2 * IP multicast routing support for mrouted 3.6/3.8
4 * (c) 1995 Alan Cox, <alan@redhat.com>
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.
12 * Version: $Id: ipmr.c,v 1.65 2001/10/31 21:55:54 davem Exp $
15 * Michael Chastain : Incorrect size of copying.
16 * Alan Cox : Added the cache manager code
17 * Alan Cox : Fixed the clone/copy bug and device race.
18 * Mike McLagan : Routing by source
19 * Malcolm Beattie : Buffer handling fixes.
20 * Alexey Kuznetsov : Double buffer free and other fixes.
21 * SVR Anand : Fixed several multicast bugs and problems.
22 * Alexey Kuznetsov : Status, optimisations and more.
23 * Brad Parker : Better behaviour on mrouted upcall
25 * Carlos Picoto : PIMv1 Support
26 * Pavlin Ivanov Radoslavov: PIMv2 Registers must checksum only PIM header
27 * Relax this requrement to work with older peers.
31 #include <linux/config.h>
32 #include <asm/system.h>
33 #include <asm/uaccess.h>
34 #include <linux/types.h>
35 #include <linux/sched.h>
36 #include <linux/errno.h>
37 #include <linux/timer.h>
39 #include <linux/kernel.h>
40 #include <linux/fcntl.h>
41 #include <linux/stat.h>
42 #include <linux/socket.h>
44 #include <linux/inet.h>
45 #include <linux/netdevice.h>
46 #include <linux/inetdevice.h>
47 #include <linux/igmp.h>
48 #include <linux/proc_fs.h>
49 #include <linux/mroute.h>
50 #include <linux/init.h>
52 #include <net/protocol.h>
53 #include <linux/skbuff.h>
58 #include <linux/notifier.h>
59 #include <linux/if_arp.h>
60 #include <linux/netfilter_ipv4.h>
62 #include <net/checksum.h>
64 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
65 #define CONFIG_IP_PIMSM 1
68 static struct sock *mroute_socket;
71 /* Big lock, protecting vif table, mrt cache and mroute socket state.
72 Note that the changes are semaphored via rtnl_lock.
75 static rwlock_t mrt_lock = RW_LOCK_UNLOCKED;
78 * Multicast router control variables
81 static struct vif_device vif_table[MAXVIFS]; /* Devices */
84 #define VIF_EXISTS(idx) (vif_table[idx].dev != NULL)
86 int mroute_do_assert; /* Set in PIM assert */
89 static struct mfc_cache *mfc_cache_array[MFC_LINES]; /* Forwarding cache */
91 static struct mfc_cache *mfc_unres_queue; /* Queue of unresolved entries */
92 atomic_t cache_resolve_queue_len; /* Size of unresolved */
94 /* Special spinlock for queue of unresolved entries */
95 static spinlock_t mfc_unres_lock = SPIN_LOCK_UNLOCKED;
97 /* We return to original Alan's scheme. Hash table of resolved
98 entries is changed only in process context and protected
99 with weak lock mrt_lock. Queue of unresolved entries is protected
100 with strong spinlock mfc_unres_lock.
102 In this case data path is free of exclusive locks at all.
105 kmem_cache_t *mrt_cachep;
107 static int ip_mr_forward(struct sk_buff *skb, struct mfc_cache *cache, int local);
108 static int ipmr_cache_report(struct sk_buff *pkt, vifi_t vifi, int assert);
109 static int ipmr_fill_mroute(struct sk_buff *skb, struct mfc_cache *c, struct rtmsg *rtm);
111 extern struct inet_protocol pim_protocol;
113 static struct timer_list ipmr_expire_timer;
115 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
118 struct net_device *ipmr_new_tunnel(struct vifctl *v)
120 struct net_device *dev;
122 dev = __dev_get_by_name("tunl0");
128 struct ip_tunnel_parm p;
129 struct in_device *in_dev;
131 memset(&p, 0, sizeof(p));
132 p.iph.daddr = v->vifc_rmt_addr.s_addr;
133 p.iph.saddr = v->vifc_lcl_addr.s_addr;
136 p.iph.protocol = IPPROTO_IPIP;
137 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
138 ifr.ifr_ifru.ifru_data = (void*)&p;
140 oldfs = get_fs(); set_fs(KERNEL_DS);
141 err = dev->do_ioctl(dev, &ifr, SIOCADDTUNNEL);
146 if (err == 0 && (dev = __dev_get_by_name(p.name)) != NULL) {
147 dev->flags |= IFF_MULTICAST;
149 in_dev = __in_dev_get(dev);
150 if (in_dev == NULL && (in_dev = inetdev_init(dev)) == NULL)
152 in_dev->cnf.rp_filter = 0;
161 unregister_netdevice(dev);
165 #ifdef CONFIG_IP_PIMSM
167 static int reg_vif_num = -1;
169 static int reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
171 read_lock(&mrt_lock);
172 ((struct net_device_stats*)dev->priv)->tx_bytes += skb->len;
173 ((struct net_device_stats*)dev->priv)->tx_packets++;
