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
31 * This code implements bridging in FreeBSD. It only acts on ethernet
32 * interfaces, including VLANs (others are still usable for routing).
33 * A FreeBSD host can implement multiple logical bridges, called
34 * "clusters". Each cluster is made of a set of interfaces, and
35 * identified by a "cluster-id" which is a number in the range 1..2^16-1.
37 * Bridging is enabled by the sysctl variable
38 * net.link.ether.bridge.enable
39 * the grouping of interfaces into clusters is done with
40 * net.link.ether.bridge.config
41 * containing a list of interfaces each optionally followed by
42 * a colon and the cluster it belongs to (1 is the default).
43 * Separators can be * spaces, commas or tabs, e.g.
44 * net.link.ether.bridge.config="fxp0:2 fxp1:2 dc0 dc1:1"
45 * Optionally bridged packets can be passed through the firewall,
46 * this is controlled by the variable
47 * net.link.ether.bridge.ipfw
49 * For each cluster there is a descriptor (cluster_softc) storing
50 * the following data structures:
51 * - a hash table with the MAC address and destination interface for each
52 * known node. The table is indexed using a hash of the source address.
53 * - an array with the MAC addresses of the interfaces used in the cluster.
55 * Input packets are tapped near the beginning of ether_input(), and
56 * analysed by bridge_in(). Depending on the result, the packet
57 * can be forwarded to one or more output interfaces using bdg_forward(),
58 * and/or sent to the upper layer (e.g. in case of multicast).
60 * Output packets are intercepted near the end of ether_output().
61 * The correct destination is selected by bridge_dst_lookup(),
62 * and then forwarding is done by bdg_forward().
64 * The arp code is also modified to let a machine answer to requests
65 * irrespective of the port the request came from.
67 * In case of loops in the bridging topology, the bridge detects this
68 * event and temporarily mutes output bridging on one of the ports.
69 * Periodically, interfaces are unmuted by bdg_timeout().
70 * Muting is only implemented as a safety measure, and also as
71 * a mechanism to support a user-space implementation of the spanning
74 * To build a bridging kernel, use the following option
76 * and then at runtime set the sysctl variable to enable bridging.
78 * Only one interface per cluster is supposed to have addresses set (but
79 * there are no substantial problems if you set addresses for none or
80 * for more than one interface).
81 * Bridging will act before routing, but nothing prevents a machine
82 * from doing both (modulo bugs in the implementation...).
85 * - bridging is incompatible with multicast routing on the same
86 * machine. There is not an easy fix to this.
87 * - be very careful when bridging VLANs
88 * - loop detection is still not very robust.
90 #include "opt_pfil_hooks.h"
92 #include <sys/param.h>
94 #include <sys/malloc.h>
95 #include <sys/protosw.h>
96 #include <sys/systm.h>
97 #include <sys/socket.h> /* for net/if.h */
98 #include <sys/ctype.h> /* string functions */
99 #include <sys/kernel.h>
100 #include <sys/sysctl.h>
103 #include <net/if_types.h>
104 #include <net/if_var.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/if_ether.h> /* for struct arpcom */
113 #include <net/pfil.h>
114 #include <netinet/ip_var.h>
117 #include <net/route.h>
118 #include <netinet/ip_fw.h>
119 #include <netinet/ip_dummynet.h>
120 #include <net/bridge.h>
122 /*--------------------*/
124 #define ETHER_ADDR_COPY(_dst,_src) bcopy(_src, _dst, ETHER_ADDR_LEN)
125 #define ETHER_ADDR_EQ(_a1,_a2) (bcmp(_a1, _a2, ETHER_ADDR_LEN) == 0)
128 * For each cluster, source MAC addresses are stored into a hash
129 * table which locates the port they reside on.
131 #define HASH_SIZE 8192 /* Table size, must be a power of 2 */
133 typedef struct hash_table { /* each entry. */
135 u_char etheraddr[ETHER_ADDR_LEN];
136 u_int16_t used; /* also, padding */
140 * The hash function applied to MAC addresses. Out of the 6 bytes,
141 * the last ones tend to vary more. Since we are on a little endian machine,
142 * we have to do some gimmick...
144 #define HASH_FN(addr) ( \
145 ntohs( ((u_int16_t *)addr)[1] ^ ((u_int16_t *)addr)[2] ) & (HASH_SIZE -1))
148 * This is the data structure where local addresses are stored.
151 u_char etheraddr[ETHER_ADDR_LEN];
156 * The configuration of each cluster includes the cluster id, a pointer to
157 * the hash table, and an array of local MAC addresses (of size "ports").
