2 * Copyright (c) 1998-2002 Luigi Rizzo
4 * Work partly supported by: Cisco Systems, Inc. - NSITE lab, RTP, NC
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * $FreeBSD: src/sys/net/bridge.c,v 1.16.2.25 2003/01/23 21:06:44 sam Exp $
28 * $DragonFly: src/sys/net/bridge/Attic/bridge.c,v 1.11 2004/09/15 19:29:35 joerg Exp $
32 * This code implements bridging in FreeBSD. It only acts on ethernet
33 * interfaces, including VLANs (others are still usable for routing).
34 * A FreeBSD host can implement multiple logical bridges, called
35 * "clusters". Each cluster is made of a set of interfaces, and
36 * identified by a "cluster-id" which is a number in the range 1..2^16-1.
38 * Bridging is enabled by the sysctl variable
39 * net.link.ether.bridge
40 * the grouping of interfaces into clusters is done with
41 * net.link.ether.bridge_cfg
42 * containing a list of interfaces each optionally followed by
43 * a colon and the cluster it belongs to (1 is the default).
44 * Separators can be * spaces, commas or tabs, e.g.
45 * net.link.ether.bridge_cfg="fxp0:2 fxp1:2 dc0 dc1:1"
46 * Optionally bridged packets can be passed through the firewall,
47 * this is controlled by the variable
48 * net.link.ether.bridge_ipfw
50 * For each cluster there is a descriptor (cluster_softc) storing
51 * the following data structures:
52 * - a hash table with the MAC address and destination interface for each
53 * known node. The table is indexed using a hash of the source address.
54 * - an array with the MAC addresses of the interfaces used in the cluster.
56 * Input packets are tapped near the beginning of ether_input(), and
57 * analysed by bridge_in(). Depending on the result, the packet
58 * can be forwarded to one or more output interfaces using bdg_forward(),
59 * and/or sent to the upper layer (e.g. in case of multicast).
61 * Output packets are intercepted near the end of ether_output().
62 * The correct destination is selected by bridge_dst_lookup(),
63 * and then forwarding is done by bdg_forward().
65 * The arp code is also modified to let a machine answer to requests
66 * irrespective of the port the request came from.
68 * In case of loops in the bridging topology, the bridge detects this
69 * event and temporarily mutes output bridging on one of the ports.
70 * Periodically, interfaces are unmuted by bdg_timeout().
71 * Muting is only implemented as a safety measure, and also as
72 * a mechanism to support a user-space implementation of the spanning
75 * To build a bridging kernel, use the following option
77 * and then at runtime set the sysctl variable to enable bridging.
79 * Only one interface per cluster is supposed to have addresses set (but
80 * there are no substantial problems if you set addresses for none or
81 * for more than one interface).
82 * Bridging will act before routing, but nothing prevents a machine
83 * from doing both (modulo bugs in the implementation...).
86 * - bridging is incompatible with multicast routing on the same
87 * machine. There is not an easy fix to this.
88 * - be very careful when bridging VLANs
89 * - loop detection is still not very robust.
92 #include <sys/param.h>
94 #include <sys/malloc.h>
95 #include <sys/systm.h>
96 #include <sys/socket.h> /* for net/if.h */
97 #include <sys/ctype.h> /* string functions */
98 #include <sys/kernel.h>
99 #include <sys/sysctl.h>
102 #include <net/if_types.h>
103 #include <net/if_var.h>
104 #include <net/pfil.h>
106 #include <netinet/in.h> /* for struct arpcom */
107 #include <netinet/in_systm.h>
108 #include <netinet/in_var.h>
109 #include <netinet/ip.h>
110 #include <netinet/ip_var.h>
111 #include <netinet/if_ether.h> /* for struct arpcom */
113 #include <net/route.h>
114 #include <net/ipfw/ip_fw.h>
115 #include <net/dummynet/ip_dummynet.h>
118 /*--------------------*/
121 * For each cluster, source MAC addresses are stored into a hash
122 * table which locates the port they reside on.
