Create another entry point into ip_input() so MT_TAGs will work.
[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. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *      This product includes software developed by the University of
16  *      California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
34  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
35  * $DragonFly: src/sys/netinet/ip_output.c,v 1.12 2004/04/13 00:14:01 hsu Exp $
36  */
37
38 #define _IP_VHL
39
40 #include "opt_ipfw.h"
41 #include "opt_ipdn.h"
42 #include "opt_ipdivert.h"
43 #include "opt_ipfilter.h"
44 #include "opt_ipsec.h"
45 #include "opt_pfil_hooks.h"
46 #include "opt_random_ip_id.h"
47 #include "opt_mbuf_stress_test.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/protosw.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/proc.h>
58 #include <sys/sysctl.h>
59 #include <sys/in_cksum.h>
60
61 #include <net/if.h>
62 #include <net/netisr.h>
63 #ifdef PFIL_HOOKS
64 #include <net/pfil.h>
65 #endif
66 #include <net/route.h>
67
68 #include <netinet/in.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/ip.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_var.h>
73 #include <netinet/ip_var.h>
74
75 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
76
77 #ifdef IPSEC
78 #include <netinet6/ipsec.h>
79 #include <netproto/key/key.h>
80 #ifdef IPSEC_DEBUG
81 #include <netproto/key/key_debug.h>
82 #else
83 #define KEYDEBUG(lev,arg)
84 #endif
85 #endif /*IPSEC*/
86
87 #ifdef FAST_IPSEC
88 #include <netipsec/ipsec.h>
89 #include <netipsec/xform.h>
90 #include <netipsec/key.h>
91 #endif /*FAST_IPSEC*/
92
93 #include <net/ipfw/ip_fw.h>
94 #include <net/dummynet/ip_dummynet.h>
95
96 #define print_ip(x, a, y)        printf("%s %d.%d.%d.%d%s",\
97                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
98                                   (ntohl(a.s_addr)>>16)&0xFF,\
99                                   (ntohl(a.s_addr)>>8)&0xFF,\
100                                   (ntohl(a.s_addr))&0xFF, y);
101
102 u_short ip_id;
103
104 #ifdef MBUF_STRESS_TEST
105 int mbuf_frag_size = 0;
106 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
107         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
108 #endif
109
110 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
111 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
112 static void     ip_mloopback
113         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
114 static int      ip_getmoptions
115         (struct sockopt *, struct ip_moptions *);
116 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
117 static int      ip_setmoptions
118         (struct sockopt *, struct ip_moptions **);
119
120 int     ip_optcopy(struct ip *, struct ip *);
121 extern int (*fr_checkp) (struct ip *, int, struct ifnet *, int, struct mbuf **);
122
123
124 extern  struct protosw inetsw[];
125
126 /*
127  * IP output.  The packet in mbuf chain m contains a skeletal IP
128  * header (with len, off, ttl, proto, tos, src, dst).
129  * The mbuf chain containing the packet will be freed.
130  * The mbuf opt, if present, will not be freed.
131  */
132 int
133 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
134         int flags, struct ip_moptions *imo, struct inpcb *inp)
135 {
136         struct ip *ip;
137         struct ifnet *ifp = NULL;       /* keep compiler happy */
138         struct mbuf *m;
139         int hlen = sizeof (struct ip);
140         int len, off, error = 0;
141         struct sockaddr_in *dst = NULL; /* keep compiler happy */
142         struct in_ifaddr *ia = NULL;
143         int isbroadcast, sw_csum;
144         struct in_addr pkt_dst;
145 #ifdef IPSEC
146         struct route iproute;
147         struct secpolicy *sp = NULL;
148         struct socket *so = inp ? inp->inp_socket : NULL;
149 #endif
150 #ifdef FAST_IPSEC
151         struct route iproute;
152         struct m_tag *mtag;
153         struct secpolicy *sp = NULL;
154         struct tdb_ident *tdbi;
155         int s;
156 #endif /* FAST_IPSEC */
157         struct ip_fw_args args;
158         int src_was_INADDR_ANY = 0;     /* as the name says... */
159
160         args.eh = NULL;
161         args.rule = NULL;
162         args.next_hop = NULL;
163         args.divert_rule = 0;                   /* divert cookie */
164
165         /* Grab info from MT_TAG mbufs prepended to the chain. */
166         for (; m0 && m0->m_type == MT_TAG; m0 = m0->m_next) {
167                 switch(m0->_m_tag_id) {
168                 default:
169                         printf("ip_output: unrecognised MT_TAG tag %d\n",
170                             m0->_m_tag_id);
171                         break;
172
173                 case PACKET_TAG_DUMMYNET:
174                         /*
175                          * the packet was already tagged, so part of the
176                          * processing was already done, and we need to go down.
177                          * Get parameters from the header.
178                          */
179                         args.rule = ((struct dn_pkt *)m0)->rule;
180                         opt = NULL ;
181                         ro = & ( ((struct dn_pkt *)m0)->ro ) ;
182                         imo = NULL ;
183                         dst = ((struct dn_pkt *)m0)->dn_dst ;
184                         ifp = ((struct dn_pkt *)m0)->ifp ;
185                         flags = ((struct dn_pkt *)m0)->flags ;
186                         break;
187
188                 case PACKET_TAG_DIVERT:
189                         args.divert_rule = (int)m0->m_data & 0xffff;
190                         break;
191
192                 case PACKET_TAG_IPFORWARD:
193                         args.next_hop = (struct sockaddr_in *)m0->m_data;
194                         break;
195                 }
196         }
197         m = m0;
198
199         KASSERT(!m || (m->m_flags & M_PKTHDR) != 0, ("ip_output: no HDR"));
200 #ifndef FAST_IPSEC
201         KASSERT(ro != NULL, ("ip_output: no route, proto %d",
202             mtod(m, struct ip *)->ip_p));
203 #endif
204
205         if (args.rule != NULL) {        /* dummynet already saw us */
206                 ip = mtod(m, struct ip *);
207                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
208                 if (ro->ro_rt)
209                         ia = ifatoia(ro->ro_rt->rt_ifa);
210                 goto sendit;
211         }
212
213         if (opt) {
214                 len = 0;
215                 m = ip_insertoptions(m, opt, &len);
216                 if (len != 0)
217                         hlen = len;
218         }
219         ip = mtod(m, struct ip *);
220         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
221
222         /*
223          * Fill in IP header.
224          */
225         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
226                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
227                 ip->ip_off &= IP_DF;
228 #ifdef RANDOM_IP_ID
229                 ip->ip_id = ip_randomid();
230 #else
231                 ip->ip_id = htons(ip_id++);
232 #endif
233                 ipstat.ips_localout++;
234         } else {
235                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
236         }
237
238 #ifdef FAST_IPSEC
239         if (ro == NULL) {
240                 ro = &iproute;
241                 bzero(ro, sizeof (*ro));
242         }
243 #endif /* FAST_IPSEC */
244         dst = (struct sockaddr_in *)&ro->ro_dst;
245         /*
246          * If there is a cached route,
247          * check that it is to the same destination
248          * and is still up.  If not, free it and try again.
249          * The address family should also be checked in case of sharing the
250          * cache with IPv6.
251          */
252         if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
253                           dst->sin_family != AF_INET ||
254                           dst->sin_addr.s_addr != pkt_dst.s_addr)) {
255                 RTFREE(ro->ro_rt);
256                 ro->ro_rt = (struct rtentry *)0;
257         }
258         if (ro->ro_rt == 0) {
259                 bzero(dst, sizeof(*dst));
260                 dst->sin_family = AF_INET;
261                 dst->sin_len = sizeof(*dst);
262                 dst->sin_addr = pkt_dst;
263         }
264         /*
265          * If routing to interface only,
266          * short circuit routing lookup.
267          */
268         if (flags & IP_ROUTETOIF) {
269                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 &&
270                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) {
271                         ipstat.ips_noroute++;
272                         error = ENETUNREACH;
273                         goto bad;
274                 }
275                 ifp = ia->ia_ifp;
276                 ip->ip_ttl = 1;
277                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
278         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
279             imo != NULL && imo->imo_multicast_ifp != NULL) {
280                 /*
281                  * Bypass the normal routing lookup for multicast
282                  * packets if the interface is specified.
283                  */
284                 ifp = imo->imo_multicast_ifp;
285                 IFP_TO_IA(ifp, ia);
286                 isbroadcast = 0;        /* fool gcc */
287         } else {
288                 /*
289                  * If this is the case, we probably don't want to allocate
290                  * a protocol-cloned route since we didn't get one from the
291                  * ULP.  This lets TCP do its thing, while not burdening
292                  * forwarding or ICMP with the overhead of cloning a route.
293                  * Of course, we still want to do any cloning requested by
294                  * the link layer, as this is probably required in all cases
295                  * for correct operation (as it is for ARP).