159 struct cluster_softc {
160 u_int16_t cluster_id;
163 struct bdg_addr *my_macs; /* local MAC addresses */
167 extern struct protosw inetsw[]; /* from netinet/ip_input.c */
169 static int n_clusters; /* number of clusters */
170 static struct cluster_softc *clusters;
172 static struct mtx bdg_mtx;
173 #define BDG_LOCK_INIT() mtx_init(&bdg_mtx, "bridge", NULL, MTX_DEF)
174 #define BDG_LOCK_DESTROY() mtx_destroy(&bdg_mtx)
175 #define BDG_LOCK() mtx_lock(&bdg_mtx)
176 #define BDG_UNLOCK() mtx_unlock(&bdg_mtx)
177 #define BDG_LOCK_ASSERT() mtx_assert(&bdg_mtx, MA_OWNED)
179 #define BDG_MUTED(ifp) (ifp2sc[ifp->if_index].flags & IFF_MUTE)
180 #define BDG_MUTE(ifp) ifp2sc[ifp->if_index].flags |= IFF_MUTE
181 #define BDG_CLUSTER(ifp) (ifp2sc[ifp->if_index].cluster)
183 #define BDG_SAMECLUSTER(ifp,src) \
184 (src == NULL || BDG_CLUSTER(ifp) == BDG_CLUSTER(src) )
187 #define BDG_MATCH(a,b) ( \
188 ((u_int16_t *)(a))[2] == ((u_int16_t *)(b))[2] && \
189 *((u_int32_t *)(a)) == *((u_int32_t *)(b)) )
190 #define IS_ETHER_BROADCAST(a) ( \
191 *((u_int32_t *)(a)) == 0xffffffff && \
192 ((u_int16_t *)(a))[2] == 0xffff )
194 /* for machines that do not support unaligned access */
195 #define BDG_MATCH(a,b) ETHER_ADDR_EQ(a,b)
196 #define IS_ETHER_BROADCAST(a) ETHER_ADDR_EQ(a,"\377\377\377\377\377\377")
199 SYSCTL_DECL(_net_link_ether);
200 SYSCTL_NODE(_net_link_ether, OID_AUTO, bridge, CTLFLAG_RD, 0,
201 "Bridge parameters");
202 static char bridge_version[] = "$Revision$ $Date$";
203 SYSCTL_STRING(_net_link_ether_bridge, OID_AUTO, version, CTLFLAG_RD,
204 bridge_version, 0, "software version");
208 int bridge_debug = 0;
209 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, debug, CTLFLAG_RW, &bridge_debug,
210 0, "control debugging printfs");
211 #define DPRINTF(X) if (bridge_debug) printf X
218 * For timing-related debugging, you can use the following macros.
219 * remember, rdtsc() only works on Pentium-class machines
222 DDB(ticks = rdtsc();)
223 ... interesting code ...
224 DDB(bdg_fw_ticks += (u_long)(rdtsc() - ticks) ; bdg_fw_count++ ;)
230 static int bdg_fw_avg;
231 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, fw_avg, CTLFLAG_RW,
232 &bdg_fw_avg, 0,"Cycle counter avg");
233 static int bdg_fw_ticks;
234 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, fw_ticks, CTLFLAG_RW,
235 &bdg_fw_ticks, 0,"Cycle counter item");
236 static int bdg_fw_count;
237 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, fw_count, CTLFLAG_RW,
238 &bdg_fw_count, 0,"Cycle counter count");
243 static int bdginit(void);
244 static void parse_bdg_cfg(void);
246 static int bdg_ipf; /* IPFilter enabled in bridge */
247 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, ipf, CTLFLAG_RW,
248 &bdg_ipf, 0,"Pass bridged pkts through IPFilter");
250 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, ipfw, CTLFLAG_RW,
251 &bdg_ipfw,0,"Pass bridged pkts through firewall");
254 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, copy, CTLFLAG_RW,
255 &bdg_copy, 0, "Force packet copy in bdg_forward");
258 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, ipfw_drop,
259 CTLFLAG_RW, &bdg_ipfw_drops,0,"");
261 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, ipfw_collisions,
262 CTLFLAG_RW, &bdg_ipfw_colls,0,"");
265 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, packets, CTLFLAG_RW,
266 &bdg_thru, 0, "Packets through bridge");
267 static int bdg_dropped;
268 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, dropped, CTLFLAG_RW,
269 &bdg_dropped, 0, "Packets dropped in bdg_forward");
270 static int bdg_predict;
271 SYSCTL_INT(_net_link_ether_bridge, OID_AUTO, predict, CTLFLAG_RW,
272 &bdg_predict, 0, "Correctly predicted header location");
275 static char *bdg_dst_names[] = {
285 #endif /* BRIDGE_DEBUG */
288 * System initialization
290 static struct bdg_stats bdg_stats ;
291 SYSCTL_STRUCT(_net_link_ether_bridge, OID_AUTO, stats, CTLFLAG_RD,
292 &bdg_stats, bdg_stats, "bridge statistics");
294 static struct callout bdg_callout;
297 * Add an interface to a cluster, possibly creating a new entry in
298 * the cluster table. This requires reallocation of the table and
299 * updating pointers in ifp2sc.