124 #define HASH_SIZE 8192 /* Table size, must be a power of 2 */
126 typedef struct hash_table { /* each entry. */
129 u_int16_t used; /* also, padding */
133 * The hash function applied to MAC addresses. Out of the 6 bytes,
134 * the last ones tend to vary more. Since we are on a little endian machine,
135 * we have to do some gimmick...
137 #define HASH_FN(addr) ( \
138 ntohs( ((u_int16_t *)addr)[1] ^ ((u_int16_t *)addr)[2] ) & (HASH_SIZE -1))
141 * This is the data structure where local addresses are stored.
144 u_char etheraddr[6] ;
149 * The configuration of each cluster includes the cluster id, a pointer to
150 * the hash table, and an array of local MAC addresses (of size "ports").
152 struct cluster_softc {
153 u_int16_t cluster_id;
156 struct bdg_addr *my_macs; /* local MAC addresses */
160 extern struct protosw inetsw[]; /* from netinet/ip_input.c */
162 static int n_clusters; /* number of clusters */
163 static struct cluster_softc *clusters;
165 #define BDG_MUTED(ifp) (ifp2sc[ifp->if_index].flags & IFF_MUTE)
166 #define BDG_MUTE(ifp) ifp2sc[ifp->if_index].flags |= IFF_MUTE
167 #define BDG_CLUSTER(ifp) (ifp2sc[ifp->if_index].cluster)
169 #define BDG_SAMECLUSTER(ifp,src) \
170 (src == NULL || BDG_CLUSTER(ifp) == BDG_CLUSTER(src) )
173 #define BDG_MATCH(a,b) ( \
174 ((u_int16_t *)(a))[2] == ((u_int16_t *)(b))[2] && \
175 *((u_int32_t *)(a)) == *((u_int32_t *)(b)) )
176 #define IS_ETHER_BROADCAST(a) ( \
177 *((u_int32_t *)(a)) == 0xffffffff && \
178 ((u_int16_t *)(a))[2] == 0xffff )
180 /* for machines that do not support unaligned access */
181 #define BDG_MATCH(a,b) (!bcmp(a, b, ETHER_ADDR_LEN) )
182 #define IS_ETHER_BROADCAST(a) (!bcmp(a, "\377\377\377\377\377\377", 6))
187 * For timing-related debugging, you can use the following macros.
188 * remember, rdtsc() only works on Pentium-class machines
191 DDB(ticks = rdtsc();)
192 ... interesting code ...
193 DDB(bdg_fw_ticks += (u_long)(rdtsc() - ticks) ; bdg_fw_count++ ;)
201 static int bdginit(void);
202 static void parse_bdg_cfg(void);
204 static int bdg_pfil; /* PFIL hooks enabled in bridge */
207 #if 0 /* debugging only */
208 static char *bdg_dst_names[] = {
220 * System initialization
223 static struct bdg_stats bdg_stats ;
224 static struct callout bdg_timeout_h ;
227 * Add an interface to a cluster, possibly creating a new entry in
228 * the cluster table. This requires reallocation of the table and
229 * updating pointers in ifp2sc.
231 static struct cluster_softc *
232 add_cluster(u_int16_t cluster_id, struct arpcom *ac)
234 struct cluster_softc *c = NULL;
237 for (i = 0; i < n_clusters ; i++)
238 if (clusters[i].cluster_id == cluster_id)
241 /* Not found, need to reallocate */
242 c = malloc((1+n_clusters) * sizeof (*c), M_IFADDR, M_WAITOK | M_ZERO);
243 c[n_clusters].ht = malloc(HASH_SIZE * sizeof(struct hash_table),
244 M_IFADDR, M_WAITOK | M_ZERO);
245 c[n_clusters].my_macs = malloc(BDG_MAX_PORTS * sizeof(struct bdg_addr),
246 M_IFADDR, M_WAITOK | M_ZERO);
248 c[n_clusters].cluster_id = cluster_id;
249 c[n_clusters].ports = 0;
251 * now copy old descriptors here
253 if (n_clusters > 0) {
254 for (i=0; i < n_clusters; i++)
257 * and finally update pointers in ifp2sc
259 for (i = 0 ; i < if_index && i < BDG_MAX_PORTS; i++)
260 if (ifp2sc[i].cluster != NULL)
261 ifp2sc[i].cluster = c + (ifp2sc[i].cluster - clusters);
262 free(clusters, M_IFADDR);
265 i = n_clusters; /* index of cluster entry */
268 c = clusters + i; /* the right cluster ... */
269 bcopy(ac->ac_enaddr, &(c->my_macs[c->ports]), 6);
276 * Turn off bridging, by clearing promisc mode on the interface,
277 * marking the interface as unused, and clearing the name in the
279 * Also dispose the hash tables associated with the clusters.