296                  */
297                 if (ro->ro_rt == 0)
298                         rtalloc_ign(ro, RTF_PRCLONING);
299                 if (ro->ro_rt == 0) {
300                         ipstat.ips_noroute++;
301                         error = EHOSTUNREACH;
302                         goto bad;
303                 }
304                 ia = ifatoia(ro->ro_rt->rt_ifa);
305                 ifp = ro->ro_rt->rt_ifp;
306                 ro->ro_rt->rt_use++;
307                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
308                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
309                 if (ro->ro_rt->rt_flags & RTF_HOST)
310                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
311                 else
312                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
313         }
314         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
315                 struct in_multi *inm;
316
317                 m->m_flags |= M_MCAST;
318                 /*
319                  * IP destination address is multicast.  Make sure "dst"
320                  * still points to the address in "ro".  (It may have been
321                  * changed to point to a gateway address, above.)
322                  */
323                 dst = (struct sockaddr_in *)&ro->ro_dst;
324                 /*
325                  * See if the caller provided any multicast options
326                  */
327                 if (imo != NULL) {
328                         ip->ip_ttl = imo->imo_multicast_ttl;
329                         if (imo->imo_multicast_vif != -1)
330                                 ip->ip_src.s_addr =
331                                     ip_mcast_src ?
332                                     ip_mcast_src(imo->imo_multicast_vif) :
333                                     INADDR_ANY;
334                 } else
335                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
336                 /*
337                  * Confirm that the outgoing interface supports multicast.
338                  */
339                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
340                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
341                                 ipstat.ips_noroute++;
342                                 error = ENETUNREACH;
343                                 goto bad;
344                         }
345                 }
346                 /*
347                  * If source address not specified yet, use address
348                  * of outgoing interface.
349                  */
350                 if (ip->ip_src.s_addr == INADDR_ANY) {
351                         /* Interface may have no addresses. */
352                         if (ia != NULL)
353                                 ip->ip_src = IA_SIN(ia)->sin_addr;
354                 }
355
356                 if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
357                         /*
358                          * XXX
359                          * delayed checksums are not currently
360                          * compatible with IP multicast routing
361                          */
362                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
363                                 in_delayed_cksum(m);
364                                 m->m_pkthdr.csum_flags &=
365                                         ~CSUM_DELAY_DATA;
366                         }
367                 }
368                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
369                 if (inm != NULL &&
370                    (imo == NULL || imo->imo_multicast_loop)) {
371                         /*
372                          * If we belong to the destination multicast group
373                          * on the outgoing interface, and the caller did not
374                          * forbid loopback, loop back a copy.
375                          */
376                         ip_mloopback(ifp, m, dst, hlen);
377                 }
378                 else {
379                         /*
380                          * If we are acting as a multicast router, perform
381                          * multicast forwarding as if the packet had just
382                          * arrived on the interface to which we are about
383                          * to send.  The multicast forwarding function
384                          * recursively calls this function, using the
385                          * IP_FORWARDING flag to prevent infinite recursion.
386                          *
387                          * Multicasts that are looped back by ip_mloopback(),
388                          * above, will be forwarded by the ip_input() routine,
389                          * if necessary.
390                          */
391                         if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
392                                 /*
393                                  * If rsvp daemon is not running, do not
394                                  * set ip_moptions. This ensures that the packet
395                                  * is multicast and not just sent down one link
396                                  * as prescribed by rsvpd.
397                                  */
398                                 if (!rsvp_on)
399                                         imo = NULL;
400                                 if (ip_mforward &&
401                                     ip_mforward(ip, ifp, m, imo) != 0) {
402                                         m_freem(m);
403                                         goto done;
404                                 }
405                         }
406                 }
407
408                 /*
409                  * Multicasts with a time-to-live of zero may be looped-
410                  * back, above, but must not be transmitted on a network.
411                  * Also, multicasts addressed to the loopback interface
412                  * are not sent -- the above call to ip_mloopback() will
413                  * loop back a copy if this host actually belongs to the
414                  * destination group on the loopback interface.
415                  */
416                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
417                         m_freem(m);
418                         goto done;
419                 }
420
421                 goto sendit;
422         }
423 #ifndef notdef
424         /*
425          * If the source address is not specified yet, use the address
426          * of the outoing interface. In case, keep note we did that, so
427          * if the the firewall changes the next-hop causing the output
428          * interface to change, we can fix that.
429          */
430         if (ip->ip_src.s_addr == INADDR_ANY) {
431                 /* Interface may have no addresses. */
432                 if (ia != NULL) {
433                         ip->ip_src = IA_SIN(ia)->sin_addr;
434                         src_was_INADDR_ANY = 1;
435                 }
436         }
437 #endif /* notdef */
438         /*
439          * Verify that we have any chance at all of being able to queue
440          *      the packet or packet fragments
441          */
442         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
443                 ifp->if_snd.ifq_maxlen) {
444                         error = ENOBUFS;
445                         ipstat.ips_odropped++;
446                         goto bad;
447         }
448
449         /*
450          * Look for broadcast address and
451          * verify user is allowed to send
452          * such a packet.
453          */
454         if (isbroadcast) {
455                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
456                         error = EADDRNOTAVAIL;
457                         goto bad;
458                 }
459                 if ((flags & IP_ALLOWBROADCAST) == 0) {
460                         error = EACCES;
461                         goto bad;
462                 }
463                 /* don't allow broadcast messages to be fragmented */
464                 if (ip->ip_len > ifp->if_mtu) {
465                         error = EMSGSIZE;
466                         goto bad;
467                 }
468                 m->m_flags |= M_BCAST;
469         } else {
470                 m->m_flags &= ~M_BCAST;
471         }
472
473 sendit:
474 #ifdef IPSEC
475         /* get SP for this packet */
476         if (so == NULL)
477                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
478         else
479                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
480
481         if (sp == NULL) {
482                 ipsecstat.out_inval++;
483                 goto bad;
484         }
485
486         error = 0;
487
488         /* check policy */
489         switch (sp->policy) {
490         case IPSEC_POLICY_DISCARD:
491                 /*
492                  * This packet is just discarded.
493                  */
494                 ipsecstat.out_polvio++;
495                 goto bad;
496
497         case IPSEC_POLICY_BYPASS:
498         case IPSEC_POLICY_NONE:
499                 /* no need to do IPsec. */
500                 goto skip_ipsec;
501         
502         case IPSEC_POLICY_IPSEC:
503                 if (sp->req == NULL) {
504                         /* acquire a policy */
505                         error = key_spdacquire(sp);
506                         goto bad;
507                 }
508                 break;
509
510         case IPSEC_POLICY_ENTRUST:
511         default:
512                 printf("ip_output: Invalid policy found. %d\n", sp->policy);
513         }
514     {
515         struct ipsec_output_state state;
516         bzero(&state, sizeof(state));
517         state.m = m;
518         if (flags & IP_ROUTETOIF) {
519                 state.ro = &iproute;
520                 bzero(&iproute, sizeof(iproute));
521         } else
522                 state.ro = ro;
523         state.dst = (struct sockaddr *)dst;
524
525         ip->ip_sum = 0;
526
527         /*
528          * XXX
529          * delayed checksums are not currently compatible with IPsec
530          */
531         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
532                 in_delayed_cksum(m);
533                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
534         }
535
536         ip->ip_len = htons(ip->ip_len);
537         ip->ip_off = htons(ip->ip_off);
538
539         error = ipsec4_output(&state, sp, flags);
540
541         m = state.m;
542         if (flags & IP_ROUTETOIF) {
543                 /*
544                  * if we have tunnel mode SA, we may need to ignore
545                  * IP_ROUTETOIF.
546                  */
547                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
548                         flags &= ~IP_ROUTETOIF;
549                         ro = state.ro;
550                 }
551         } else
552                 ro = state.ro;
553         dst = (struct sockaddr_in *)state.dst;
554         if (error) {
555                 /* mbuf is already reclaimed in ipsec4_output. */
556                 m0 = NULL;
557                 switch (error) {
558                 case EHOSTUNREACH:
559                 case ENETUNREACH:
560                 case EMSGSIZE:
561                 case ENOBUFS:
562                 case ENOMEM:
563                         break;
564                 default:
565                         printf("ip4_output (ipsec): error code %d\n", error);
566                         /*fall through*/
567                 case ENOENT:
568                         /* don't show these error codes to the user */
569                         error = 0;
570                         break;
571                 }
572                 goto bad;
573         }
574     }
575
576         /* be sure to update variables that are affected by ipsec4_output() */
577         ip = mtod(m, struct ip *);
578 #ifdef _IP_VHL
579         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
580 #else
581         hlen = ip->ip_hl << 2;
582 #endif
583         if (ro->ro_rt == NULL) {
584                 if ((flags & IP_ROUTETOIF) == 0) {
585                         printf("ip_output: "
586                                 "can't update route after IPsec processing\n");
587                         error = EHOSTUNREACH;   /*XXX*/
588                         goto bad;
589                 }
590         } else {
591                 ia = ifatoia(ro->ro_rt->rt_ifa);
592                 ifp = ro->ro_rt->rt_ifp;
593         }
594
595         /* make it flipped, again. */
596         ip->ip_len = ntohs(ip->ip_len);
597         ip->ip_off = ntohs(ip->ip_off);
598 skip_ipsec:
599 #endif /*IPSEC*/
600 #ifdef FAST_IPSEC
601         /*
602          * Check the security policy (SP) for the packet and, if
603          * required, do IPsec-related processing.  There are two
604          * cases here; the first time a packet is sent through
605          * it will be untagged and handled by ipsec4_checkpolicy.
