cab77a7fd1edfc56b529b7cdca2cc47c848f86b0
[dragonfly.git] / sys / netinet / ip_output.c
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.49 2008/08/22 09:14:16 sephe Exp $
32  */
33
34 #define _IP_VHL
35
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipfilter.h"
40 #include "opt_ipsec.h"
41 #include "opt_mbuf_stress_test.h"
42 #include "opt_mpls.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/thread2.h>
55 #include <sys/in_cksum.h>
56
57 #include <net/if.h>
58 #include <net/netisr.h>
59 #include <net/pfil.h>
60 #include <net/route.h>
61
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip_var.h>
68
69 #include <netproto/mpls/mpls_var.h>
70
71 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
72
73 #ifdef IPSEC
74 #include <netinet6/ipsec.h>
75 #include <netproto/key/key.h>
76 #ifdef IPSEC_DEBUG
77 #include <netproto/key/key_debug.h>
78 #else
79 #define KEYDEBUG(lev,arg)
80 #endif
81 #endif /*IPSEC*/
82
83 #ifdef FAST_IPSEC
84 #include <netproto/ipsec/ipsec.h>
85 #include <netproto/ipsec/xform.h>
86 #include <netproto/ipsec/key.h>
87 #endif /*FAST_IPSEC*/
88
89 #include <net/ipfw/ip_fw.h>
90 #include <net/dummynet/ip_dummynet.h>
91
92 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
93                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
94                                   (ntohl(a.s_addr)>>16)&0xFF,\
95                                   (ntohl(a.s_addr)>>8)&0xFF,\
96                                   (ntohl(a.s_addr))&0xFF, y);
97
98 u_short ip_id;
99
100 #ifdef MBUF_STRESS_TEST
101 int mbuf_frag_size = 0;
102 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
103         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
104 #endif
105
106 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
107 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
108 static void     ip_mloopback
109         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
110 static int      ip_getmoptions
111         (struct sockopt *, struct ip_moptions *);
112 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
113 static int      ip_setmoptions
114         (struct sockopt *, struct ip_moptions **);
115
116 int     ip_optcopy(struct ip *, struct ip *);
117
118
119 extern  struct protosw inetsw[];
120
121 /*
122  * IP output.  The packet in mbuf chain m contains a skeletal IP
123  * header (with len, off, ttl, proto, tos, src, dst).
124  * The mbuf chain containing the packet will be freed.
125  * The mbuf opt, if present, will not be freed.
126  */
127 int
128 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
129           int flags, struct ip_moptions *imo, struct inpcb *inp)
130 {
131         struct ip *ip;
132         struct ifnet *ifp = NULL;       /* keep compiler happy */
133         struct mbuf *m;
134         int hlen = sizeof(struct ip);
135         int len, off, error = 0;
136         struct sockaddr_in *dst = NULL; /* keep compiler happy */
137         struct in_ifaddr *ia = NULL;
138         int isbroadcast, sw_csum;
139         struct in_addr pkt_dst;
140         struct route iproute;
141         struct m_tag *dn_mtag, *mtag;
142 #ifdef IPSEC
143         struct secpolicy *sp = NULL;
144         struct socket *so = inp ? inp->inp_socket : NULL;
145 #endif
146 #ifdef FAST_IPSEC
147         struct secpolicy *sp = NULL;
148         struct tdb_ident *tdbi;
149 #endif /* FAST_IPSEC */
150         struct ip_fw_args args;
151         struct sockaddr_in *next_hop = NULL;
152         int src_was_INADDR_ANY = 0;     /* as the name says... */
153
154         args.eh = NULL;
155         args.rule = NULL;
156
157         m = m0;
158         M_ASSERTPKTHDR(m);
159
160         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
161                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
162                 KKASSERT(mtag != NULL);
163                 next_hop = m_tag_data(mtag);
164         }
165
166         /* Extract info from dummynet tag */
167         dn_mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
168         if (dn_mtag != NULL) {
169                 struct dn_pkt *dn_pkt = m_tag_data(dn_mtag);
170
171                 /*
172                  * The packet was already tagged, so part of the
173                  * processing was already done, and we need to go down.
174                  * Get parameters from the tag.
175                  */
176                 args.rule = dn_pkt->dn_priv;
177                 opt = NULL;
178                 ro = &dn_pkt->ro;
179                 imo = NULL;
180                 dst = dn_pkt->dn_dst;
181                 ifp = dn_pkt->ifp;
182                 flags = dn_pkt->flags;
183
184                 /*
185                  * Don't delete the dummynet tag here, just unlink it,
186                  * since some local variables (like 'ro' and 'dst') are
187                  * still referencing certain parts of it.
188                  * The dummynet tag will be freed at the end of the
189                  * output process.
190                  */
191                 m_tag_unlink(m, dn_mtag);
192         }
193
194         if (ro == NULL) {
195                 ro = &iproute;
196                 bzero(ro, sizeof *ro);
197         }
198
199         if (args.rule != NULL) {        /* dummynet already saw us */
200                 ip = mtod(m, struct ip *);
201                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
202                 if (ro->ro_rt)
203                         ia = ifatoia(ro->ro_rt->rt_ifa);
204                 goto sendit;
205         }
206
207         if (opt) {
208                 len = 0;
209                 m = ip_insertoptions(m, opt, &len);
210                 if (len != 0)
211                         hlen = len;
212         }
213         ip = mtod(m, struct ip *);
214         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
215
216         /*
217          * Fill in IP header.
218          */
219         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
220                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
221                 ip->ip_off &= IP_DF;
222                 ip->ip_id = ip_newid();
223                 ipstat.ips_localout++;
224         } else {
225                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
226         }
227
228         dst = (struct sockaddr_in *)&ro->ro_dst;
229         /*
230          * If there is a cached route,
231          * check that it is to the same destination
232          * and is still up.  If not, free it and try again.
233          * The address family should also be checked in case of sharing the
234          * cache with IPv6.