287 DEB(printf("bridge_off: n_clusters %d\n", n_clusters);)
288 TAILQ_FOREACH(ifp, &ifnet, if_link) {
291 if (ifp->if_index >= BDG_MAX_PORTS)
292 continue; /* make sure we do not go beyond the end */
293 b = &(ifp2sc[ifp->if_index]);
295 if ( b->flags & IFF_BDG_PROMISC ) {
299 b->flags &= ~(IFF_BDG_PROMISC|IFF_MUTE) ;
300 DEB(printf(">> now %s promisc OFF if_flags 0x%x bdg_flags 0x%x\n",
302 ifp->if_flags, b->flags);)
304 b->flags &= ~(IFF_USED) ;
306 bdg_stats.s[ifp->if_index].name[0] = '\0';
311 for (i=0; i < n_clusters; i++) {
312 free(clusters[i].ht, M_IFADDR);
313 free(clusters[i].my_macs, M_IFADDR);
315 if (clusters != NULL)
316 free(clusters, M_IFADDR);
323 * set promisc mode on the interfaces we use.
331 TAILQ_FOREACH(ifp, &ifnet, if_link) {
332 struct bdg_softc *b = &ifp2sc[ifp->if_index];
334 if ( !(b->flags & IFF_USED) )
336 if ( !( ifp->if_flags & IFF_UP) ) {
341 if ( !(b->flags & IFF_BDG_PROMISC) ) {
344 ret = ifpromisc(ifp, 1);
346 b->flags |= IFF_BDG_PROMISC ;
347 DEB(printf(">> now %s promisc ON if_flags 0x%x bdg_flags 0x%x\n",
349 ifp->if_flags, b->flags);)
351 if (b->flags & IFF_MUTE) {
352 DEB(printf(">> unmuting %s\n", ifp->if_xname);)
353 b->flags &= ~IFF_MUTE;
359 * reconfigure bridge.
360 * This is also done every time we attach or detach an interface.
361 * Main use is to make sure that we do not bridge on some old
362 * (ejected) device. So, it would be really useful to have a
363 * pointer to the modified device as an argument. Without it, we
364 * have to scan all interfaces.
367 reconfigure_bridge(void)
371 if (if_index >= BDG_MAX_PORTS) {
372 printf("-- sorry too many interfaces (%d, max is %d),"
373 " disabling bridging\n", if_index, BDG_MAX_PORTS);
382 static char bridge_cfg[1024]; /* in BSS so initialized to all NULs */
385 * parse the config string, set IFF_USED, name and cluster_id
386 * for all interfaces found.