606          * If the packet is resubmitted to ip_output (e.g. after
607          * AH, ESP, etc. processing), there will be a tag to bypass
608          * the lookup and related policy checking.
609          */
610         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
611         s = splnet();
612         if (mtag != NULL) {
613                 tdbi = (struct tdb_ident *)(mtag + 1);
614                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
615                 if (sp == NULL)
616                         error = -EINVAL;        /* force silent drop */
617                 m_tag_delete(m, mtag);
618         } else {
619                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
620                                         &error, inp);
621         }
622         /*
623          * There are four return cases:
624          *    sp != NULL                    apply IPsec policy
625          *    sp == NULL, error == 0        no IPsec handling needed
626          *    sp == NULL, error == -EINVAL  discard packet w/o error
627          *    sp == NULL, error != 0        discard packet, report error
628          */
629         if (sp != NULL) {
630                 /* Loop detection, check if ipsec processing already done */
631                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
632                 for (mtag = m_tag_first(m); mtag != NULL;
633                      mtag = m_tag_next(m, mtag)) {
634                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
635                                 continue;
636                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
637                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
638                                 continue;
639                         /*
640                          * Check if policy has an SA associated with it.
641                          * This can happen when an SP has yet to acquire
642                          * an SA; e.g. on first reference.  If it occurs,
643                          * then we let ipsec4_process_packet do its thing.
644                          */
645                         if (sp->req->sav == NULL)
646                                 break;
647                         tdbi = (struct tdb_ident *)(mtag + 1);
648                         if (tdbi->spi == sp->req->sav->spi &&
649                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
650                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
651                                  sizeof (union sockaddr_union)) == 0) {
652                                 /*
653                                  * No IPsec processing is needed, free
654                                  * reference to SP.
655                                  *
656                                  * NB: null pointer to avoid free at
657                                  *     done: below.
658                                  */
659                                 KEY_FREESP(&sp), sp = NULL;
660                                 splx(s);
661                                 goto spd_done;
662                         }
663                 }
664
665                 /*
666                  * Do delayed checksums now because we send before
667                  * this is done in the normal processing path.
668                  */
669                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
670                         in_delayed_cksum(m);
671                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
672                 }
673
674                 ip->ip_len = htons(ip->ip_len);
675                 ip->ip_off = htons(ip->ip_off);
676
677                 /* NB: callee frees mbuf */
678                 error = ipsec4_process_packet(m, sp->req, flags, 0);
679                 /*
680                  * Preserve KAME behaviour: ENOENT can be returned
681                  * when an SA acquire is in progress.  Don't propagate
682                  * this to user-level; it confuses applications.
683                  *
684                  * XXX this will go away when the SADB is redone.
685                  */
686                 if (error == ENOENT)
687                         error = 0;
688                 splx(s);
689                 goto done;
690         } else {
691                 splx(s);
692
693                 if (error != 0) {
694                         /*
695                          * Hack: -EINVAL is used to signal that a packet
696                          * should be silently discarded.  This is typically
697                          * because we asked key management for an SA and
698                          * it was delayed (e.g. kicked up to IKE).
699                          */
700                         if (error == -EINVAL)
701                                 error = 0;
702                         goto bad;
703                 } else {
704                         /* No IPsec processing for this packet. */
705                 }
706 #ifdef notyet
707                 /*
708                  * If deferred crypto processing is needed, check that
709                  * the interface supports it.
710                  */ 
711                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
712                 if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) {
713                         /* notify IPsec to do its own crypto */
714                         ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
715                         error = EHOSTUNREACH;
716                         goto bad;
717                 }
718 #endif
719         }
720 spd_done:
721 #endif /* FAST_IPSEC */
722         /*
723          * IpHack's section.
724          * - Xlate: translate packet's addr/port (NAT).
725          * - Firewall: deny/allow/etc.
726          * - Wrap: fake packet's addr/port <unimpl.>
727          * - Encapsulate: put it in another IP and send out. <unimp.>
728          */ 
729 #ifdef PFIL_HOOKS
730         /*
731          * Run through list of hooks for output packets.
732          */
733         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
734         if (error != 0 || m == NULL)
735                 goto done;
736         ip = mtod(m, struct ip *);
737 #endif /* PFIL_HOOKS */
738
739         /*
740          * Check with the firewall...
741          * but not if we are already being fwd'd from a firewall.
742          */
743         if (fw_enable && IPFW_LOADED && !args.next_hop) {
744                 struct sockaddr_in *old = dst;
745
746                 args.m = m;
747                 args.next_hop = dst;
748                 args.oif = ifp;
749                 off = ip_fw_chk_ptr(&args);
750                 m = args.m;
751                 dst = args.next_hop;
752
753                 /*
754                  * On return we must do the following:
755                  * m == NULL    -> drop the pkt (old interface, deprecated)
756                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
757                  * 1<=off<= 0xffff              -> DIVERT
758                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
759                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
760                  * dst != old                   -> IPFIREWALL_FORWARD
761                  * off==0, dst==old             -> accept
762                  * If some of the above modules are not compiled in, then
763                  * we should't have to check the corresponding condition
764                  * (because the ipfw control socket should not accept
765                  * unsupported rules), but better play safe and drop
766                  * packets in case of doubt.
767                  */
768                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
769                         if (m)
770                                 m_freem(m);
771                         error = EACCES;
772                         goto done;
773                 }
774                 ip = mtod(m, struct ip *);
775                 if (off == 0 && dst == old)             /* common case */
776                         goto pass;
777                 if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) {
778                         /*
779                          * pass the pkt to dummynet. Need to include
780                          * pipe number, m, ifp, ro, dst because these are
781                          * not recomputed in the next pass.
782                          * All other parameters have been already used and
783                          * so they are not needed anymore. 
784                          * XXX note: if the ifp or ro entry are deleted
785                          * while a pkt is in dummynet, we are in trouble!
786                          */ 
787                         args.ro = ro;
788                         args.dst = dst;
789                         args.flags = flags;
790
791                         error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT,
792                                 &args);
793                         goto done;
794                 }
795 #ifdef IPDIVERT
796                 if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) {
797                         struct mbuf *clone = NULL;
798
799                         /* Clone packet if we're doing a 'tee' */
800                         if ((off & IP_FW_PORT_TEE_FLAG) != 0)
801                                 clone = m_dup(m, M_DONTWAIT);
802
803                         /*
804                          * XXX
805                          * delayed checksums are not currently compatible
806                          * with divert sockets.
807                          */
808                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
809                                 in_delayed_cksum(m);
810                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
811                         }
812
813                         /* Restore packet header fields to original values */
814                         ip->ip_len = htons(ip->ip_len);
815                         ip->ip_off = htons(ip->ip_off);
816
817                         /* Deliver packet to divert input routine */
818                         divert_packet(m, 0, off & 0xffff, args.divert_rule);
819
820                         /* If 'tee', continue with original packet */
821                         if (clone != NULL) {
822                                 m = clone;
823                                 ip = mtod(m, struct ip *);
824                                 goto pass;
825                         }
826                         goto done;
827                 }
828 #endif
829
830                 /* IPFIREWALL_FORWARD */
831                 /*
832                  * Check dst to make sure it is directly reachable on the
833                  * interface we previously thought it was.
834                  * If it isn't (which may be likely in some situations) we have
835                  * to re-route it (ie, find a route for the next-hop and the
836                  * associated interface) and set them here. This is nested
837                  * forwarding which in most cases is undesirable, except where
838                  * such control is nigh impossible. So we do it here.
839                  * And I'm babbling.
840                  */
841                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
842 #if 0
843                         /*
844                          * XXX To improve readability, this block should be
845                          * changed into a function call as below:
846                          */
847                         error = ip_ipforward(&m, &dst, &ifp);
848                         if (error)
849                                 goto bad;
850                         if (m == NULL) /* ip_input consumed the mbuf */
851                                 goto done;
852 #else
853                         struct in_ifaddr *ia;
854
855                         /*
856                          * XXX sro_fwd below is static, and a pointer
857                          * to it gets passed to routines downstream.
858                          * This could have surprisingly bad results in
859                          * practice, because its content is overwritten
860                          * by subsequent packets.