174 ipmr_cache_report(skb, reg_vif_num, IGMPMSG_WHOLEPKT);
175 read_unlock(&mrt_lock);
180 static struct net_device_stats *reg_vif_get_stats(struct net_device *dev)
182 return (struct net_device_stats*)dev->priv;
186 struct net_device *ipmr_reg_vif(struct vifctl *v)
188 struct net_device *dev;
189 struct in_device *in_dev;
192 size = sizeof(*dev) + sizeof(struct net_device_stats);
193 dev = kmalloc(size, GFP_KERNEL);
197 memset(dev, 0, size);
201 strcpy(dev->name, "pimreg");
203 dev->type = ARPHRD_PIMREG;
204 dev->mtu = 1500 - sizeof(struct iphdr) - 8;
205 dev->flags = IFF_NOARP;
206 dev->hard_start_xmit = reg_vif_xmit;
207 dev->get_stats = reg_vif_get_stats;
208 dev->features |= NETIF_F_DYNALLOC;
210 if (register_netdevice(dev)) {
216 if ((in_dev = inetdev_init(dev)) == NULL)
219 in_dev->cnf.rp_filter = 0;
227 unregister_netdevice(dev);
236 static int vif_delete(int vifi)
238 struct vif_device *v;
239 struct net_device *dev;
240 struct in_device *in_dev;
242 if (vifi < 0 || vifi >= maxvif)
243 return -EADDRNOTAVAIL;
245 v = &vif_table[vifi];
247 write_lock_bh(&mrt_lock);
252 write_unlock_bh(&mrt_lock);
253 return -EADDRNOTAVAIL;
256 #ifdef CONFIG_IP_PIMSM
257 if (vifi == reg_vif_num)
261 if (vifi+1 == maxvif) {
263 for (tmp=vifi-1; tmp>=0; tmp--) {
270 write_unlock_bh(&mrt_lock);
272 dev_set_allmulti(dev, -1);
274 if ((in_dev = __in_dev_get(dev)) != NULL) {
275 in_dev->cnf.mc_forwarding--;
276 ip_rt_multicast_event(in_dev);
279 if (v->flags&(VIFF_TUNNEL|VIFF_REGISTER))
280 unregister_netdevice(dev);
286 /* Destroy an unresolved cache entry, killing queued skbs
287 and reporting error to netlink readers.
290 static void ipmr_destroy_unres(struct mfc_cache *c)
294 atomic_dec(&cache_resolve_queue_len);
296 while((skb=skb_dequeue(&c->mfc_un.unres.unresolved))) {
297 if (skb->nh.iph->version == 0) {
298 struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
299 nlh->nlmsg_type = NLMSG_ERROR;
300 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
301 skb_trim(skb, nlh->nlmsg_len);
302 ((struct nlmsgerr*)NLMSG_DATA(nlh))->error = -ETIMEDOUT;
303 netlink_unicast(rtnl, skb, NETLINK_CB(skb).dst_pid, MSG_DONTWAIT);
308 kmem_cache_free(mrt_cachep, c);
312 /* Single timer process for all the unresolved queue. */
314 void ipmr_expire_process(unsigned long dummy)
317 unsigned long expires;
318 struct mfc_cache *c, **cp;
320 if (!spin_trylock(&mfc_unres_lock)) {
321 mod_timer(&ipmr_expire_timer, jiffies+HZ/10);
325 if (atomic_read(&cache_resolve_queue_len) == 0)
330 cp = &mfc_unres_queue;
332 while ((c=*cp) != NULL) {
333 long interval = c->mfc_un.unres.expires - now;
336 if (interval < expires)
344 ipmr_destroy_unres(c);
347 if (atomic_read(&cache_resolve_queue_len))
348 mod_timer(&ipmr_expire_timer, jiffies + expires);
351 spin_unlock(&mfc_unres_lock);
354 /* Fill oifs list. It is called under write locked mrt_lock. */
356 static void ipmr_update_threshoulds(struct mfc_cache *cache, unsigned char *ttls)
360 cache->mfc_un.res.minvif = MAXVIFS;
361 cache->mfc_un.res.maxvif = 0;
362 memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
364 for (vifi=0; vifi<maxvif; vifi++) {
365 if (VIF_EXISTS(vifi) && ttls[vifi] && ttls[vifi] < 255) {
366 cache->mfc_un.res.ttls[vifi] = ttls[vifi];
367 if (cache->mfc_un.res.minvif > vifi)
368 cache->mfc_un.res.minvif = vifi;
369 if (cache->mfc_un.res.maxvif <= vifi)
370 cache->mfc_un.res.maxvif = vifi + 1;
375 static int vif_add(struct vifctl *vifc, int mrtsock)
377 int vifi = vifc->vifc_vifi;
378 struct vif_device *v = &vif_table[vifi];
379 struct net_device *dev;
380 struct in_device *in_dev;
383 if (VIF_EXISTS(vifi))
386 switch (vifc->vifc_flags) {
387 #ifdef CONFIG_IP_PIMSM
390 * Special Purpose VIF in PIM
391 * All the packets will be sent to the daemon
393 if (reg_vif_num >= 0)
395 dev = ipmr_reg_vif(vifc);
401 dev = ipmr_new_tunnel(vifc);
406 dev=ip_dev_find(vifc->vifc_lcl_addr.s_addr);
408 return -EADDRNOTAVAIL;
415 if ((in_dev = __in_dev_get(dev)) == NULL)
416 return -EADDRNOTAVAIL;
417 in_dev->cnf.mc_forwarding++;
418 dev_set_allmulti(dev, +1);
419 ip_rt_multicast_event(in_dev);
422 * Fill in the VIF structures
424 v->rate_limit=vifc->vifc_rate_limit;