235          */
236         if (ro->ro_rt &&
237             (!(ro->ro_rt->rt_flags & RTF_UP) ||
238              dst->sin_family != AF_INET ||
239              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
240                 rtfree(ro->ro_rt);
241                 ro->ro_rt = (struct rtentry *)NULL;
242         }
243         if (ro->ro_rt == NULL) {
244                 bzero(dst, sizeof *dst);
245                 dst->sin_family = AF_INET;
246                 dst->sin_len = sizeof *dst;
247                 dst->sin_addr = pkt_dst;
248         }
249         /*
250          * If routing to interface only,
251          * short circuit routing lookup.
252          */
253         if (flags & IP_ROUTETOIF) {
254                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
255                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
256                         ipstat.ips_noroute++;
257                         error = ENETUNREACH;
258                         goto bad;
259                 }
260                 ifp = ia->ia_ifp;
261                 ip->ip_ttl = 1;
262                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
263         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
264                    imo != NULL && imo->imo_multicast_ifp != NULL) {
265                 /*
266                  * Bypass the normal routing lookup for multicast
267                  * packets if the interface is specified.
268                  */
269                 ifp = imo->imo_multicast_ifp;
270                 ia = IFP_TO_IA(ifp);
271                 isbroadcast = 0;        /* fool gcc */
272         } else {
273                 /*
274                  * If this is the case, we probably don't want to allocate
275                  * a protocol-cloned route since we didn't get one from the
276                  * ULP.  This lets TCP do its thing, while not burdening
277                  * forwarding or ICMP with the overhead of cloning a route.
278                  * Of course, we still want to do any cloning requested by
279                  * the link layer, as this is probably required in all cases
280                  * for correct operation (as it is for ARP).
281                  */
282                 if (ro->ro_rt == NULL)
283                         rtalloc_ign(ro, RTF_PRCLONING);
284                 if (ro->ro_rt == NULL) {
285                         ipstat.ips_noroute++;
286                         error = EHOSTUNREACH;
287                         goto bad;
288                 }
289                 ia = ifatoia(ro->ro_rt->rt_ifa);
290                 ifp = ro->ro_rt->rt_ifp;
291                 ro->ro_rt->rt_use++;
292                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
293                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
294                 if (ro->ro_rt->rt_flags & RTF_HOST)
295                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
296                 else
297                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
298         }
299         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
300                 struct in_multi *inm;
301
302                 m->m_flags |= M_MCAST;
303                 /*
304                  * IP destination address is multicast.  Make sure "dst"
305                  * still points to the address in "ro".  (It may have been
306                  * changed to point to a gateway address, above.)
307                  */
308                 dst = (struct sockaddr_in *)&ro->ro_dst;
309                 /*
310                  * See if the caller provided any multicast options
311                  */
312                 if (imo != NULL) {
313                         ip->ip_ttl = imo->imo_multicast_ttl;
314                         if (imo->imo_multicast_vif != -1)
315                                 ip->ip_src.s_addr =
316                                     ip_mcast_src ?
317                                     ip_mcast_src(imo->imo_multicast_vif) :
318                                     INADDR_ANY;
319                 } else
320                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
321                 /*
322                  * Confirm that the outgoing interface supports multicast.
323                  */
324                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
325                         if (!(ifp->if_flags & IFF_MULTICAST)) {
326                                 ipstat.ips_noroute++;
327                                 error = ENETUNREACH;
328                                 goto bad;
329                         }
330                 }
331                 /*
332                  * If source address not specified yet, use address
333                  * of outgoing interface.
334                  */
335                 if (ip->ip_src.s_addr == INADDR_ANY) {
336                         /* Interface may have no addresses. */
337                         if (ia != NULL)
338                                 ip->ip_src = IA_SIN(ia)->sin_addr;
339                 }
340
341                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
342                 if (inm != NULL &&
343                    (imo == NULL || imo->imo_multicast_loop)) {
344                         /*
345                          * If we belong to the destination multicast group
346                          * on the outgoing interface, and the caller did not
347                          * forbid loopback, loop back a copy.
348                          */
349                         ip_mloopback(ifp, m, dst, hlen);
350                 }
351                 else {
352                         /*
353                          * If we are acting as a multicast router, perform
354                          * multicast forwarding as if the packet had just
355                          * arrived on the interface to which we are about
356                          * to send.  The multicast forwarding function
357                          * recursively calls this function, using the
358                          * IP_FORWARDING flag to prevent infinite recursion.
359                          *
360                          * Multicasts that are looped back by ip_mloopback(),
361                          * above, will be forwarded by the ip_input() routine,
362                          * if necessary.
363                          */
364                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
365                                 /*
366                                  * If rsvp daemon is not running, do not
367                                  * set ip_moptions. This ensures that the packet
368                                  * is multicast and not just sent down one link
369                                  * as prescribed by rsvpd.
370                                  */
371                                 if (!rsvp_on)
372                                         imo = NULL;
373                                 if (ip_mforward &&
374                                     ip_mforward(ip, ifp, m, imo) != 0) {
375                                         m_freem(m);
376                                         goto done;
377                                 }
378                         }
379                 }
380
381                 /*
382                  * Multicasts with a time-to-live of zero may be looped-
383                  * back, above, but must not be transmitted on a network.
384                  * Also, multicasts addressed to the loopback interface
385                  * are not sent -- the above call to ip_mloopback() will
386                  * loop back a copy if this host actually belongs to the
387                  * destination group on the loopback interface.
388                  */
389                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
390                         m_freem(m);
391                         goto done;
392                 }
393
394                 goto sendit;
395         }
396 #ifndef notdef
397         /*
398          * If the source address is not specified yet, use the address
399          * of the outoing interface. In case, keep note we did that, so
400          * if the the firewall changes the next-hop causing the output
401          * interface to change, we can fix that.
