Clean up the networking code before I parallelize the routing code.
[dragonfly.git] / sys / netinet / ip_output.c
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.24 2004/12/28 08:09:59 hsu Exp $
32  */
33
34 #define _IP_VHL
35
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipfilter.h"
40 #include "opt_ipsec.h"
41 #include "opt_random_ip_id.h"
42 #include "opt_mbuf_stress_test.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/in_cksum.h>
55
56 #include <net/if.h>
57 #include <net/netisr.h>
58 #include <net/pfil.h>
59 #include <net/route.h>
60
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67
68 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
69
70 #ifdef IPSEC
71 #include <netinet6/ipsec.h>
72 #include <netproto/key/key.h>
73 #ifdef IPSEC_DEBUG
74 #include <netproto/key/key_debug.h>
75 #else
76 #define KEYDEBUG(lev,arg)
77 #endif
78 #endif /*IPSEC*/
79
80 #ifdef FAST_IPSEC
81 #include <netproto/ipsec/ipsec.h>
82 #include <netproto/ipsec/xform.h>
83 #include <netproto/ipsec/key.h>
84 #endif /*FAST_IPSEC*/
85
86 #include <net/ipfw/ip_fw.h>
87 #include <net/dummynet/ip_dummynet.h>
88
89 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
90                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
91                                   (ntohl(a.s_addr)>>16)&0xFF,\
92                                   (ntohl(a.s_addr)>>8)&0xFF,\
93                                   (ntohl(a.s_addr))&0xFF, y);
94
95 u_short ip_id;
96
97 #ifdef MBUF_STRESS_TEST
98 int mbuf_frag_size = 0;
99 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
100         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
101 #endif
102
103 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
104 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
105 static void     ip_mloopback
106         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
107 static int      ip_getmoptions
108         (struct sockopt *, struct ip_moptions *);
109 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
110 static int      ip_setmoptions
111         (struct sockopt *, struct ip_moptions **);
112
113 int     ip_optcopy(struct ip *, struct ip *);
114
115
116 extern  struct protosw inetsw[];
117
118 /*
119  * IP output.  The packet in mbuf chain m contains a skeletal IP
120  * header (with len, off, ttl, proto, tos, src, dst).
121  * The mbuf chain containing the packet will be freed.
122  * The mbuf opt, if present, will not be freed.
123  */
124 int
125 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
126           int flags, struct ip_moptions *imo, struct inpcb *inp)
127 {
128         struct ip *ip;
129         struct ifnet *ifp = NULL;       /* keep compiler happy */
130         struct mbuf *m;
131         int hlen = sizeof(struct ip);
132         int len, off, error = 0;
133         struct sockaddr_in *dst = NULL; /* keep compiler happy */
134         struct in_ifaddr *ia = NULL;
135         int isbroadcast, sw_csum;
136         struct in_addr pkt_dst;
137         struct route iproute;
138 #ifdef IPSEC
139         struct secpolicy *sp = NULL;
140         struct socket *so = inp ? inp->inp_socket : NULL;
141 #endif
142 #ifdef FAST_IPSEC
143         struct m_tag *mtag;
144         struct secpolicy *sp = NULL;
145         struct tdb_ident *tdbi;
146         int s;
147 #endif /* FAST_IPSEC */
148         struct ip_fw_args args;
149         int src_was_INADDR_ANY = 0;     /* as the name says... */
150
151         args.eh = NULL;
152         args.rule = NULL;
153         args.next_hop = NULL;
154         args.divert_rule = 0;                   /* divert cookie */
155
156         /* Grab info from MT_TAG mbufs prepended to the chain. */
157         while (m0 != NULL && m0->m_type == MT_TAG) {
158                 switch(m0->_m_tag_id) {
159                 case PACKET_TAG_DUMMYNET:
160                         /*
161                          * the packet was already tagged, so part of the
162                          * processing was already done, and we need to go down.
163                          * Get parameters from the header.
164                          */
165                         args.rule = ((struct dn_pkt *)m0)->rule;
166                         opt = NULL ;
167                         ro = &((struct dn_pkt *)m0)->ro;
168                         imo = NULL ;
169                         dst = ((struct dn_pkt *)m0)->dn_dst ;
170                         ifp = ((struct dn_pkt *)m0)->ifp ;
171                         flags = ((struct dn_pkt *)m0)->flags ;
172                         break;
173                 case PACKET_TAG_DIVERT:
174                         args.divert_rule = (int)m0->m_data & 0xffff;
175                         break;
176                 case PACKET_TAG_IPFORWARD:
177                         args.next_hop = (struct sockaddr_in *)m0->m_data;
178                         break;
179                 default:
180                         printf("ip_output: unrecognised MT_TAG tag %d\n",
181                             m0->_m_tag_id);
182                         break;
183                 }
184                 m0 = m0->m_next;
185         }
186         m = m0;
187         KASSERT(m != NULL && (m->m_flags & M_PKTHDR), ("ip_output: no HDR"));
188
189         if (ro == NULL) {
190                 ro = &iproute;
191                 bzero(ro, sizeof *ro);
192         }
193
194         if (args.rule != NULL) {        /* dummynet already saw us */
195                 ip = mtod(m, struct ip *);
196                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
197                 if (ro->ro_rt)
198                         ia = ifatoia(ro->ro_rt->rt_ifa);
199                 goto sendit;
200         }
201
202         if (opt) {
203                 len = 0;
204                 m = ip_insertoptions(m, opt, &len);
205                 if (len != 0)
206                         hlen = len;
207         }
208         ip = mtod(m, struct ip *);
209         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
210
211         /*
212          * Fill in IP header.
213          */
214         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
215                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
216                 ip->ip_off &= IP_DF;
217 #ifdef RANDOM_IP_ID
218                 ip->ip_id = ip_randomid();
219 #else
220                 ip->ip_id = htons(ip_id++);
221 #endif
222                 ipstat.ips_localout++;
223         } else {
224                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
225         }
226
227         dst = (struct sockaddr_in *)&ro->ro_dst;
228         /*
229          * If there is a cached route,
230          * check that it is to the same destination
231          * and is still up.  If not, free it and try again.
232          * The address family should also be checked in case of sharing the
233          * cache with IPv6.
234          */
235         if (ro->ro_rt &&
236             (!(ro->ro_rt->rt_flags & RTF_UP) ||
237              dst->sin_family != AF_INET ||
238              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
239                 rtfree(ro->ro_rt);
240                 ro->ro_rt = (struct rtentry *)NULL;
241         }
242         if (ro->ro_rt == NULL) {
243                 bzero(dst, sizeof *dst);
244                 dst->sin_family = AF_INET;
245                 dst->sin_len = sizeof *dst;
246                 dst->sin_addr = pkt_dst;
247         }
248         /*
249          * If routing to interface only,
250          * short circuit routing lookup.
251          */
252         if (flags & IP_ROUTETOIF) {
253                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
254                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
255                         ipstat.ips_noroute++;
256                         error = ENETUNREACH;
257                         goto bad;
258                 }
259                 ifp = ia->ia_ifp;
260                 ip->ip_ttl = 1;
261                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
262         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
263                    imo != NULL && imo->imo_multicast_ifp != NULL) {
264                 /*
265                  * Bypass the normal routing lookup for multicast
266                  * packets if the interface is specified.
267                  */
268                 ifp = imo->imo_multicast_ifp;
269                 IFP_TO_IA(ifp, ia);
270                 isbroadcast = 0;        /* fool gcc */
271         } else {
272                 /*
273                  * If this is the case, we probably don't want to allocate
274                  * a protocol-cloned route since we didn't get one from the
275                  * ULP.  This lets TCP do its thing, while not burdening
276                  * forwarding or ICMP with the overhead of cloning a route.
277                  * Of course, we still want to do any cloning requested by
278                  * the link layer, as this is probably required in all cases
279                  * for correct operation (as it is for ARP).
280                  */
281                 if (ro->ro_rt == NULL)
282                         rtalloc_ign(ro, RTF_PRCLONING);
283                 if (ro->ro_rt == NULL) {
284                         ipstat.ips_noroute++;
285                         error = EHOSTUNREACH;
286                         goto bad;
287                 }
288                 ia = ifatoia(ro->ro_rt->rt_ifa);
289                 ifp = ro->ro_rt->rt_ifp;
290                 ro->ro_rt->rt_use++;
291                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
292                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
293                 if (ro->ro_rt->rt_flags & RTF_HOST)
294                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
295                 else
296                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
297         }
298         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
299                 struct in_multi *inm;
300
301                 m->m_flags |= M_MCAST;
302                 /*
303                  * IP destination address is multicast.  Make sure "dst"
304                  * still points to the address in "ro".  (It may have been
305                  * changed to point to a gateway address, above.)