861                          */
862                         /* There must be a better way to do this next line... */
863                         static struct route sro_fwd;
864                         struct route *ro_fwd = &sro_fwd;
865
866 #if 0
867                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
868                             dst->sin_addr, "\n");
869 #endif
870
871                         /*
872                          * We need to figure out if we have been forwarded
873                          * to a local socket. If so, then we should somehow 
874                          * "loop back" to ip_input, and get directed to the
875                          * PCB as if we had received this packet. This is
876                          * because it may be dificult to identify the packets
877                          * you want to forward until they are being output
878                          * and have selected an interface. (e.g. locally
879                          * initiated packets) If we used the loopback inteface,
880                          * we would not be able to control what happens 
881                          * as the packet runs through ip_input() as
882                          * it is done through a ISR.
883                          */
884                         LIST_FOREACH(ia,
885                             INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
886                                 /*
887                                  * If the addr to forward to is one
888                                  * of ours, we pretend to
889                                  * be the destination for this packet.
890                                  */
891                                 if (IA_SIN(ia)->sin_addr.s_addr ==
892                                                  dst->sin_addr.s_addr)
893                                         break;
894                         }
895                         if (ia) {       /* tell ip_input "dont filter" */
896                                 struct m_hdr tag;
897
898                                 tag.mh_type = MT_TAG;
899                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
900                                 tag.mh_data = (caddr_t)args.next_hop;
901                                 tag.mh_next = m;
902
903                                 if (m->m_pkthdr.rcvif == NULL)
904                                         m->m_pkthdr.rcvif = ifunit("lo0");
905                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
906                                         m->m_pkthdr.csum_flags |=
907                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
908                                         m0->m_pkthdr.csum_data = 0xffff;
909                                 }
910                                 m->m_pkthdr.csum_flags |=
911                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
912                                 ip->ip_len = htons(ip->ip_len);
913                                 ip->ip_off = htons(ip->ip_off);
914                                 ip_input((struct mbuf *)&tag);
915                                 goto done;
916                         }
917                         /* Some of the logic for this was
918                          * nicked from above.
919                          *
920                          * This rewrites the cached route in a local PCB.
921                          * Is this what we want to do?
922                          */
923                         bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
924
925                         ro_fwd->ro_rt = 0;
926                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
927
928                         if (ro_fwd->ro_rt == 0) {
929                                 ipstat.ips_noroute++;
930                                 error = EHOSTUNREACH;
931                                 goto bad;
932                         }
933
934                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
935                         ifp = ro_fwd->ro_rt->rt_ifp;
936                         ro_fwd->ro_rt->rt_use++;
937                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
938                                 dst = (struct sockaddr_in *)
939                                         ro_fwd->ro_rt->rt_gateway;
940                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
941                                 isbroadcast =
942                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
943                         else
944                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
945                         if (ro->ro_rt)
946                                 RTFREE(ro->ro_rt);
947                         ro->ro_rt = ro_fwd->ro_rt;
948                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
949
950 #endif  /* ... block to be put into a function */
951                         /*
952                          * If we added a default src ip earlier,
953                          * which would have been gotten from the-then
954                          * interface, do it again, from the new one.
955                          */
956                         if (src_was_INADDR_ANY)
957                                 ip->ip_src = IA_SIN(ia)->sin_addr;
958                         goto pass ;
959                 }
960
961                 /*
962                  * if we get here, none of the above matches, and 
963                  * we have to drop the pkt
964                  */
965                 m_freem(m);
966                 error = EACCES; /* not sure this is the right error msg */
967                 goto done;
968         }
969
970 pass:
971         /* 127/8 must not appear on wire - RFC1122. */
972         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
973             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
974                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
975                         ipstat.ips_badaddr++;
976                         error = EADDRNOTAVAIL;
977                         goto bad;
978                 }
979         }
980
981         m->m_pkthdr.csum_flags |= CSUM_IP;
982         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
983         if (sw_csum & CSUM_DELAY_DATA) {
984                 in_delayed_cksum(m);
985                 sw_csum &= ~CSUM_DELAY_DATA;
986         }
987         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
988
989         /*
990          * If small enough for interface, or the interface will take
991          * care of the fragmentation for us, can just send directly.
992          */
993         if (ip->ip_len <= ifp->if_mtu || ifp->if_hwassist & CSUM_FRAGMENT) {
994                 ip->ip_len = htons(ip->ip_len);
995                 ip->ip_off = htons(ip->ip_off);
996                 ip->ip_sum = 0;
997                 if (sw_csum & CSUM_DELAY_IP) {
998                         if (ip->ip_vhl == IP_VHL_BORING) {
999                                 ip->ip_sum = in_cksum_hdr(ip);
1000                         } else {
1001                                 ip->ip_sum = in_cksum(m, hlen);
1002                         }
1003                 }
1004
1005                 /* Record statistics for this interface address. */
1006                 if (!(flags & IP_FORWARDING) && ia) {
1007                         ia->ia_ifa.if_opackets++;
1008                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1009                 }
1010
1011 #ifdef IPSEC
1012                 /* clean ipsec history once it goes out of the node */
1013                 ipsec_delaux(m);
1014 #endif
1015
1016 #ifdef MBUF_STRESS_TEST
1017                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1018                         struct mbuf *m1, *m2;
1019                         int length, tmp;
1020
1021                         tmp = length = m->m_pkthdr.len;
1022
1023                         while ((length -= mbuf_frag_size) >= 1) {
1024                                 m1 = m_split(m, length, M_DONTWAIT);
1025                                 if (m1 == NULL)
1026                                         break;
1027                                 m1->m_flags &= ~M_PKTHDR;
1028                                 m2 = m;
1029                                 while (m2->m_next != NULL)
1030                                         m2 = m2->m_next;
1031                                 m2->m_next = m1;
1032                         }
1033                         m->m_pkthdr.len = tmp;
1034                 }
1035 #endif
1036                 error = (*ifp->if_output)(ifp, m,
1037                                 (struct sockaddr *)dst, ro->ro_rt);
1038                 goto done;
1039         }
1040
1041         if (ip->ip_off & IP_DF) {
1042                 error = EMSGSIZE;
1043                 /*
1044                  * This case can happen if the user changed the MTU
1045                  * of an interface after enabling IP on it.  Because
1046                  * most netifs don't keep track of routes pointing to
1047                  * them, there is no way for one to update all its
1048                  * routes when the MTU is changed.
1049                  */
1050                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1051                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1052                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1053                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1054                 }
1055                 ipstat.ips_cantfrag++;
1056                 goto bad;
1057         }
1058
1059         /*
1060          * Too large for interface; fragment if possible. If successful,
1061          * on return, m will point to a list of packets to be sent.
1062          */
1063         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1064         if (error)
1065                 goto bad;
1066         for (; m; m = m0) {
1067                 m0 = m->m_nextpkt;
1068                 m->m_nextpkt = 0;
1069 #ifdef IPSEC
1070                 /* clean ipsec history once it goes out of the node */
1071                 ipsec_delaux(m);
1072 #endif
1073                 if (error == 0) {
1074                         /* Record statistics for this interface address. */
1075                         if (ia != NULL) {
1076                                 ia->ia_ifa.if_opackets++;
1077                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1078                         }
1079                         
1080                         error = (*ifp->if_output)(ifp, m,
1081                             (struct sockaddr *)dst, ro->ro_rt);
1082                 } else
1083                         m_freem(m);
1084         }
1085
1086         if (error == 0)
1087                 ipstat.ips_fragmented++;
1088
1089 done:
1090 #ifdef IPSEC
1091         if (ro == &iproute && ro->ro_rt) {
1092                 RTFREE(ro->ro_rt);
1093                 ro->ro_rt = NULL;
1094         }
1095         if (sp != NULL) {
1096                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1097                         printf("DP ip_output call free SP:%p\n", sp));
1098                 key_freesp(sp);
1099         }
1100 #endif
1101 #ifdef FAST_IPSEC
1102         if (ro == &iproute && ro->ro_rt) {
1103                 RTFREE(ro->ro_rt);
1104                 ro->ro_rt = NULL;
1105         }
1106         if (sp != NULL)
1107                 KEY_FREESP(&sp);
1108 #endif
1109         return (error);
1110 bad:
1111         m_freem(m);
1112         goto done;
1113 }
1114
1115 /*
1116  * Create a chain of fragments which fit the given mtu. m_frag points to the
1117  * mbuf to be fragmented; on return it points to the chain with the fragments.
1118  * Return 0 if no error. If error, m_frag may contain a partially built
1119  * chain of fragments that should be freed by the caller.
1120  *
1121  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1122  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1123  */
1124 int
1125 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1126             u_long if_hwassist_flags, int sw_csum)
1127 {
1128         int error = 0;
1129         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1130         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1131         int off;
1132         struct mbuf *m0 = *m_frag;      /* the original packet          */
1133         int firstlen;
1134         struct mbuf **mnext;
1135         int nfrags;
1136
1137         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1138                 ipstat.ips_cantfrag++;
1139                 return EMSGSIZE;
1140         }
1141
1142         /*
1143          * Must be able to put at least 8 bytes per fragment.