402          */
403         if (ip->ip_src.s_addr == INADDR_ANY) {
404                 /* Interface may have no addresses. */
405                 if (ia != NULL) {
406                         ip->ip_src = IA_SIN(ia)->sin_addr;
407                         src_was_INADDR_ANY = 1;
408                 }
409         }
410 #endif /* notdef */
411 #ifdef ALTQ
412         /*
413          * Disable packet drop hack.
414          * Packetdrop should be done by queueing.
415          */
416 #else /* !ALTQ */
417         /*
418          * Verify that we have any chance at all of being able to queue
419          *      the packet or packet fragments
420          */
421         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
422                 ifp->if_snd.ifq_maxlen) {
423                         error = ENOBUFS;
424                         ipstat.ips_odropped++;
425                         goto bad;
426         }
427 #endif /* !ALTQ */
428
429         /*
430          * Look for broadcast address and
431          * verify user is allowed to send
432          * such a packet.
433          */
434         if (isbroadcast) {
435                 if (!(ifp->if_flags & IFF_BROADCAST)) {
436                         error = EADDRNOTAVAIL;
437                         goto bad;
438                 }
439                 if (!(flags & IP_ALLOWBROADCAST)) {
440                         error = EACCES;
441                         goto bad;
442                 }
443                 /* don't allow broadcast messages to be fragmented */
444                 if (ip->ip_len > ifp->if_mtu) {
445                         error = EMSGSIZE;
446                         goto bad;
447                 }
448                 m->m_flags |= M_BCAST;
449         } else {
450                 m->m_flags &= ~M_BCAST;
451         }
452
453 sendit:
454 #ifdef IPSEC
455         /* get SP for this packet */
456         if (so == NULL)
457                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
458         else
459                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
460
461         if (sp == NULL) {
462                 ipsecstat.out_inval++;
463                 goto bad;
464         }
465
466         error = 0;
467
468         /* check policy */
469         switch (sp->policy) {
470         case IPSEC_POLICY_DISCARD:
471                 /*
472                  * This packet is just discarded.
473                  */
474                 ipsecstat.out_polvio++;
475                 goto bad;
476
477         case IPSEC_POLICY_BYPASS:
478         case IPSEC_POLICY_NONE:
479                 /* no need to do IPsec. */
480                 goto skip_ipsec;
481
482         case IPSEC_POLICY_IPSEC:
483                 if (sp->req == NULL) {
484                         /* acquire a policy */
485                         error = key_spdacquire(sp);
486                         goto bad;
487                 }
488                 break;
489
490         case IPSEC_POLICY_ENTRUST:
491         default:
492                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
493         }
494     {
495         struct ipsec_output_state state;
496         bzero(&state, sizeof state);
497         state.m = m;
498         if (flags & IP_ROUTETOIF) {
499                 state.ro = &iproute;
500                 bzero(&iproute, sizeof iproute);
501         } else
502                 state.ro = ro;
503         state.dst = (struct sockaddr *)dst;
504
505         ip->ip_sum = 0;
506
507         /*
508          * XXX
509          * delayed checksums are not currently compatible with IPsec
510          */
511         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
512                 in_delayed_cksum(m);
513                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
514         }
515
516         ip->ip_len = htons(ip->ip_len);
517         ip->ip_off = htons(ip->ip_off);
518
519         error = ipsec4_output(&state, sp, flags);
520
521         m = state.m;
522         if (flags & IP_ROUTETOIF) {
523                 /*
524                  * if we have tunnel mode SA, we may need to ignore
525                  * IP_ROUTETOIF.
526                  */
527                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
528                         flags &= ~IP_ROUTETOIF;
529                         ro = state.ro;
530                 }
531         } else
532                 ro = state.ro;
533         dst = (struct sockaddr_in *)state.dst;
534         if (error) {
535                 /* mbuf is already reclaimed in ipsec4_output. */
536                 m0 = NULL;
537                 switch (error) {
538                 case EHOSTUNREACH:
539                 case ENETUNREACH:
540                 case EMSGSIZE:
541                 case ENOBUFS:
542                 case ENOMEM:
543                         break;
544                 default:
545                         kprintf("ip4_output (ipsec): error code %d\n", error);
546                         /*fall through*/
547                 case ENOENT:
548                         /* don't show these error codes to the user */
549                         error = 0;
550                         break;
551                 }
552                 goto bad;
553         }
554     }
555
556         /* be sure to update variables that are affected by ipsec4_output() */
557         ip = mtod(m, struct ip *);
558 #ifdef _IP_VHL
559         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
560 #else
561         hlen = ip->ip_hl << 2;
562 #endif
563         if (ro->ro_rt == NULL) {
564                 if (!(flags & IP_ROUTETOIF)) {
565                         kprintf("ip_output: "
566                                 "can't update route after IPsec processing\n");
567                         error = EHOSTUNREACH;   /*XXX*/
568                         goto bad;
569                 }
570         } else {
571                 ia = ifatoia(ro->ro_rt->rt_ifa);
572                 ifp = ro->ro_rt->rt_ifp;
573         }
574
575         /* make it flipped, again. */
576         ip->ip_len = ntohs(ip->ip_len);
577         ip->ip_off = ntohs(ip->ip_off);
578 skip_ipsec:
579 #endif /*IPSEC*/
580 #ifdef FAST_IPSEC
581         /*
582          * Check the security policy (SP) for the packet and, if
583          * required, do IPsec-related processing.  There are two
584          * cases here; the first time a packet is sent through
585          * it will be untagged and handled by ipsec4_checkpolicy.
586          * If the packet is resubmitted to ip_output (e.g. after
587          * AH, ESP, etc. processing), there will be a tag to bypass
588          * the lookup and related policy checking.