425 v->local=vifc->vifc_lcl_addr.s_addr;
426 v->remote=vifc->vifc_rmt_addr.s_addr;
427 v->flags=vifc->vifc_flags;
429 v->flags |= VIFF_STATIC;
430 v->threshold=vifc->vifc_threshold;
435 v->link = dev->ifindex;
436 if (v->flags&(VIFF_TUNNEL|VIFF_REGISTER))
437 v->link = dev->iflink;
439 /* And finish update writing critical data */
440 write_lock_bh(&mrt_lock);
443 #ifdef CONFIG_IP_PIMSM
444 if (v->flags&VIFF_REGISTER)
449 write_unlock_bh(&mrt_lock);
453 static struct mfc_cache *ipmr_cache_find(__u32 origin, __u32 mcastgrp)
455 int line=MFC_HASH(mcastgrp,origin);
458 for (c=mfc_cache_array[line]; c; c = c->next) {
459 if (c->mfc_origin==origin && c->mfc_mcastgrp==mcastgrp)
466 * Allocate a multicast cache entry
468 static struct mfc_cache *ipmr_cache_alloc(void)
470 struct mfc_cache *c=kmem_cache_alloc(mrt_cachep, GFP_KERNEL);
473 memset(c, 0, sizeof(*c));
474 c->mfc_un.res.minvif = MAXVIFS;
478 static struct mfc_cache *ipmr_cache_alloc_unres(void)
480 struct mfc_cache *c=kmem_cache_alloc(mrt_cachep, GFP_ATOMIC);
483 memset(c, 0, sizeof(*c));
484 skb_queue_head_init(&c->mfc_un.unres.unresolved);
485 c->mfc_un.unres.expires = jiffies + 10*HZ;
490 * A cache entry has gone into a resolved state from queued
493 static void ipmr_cache_resolve(struct mfc_cache *uc, struct mfc_cache *c)
498 * Play the pending entries through our router
501 while((skb=__skb_dequeue(&uc->mfc_un.unres.unresolved))) {
502 if (skb->nh.iph->version == 0) {
504 struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
506 if (ipmr_fill_mroute(skb, c, NLMSG_DATA(nlh)) > 0) {
507 nlh->nlmsg_len = skb->tail - (u8*)nlh;
509 nlh->nlmsg_type = NLMSG_ERROR;
510 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
511 skb_trim(skb, nlh->nlmsg_len);
512 ((struct nlmsgerr*)NLMSG_DATA(nlh))->error = -EMSGSIZE;
514 err = netlink_unicast(rtnl, skb, NETLINK_CB(skb).dst_pid, MSG_DONTWAIT);
516 ip_mr_forward(skb, c, 0);
521 * Bounce a cache query up to mrouted. We could use netlink for this but mrouted
522 * expects the following bizarre scheme.
524 * Called under mrt_lock.
527 static int ipmr_cache_report(struct sk_buff *pkt, vifi_t vifi, int assert)
530 int ihl = pkt->nh.iph->ihl<<2;
531 struct igmphdr *igmp;
535 #ifdef CONFIG_IP_PIMSM
536 if (assert == IGMPMSG_WHOLEPKT)
537 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
540 skb = alloc_skb(128, GFP_ATOMIC);
545 #ifdef CONFIG_IP_PIMSM
546 if (assert == IGMPMSG_WHOLEPKT) {
547 /* Ugly, but we have no choice with this interface.
548 Duplicate old header, fix ihl, length etc.
549 And all this only to mangle msg->im_msgtype and
550 to set msg->im_mbz to "mbz" :-)
552 msg = (struct igmpmsg*)skb_push(skb, sizeof(struct iphdr));
553 skb->nh.raw = skb->h.raw = (u8*)msg;
554 memcpy(msg, pkt->nh.raw, sizeof(struct iphdr));
555 msg->im_msgtype = IGMPMSG_WHOLEPKT;
557 msg->im_vif = reg_vif_num;
558 skb->nh.iph->ihl = sizeof(struct iphdr) >> 2;
559 skb->nh.iph->tot_len = htons(ntohs(pkt->nh.iph->tot_len) + sizeof(struct iphdr));
568 skb->nh.iph = (struct iphdr *)skb_put(skb, ihl);
569 memcpy(skb->data,pkt->data,ihl);
570 skb->nh.iph->protocol = 0; /* Flag to the kernel this is a route add */
571 msg = (struct igmpmsg*)skb->nh.iph;
573 skb->dst = dst_clone(pkt->dst);
579 igmp=(struct igmphdr *)skb_put(skb,sizeof(struct igmphdr));
581 msg->im_msgtype = assert;
583 skb->nh.iph->tot_len=htons(skb->len); /* Fix the length */
584 skb->h.raw = skb->nh.raw;
587 if (mroute_socket == NULL) {
595 if ((ret=sock_queue_rcv_skb(mroute_socket,skb))<0) {
597 printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
605 * Queue a packet for resolution. It gets locked cache entry!
609 ipmr_cache_unresolved(vifi_t vifi, struct sk_buff *skb)
614 spin_lock_bh(&mfc_unres_lock);
615 for (c=mfc_unres_queue; c; c=c->next) {
616 if (c->mfc_mcastgrp == skb->nh.iph->daddr &&
617 c->mfc_origin == skb->nh.iph->saddr)
623 * Create a new entry if allowable
626 if (atomic_read(&cache_resolve_queue_len)>=10 ||
627 (c=ipmr_cache_alloc_unres())==NULL) {
628 spin_unlock_bh(&mfc_unres_lock);
635 * Fill in the new cache entry
638 c->mfc_origin=skb->nh.iph->saddr;
639 c->mfc_mcastgrp=skb->nh.iph->daddr;
642 * Reflect first query at mrouted.
