FreeBSD/Linux Kernel Cross Reference
sys/net/bridge.c
1 /*
2 * Copyright (c) 1998-2000 Luigi Rizzo
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
12 *
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23 * SUCH DAMAGE.
24 *
25 * $FreeBSD$
26 */
27
28 /*
29 * This code implements bridging in FreeBSD. It only acts on ethernet
30 * type of interfaces (others are still usable for routing).
31 * A bridging table holds the source MAC address/dest. interface for each
32 * known node. The table is indexed using an hash of the source address.
33 *
34 * Input packets are tapped near the end of the input routine in each
35 * driver (near the call to bpf_mtap, or before the call to ether_input)
36 * and analysed calling bridge_in(). Depending on the result, the packet
37 * can be forwarded to one or more output interfaces using bdg_forward(),
38 * and/or sent to the upper layer (e.g. in case of multicast).
39 *
40 * Output packets are intercepted near the end of ether_output(),
41 * the correct destination is selected calling bdg_dst_lookup(),
42 * and then forwarding is done using bdg_forward().
43 * Bridging is controlled by the sysctl variable net.link.ether.bridge
44 *
45 * The arp code is also modified to let a machine answer to requests
46 * irrespective of the port the request came from.
47 *
48 * In case of loops in the bridging topology, the bridge detects this
49 * event and temporarily mutes output bridging on one of the ports.
50 * Periodically, interfaces are unmuted by bdg_timeout().
51 * Muting is only implemented as a safety measure, and also as
52 * a mechanism to support a user-space implementation of the spanning
53 * tree algorithm. In the final release, unmuting will only occur
54 * because of explicit action of the user-level daemon.
55 *
56 * To build a bridging kernel, use the following option
57 * option BRIDGE
58 * and then at runtime set the sysctl variable to enable bridging.
59 *
60 * Only one interface is supposed to have addresses set (but
61 * there are no problems in practice if you set addresses for more
62 * than one interface).
63 * Bridging will act before routing, but nothing prevents a machine
64 * from doing both (modulo bugs in the implementation...).
65 *
66 * THINGS TO REMEMBER
67 * - bridging requires some (small) modifications to the interface
68 * driver. Not all of them have been changed, see the "ed" and "de"
69 * drivers as examples on how to operate.
70 * - bridging is incompatible with multicast routing on the same
71 * machine. There is not an easy fix to this.
72 * - loop detection is still not very robust.
73 * - the interface of bdg_forward() could be improved.
74 */
75
76 #include <sys/param.h>
77 #include <sys/mbuf.h>
78 #include <sys/malloc.h>
79 #include <sys/systm.h>
80 #include <sys/socket.h> /* for net/if.h */
81 #include <sys/kernel.h>
82 #include <sys/sysctl.h>
83
84 #include <net/if.h>
85 #include <net/if_types.h>
86
87 #include <netinet/in.h> /* for struct arpcom */
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/if_ether.h> /* for struct arpcom */
92
93 #include "opt_ipfw.h"
94 #include "opt_ipdn.h"
95
96 #if defined(IPFIREWALL)
97 #include <net/route.h>
98 #include <netinet/ip_fw.h>
99 #if defined(DUMMYNET)
100 #include <netinet/ip_dummynet.h>
101 #endif
102 #endif
103
104 #include <net/bridge.h>
105
106 /*
107 * For debugging, you can use the following macros.
108 * remember, rdtsc() only works on Pentium-class machines
109
110 quad_t ticks;
111 DDB(ticks = rdtsc();)
112 ... interesting code ...
113 DDB(bdg_fw_ticks += (u_long)(rdtsc() - ticks) ; bdg_fw_count++ ;)
114
115 *
116 */
117
118 #define DDB(x) x
119 #define DEB(x)
120
121 static void bdginit(void *);
122 static void flush_table(void);
123 static void bdg_promisc_on(void);
124 static void parse_bdg_cfg(void);
125
126 static int bdg_ipfw = 0 ;
127 int do_bridge = 0;
128 bdg_hash_table *bdg_table = NULL ;
129
130 /*
131 * System initialization
132 */
133
134 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, bdginit, NULL)
135
136 /*
137 * We need additional info for the bridge. The bdg_ifp2sc[] array
138 * provides a pointer to this struct using the if_index.