589          */
590         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
591         crit_enter();
592         if (mtag != NULL) {
593                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
594                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
595                 if (sp == NULL)
596                         error = -EINVAL;        /* force silent drop */
597                 m_tag_delete(m, mtag);
598         } else {
599                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
600                                         &error, inp);
601         }
602         /*
603          * There are four return cases:
604          *    sp != NULL                    apply IPsec policy
605          *    sp == NULL, error == 0        no IPsec handling needed
606          *    sp == NULL, error == -EINVAL  discard packet w/o error
607          *    sp == NULL, error != 0        discard packet, report error
608          */
609         if (sp != NULL) {
610                 /* Loop detection, check if ipsec processing already done */
611                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
612                 for (mtag = m_tag_first(m); mtag != NULL;
613                      mtag = m_tag_next(m, mtag)) {
614                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
615                                 continue;
616                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
617                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
618                                 continue;
619                         /*
620                          * Check if policy has an SA associated with it.
621                          * This can happen when an SP has yet to acquire
622                          * an SA; e.g. on first reference.  If it occurs,
623                          * then we let ipsec4_process_packet do its thing.
624                          */
625                         if (sp->req->sav == NULL)
626                                 break;
627                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
628                         if (tdbi->spi == sp->req->sav->spi &&
629                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
630                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
631                                  sizeof(union sockaddr_union)) == 0) {
632                                 /*
633                                  * No IPsec processing is needed, free
634                                  * reference to SP.
635                                  *
636                                  * NB: null pointer to avoid free at
637                                  *     done: below.
638                                  */
639                                 KEY_FREESP(&sp), sp = NULL;
640                                 crit_exit();
641                                 goto spd_done;
642                         }
643                 }
644
645                 /*
646                  * Do delayed checksums now because we send before
647                  * this is done in the normal processing path.
648                  */
649                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
650                         in_delayed_cksum(m);
651                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
652                 }
653
654                 ip->ip_len = htons(ip->ip_len);
655                 ip->ip_off = htons(ip->ip_off);
656
657                 /* NB: callee frees mbuf */
658                 error = ipsec4_process_packet(m, sp->req, flags, 0);
659                 /*
660                  * Preserve KAME behaviour: ENOENT can be returned
661                  * when an SA acquire is in progress.  Don't propagate
662                  * this to user-level; it confuses applications.
663                  *
664                  * XXX this will go away when the SADB is redone.
665                  */
666                 if (error == ENOENT)
667                         error = 0;
668                 crit_exit();
669                 goto done;
670         } else {
671                 crit_exit();
672
673                 if (error != 0) {
674                         /*
675                          * Hack: -EINVAL is used to signal that a packet
676                          * should be silently discarded.  This is typically
677                          * because we asked key management for an SA and
678                          * it was delayed (e.g. kicked up to IKE).
679                          */
680                         if (error == -EINVAL)
681                                 error = 0;
682                         goto bad;
683                 } else {
684                         /* No IPsec processing for this packet. */
685                 }
686 #ifdef notyet
687                 /*
688                  * If deferred crypto processing is needed, check that
689                  * the interface supports it.
690                  */
691                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
692                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
693                         /* notify IPsec to do its own crypto */
694                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
695                         error = EHOSTUNREACH;
696                         goto bad;
697                 }
698 #endif
699         }
700 spd_done:
701 #endif /* FAST_IPSEC */
702         /*
703          * IpHack's section.
704          * - Xlate: translate packet's addr/port (NAT).
705          * - Firewall: deny/allow/etc.
706          * - Wrap: fake packet's addr/port <unimpl.>
707          * - Encapsulate: put it in another IP and send out. <unimp.>
708          */
709
710         /*
711          * Run through list of hooks for output packets.
712          */
713         if (pfil_has_hooks(&inet_pfil_hook)) {
714                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
715                 if (error != 0 || m == NULL)
716                         goto done;
717                 ip = mtod(m, struct ip *);
718         }
719
720         /*
721          * Check with the firewall...
722          * but not if we are already being fwd'd from a firewall.
723          */
724         if (fw_enable && IPFW_LOADED && !next_hop) {
725                 struct sockaddr_in *old = dst;
726
727                 args.m = m;
728                 args.oif = ifp;
729                 off = ip_fw_chk_ptr(&args);
730                 m = args.m;
731
732                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
733                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
734                         KKASSERT(mtag != NULL);
735                         next_hop = m_tag_data(mtag);
736                         dst = next_hop;
737                 }
738
739                 /*
740                  * On return we must do the following:
741                  * m == NULL    -> drop the pkt (old interface, deprecated)
742                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
743                  * 1<=off<= 0xffff              -> DIVERT
744                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
745                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
746                  * dst != old                   -> IPFIREWALL_FORWARD
747                  * off==0, dst==old             -> accept
748                  * If some of the above modules are not compiled in, then
749                  * we should't have to check the corresponding condition
750                  * (because the ipfw control socket should not accept
751                  * unsupported rules), but better play safe and drop
752                  * packets in case of doubt.
753                  */
754                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
755                         if (m)
756                                 m_freem(m);
757                         error = EACCES;
758                         goto done;
759                 }
760                 ip = mtod(m, struct ip *);
761                 if (off == 0 && dst == old)             /* common case */
762                         goto pass;
763                 if (off & IP_FW_PORT_DYNT_FLAG) {
764                         /*
765                          * pass the pkt to dummynet. Need to include
766                          * pipe number, m, ifp, ro, dst because these are
767                          * not recomputed in the next pass.
768                          * All other parameters have been already used and
769                          * so they are not needed anymore.
770                          * XXX note: if the ifp or ro entry are deleted
771                          * while a pkt is in dummynet, we are in trouble!
772                          */
773                         args.ro = ro;
774                         args.dst = dst;
775                         args.flags = flags;
776
777                         error = 0;
778                         ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
779                         goto done;
780                 }
781 #ifdef IPDIVERT
782                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
783                         struct mbuf *clone = NULL;
784
785                         /* Clone packet if we're doing a 'tee' */
786                         if ((off & IP_FW_PORT_TEE_FLAG))
787                                 clone = m_dup(m, MB_DONTWAIT);
788
789                         /*
790                          * XXX
791                          * delayed checksums are not currently compatible
792                          * with divert sockets.