387 * The config string is a list of "if[:cluster]" with
388 * a number of possible separators (see "sep"). In particular the
389 * use of the space lets you set bridge_cfg with the output from
397 static char *sep = ", \t";
399 for (p = bridge_cfg; *p ; p++) {
404 if (index(sep, *p)) /* skip separators */
406 /* names are lowercase and digits */
407 for ( beg = p ; islower(*p) || isdigit(*p) ; p++ )
409 l = p - beg ; /* length of name string */
410 if (l == 0) /* invalid name */
412 if ( *p != ':' ) /* no ':', assume default cluster 1 */
414 else /* fetch cluster */
415 cluster = strtoul( p+1, &p, 10);
419 * now search in interface list for a matching name
421 TAILQ_FOREACH(ifp, &ifnet, if_link) {
424 snprintf(buf, sizeof(buf), "%s", ifp->if_xname);
425 if (!strncmp(beg, buf, max(l, strlen(buf)))) {
426 struct bdg_softc *b = &ifp2sc[ifp->if_index];
427 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) {
428 printf("%s is not an ethernet, continue\n", buf);
431 if (b->flags & IFF_USED) {
432 printf("%s already used, skipping\n", buf);
435 b->cluster = add_cluster(htons(cluster), (struct arpcom *)ifp);
436 b->flags |= IFF_USED ;
437 sprintf(bdg_stats.s[ifp->if_index].name,
438 "%s:%d", ifp->if_xname, cluster);
440 DEB(printf("--++ found %s next c %d\n",
441 bdg_stats.s[ifp->if_index].name, c);)
447 printf("interface %s Not found in bridge\n", beg);
456 * handler for net.link.ether.bridge
459 sysctl_bdg(SYSCTL_HANDLER_ARGS)
461 int error, oldval = do_bridge ;
463 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
464 DEB( printf("called sysctl for bridge name %s arg2 %d val %d->%d\n",
465 oidp->oid_name, oidp->oid_arg2,
466 oldval, do_bridge); )
468 if (oldval != do_bridge)
469 reconfigure_bridge();
474 * handler for net.link.ether.bridge_cfg
477 sysctl_bdg_cfg(SYSCTL_HANDLER_ARGS)
482 strcpy(old_cfg, bridge_cfg) ;
484 error = sysctl_handle_string(oidp, bridge_cfg, oidp->oid_arg2, req);
486 printf("called sysctl for bridge name %s arg2 %d err %d val %s->%s\n",
487 oidp->oid_name, oidp->oid_arg2,
489 old_cfg, bridge_cfg);
491 if (strcmp(old_cfg, bridge_cfg))
492 reconfigure_bridge();
497 sysctl_refresh(SYSCTL_HANDLER_ARGS)
500 reconfigure_bridge();
506 SYSCTL_DECL(_net_link_ether);
507 SYSCTL_PROC(_net_link_ether, OID_AUTO, bridge_cfg, CTLTYPE_STRING|CTLFLAG_RW,
508 &bridge_cfg, sizeof(bridge_cfg), &sysctl_bdg_cfg, "A",
509 "Bridge configuration");
511 SYSCTL_PROC(_net_link_ether, OID_AUTO, bridge, CTLTYPE_INT|CTLFLAG_RW,
512 &do_bridge, 0, &sysctl_bdg, "I", "Bridging");
514 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_ipfw, CTLFLAG_RW,
515 &bdg_ipfw,0,"Pass bridged pkts through IPFW");
517 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_pfil, CTLFLAG_RW,
518 &bdg_pfil, 0,"Pass bridged pkts through PFIL hooks");
521 * The follow macro declares a variable, and maps it to
522 * a SYSCTL_INT entry with the same name.
524 #define SY(parent, var, comment) \
526 SYSCTL_INT(parent, OID_AUTO, var, CTLFLAG_RW, &(var), 0, comment);
529 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_ipfw_drop,
530 CTLFLAG_RW, &bdg_ipfw_drops,0,"");
533 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_ipfw_collisions,
534 CTLFLAG_RW, &bdg_ipfw_colls,0,"");
536 SYSCTL_PROC(_net_link_ether, OID_AUTO, bridge_refresh, CTLTYPE_INT|CTLFLAG_WR,
537 NULL, 0, &sysctl_refresh, "I", "iface refresh");
539 #if 1 /* diagnostic vars */
541 SY(_net_link_ether, verbose, "Be verbose");
542 SY(_net_link_ether, bdg_split_pkts, "Packets split in bdg_forward");
544 SY(_net_link_ether, bdg_thru, "Packets through bridge");
546 SY(_net_link_ether, bdg_copied, "Packets copied in bdg_forward");
548 SY(_net_link_ether, bdg_copy, "Force copy in bdg_forward");
549 SY(_net_link_ether, bdg_predict, "Correctly predicted header location");
551 SY(_net_link_ether, bdg_fw_avg, "Cycle counter avg");
552 SY(_net_link_ether, bdg_fw_ticks, "Cycle counter item");
553 SY(_net_link_ether, bdg_fw_count, "Cycle counter count");
556 SYSCTL_STRUCT(_net_link_ether, PF_BDG, bdgstats,
557 CTLFLAG_RD, &bdg_stats , bdg_stats, "bridge statistics");
559 static int bdg_loops ;
562 * called periodically to flush entries etc.