301 static struct cluster_softc *
302 add_cluster(u_int16_t cluster_id, struct arpcom *ac)
304 struct cluster_softc *c = NULL;
309 for (i = 0; i < n_clusters ; i++)
310 if (clusters[i].cluster_id == cluster_id)
313 /* Not found, need to reallocate */
314 c = malloc((1+n_clusters) * sizeof (*c), M_IFADDR, M_NOWAIT | M_ZERO);
315 if (c == NULL) {/* malloc failure */
316 printf("-- bridge: cannot add new cluster\n");
319 c[n_clusters].ht = (struct hash_table *)
320 malloc(HASH_SIZE * sizeof(struct hash_table),
321 M_IFADDR, M_NOWAIT | M_ZERO);
322 if (c[n_clusters].ht == NULL) {
323 printf("-- bridge: cannot allocate hash table for new cluster\n");
326 c[n_clusters].my_macs = (struct bdg_addr *)
327 malloc(BDG_MAX_PORTS * sizeof(struct bdg_addr),
328 M_IFADDR, M_NOWAIT | M_ZERO);
329 if (c[n_clusters].my_macs == NULL) {
330 printf("-- bridge: cannot allocate mac addr table for new cluster\n");
331 free(c[n_clusters].ht, M_IFADDR);
335 c[n_clusters].cluster_id = cluster_id;
336 c[n_clusters].ports = 0;
338 * now copy old descriptors here
340 if (n_clusters > 0) {
341 for (i=0; i < n_clusters; i++)
344 * and finally update pointers in ifp2sc
346 for (i = 0 ; i < if_index && i < BDG_MAX_PORTS; i++)
347 if (ifp2sc[i].cluster != NULL)
348 ifp2sc[i].cluster = c + (ifp2sc[i].cluster - clusters);
349 free(clusters, M_IFADDR);
352 i = n_clusters; /* index of cluster entry */
355 c = clusters + i; /* the right cluster ... */
356 ETHER_ADDR_COPY(c->my_macs[c->ports].etheraddr, ac->ac_enaddr);
367 * Turn off bridging, by clearing promisc mode on the interface,
368 * marking the interface as unused, and clearing the name in the
370 * Also dispose the hash tables associated with the clusters.
380 DPRINTF(("%s: n_clusters %d\n", __func__, n_clusters));
383 TAILQ_FOREACH(ifp, &ifnet, if_link) {
386 if (ifp->if_index >= BDG_MAX_PORTS)
387 continue; /* make sure we do not go beyond the end */
388 b = &ifp2sc[ifp->if_index];
390 if ( b->flags & IFF_BDG_PROMISC ) {
392 b->flags &= ~(IFF_BDG_PROMISC|IFF_MUTE) ;
393 DPRINTF(("%s: %s%d promisc OFF if_flags 0x%x "
394 "bdg_flags 0x%x\n", __func__, ifp->if_name, ifp->if_unit,
395 ifp->if_flags, b->flags));
397 b->flags &= ~(IFF_USED) ;
399 bdg_stats.s[ifp->if_index].name[0] = '\0';
404 for (i=0; i < n_clusters; i++) {
405 free(clusters[i].ht, M_IFADDR);
406 free(clusters[i].my_macs, M_IFADDR);
408 if (clusters != NULL)
409 free(clusters, M_IFADDR);
415 * set promisc mode on the interfaces we use.
425 TAILQ_FOREACH(ifp, &ifnet, if_link) {
426 struct bdg_softc *b = &ifp2sc[ifp->if_index];
428 if ( !(b->flags & IFF_USED) )
430 if ( !( ifp->if_flags & IFF_UP) ) {
433 if ( !(b->flags & IFF_BDG_PROMISC) ) {
434 (void) ifpromisc(ifp, 1);
435 b->flags |= IFF_BDG_PROMISC ;
436 DPRINTF(("%s: %s%d promisc ON if_flags 0x%x bdg_flags 0x%x\n",
437 __func__, ifp->if_name, ifp->if_unit, ifp->if_flags, b->flags));
439 if (b->flags & IFF_MUTE) {
440 DPRINTF(("%s: unmuting %s%d\n", __func__,
441 ifp->if_name, ifp->if_unit));
442 b->flags &= ~IFF_MUTE;
448 static char bridge_cfg[1024]; /* NB: in BSS so initialized to zero */
451 * reconfigure bridge.
452 * This is also done every time we attach or detach an interface.