793                          */
794                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
795                                 in_delayed_cksum(m);
796                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
797                         }
798
799                         /* Restore packet header fields to original values */
800                         ip->ip_len = htons(ip->ip_len);
801                         ip->ip_off = htons(ip->ip_off);
802
803                         /* Deliver packet to divert input routine */
804                         divert_packet(m, 0, off & 0xffff);
805
806                         /* If 'tee', continue with original packet */
807                         if (clone != NULL) {
808                                 m = clone;
809                                 ip = mtod(m, struct ip *);
810                                 goto pass;
811                         }
812                         goto done;
813                 }
814 #endif
815
816                 /* IPFIREWALL_FORWARD */
817                 /*
818                  * Check dst to make sure it is directly reachable on the
819                  * interface we previously thought it was.
820                  * If it isn't (which may be likely in some situations) we have
821                  * to re-route it (ie, find a route for the next-hop and the
822                  * associated interface) and set them here. This is nested
823                  * forwarding which in most cases is undesirable, except where
824                  * such control is nigh impossible. So we do it here.
825                  * And I'm babbling.
826                  */
827                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
828 #if 0
829                         /*
830                          * XXX To improve readability, this block should be
831                          * changed into a function call as below:
832                          */
833                         error = ip_ipforward(&m, &dst, &ifp);
834                         if (error)
835                                 goto bad;
836                         if (m == NULL) /* ip_input consumed the mbuf */
837                                 goto done;
838 #else
839                         struct in_ifaddr *ia;
840                         struct in_ifaddr_container *iac;
841
842                         /*
843                          * XXX sro_fwd below is static, and a pointer
844                          * to it gets passed to routines downstream.
845                          * This could have surprisingly bad results in
846                          * practice, because its content is overwritten
847                          * by subsequent packets.
848                          */
849                         /* There must be a better way to do this next line... */
850                         static struct route sro_fwd;
851                         struct route *ro_fwd = &sro_fwd;
852
853 #if 0
854                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
855                             dst->sin_addr, "\n");
856 #endif
857
858                         /*
859                          * We need to figure out if we have been forwarded
860                          * to a local socket. If so, then we should somehow
861                          * "loop back" to ip_input, and get directed to the
862                          * PCB as if we had received this packet. This is
863                          * because it may be dificult to identify the packets
864                          * you want to forward until they are being output
865                          * and have selected an interface. (e.g. locally
866                          * initiated packets) If we used the loopback inteface,
867                          * we would not be able to control what happens
868                          * as the packet runs through ip_input() as
869                          * it is done through a ISR.
870                          */
871                         ia = NULL;
872                         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr),
873                                      ia_hash) {
874                                 /*
875                                  * If the addr to forward to is one
876                                  * of ours, we pretend to
877                                  * be the destination for this packet.
878                                  */
879                                 if (IA_SIN(iac->ia)->sin_addr.s_addr ==
880                                     dst->sin_addr.s_addr) {
881                                         ia = iac->ia;
882                                         break;
883                                 }
884                         }
885                         if (ia != NULL) {    /* tell ip_input "dont filter" */
886                                 if (m->m_pkthdr.rcvif == NULL)
887                                         m->m_pkthdr.rcvif = ifunit("lo0");
888                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
889                                         m->m_pkthdr.csum_flags |=
890                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
891                                         m->m_pkthdr.csum_data = 0xffff;
892                                 }
893                                 m->m_pkthdr.csum_flags |=
894                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
895                                 ip->ip_len = htons(ip->ip_len);
896                                 ip->ip_off = htons(ip->ip_off);
897                                 ip_input(m);
898                                 goto done;
899                         }
900                         /* Some of the logic for this was nicked from above.
901                          *
902                          * This rewrites the cached route in a local PCB.
903                          * Is this what we want to do?
904                          */
905                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
906                         ro_fwd->ro_rt = NULL;
907
908                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
909                         if (ro_fwd->ro_rt == NULL) {
910                                 ipstat.ips_noroute++;
911                                 error = EHOSTUNREACH;
912                                 goto bad;
913                         }
914
915                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
916                         ifp = ro_fwd->ro_rt->rt_ifp;
917                         ro_fwd->ro_rt->rt_use++;
918                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
919                                 dst = (struct sockaddr_in *)
920                                     ro_fwd->ro_rt->rt_gateway;
921                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
922                                 isbroadcast =
923                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
924                         else
925                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
926                         if (ro->ro_rt != NULL)
927                                 rtfree(ro->ro_rt);
928                         ro->ro_rt = ro_fwd->ro_rt;
929                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
930
931 #endif  /* ... block to be put into a function */
932                         /*
933                          * If we added a default src ip earlier,
934                          * which would have been gotten from the-then
935                          * interface, do it again, from the new one.
936                          */
937                         if (src_was_INADDR_ANY)
938                                 ip->ip_src = IA_SIN(ia)->sin_addr;
939                         goto pass ;
940                 }
941
942                 /*
943                  * if we get here, none of the above matches, and
944                  * we have to drop the pkt
945                  */
946                 m_freem(m);
947                 error = EACCES; /* not sure this is the right error msg */
948                 goto done;
949         }
950
951 pass:
952         /* 127/8 must not appear on wire - RFC1122. */
953         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
954             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
955                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
956                         ipstat.ips_badaddr++;
957                         error = EADDRNOTAVAIL;
958                         goto bad;
959                 }
960         }
961
962         m->m_pkthdr.csum_flags |= CSUM_IP;
963         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
964         if (sw_csum & CSUM_DELAY_DATA) {
965                 in_delayed_cksum(m);
966                 sw_csum &= ~CSUM_DELAY_DATA;
967         }
968         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
969
970         /*
971          * If small enough for interface, or the interface will take
972          * care of the fragmentation for us, can just send directly.
