 /* Fill oifs list. It is called under write locked mrt_lock. */
574 static void ip6mr_update_thresholds(struct mfc6_cache *cache, unsigned char *ttls)
578 cache->mfc_un.res.minvif = MAXMIFS;
579 cache->mfc_un.res.maxvif = 0;
580 memset(cache->mfc_un.res.ttls, 255, MAXMIFS);
582 for (vifi = 0; vifi < maxvif; vifi++) {
583 if (MIF_EXISTS(vifi) && ttls[vifi] && ttls[vifi] < 255) {
584 cache->mfc_un.res.ttls[vifi] = ttls[vifi];
585 if (cache->mfc_un.res.minvif > vifi)
586 cache->mfc_un.res.minvif = vifi;
587 if (cache->mfc_un.res.maxvif <= vifi)
588 cache->mfc_un.res.maxvif = vifi + 1;
593 static int mif6_add(struct mif6ctl *vifc, int mrtsock)
595 int vifi = vifc->mif6c_mifi;
596 struct mif_device *v = &vif6_table[vifi];
597 struct net_device *dev;
600 if (MIF_EXISTS(vifi))
603 switch (vifc->mif6c_flags) {
604 #ifdef CONFIG_IPV6_PIMSM_V2
607 * Special Purpose VIF in PIM
608 * All the packets will be sent to the daemon
610 if (reg_vif_num >= 0)
612 dev = ip6mr_reg_vif();
618 dev = dev_get_by_index(&init_net, vifc->mif6c_pifi);
620 return -EADDRNOTAVAIL;
627 dev_set_allmulti(dev, 1);
630 * Fill in the VIF structures
632 v->rate_limit = vifc->vifc_rate_limit;
633 v->flags = vifc->mif6c_flags;
635 v->flags |= VIFF_STATIC;
636 v->threshold = vifc->vifc_threshold;
641 v->link = dev->ifindex;
642 if (v->flags & MIFF_REGISTER)
643 v->link = dev->iflink;
645 /* And finish update writing critical data */
646 write_lock_bh(&mrt_lock);
649 #ifdef CONFIG_IPV6_PIMSM_V2
650 if (v->flags & MIFF_REGISTER)
653 if (vifi + 1 > maxvif)
655 write_unlock_bh(&mrt_lock);
659 static struct mfc6_cache *ip6mr_cache_find(struct in6_addr *origin, struct in6_addr *mcastgrp)
661 int line = MFC6_HASH(mcastgrp, origin);
662 struct mfc6_cache *c;
664 for (c = mfc6_cache_array[line]; c; c = c->next) {
665 if (ipv6_addr_equal(&c->mf6c_origin, origin) &&
666 ipv6_addr_equal(&c->mf6c_mcastgrp, mcastgrp))
673 * Allocate a multicast cache entry
675 static struct mfc6_cache *ip6mr_cache_alloc(void)
677 struct mfc6_cache *c = kmem_cache_alloc(mrt_cachep, GFP_KERNEL);
680 memset(c, 0, sizeof(*c));
681 c->mfc_un.res.minvif = MAXMIFS;
685 static struct mfc6_cache *ip6mr_cache_alloc_unres(void)
687 struct mfc6_cache *c = kmem_cache_alloc(mrt_cachep, GFP_ATOMIC);
690 memset(c, 0, sizeof(*c));
691 skb_queue_head_init(&c->mfc_un.unres.unresolved);
692 c->mfc_un.unres.expires = jiffies + 10 * HZ;
697 * A cache entry has gone into a resolved state from queued
700 static void ip6mr_cache_resolve(struct mfc6_cache *uc, struct mfc6_cache *c)
705 * Play the pending entries through our router
708 while((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
709 if (ipv6_hdr(skb)->version == 0) {
711 struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct ipv6hdr));
713 if (ip6mr_fill_mroute(skb, c, NLMSG_DATA(nlh)) > 0) {
714 nlh->nlmsg_len = skb_tail_pointer(skb) - (u8 *)nlh;
716 nlh->nlmsg_type = NLMSG_ERROR;
717 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
718 skb_trim(skb, nlh->nlmsg_len);
719 ((struct nlmsgerr *)NLMSG_DATA(nlh))->error = -EMSGSIZE;
721 err = rtnl_unicast(skb, &init_net, NETLINK_CB(skb).pid);
723 ip6_mr_forward(skb, c);
728 * Bounce a cache query up to pim6sd. We could use netlink for this but pim6sd
729 * expects the following bizarre scheme.