644 if ((err = ipmr_cache_report(skb, vifi, IGMPMSG_NOCACHE))<0) {
645 /* If the report failed throw the cache entry
648 spin_unlock_bh(&mfc_unres_lock);
650 kmem_cache_free(mrt_cachep, c);
655 atomic_inc(&cache_resolve_queue_len);
656 c->next = mfc_unres_queue;
659 mod_timer(&ipmr_expire_timer, c->mfc_un.unres.expires);
663 * See if we can append the packet
665 if (c->mfc_un.unres.unresolved.qlen>3) {
669 skb_queue_tail(&c->mfc_un.unres.unresolved,skb);
673 spin_unlock_bh(&mfc_unres_lock);
678 * MFC cache manipulation by user space mroute daemon
681 int ipmr_mfc_delete(struct mfcctl *mfc)
684 struct mfc_cache *c, **cp;
686 line=MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
688 for (cp=&mfc_cache_array[line]; (c=*cp) != NULL; cp = &c->next) {
689 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
690 c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
691 write_lock_bh(&mrt_lock);
693 write_unlock_bh(&mrt_lock);
695 kmem_cache_free(mrt_cachep, c);
702 int ipmr_mfc_add(struct mfcctl *mfc, int mrtsock)
705 struct mfc_cache *uc, *c, **cp;
707 line=MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
709 for (cp=&mfc_cache_array[line]; (c=*cp) != NULL; cp = &c->next) {
710 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
711 c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr)
716 write_lock_bh(&mrt_lock);
717 c->mfc_parent = mfc->mfcc_parent;
718 ipmr_update_threshoulds(c, mfc->mfcc_ttls);
720 c->mfc_flags |= MFC_STATIC;
721 write_unlock_bh(&mrt_lock);
725 if(!MULTICAST(mfc->mfcc_mcastgrp.s_addr))
728 c=ipmr_cache_alloc();
732 c->mfc_origin=mfc->mfcc_origin.s_addr;
733 c->mfc_mcastgrp=mfc->mfcc_mcastgrp.s_addr;
734 c->mfc_parent=mfc->mfcc_parent;
735 ipmr_update_threshoulds(c, mfc->mfcc_ttls);
737 c->mfc_flags |= MFC_STATIC;
739 write_lock_bh(&mrt_lock);
740 c->next = mfc_cache_array[line];
741 mfc_cache_array[line] = c;
742 write_unlock_bh(&mrt_lock);
745 * Check to see if we resolved a queued list. If so we
746 * need to send on the frames and tidy up.
748 spin_lock_bh(&mfc_unres_lock);
749 for (cp = &mfc_unres_queue; (uc=*cp) != NULL;
751 if (uc->mfc_origin == c->mfc_origin &&
752 uc->mfc_mcastgrp == c->mfc_mcastgrp) {
754 if (atomic_dec_and_test(&cache_resolve_queue_len))
755 del_timer(&ipmr_expire_timer);
759 spin_unlock_bh(&mfc_unres_lock);
762 ipmr_cache_resolve(uc, c);
763 kmem_cache_free(mrt_cachep, uc);
769 * Close the multicast socket, and clear the vif tables etc
772 static void mroute_clean_tables(struct sock *sk)
777 * Shut down all active vif entries
779 for(i=0; i<maxvif; i++) {
780 if (!(vif_table[i].flags&VIFF_STATIC))
787 for (i=0;i<MFC_LINES;i++) {
788 struct mfc_cache *c, **cp;
790 cp = &mfc_cache_array[i];
791 while ((c = *cp) != NULL) {
792 if (c->mfc_flags&MFC_STATIC) {
796 write_lock_bh(&mrt_lock);
798 write_unlock_bh(&mrt_lock);
800 kmem_cache_free(mrt_cachep, c);
804 if (atomic_read(&cache_resolve_queue_len) != 0) {
807 spin_lock_bh(&mfc_unres_lock);
808 while (mfc_unres_queue != NULL) {
810 mfc_unres_queue = c->next;
811 spin_unlock_bh(&mfc_unres_lock);
813 ipmr_destroy_unres(c);
815 spin_lock_bh(&mfc_unres_lock);
817 spin_unlock_bh(&mfc_unres_lock);
821 static void mrtsock_destruct(struct sock *sk)
824 if (sk == mroute_socket) {
825 ipv4_devconf.mc_forwarding--;
827 write_lock_bh(&mrt_lock);
829 write_unlock_bh(&mrt_lock);
831 mroute_clean_tables(sk);
837 * Socket options and virtual interface manipulation. The whole
838 * virtual interface system is a complete heap, but unfortunately
839 * that's how BSD mrouted happens to think. Maybe one day with a proper
840 * MOSPF/PIM router set up we can clean this up.
843 int ip_mroute_setsockopt(struct sock *sk,int optname,char *optval,int optlen)
849 if(optname!=MRT_INIT)
851 if(sk!=mroute_socket && !capable(CAP_NET_ADMIN))
858 if(sk->type!=SOCK_RAW || sk->num!=IPPROTO_IGMP)
860 if(optlen!=sizeof(int))
869 ret = ip_ra_control(sk, 1, mrtsock_destruct);
871 write_lock_bh(&mrt_lock);
873 write_unlock_bh(&mrt_lock);
875 ipv4_devconf.mc_forwarding++;
880 if (sk!=mroute_socket)
882 return ip_ra_control(sk, 0, NULL);
885 if(optlen!=sizeof(vif))
887 if (copy_from_user(&vif,optval,sizeof(vif)))
889 if(vif.vifc_vifi >= MAXVIFS)
892 if (optname==MRT_ADD_VIF) {
893 ret = vif_add(&vif, sk==mroute_socket);
895 ret = vif_delete(vif.vifc_vifi);
901 * Manipulate the forwarding caches. These live
902 * in a sort of kernel/user symbiosis.