973          */
974         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
975             !(ip->ip_off & IP_DF))) {
976                 ip->ip_len = htons(ip->ip_len);
977                 ip->ip_off = htons(ip->ip_off);
978                 ip->ip_sum = 0;
979                 if (sw_csum & CSUM_DELAY_IP) {
980                         if (ip->ip_vhl == IP_VHL_BORING) {
981                                 ip->ip_sum = in_cksum_hdr(ip);
982                         } else {
983                                 ip->ip_sum = in_cksum(m, hlen);
984                         }
985                 }
986
987                 /* Record statistics for this interface address. */
988                 if (!(flags & IP_FORWARDING) && ia) {
989                         ia->ia_ifa.if_opackets++;
990                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
991                 }
992
993 #ifdef IPSEC
994                 /* clean ipsec history once it goes out of the node */
995                 ipsec_delaux(m);
996 #endif
997
998 #ifdef MBUF_STRESS_TEST
999                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1000                         struct mbuf *m1, *m2;
1001                         int length, tmp;
1002
1003                         tmp = length = m->m_pkthdr.len;
1004
1005                         while ((length -= mbuf_frag_size) >= 1) {
1006                                 m1 = m_split(m, length, MB_DONTWAIT);
1007                                 if (m1 == NULL)
1008                                         break;
1009                                 m2 = m;
1010                                 while (m2->m_next != NULL)
1011                                         m2 = m2->m_next;
1012                                 m2->m_next = m1;
1013                         }
1014                         m->m_pkthdr.len = tmp;
1015                 }
1016 #endif
1017
1018 #ifdef MPLS
1019                 if (!mpls_output_process(m, ro->ro_rt))
1020                         goto done;
1021 #endif
1022                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1023                                        ro->ro_rt);
1024                 goto done;
1025         }
1026
1027         if (ip->ip_off & IP_DF) {
1028                 error = EMSGSIZE;
1029                 /*
1030                  * This case can happen if the user changed the MTU
1031                  * of an interface after enabling IP on it.  Because
1032                  * most netifs don't keep track of routes pointing to
1033                  * them, there is no way for one to update all its
1034                  * routes when the MTU is changed.
1035                  */
1036                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1037                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1038                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1039                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1040                 }
1041                 ipstat.ips_cantfrag++;
1042                 goto bad;
1043         }
1044
1045         /*
1046          * Too large for interface; fragment if possible. If successful,
1047          * on return, m will point to a list of packets to be sent.
1048          */
1049         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1050         if (error)
1051                 goto bad;
1052         for (; m; m = m0) {
1053                 m0 = m->m_nextpkt;
1054                 m->m_nextpkt = NULL;
1055 #ifdef IPSEC
1056                 /* clean ipsec history once it goes out of the node */
1057                 ipsec_delaux(m);
1058 #endif
1059                 if (error == 0) {
1060                         /* Record statistics for this interface address. */
1061                         if (ia != NULL) {
1062                                 ia->ia_ifa.if_opackets++;
1063                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1064                         }
1065 #ifdef MPLS
1066                         if (!mpls_output_process(m, ro->ro_rt))
1067                                 continue;
1068 #endif
1069                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1070                                                ro->ro_rt);
1071                 } else {
1072                         m_freem(m);
1073                 }
1074         }
1075
1076         if (error == 0)
1077                 ipstat.ips_fragmented++;
1078
1079 done:
1080         if (ro == &iproute && ro->ro_rt != NULL) {
1081                 RTFREE(ro->ro_rt);
1082                 ro->ro_rt = NULL;
1083         }
1084 #ifdef IPSEC
1085         if (sp != NULL) {
1086                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1087                         kprintf("DP ip_output call free SP:%p\n", sp));
1088                 key_freesp(sp);
1089         }
1090 #endif
1091 #ifdef FAST_IPSEC
1092         if (sp != NULL)
1093                 KEY_FREESP(&sp);
1094 #endif
1095         if (dn_mtag != NULL)
1096                 m_tag_free(dn_mtag);
1097
1098         return (error);
1099 bad:
1100         m_freem(m);
1101         goto done;
1102 }
1103
1104 /*
1105  * Create a chain of fragments which fit the given mtu. m_frag points to the
1106  * mbuf to be fragmented; on return it points to the chain with the fragments.
1107  * Return 0 if no error. If error, m_frag may contain a partially built
1108  * chain of fragments that should be freed by the caller.
1109  *
1110  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1111  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1112  */
1113 int
1114 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1115             u_long if_hwassist_flags, int sw_csum)
1116 {
1117         int error = 0;
1118         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1119         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1120         int off;
1121         struct mbuf *m0 = *m_frag;      /* the original packet          */
1122         int firstlen;
1123         struct mbuf **mnext;
1124         int nfrags;
1125
1126         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1127                 ipstat.ips_cantfrag++;
1128                 return EMSGSIZE;
1129         }
1130
1131         /*
1132          * Must be able to put at least 8 bytes per fragment.
1133          */
1134         if (len < 8)
1135                 return EMSGSIZE;
1136
1137         /*
1138          * If the interface will not calculate checksums on
1139          * fragmented packets, then do it here.
1140          */
1141         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1142             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1143                 in_delayed_cksum(m0);
1144                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1145         }
1146
1147         if (len > PAGE_SIZE) {
1148                 /*
1149                  * Fragment large datagrams such that each segment
1150                  * contains a multiple of PAGE_SIZE amount of data,
1151                  * plus headers. This enables a receiver to perform
1152                  * page-flipping zero-copy optimizations.
1153                  *
1154                  * XXX When does this help given that sender and receiver
1155                  * could have different page sizes, and also mtu could
1156                  * be less than the receiver's page size ?
1157                  */
1158                 int newlen;
1159                 struct mbuf *m;
1160
1161                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1162                         off += m->m_len;
1163
1164                 /*
1165                  * firstlen (off - hlen) must be aligned on an
1166                  * 8-byte boundary
1167                  */
1168                 if (off < hlen)
1169                         goto smart_frag_failure;
1170                 off = ((off - hlen) & ~7) + hlen;
1171                 newlen = (~PAGE_MASK) & mtu;
1172                 if ((newlen + sizeof(struct ip)) > mtu) {
1173                         /* we failed, go back the default */
1174 smart_frag_failure:
1175                         newlen = len;
1176                         off = hlen + len;
1177                 }
1178                 len = newlen;
1179
1180         } else {
1181                 off = hlen + len;
1182         }
1183
1184         firstlen = off - hlen;
1185         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1186
1187         /*
1188          * Loop through length of segment after first fragment,
1189          * make new header and copy data of each part and link onto chain.