731 * Called under mrt_lock.
734 static int ip6mr_cache_report(struct sk_buff *pkt, mifi_t mifi, int assert)
740 #ifdef CONFIG_IPV6_PIMSM_V2
741 if (assert == MRT6MSG_WHOLEPKT)
742 skb = skb_realloc_headroom(pkt, -skb_network_offset(pkt)
746 skb = alloc_skb(sizeof(struct ipv6hdr) + sizeof(*msg), GFP_ATOMIC);
751 /* I suppose that internal messages
752 * do not require checksums */
754 skb->ip_summed = CHECKSUM_UNNECESSARY;
756 #ifdef CONFIG_IPV6_PIMSM_V2
757 if (assert == MRT6MSG_WHOLEPKT) {
758 /* Ugly, but we have no choice with this interface.
759 Duplicate old header, fix length etc.
760 And all this only to mangle msg->im6_msgtype and
761 to set msg->im6_mbz to "mbz" :-)
763 skb_push(skb, -skb_network_offset(pkt));
765 skb_push(skb, sizeof(*msg));
766 skb_reset_transport_header(skb);
767 msg = (struct mrt6msg *)skb_transport_header(skb);
769 msg->im6_msgtype = MRT6MSG_WHOLEPKT;
770 msg->im6_mif = reg_vif_num;
772 ipv6_addr_copy(&msg->im6_src, &ipv6_hdr(pkt)->saddr);
773 ipv6_addr_copy(&msg->im6_dst, &ipv6_hdr(pkt)->daddr);
775 skb->ip_summed = CHECKSUM_UNNECESSARY;
783 skb_put(skb, sizeof(struct ipv6hdr));
784 skb_reset_network_header(skb);
785 skb_copy_to_linear_data(skb, ipv6_hdr(pkt), sizeof(struct ipv6hdr));
790 skb_put(skb, sizeof(*msg));
791 skb_reset_transport_header(skb);
792 msg = (struct mrt6msg *)skb_transport_header(skb);
795 msg->im6_msgtype = assert;
798 ipv6_addr_copy(&msg->im6_src, &ipv6_hdr(pkt)->saddr);
799 ipv6_addr_copy(&msg->im6_dst, &ipv6_hdr(pkt)->daddr);
801 skb->dst = dst_clone(pkt->dst);
802 skb->ip_summed = CHECKSUM_UNNECESSARY;
804 skb_pull(skb, sizeof(struct ipv6hdr));
807 if (mroute6_socket == NULL) {
813 * Deliver to user space multicast routing algorithms
815 if ((ret = sock_queue_rcv_skb(mroute6_socket, skb)) < 0) {
817 printk(KERN_WARNING "mroute6: pending queue full, dropping entries.\n");
825 * Queue a packet for resolution. It gets locked cache entry!