306                  */
307                 dst = (struct sockaddr_in *)&ro->ro_dst;
308                 /*
309                  * See if the caller provided any multicast options
310                  */
311                 if (imo != NULL) {
312                         ip->ip_ttl = imo->imo_multicast_ttl;
313                         if (imo->imo_multicast_vif != -1)
314                                 ip->ip_src.s_addr =
315                                     ip_mcast_src ?
316                                     ip_mcast_src(imo->imo_multicast_vif) :
317                                     INADDR_ANY;
318                 } else
319                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
320                 /*
321                  * Confirm that the outgoing interface supports multicast.
322                  */
323                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
324                         if (!(ifp->if_flags & IFF_MULTICAST)) {
325                                 ipstat.ips_noroute++;
326                                 error = ENETUNREACH;
327                                 goto bad;
328                         }
329                 }
330                 /*
331                  * If source address not specified yet, use address
332                  * of outgoing interface.
333                  */
334                 if (ip->ip_src.s_addr == INADDR_ANY) {
335                         /* Interface may have no addresses. */
336                         if (ia != NULL)
337                                 ip->ip_src = IA_SIN(ia)->sin_addr;
338                 }
339
340                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
341                 if (inm != NULL &&
342                    (imo == NULL || imo->imo_multicast_loop)) {
343                         /*
344                          * If we belong to the destination multicast group
345                          * on the outgoing interface, and the caller did not
346                          * forbid loopback, loop back a copy.
347                          */
348                         ip_mloopback(ifp, m, dst, hlen);
349                 }
350                 else {
351                         /*
352                          * If we are acting as a multicast router, perform
353                          * multicast forwarding as if the packet had just
354                          * arrived on the interface to which we are about
355                          * to send.  The multicast forwarding function
356                          * recursively calls this function, using the
357                          * IP_FORWARDING flag to prevent infinite recursion.
358                          *
359                          * Multicasts that are looped back by ip_mloopback(),
360                          * above, will be forwarded by the ip_input() routine,
361                          * if necessary.
362                          */
363                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
364                                 /*
365                                  * If rsvp daemon is not running, do not
366                                  * set ip_moptions. This ensures that the packet
367                                  * is multicast and not just sent down one link
368                                  * as prescribed by rsvpd.
369                                  */
370                                 if (!rsvp_on)
371                                         imo = NULL;
372                                 if (ip_mforward &&
373                                     ip_mforward(ip, ifp, m, imo) != 0) {
374                                         m_freem(m);
375                                         goto done;
376                                 }
377                         }
378                 }
379
380                 /*
381                  * Multicasts with a time-to-live of zero may be looped-
382                  * back, above, but must not be transmitted on a network.
383                  * Also, multicasts addressed to the loopback interface
384                  * are not sent -- the above call to ip_mloopback() will
385                  * loop back a copy if this host actually belongs to the
386                  * destination group on the loopback interface.
387                  */
388                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
389                         m_freem(m);
390                         goto done;
391                 }
392
393                 goto sendit;
394         }
395 #ifndef notdef
396         /*
397          * If the source address is not specified yet, use the address
398          * of the outoing interface. In case, keep note we did that, so
399          * if the the firewall changes the next-hop causing the output
400          * interface to change, we can fix that.
401          */
402         if (ip->ip_src.s_addr == INADDR_ANY) {
403                 /* Interface may have no addresses. */
404                 if (ia != NULL) {
405                         ip->ip_src = IA_SIN(ia)->sin_addr;
406                         src_was_INADDR_ANY = 1;
407                 }
408         }
409 #endif /* notdef */
410         /*
411          * Verify that we have any chance at all of being able to queue
412          *      the packet or packet fragments
413          */
414         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
415                 ifp->if_snd.ifq_maxlen) {
416                         error = ENOBUFS;
417                         ipstat.ips_odropped++;
418                         goto bad;
419         }
420
421         /*
422          * Look for broadcast address and
423          * verify user is allowed to send
424          * such a packet.
425          */
426         if (isbroadcast) {
427                 if (!(ifp->if_flags & IFF_BROADCAST)) {
428                         error = EADDRNOTAVAIL;
429                         goto bad;
430                 }
431                 if (!(flags & IP_ALLOWBROADCAST)) {
432                         error = EACCES;
433                         goto bad;
434                 }
435                 /* don't allow broadcast messages to be fragmented */
436                 if (ip->ip_len > ifp->if_mtu) {
437                         error = EMSGSIZE;
438                         goto bad;
439                 }
440                 m->m_flags |= M_BCAST;
441         } else {
442                 m->m_flags &= ~M_BCAST;
443         }
444
445 sendit:
446 #ifdef IPSEC
447         /* get SP for this packet */
448         if (so == NULL)
449                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
450         else
451                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
452
453         if (sp == NULL) {
454                 ipsecstat.out_inval++;
455                 goto bad;
456         }
457
458         error = 0;
459
460         /* check policy */
461         switch (sp->policy) {
462         case IPSEC_POLICY_DISCARD:
463                 /*
464                  * This packet is just discarded.
465                  */
466                 ipsecstat.out_polvio++;
467                 goto bad;
468
469         case IPSEC_POLICY_BYPASS:
470         case IPSEC_POLICY_NONE:
471                 /* no need to do IPsec. */
472                 goto skip_ipsec;
473
474         case IPSEC_POLICY_IPSEC:
475                 if (sp->req == NULL) {
476                         /* acquire a policy */
477                         error = key_spdacquire(sp);
478                         goto bad;
479                 }
480                 break;
481
482         case IPSEC_POLICY_ENTRUST:
483         default:
484                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
485         }
486     {
487         struct ipsec_output_state state;
488         bzero(&state, sizeof state);
489         state.m = m;
490         if (flags & IP_ROUTETOIF) {
491                 state.ro = &iproute;
492                 bzero(&iproute, sizeof iproute);
493         } else
494                 state.ro = ro;
495         state.dst = (struct sockaddr *)dst;
496
497         ip->ip_sum = 0;
498
499         /*
500          * XXX
501          * delayed checksums are not currently compatible with IPsec
502          */
503         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
504                 in_delayed_cksum(m);
505                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
506         }
507
508         ip->ip_len = htons(ip->ip_len);
509         ip->ip_off = htons(ip->ip_off);
510
511         error = ipsec4_output(&state, sp, flags);
512
513         m = state.m;
514         if (flags & IP_ROUTETOIF) {
515                 /*
516                  * if we have tunnel mode SA, we may need to ignore
517                  * IP_ROUTETOIF.
518                  */
519                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
520                         flags &= ~IP_ROUTETOIF;
521                         ro = state.ro;
522                 }
523         } else
524                 ro = state.ro;
525         dst = (struct sockaddr_in *)state.dst;
526         if (error) {
527                 /* mbuf is already reclaimed in ipsec4_output. */
528                 m0 = NULL;
529                 switch (error) {
530                 case EHOSTUNREACH:
531                 case ENETUNREACH:
532                 case EMSGSIZE:
533                 case ENOBUFS:
534                 case ENOMEM:
535                         break;
536                 default:
537                         printf("ip4_output (ipsec): error code %d\n", error);
538                         /*fall through*/
539                 case ENOENT:
540                         /* don't show these error codes to the user */
541                         error = 0;
542                         break;
543                 }
544                 goto bad;
545         }
546     }
547
548         /* be sure to update variables that are affected by ipsec4_output() */
549         ip = mtod(m, struct ip *);
550 #ifdef _IP_VHL
551         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
552 #else
553         hlen = ip->ip_hl << 2;
554 #endif
555         if (ro->ro_rt == NULL) {
556                 if (!(flags & IP_ROUTETOIF)) {
557                         printf("ip_output: "
558                                 "can't update route after IPsec processing\n");
559                         error = EHOSTUNREACH;   /*XXX*/
560                         goto bad;
561                 }
562         } else {
563                 ia = ifatoia(ro->ro_rt->rt_ifa);
564                 ifp = ro->ro_rt->rt_ifp;
565         }
566
567         /* make it flipped, again. */
568         ip->ip_len = ntohs(ip->ip_len);
569         ip->ip_off = ntohs(ip->ip_off);
570 skip_ipsec:
571 #endif /*IPSEC*/
572 #ifdef FAST_IPSEC
573         /*
574          * Check the security policy (SP) for the packet and, if
575          * required, do IPsec-related processing.  There are two
576          * cases here; the first time a packet is sent through
577          * it will be untagged and handled by ipsec4_checkpolicy.