906 if(optlen!=sizeof(mfc))
908 if (copy_from_user(&mfc,optval, sizeof(mfc)))
911 if (optname==MRT_DEL_MFC)
912 ret = ipmr_mfc_delete(&mfc);
914 ret = ipmr_mfc_add(&mfc, sk==mroute_socket);
918 * Control PIM assert.
923 if(get_user(v,(int *)optval))
925 mroute_do_assert=(v)?1:0;
928 #ifdef CONFIG_IP_PIMSM
932 if(get_user(v,(int *)optval))
936 if (v != mroute_do_pim) {
938 mroute_do_assert = v;
939 #ifdef CONFIG_IP_PIMSM_V2
941 inet_add_protocol(&pim_protocol);
943 inet_del_protocol(&pim_protocol);
951 * Spurious command, or MRT_VERSION which you cannot
960 * Getsock opt support for the multicast routing system.
963 int ip_mroute_getsockopt(struct sock *sk,int optname,char *optval,int *optlen)
968 if(optname!=MRT_VERSION &&
969 #ifdef CONFIG_IP_PIMSM
975 if (get_user(olr, optlen))
978 olr = min_t(unsigned int, olr, sizeof(int));
982 if(put_user(olr,optlen))
984 if(optname==MRT_VERSION)
986 #ifdef CONFIG_IP_PIMSM
987 else if(optname==MRT_PIM)
991 val=mroute_do_assert;
992 if(copy_to_user(optval,&val,olr))
998 * The IP multicast ioctl support routines.
1001 int ipmr_ioctl(struct sock *sk, int cmd, unsigned long arg)
1003 struct sioc_sg_req sr;
1004 struct sioc_vif_req vr;
1005 struct vif_device *vif;
1006 struct mfc_cache *c;
1011 if (copy_from_user(&vr,(void *)arg,sizeof(vr)))
1015 read_lock(&mrt_lock);
1016 vif=&vif_table[vr.vifi];
1017 if(VIF_EXISTS(vr.vifi)) {
1018 vr.icount=vif->pkt_in;
1019 vr.ocount=vif->pkt_out;
1020 vr.ibytes=vif->bytes_in;
1021 vr.obytes=vif->bytes_out;
1022 read_unlock(&mrt_lock);
1024 if (copy_to_user((void *)arg,&vr,sizeof(vr)))
1028 read_unlock(&mrt_lock);
1029 return -EADDRNOTAVAIL;
1031 if (copy_from_user(&sr,(void *)arg,sizeof(sr)))
1034 read_lock(&mrt_lock);
1035 c = ipmr_cache_find(sr.src.s_addr, sr.grp.s_addr);
1037 sr.pktcnt = c->mfc_un.res.pkt;
1038 sr.bytecnt = c->mfc_un.res.bytes;
1039 sr.wrong_if = c->mfc_un.res.wrong_if;
1040 read_unlock(&mrt_lock);
1042 if (copy_to_user((void *)arg,&sr,sizeof(sr)))
1046 read_unlock(&mrt_lock);
1047 return -EADDRNOTAVAIL;
1049 return -ENOIOCTLCMD;
1054 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1056 struct vif_device *v;
1058 if (event != NETDEV_UNREGISTER)
1061 for(ct=0;ct<maxvif;ct++,v++) {
1069 static struct notifier_block ip_mr_notifier={
1076 * Encapsulate a packet by attaching a valid IPIP header to it.
1077 * This avoids tunnel drivers and other mess and gives us the speed so
1078 * important for multicast video.
1081 static void ip_encap(struct sk_buff *skb, u32 saddr, u32 daddr)
1083 struct iphdr *iph = (struct iphdr *)skb_push(skb,sizeof(struct iphdr));
1086 iph->tos = skb->nh.iph->tos;
1087 iph->ttl = skb->nh.iph->ttl;
1091 iph->protocol = IPPROTO_IPIP;
1093 iph->tot_len = htons(skb->len);
1094 ip_select_ident(iph, skb->dst, NULL);
1097 skb->h.ipiph = skb->nh.iph;
1099 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1103 static inline int ipmr_forward_finish(struct sk_buff *skb)
1105 struct ip_options *opt = &(IPCB(skb)->opt);
1106 struct dst_entry *dst = skb->dst;
1108 if (unlikely(opt->optlen))
1109 ip_forward_options(skb);
1111 if (skb->len <= dst->pmtu)
1112 return dst->output(skb);
1114 return ip_fragment(skb, dst->output);
1118 * Processing handlers for ipmr_forward
1121 static void ipmr_queue_xmit(struct sk_buff *skb, struct mfc_cache *c,
1124 struct iphdr *iph = skb->nh.iph;
1125 struct vif_device *vif = &vif_table[vifi];
1126 struct net_device *dev;
1129 struct sk_buff *skb2;
1131 if (vif->dev == NULL)
1134 #ifdef CONFIG_IP_PIMSM
1135 if (vif->flags & VIFF_REGISTER) {
1137 vif->bytes_out+=skb->len;
1138 ((struct net_device_stats*)vif->dev->priv)->tx_bytes += skb->len;
1139 ((struct net_device_stats*)vif->dev->priv)->tx_packets++;
1140 ipmr_cache_report(skb, vifi, IGMPMSG_WHOLEPKT);
1145 if (vif->flags&VIFF_TUNNEL) {
1146 if (ip_route_output(&rt, vif->remote, vif->local, RT_TOS(iph->tos), vif->link))
1148 encap = sizeof(struct iphdr);
1150 if (ip_route_output(&rt, iph->daddr, 0, RT_TOS(iph->tos), vif->link))
1154 dev = rt->u.dst.dev;
1156 if (skb->len+encap > rt->u.dst.pmtu && (ntohs(iph->frag_off) & IP_DF)) {
1157 /* Do not fragment multicasts. Alas, IPv4 does not
1158 allow to send ICMP, so that packets will disappear
1162 IP_INC_STATS_BH(IpFragFails);
1167 encap += dev->hard_header_len;
1169 if (skb_headroom(skb) < encap || skb_cloned(skb) || !last)
1170 skb2 = skb_realloc_headroom(skb, (encap + 15)&~15);
1171 else if (atomic_read(&skb->users) != 1)
1172 skb2 = skb_clone(skb, GFP_ATOMIC);
1174 atomic_inc(&skb->users);
1184 vif->bytes_out+=skb->len;
1186 dst_release(skb2->dst);
1187 skb2->dst = &rt->u.dst;
1189 ip_decrease_ttl(iph);
1191 /* FIXME: forward and output firewalls used to be called here.