829 ip6mr_cache_unresolved(mifi_t mifi, struct sk_buff *skb)
832 struct mfc6_cache *c;
834 spin_lock_bh(&mfc_unres_lock);
835 for (c = mfc_unres_queue; c; c = c->next) {
836 if (ipv6_addr_equal(&c->mf6c_mcastgrp, &ipv6_hdr(skb)->daddr) &&
837 ipv6_addr_equal(&c->mf6c_origin, &ipv6_hdr(skb)->saddr))
843 * Create a new entry if allowable
846 if (atomic_read(&cache_resolve_queue_len) >= 10 ||
847 (c = ip6mr_cache_alloc_unres()) == NULL) {
848 spin_unlock_bh(&mfc_unres_lock);
855 * Fill in the new cache entry
858 c->mf6c_origin = ipv6_hdr(skb)->saddr;
859 c->mf6c_mcastgrp = ipv6_hdr(skb)->daddr;
862 * Reflect first query at pim6sd
864 if ((err = ip6mr_cache_report(skb, mifi, MRT6MSG_NOCACHE)) < 0) {
865 /* If the report failed throw the cache entry
868 spin_unlock_bh(&mfc_unres_lock);
870 kmem_cache_free(mrt_cachep, c);
875 atomic_inc(&cache_resolve_queue_len);
876 c->next = mfc_unres_queue;
879 ipmr_do_expire_process(1);
883 * See if we can append the packet
885 if (c->mfc_un.unres.unresolved.qlen > 3) {
889 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
893 spin_unlock_bh(&mfc_unres_lock);
898 * MFC6 cache manipulation by user space
901 static int ip6mr_mfc_delete(struct mf6cctl *mfc)
904 struct mfc6_cache *c, **cp;
906 line = MFC6_HASH(&mfc->mf6cc_mcastgrp.sin6_addr, &mfc->mf6cc_origin.sin6_addr);
908 for (cp = &mfc6_cache_array[line]; (c = *cp) != NULL; cp = &c->next) {
909 if (ipv6_addr_equal(&c->mf6c_origin, &mfc->mf6cc_origin.sin6_addr) &&
910 ipv6_addr_equal(&c->mf6c_mcastgrp, &mfc->mf6cc_mcastgrp.sin6_addr)) {
911 write_lock_bh(&mrt_lock);
913 write_unlock_bh(&mrt_lock);
915 kmem_cache_free(mrt_cachep, c);
922 static int ip6mr_device_event(struct notifier_block *this,
923 unsigned long event, void *ptr)
925 struct net_device *dev = ptr;
926 struct mif_device *v;
929 if (dev_net(dev) != &init_net)
932 if (event != NETDEV_UNREGISTER)
936 for (ct = 0; ct < maxvif; ct++, v++) {
943 static struct notifier_block ip6_mr_notifier = {
944 .notifier_call = ip6mr_device_event
948 * Setup for IP multicast routing
951 void __init ip6_mr_init(void)
953 mrt_cachep = kmem_cache_create("ip6_mrt_cache",
954 sizeof(struct mfc6_cache),
955 0, SLAB_HWCACHE_ALIGN,
958 panic("cannot allocate ip6_mrt_cache");
960 setup_timer(&ipmr_expire_timer, ipmr_expire_process, 0);
961 register_netdevice_notifier(&ip6_mr_notifier);
962 #ifdef CONFIG_PROC_FS
963 proc_net_fops_create(&init_net, "ip6_mr_vif", 0, &ip6mr_vif_fops);
964 proc_net_fops_create(&init_net, "ip6_mr_cache", 0, &ip6mr_mfc_fops);
969 static int ip6mr_mfc_add(struct mf6cctl *mfc, int mrtsock)
972 struct mfc6_cache *uc, *c, **cp;
973 unsigned char ttls[MAXMIFS];
976 memset(ttls, 255, MAXMIFS);
977 for (i = 0; i < MAXMIFS; i++) {
978 if (IF_ISSET(i, &mfc->mf6cc_ifset))
983 line = MFC6_HASH(&mfc->mf6cc_mcastgrp.sin6_addr, &mfc->mf6cc_origin.sin6_addr);