578          * If the packet is resubmitted to ip_output (e.g. after
579          * AH, ESP, etc. processing), there will be a tag to bypass
580          * the lookup and related policy checking.
581          */
582         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
583         s = splnet();
584         if (mtag != NULL) {
585                 tdbi = (struct tdb_ident *)(mtag + 1);
586                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
587                 if (sp == NULL)
588                         error = -EINVAL;        /* force silent drop */
589                 m_tag_delete(m, mtag);
590         } else {
591                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
592                                         &error, inp);
593         }
594         /*
595          * There are four return cases:
596          *    sp != NULL                    apply IPsec policy
597          *    sp == NULL, error == 0        no IPsec handling needed
598          *    sp == NULL, error == -EINVAL  discard packet w/o error
599          *    sp == NULL, error != 0        discard packet, report error
600          */
601         if (sp != NULL) {
602                 /* Loop detection, check if ipsec processing already done */
603                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
604                 for (mtag = m_tag_first(m); mtag != NULL;
605                      mtag = m_tag_next(m, mtag)) {
606                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
607                                 continue;
608                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
609                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
610                                 continue;
611                         /*
612                          * Check if policy has an SA associated with it.
613                          * This can happen when an SP has yet to acquire
614                          * an SA; e.g. on first reference.  If it occurs,
615                          * then we let ipsec4_process_packet do its thing.
616                          */
617                         if (sp->req->sav == NULL)
618                                 break;
619                         tdbi = (struct tdb_ident *)(mtag + 1);
620                         if (tdbi->spi == sp->req->sav->spi &&
621                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
622                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
623                                  sizeof(union sockaddr_union)) == 0) {
624                                 /*
625                                  * No IPsec processing is needed, free
626                                  * reference to SP.
627                                  *
628                                  * NB: null pointer to avoid free at
629                                  *     done: below.
630                                  */
631                                 KEY_FREESP(&sp), sp = NULL;
632                                 splx(s);
633                                 goto spd_done;
634                         }
635                 }
636
637                 /*
638                  * Do delayed checksums now because we send before
639                  * this is done in the normal processing path.
640                  */
641                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
642                         in_delayed_cksum(m);
643                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
644                 }
645
646                 ip->ip_len = htons(ip->ip_len);
647                 ip->ip_off = htons(ip->ip_off);
648
649                 /* NB: callee frees mbuf */
650                 error = ipsec4_process_packet(m, sp->req, flags, 0);
651                 /*
652                  * Preserve KAME behaviour: ENOENT can be returned
653                  * when an SA acquire is in progress.  Don't propagate
654                  * this to user-level; it confuses applications.
655                  *
656                  * XXX this will go away when the SADB is redone.
657                  */
658                 if (error == ENOENT)
659                         error = 0;
660                 splx(s);
661                 goto done;
662         } else {
663                 splx(s);
664
665                 if (error != 0) {
666                         /*
667                          * Hack: -EINVAL is used to signal that a packet
668                          * should be silently discarded.  This is typically
669                          * because we asked key management for an SA and
670                          * it was delayed (e.g. kicked up to IKE).
671                          */
672                         if (error == -EINVAL)
673                                 error = 0;
674                         goto bad;
675                 } else {
676                         /* No IPsec processing for this packet. */
677                 }
678 #ifdef notyet
679                 /*
680                  * If deferred crypto processing is needed, check that
681                  * the interface supports it.
682                  */
683                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
684                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
685                         /* notify IPsec to do its own crypto */
686                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
687                         error = EHOSTUNREACH;
688                         goto bad;
689                 }
690 #endif
691         }
692 spd_done:
693 #endif /* FAST_IPSEC */
694         /*
695          * IpHack's section.
696          * - Xlate: translate packet's addr/port (NAT).
697          * - Firewall: deny/allow/etc.
698          * - Wrap: fake packet's addr/port <unimpl.>
699          * - Encapsulate: put it in another IP and send out. <unimp.>
700          */
701
702         /*
703          * Run through list of hooks for output packets.
704          */
705         if (pfil_has_hooks(&inet_pfil_hook)) {
706                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
707                 if (error != 0 || m == NULL)
708                         goto done;
709                 ip = mtod(m, struct ip *);
710         }
711
712         /*
713          * Check with the firewall...
714          * but not if we are already being fwd'd from a firewall.
715          */
716         if (fw_enable && IPFW_LOADED && !args.next_hop) {
717                 struct sockaddr_in *old = dst;
718
719                 args.m = m;
720                 args.next_hop = dst;
721                 args.oif = ifp;
722                 off = ip_fw_chk_ptr(&args);
723                 m = args.m;
724                 dst = args.next_hop;
725
726                 /*
727                  * On return we must do the following:
728                  * m == NULL    -> drop the pkt (old interface, deprecated)
729                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
730                  * 1<=off<= 0xffff              -> DIVERT
731                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
732                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
733                  * dst != old                   -> IPFIREWALL_FORWARD
734                  * off==0, dst==old             -> accept
735                  * If some of the above modules are not compiled in, then
736                  * we should't have to check the corresponding condition
737                  * (because the ipfw control socket should not accept
738                  * unsupported rules), but better play safe and drop
739                  * packets in case of doubt.
740                  */
741                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
742                         if (m)
743                                 m_freem(m);
744                         error = EACCES;
745                         goto done;
746                 }
747                 ip = mtod(m, struct ip *);
748                 if (off == 0 && dst == old)             /* common case */
749                         goto pass;
750                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG)) {
751                         /*
752                          * pass the pkt to dummynet. Need to include
753                          * pipe number, m, ifp, ro, dst because these are
754                          * not recomputed in the next pass.
755                          * All other parameters have been already used and
756                          * so they are not needed anymore.
757                          * XXX note: if the ifp or ro entry are deleted
758                          * while a pkt is in dummynet, we are in trouble!
759                          */
760                         args.ro = ro;
761                         args.dst = dst;
762                         args.flags = flags;
763
764                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
765                                 &args);
766                         goto done;
767                 }
768 #ifdef IPDIVERT
769                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
770                         struct mbuf *clone = NULL;
771
772                         /* Clone packet if we're doing a 'tee' */
773                         if ((off & IP_FW_PORT_TEE_FLAG))
774                                 clone = m_dup(m, MB_DONTWAIT);
775
776                         /*
777                          * XXX
778                          * delayed checksums are not currently compatible
779                          * with divert sockets.
780                          */
781                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
782                                 in_delayed_cksum(m);
783                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
784                         }
785
786                         /* Restore packet header fields to original values */
787                         ip->ip_len = htons(ip->ip_len);
788                         ip->ip_off = htons(ip->ip_off);
789
790                         /* Deliver packet to divert input routine */
791                         divert_packet(m, 0, off & 0xffff, args.divert_rule);
792
793                         /* If 'tee', continue with original packet */
794                         if (clone != NULL) {
795                                 m = clone;
796                                 ip = mtod(m, struct ip *);
797                                 goto pass;
798                         }
799                         goto done;
800                 }
801 #endif
802
803                 /* IPFIREWALL_FORWARD */
804                 /*
805                  * Check dst to make sure it is directly reachable on the
806                  * interface we previously thought it was.
807                  * If it isn't (which may be likely in some situations) we have
808                  * to re-route it (ie, find a route for the next-hop and the
809                  * associated interface) and set them here. This is nested
810                  * forwarding which in most cases is undesirable, except where
811                  * such control is nigh impossible. So we do it here.
812                  * And I'm babbling.
813                  */
814                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
815 #if 0
816                         /*
817                          * XXX To improve readability, this block should be
818                          * changed into a function call as below:
819                          */
820                         error = ip_ipforward(&m, &dst, &ifp);
821                         if (error)
822                                 goto bad;
823                         if (m == NULL) /* ip_input consumed the mbuf */
824                                 goto done;
825 #else
826                         struct in_ifaddr *ia;
827
828                         /*
829                          * XXX sro_fwd below is static, and a pointer
830                          * to it gets passed to routines downstream.
831                          * This could have surprisingly bad results in
832                          * practice, because its content is overwritten
833                          * by subsequent packets.