139 * bdg_softc has a backpointer to the struct ifnet, the bridge
140 * flags, and a cluster (bridging occurs only between port of the
141 * same cluster).
142 */
143 struct bdg_softc {
144 struct ifnet *ifp ;
145 /* ((struct arpcom *)ifp)->ac_enaddr is the eth. addr */
146 int flags ;
147 #define IFF_BDG_PROMISC 0x0001 /* set promisc mode on this if. */
148 #define IFF_MUTE 0x0002 /* mute this if for bridging. */
149 #define IFF_USED 0x0004 /* use this if for bridging. */
150 short cluster_id ; /* in network format */
151 u_long magic;
152 } ;
153
154 static struct bdg_stats bdg_stats ;
155 static struct bdg_softc *ifp2sc = NULL ;
156 /* XXX make it static of size BDG_MAX_PORTS */
157
158 #define USED(ifp) (ifp2sc[ifp->if_index].flags & IFF_USED)
159 #define MUTED(ifp) (ifp2sc[ifp->if_index].flags & IFF_MUTE)
160 #define MUTE(ifp) ifp2sc[ifp->if_index].flags |= IFF_MUTE
161 #define UNMUTE(ifp) ifp2sc[ifp->if_index].flags &= ~IFF_MUTE
162 #define CLUSTER(ifp) (ifp2sc[ifp->if_index].cluster_id)
163 #define IFP_CHK(ifp, x) \
164 if (ifp2sc[ifp->if_index].magic != 0xDEADBEEF) { x ; }
165
166 #define SAMECLUSTER(ifp,src,eh) \
167 (src == NULL || CLUSTER(ifp) == CLUSTER(src) )
168
169 /*
170 * turn off promisc mode, optionally clear the IFF_USED flag
171 */
172 static void
173 bdg_promisc_off(int clear_used)
174 {
175 struct ifnet *ifp ;
176 for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next ) {
177 if ( (ifp2sc[ifp->if_index].flags & IFF_BDG_PROMISC) ) {
178 int s, ret ;
179 s = splimp();
180 ret = ifpromisc(ifp, 0);
181 splx(s);
182 ifp2sc[ifp->if_index].flags &= ~(IFF_BDG_PROMISC|IFF_MUTE) ;
183 if (clear_used)
184 ifp2sc[ifp->if_index].flags &= ~(IFF_USED) ;
185 printf(">> now %s%d promisc ON if_flags 0x%x bdg_flags 0x%x\n",
186 ifp->if_name, ifp->if_unit,
187 ifp->if_flags, ifp2sc[ifp->if_index].flags);
188 }
189 }
190 }
191
192 /*
193 * set promisc mode on the interfaces we use.
194 */
195 static void
196 bdg_promisc_on()
197 {
198 struct ifnet *ifp ;
199 int s ;
200
201 for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next ) {
202 if ( !USED(ifp) )
203 continue ;
204 if ( 0 == ( ifp->if_flags & IFF_UP) ) {
205 s = splimp();
206 if_up(ifp);
207 splx(s);
208 }
209 if ( !(ifp2sc[ifp->if_index].flags & IFF_BDG_PROMISC) ) {
210 int ret ;
211 s = splimp();
212 ret = ifpromisc(ifp, 1);
213 splx(s);
214 ifp2sc[ifp->if_index].flags |= IFF_BDG_PROMISC ;
215 printf(">> now %s%d promisc ON if_flags 0x%x bdg_flags 0x%x\n",
216 ifp->if_name, ifp->if_unit,
217 ifp->if_flags, ifp2sc[ifp->if_index].flags);
218 }
219 if (MUTED(ifp)) {
220 printf(">> unmuting %s%d\n", ifp->if_name, ifp->if_unit);
221 UNMUTE(ifp) ;
222 }
223 }
224 }
225
226 static int
227 sysctl_bdg SYSCTL_HANDLER_ARGS
228 {
229 int error, oldval = do_bridge ;
230
231 error = sysctl_handle_int(oidp,
232 oidp->oid_arg1, oidp->oid_arg2, req);
233 DEB( printf("called sysctl for bridge name %s arg2 %d val %d->%d\n",
234 oidp->oid_name, oidp->oid_arg2,
235 oldval, do_bridge); )
236
237 if (bdg_table == NULL)
238 do_bridge = 0 ;
239 if (oldval != do_bridge) {
240 flush_table();
241 if (do_bridge == 0)
242 bdg_promisc_off( 0 ); /* leave IFF_USED set */
243 else
244 bdg_promisc_on();
245 }
246 return error ;
247 }
248
249 static char bridge_cfg[256] = { "" } ;
250
251 /*
252 * parse the config string, set IFF_USED, name and cluster_id
253 * for all interfaces found.