565 bdg_timeout(void *dummy)
567 static int slowtimer; /* in BSS so initialized to 0 */
570 static int age_index = 0 ; /* index of table position to age */
571 int l = age_index + HASH_SIZE/4 ;
574 * age entries in the forwarding table.
579 for (i=0; i<n_clusters; i++) {
580 bdg_hash_table *bdg_table = clusters[i].ht;
581 for (; age_index < l ; age_index++)
582 if (bdg_table[age_index].used)
583 bdg_table[age_index].used = 0 ;
584 else if (bdg_table[age_index].name) {
585 /* printf("xx flushing stale entry %d\n", age_index); */
586 bdg_table[age_index].name = NULL ;
589 if (age_index >= HASH_SIZE)
592 if (--slowtimer <= 0 ) {
595 bridge_on() ; /* we just need unmute, really */
599 callout_reset(&bdg_timeout_h, 2*hz, bdg_timeout, NULL);
603 * Find the right pkt destination:
604 * BDG_BCAST is a broadcast
605 * BDG_MCAST is a multicast
606 * BDG_LOCAL is for a local address
607 * BDG_DROP must be dropped
608 * other ifp of the dest. interface (incl.self)
610 * We assume this is only called for interfaces for which bridging
611 * is enabled, i.e. BDG_USED(ifp) is true.
615 bridge_dst_lookup(struct ether_header *eh, struct cluster_softc *c)
620 bdg_hash_table *bt; /* pointer to entry in hash table */
622 if (IS_ETHER_BROADCAST(eh->ether_dhost))
624 if (eh->ether_dhost[0] & 1)
627 * Lookup local addresses in case one matches.
629 for (index = c->ports, p = c->my_macs; index ; index--, p++ )
630 if (BDG_MATCH(p->etheraddr, eh->ether_dhost) )
633 * Look for a possible destination in table
635 index= HASH_FN( eh->ether_dhost );
636 bt = &(c->ht[index]);
638 if ( dst && BDG_MATCH( bt->etheraddr, eh->ether_dhost) )
645 * bridge_in() is invoked to perform bridging decision on input packets.
648 * eh Ethernet header of the incoming packet.
649 * ifp interface the packet is coming from.
651 * On Return: destination of packet, one of
652 * BDG_BCAST broadcast
653 * BDG_MCAST multicast
654 * BDG_LOCAL is only for a local address (do not forward)
655 * BDG_DROP drop the packet
656 * ifp ifp of the destination interface.
658 * Forwarding is not done directly to give a chance to some drivers
659 * to fetch more of the packet, or simply drop it completely.
662 static struct ifnet *
663 bridge_in(struct ifnet *ifp, struct ether_header *eh)
666 struct ifnet *dst , *old ;
667 bdg_hash_table *bt; /* location in hash table */
668 int dropit = BDG_MUTED(ifp) ;
671 * hash the source address
673 index= HASH_FN(eh->ether_shost);
674 bt = &(ifp2sc[ifp->if_index].cluster->ht[index]);
677 if ( old ) { /* the entry is valid. */
678 if (!BDG_MATCH( eh->ether_shost, bt->etheraddr) ) {
681 } else if (old != ifp) {
683 * Found a loop. Either a machine has moved, or there
684 * is a misconfiguration/reconfiguration of the network.
685 * First, do not forward this packet!
686 * Record the relocation anyways; then, if loops persist,
687 * suspect a reconfiguration and disable forwarding
688 * from the old interface.
690 bt->name = ifp ; /* relocate address */
691 printf("-- loop (%d) %6D to %s from %s (%s)\n",
692 bdg_loops, eh->ether_shost, ".",
693 ifp->if_xname, old->if_xname,
694 BDG_MUTED(old) ? "muted":"active");
696 if ( !BDG_MUTED(old) ) {
697 if (++bdg_loops > 10)
704 * now write the source address into the table
706 if (bt->name == NULL) {
707 DEB(printf("new addr %6D at %d for %s\n",
708 eh->ether_shost, ".", index, ifp->if_xname);)
709 bcopy(eh->ether_shost, bt->etheraddr, 6);
712 dst = bridge_dst_lookup(eh, ifp2sc[ifp->if_index].cluster);
714 * bridge_dst_lookup can return the following values:
715 * BDG_BCAST, BDG_MCAST, BDG_LOCAL, BDG_UNKNOWN, BDG_DROP, ifp.