1190          * Here, m0 is the original packet, m is the fragment being created.
1191          * The fragments are linked off the m_nextpkt of the original
1192          * packet, which after processing serves as the first fragment.
1193          */
1194         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1195                 struct ip *mhip;        /* ip header on the fragment */
1196                 struct mbuf *m;
1197                 int mhlen = sizeof(struct ip);
1198
1199                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1200                 if (m == NULL) {
1201                         error = ENOBUFS;
1202                         ipstat.ips_odropped++;
1203                         goto done;
1204                 }
1205                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1206                 /*
1207                  * In the first mbuf, leave room for the link header, then
1208                  * copy the original IP header including options. The payload
1209                  * goes into an additional mbuf chain returned by m_copy().
1210                  */
1211                 m->m_data += max_linkhdr;
1212                 mhip = mtod(m, struct ip *);
1213                 *mhip = *ip;
1214                 if (hlen > sizeof(struct ip)) {
1215                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1216                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1217                 }
1218                 m->m_len = mhlen;
1219                 /* XXX do we need to add ip->ip_off below ? */
1220                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1221                 if (off + len >= ip->ip_len) {  /* last fragment */
1222                         len = ip->ip_len - off;
1223                         m->m_flags |= M_LASTFRAG;
1224                 } else
1225                         mhip->ip_off |= IP_MF;
1226                 mhip->ip_len = htons((u_short)(len + mhlen));
1227                 m->m_next = m_copy(m0, off, len);
1228                 if (m->m_next == NULL) {                /* copy failed */
1229                         m_free(m);
1230                         error = ENOBUFS;        /* ??? */
1231                         ipstat.ips_odropped++;
1232                         goto done;
1233                 }
1234                 m->m_pkthdr.len = mhlen + len;
1235                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1236                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1237                 mhip->ip_off = htons(mhip->ip_off);
1238                 mhip->ip_sum = 0;
1239                 if (sw_csum & CSUM_DELAY_IP)
1240                         mhip->ip_sum = in_cksum(m, mhlen);
1241                 *mnext = m;
1242                 mnext = &m->m_nextpkt;
1243         }
1244         ipstat.ips_ofragments += nfrags;
1245
1246         /* set first marker for fragment chain */
1247         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1248         m0->m_pkthdr.csum_data = nfrags;
1249
1250         /*
1251          * Update first fragment by trimming what's been copied out
1252          * and updating header.
1253          */
1254         m_adj(m0, hlen + firstlen - ip->ip_len);
1255         m0->m_pkthdr.len = hlen + firstlen;
1256         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1257         ip->ip_off |= IP_MF;
1258         ip->ip_off = htons(ip->ip_off);
1259         ip->ip_sum = 0;
1260         if (sw_csum & CSUM_DELAY_IP)
1261                 ip->ip_sum = in_cksum(m0, hlen);
1262
1263 done:
1264         *m_frag = m0;
1265         return error;
1266 }
1267
1268 void
1269 in_delayed_cksum(struct mbuf *m)
1270 {
1271         struct ip *ip;
1272         u_short csum, offset;
1273
1274         ip = mtod(m, struct ip *);
1275         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1276         csum = in_cksum_skip(m, ip->ip_len, offset);
1277         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1278                 csum = 0xffff;
1279         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1280
1281         if (offset + sizeof(u_short) > m->m_len) {
1282                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1283                     m->m_len, offset, ip->ip_p);
1284                 /*
1285                  * XXX
1286                  * this shouldn't happen, but if it does, the
1287                  * correct behavior may be to insert the checksum
1288                  * in the existing chain instead of rearranging it.
1289                  */
1290                 m = m_pullup(m, offset + sizeof(u_short));
1291         }
1292         *(u_short *)(m->m_data + offset) = csum;
1293 }
1294
1295 /*
1296  * Insert IP options into preformed packet.
1297  * Adjust IP destination as required for IP source routing,
1298  * as indicated by a non-zero in_addr at the start of the options.
1299  *
1300  * XXX This routine assumes that the packet has no options in place.
1301  */
1302 static struct mbuf *
1303 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1304 {
1305         struct ipoption *p = mtod(opt, struct ipoption *);
1306         struct mbuf *n;
1307         struct ip *ip = mtod(m, struct ip *);
1308         unsigned optlen;
1309
1310         optlen = opt->m_len - sizeof p->ipopt_dst;
1311         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1312                 *phlen = 0;
1313                 return (m);             /* XXX should fail */
1314         }
1315         if (p->ipopt_dst.s_addr)
1316                 ip->ip_dst = p->ipopt_dst;
1317         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1318                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1319                 if (n == NULL) {
1320                         *phlen = 0;
1321                         return (m);
1322                 }
1323                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1324                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1325                 m->m_len -= sizeof(struct ip);
1326                 m->m_data += sizeof(struct ip);
1327                 n->m_next = m;
1328                 m = n;
1329                 m->m_len = optlen + sizeof(struct ip);
1330                 m->m_data += max_linkhdr;
1331                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1332         } else {
1333                 m->m_data -= optlen;
1334                 m->m_len += optlen;
1335                 m->m_pkthdr.len += optlen;
1336                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1337         }
1338         ip = mtod(m, struct ip *);
1339         bcopy(p->ipopt_list, ip + 1, optlen);
1340         *phlen = sizeof(struct ip) + optlen;
1341         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1342         ip->ip_len += optlen;
1343         return (m);
1344 }
1345
1346 /*
1347  * Copy options from ip to jp,
1348  * omitting those not copied during fragmentation.