254 */
255 static void
256 parse_bdg_cfg()
257 {
258 char *p, *beg ;
259 int i, l, cluster;
260 struct bdg_softc *b;
261
262 for (p= bridge_cfg; *p ; p++) {
263 /* interface names begin with [a-z] and continue up to ':' */
264 if (*p < 'a' || *p > 'z')
265 continue ;
266 for ( beg = p ; *p && *p != ':' ; p++ )
267 ;
268 if (*p == 0) /* end of string, ':' not found */
269 return ;
270 l = p - beg ; /* length of name string */
271 p++ ;
272 DDB(printf("-- match beg(%d) <%s> p <%s>\n", l, beg, p);)
273 for (cluster = 0 ; *p && *p >= '' && *p <= '9' ; p++)
274 cluster = cluster*10 + (*p -'');
275 /*
276 * now search in bridge strings
277 */
278 for (i=0, b = ifp2sc ; i < if_index ; i++, b++) {
279 char buf[32];
280 struct ifnet *ifp = b->ifp ;
281
282 if (ifp == NULL)
283 continue;
284 sprintf(buf, "%s%d", ifp->if_name, ifp->if_unit);
285 if (!strncmp(beg, buf, l)) { /* XXX not correct for >10 if! */
286 b->cluster_id = htons(cluster) ;
287 b->flags |= IFF_USED ;
288 sprintf(bdg_stats.s[ifp->if_index].name+l,
289 ":%d", cluster);
290
291 DDB(printf("--++ found %s\n",
292 bdg_stats.s[ifp->if_index].name);)
293 break ;
294 }
295 }
296 }
297 }
298
299 static int
300 sysctl_bdg_cfg SYSCTL_HANDLER_ARGS
301 {
302 int error = 0 ;
303 char oldval[256] ;
304
305 strcpy(oldval, bridge_cfg) ;
306
307 error = sysctl_handle_string(oidp,
308 bridge_cfg, oidp->oid_arg2, req);
309 DEB(
310 printf("called sysctl for bridge name %s arg2 %d err %d val %s->%s\n",
311 oidp->oid_name, oidp->oid_arg2,
312 error,
313 oldval, bridge_cfg);
314 )
315 if (strcmp(oldval, bridge_cfg)) {
316 bdg_promisc_off( 1 ); /* reset previously-used interfaces */
317 flush_table();
318 parse_bdg_cfg(); /* and set new ones... */
319 if (do_bridge)
320 bdg_promisc_on(); /* re-enable interfaces */
321 }
322 return error ;
323 }
324
325 SYSCTL_PROC(_net_link_ether, OID_AUTO, bridge_cfg, CTLTYPE_STRING|CTLFLAG_RW,
326 &bridge_cfg, sizeof(bridge_cfg), &sysctl_bdg_cfg, "A",
327 "Bridge configuration");
328
329 SYSCTL_PROC(_net_link_ether, OID_AUTO, bridge, CTLTYPE_INT|CTLFLAG_RW,
330 &do_bridge, 0, &sysctl_bdg, "I", "Bridging");
331
332 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_ipfw, CTLFLAG_RW,
333 &bdg_ipfw,0,"Pass bridged pkts through firewall");
334
335 int bdg_ipfw_drops;