1192 * What do we do with netfilter? -- RR */
1193 if (vif->flags & VIFF_TUNNEL) {
1194 ip_encap(skb2, vif->local, vif->remote);
1195 /* FIXME: extra output firewall step used to be here. --RR */
1196 ((struct ip_tunnel *)vif->dev->priv)->stat.tx_packets++;
1197 ((struct ip_tunnel *)vif->dev->priv)->stat.tx_bytes+=skb2->len;
1200 IPCB(skb2)->flags |= IPSKB_FORWARDED;
1203 * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1204 * not only before forwarding, but after forwarding on all output
1205 * interfaces. It is clear, if mrouter runs a multicasting
1206 * program, it should receive packets not depending to what interface
1207 * program is joined.
1208 * If we will not make it, the program will have to join on all
1209 * interfaces. On the other hand, multihoming host (or router, but
1210 * not mrouter) cannot join to more than one interface - it will
1211 * result in receiving multiple packets.
1213 NF_HOOK(PF_INET, NF_IP_FORWARD, skb2, skb->dev, dev,
1214 ipmr_forward_finish);
1217 int ipmr_find_vif(struct net_device *dev)
1220 for (ct=maxvif-1; ct>=0; ct--) {
1221 if (vif_table[ct].dev == dev)
1227 /* "local" means that we should preserve one skb (for local delivery) */
1229 int ip_mr_forward(struct sk_buff *skb, struct mfc_cache *cache, int local)
1234 vif = cache->mfc_parent;
1235 cache->mfc_un.res.pkt++;
1236 cache->mfc_un.res.bytes += skb->len;
1239 * Wrong interface: drop packet and (maybe) send PIM assert.
1241 if (vif_table[vif].dev != skb->dev) {
1244 if (((struct rtable*)skb->dst)->key.iif == 0) {
1245 /* It is our own packet, looped back.
1246 Very complicated situation...
1248 The best workaround until routing daemons will be
1249 fixed is not to redistribute packet, if it was
1250 send through wrong interface. It means, that
1251 multicast applications WILL NOT work for
1252 (S,G), which have default multicast route pointing
1253 to wrong oif. In any case, it is not a good
1254 idea to use multicasting applications on router.
1259 cache->mfc_un.res.wrong_if++;
1260 true_vifi = ipmr_find_vif(skb->dev);
1262 if (true_vifi >= 0 && mroute_do_assert &&
1263 /* pimsm uses asserts, when switching from RPT to SPT,
1264 so that we cannot check that packet arrived on an oif.
1265 It is bad, but otherwise we would need to move pretty
1266 large chunk of pimd to kernel. Ough... --ANK
1268 (mroute_do_pim || cache->mfc_un.res.ttls[true_vifi] < 255) &&
1269 jiffies - cache->mfc_un.res.last_assert > MFC_ASSERT_THRESH) {
1270 cache->mfc_un.res.last_assert = jiffies;
1271 ipmr_cache_report(skb, true_vifi, IGMPMSG_WRONGVIF);
1276 vif_table[vif].pkt_in++;
1277 vif_table[vif].bytes_in+=skb->len;
1282 for (ct = cache->mfc_un.res.maxvif-1; ct >= cache->mfc_un.res.minvif; ct--) {
1283 if (skb->nh.iph->ttl > cache->mfc_un.res.ttls[ct]) {
1285 ipmr_queue_xmit(skb, cache, psend, 0);
1290 ipmr_queue_xmit(skb, cache, psend, !local);
1300 * Multicast packets for forwarding arrive here
1303 int ip_mr_input(struct sk_buff *skb)
1305 struct mfc_cache *cache;
1306 int local = ((struct rtable*)skb->dst)->rt_flags&RTCF_LOCAL;
1308 /* Packet is looped back after forward, it should not be
1309 forwarded second time, but still can be delivered locally.
1311 if (IPCB(skb)->flags&IPSKB_FORWARDED)
1315 if (IPCB(skb)->opt.router_alert) {
1316 if (ip_call_ra_chain(skb))
1318 } else if (skb->nh.iph->protocol == IPPROTO_IGMP){
1319 /* IGMPv1 (and broken IGMPv2 implementations sort of
1320 Cisco IOS <= 11.2(8)) do not put router alert
1321 option to IGMP packets destined to routable
1322 groups. It is very bad, because it means
1323 that we can forward NO IGMP messages.