985 for (cp = &mfc6_cache_array[line]; (c = *cp) != NULL; cp = &c->next) {
986 if (ipv6_addr_equal(&c->mf6c_origin, &mfc->mf6cc_origin.sin6_addr) &&
987 ipv6_addr_equal(&c->mf6c_mcastgrp, &mfc->mf6cc_mcastgrp.sin6_addr))
992 write_lock_bh(&mrt_lock);
993 c->mf6c_parent = mfc->mf6cc_parent;
994 ip6mr_update_thresholds(c, ttls);
996 c->mfc_flags |= MFC_STATIC;
997 write_unlock_bh(&mrt_lock);
1001 if (!ipv6_addr_is_multicast(&mfc->mf6cc_mcastgrp.sin6_addr))
1004 c = ip6mr_cache_alloc();
1008 c->mf6c_origin = mfc->mf6cc_origin.sin6_addr;
1009 c->mf6c_mcastgrp = mfc->mf6cc_mcastgrp.sin6_addr;
1010 c->mf6c_parent = mfc->mf6cc_parent;
1011 ip6mr_update_thresholds(c, ttls);
1013 c->mfc_flags |= MFC_STATIC;
1015 write_lock_bh(&mrt_lock);
1016 c->next = mfc6_cache_array[line];
1017 mfc6_cache_array[line] = c;
1018 write_unlock_bh(&mrt_lock);
1021 * Check to see if we resolved a queued list. If so we
1022 * need to send on the frames and tidy up.
1024 spin_lock_bh(&mfc_unres_lock);
1025 for (cp = &mfc_unres_queue; (uc = *cp) != NULL;
1027 if (ipv6_addr_equal(&uc->mf6c_origin, &c->mf6c_origin) &&
1028 ipv6_addr_equal(&uc->mf6c_mcastgrp, &c->mf6c_mcastgrp)) {
1030 if (atomic_dec_and_test(&cache_resolve_queue_len))
1031 del_timer(&ipmr_expire_timer);
1035 spin_unlock_bh(&mfc_unres_lock);
1038 ip6mr_cache_resolve(uc, c);
1039 kmem_cache_free(mrt_cachep, uc);
1045 * Close the multicast socket, and clear the vif tables etc
1048 static void mroute_clean_tables(struct sock *sk)
1053 * Shut down all active vif entries
1055 for (i = 0; i < maxvif; i++) {
1056 if (!(vif6_table[i].flags & VIFF_STATIC))
1063 for (i = 0; i < ARRAY_SIZE(mfc6_cache_array); i++) {
1064 struct mfc6_cache *c, **cp;
1066 cp = &mfc6_cache_array[i];
1067 while ((c = *cp) != NULL) {
1068 if (c->mfc_flags & MFC_STATIC) {
1072 write_lock_bh(&mrt_lock);
1074 write_unlock_bh(&mrt_lock);
1076 kmem_cache_free(mrt_cachep, c);
1080 if (atomic_read(&cache_resolve_queue_len) != 0) {
1081 struct mfc6_cache *c;
1083 spin_lock_bh(&mfc_unres_lock);
1084 while (mfc_unres_queue != NULL) {
1085 c = mfc_unres_queue;
1086 mfc_unres_queue = c->next;
1087 spin_unlock_bh(&mfc_unres_lock);
1089 ip6mr_destroy_unres(c);
1091 spin_lock_bh(&mfc_unres_lock);
1093 spin_unlock_bh(&mfc_unres_lock);
1097 static int ip6mr_sk_init(struct sock *sk)
1102 write_lock_bh(&mrt_lock);
1103 if (likely(mroute6_socket == NULL))
1104 mroute6_socket = sk;
1107 write_unlock_bh(&mrt_lock);
1114 int ip6mr_sk_done(struct sock *sk)
1119 if (sk == mroute6_socket) {
1120 write_lock_bh(&mrt_lock);
1121 mroute6_socket = NULL;
1122 write_unlock_bh(&mrt_lock);
1124 mroute_clean_tables(sk);
1133 * Socket options and virtual interface manipulation. The whole
1134 * virtual interface system is a complete heap, but unfortunately
1135 * that's how BSD mrouted happens to think. Maybe one day with a proper
1136 * MOSPF/PIM router set up we can clean this up.