453 * Main use is to make sure that we do not bridge on some old
454 * (ejected) device. So, it would be really useful to have a
455 * pointer to the modified device as an argument. Without it, we
456 * have to scan all interfaces.
459 reconfigure_bridge_locked(void)
465 if (if_index >= BDG_MAX_PORTS) {
466 printf("-- sorry too many interfaces (%d, max is %d),"
467 " disabling bridging\n", if_index, BDG_MAX_PORTS);
477 reconfigure_bridge(void)
480 reconfigure_bridge_locked();
485 * parse the config string, set IFF_USED, name and cluster_id
486 * for all interfaces found.
487 * The config string is a list of "if[:cluster]" with
488 * a number of possible separators (see "sep"). In particular the
489 * use of the space lets you set bridge_cfg with the output from
497 static const char *sep = ", \t";
501 for (p = bridge_cfg; *p ; p++) {
506 if (index(sep, *p)) /* skip separators */
508 /* names are lowercase and digits */
509 for ( beg = p ; islower(*p) || isdigit(*p) ; p++ )
511 l = p - beg ; /* length of name string */
512 if (l == 0) /* invalid name */
514 if ( *p != ':' ) /* no ':', assume default cluster 1 */
516 else /* fetch cluster */
517 cluster = strtoul( p+1, &p, 10);
521 * now search in interface list for a matching name
523 IFNET_RLOCK(); /* could sleep XXX */
524 TAILQ_FOREACH(ifp, &ifnet, if_link) {
527 snprintf(buf, sizeof(buf), "%s%d", ifp->if_name, ifp->if_unit);
528 if (!strncmp(beg, buf, max(l, strlen(buf)))) {
529 struct bdg_softc *b = &ifp2sc[ifp->if_index];
530 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) {
531 printf("%s is not an ethernet, continue\n", buf);
534 if (b->flags & IFF_USED) {
535 printf("%s already used, skipping\n", buf);
538 b->cluster = add_cluster(htons(cluster), (struct arpcom *)ifp);
539 b->flags |= IFF_USED ;
540 sprintf(bdg_stats.s[ifp->if_index].name,
541 "%s%d:%d", ifp->if_name, ifp->if_unit, cluster);
543 DPRINTF(("%s: found %s next c %d\n", __func__,
544 bdg_stats.s[ifp->if_index].name, c));
551 printf("interface %s Not found in bridge\n", beg);
559 * handler for net.link.ether.bridge
562 sysctl_bdg(SYSCTL_HANDLER_ARGS)
564 int enable = do_bridge;
567 error = sysctl_handle_int(oidp, &enable, 0, req);
569 if (enable != do_bridge) {
571 reconfigure_bridge_locked();
576 SYSCTL_PROC(_net_link_ether_bridge, OID_AUTO, enable, CTLTYPE_INT|CTLFLAG_RW,
577 &do_bridge, 0, &sysctl_bdg, "I", "Bridging");
580 * handler for net.link.ether.bridge_cfg
583 sysctl_bdg_cfg(SYSCTL_HANDLER_ARGS)
588 new_cfg = malloc(sizeof(bridge_cfg), M_TEMP, M_WAITOK);
589 bcopy(bridge_cfg, new_cfg, sizeof(bridge_cfg));
591 error = sysctl_handle_string(oidp, new_cfg, oidp->oid_arg2, req);
594 if (strcmp(new_cfg, bridge_cfg)) {
595 bcopy(new_cfg, bridge_cfg, sizeof(bridge_cfg));
596 reconfigure_bridge_locked();
601 free(new_cfg, M_TEMP);
605 SYSCTL_PROC(_net_link_ether_bridge, OID_AUTO, config, CTLTYPE_STRING|CTLFLAG_RW,
606 &bridge_cfg, sizeof(bridge_cfg), &sysctl_bdg_cfg, "A",
607 "Bridge configuration");
610 sysctl_refresh(SYSCTL_HANDLER_ARGS)
613 reconfigure_bridge();
617 SYSCTL_PROC(_net_link_ether_bridge, OID_AUTO, refresh, CTLTYPE_INT|CTLFLAG_WR,
618 NULL, 0, &sysctl_refresh, "I", "iface refresh");
621 #define SYSCTL_OID_COMPAT(parent, nbr, name, kind, a1, a2, handler, fmt, descr)\
622 static struct sysctl_oid sysctl__##parent##_##name##_compat = { \
623 &sysctl_##parent##_children, { 0 }, \
624 nbr, kind, a1, a2, #name, handler, fmt, 0, descr }; \
625 DATA_SET(sysctl_set, sysctl__##parent##_##name##_compat)
626 #define SYSCTL_INT_COMPAT(parent, nbr, name, access, ptr, val, descr) \