716 * For muted interfaces, or when we detect a loop, the first 3 are
717 * changed in BDG_LOCAL (we still listen to incoming traffic),
718 * and others to BDG_DROP (no use for the local host).
719 * Also, for incoming packets, ifp is changed to BDG_DROP if ifp == src.
720 * These changes are not necessary for outgoing packets from ether_output().
722 BDG_STAT(ifp, BDG_IN);
723 switch ((uintptr_t)dst) {
724 case (uintptr_t)BDG_BCAST:
725 case (uintptr_t)BDG_MCAST:
726 case (uintptr_t)BDG_LOCAL:
727 case (uintptr_t)BDG_UNKNOWN:
728 case (uintptr_t)BDG_DROP:
732 if (dst == ifp || dropit)
733 BDG_STAT(ifp, BDG_DROP);
735 BDG_STAT(ifp, BDG_FORWARD);
740 if (dst == BDG_BCAST || dst == BDG_MCAST || dst == BDG_LOCAL)
748 DEB(printf("bridge_in %6D ->%6D ty 0x%04x dst %s\n",
749 eh->ether_shost, ".",
750 eh->ether_dhost, ".",
751 ntohs(eh->ether_type),
752 (dst <= BDG_FORWARD) ? bdg_dst_names[(int)dst]"0" :
759 * Forward a packet to dst -- which can be a single interface or
760 * an entire cluster. The src port and muted interfaces are excluded.
762 * If src == NULL, the pkt comes from ether_output, and dst is the real
763 * interface the packet is originally sent to. In this case, we must forward
764 * it to the whole cluster.
765 * We never call bdg_forward from ether_output on interfaces which are
766 * not part of a cluster.
768 * If possible (i.e. we can determine that the caller does not need
769 * a copy), the packet is consumed here, and bdg_forward returns NULL.
770 * Otherwise, a pointer to a copy of the packet is returned.
772 * XXX be careful with eh, it can be a pointer into *m
775 bdg_forward(struct mbuf *m0, struct ether_header *const eh, struct ifnet *dst)
778 struct ifnet *ifp, *last;
779 int shared = bdg_copy ; /* someone else is using the mbuf */
780 int once = 0; /* loop only once */
781 struct ifnet *real_dst = dst ; /* real dst from ether_output */
782 struct ip_fw_args args;
785 * XXX eh is usually a pointer within the mbuf (some ethernet drivers
786 * do that), so we better copy it before doing anything with the mbuf,
787 * or we might corrupt the header.
789 struct ether_header save_eh = *eh ;
791 DEB(quad_t ticks; ticks = rdtsc();)
793 args.rule = NULL; /* did we match a firewall rule ? */
794 /* Fetch state from dummynet tag, ignore others */
795 for (;m0->m_type == MT_TAG; m0 = m0->m_next)
796 if (m0->_m_tag_id == PACKET_TAG_DUMMYNET) {
797 args.rule = ((struct dn_pkt *)m0)->rule;
798 shared = 0; /* For sure this is our own mbuf. */
800 if (args.rule == NULL)
801 bdg_thru++; /* first time through bdg_forward, count packet */
803 src = m0->m_pkthdr.rcvif;
804 if (src == NULL) /* packet from ether_output */
805 dst = bridge_dst_lookup(eh, ifp2sc[real_dst->if_index].cluster);
807 if (dst == BDG_DROP) { /* this should not happen */
808 printf("xx bdg_forward for BDG_DROP\n");
812 if (dst == BDG_LOCAL) { /* this should not happen as well */
813 printf("xx ouch, bdg_forward for local pkt\n");
816 if (dst == BDG_BCAST || dst == BDG_MCAST || dst == BDG_UNKNOWN) {
817 ifp = TAILQ_FIRST(&ifnet) ; /* scan all ports */
819 if (dst != BDG_UNKNOWN) /* need a copy for the local stack */
825 if ( (uintptr_t)(ifp) <= (u_int)BDG_FORWARD )
826 panic("bdg_forward: bad dst");
829 * Do filtering in a very similar way to what is done in ip_output.