1144          */
1145         if (len < 8)
1146                 return EMSGSIZE;
1147
1148         /*
1149          * If the interface will not calculate checksums on
1150          * fragmented packets, then do it here.
1151          */
1152         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
1153             (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
1154                 in_delayed_cksum(m0);
1155                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1156         }
1157
1158         if (len > PAGE_SIZE) {
1159                 /* 
1160                  * Fragment large datagrams such that each segment 
1161                  * contains a multiple of PAGE_SIZE amount of data, 
1162                  * plus headers. This enables a receiver to perform 
1163                  * page-flipping zero-copy optimizations.
1164                  *
1165                  * XXX When does this help given that sender and receiver
1166                  * could have different page sizes, and also mtu could
1167                  * be less than the receiver's page size ?
1168                  */
1169                 int newlen;
1170                 struct mbuf *m;
1171
1172                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1173                         off += m->m_len;
1174
1175                 /*
1176                  * firstlen (off - hlen) must be aligned on an 
1177                  * 8-byte boundary
1178                  */
1179                 if (off < hlen)
1180                         goto smart_frag_failure;
1181                 off = ((off - hlen) & ~7) + hlen;
1182                 newlen = (~PAGE_MASK) & mtu;
1183                 if ((newlen + sizeof (struct ip)) > mtu) {
1184                         /* we failed, go back the default */
1185 smart_frag_failure:
1186                         newlen = len;
1187                         off = hlen + len;
1188                 }
1189                 len = newlen;
1190
1191         } else {
1192                 off = hlen + len;
1193         }
1194
1195         firstlen = off - hlen;
1196         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1197
1198         /*
1199          * Loop through length of segment after first fragment,
1200          * make new header and copy data of each part and link onto chain.
1201          * Here, m0 is the original packet, m is the fragment being created.
1202          * The fragments are linked off the m_nextpkt of the original
1203          * packet, which after processing serves as the first fragment.
1204          */
1205         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1206                 struct ip *mhip;        /* ip header on the fragment */
1207                 struct mbuf *m;
1208                 int mhlen = sizeof (struct ip);
1209
1210                 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1211                 if (m == 0) {
1212                         error = ENOBUFS;
1213                         ipstat.ips_odropped++;
1214                         goto done;
1215                 }
1216                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1217                 /*
1218                  * In the first mbuf, leave room for the link header, then
1219                  * copy the original IP header including options. The payload
1220                  * goes into an additional mbuf chain returned by m_copy().
1221                  */
1222                 m->m_data += max_linkhdr;
1223                 mhip = mtod(m, struct ip *);
1224                 *mhip = *ip;
1225                 if (hlen > sizeof (struct ip)) {
1226                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
1227                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1228                 }
1229                 m->m_len = mhlen;
1230                 /* XXX do we need to add ip->ip_off below ? */
1231                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1232                 if (off + len >= ip->ip_len) {  /* last fragment */
1233                         len = ip->ip_len - off;
1234                         m->m_flags |= M_LASTFRAG;
1235                 } else
1236                         mhip->ip_off |= IP_MF;
1237                 mhip->ip_len = htons((u_short)(len + mhlen));
1238                 m->m_next = m_copy(m0, off, len);
1239                 if (m->m_next == 0) {           /* copy failed */
1240                         m_free(m);
1241                         error = ENOBUFS;        /* ??? */
1242                         ipstat.ips_odropped++;
1243                         goto done;
1244                 }
1245                 m->m_pkthdr.len = mhlen + len;
1246                 m->m_pkthdr.rcvif = (struct ifnet *)0;
1247                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1248                 mhip->ip_off = htons(mhip->ip_off);
1249                 mhip->ip_sum = 0;
1250                 if (sw_csum & CSUM_DELAY_IP)
1251                         mhip->ip_sum = in_cksum(m, mhlen);
1252                 *mnext = m;
1253                 mnext = &m->m_nextpkt;
1254         }
1255         ipstat.ips_ofragments += nfrags;
1256
1257         /* set first marker for fragment chain */
1258         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1259         m0->m_pkthdr.csum_data = nfrags;
1260
1261         /*
1262          * Update first fragment by trimming what's been copied out
1263          * and updating header.
1264          */
1265         m_adj(m0, hlen + firstlen - ip->ip_len);
1266         m0->m_pkthdr.len = hlen + firstlen;
1267         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1268         ip->ip_off |= IP_MF;
1269         ip->ip_off = htons(ip->ip_off);
1270         ip->ip_sum = 0;
1271         if (sw_csum & CSUM_DELAY_IP)
1272                 ip->ip_sum = in_cksum(m0, hlen);
1273
1274 done:
1275         *m_frag = m0;
1276         return error;
1277 }
1278
1279 void
1280 in_delayed_cksum(struct mbuf *m)
1281 {
1282         struct ip *ip;
1283         u_short csum, offset;
1284
1285         ip = mtod(m, struct ip *);
1286         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1287         csum = in_cksum_skip(m, ip->ip_len, offset);
1288         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1289                 csum = 0xffff;
1290         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1291
1292         if (offset + sizeof(u_short) > m->m_len) {
1293                 printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1294                     m->m_len, offset, ip->ip_p);
1295                 /*
1296                  * XXX
1297                  * this shouldn't happen, but if it does, the
1298                  * correct behavior may be to insert the checksum
1299                  * in the existing chain instead of rearranging it.
1300                  */
1301                 m = m_pullup(m, offset + sizeof(u_short));
1302         }
1303         *(u_short *)(m->m_data + offset) = csum;
1304 }
1305
1306 /*
1307  * Insert IP options into preformed packet.
1308  * Adjust IP destination as required for IP source routing,
1309  * as indicated by a non-zero in_addr at the start of the options.
1310  *
1311  * XXX This routine assumes that the packet has no options in place.
1312  */
1313 static struct mbuf *
1314 ip_insertoptions(m, opt, phlen)
1315         struct mbuf *m;
1316         struct mbuf *opt;
1317         int *phlen;
1318 {
1319         struct ipoption *p = mtod(opt, struct ipoption *);
1320         struct mbuf *n;
1321         struct ip *ip = mtod(m, struct ip *);
1322         unsigned optlen;
1323
1324         optlen = opt->m_len - sizeof(p->ipopt_dst);
1325         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1326                 *phlen = 0;
1327                 return (m);             /* XXX should fail */
1328         }
1329         if (p->ipopt_dst.s_addr)
1330                 ip->ip_dst = p->ipopt_dst;
1331         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1332                 MGETHDR(n, M_DONTWAIT, MT_HEADER);
1333                 if (n == 0) {
1334                         *phlen = 0;
1335                         return (m);
1336                 }
1337                 n->m_pkthdr.rcvif = (struct ifnet *)0;
1338                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1339                 m->m_len -= sizeof(struct ip);
1340                 m->m_data += sizeof(struct ip);
1341                 n->m_next = m;
1342                 m = n;
1343                 m->m_len = optlen + sizeof(struct ip);
1344                 m->m_data += max_linkhdr;
1345                 (void)memcpy(mtod(m, void *), ip, sizeof(struct ip));
1346         } else {
1347                 m->m_data -= optlen;
1348                 m->m_len += optlen;
1349                 m->m_pkthdr.len += optlen;
1350                 ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1351         }
1352         ip = mtod(m, struct ip *);
1353         bcopy(p->ipopt_list, ip + 1, optlen);
1354         *phlen = sizeof(struct ip) + optlen;
1355         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1356         ip->ip_len += optlen;
1357         return (m);
1358 }
1359
1360 /*
1361  * Copy options from ip to jp,
1362  * omitting those not copied during fragmentation.
1363  */
1364 int
1365 ip_optcopy(ip, jp)
1366         struct ip *ip, *jp;
1367 {
1368         u_char *cp, *dp;
1369         int opt, optlen, cnt;
1370
1371         cp = (u_char *)(ip + 1);
1372         dp = (u_char *)(jp + 1);
1373         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
1374         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1375                 opt = cp[0];
1376                 if (opt == IPOPT_EOL)
1377                         break;
1378                 if (opt == IPOPT_NOP) {
1379                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1380                         *dp++ = IPOPT_NOP;
1381                         optlen = 1;
1382                         continue;
1383                 }
1384
1385                 KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp),
1386                     ("ip_optcopy: malformed ipv4 option"));
1387                 optlen = cp[IPOPT_OLEN];
1388                 KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt,
1389                     ("ip_optcopy: malformed ipv4 option"));
1390
1391                 /* bogus lengths should have been caught by ip_dooptions */
1392                 if (optlen > cnt)
1393                         optlen = cnt;
1394                 if (IPOPT_COPIED(opt)) {
1395                         bcopy(cp, dp, optlen);
1396                         dp += optlen;
1397                 }
1398         }
1399         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1400                 *dp++ = IPOPT_EOL;
1401         return (optlen);
1402 }
1403
1404 /*
1405  * IP socket option processing.