627 SYSCTL_OID_COMPAT(parent, nbr, name, CTLTYPE_INT|(access), \
628 ptr, val, sysctl_handle_int, "I", descr)
629 #define SYSCTL_STRUCT_COMPAT(parent, nbr, name, access, ptr, type, descr)\
630 SYSCTL_OID_COMPAT(parent, nbr, name, CTLTYPE_OPAQUE|(access), \
631 ptr, sizeof(struct type), sysctl_handle_opaque, \
633 #define SYSCTL_PROC_COMPAT(parent, nbr, name, access, ptr, arg, handler, fmt, descr) \
634 SYSCTL_OID_COMPAT(parent, nbr, name, (access), \
635 ptr, arg, handler, fmt, descr)
637 SYSCTL_INT_COMPAT(_net_link_ether, OID_AUTO, bridge_ipf, CTLFLAG_RW,
638 &bdg_ipf, 0,"Pass bridged pkts through IPFilter");
639 SYSCTL_INT_COMPAT(_net_link_ether, OID_AUTO, bridge_ipfw, CTLFLAG_RW,
640 &bdg_ipfw,0,"Pass bridged pkts through firewall");
641 SYSCTL_STRUCT_COMPAT(_net_link_ether, PF_BDG, bdgstats, CTLFLAG_RD,
642 &bdg_stats, bdg_stats, "bridge statistics");
643 SYSCTL_PROC_COMPAT(_net_link_ether, OID_AUTO, bridge_cfg,
644 CTLTYPE_STRING|CTLFLAG_RW,
645 &bridge_cfg, sizeof(bridge_cfg), &sysctl_bdg_cfg, "A",
646 "Bridge configuration");
647 SYSCTL_PROC_COMPAT(_net_link_ether, OID_AUTO, bridge_refresh,
648 CTLTYPE_INT|CTLFLAG_WR,
649 NULL, 0, &sysctl_refresh, "I", "iface refresh");
652 static int bdg_loops;
653 static int bdg_slowtimer = 0;
654 static int bdg_age_index = 0; /* index of table position to age */
657 * called periodically to flush entries etc.
660 bdg_timeout(void *dummy)
667 * age entries in the forwarding table.
669 l = bdg_age_index + HASH_SIZE/4 ;
673 for (i = 0; i < n_clusters; i++) {
674 bdg_hash_table *bdg_table = clusters[i].ht;
675 for (; bdg_age_index < l; bdg_age_index++)
676 if (bdg_table[bdg_age_index].used)
677 bdg_table[bdg_age_index].used = 0;
678 else if (bdg_table[bdg_age_index].name) {
679 DPRINTF(("%s: flushing stale entry %d\n",
680 __func__, bdg_age_index));
681 bdg_table[bdg_age_index].name = NULL;
684 if (bdg_age_index >= HASH_SIZE)
687 if (--bdg_slowtimer <= 0 ) {
690 bridge_on(); /* we just need unmute, really */
695 callout_reset(&bdg_callout, 2*hz, bdg_timeout, NULL);
699 * Find the right pkt destination:
700 * BDG_BCAST is a broadcast
701 * BDG_MCAST is a multicast
702 * BDG_LOCAL is for a local address
703 * BDG_DROP must be dropped
704 * other ifp of the dest. interface (incl.self)
706 * We assume this is only called for interfaces for which bridging
707 * is enabled, i.e. BDG_USED(ifp) is true.
709 static __inline struct ifnet *
710 bridge_dst_lookup(struct ether_header *eh, struct cluster_softc *c)
712 bdg_hash_table *bt; /* pointer to entry in hash table */
716 if (ETHER_IS_MULTICAST(eh->ether_dhost))
717 return IS_ETHER_BROADCAST(eh->ether_dhost) ? BDG_BCAST : BDG_MCAST;
719 * Lookup local addresses in case one matches. We optimize
720 * for the common case of two interfaces.
722 KASSERT(c->ports != 0, ("lookup with no ports!"));
726 for (i = c->ports-1; i > 1; i--) {
727 if (ETHER_ADDR_EQ(c->my_macs[i].etheraddr, eh->ether_dhost))
732 if (ETHER_ADDR_EQ(c->my_macs[1].etheraddr, eh->ether_dhost))
735 if (ETHER_ADDR_EQ(c->my_macs[0].etheraddr, eh->ether_dhost))
739 * Look for a possible destination in table
741 bt = &c->ht[HASH_FN(eh->ether_dhost)];
742 if (bt->name && ETHER_ADDR_EQ(bt->etheraddr, eh->ether_dhost))
749 * bridge_in() is invoked to perform bridging decision on input packets.
752 * eh Ethernet header of the incoming packet.
753 * ifp interface the packet is coming from.
755 * On Return: destination of packet, one of
756 * BDG_BCAST broadcast
757 * BDG_MCAST multicast
758 * BDG_LOCAL is only for a local address (do not forward)
759 * BDG_DROP drop the packet
760 * ifp ifp of the destination interface.