1139 int ip6_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, int optlen)
1146 if (optname != MRT6_INIT) {
1147 if (sk != mroute6_socket && !capable(CAP_NET_ADMIN))
1153 if (sk->sk_type != SOCK_RAW ||
1154 inet_sk(sk)->num != IPPROTO_ICMPV6)
1156 if (optlen < sizeof(int))
1159 return ip6mr_sk_init(sk);
1162 return ip6mr_sk_done(sk);
1165 if (optlen < sizeof(vif))
1167 if (copy_from_user(&vif, optval, sizeof(vif)))
1169 if (vif.mif6c_mifi >= MAXMIFS)
1172 ret = mif6_add(&vif, sk == mroute6_socket);
1177 if (optlen < sizeof(mifi_t))
1179 if (copy_from_user(&mifi, optval, sizeof(mifi_t)))
1182 ret = mif6_delete(mifi);
1187 * Manipulate the forwarding caches. These live
1188 * in a sort of kernel/user symbiosis.
1192 if (optlen < sizeof(mfc))
1194 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1197 if (optname == MRT6_DEL_MFC)
1198 ret = ip6mr_mfc_delete(&mfc);
1200 ret = ip6mr_mfc_add(&mfc, sk == mroute6_socket);
1205 * Control PIM assert (to activate pim will activate assert)
1210 if (get_user(v, (int __user *)optval))
1212 mroute_do_assert = !!v;
1216 #ifdef CONFIG_IPV6_PIMSM_V2
1220 if (get_user(v, (int __user *)optval))
1225 if (v != mroute_do_pim) {
1227 mroute_do_assert = v;
1229 ret = inet6_add_protocol(&pim6_protocol,
1232 ret = inet6_del_protocol(&pim6_protocol,
1243 * Spurious command, or MRT_VERSION which you cannot
1247 return -ENOPROTOOPT;
1252 * Getsock opt support for the multicast routing system.
1255 int ip6_mroute_getsockopt(struct sock *sk, int optname, char __user *optval,
1265 #ifdef CONFIG_IPV6_PIMSM_V2
1267 val = mroute_do_pim;
1271 val = mroute_do_assert;
1274 return -ENOPROTOOPT;
1277 if (get_user(olr, optlen))
1280 olr = min_t(int, olr, sizeof(int));
1284 if (put_user(olr, optlen))
1286 if (copy_to_user(optval, &val, olr))
1292 * The IP multicast ioctl support routines.
1295 int ip6mr_ioctl(struct sock *sk, int cmd, void __user *arg)
1297 struct sioc_sg_req6 sr;
1298 struct sioc_mif_req6 vr;
1299 struct mif_device *vif;
1300 struct mfc6_cache *c;
1303 case SIOCGETMIFCNT_IN6:
1304 if (copy_from_user(&vr, arg, sizeof(vr)))
1306 if (vr.mifi >= maxvif)
1308 read_lock(&mrt_lock);
1309 vif = &vif6_table[vr.mifi];
1310 if (MIF_EXISTS(vr.mifi)) {
1311 vr.icount = vif->pkt_in;
1312 vr.ocount = vif->pkt_out;
1313 vr.ibytes = vif->bytes_in;
1314 vr.obytes = vif->bytes_out;
1315 read_unlock(&mrt_lock);
1317 if (copy_to_user(arg, &vr, sizeof(vr)))
1321 read_unlock(&mrt_lock);
1322 return -EADDRNOTAVAIL;
1323 case SIOCGETSGCNT_IN6:
1324 if (copy_from_user(&sr, arg, sizeof(sr)))
1327 read_lock(&mrt_lock);
1328 c = ip6mr_cache_find(&sr.src.sin6_addr, &sr.grp.sin6_addr);