834                          */
835                         /* There must be a better way to do this next line... */
836                         static struct route sro_fwd;
837                         struct route *ro_fwd = &sro_fwd;
838
839 #if 0
840                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
841                             dst->sin_addr, "\n");
842 #endif
843
844                         /*
845                          * We need to figure out if we have been forwarded
846                          * to a local socket. If so, then we should somehow
847                          * "loop back" to ip_input, and get directed to the
848                          * PCB as if we had received this packet. This is
849                          * because it may be dificult to identify the packets
850                          * you want to forward until they are being output
851                          * and have selected an interface. (e.g. locally
852                          * initiated packets) If we used the loopback inteface,
853                          * we would not be able to control what happens
854                          * as the packet runs through ip_input() as
855                          * it is done through a ISR.
856                          */
857                         LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
858                                      ia_hash) {
859                                 /*
860                                  * If the addr to forward to is one
861                                  * of ours, we pretend to
862                                  * be the destination for this packet.
863                                  */
864                                 if (IA_SIN(ia)->sin_addr.s_addr ==
865                                                  dst->sin_addr.s_addr)
866                                         break;
867                         }
868                         if (ia != NULL) {    /* tell ip_input "dont filter" */
869                                 struct m_hdr tag;
870
871                                 tag.mh_type = MT_TAG;
872                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
873                                 tag.mh_data = (caddr_t)args.next_hop;
874                                 tag.mh_next = m;
875
876                                 if (m->m_pkthdr.rcvif == NULL)
877                                         m->m_pkthdr.rcvif = ifunit("lo0");
878                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
879                                         m->m_pkthdr.csum_flags |=
880                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
881                                         m0->m_pkthdr.csum_data = 0xffff;
882                                 }
883                                 m->m_pkthdr.csum_flags |=
884                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
885                                 ip->ip_len = htons(ip->ip_len);
886                                 ip->ip_off = htons(ip->ip_off);
887                                 ip_input((struct mbuf *)&tag);
888                                 goto done;
889                         }
890                         /* Some of the logic for this was nicked from above.
891                          *
892                          * This rewrites the cached route in a local PCB.
893                          * Is this what we want to do?
894                          */
895                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
896                         ro_fwd->ro_rt = NULL;
897
898                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
899                         if (ro_fwd->ro_rt == NULL) {
900                                 ipstat.ips_noroute++;
901                                 error = EHOSTUNREACH;
902                                 goto bad;
903                         }
904
905                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
906                         ifp = ro_fwd->ro_rt->rt_ifp;
907                         ro_fwd->ro_rt->rt_use++;
908                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
909                                 dst = (struct sockaddr_in *)
910                                     ro_fwd->ro_rt->rt_gateway;
911                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
912                                 isbroadcast =
913                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
914                         else
915                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
916                         if (ro->ro_rt != NULL)
917                                 rtfree(ro->ro_rt);
918                         ro->ro_rt = ro_fwd->ro_rt;
919                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
920
921 #endif  /* ... block to be put into a function */
922                         /*
923                          * If we added a default src ip earlier,
924                          * which would have been gotten from the-then
925                          * interface, do it again, from the new one.
926                          */
927                         if (src_was_INADDR_ANY)
928                                 ip->ip_src = IA_SIN(ia)->sin_addr;
929                         goto pass ;
930                 }
931
932                 /*
933                  * if we get here, none of the above matches, and
934                  * we have to drop the pkt
935                  */
936                 m_freem(m);
937                 error = EACCES; /* not sure this is the right error msg */
938                 goto done;
939         }
940
941 pass:
942         /* 127/8 must not appear on wire - RFC1122. */
943         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
944             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
945                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
946                         ipstat.ips_badaddr++;
947                         error = EADDRNOTAVAIL;
948                         goto bad;
949                 }
950         }
951
952         m->m_pkthdr.csum_flags |= CSUM_IP;
953         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
954         if (sw_csum & CSUM_DELAY_DATA) {
955                 in_delayed_cksum(m);
956                 sw_csum &= ~CSUM_DELAY_DATA;
957         }
958         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
959
960         /*
961          * If small enough for interface, or the interface will take
962          * care of the fragmentation for us, can just send directly.
963          */
964         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
965             !(ip->ip_off & IP_DF))) {
966                 ip->ip_len = htons(ip->ip_len);
967                 ip->ip_off = htons(ip->ip_off);
968                 ip->ip_sum = 0;
969                 if (sw_csum & CSUM_DELAY_IP) {
970                         if (ip->ip_vhl == IP_VHL_BORING) {
971                                 ip->ip_sum = in_cksum_hdr(ip);
972                         } else {
973                                 ip->ip_sum = in_cksum(m, hlen);
974                         }
975                 }
976
977                 /* Record statistics for this interface address. */
978                 if (!(flags & IP_FORWARDING) && ia) {
979                         ia->ia_ifa.if_opackets++;
980                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
981                 }
982
983 #ifdef IPSEC
984                 /* clean ipsec history once it goes out of the node */
985                 ipsec_delaux(m);
986 #endif
987
988 #ifdef MBUF_STRESS_TEST
989                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
990                         struct mbuf *m1, *m2;
991                         int length, tmp;
992
993                         tmp = length = m->m_pkthdr.len;
994
995                         while ((length -= mbuf_frag_size) >= 1) {
996                                 m1 = m_split(m, length, MB_DONTWAIT);
997                                 if (m1 == NULL)
998                                         break;
999                                 m1->m_flags &= ~M_PKTHDR;
1000                                 m2 = m;
1001                                 while (m2->m_next != NULL)
1002                                         m2 = m2->m_next;
1003                                 m2->m_next = m1;
1004                         }
1005                         m->m_pkthdr.len = tmp;
1006                 }
1007 #endif
1008                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1009                                           ro->ro_rt);
1010                 goto done;
1011         }
1012
1013         if (ip->ip_off & IP_DF) {
1014                 error = EMSGSIZE;
1015                 /*
1016                  * This case can happen if the user changed the MTU
1017                  * of an interface after enabling IP on it.  Because
1018                  * most netifs don't keep track of routes pointing to
1019                  * them, there is no way for one to update all its
1020                  * routes when the MTU is changed.
1021                  */
1022                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1023                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1024                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1025                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1026                 }
1027                 ipstat.ips_cantfrag++;
1028                 goto bad;
1029         }
1030
1031         /*
1032          * Too large for interface; fragment if possible. If successful,
1033          * on return, m will point to a list of packets to be sent.
1034          */
1035         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1036         if (error)
1037                 goto bad;
1038         for (; m; m = m0) {
1039                 m0 = m->m_nextpkt;
1040                 m->m_nextpkt = NULL;
1041 #ifdef IPSEC
1042                 /* clean ipsec history once it goes out of the node */
1043                 ipsec_delaux(m);
1044 #endif
1045                 if (error == 0) {
1046                         /* Record statistics for this interface address. */
1047                         if (ia != NULL) {
1048                                 ia->ia_ifa.if_opackets++;
1049                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1050                         }
1051                         error = (*ifp->if_output)(ifp, m,
1052                                                   (struct sockaddr *)dst,
1053                                                   ro->ro_rt);
1054                 } else
1055                         m_freem(m);
1056         }
1057
1058         if (error == 0)
1059                 ipstat.ips_fragmented++;
1060
1061 done:
1062         if (ro == &iproute && ro->ro_rt != NULL) {
1063                 RTFREE(ro->ro_rt);
1064                 ro->ro_rt = NULL;
1065         }
1066 #ifdef IPSEC
1067         if (sp != NULL) {
1068                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1069                         printf("DP ip_output call free SP:%p\n", sp));
1070                 key_freesp(sp);
1071         }
1072 #endif
1073 #ifdef FAST_IPSEC
1074         if (sp != NULL)
1075                 KEY_FREESP(&sp);
1076 #endif
1077         return (error);
1078 bad:
1079         m_freem(m);
1080         goto done;
1081 }
1082
1083 /*
1084  * Create a chain of fragments which fit the given mtu. m_frag points to the
1085  * mbuf to be fragmented; on return it points to the chain with the fragments.
1086  * Return 0 if no error. If error, m_frag may contain a partially built
1087  * chain of fragments that should be freed by the caller.
1088  *
1089  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1090  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1091  */
1092 int
1093 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1094             u_long if_hwassist_flags, int sw_csum)
1095 {
1096         int error = 0;
1097         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1098         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1099         int off;
1100         struct mbuf *m0 = *m_frag;      /* the original packet          */
1101         int firstlen;
1102         struct mbuf **mnext;
1103         int nfrags;
1104
1105         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1106                 ipstat.ips_cantfrag++;
1107                 return EMSGSIZE;
1108         }
1109
1110         /*
1111          * Must be able to put at least 8 bytes per fragment.
1112          */
1113         if (len < 8)
1114                 return EMSGSIZE;
1115
1116         /*
1117          * If the interface will not calculate checksums on
1118          * fragmented packets, then do it here.