762 * Forwarding is not done directly to give a chance to some drivers
763 * to fetch more of the packet, or simply drop it completely.
766 static struct ifnet *
767 bridge_in(struct ifnet *ifp, struct ether_header *eh)
770 struct ifnet *dst, *old;
771 bdg_hash_table *bt; /* location in hash table */
772 int dropit = BDG_MUTED(ifp);
775 * hash the source address
778 index = HASH_FN(eh->ether_shost);
779 bt = &BDG_CLUSTER(ifp)->ht[index];
782 if (old) { /* the entry is valid */
783 if (!ETHER_ADDR_EQ(eh->ether_shost, bt->etheraddr)) {
785 bt->name = NULL; /* NB: will overwrite below */
786 } else if (old != ifp) {
788 * Found a loop. Either a machine has moved, or there
789 * is a misconfiguration/reconfiguration of the network.
790 * First, do not forward this packet!
791 * Record the relocation anyways; then, if loops persist,
792 * suspect a reconfiguration and disable forwarding
793 * from the old interface.
795 bt->name = ifp; /* relocate address */
796 printf("-- loop (%d) %6D to %s%d from %s%d (%s)\n",
797 bdg_loops, eh->ether_shost, ".",
798 ifp->if_name, ifp->if_unit,
799 old->if_name, old->if_unit,
800 BDG_MUTED(old) ? "muted":"active");
802 if (!BDG_MUTED(old)) {
803 if (bdg_loops++ > 10)
810 * now write the source address into the table
812 if (bt->name == NULL) {
813 DPRINTF(("%s: new addr %6D at %d for %s%d\n",
814 __func__, eh->ether_shost, ".", index, ifp->if_name, ifp->if_unit));
815 ETHER_ADDR_COPY(bt->etheraddr, eh->ether_shost);
818 dst = bridge_dst_lookup(eh, BDG_CLUSTER(ifp));
822 * bridge_dst_lookup can return the following values:
823 * BDG_BCAST, BDG_MCAST, BDG_LOCAL, BDG_UNKNOWN, BDG_DROP, ifp.
824 * For muted interfaces, or when we detect a loop, the first 3 are
825 * changed in BDG_LOCAL (we still listen to incoming traffic),
826 * and others to BDG_DROP (no use for the local host).
827 * Also, for incoming packets, ifp is changed to BDG_DROP if ifp == src.
828 * These changes are not necessary for outgoing packets from ether_output().
830 BDG_STAT(ifp, BDG_IN);
831 switch ((uintptr_t)dst) {
832 case (uintptr_t)BDG_BCAST:
833 case (uintptr_t)BDG_MCAST:
834 case (uintptr_t)BDG_LOCAL:
835 case (uintptr_t)BDG_UNKNOWN:
836 case (uintptr_t)BDG_DROP:
840 if (dst == ifp || dropit)
841 BDG_STAT(ifp, BDG_DROP);
843 BDG_STAT(ifp, BDG_FORWARD);
848 if (dst == BDG_BCAST || dst == BDG_MCAST || dst == BDG_LOCAL)
856 DPRINTF(("%s: %6D ->%6D ty 0x%04x dst %s%d\n", __func__,
857 eh->ether_shost, ".",
858 eh->ether_dhost, ".",
859 ntohs(eh->ether_type),
860 (dst <= BDG_FORWARD) ? bdg_dst_names[(uintptr_t)dst] :
862 (dst <= BDG_FORWARD) ? 0 : dst->if_unit));
868 * Return 1 if it's ok to send a packet out the specified interface.
869 * The interface must be:
874 * not the source interface, and
875 * belong to the same cluster as the 'real_dst'.
878 bridge_ifok(struct ifnet *ifp, struct ifnet *src, struct ifnet *dst)
880 return (BDG_USED(ifp)
882 && !_IF_QFULL(&ifp->if_snd)
883 && (ifp->if_flags & (IFF_UP|IFF_RUNNING)) == (IFF_UP|IFF_RUNNING)
885 && BDG_SAMECLUSTER(ifp, dst));
889 * Forward a packet to dst -- which can be a single interface or
890 * an entire cluster. The src port and muted interfaces are excluded.
892 * If src == NULL, the pkt comes from ether_output, and dst is the real
893 * interface the packet is originally sent to. In this case, we must forward
894 * it to the whole cluster.
895 * We never call bdg_forward from ether_output on interfaces which are
896 * not part of a cluster.
898 * If possible (i.e. we can determine that the caller does not need
899 * a copy), the packet is consumed here, and bdg_forward returns NULL.
900 * Otherwise, a pointer to a copy of the packet is returned.