336 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_ipfw_drop,
337 CTLFLAG_RW, &bdg_ipfw_drops,0,"");
338
339 int bdg_ipfw_colls;
340 SYSCTL_INT(_net_link_ether, OID_AUTO, bridge_ipfw_collisions,
341 CTLFLAG_RW, &bdg_ipfw_colls,0,"");
342
343 #if 1 /* diagnostic vars */
344 int bdg_in_count = 0 , bdg_in_ticks = 0 , bdg_fw_count = 0, bdg_fw_ticks = 0 ;
345 SYSCTL_INT(_net_link_ether, OID_AUTO, bdginc, CTLFLAG_RW, &bdg_in_count,0,"");
346 SYSCTL_INT(_net_link_ether, OID_AUTO, bdgint, CTLFLAG_RW, &bdg_in_ticks,0,"");
347 SYSCTL_INT(_net_link_ether, OID_AUTO, bdgfwc, CTLFLAG_RW, &bdg_fw_count,0,"");
348 SYSCTL_INT(_net_link_ether, OID_AUTO, bdgfwt, CTLFLAG_RW, &bdg_fw_ticks,0,"");
349 #endif
350
351 SYSCTL_STRUCT(_net_link_ether, PF_BDG, bdgstats,
352 CTLFLAG_RD, &bdg_stats , bdg_stats, "bridge statistics");
353
354 static int bdg_loops ;
355
356 /*
357 * completely flush the bridge table.
358 */
359 static void
360 flush_table()
361 {
362 int s,i;
363
364 if (bdg_table == NULL)
365 return ;
366 s = splimp();
367 for (i=0; i< HASH_SIZE; i++)
368 bdg_table[i].name= NULL; /* clear table */
369 splx(s);
370 }
371
372 /*
373 * called periodically to flush entries etc.
374 */
375 static void
376 bdg_timeout(void *dummy)
377 {
378 static int slowtimer = 0 ;
379
380 if (do_bridge) {
381 static int age_index = 0 ; /* index of table position to age */
382 int l = age_index + HASH_SIZE/4 ;
383 /*
384 * age entries in the forwarding table.
385 */
386 if (l > HASH_SIZE)
387 l = HASH_SIZE ;
388 for (; age_index < l ; age_index++)
389 if (bdg_table[age_index].used)
390 bdg_table[age_index].used = 0 ;
391 else if (bdg_table[age_index].name) {
392 /* printf("xx flushing stale entry %d\n", age_index); */
393 bdg_table[age_index].name = NULL ;
394 }
395 if (age_index >= HASH_SIZE)
396 age_index = 0 ;
397
398 if (--slowtimer <= 0 ) {
399 slowtimer = 5 ;
400
401 bdg_promisc_on() ; /* we just need unmute, really */
402 bdg_loops = 0 ;
403 }
404 }
405 timeout(bdg_timeout, (void *)0, 2*hz );
406 }
407
408 /*
409 * local MAC addresses are held in a small array. This makes comparisons
410 * much faster.
411 */
412 unsigned char bdg_addresses[6*BDG_MAX_PORTS];
413 int bdg_ports ;
414
415 /*
416 * initialization of bridge code. This needs to be done after all
417 * interfaces have been configured.