830 * Only if firewall is loaded, enabled, and the packet is not
831 * from ether_output() (src==NULL, or we would filter it twice).
832 * Additional restrictions may apply e.g. non-IP, short packets,
833 * and pkts already gone through a pipe.
836 (bdg_pfil && pfil_has_hooks(&inet_pfil_hook)) ||
837 (IPFW_LOADED && bdg_ipfw != 0))) {
841 if (args.rule != NULL && fw_one_pass)
842 goto forward; /* packet already partially processed */
844 * i need some amt of data to be contiguous, and in case others need
845 * the packet (shared==1) also better be in the first mbuf.
847 i = min(m0->m_pkthdr.len, max_protohdr) ;
848 if ( shared || m0->m_len < i) {
849 m0 = m_pullup(m0, i) ;
851 printf("-- bdg: pullup failed.\n") ;
857 * PFIL hooks processing
859 if (bdg_pfil && pfil_has_hooks(&inet_pfil_hook) &&
860 m0->m_pkthdr.len >= sizeof(struct ip) &&
861 ntohs(save_eh.ether_type) == ETHERTYPE_IP) {
863 * before calling the firewall, swap fields the same as IP does.
864 * here we assume the pkt is an IP one and the header is contiguous
866 struct ip *ip = mtod(m0, struct ip *);
869 ip->ip_len = ntohs(ip->ip_len);
870 ip->ip_off = ntohs(ip->ip_off);
873 * XXX I think this needs to be reworked to enable stateful
874 * filtering. Think about this. For now, declare the packet as incoming.
876 error = pfil_run_hooks(&inet_pfil_hook, &m0, ifp, PFIL_IN);
877 if (error != 0 || m0 == NULL)
881 * If we get here, the firewall has passed the pkt, but the mbuf
882 * pointer might have changed. Restore ip and the fields ntohs()'d.
884 ip = mtod(m0, struct ip *);
885 ip->ip_len = htons(ip->ip_len);
886 ip->ip_off = htons(ip->ip_off);
890 * Prepare arguments and call the firewall.
892 if (!IPFW_LOADED || bdg_ipfw == 0)
893 goto forward; /* not using ipfw, accept the packet */
896 * XXX The following code is very similar to the one in
897 * if_ethersubr.c:ether_ipfw_chk()
900 args.m = m0; /* the packet we are looking at */
901 args.oif = NULL; /* this is an input packet */
902 args.divert_rule = 0; /* we do not support divert yet */
903 args.next_hop = NULL; /* we do not support forward yet */
904 args.eh = &save_eh; /* MAC header for bridged/MAC packets */
905 i = ip_fw_chk_ptr(&args);
906 m0 = args.m; /* in case the firewall used the mbuf */
908 if ( (i & IP_FW_PORT_DENY_FLAG) || m0 == NULL) /* drop */
911 if (i == 0) /* a PASS rule. */
913 if (DUMMYNET_LOADED && (i & IP_FW_PORT_DYNT_FLAG)) {
915 * Pass the pkt to dummynet, which consumes it.
916 * If shared, make a copy and keep the original.
921 m = m_copypacket(m0, MB_DONTWAIT);
922 if (m == NULL) /* copy failed, give up */
925 m = m0 ; /* pass the original to dummynet */
926 m0 = NULL ; /* and nothing back to the caller */
929 * Prepend the header, optimize for the common case of
930 * eh pointing into the mbuf.
932 if ( (void *)(eh + 1) == (void *)m->m_data) {
933 m->m_data -= ETHER_HDR_LEN ;
934 m->m_len += ETHER_HDR_LEN ;
935 m->m_pkthdr.len += ETHER_HDR_LEN ;
938 M_PREPEND(m, ETHER_HDR_LEN, MB_DONTWAIT);
939 if (m == NULL) /* nope... */
941 bcopy(&save_eh, mtod(m, struct ether_header *), ETHER_HDR_LEN);
945 ip_dn_io_ptr(m, (i & 0xffff),DN_TO_BDG_FWD, &args);
949 * XXX at some point, add support for divert/forward actions.