903 bdg_forward(struct mbuf *m0, struct ifnet *dst)
905 #define EH_RESTORE(_m) do { \
906 M_PREPEND((_m), ETHER_HDR_LEN, M_DONTWAIT); \
907 if ((_m) == NULL) { \
911 if (eh != mtod((_m), struct ether_header *)) \
912 bcopy(&save_eh, mtod((_m), struct ether_header *), ETHER_HDR_LEN); \
916 struct ether_header *eh;
918 struct ifnet *ifp, *last;
919 int shared = bdg_copy; /* someone else is using the mbuf */
920 struct ifnet *real_dst = dst; /* real dst from ether_output */
921 struct ip_fw_args args;
922 struct ether_header save_eh;
925 DDB(quad_t ticks; ticks = rdtsc();)
927 args.rule = NULL; /* did we match a firewall rule ? */
928 /* Fetch state from dummynet tag, ignore others */
929 for (;m0->m_type == MT_TAG; m0 = m0->m_next)
930 if (m0->_m_tag_id == PACKET_TAG_DUMMYNET) {
931 args.rule = ((struct dn_pkt *)m0)->rule;
932 shared = 0; /* For sure this is our own mbuf. */
934 if (args.rule == NULL)
935 bdg_thru++; /* count 1st time through bdg_forward */
938 * The packet arrives with the Ethernet header at the front.
940 eh = mtod(m0, struct ether_header *);
942 src = m0->m_pkthdr.rcvif;
943 if (src == NULL) { /* packet from ether_output */
945 dst = bridge_dst_lookup(eh, BDG_CLUSTER(real_dst));
949 if (dst == BDG_DROP) { /* this should not happen */
950 printf("xx bdg_forward for BDG_DROP\n");
955 if (dst == BDG_LOCAL) { /* this should not happen as well */
956 printf("xx ouch, bdg_forward for local pkt\n");
959 if (dst == BDG_BCAST || dst == BDG_MCAST) {
960 /* need a copy for the local stack */
965 * Do filtering in a very similar way to what is done in ip_output.
966 * Only if firewall is loaded, enabled, and the packet is not
967 * from ether_output() (src==NULL, or we would filter it twice).
968 * Additional restrictions may apply e.g. non-IP, short packets,
969 * and pkts already gone through a pipe.
973 (inet_pfil_hook.ph_busy_count >= 0 && bdg_ipf != 0) ||
975 (IPFW_LOADED && bdg_ipfw != 0))) {
979 if (args.rule != NULL && fw_one_pass)
980 goto forward; /* packet already partially processed */
982 * i need some amt of data to be contiguous, and in case others need
983 * the packet (shared==1) also better be in the first mbuf.
985 i = min(m0->m_pkthdr.len, max_protohdr) ;
986 if (shared || m0->m_len < i) {
987 m0 = m_pullup(m0, i);
989 printf("%s: m_pullup failed\n", __func__); /* XXXDPRINTF*/
993 eh = mtod(m0, struct ether_header *);
997 * Processing below expects the Ethernet header is stripped.
998 * Furthermore, the mbuf chain might be replaced at various
999 * places. To deal with this we copy the header to a temporary
1000 * location, strip the header, and restore it as needed.
1002 bcopy(eh, &save_eh, ETHER_HDR_LEN); /* local copy for restore */
1003 m_adj(m0, ETHER_HDR_LEN); /* temporarily strip header */
1007 * NetBSD-style generic packet filter, pfil(9), hooks.
1008 * Enables ipf(8) in bridging.
1010 if (inet_pfil_hook.ph_busy_count >= 0 &&
1011 m0->m_pkthdr.len >= sizeof(struct ip) &&
1012 ntohs(save_eh.ether_type) == ETHERTYPE_IP) {
1014 * before calling the firewall, swap fields the same as IP does.
1015 * here we assume the pkt is an IP one and the header is contiguous
1017 struct ip *ip = mtod(m0, struct ip *);
1019 ip->ip_len = ntohs(ip->ip_len);
1020 ip->ip_off = ntohs(ip->ip_off);
1022 if (pfil_run_hooks(&inet_pfil_hook, &m0, src, PFIL_IN) != 0) {
1023 /* NB: hook should consume packet */
1026 if (m0 == NULL) /* consumed by filter */
1029 * If we get here, the firewall has passed the pkt, but the mbuf
1030 * pointer might have changed. Restore ip and the fields ntohs()'d.
1032 ip = mtod(m0, struct ip *);
1033 ip->ip_len = htons(ip->ip_len);
1034 ip->ip_off = htons(ip->ip_off);
1036 #endif /* PFIL_HOOKS */
1039 * Prepare arguments and call the firewall.