418 */
419 static void
420 bdginit(void *dummy)
421 {
422 int i ;
423 struct ifnet *ifp;
424 struct arpcom *ac ;
425 u_char *eth_addr ;
426 struct bdg_softc *bp;
427
428 if (bdg_table == NULL)
429 bdg_table = (struct hash_table *)
430 malloc(HASH_SIZE * sizeof(struct hash_table),
431 M_IFADDR, M_WAITOK);
432 flush_table();
433
434 ifp2sc = malloc(BDG_MAX_PORTS * sizeof(struct bdg_softc),
435 M_IFADDR, M_WAITOK );
436 bzero(ifp2sc, BDG_MAX_PORTS * sizeof(struct bdg_softc) );
437
438 bzero(&bdg_stats, sizeof(bdg_stats) );
439 bdg_ports = 0 ;
440 eth_addr = bdg_addresses ;
441
442 printf("BRIDGE 990810, have %d interfaces\n", if_index);
443 for (i = 0 , ifp = ifnet.tqh_first ; i < if_index ;
444 i++, ifp = ifp->if_link.tqe_next)
445 if (ifp->if_type == IFT_ETHER) { /* ethernet ? */
446 bp = &ifp2sc[ifp->if_index] ;
447 ac = (struct arpcom *)ifp;
448 sprintf(bridge_cfg + strlen(bridge_cfg),
449 "%s%d:1,", ifp->if_name, ifp->if_unit);
450 printf("-- index %d %s type %d phy %d addrl %d addr %6D\n",
451 ifp->if_index,
452 bdg_stats.s[ifp->if_index].name,
453 (int)ifp->if_type, (int) ifp->if_physical,
454 (int)ifp->if_addrlen,
455 ac->ac_enaddr, "." );
456 bcopy(ac->ac_enaddr, eth_addr, 6);
457 eth_addr += 6 ;
458 bp->ifp = ifp ;
459 bp->flags = IFF_USED ;
460 bp->cluster_id = htons(1) ;
461 bp->magic = 0xDEADBEEF ;
462
463 sprintf(bdg_stats.s[ifp->if_index].name,
464 "%s%d:%d", ifp->if_name, ifp->if_unit,
465 ntohs(bp->cluster_id));
466 bdg_ports ++ ;
467 }
468
469 bdg_timeout(0);
470 do_bridge=0;
471 }
472
473 /*
474 * bridge_in() is invoked to perform bridging decision on input packets.
475 * On Input:
476 * m packet to be bridged. The mbuf need not to hold the
477 * whole packet, only the first 14 bytes suffice. We
478 * assume them to be contiguous. No alignment assumptions
479 * because they are not a problem on i386 class machines.
480 *
481 * On Return: destination of packet, one of
482 * BDG_BCAST broadcast
483 * BDG_MCAST multicast
484 * BDG_LOCAL is only for a local address (do not forward)
485 * BDG_DROP drop the packet
486 * ifp ifp of the destination interface.
487 *
488 * Forwarding is not done directly to give a chance to some drivers
489 * to fetch more of the packet, or simply drop it completely.
490 */
491
492
493 struct ifnet *
494 bridge_in(struct mbuf *m)
495 {
496 int index;
497 struct ifnet *ifp = m->m_pkthdr.rcvif, *dst , *old ;
498 int dropit = MUTED(ifp) ;
499 struct ether_header *eh;
500
501 eh = mtod(m, struct ether_header *);
502
503 /*
504 * hash the source address
505 */
506 index= HASH_FN(eh->ether_shost);
507 bdg_table[index].used = 1 ;
508 old = bdg_table[index].name ;
509 if ( old ) { /* the entry is valid. */
510 IFP_CHK(old, printf("bridge_in-- reading table\n") );
511
512 if (!BDG_MATCH( eh->ether_shost, bdg_table[index].etheraddr) ) {
513 bdg_ipfw_colls++ ;
514 bdg_table[index].name = NULL ;
515 } else if (old != ifp) {
516 /*
517 * found a loop. Either a machine has moved, or there
518 * is a misconfiguration/reconfiguration of the network.
519 * First, do not forward this packet!
520 * Record the relocation anyways; then, if loops persist,
521 * suspect a reconfiguration and disable forwarding
522 * from the old interface.