1119          */
1120         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1121             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1122                 in_delayed_cksum(m0);
1123                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1124         }
1125
1126         if (len > PAGE_SIZE) {
1127                 /*
1128                  * Fragment large datagrams such that each segment
1129                  * contains a multiple of PAGE_SIZE amount of data,
1130                  * plus headers. This enables a receiver to perform
1131                  * page-flipping zero-copy optimizations.
1132                  *
1133                  * XXX When does this help given that sender and receiver
1134                  * could have different page sizes, and also mtu could
1135                  * be less than the receiver's page size ?
1136                  */
1137                 int newlen;
1138                 struct mbuf *m;
1139
1140                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1141                         off += m->m_len;
1142
1143                 /*
1144                  * firstlen (off - hlen) must be aligned on an
1145                  * 8-byte boundary
1146                  */
1147                 if (off < hlen)
1148                         goto smart_frag_failure;
1149                 off = ((off - hlen) & ~7) + hlen;
1150                 newlen = (~PAGE_MASK) & mtu;
1151                 if ((newlen + sizeof(struct ip)) > mtu) {
1152                         /* we failed, go back the default */
1153 smart_frag_failure:
1154                         newlen = len;
1155                         off = hlen + len;
1156                 }
1157                 len = newlen;
1158
1159         } else {
1160                 off = hlen + len;
1161         }
1162
1163         firstlen = off - hlen;
1164         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1165
1166         /*
1167          * Loop through length of segment after first fragment,
1168          * make new header and copy data of each part and link onto chain.
1169          * Here, m0 is the original packet, m is the fragment being created.
1170          * The fragments are linked off the m_nextpkt of the original
1171          * packet, which after processing serves as the first fragment.
1172          */
1173         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1174                 struct ip *mhip;        /* ip header on the fragment */
1175                 struct mbuf *m;
1176                 int mhlen = sizeof(struct ip);
1177
1178                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1179                 if (m == NULL) {
1180                         error = ENOBUFS;
1181                         ipstat.ips_odropped++;
1182                         goto done;
1183                 }
1184                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1185                 /*
1186                  * In the first mbuf, leave room for the link header, then
1187                  * copy the original IP header including options. The payload
1188                  * goes into an additional mbuf chain returned by m_copy().
1189                  */
1190                 m->m_data += max_linkhdr;
1191                 mhip = mtod(m, struct ip *);
1192                 *mhip = *ip;
1193                 if (hlen > sizeof(struct ip)) {
1194                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1195                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1196                 }
1197                 m->m_len = mhlen;
1198                 /* XXX do we need to add ip->ip_off below ? */
1199                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1200                 if (off + len >= ip->ip_len) {  /* last fragment */
1201                         len = ip->ip_len - off;
1202                         m->m_flags |= M_LASTFRAG;
1203                 } else
1204                         mhip->ip_off |= IP_MF;
1205                 mhip->ip_len = htons((u_short)(len + mhlen));
1206                 m->m_next = m_copy(m0, off, len);
1207                 if (m->m_next == NULL) {                /* copy failed */
1208                         m_free(m);
1209                         error = ENOBUFS;        /* ??? */
1210                         ipstat.ips_odropped++;
1211                         goto done;
1212                 }
1213                 m->m_pkthdr.len = mhlen + len;
1214                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1215                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1216                 mhip->ip_off = htons(mhip->ip_off);
1217                 mhip->ip_sum = 0;
1218                 if (sw_csum & CSUM_DELAY_IP)
1219                         mhip->ip_sum = in_cksum(m, mhlen);
1220                 *mnext = m;
1221                 mnext = &m->m_nextpkt;
1222         }
1223         ipstat.ips_ofragments += nfrags;
1224
1225         /* set first marker for fragment chain */
1226         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1227         m0->m_pkthdr.csum_data = nfrags;
1228
1229         /*
1230          * Update first fragment by trimming what's been copied out
1231          * and updating header.
1232          */
1233         m_adj(m0, hlen + firstlen - ip->ip_len);
1234         m0->m_pkthdr.len = hlen + firstlen;
1235         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1236         ip->ip_off |= IP_MF;
1237         ip->ip_off = htons(ip->ip_off);
1238         ip->ip_sum = 0;
1239         if (sw_csum & CSUM_DELAY_IP)
1240                 ip->ip_sum = in_cksum(m0, hlen);
1241
1242 done:
1243         *m_frag = m0;
1244         return error;
1245 }
1246
1247 void
1248 in_delayed_cksum(struct mbuf *m)
1249 {
1250         struct ip *ip;
1251         u_short csum, offset;
1252
1253         ip = mtod(m, struct ip *);
1254         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1255         csum = in_cksum_skip(m, ip->ip_len, offset);
1256         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1257                 csum = 0xffff;
1258         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1259
1260         if (offset + sizeof(u_short) > m->m_len) {
1261                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1262                     m->m_len, offset, ip->ip_p);
1263                 /*
1264                  * XXX
1265                  * this shouldn't happen, but if it does, the
1266                  * correct behavior may be to insert the checksum
1267                  * in the existing chain instead of rearranging it.
1268                  */
1269                 m = m_pullup(m, offset + sizeof(u_short));
1270         }
1271         *(u_short *)(m->m_data + offset) = csum;
1272 }
1273
1274 /*
1275  * Insert IP options into preformed packet.
1276  * Adjust IP destination as required for IP source routing,
1277  * as indicated by a non-zero in_addr at the start of the options.
1278  *
1279  * XXX This routine assumes that the packet has no options in place.
1280  */
1281 static struct mbuf *
1282 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1283 {
1284         struct ipoption *p = mtod(opt, struct ipoption *);
1285         struct mbuf *n;
1286         struct ip *ip = mtod(m, struct ip *);
1287         unsigned optlen;
1288
1289         optlen = opt->m_len - sizeof p->ipopt_dst;
1290         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1291                 *phlen = 0;
1292                 return (m);             /* XXX should fail */
1293         }
1294         if (p->ipopt_dst.s_addr)
1295                 ip->ip_dst = p->ipopt_dst;
1296         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1297                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1298                 if (n == NULL) {
1299                         *phlen = 0;
1300                         return (m);
1301                 }
1302                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1303                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1304                 m->m_len -= sizeof(struct ip);
1305                 m->m_data += sizeof(struct ip);
1306                 n->m_next = m;
1307                 m = n;
1308                 m->m_len = optlen + sizeof(struct ip);
1309                 m->m_data += max_linkhdr;
1310                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1311         } else {
1312                 m->m_data -= optlen;
1313                 m->m_len += optlen;
1314                 m->m_pkthdr.len += optlen;
1315                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1316         }
1317         ip = mtod(m, struct ip *);
1318         bcopy(p->ipopt_list, ip + 1, optlen);
1319         *phlen = sizeof(struct ip) + optlen;
1320         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1321         ip->ip_len += optlen;
1322         return (m);
1323 }
1324
1325 /*
1326  * Copy options from ip to jp,
1327  * omitting those not copied during fragmentation.
1328  */
1329 int
1330 ip_optcopy(struct ip *ip, struct ip *jp)
1331 {
1332         u_char *cp, *dp;
1333         int opt, optlen, cnt;
1334
1335         cp = (u_char *)(ip + 1);
1336         dp = (u_char *)(jp + 1);
1337         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1338         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1339                 opt = cp[0];
1340                 if (opt == IPOPT_EOL)
1341                         break;
1342                 if (opt == IPOPT_NOP) {
1343                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1344                         *dp++ = IPOPT_NOP;
1345                         optlen = 1;
1346                         continue;
1347                 }
1348
1349                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1350                     ("ip_optcopy: malformed ipv4 option"));
1351                 optlen = cp[IPOPT_OLEN];
1352                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1353                     ("ip_optcopy: malformed ipv4 option"));
1354
1355                 /* bogus lengths should have been caught by ip_dooptions */
1356                 if (optlen > cnt)
1357                         optlen = cnt;
1358                 if (IPOPT_COPIED(opt)) {
1359                         bcopy(cp, dp, optlen);
1360                         dp += optlen;
1361                 }
1362         }
1363         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1364                 *dp++ = IPOPT_EOL;
1365         return (optlen);
1366 }
1367
1368 /*
1369  * IP socket option processing.