1325 read_lock(&mrt_lock);
1326 if (mroute_socket) {
1327 raw_rcv(mroute_socket, skb);
1328 read_unlock(&mrt_lock);
1331 read_unlock(&mrt_lock);
1335 read_lock(&mrt_lock);
1336 cache = ipmr_cache_find(skb->nh.iph->saddr, skb->nh.iph->daddr);
1339 * No usable cache entry
1345 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1346 ip_local_deliver(skb);
1348 read_unlock(&mrt_lock);
1354 vif = ipmr_find_vif(skb->dev);
1356 int err = ipmr_cache_unresolved(vif, skb);
1357 read_unlock(&mrt_lock);
1361 read_unlock(&mrt_lock);
1366 ip_mr_forward(skb, cache, local);
1368 read_unlock(&mrt_lock);
1371 return ip_local_deliver(skb);
1377 return ip_local_deliver(skb);
1382 #ifdef CONFIG_IP_PIMSM_V1
1384 * Handle IGMP messages of PIMv1
1387 int pim_rcv_v1(struct sk_buff * skb)
1389 struct igmphdr *pim = (struct igmphdr*)skb->h.raw;
1390 struct iphdr *encap;
1391 struct net_device *reg_dev = NULL;
1393 if (skb_is_nonlinear(skb)) {
1394 if (skb_linearize(skb, GFP_ATOMIC) != 0) {
1398 pim = (struct igmphdr*)skb->h.raw;
1401 if (!mroute_do_pim ||
1402 skb->len < sizeof(*pim) + sizeof(*encap) ||
1403 pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER) {
1408 encap = (struct iphdr*)(skb->h.raw + sizeof(struct igmphdr));
1411 a. packet is really destinted to a multicast group
1412 b. packet is not a NULL-REGISTER
1413 c. packet is not truncated
1415 if (!MULTICAST(encap->daddr) ||
1416 ntohs(encap->tot_len) == 0 ||
1417 ntohs(encap->tot_len) + sizeof(*pim) > skb->len) {
1422 read_lock(&mrt_lock);
1423 if (reg_vif_num >= 0)
1424 reg_dev = vif_table[reg_vif_num].dev;
1427 read_unlock(&mrt_lock);
1429 if (reg_dev == NULL) {
1434 skb->mac.raw = skb->nh.raw;
1435 skb_pull(skb, (u8*)encap - skb->data);
1436 skb->nh.iph = (struct iphdr *)skb->data;
1438 memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
1439 skb->protocol = htons(ETH_P_IP);
1441 skb->pkt_type = PACKET_HOST;
1442 dst_release(skb->dst);
1444 ((struct net_device_stats*)reg_dev->priv)->rx_bytes += skb->len;
1445 ((struct net_device_stats*)reg_dev->priv)->rx_packets++;
1453 #ifdef CONFIG_IP_PIMSM_V2
1454 int pim_rcv(struct sk_buff * skb)
1456 struct pimreghdr *pim = (struct pimreghdr*)skb->h.raw;
1457 struct iphdr *encap;
1458 struct net_device *reg_dev = NULL;
1460 if (skb_is_nonlinear(skb)) {
1461 if (skb_linearize(skb, GFP_ATOMIC) != 0) {
1465 pim = (struct pimreghdr*)skb->h.raw;
1468 if (skb->len < sizeof(*pim) + sizeof(*encap) ||
1469 pim->type != ((PIM_VERSION<<4)|(PIM_REGISTER)) ||
1470 (pim->flags&PIM_NULL_REGISTER) ||
1471 (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
1472 ip_compute_csum((void *)pim, skb->len))) {
1477 /* check if the inner packet is destined to mcast group */
1478 encap = (struct iphdr*)(skb->h.raw + sizeof(struct pimreghdr));
1479 if (!MULTICAST(encap->daddr) ||
1480 ntohs(encap->tot_len) == 0 ||
1481 ntohs(encap->tot_len) + sizeof(*pim) > skb->len) {
1486 read_lock(&mrt_lock);
1487 if (reg_vif_num >= 0)
1488 reg_dev = vif_table[reg_vif_num].dev;
1491 read_unlock(&mrt_lock);
1493 if (reg_dev == NULL) {
1498 skb->mac.raw = skb->nh.raw;
1499 skb_pull(skb, (u8*)encap - skb->data);
1500 skb->nh.iph = (struct iphdr *)skb->data;
1502 memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
1503 skb->protocol = htons(ETH_P_IP);
1505 skb->pkt_type = PACKET_HOST;
1506 dst_release(skb->dst);
1507 ((struct net_device_stats*)reg_dev->priv)->rx_bytes += skb->len;
1508 ((struct net_device_stats*)reg_dev->priv)->rx_packets++;
1518 ipmr_fill_mroute(struct sk_buff *skb, struct mfc_cache *c, struct rtmsg *rtm)
1521 struct rtnexthop *nhp;
1522 struct net_device *dev = vif_table[c->mfc_parent].dev;
1524 struct rtattr *mp_head;
1527 RTA_PUT(skb, RTA_IIF, 4, &dev->ifindex);
1529 mp_head = (struct rtattr*)skb_put(skb, RTA_LENGTH(0));
1531 for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
1532 if (c->mfc_un.res.ttls[ct] < 255) {
1533 if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
1534 goto rtattr_failure;