523 */
524 bdg_table[index].name = ifp ; /* relocate address */
525 printf("-- loop (%d) %6D to %s%d from %s%d (%s)\n",
526 bdg_loops, eh->ether_shost, ".",
527 ifp->if_name, ifp->if_unit,
528 old->if_name, old->if_unit,
529 MUTED(old) ? "muted":"active");
530 dropit = 1 ;
531 if ( !MUTED(old) ) {
532 if (++bdg_loops > 10)
533 MUTE(old) ;
534 }
535 }
536 }
537
538 /*
539 * now write the source address into the table
540 */
541 if (bdg_table[index].name == NULL) {
542 DEB(printf("new addr %6D at %d for %s%d\n",
543 eh->ether_shost, ".", index, ifp->if_name, ifp->if_unit);)
544 bcopy(eh->ether_shost, bdg_table[index].etheraddr, 6);
545 bdg_table[index].name = ifp ;
546 }
547 dst = bridge_dst_lookup(m);
548 /* Return values:
549 * BDG_BCAST, BDG_MCAST, BDG_LOCAL, BDG_UNKNOWN, BDG_DROP, ifp.
550 * For muted interfaces, the first 3 are changed in BDG_LOCAL,
551 * and others to BDG_DROP. Also, for incoming packets, ifp is changed
552 * to BDG_DROP in case ifp == src . These mods are not necessary
553 * for outgoing packets from ether_output().
554 */
555 BDG_STAT(ifp, BDG_IN);
556 switch ((int)dst) {
557 case (int)BDG_BCAST:
558 case (int)BDG_MCAST:
559 case (int)BDG_LOCAL:
560 case (int)BDG_UNKNOWN:
561 case (int)BDG_DROP:
562 BDG_STAT(ifp, dst);
563 break ;
564 default :
565 if (dst == ifp || dropit )
566 BDG_STAT(ifp, BDG_DROP);
567 else
568 BDG_STAT(ifp, BDG_FORWARD);
569 break ;
570 }
571
572 if ( dropit ) {
573 if (dst == BDG_BCAST || dst == BDG_MCAST || dst == BDG_LOCAL)
574 return BDG_LOCAL ;
575 else
576 return BDG_DROP ;
577 } else {
578 return (dst == ifp ? BDG_DROP : dst ) ;
579 }
580 }
581
582 /*
583 * Forward to dst, excluding src port and muted interfaces.
584 * The packet is freed if possible (i.e. surely not of interest for
585 * the upper layer), otherwise a copy is left for use by the caller
586 * (pointer in *m0).
587 *
588 * It would be more efficient to make bdg_forward() always consume
589 * the packet, leaving to the caller the task to check if it needs a copy
590 * and get one in case. As it is now, bdg_forward() can sometimes make
591 * a copy whereas it is not necessary.
592 *
593 * INPUT:
594 * *m0 -- ptr to pkt (not null at call time)
595 * *dst -- destination (special value or ifnet *)
596 * (*m0)->m_pkthdr.rcvif -- NULL only for output pkts.
597 * OUTPUT:
598 * *m0 -- pointer to the packet (NULL if still existent)
599 */
600 int
601 bdg_forward (struct mbuf **m0, struct ifnet *dst)
602 {
603 struct ifnet *src = (*m0)->m_pkthdr.rcvif; /* could be NULL in output */
604 struct ifnet *ifp, *last = NULL ;
605 int error=0, s ;
606 int canfree = 0 ; /* can free the buf at the end if set */
607 int once = 0; /* loop only once */
608 struct mbuf *m ;
609
610 struct ether_header *eh = mtod(*m0, struct ether_header *); /* XXX */
611
612 if (dst == BDG_DROP) { /* this should not happen */
613 printf("xx bdg_forward for BDG_DROP\n");
614 m_freem(*m0) ;
615 *m0 = NULL ;
616 return 0;
617 }
618 if (dst == BDG_LOCAL) { /* this should not happen as well */
619 printf("xx ouch, bdg_forward for local pkt\n");
620 return 0;
621 }
622 if (dst == BDG_BCAST || dst == BDG_MCAST || dst == BDG_UNKNOWN) {
623 ifp = ifnet.tqh_first ; /* scan all ports */
624 once = 0 ;
625 if (dst != BDG_UNKNOWN) /* need a copy for the local stack */
626 canfree = 0 ;
627 } else {
628 ifp = dst ;
629 once = 1 ;
630 }
631 if ( (u_int)(ifp) <= (u_int)BDG_FORWARD )
632 panic("bdg_forward: bad dst");
633
634 #ifdef IPFIREWALL
635 /*
636 * do filtering in a very similar way to what is done
637 * in ip_output. Only for IP packets, and only pass/fail/dummynet
638 * is supported. The tricky thing is to make sure that enough of
639 * the packet (basically, Eth+IP+TCP/UDP headers) is contiguous
640 * so that calls to m_pullup in ip_fw_chk will not kill the
641 * ethernet header.