1370  */
1371 int
1372 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1373 {
1374         struct  inpcb *inp = sotoinpcb(so);
1375         int     error, optval;
1376
1377         error = optval = 0;
1378         if (sopt->sopt_level != IPPROTO_IP) {
1379                 return (EINVAL);
1380         }
1381
1382         switch (sopt->sopt_dir) {
1383         case SOPT_SET:
1384                 switch (sopt->sopt_name) {
1385                 case IP_OPTIONS:
1386 #ifdef notyet
1387                 case IP_RETOPTS:
1388 #endif
1389                 {
1390                         struct mbuf *m;
1391                         if (sopt->sopt_valsize > MLEN) {
1392                                 error = EMSGSIZE;
1393                                 break;
1394                         }
1395                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1396                         if (m == NULL) {
1397                                 error = ENOBUFS;
1398                                 break;
1399                         }
1400                         m->m_len = sopt->sopt_valsize;
1401                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1402                                             m->m_len);
1403
1404                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1405                                            m));
1406                 }
1407
1408                 case IP_TOS:
1409                 case IP_TTL:
1410                 case IP_RECVOPTS:
1411                 case IP_RECVRETOPTS:
1412                 case IP_RECVDSTADDR:
1413                 case IP_RECVIF:
1414                 case IP_FAITH:
1415                         error = sooptcopyin(sopt, &optval, sizeof optval,
1416                                             sizeof optval);
1417                         if (error)
1418                                 break;
1419
1420                         switch (sopt->sopt_name) {
1421                         case IP_TOS:
1422                                 inp->inp_ip_tos = optval;
1423                                 break;
1424
1425                         case IP_TTL:
1426                                 inp->inp_ip_ttl = optval;
1427                                 break;
1428 #define OPTSET(bit) \
1429         if (optval) \
1430                 inp->inp_flags |= bit; \
1431         else \
1432                 inp->inp_flags &= ~bit;
1433
1434                         case IP_RECVOPTS:
1435                                 OPTSET(INP_RECVOPTS);
1436                                 break;
1437
1438                         case IP_RECVRETOPTS:
1439                                 OPTSET(INP_RECVRETOPTS);
1440                                 break;
1441
1442                         case IP_RECVDSTADDR:
1443                                 OPTSET(INP_RECVDSTADDR);
1444                                 break;
1445
1446                         case IP_RECVIF:
1447                                 OPTSET(INP_RECVIF);
1448                                 break;
1449
1450                         case IP_FAITH:
1451                                 OPTSET(INP_FAITH);
1452                                 break;
1453                         }
1454                         break;
1455 #undef OPTSET
1456
1457                 case IP_MULTICAST_IF:
1458                 case IP_MULTICAST_VIF:
1459                 case IP_MULTICAST_TTL:
1460                 case IP_MULTICAST_LOOP:
1461                 case IP_ADD_MEMBERSHIP:
1462                 case IP_DROP_MEMBERSHIP:
1463                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1464                         break;
1465
1466                 case IP_PORTRANGE:
1467                         error = sooptcopyin(sopt, &optval, sizeof optval,
1468                                             sizeof optval);
1469                         if (error)
1470                                 break;
1471
1472                         switch (optval) {
1473                         case IP_PORTRANGE_DEFAULT:
1474                                 inp->inp_flags &= ~(INP_LOWPORT);
1475                                 inp->inp_flags &= ~(INP_HIGHPORT);
1476                                 break;
1477
1478                         case IP_PORTRANGE_HIGH:
1479                                 inp->inp_flags &= ~(INP_LOWPORT);
1480                                 inp->inp_flags |= INP_HIGHPORT;
1481                                 break;
1482
1483                         case IP_PORTRANGE_LOW:
1484                                 inp->inp_flags &= ~(INP_HIGHPORT);
1485                                 inp->inp_flags |= INP_LOWPORT;
1486                                 break;
1487
1488                         default:
1489                                 error = EINVAL;
1490                                 break;
1491                         }
1492                         break;
1493
1494 #if defined(IPSEC) || defined(FAST_IPSEC)
1495                 case IP_IPSEC_POLICY:
1496                 {
1497                         caddr_t req;
1498                         size_t len = 0;
1499                         int priv;
1500                         struct mbuf *m;
1501                         int optname;
1502
1503                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1504                                 break;
1505                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1506                                 break;
1507                         priv = (sopt->sopt_td != NULL &&
1508                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1509                         req = mtod(m, caddr_t);
1510                         len = m->m_len;
1511                         optname = sopt->sopt_name;
1512                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1513                         m_freem(m);
1514                         break;
1515                 }
1516 #endif /*IPSEC*/
1517
1518                 default:
1519                         error = ENOPROTOOPT;
1520                         break;
1521                 }
1522                 break;
1523
1524         case SOPT_GET:
1525                 switch (sopt->sopt_name) {
1526                 case IP_OPTIONS:
1527                 case IP_RETOPTS:
1528                         if (inp->inp_options)
1529                                 error = sooptcopyout(sopt,
1530                                                      mtod(inp->inp_options,
1531                                                           char *),
1532                                                      inp->inp_options->m_len);
1533                         else
1534                                 sopt->sopt_valsize = 0;
1535                         break;
1536
1537                 case IP_TOS:
1538                 case IP_TTL:
1539                 case IP_RECVOPTS:
1540                 case IP_RECVRETOPTS:
1541                 case IP_RECVDSTADDR:
1542                 case IP_RECVIF:
1543                 case IP_PORTRANGE:
1544                 case IP_FAITH:
1545                         switch (sopt->sopt_name) {
1546
1547                         case IP_TOS:
1548                                 optval = inp->inp_ip_tos;
1549                                 break;
1550
1551                         case IP_TTL:
1552                                 optval = inp->inp_ip_ttl;
1553                                 break;
1554
1555 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1556
1557                         case IP_RECVOPTS:
1558                                 optval = OPTBIT(INP_RECVOPTS);
1559                                 break;
1560
1561                         case IP_RECVRETOPTS:
1562                                 optval = OPTBIT(INP_RECVRETOPTS);
1563                                 break;
1564
1565                         case IP_RECVDSTADDR:
1566                                 optval = OPTBIT(INP_RECVDSTADDR);
1567                                 break;
1568
1569                         case IP_RECVIF:
1570                                 optval = OPTBIT(INP_RECVIF);
1571                                 break;
1572
1573                         case IP_PORTRANGE:
1574                                 if (inp->inp_flags & INP_HIGHPORT)
1575                                         optval = IP_PORTRANGE_HIGH;
1576                                 else if (inp->inp_flags & INP_LOWPORT)
1577                                         optval = IP_PORTRANGE_LOW;
1578                                 else
1579                                         optval = 0;
1580                                 break;
1581
1582                         case IP_FAITH:
1583                                 optval = OPTBIT(INP_FAITH);
1584                                 break;
1585                         }
1586                         error = sooptcopyout(sopt, &optval, sizeof optval);
1587                         break;
1588
1589                 case IP_MULTICAST_IF:
1590                 case IP_MULTICAST_VIF:
1591                 case IP_MULTICAST_TTL:
1592                 case IP_MULTICAST_LOOP:
1593                 case IP_ADD_MEMBERSHIP:
1594                 case IP_DROP_MEMBERSHIP:
1595                         error = ip_getmoptions(sopt, inp->inp_moptions);
1596                         break;
1597
1598 #if defined(IPSEC) || defined(FAST_IPSEC)
1599                 case IP_IPSEC_POLICY:
1600                 {
1601                         struct mbuf *m = NULL;
1602                         caddr_t req = NULL;
1603                         size_t len = 0;
1604
1605                         if (m != NULL) {
1606                                 req = mtod(m, caddr_t);
1607                                 len = m->m_len;
1608                         }
1609                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1610                         if (error == 0)
1611                                 error = soopt_mcopyout(sopt, m); /* XXX */
1612                         if (error == 0)
1613                                 m_freem(m);
1614                         break;
1615                 }
1616 #endif /*IPSEC*/
1617
1618                 default:
1619                         error = ENOPROTOOPT;
1620                         break;
1621                 }
1622                 break;
1623         }
1624         return (error);
1625 }
1626
1627 /*
1628  * Set up IP options in pcb for insertion in output packets.
1629  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1630  * with destination address if source routed.
1631  */
1632 static int
1633 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1634 {
1635         int cnt, optlen;
1636         u_char *cp;
1637         u_char opt;
1638
1639         /* turn off any old options */
1640         if (*pcbopt)
1641                 m_free(*pcbopt);
1642         *pcbopt = 0;
1643         if (m == NULL || m->m_len == 0) {
1644                 /*
1645                  * Only turning off any previous options.