1406  */
1407 int
1408 ip_ctloutput(so, sopt)
1409         struct socket *so;
1410         struct sockopt *sopt;
1411 {
1412         struct  inpcb *inp = sotoinpcb(so);
1413         int     error, optval;
1414
1415         error = optval = 0;
1416         if (sopt->sopt_level != IPPROTO_IP) {
1417                 return (EINVAL);
1418         }
1419
1420         switch (sopt->sopt_dir) {
1421         case SOPT_SET:
1422                 switch (sopt->sopt_name) {
1423                 case IP_OPTIONS:
1424 #ifdef notyet
1425                 case IP_RETOPTS:
1426 #endif
1427                 {
1428                         struct mbuf *m;
1429                         if (sopt->sopt_valsize > MLEN) {
1430                                 error = EMSGSIZE;
1431                                 break;
1432                         }
1433                         MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_HEADER);
1434                         if (m == 0) {
1435                                 error = ENOBUFS;
1436                                 break;
1437                         }
1438                         m->m_len = sopt->sopt_valsize;
1439                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1440                                             m->m_len);
1441                         
1442                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1443                                            m));
1444                 }
1445
1446                 case IP_TOS:
1447                 case IP_TTL:
1448                 case IP_RECVOPTS:
1449                 case IP_RECVRETOPTS:
1450                 case IP_RECVDSTADDR:
1451                 case IP_RECVIF:
1452                 case IP_FAITH:
1453                         error = sooptcopyin(sopt, &optval, sizeof optval,
1454                                             sizeof optval);
1455                         if (error)
1456                                 break;
1457
1458                         switch (sopt->sopt_name) {
1459                         case IP_TOS:
1460                                 inp->inp_ip_tos = optval;
1461                                 break;
1462
1463                         case IP_TTL:
1464                                 inp->inp_ip_ttl = optval;
1465                                 break;
1466 #define OPTSET(bit) \
1467         if (optval) \
1468                 inp->inp_flags |= bit; \
1469         else \
1470                 inp->inp_flags &= ~bit;
1471
1472                         case IP_RECVOPTS:
1473                                 OPTSET(INP_RECVOPTS);
1474                                 break;
1475
1476                         case IP_RECVRETOPTS:
1477                                 OPTSET(INP_RECVRETOPTS);
1478                                 break;
1479
1480                         case IP_RECVDSTADDR:
1481                                 OPTSET(INP_RECVDSTADDR);
1482                                 break;
1483
1484                         case IP_RECVIF:
1485                                 OPTSET(INP_RECVIF);
1486                                 break;
1487
1488                         case IP_FAITH:
1489                                 OPTSET(INP_FAITH);
1490                                 break;
1491                         }
1492                         break;
1493 #undef OPTSET
1494
1495                 case IP_MULTICAST_IF:
1496                 case IP_MULTICAST_VIF:
1497                 case IP_MULTICAST_TTL:
1498                 case IP_MULTICAST_LOOP:
1499                 case IP_ADD_MEMBERSHIP:
1500                 case IP_DROP_MEMBERSHIP:
1501                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1502                         break;
1503
1504                 case IP_PORTRANGE:
1505                         error = sooptcopyin(sopt, &optval, sizeof optval,
1506                                             sizeof optval);
1507                         if (error)
1508                                 break;
1509
1510                         switch (optval) {
1511                         case IP_PORTRANGE_DEFAULT:
1512                                 inp->inp_flags &= ~(INP_LOWPORT);
1513                                 inp->inp_flags &= ~(INP_HIGHPORT);
1514                                 break;
1515
1516                         case IP_PORTRANGE_HIGH:
1517                                 inp->inp_flags &= ~(INP_LOWPORT);
1518                                 inp->inp_flags |= INP_HIGHPORT;
1519                                 break;
1520
1521                         case IP_PORTRANGE_LOW:
1522                                 inp->inp_flags &= ~(INP_HIGHPORT);
1523                                 inp->inp_flags |= INP_LOWPORT;
1524                                 break;
1525
1526                         default:
1527                                 error = EINVAL;
1528                                 break;
1529                         }
1530                         break;
1531
1532 #if defined(IPSEC) || defined(FAST_IPSEC)
1533                 case IP_IPSEC_POLICY:
1534                 {
1535                         caddr_t req;
1536                         size_t len = 0;
1537                         int priv;
1538                         struct mbuf *m;
1539                         int optname;
1540
1541                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1542                                 break;
1543                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1544                                 break;
1545                         priv = (sopt->sopt_td != NULL &&
1546                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1547                         req = mtod(m, caddr_t);
1548                         len = m->m_len;
1549                         optname = sopt->sopt_name;
1550                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1551                         m_freem(m);
1552                         break;
1553                 }
1554 #endif /*IPSEC*/
1555
1556                 default:
1557                         error = ENOPROTOOPT;
1558                         break;
1559                 }
1560                 break;
1561
1562         case SOPT_GET:
1563                 switch (sopt->sopt_name) {
1564                 case IP_OPTIONS:
1565                 case IP_RETOPTS:
1566                         if (inp->inp_options)
1567                                 error = sooptcopyout(sopt, 
1568                                                      mtod(inp->inp_options,
1569                                                           char *),
1570                                                      inp->inp_options->m_len);
1571                         else
1572                                 sopt->sopt_valsize = 0;
1573                         break;
1574
1575                 case IP_TOS:
1576                 case IP_TTL:
1577                 case IP_RECVOPTS:
1578                 case IP_RECVRETOPTS:
1579                 case IP_RECVDSTADDR:
1580                 case IP_RECVIF:
1581                 case IP_PORTRANGE:
1582                 case IP_FAITH:
1583                         switch (sopt->sopt_name) {
1584
1585                         case IP_TOS:
1586                                 optval = inp->inp_ip_tos;
1587                                 break;
1588
1589                         case IP_TTL:
1590                                 optval = inp->inp_ip_ttl;
1591                                 break;
1592
1593 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1594
1595                         case IP_RECVOPTS:
1596                                 optval = OPTBIT(INP_RECVOPTS);
1597                                 break;
1598
1599                         case IP_RECVRETOPTS:
1600                                 optval = OPTBIT(INP_RECVRETOPTS);
1601                                 break;
1602
1603                         case IP_RECVDSTADDR:
1604                                 optval = OPTBIT(INP_RECVDSTADDR);
1605                                 break;
1606
1607                         case IP_RECVIF:
1608                                 optval = OPTBIT(INP_RECVIF);
1609                                 break;
1610
1611                         case IP_PORTRANGE:
1612                                 if (inp->inp_flags & INP_HIGHPORT)
1613                                         optval = IP_PORTRANGE_HIGH;
1614                                 else if (inp->inp_flags & INP_LOWPORT)
1615                                         optval = IP_PORTRANGE_LOW;
1616                                 else
1617                                         optval = 0;
1618                                 break;
1619
1620                         case IP_FAITH:
1621                                 optval = OPTBIT(INP_FAITH);
1622                                 break;
1623                         }
1624                         error = sooptcopyout(sopt, &optval, sizeof optval);
1625                         break;
1626
1627                 case IP_MULTICAST_IF:
1628                 case IP_MULTICAST_VIF:
1629                 case IP_MULTICAST_TTL:
1630                 case IP_MULTICAST_LOOP:
1631                 case IP_ADD_MEMBERSHIP:
1632                 case IP_DROP_MEMBERSHIP:
1633                         error = ip_getmoptions(sopt, inp->inp_moptions);
1634                         break;
1635
1636 #if defined(IPSEC) || defined(FAST_IPSEC)
1637                 case IP_IPSEC_POLICY:
1638                 {
1639                         struct mbuf *m = NULL;
1640                         caddr_t req = NULL;
1641                         size_t len = 0;
1642
1643                         if (m != 0) {
1644                                 req = mtod(m, caddr_t);
1645                                 len = m->m_len;
1646                         }
1647                         error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1648                         if (error == 0)
1649                                 error = soopt_mcopyout(sopt, m); /* XXX */
1650                         if (error == 0)
1651                                 m_freem(m);
1652                         break;
1653                 }
1654 #endif /*IPSEC*/
1655
1656                 default:
1657                         error = ENOPROTOOPT;
1658                         break;
1659                 }
1660                 break;
1661         }
1662         return (error);
1663 }
1664
1665 /*
1666  * Set up IP options in pcb for insertion in output packets.
1667  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1668  * with destination address if source routed.
1669  */
1670 static int
1671 ip_pcbopts(optname, pcbopt, m)
1672         int optname;
1673         struct mbuf **pcbopt;
1674         struct mbuf *m;
1675 {
1676         int cnt, optlen;
1677         u_char *cp;
1678         u_char opt;
1679
1680         /* turn off any old options */
1681         if (*pcbopt)
1682                 (void)m_free(*pcbopt);
1683         *pcbopt = 0;
1684         if (m == (struct mbuf *)0 || m->m_len == 0) {
1685                 /*
1686                  * Only turning off any previous options.