1041 if (!IPFW_LOADED || bdg_ipfw == 0) {
1042 EH_RESTORE(m0); /* restore Ethernet header */
1043 goto forward; /* not using ipfw, accept the packet */
1047 * XXX The following code is very similar to the one in
1048 * if_ethersubr.c:ether_ipfw_chk()
1051 args.m = m0; /* the packet we are looking at */
1052 args.oif = NULL; /* this is an input packet */
1053 args.divert_rule = 0; /* we do not support divert yet */
1054 args.next_hop = NULL; /* we do not support forward yet */
1055 args.eh = &save_eh; /* MAC header for bridged/MAC packets */
1056 i = ip_fw_chk_ptr(&args);
1057 m0 = args.m; /* in case the firewall used the mbuf */
1060 EH_RESTORE(m0); /* restore Ethernet header */
1062 if ( (i & IP_FW_PORT_DENY_FLAG) || m0 == NULL) /* drop */
1065 if (i == 0) /* a PASS rule. */
1067 if (DUMMYNET_LOADED && (i & IP_FW_PORT_DYNT_FLAG)) {
1069 * Pass the pkt to dummynet, which consumes it.
1070 * If shared, make a copy and keep the original.
1073 m = m_copypacket(m0, M_DONTWAIT);
1074 if (m == NULL) { /* copy failed, give up */
1079 m = m0 ; /* pass the original to dummynet */
1080 m0 = NULL ; /* and nothing back to the caller */
1083 args.oif = real_dst;
1084 ip_dn_io_ptr(m, (i & 0xffff),DN_TO_BDG_FWD, &args);
1088 * XXX at some point, add support for divert/forward actions.
1089 * If none of the above matches, we have to drop the packet.
1096 * Again, bring up the headers in case of shared bufs to avoid
1097 * corruptions in the future.
1100 int i = min(m0->m_pkthdr.len, max_protohdr);
1102 m0 = m_pullup(m0, i);
1107 /* NB: eh is not used below; no need to recalculate it */
1111 * now real_dst is used to determine the cluster where to forward.
1112 * For packets coming from ether_input, this is the one of the 'src'
1113 * interface, whereas for locally generated packets (src==NULL) it
1114 * is the cluster of the original destination interface, which
1115 * was already saved into real_dst.
1121 if (dst == BDG_BCAST || dst == BDG_MCAST || dst == BDG_UNKNOWN) {
1123 * Scan all ports and send copies to all but the last.
1125 IFNET_RLOCK(); /* XXX replace with generation # */
1126 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1127 if (bridge_ifok(ifp, src, real_dst)) {
1130 * At this point we know two interfaces need a copy
1131 * of the packet (last + ifp) so we must create a
1132 * copy to handoff to last.
1134 m = m_copypacket(m0, M_DONTWAIT);
1137 printf("%s: , m_copypacket failed!\n", __func__);
1139 return m0; /* the original is still there... */
1141 if (IF_HANDOFF(&last->if_snd, m, last))
1142 BDG_STAT(last, BDG_OUT);
1151 if (bridge_ifok(dst, src, real_dst))
1155 if (shared) { /* need to copy */
1156 m = m_copypacket(m0, M_DONTWAIT);
1158 printf("%s: sorry, m_copypacket failed!\n", __func__);
1160 return m0; /* the original is still there... */
1162 } else { /* consume original */
1165 if (IF_HANDOFF(&last->if_snd, m, last))
1166 BDG_STAT(last, BDG_OUT);
1171 DDB(bdg_fw_ticks += (u_long)(rdtsc() - ticks) ; bdg_fw_count++ ;
1172 if (bdg_fw_count != 0) bdg_fw_avg = bdg_fw_ticks/bdg_fw_count; )
1178 * initialization of bridge code.
1184 printf("BRIDGE %s loaded\n", bridge_version);
1186 ifp2sc = malloc(BDG_MAX_PORTS * sizeof(struct bdg_softc),
1187 M_IFADDR, M_WAITOK | M_ZERO );
1197 bzero(&bdg_stats, sizeof(bdg_stats));
1199 bridge_in_ptr = bridge_in;
1200 bdg_forward_ptr = bdg_forward;
1201 bdgtakeifaces_ptr = reconfigure_bridge;
1203 bdgtakeifaces_ptr(); /* XXX does this do anything? */
1205 callout_init(&bdg_callout, CALLOUT_MPSAFE);
1213 bridge_in_ptr = NULL;
1214 bdg_forward_ptr = NULL;
1215 bdgtakeifaces_ptr = NULL;
1217 callout_stop(&bdg_callout);
1222 free(ifp2sc, M_IFADDR);
1229 * initialization code, both for static and dynamic loading.
1232 bridge_modevent(module_t mod, int type, void *unused)
1255 static moduledata_t bridge_mod = {
1261 DECLARE_MODULE(bridge, bridge_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
1262 MODULE_VERSION(bridge, 1);