1646                  */
1647                 if (m != NULL)
1648                         m_free(m);
1649                 return (0);
1650         }
1651
1652         if (m->m_len % sizeof(int32_t))
1653                 goto bad;
1654         /*
1655          * IP first-hop destination address will be stored before
1656          * actual options; move other options back
1657          * and clear it when none present.
1658          */
1659         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1660                 goto bad;
1661         cnt = m->m_len;
1662         m->m_len += sizeof(struct in_addr);
1663         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1664         ovbcopy(mtod(m, caddr_t), cp, cnt);
1665         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1666
1667         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1668                 opt = cp[IPOPT_OPTVAL];
1669                 if (opt == IPOPT_EOL)
1670                         break;
1671                 if (opt == IPOPT_NOP)
1672                         optlen = 1;
1673                 else {
1674                         if (cnt < IPOPT_OLEN + sizeof *cp)
1675                                 goto bad;
1676                         optlen = cp[IPOPT_OLEN];
1677                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1678                                 goto bad;
1679                 }
1680                 switch (opt) {
1681
1682                 default:
1683                         break;
1684
1685                 case IPOPT_LSRR:
1686                 case IPOPT_SSRR:
1687                         /*
1688                          * user process specifies route as:
1689                          *      ->A->B->C->D
1690                          * D must be our final destination (but we can't
1691                          * check that since we may not have connected yet).
1692                          * A is first hop destination, which doesn't appear in
1693                          * actual IP option, but is stored before the options.
1694                          */
1695                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1696                                 goto bad;
1697                         m->m_len -= sizeof(struct in_addr);
1698                         cnt -= sizeof(struct in_addr);
1699                         optlen -= sizeof(struct in_addr);
1700                         cp[IPOPT_OLEN] = optlen;
1701                         /*
1702                          * Move first hop before start of options.
1703                          */
1704                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1705                               sizeof(struct in_addr));
1706                         /*
1707                          * Then copy rest of options back
1708                          * to close up the deleted entry.
1709                          */
1710                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1711                                 &cp[IPOPT_OFFSET+1],
1712                                 cnt - (IPOPT_MINOFF - 1));
1713                         break;
1714                 }
1715         }
1716         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1717                 goto bad;
1718         *pcbopt = m;
1719         return (0);
1720
1721 bad:
1722         m_free(m);
1723         return (EINVAL);
1724 }
1725
1726 /*
1727  * XXX
1728  * The whole multicast option thing needs to be re-thought.
1729  * Several of these options are equally applicable to non-multicast
1730  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1731  * standard option (IP_TTL).
1732  */
1733
1734 /*
1735  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1736  */
1737 static struct ifnet *
1738 ip_multicast_if(struct in_addr *a, int *ifindexp)
1739 {
1740         int ifindex;
1741         struct ifnet *ifp;
1742
1743         if (ifindexp)
1744                 *ifindexp = 0;
1745         if (ntohl(a->s_addr) >> 24 == 0) {
1746                 ifindex = ntohl(a->s_addr) & 0xffffff;
1747                 if (ifindex < 0 || if_index < ifindex)
1748                         return NULL;
1749                 ifp = ifindex2ifnet[ifindex];
1750                 if (ifindexp)
1751                         *ifindexp = ifindex;
1752         } else {
1753                 INADDR_TO_IFP(*a, ifp);
1754         }
1755         return ifp;
1756 }
1757
1758 /*
1759  * Set the IP multicast options in response to user setsockopt().
1760  */
1761 static int
1762 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1763 {
1764         int error = 0;
1765         int i;
1766         struct in_addr addr;
1767         struct ip_mreq mreq;
1768         struct ifnet *ifp;
1769         struct ip_moptions *imo = *imop;
1770         struct sockaddr_in *dst;
1771         int ifindex;
1772         int s;
1773
1774         if (imo == NULL) {
1775                 /*
1776                  * No multicast option buffer attached to the pcb;
1777                  * allocate one and initialize to default values.
1778                  */
1779                 imo = malloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1780
1781                 if (imo == NULL)
1782                         return (ENOBUFS);
1783                 *imop = imo;
1784                 imo->imo_multicast_ifp = NULL;
1785                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1786                 imo->imo_multicast_vif = -1;
1787                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1788                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1789                 imo->imo_num_memberships = 0;
1790         }
1791
1792         switch (sopt->sopt_name) {
1793         /* store an index number for the vif you wanna use in the send */
1794         case IP_MULTICAST_VIF:
1795                 if (legal_vif_num == 0) {
1796                         error = EOPNOTSUPP;
1797                         break;
1798                 }
1799                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1800                 if (error)
1801                         break;
1802                 if (!legal_vif_num(i) && (i != -1)) {
1803                         error = EINVAL;
1804                         break;
1805                 }
1806                 imo->imo_multicast_vif = i;
1807                 break;
1808
1809         case IP_MULTICAST_IF:
1810                 /*
1811                  * Select the interface for outgoing multicast packets.
1812                  */
1813                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1814                 if (error)
1815                         break;
1816                 /*
1817                  * INADDR_ANY is used to remove a previous selection.
1818                  * When no interface is selected, a default one is
1819                  * chosen every time a multicast packet is sent.
1820                  */
1821                 if (addr.s_addr == INADDR_ANY) {
1822                         imo->imo_multicast_ifp = NULL;
1823                         break;
1824                 }
1825                 /*
1826                  * The selected interface is identified by its local
1827                  * IP address.  Find the interface and confirm that
1828                  * it supports multicasting.
1829                  */
1830                 s = splimp();
1831                 ifp = ip_multicast_if(&addr, &ifindex);
1832                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1833                         splx(s);
1834                         error = EADDRNOTAVAIL;
1835                         break;
1836                 }
1837                 imo->imo_multicast_ifp = ifp;
1838                 if (ifindex)
1839                         imo->imo_multicast_addr = addr;
1840                 else
1841                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1842                 splx(s);
1843                 break;
1844
1845         case IP_MULTICAST_TTL:
1846                 /*
1847                  * Set the IP time-to-live for outgoing multicast packets.
1848                  * The original multicast API required a char argument,
1849                  * which is inconsistent with the rest of the socket API.
1850                  * We allow either a char or an int.
1851                  */
1852                 if (sopt->sopt_valsize == 1) {
1853                         u_char ttl;
1854                         error = sooptcopyin(sopt, &ttl, 1, 1);
1855                         if (error)
1856                                 break;
1857                         imo->imo_multicast_ttl = ttl;
1858                 } else {
1859                         u_int ttl;
1860                         error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1861                         if (error)
1862                                 break;
1863                         if (ttl > 255)
1864                                 error = EINVAL;
1865                         else
1866                                 imo->imo_multicast_ttl = ttl;
1867                 }
1868                 break;
1869
1870         case IP_MULTICAST_LOOP:
1871                 /*
1872                  * Set the loopback flag for outgoing multicast packets.
1873                  * Must be zero or one.  The original multicast API required a
1874                  * char argument, which is inconsistent with the rest
1875                  * of the socket API.  We allow either a char or an int.
1876                  */
1877                 if (sopt->sopt_valsize == 1) {
1878                         u_char loop;
1879
1880                         error = sooptcopyin(sopt, &loop, 1, 1);
1881                         if (error)
1882                                 break;
1883                         imo->imo_multicast_loop = !!loop;
1884                 } else {
1885                         u_int loop;
1886
1887                         error = sooptcopyin(sopt, &loop, sizeof loop,
1888                                             sizeof loop);
1889                         if (error)
1890                                 break;
1891                         imo->imo_multicast_loop = !!loop;
1892                 }
1893                 break;
1894
1895         case IP_ADD_MEMBERSHIP:
1896                 /*
1897                  * Add a multicast group membership.
1898                  * Group must be a valid IP multicast address.
1899                  */
1900                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1901                 if (error)
1902                         break;
1903
1904                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1905                         error = EINVAL;
1906                         break;
1907                 }
1908                 s = splimp();
1909                 /*
1910                  * If no interface address was provided, use the interface of
1911                  * the route to the given multicast address.
1912                  */
1913                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1914                         struct route ro;
1915
1916                         bzero(&ro, sizeof ro);
1917                         dst = (struct sockaddr_in *)&ro.ro_dst;
1918                         dst->sin_len = sizeof *dst;
1919                         dst->sin_family = AF_INET;
1920                         dst->sin_addr = mreq.imr_multiaddr;
1921                         rtalloc(&ro);
1922                         if (ro.ro_rt == NULL) {
1923                                 error = EADDRNOTAVAIL;
1924                                 splx(s);
1925                                 break;
1926                         }
1927                         --ro.ro_rt->rt_refcnt;
1928                         ifp = ro.ro_rt->rt_ifp;
1929                 }
1930                 else {
1931                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1932                 }
1933
1934                 /*
1935                  * See if we found an interface, and confirm that it
1936                  * supports multicast.