1687                  */
1688                 if (m)
1689                         (void)m_free(m);
1690                 return (0);
1691         }
1692
1693         if (m->m_len % sizeof(int32_t))
1694                 goto bad;
1695         /*
1696          * IP first-hop destination address will be stored before
1697          * actual options; move other options back
1698          * and clear it when none present.
1699          */
1700         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1701                 goto bad;
1702         cnt = m->m_len;
1703         m->m_len += sizeof(struct in_addr);
1704         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1705         ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
1706         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1707
1708         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1709                 opt = cp[IPOPT_OPTVAL];
1710                 if (opt == IPOPT_EOL)
1711                         break;
1712                 if (opt == IPOPT_NOP)
1713                         optlen = 1;
1714                 else {
1715                         if (cnt < IPOPT_OLEN + sizeof(*cp))
1716                                 goto bad;
1717                         optlen = cp[IPOPT_OLEN];
1718                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt)
1719                                 goto bad;
1720                 }
1721                 switch (opt) {
1722
1723                 default:
1724                         break;
1725
1726                 case IPOPT_LSRR:
1727                 case IPOPT_SSRR:
1728                         /*
1729                          * user process specifies route as:
1730                          *      ->A->B->C->D
1731                          * D must be our final destination (but we can't
1732                          * check that since we may not have connected yet).
1733                          * A is first hop destination, which doesn't appear in
1734                          * actual IP option, but is stored before the options.
1735                          */
1736                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1737                                 goto bad;
1738                         m->m_len -= sizeof(struct in_addr);
1739                         cnt -= sizeof(struct in_addr);
1740                         optlen -= sizeof(struct in_addr);
1741                         cp[IPOPT_OLEN] = optlen;
1742                         /*
1743                          * Move first hop before start of options.
1744                          */
1745                         bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1746                             sizeof(struct in_addr));
1747                         /*
1748                          * Then copy rest of options back
1749                          * to close up the deleted entry.
1750                          */
1751                         ovbcopy((caddr_t)(&cp[IPOPT_OFFSET+1] +
1752                             sizeof(struct in_addr)),
1753                             (caddr_t)&cp[IPOPT_OFFSET+1],
1754                             (unsigned)cnt + sizeof(struct in_addr));
1755                         break;
1756                 }
1757         }
1758         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1759                 goto bad;
1760         *pcbopt = m;
1761         return (0);
1762
1763 bad:
1764         (void)m_free(m);
1765         return (EINVAL);
1766 }
1767
1768 /*
1769  * XXX
1770  * The whole multicast option thing needs to be re-thought.
1771  * Several of these options are equally applicable to non-multicast
1772  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1773  * standard option (IP_TTL).
1774  */
1775
1776 /*
1777  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1778  */
1779 static struct ifnet *
1780 ip_multicast_if(a, ifindexp)
1781         struct in_addr *a;
1782         int *ifindexp;
1783 {
1784         int ifindex;
1785         struct ifnet *ifp;
1786
1787         if (ifindexp)
1788                 *ifindexp = 0;
1789         if (ntohl(a->s_addr) >> 24 == 0) {
1790                 ifindex = ntohl(a->s_addr) & 0xffffff;
1791                 if (ifindex < 0 || if_index < ifindex)
1792                         return NULL;
1793                 ifp = ifindex2ifnet[ifindex];
1794                 if (ifindexp)
1795                         *ifindexp = ifindex;
1796         } else {
1797                 INADDR_TO_IFP(*a, ifp);
1798         }
1799         return ifp;
1800 }
1801
1802 /*
1803  * Set the IP multicast options in response to user setsockopt().
1804  */
1805 static int
1806 ip_setmoptions(sopt, imop)
1807         struct sockopt *sopt;
1808         struct ip_moptions **imop;
1809 {
1810         int error = 0;
1811         int i;
1812         struct in_addr addr;
1813         struct ip_mreq mreq;
1814         struct ifnet *ifp;
1815         struct ip_moptions *imo = *imop;
1816         struct route ro;
1817         struct sockaddr_in *dst;
1818         int ifindex;
1819         int s;
1820
1821         if (imo == NULL) {
1822                 /*
1823                  * No multicast option buffer attached to the pcb;
1824                  * allocate one and initialize to default values.
1825                  */
1826                 imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS,
1827                     M_WAITOK);
1828
1829                 if (imo == NULL)
1830                         return (ENOBUFS);
1831                 *imop = imo;
1832                 imo->imo_multicast_ifp = NULL;
1833                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1834                 imo->imo_multicast_vif = -1;
1835                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1836                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1837                 imo->imo_num_memberships = 0;
1838         }
1839
1840         switch (sopt->sopt_name) {
1841         /* store an index number for the vif you wanna use in the send */
1842         case IP_MULTICAST_VIF:
1843                 if (legal_vif_num == 0) {
1844                         error = EOPNOTSUPP;
1845                         break;
1846                 }
1847                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1848                 if (error)
1849                         break;
1850                 if (!legal_vif_num(i) && (i != -1)) {
1851                         error = EINVAL;
1852                         break;
1853                 }
1854                 imo->imo_multicast_vif = i;
1855                 break;
1856
1857         case IP_MULTICAST_IF:
1858                 /*
1859                  * Select the interface for outgoing multicast packets.
1860                  */
1861                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1862                 if (error)
1863                         break;
1864                 /*
1865                  * INADDR_ANY is used to remove a previous selection.
1866                  * When no interface is selected, a default one is
1867                  * chosen every time a multicast packet is sent.
1868                  */
1869                 if (addr.s_addr == INADDR_ANY) {
1870                         imo->imo_multicast_ifp = NULL;
1871                         break;
1872                 }
1873                 /*
1874                  * The selected interface is identified by its local
1875                  * IP address.  Find the interface and confirm that
1876                  * it supports multicasting.
1877                  */
1878                 s = splimp();
1879                 ifp = ip_multicast_if(&addr, &ifindex);
1880                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1881                         splx(s);
1882                         error = EADDRNOTAVAIL;
1883                         break;
1884                 }
1885                 imo->imo_multicast_ifp = ifp;
1886                 if (ifindex)
1887                         imo->imo_multicast_addr = addr;
1888                 else
1889                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1890                 splx(s);
1891                 break;
1892
1893         case IP_MULTICAST_TTL:
1894                 /*
1895                  * Set the IP time-to-live for outgoing multicast packets.
1896                  * The original multicast API required a char argument,
1897                  * which is inconsistent with the rest of the socket API.
1898                  * We allow either a char or an int.
1899                  */
1900                 if (sopt->sopt_valsize == 1) {
1901                         u_char ttl;
1902                         error = sooptcopyin(sopt, &ttl, 1, 1);
1903                         if (error)
1904                                 break;
1905                         imo->imo_multicast_ttl = ttl;
1906                 } else {
1907                         u_int ttl;
1908                         error = sooptcopyin(sopt, &ttl, sizeof ttl, 
1909                                             sizeof ttl);
1910                         if (error)
1911                                 break;
1912                         if (ttl > 255)
1913                                 error = EINVAL;
1914                         else
1915                                 imo->imo_multicast_ttl = ttl;
1916                 }
1917                 break;
1918
1919         case IP_MULTICAST_LOOP:
1920                 /*
1921                  * Set the loopback flag for outgoing multicast packets.
1922                  * Must be zero or one.  The original multicast API required a
1923                  * char argument, which is inconsistent with the rest
1924                  * of the socket API.  We allow either a char or an int.
1925                  */
1926                 if (sopt->sopt_valsize == 1) {
1927                         u_char loop;
1928                         error = sooptcopyin(sopt, &loop, 1, 1);
1929                         if (error)
1930                                 break;
1931                         imo->imo_multicast_loop = !!loop;
1932                 } else {
1933                         u_int loop;
1934                         error = sooptcopyin(sopt, &loop, sizeof loop,
1935                                             sizeof loop);
1936                         if (error)
1937                                 break;
1938                         imo->imo_multicast_loop = !!loop;
1939                 }
1940                 break;
1941
1942         case IP_ADD_MEMBERSHIP:
1943                 /*
1944                  * Add a multicast group membership.
1945                  * Group must be a valid IP multicast address.
1946                  */
1947                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1948                 if (error)
1949                         break;
1950
1951                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1952                         error = EINVAL;
1953                         break;
1954                 }
1955                 s = splimp();
1956                 /*
1957                  * If no interface address was provided, use the interface of
1958                  * the route to the given multicast address.