1535 nhp = (struct rtnexthop*)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
1536 nhp->rtnh_flags = 0;
1537 nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
1538 nhp->rtnh_ifindex = vif_table[ct].dev->ifindex;
1539 nhp->rtnh_len = sizeof(*nhp);
1542 mp_head->rta_type = RTA_MULTIPATH;
1543 mp_head->rta_len = skb->tail - (u8*)mp_head;
1544 rtm->rtm_type = RTN_MULTICAST;
1548 skb_trim(skb, b - skb->data);
1552 int ipmr_get_route(struct sk_buff *skb, struct rtmsg *rtm, int nowait)
1555 struct mfc_cache *cache;
1556 struct rtable *rt = (struct rtable*)skb->dst;
1558 read_lock(&mrt_lock);
1559 cache = ipmr_cache_find(rt->rt_src, rt->rt_dst);
1562 struct net_device *dev;
1566 read_unlock(&mrt_lock);
1571 if (dev == NULL || (vif = ipmr_find_vif(dev)) < 0) {
1572 read_unlock(&mrt_lock);
1575 skb->nh.raw = skb_push(skb, sizeof(struct iphdr));
1576 skb->nh.iph->ihl = sizeof(struct iphdr)>>2;
1577 skb->nh.iph->saddr = rt->rt_src;
1578 skb->nh.iph->daddr = rt->rt_dst;
1579 skb->nh.iph->version = 0;
1580 err = ipmr_cache_unresolved(vif, skb);
1581 read_unlock(&mrt_lock);
1585 if (!nowait && (rtm->rtm_flags&RTM_F_NOTIFY))
1586 cache->mfc_flags |= MFC_NOTIFY;
1587 err = ipmr_fill_mroute(skb, cache, rtm);
1588 read_unlock(&mrt_lock);
1592 #ifdef CONFIG_PROC_FS
1594 * The /proc interfaces to multicast routing /proc/ip_mr_cache /proc/ip_mr_vif
1597 static int ipmr_vif_info(char *buffer, char **start, off_t offset, int length)
1599 struct vif_device *vif;
1606 len += sprintf(buffer,
1607 "Interface BytesIn PktsIn BytesOut PktsOut Flags Local Remote\n");
1610 read_lock(&mrt_lock);
1611 for (ct=0;ct<maxvif;ct++)
1613 char *name = "none";
1618 name = vif->dev->name;
1619 size = sprintf(buffer+len, "%2d %-10s %8ld %7ld %8ld %7ld %05X %08X %08X\n",
1620 ct, name, vif->bytes_in, vif->pkt_in, vif->bytes_out, vif->pkt_out,
1621 vif->flags, vif->local, vif->remote);
1629 if(pos>offset+length)
1632 read_unlock(&mrt_lock);
1634 *start=buffer+(offset-begin);
1635 len-=(offset-begin);
1643 static int ipmr_mfc_info(char *buffer, char **start, off_t offset, int length)
1645 struct mfc_cache *mfc;
1652 len += sprintf(buffer,
1653 "Group Origin Iif Pkts Bytes Wrong Oifs\n");
1656 read_lock(&mrt_lock);
1657 for (ct=0;ct<MFC_LINES;ct++)
1659 for(mfc=mfc_cache_array[ct]; mfc; mfc=mfc->next)
1664 * Interface forwarding map
1666 size = sprintf(buffer+len, "%08lX %08lX %-3d %8ld %8ld %8ld",
1667 (unsigned long)mfc->mfc_mcastgrp,
1668 (unsigned long)mfc->mfc_origin,
1670 mfc->mfc_un.res.pkt,
1671 mfc->mfc_un.res.bytes,
1672 mfc->mfc_un.res.wrong_if);
1673 for(n=mfc->mfc_un.res.minvif;n<mfc->mfc_un.res.maxvif;n++)
1675 if(VIF_EXISTS(n) && mfc->mfc_un.res.ttls[n] < 255)
1676 size += sprintf(buffer+len+size, " %2d:%-3d", n, mfc->mfc_un.res.ttls[n]);
1678 size += sprintf(buffer+len+size, "\n");
1686 if(pos>offset+length)
1691 spin_lock_bh(&mfc_unres_lock);
1692 for(mfc=mfc_unres_queue; mfc; mfc=mfc->next) {
1693 size = sprintf(buffer+len, "%08lX %08lX %-3d %8ld %8ld %8ld\n",
1694 (unsigned long)mfc->mfc_mcastgrp,
1695 (unsigned long)mfc->mfc_origin,
1697 (long)mfc->mfc_un.unres.unresolved.qlen,
1706 if(pos>offset+length)
1709 spin_unlock_bh(&mfc_unres_lock);
1712 read_unlock(&mrt_lock);
1713 *start=buffer+(offset-begin);
1714 len-=(offset-begin);
1725 #ifdef CONFIG_IP_PIMSM_V2
1726 struct inet_protocol pim_protocol =
1728 pim_rcv, /* PIM handler */
1729 NULL, /* PIM error control */
1731 IPPROTO_PIM, /* protocol ID */
1740 * Setup for IP multicast routing
1743 void __init ip_mr_init(void)
1745 printk(KERN_INFO "Linux IP multicast router 0.06 plus PIM-SM\n");
1746 mrt_cachep = kmem_cache_create("ip_mrt_cache",
1747 sizeof(struct mfc_cache),
1748 0, SLAB_HWCACHE_ALIGN,
1750 init_timer(&ipmr_expire_timer);
1751 ipmr_expire_timer.function=ipmr_expire_process;
1752 register_netdevice_notifier(&ip_mr_notifier);
1753 #ifdef CONFIG_PROC_FS
1754 proc_net_create("ip_mr_vif",0,ipmr_vif_info);
1755 proc_net_create("ip_mr_cache",0,ipmr_mfc_info);