1937                  */
1938                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1939                         error = EADDRNOTAVAIL;
1940                         splx(s);
1941                         break;
1942                 }
1943                 /*
1944                  * See if the membership already exists or if all the
1945                  * membership slots are full.
1946                  */
1947                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1948                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1949                             imo->imo_membership[i]->inm_addr.s_addr
1950                                                 == mreq.imr_multiaddr.s_addr)
1951                                 break;
1952                 }
1953                 if (i < imo->imo_num_memberships) {
1954                         error = EADDRINUSE;
1955                         splx(s);
1956                         break;
1957                 }
1958                 if (i == IP_MAX_MEMBERSHIPS) {
1959                         error = ETOOMANYREFS;
1960                         splx(s);
1961                         break;
1962                 }
1963                 /*
1964                  * Everything looks good; add a new record to the multicast
1965                  * address list for the given interface.
1966                  */
1967                 if ((imo->imo_membership[i] =
1968                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1969                         error = ENOBUFS;
1970                         splx(s);
1971                         break;
1972                 }
1973                 ++imo->imo_num_memberships;
1974                 splx(s);
1975                 break;
1976
1977         case IP_DROP_MEMBERSHIP:
1978                 /*
1979                  * Drop a multicast group membership.
1980                  * Group must be a valid IP multicast address.
1981                  */
1982                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1983                 if (error)
1984                         break;
1985
1986                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1987                         error = EINVAL;
1988                         break;
1989                 }
1990
1991                 s = splimp();
1992                 /*
1993                  * If an interface address was specified, get a pointer
1994                  * to its ifnet structure.
1995                  */
1996                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1997                         ifp = NULL;
1998                 else {
1999                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2000                         if (ifp == NULL) {
2001                                 error = EADDRNOTAVAIL;
2002                                 splx(s);
2003                                 break;
2004                         }
2005                 }
2006                 /*
2007                  * Find the membership in the membership array.
2008                  */
2009                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2010                         if ((ifp == NULL ||
2011                              imo->imo_membership[i]->inm_ifp == ifp) &&
2012                             imo->imo_membership[i]->inm_addr.s_addr ==
2013                             mreq.imr_multiaddr.s_addr)
2014                                 break;
2015                 }
2016                 if (i == imo->imo_num_memberships) {
2017                         error = EADDRNOTAVAIL;
2018                         splx(s);
2019                         break;
2020                 }
2021                 /*
2022                  * Give up the multicast address record to which the
2023                  * membership points.
2024                  */
2025                 in_delmulti(imo->imo_membership[i]);
2026                 /*
2027                  * Remove the gap in the membership array.
2028                  */
2029                 for (++i; i < imo->imo_num_memberships; ++i)
2030                         imo->imo_membership[i-1] = imo->imo_membership[i];
2031                 --imo->imo_num_memberships;
2032                 splx(s);
2033                 break;
2034
2035         default:
2036                 error = EOPNOTSUPP;
2037                 break;
2038         }
2039
2040         /*
2041          * If all options have default values, no need to keep the mbuf.
2042          */
2043         if (imo->imo_multicast_ifp == NULL &&
2044             imo->imo_multicast_vif == -1 &&
2045             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2046             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2047             imo->imo_num_memberships == 0) {
2048                 free(*imop, M_IPMOPTS);
2049                 *imop = NULL;
2050         }
2051
2052         return (error);
2053 }
2054
2055 /*
2056  * Return the IP multicast options in response to user getsockopt().
2057  */
2058 static int
2059 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2060 {
2061         struct in_addr addr;
2062         struct in_ifaddr *ia;
2063         int error, optval;
2064         u_char coptval;
2065
2066         error = 0;
2067         switch (sopt->sopt_name) {
2068         case IP_MULTICAST_VIF:
2069                 if (imo != NULL)
2070                         optval = imo->imo_multicast_vif;
2071                 else
2072                         optval = -1;
2073                 error = sooptcopyout(sopt, &optval, sizeof optval);
2074                 break;
2075
2076         case IP_MULTICAST_IF:
2077                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2078                         addr.s_addr = INADDR_ANY;
2079                 else if (imo->imo_multicast_addr.s_addr) {
2080                         /* return the value user has set */
2081                         addr = imo->imo_multicast_addr;
2082                 } else {
2083                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2084                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2085                                 : IA_SIN(ia)->sin_addr.s_addr;
2086                 }
2087                 error = sooptcopyout(sopt, &addr, sizeof addr);
2088                 break;
2089
2090         case IP_MULTICAST_TTL:
2091                 if (imo == NULL)
2092                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2093                 else
2094                         optval = coptval = imo->imo_multicast_ttl;
2095                 if (sopt->sopt_valsize == 1)
2096                         error = sooptcopyout(sopt, &coptval, 1);
2097                 else
2098                         error = sooptcopyout(sopt, &optval, sizeof optval);
2099                 break;
2100
2101         case IP_MULTICAST_LOOP:
2102                 if (imo == NULL)
2103                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2104                 else
2105                         optval = coptval = imo->imo_multicast_loop;
2106                 if (sopt->sopt_valsize == 1)
2107                         error = sooptcopyout(sopt, &coptval, 1);
2108                 else
2109                         error = sooptcopyout(sopt, &optval, sizeof optval);
2110                 break;
2111
2112         default:
2113                 error = ENOPROTOOPT;
2114                 break;
2115         }
2116         return (error);
2117 }
2118
2119 /*
2120  * Discard the IP multicast options.
2121  */
2122 void
2123 ip_freemoptions(struct ip_moptions *imo)
2124 {
2125         int i;
2126
2127         if (imo != NULL) {
2128                 for (i = 0; i < imo->imo_num_memberships; ++i)
2129                         in_delmulti(imo->imo_membership[i]);
2130                 free(imo, M_IPMOPTS);
2131         }
2132 }
2133
2134 /*
2135  * Routine called from ip_output() to loop back a copy of an IP multicast
2136  * packet to the input queue of a specified interface.  Note that this
2137  * calls the output routine of the loopback "driver", but with an interface
2138  * pointer that might NOT be a loopback interface -- evil, but easier than
2139  * replicating that code here.
2140  */
2141 static void
2142 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2143              int hlen)
2144 {
2145         struct ip *ip;
2146         struct mbuf *copym;
2147
2148         copym = m_copypacket(m, MB_DONTWAIT);
2149         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2150                 copym = m_pullup(copym, hlen);
2151         if (copym != NULL) {
2152                 /*
2153                  * if the checksum hasn't been computed, mark it as valid
2154                  */
2155                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2156                         in_delayed_cksum(copym);
2157                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2158                         copym->m_pkthdr.csum_flags |=
2159                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2160                         copym->m_pkthdr.csum_data = 0xffff;
2161                 }
2162                 /*
2163                  * We don't bother to fragment if the IP length is greater
2164                  * than the interface's MTU.  Can this possibly matter?
2165                  */
2166                 ip = mtod(copym, struct ip *);
2167                 ip->ip_len = htons(ip->ip_len);
2168                 ip->ip_off = htons(ip->ip_off);
2169                 ip->ip_sum = 0;
2170                 if (ip->ip_vhl == IP_VHL_BORING) {
2171                         ip->ip_sum = in_cksum_hdr(ip);
2172                 } else {
2173                         ip->ip_sum = in_cksum(copym, hlen);
2174                 }
2175                 /*
2176                  * NB:
2177                  * It's not clear whether there are any lingering
2178                  * reentrancy problems in other areas which might
2179                  * be exposed by using ip_input directly (in
2180                  * particular, everything which modifies the packet
2181                  * in-place).  Yet another option is using the
2182                  * protosw directly to deliver the looped back
2183                  * packet.  For the moment, we'll err on the side
2184                  * of safety by using if_simloop().
2185                  */
2186 #if 1 /* XXX */
2187                 if (dst->sin_family != AF_INET) {
2188                         printf("ip_mloopback: bad address family %d\n",
2189                                                 dst->sin_family);
2190                         dst->sin_family = AF_INET;
2191                 }
2192 #endif
2193
2194 #ifdef notdef
2195                 copym->m_pkthdr.rcvif = ifp;
2196                 ip_input(copym);
2197 #else
2198                 if_simloop(ifp, copym, dst->sin_family, 0);
2199 #endif
2200         }
2201 }