1959                  */
1960                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1961                         bzero((caddr_t)&ro, sizeof(ro));
1962                         dst = (struct sockaddr_in *)&ro.ro_dst;
1963                         dst->sin_len = sizeof(*dst);
1964                         dst->sin_family = AF_INET;
1965                         dst->sin_addr = mreq.imr_multiaddr;
1966                         rtalloc(&ro);
1967                         if (ro.ro_rt == NULL) {
1968                                 error = EADDRNOTAVAIL;
1969                                 splx(s);
1970                                 break;
1971                         }
1972                         ifp = ro.ro_rt->rt_ifp;
1973                         rtfree(ro.ro_rt);
1974                 }
1975                 else {
1976                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1977                 }
1978
1979                 /*
1980                  * See if we found an interface, and confirm that it
1981                  * supports multicast.
1982                  */
1983                 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
1984                         error = EADDRNOTAVAIL;
1985                         splx(s);
1986                         break;
1987                 }
1988                 /*
1989                  * See if the membership already exists or if all the
1990                  * membership slots are full.
1991                  */
1992                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1993                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1994                             imo->imo_membership[i]->inm_addr.s_addr
1995                                                 == mreq.imr_multiaddr.s_addr)
1996                                 break;
1997                 }
1998                 if (i < imo->imo_num_memberships) {
1999                         error = EADDRINUSE;
2000                         splx(s);
2001                         break;
2002                 }
2003                 if (i == IP_MAX_MEMBERSHIPS) {
2004                         error = ETOOMANYREFS;
2005                         splx(s);
2006                         break;
2007                 }
2008                 /*
2009                  * Everything looks good; add a new record to the multicast
2010                  * address list for the given interface.
2011                  */
2012                 if ((imo->imo_membership[i] =
2013                     in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2014                         error = ENOBUFS;
2015                         splx(s);
2016                         break;
2017                 }
2018                 ++imo->imo_num_memberships;
2019                 splx(s);
2020                 break;
2021
2022         case IP_DROP_MEMBERSHIP:
2023                 /*
2024                  * Drop a multicast group membership.
2025                  * Group must be a valid IP multicast address.
2026                  */
2027                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2028                 if (error)
2029                         break;
2030
2031                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2032                         error = EINVAL;
2033                         break;
2034                 }
2035
2036                 s = splimp();
2037                 /*
2038                  * If an interface address was specified, get a pointer
2039                  * to its ifnet structure.
2040                  */
2041                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2042                         ifp = NULL;
2043                 else {
2044                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2045                         if (ifp == NULL) {
2046                                 error = EADDRNOTAVAIL;
2047                                 splx(s);
2048                                 break;
2049                         }
2050                 }
2051                 /*
2052                  * Find the membership in the membership array.
2053                  */
2054                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2055                         if ((ifp == NULL ||
2056                              imo->imo_membership[i]->inm_ifp == ifp) &&
2057                              imo->imo_membership[i]->inm_addr.s_addr ==
2058                              mreq.imr_multiaddr.s_addr)
2059                                 break;
2060                 }
2061                 if (i == imo->imo_num_memberships) {
2062                         error = EADDRNOTAVAIL;
2063                         splx(s);
2064                         break;
2065                 }
2066                 /*
2067                  * Give up the multicast address record to which the
2068                  * membership points.
2069                  */
2070                 in_delmulti(imo->imo_membership[i]);
2071                 /*
2072                  * Remove the gap in the membership array.
2073                  */
2074                 for (++i; i < imo->imo_num_memberships; ++i)
2075                         imo->imo_membership[i-1] = imo->imo_membership[i];
2076                 --imo->imo_num_memberships;
2077                 splx(s);
2078                 break;
2079
2080         default:
2081                 error = EOPNOTSUPP;
2082                 break;
2083         }
2084
2085         /*
2086          * If all options have default values, no need to keep the mbuf.
2087          */
2088         if (imo->imo_multicast_ifp == NULL &&
2089             imo->imo_multicast_vif == -1 &&
2090             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2091             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2092             imo->imo_num_memberships == 0) {
2093                 free(*imop, M_IPMOPTS);
2094                 *imop = NULL;
2095         }
2096
2097         return (error);
2098 }
2099
2100 /*
2101  * Return the IP multicast options in response to user getsockopt().
2102  */
2103 static int
2104 ip_getmoptions(sopt, imo)
2105         struct sockopt *sopt;
2106         struct ip_moptions *imo;
2107 {
2108         struct in_addr addr;
2109         struct in_ifaddr *ia;
2110         int error, optval;
2111         u_char coptval;
2112
2113         error = 0;
2114         switch (sopt->sopt_name) {
2115         case IP_MULTICAST_VIF: 
2116                 if (imo != NULL)
2117                         optval = imo->imo_multicast_vif;
2118                 else
2119                         optval = -1;
2120                 error = sooptcopyout(sopt, &optval, sizeof optval);
2121                 break;
2122
2123         case IP_MULTICAST_IF:
2124                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2125                         addr.s_addr = INADDR_ANY;
2126                 else if (imo->imo_multicast_addr.s_addr) {
2127                         /* return the value user has set */
2128                         addr = imo->imo_multicast_addr;
2129                 } else {
2130                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2131                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2132                                 : IA_SIN(ia)->sin_addr.s_addr;
2133                 }
2134                 error = sooptcopyout(sopt, &addr, sizeof addr);
2135                 break;
2136
2137         case IP_MULTICAST_TTL:
2138                 if (imo == 0)
2139                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2140                 else
2141                         optval = coptval = imo->imo_multicast_ttl;
2142                 if (sopt->sopt_valsize == 1)
2143                         error = sooptcopyout(sopt, &coptval, 1);
2144                 else
2145                         error = sooptcopyout(sopt, &optval, sizeof optval);
2146                 break;
2147
2148         case IP_MULTICAST_LOOP:
2149                 if (imo == 0)
2150                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2151                 else
2152                         optval = coptval = imo->imo_multicast_loop;
2153                 if (sopt->sopt_valsize == 1)
2154                         error = sooptcopyout(sopt, &coptval, 1);
2155                 else
2156                         error = sooptcopyout(sopt, &optval, sizeof optval);
2157                 break;
2158
2159         default:
2160                 error = ENOPROTOOPT;
2161                 break;
2162         }
2163         return (error);
2164 }
2165
2166 /*
2167  * Discard the IP multicast options.
2168  */
2169 void
2170 ip_freemoptions(imo)
2171         struct ip_moptions *imo;
2172 {
2173         int i;
2174
2175         if (imo != NULL) {
2176                 for (i = 0; i < imo->imo_num_memberships; ++i)
2177                         in_delmulti(imo->imo_membership[i]);
2178                 free(imo, M_IPMOPTS);
2179         }
2180 }
2181
2182 /*
2183  * Routine called from ip_output() to loop back a copy of an IP multicast
2184  * packet to the input queue of a specified interface.  Note that this
2185  * calls the output routine of the loopback "driver", but with an interface
2186  * pointer that might NOT be a loopback interface -- evil, but easier than
2187  * replicating that code here.
2188  */
2189 static void
2190 ip_mloopback(ifp, m, dst, hlen)
2191         struct ifnet *ifp;
2192         struct mbuf *m;
2193         struct sockaddr_in *dst;
2194         int hlen;
2195 {
2196         struct ip *ip;
2197         struct mbuf *copym;
2198
2199         copym = m_copy(m, 0, M_COPYALL);
2200         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2201                 copym = m_pullup(copym, hlen);
2202         if (copym != NULL) {
2203                 /*
2204                  * We don't bother to fragment if the IP length is greater
2205                  * than the interface's MTU.  Can this possibly matter?
2206                  */
2207                 ip = mtod(copym, struct ip *);
2208                 ip->ip_len = htons(ip->ip_len);
2209                 ip->ip_off = htons(ip->ip_off);
2210                 ip->ip_sum = 0;
2211                 if (ip->ip_vhl == IP_VHL_BORING) {
2212                         ip->ip_sum = in_cksum_hdr(ip);
2213                 } else {
2214                         ip->ip_sum = in_cksum(copym, hlen);
2215                 }
2216                 /*
2217                  * NB:
2218                  * It's not clear whether there are any lingering
2219                  * reentrancy problems in other areas which might
2220                  * be exposed by using ip_input directly (in
2221                  * particular, everything which modifies the packet
2222                  * in-place).  Yet another option is using the
2223                  * protosw directly to deliver the looped back
2224                  * packet.  For the moment, we'll err on the side
2225                  * of safety by using if_simloop().
2226                  */
2227 #if 1 /* XXX */
2228                 if (dst->sin_family != AF_INET) {
2229                         printf("ip_mloopback: bad address family %d\n",
2230                                                 dst->sin_family);
2231                         dst->sin_family = AF_INET;
2232                 }
2233 #endif
2234
2235 #ifdef notdef
2236                 copym->m_pkthdr.rcvif = ifp;
2237                 ip_input(copym);
2238 #else
2239                 /* if the checksum hasn't been computed, mark it as valid */
2240                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2241                         copym->m_pkthdr.csum_flags |=
2242                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2243                         copym->m_pkthdr.csum_data = 0xffff;
2244                 }
2245                 if_simloop(ifp, copym, dst->sin_family, 0);
2246 #endif
2247         }
2248 }