1330 sr.pktcnt = c->mfc_un.res.pkt;
1331 sr.bytecnt = c->mfc_un.res.bytes;
1332 sr.wrong_if = c->mfc_un.res.wrong_if;
1333 read_unlock(&mrt_lock);
1335 if (copy_to_user(arg, &sr, sizeof(sr)))
1339 read_unlock(&mrt_lock);
1340 return -EADDRNOTAVAIL;
1342 return -ENOIOCTLCMD;
1347 static inline int ip6mr_forward2_finish(struct sk_buff *skb)
1349 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
1350 return dst_output(skb);
1354 * Processing handlers for ip6mr_forward
1357 static int ip6mr_forward2(struct sk_buff *skb, struct mfc6_cache *c, int vifi)
1359 struct ipv6hdr *ipv6h;
1360 struct mif_device *vif = &vif6_table[vifi];
1361 struct net_device *dev;
1362 struct dst_entry *dst;
1365 if (vif->dev == NULL)
1368 #ifdef CONFIG_IPV6_PIMSM_V2
1369 if (vif->flags & MIFF_REGISTER) {
1371 vif->bytes_out += skb->len;
1372 vif->dev->stats.tx_bytes += skb->len;
1373 vif->dev->stats.tx_packets++;
1374 ip6mr_cache_report(skb, vifi, MRT6MSG_WHOLEPKT);
1380 ipv6h = ipv6_hdr(skb);
1382 fl = (struct flowi) {
1385 { .daddr = ipv6h->daddr, }
1389 dst = ip6_route_output(&init_net, NULL, &fl);
1393 dst_release(skb->dst);
1397 * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1398 * not only before forwarding, but after forwarding on all output
1399 * interfaces. It is clear, if mrouter runs a multicasting
1400 * program, it should receive packets not depending to what interface
1401 * program is joined.
1402 * If we will not make it, the program will have to join on all
1403 * interfaces. On the other hand, multihoming host (or router, but
1404 * not mrouter) cannot join to more than one interface - it will
1405 * result in receiving multiple packets.
1410 vif->bytes_out += skb->len;
1412 /* We are about to write */
1413 /* XXX: extension headers? */
1414 if (skb_cow(skb, sizeof(*ipv6h) + LL_RESERVED_SPACE(dev)))
1417 ipv6h = ipv6_hdr(skb);
1420 IP6CB(skb)->flags |= IP6SKB_FORWARDED;
1422 return NF_HOOK(PF_INET6, NF_INET_FORWARD, skb, skb->dev, dev,
1423 ip6mr_forward2_finish);
1430 static int ip6mr_find_vif(struct net_device *dev)
1433 for (ct = maxvif - 1; ct >= 0; ct--) {
1434 if (vif6_table[ct].dev == dev)
1440 static int ip6_mr_forward(struct sk_buff *skb, struct mfc6_cache *cache)
1445 vif = cache->mf6c_parent;
1446 cache->mfc_un.res.pkt++;
1447 cache->mfc_un.res.bytes += skb->len;
1450 * Wrong interface: drop packet and (maybe) send PIM assert.
1452 if (vif6_table[vif].dev != skb->dev) {
1455 cache->mfc_un.res.wrong_if++;
1456 true_vifi = ip6mr_find_vif(skb->dev);
1458 if (true_vifi >= 0 && mroute_do_assert &&
1459 /* pimsm uses asserts, when switching from RPT to SPT,
1460 so that we cannot check that packet arrived on an oif.