950 * If none of the above matches, we have to drop the packet.
957 * Again, bring up the headers in case of shared bufs to avoid
958 * corruptions in the future.
961 int i = min(m0->m_pkthdr.len, max_protohdr) ;
963 m0 = m_pullup(m0, i) ;
968 * now real_dst is used to determine the cluster where to forward.
969 * For packets coming from ether_input, this is the one of the 'src'
970 * interface, whereas for locally generated packets (src==NULL) it
971 * is the cluster of the original destination interface, which
972 * was already saved into real_dst.
979 if (last) { /* need to forward packet leftover from previous loop */
981 if (shared == 0 && once ) { /* no need to copy */
983 m0 = NULL ; /* original is gone */
985 m = m_copypacket(m0, MB_DONTWAIT);
987 printf("bdg_forward: sorry, m_copypacket failed!\n");
988 return m0 ; /* the original is still there... */
992 * Add header (optimized for the common case of eh pointing
993 * already into the mbuf) and execute last part of ether_output:
994 * queue pkt and start output if interface not yet active.
996 if ( (void *)(eh + 1) == (void *)m->m_data) {
997 m->m_data -= ETHER_HDR_LEN ;
998 m->m_len += ETHER_HDR_LEN ;
999 m->m_pkthdr.len += ETHER_HDR_LEN ;
1002 M_PREPEND(m, ETHER_HDR_LEN, MB_DONTWAIT);
1004 printf("M_PREPEND failed\n");
1007 bcopy(&save_eh, mtod(m, struct ether_header *), ETHER_HDR_LEN);
1009 if (!IF_HANDOFF(&last->if_snd, m, last)) {
1011 BDG_MUTE(last); /* should I also mute ? */
1014 BDG_STAT(last, BDG_OUT);
1022 * If the interface is used for bridging, not muted, not full,
1023 * up and running, is not the source interface, and belongs to
1024 * the same cluster as the 'real_dst', then send here.
1026 if ( BDG_USED(ifp) && !BDG_MUTED(ifp) && !_IF_QFULL(&ifp->if_snd) &&
1027 (ifp->if_flags & (IFF_UP|IFF_RUNNING)) == (IFF_UP|IFF_RUNNING) &&
1028 ifp != src && BDG_SAMECLUSTER(ifp, real_dst) )
1030 ifp = TAILQ_NEXT(ifp, if_link) ;
1034 DEB(bdg_fw_ticks += (u_long)(rdtsc() - ticks) ; bdg_fw_count++ ;
1035 if (bdg_fw_count != 0) bdg_fw_avg = bdg_fw_ticks/bdg_fw_count; )
1040 * initialization of bridge code.
1045 printf("BRIDGE 020214 loaded\n");
1047 ifp2sc = malloc(BDG_MAX_PORTS * sizeof(struct bdg_softc),
1048 M_IFADDR, M_WAITOK | M_ZERO );
1052 callout_init(&bdg_timeout_h);
1054 bridge_in_ptr = bridge_in;
1055 bdg_forward_ptr = bdg_forward;
1056 bdgtakeifaces_ptr = reconfigure_bridge;
1062 bzero(&bdg_stats, sizeof(bdg_stats) );
1063 bdgtakeifaces_ptr();
1069 * initialization code, both for static and dynamic loading.
1072 bridge_modevent(module_t mod, int type, void *unused)
1088 #if !defined(KLD_MODULE)
1089 printf("bridge statically compiled, cannot unload\n");
1094 bridge_in_ptr = NULL;
1095 bdg_forward_ptr = NULL;
1096 bdgtakeifaces_ptr = NULL;
1097 callout_stop(&bdg_timeout_h);
1100 free(clusters, M_IFADDR);
1101 free(ifp2sc, M_IFADDR);
1113 static moduledata_t bridge_mod = {
1119 DECLARE_MODULE(bridge, bridge_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
1120 MODULE_VERSION(bridge, 1);