642 */
643 if (ip_fw_chk_ptr) {
644 u_int16_t dummy = 0 ;
645 struct ip_fw_chain *rule = NULL ;
646 int off;
647 struct ip *ip ;
648
649 m = *m0 ;
650 #ifdef DUMMYNET
651 if (m->m_type == MT_DUMMYNET) {
652 /*
653 * the packet was already tagged, so part of the
654 * processing was already done, and we need to go down.
655 */
656 rule = (struct ip_fw_chain *)(m->m_data) ;
657 (*m0) = m = m->m_next ;
658
659 src = m->m_pkthdr.rcvif; /* could be NULL in output */
660 eh = mtod(m, struct ether_header *); /* XXX */
661 canfree = 1 ; /* for sure, a copy is not needed later. */
662 goto forward; /* HACK! I should obey the fw_one_pass */
663 }
664 #endif
665 if (bdg_ipfw == 0) /* this test must be here. */
666 goto forward ;
667 if (src == NULL)
668 goto forward ; /* do not apply to packets from ether_output */
669 /*
670 * In this section, canfree=1 means m is the same as *m0.
671 * canfree==0 means m is a copy. We need to make a copy here
672 * (to be destroyed on exit from the firewall section) because
673 * the firewall itself might destroy the packet.
674 * (This is not very smart... i should really change ipfw to
675 * leave the pkt alive!)
676 */
677 if (canfree == 0 ) {
678 /*
679 * Need to make a copy (and for good measure, make sure that
680 * the header is contiguous). The original is still in *m0
681 */
682 int needed = min(MHLEN, 14+max_protohdr) ;
683 needed = min(needed, (*m0)->m_len ) ;
684
685 m = m_copypacket( (*m0), M_DONTWAIT);
686 if (m == NULL) {
687 printf("-- bdg: m_copypacket failed.\n") ;
688 return ENOBUFS ;
689 }
690 m = m_pullup(m, needed) ;
691 if ( m == NULL ) {
692 printf("-- bdg: pullup failed.\n") ;
693 return ENOBUFS ;
694 }
695 }
696
697 /*
698 * before calling the firewall, swap fields the same as IP does.
699 * here we assume the pkt is an IP one and the header is contiguous
700 */
701 eh = mtod(m, struct ether_header *);
702 ip = (struct ip *)(eh + 1 ) ;
703 NTOHS(ip->ip_len);
704 NTOHS(ip->ip_id);
705 NTOHS(ip->ip_off);
706
707 /*
708 * The third parameter to the firewall code is the dst. interface.
709 * Since we apply checks only on input pkts we use NULL.
710 */
711 off = (*ip_fw_chk_ptr)(NULL, 0, NULL, &dummy, &m, &rule, NULL) ;
712
713 if (m == NULL) { /* pkt discarded by firewall */
714 /*
715 * At this point, if canfree==1, m and *m0 were the same
716 * thing, so just clear ptr. Otherwise, leave it alone, the
717 * upper layer might still make use of it somewhere.
718 */
719 if (canfree)
720 *m0 = NULL ;
721 return 0 ;
722 }
723
724 /*
725 * If we get here, the firewall has passed the pkt, but the
726 * mbuf pointer might have changed. Restore eh, ip, and the
727 * fields NTOHS()'d. Then, if canfree==1, also restore *m0.