1461 It is bad, but otherwise we would need to move pretty
1462 large chunk of pimd to kernel. Ough... --ANK
1464 (mroute_do_pim || cache->mfc_un.res.ttls[true_vifi] < 255) &&
1466 cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1467 cache->mfc_un.res.last_assert = jiffies;
1468 ip6mr_cache_report(skb, true_vifi, MRT6MSG_WRONGMIF);
1473 vif6_table[vif].pkt_in++;
1474 vif6_table[vif].bytes_in += skb->len;
1479 for (ct = cache->mfc_un.res.maxvif - 1; ct >= cache->mfc_un.res.minvif; ct--) {
1480 if (ipv6_hdr(skb)->hop_limit > cache->mfc_un.res.ttls[ct]) {
1482 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1484 ip6mr_forward2(skb2, cache, psend);
1490 ip6mr_forward2(skb, cache, psend);
1501 * Multicast packets for forwarding arrive here
1504 int ip6_mr_input(struct sk_buff *skb)
1506 struct mfc6_cache *cache;
1508 read_lock(&mrt_lock);
1509 cache = ip6mr_cache_find(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr);
1512 * No usable cache entry
1514 if (cache == NULL) {
1517 vif = ip6mr_find_vif(skb->dev);
1519 int err = ip6mr_cache_unresolved(vif, skb);
1520 read_unlock(&mrt_lock);
1524 read_unlock(&mrt_lock);
1529 ip6_mr_forward(skb, cache);
1531 read_unlock(&mrt_lock);
1538 ip6mr_fill_mroute(struct sk_buff *skb, struct mfc6_cache *c, struct rtmsg *rtm)
1541 struct rtnexthop *nhp;
1542 struct net_device *dev = vif6_table[c->mf6c_parent].dev;
1543 u8 *b = skb_tail_pointer(skb);
1544 struct rtattr *mp_head;
1547 RTA_PUT(skb, RTA_IIF, 4, &dev->ifindex);
1549 mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
1551 for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
1552 if (c->mfc_un.res.ttls[ct] < 255) {
1553 if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
1554 goto rtattr_failure;
1555 nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
1556 nhp->rtnh_flags = 0;
1557 nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
1558 nhp->rtnh_ifindex = vif6_table[ct].dev->ifindex;
1559 nhp->rtnh_len = sizeof(*nhp);
1562 mp_head->rta_type = RTA_MULTIPATH;
1563 mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
1564 rtm->rtm_type = RTN_MULTICAST;
1572 int ip6mr_get_route(struct sk_buff *skb, struct rtmsg *rtm, int nowait)
1575 struct mfc6_cache *cache;
1576 struct rt6_info *rt = (struct rt6_info *)skb->dst;
1578 read_lock(&mrt_lock);
1579 cache = ip6mr_cache_find(&rt->rt6i_src.addr, &rt->rt6i_dst.addr);
1582 struct sk_buff *skb2;
1583 struct ipv6hdr *iph;
1584 struct net_device *dev;
1588 read_unlock(&mrt_lock);
1593 if (dev == NULL || (vif = ip6mr_find_vif(dev)) < 0) {
1594 read_unlock(&mrt_lock);
1598 /* really correct? */
1599 skb2 = alloc_skb(sizeof(struct ipv6hdr), GFP_ATOMIC);
1601 read_unlock(&mrt_lock);
1605 skb_reset_transport_header(skb2);
1607 skb_put(skb2, sizeof(struct ipv6hdr));
1608 skb_reset_network_header(skb2);
1610 iph = ipv6_hdr(skb2);
1613 iph->flow_lbl[0] = 0;
1614 iph->flow_lbl[1] = 0;
1615 iph->flow_lbl[2] = 0;
1616 iph->payload_len = 0;
1617 iph->nexthdr = IPPROTO_NONE;
1619 ipv6_addr_copy(&iph->saddr, &rt->rt6i_src.addr);
1620 ipv6_addr_copy(&iph->daddr, &rt->rt6i_dst.addr);
1622 err = ip6mr_cache_unresolved(vif, skb2);
1623 read_unlock(&mrt_lock);
1628 if (!nowait && (rtm->rtm_flags&RTM_F_NOTIFY))
1629 cache->mfc_flags |= MFC_NOTIFY;
1631 err = ip6mr_fill_mroute(skb, cache, rtm);
1632 read_unlock(&mrt_lock);