728 */
729 eh = mtod(m, struct ether_header *);
730 ip = (struct ip *)(eh + 1 ) ;
731 HTONS(ip->ip_len);
732 HTONS(ip->ip_id);
733 HTONS(ip->ip_off);
734 if (canfree) /* m was a reference to *m0, so update *m0 */
735 *m0 = m ;
736
737 if (off == 0) { /* a PASS rule. */
738 if (canfree == 0) /* destroy the copy */
739 m_freem(m);
740 goto forward ;
741 }
742 #ifdef DUMMYNET
743 if (off & 0x10000) {
744 /*
745 * pass the pkt to dummynet. Need to include m, dst, rule.
746 * Dummynet consumes the packet in all cases.
747 */
748 dummynet_io((off & 0xffff), DN_TO_BDG_FWD, m, dst, NULL, 0, rule);
749 if (canfree) /* dummynet has consumed the original one */
750 *m0 = NULL ;
751 return 0 ;
752 }
753 #endif
754 /* if none of the above matches, we have to drop the pkt */
755 bdg_ipfw_drops++ ;
756 m_freem(m);
757 if (canfree == 0) /* m was a copy */
758 m_freem(*m0);
759 *m0 = NULL ;
760 return 0 ;
761 }
762 forward:
763 #endif /* IPFIREWALL */
764 /*
765 * Now *m0 is the only pkt left. If canfree != 0 the pkt might be
766 * used by the upper layers which could scramble header fields.
767 * (basically ntoh*() etc.). To avoid problems, make sure that
768 * all fields that might be changed by the local network stack are not
769 * in a cluster by calling m_pullup on *m0. We lose some efficiency
770 * but better than having packets corrupt!
771 */
772 if (canfree == 0 ) {
773 int needed = min(MHLEN, 14+max_protohdr) ;
774 needed = min(needed, (*m0)->m_len ) ;
775
776 *m0 = m_pullup( *m0, needed) ;
777 if ( *m0 == NULL ) {
778 printf("-- bdg: pullup failed.\n") ;
779 return ENOBUFS ;
780 }
781 }
782 for (;;) {
783 if (last) { /* need to forward packet */
784 if (canfree && once ) { /* no need to copy */
785 m = *m0 ;
786 *m0 = NULL ; /* original is gone */
787 } else /* on a P5-90, m_copypacket takes 540 ticks */
788 m = m_copypacket(*m0, M_DONTWAIT);
789 if (m == NULL) {
790 printf("bdg_forward: sorry, m_copy failed!\n");
791 return ENOBUFS ; /* the original is still there... */
792 }
793 /*
794 * Last part of ether_output: queue pkt and start
795 * output if interface not yet active.
796 */
797 s = splimp();
798 if (IF_QFULL(&last->if_snd)) {
799 IF_DROP(&last->if_snd);
800 #if 0
801 MUTE(last); /* should I also mute ? */
802 #endif
803 splx(s);
804 m_freem(m); /* consume the pkt anyways */
805 error = ENOBUFS ;
806 } else {
807 last->if_obytes += m->m_pkthdr.len ;
808 if (m->m_flags & M_MCAST)
809 last->if_omcasts++;
810 IF_ENQUEUE(&last->if_snd, m);
811 if ((last->if_flags & IFF_OACTIVE) == 0)
812 (*last->if_start)(last);
813 splx(s);
814 }
815 BDG_STAT(last, BDG_OUT);
816 last = NULL ;
817 if (once)
818 break ;
819 }
820 if (ifp == NULL)
821 break ;
822 if (ifp != src && /* do not send to self */
823 USED(ifp) && /* if used for bridging */
824 ! IF_QFULL(&ifp->if_snd) &&
825 (ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
826 (IFF_UP|IFF_RUNNING) &&
827 SAMECLUSTER(ifp, src, eh) && !MUTED(ifp) )
828 last = ifp ;
829 ifp = ifp->if_link.tqe_next ;
830 if (ifp == NULL)
831 once = 1 ;
832 }
833
834 return error ;
835 }
Cache object: 6eef951d682c2df17802239189e3a03d
|