1349  */
1350 int
1351 ip_optcopy(struct ip *ip, struct ip *jp)
1352 {
1353         u_char *cp, *dp;
1354         int opt, optlen, cnt;
1355
1356         cp = (u_char *)(ip + 1);
1357         dp = (u_char *)(jp + 1);
1358         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1359         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1360                 opt = cp[0];
1361                 if (opt == IPOPT_EOL)
1362                         break;
1363                 if (opt == IPOPT_NOP) {
1364                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1365                         *dp++ = IPOPT_NOP;
1366                         optlen = 1;
1367                         continue;
1368                 }
1369
1370                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1371                     ("ip_optcopy: malformed ipv4 option"));
1372                 optlen = cp[IPOPT_OLEN];
1373                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1374                     ("ip_optcopy: malformed ipv4 option"));
1375
1376                 /* bogus lengths should have been caught by ip_dooptions */
1377                 if (optlen > cnt)
1378                         optlen = cnt;
1379                 if (IPOPT_COPIED(opt)) {
1380                         bcopy(cp, dp, optlen);
1381                         dp += optlen;
1382                 }
1383         }
1384         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1385                 *dp++ = IPOPT_EOL;
1386         return (optlen);
1387 }
1388
1389 /*
1390  * IP socket option processing.
1391  */
1392 int
1393 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1394 {
1395         struct  inpcb *inp = so->so_pcb;
1396         int     error, optval;
1397
1398         error = optval = 0;
1399         if (sopt->sopt_level != IPPROTO_IP) {
1400                 return (EINVAL);
1401         }
1402
1403         switch (sopt->sopt_dir) {
1404         case SOPT_SET:
1405                 switch (sopt->sopt_name) {
1406                 case IP_OPTIONS:
1407 #ifdef notyet
1408                 case IP_RETOPTS:
1409 #endif
1410                 {
1411                         struct mbuf *m;
1412                         if (sopt->sopt_valsize > MLEN) {
1413                                 error = EMSGSIZE;
1414                                 break;
1415                         }
1416                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1417                         if (m == NULL) {
1418                                 error = ENOBUFS;
1419                                 break;
1420                         }
1421                         m->m_len = sopt->sopt_valsize;
1422                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1423                                               m->m_len);
1424                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1425                                            m));
1426                 }
1427
1428                 case IP_TOS:
1429                 case IP_TTL:
1430                 case IP_MINTTL:
1431                 case IP_RECVOPTS:
1432                 case IP_RECVRETOPTS:
1433                 case IP_RECVDSTADDR:
1434                 case IP_RECVIF:
1435                 case IP_RECVTTL:
1436                 case IP_FAITH:
1437                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1438                                              sizeof optval);
1439                         if (error)
1440                                 break;
1441                         switch (sopt->sopt_name) {
1442                         case IP_TOS:
1443                                 inp->inp_ip_tos = optval;
1444                                 break;
1445
1446                         case IP_TTL:
1447                                 inp->inp_ip_ttl = optval;
1448                                 break;
1449                         case IP_MINTTL:
1450                                 if (optval > 0 && optval <= MAXTTL)
1451                                         inp->inp_ip_minttl = optval;
1452                                 else
1453                                         error = EINVAL;
1454                                 break;
1455 #define OPTSET(bit) \
1456         if (optval) \
1457                 inp->inp_flags |= bit; \
1458         else \
1459                 inp->inp_flags &= ~bit;
1460
1461                         case IP_RECVOPTS:
1462                                 OPTSET(INP_RECVOPTS);
1463                                 break;
1464
1465                         case IP_RECVRETOPTS:
1466                                 OPTSET(INP_RECVRETOPTS);
1467                                 break;
1468
1469                         case IP_RECVDSTADDR:
1470                                 OPTSET(INP_RECVDSTADDR);
1471                                 break;
1472
1473                         case IP_RECVIF:
1474                                 OPTSET(INP_RECVIF);
1475                                 break;
1476
1477                         case IP_RECVTTL:
1478                                 OPTSET(INP_RECVTTL);
1479                                 break;
1480
1481                         case IP_FAITH:
1482                                 OPTSET(INP_FAITH);
1483                                 break;
1484                         }
1485                         break;
1486 #undef OPTSET
1487
1488                 case IP_MULTICAST_IF:
1489                 case IP_MULTICAST_VIF:
1490                 case IP_MULTICAST_TTL:
1491                 case IP_MULTICAST_LOOP:
1492                 case IP_ADD_MEMBERSHIP:
1493                 case IP_DROP_MEMBERSHIP:
1494                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1495                         break;
1496
1497                 case IP_PORTRANGE:
1498                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1499                                             sizeof optval);
1500                         if (error)
1501                                 break;
1502
1503                         switch (optval) {
1504                         case IP_PORTRANGE_DEFAULT:
1505                                 inp->inp_flags &= ~(INP_LOWPORT);
1506                                 inp->inp_flags &= ~(INP_HIGHPORT);
1507                                 break;
1508
1509                         case IP_PORTRANGE_HIGH:
1510                                 inp->inp_flags &= ~(INP_LOWPORT);
1511                                 inp->inp_flags |= INP_HIGHPORT;
1512                                 break;
1513
1514                         case IP_PORTRANGE_LOW:
1515                                 inp->inp_flags &= ~(INP_HIGHPORT);
1516                                 inp->inp_flags |= INP_LOWPORT;
1517                                 break;
1518
1519                         default:
1520                                 error = EINVAL;
1521                                 break;
1522                         }
1523                         break;
1524
1525 #if defined(IPSEC) || defined(FAST_IPSEC)
1526                 case IP_IPSEC_POLICY:
1527                 {
1528                         caddr_t req;
1529                         size_t len = 0;
1530                         int priv;
1531                         struct mbuf *m;
1532                         int optname;
1533
1534                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1535                                 break;
1536                         soopt_to_mbuf(sopt, m);
1537                         priv = (sopt->sopt_td != NULL &&
1538                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1539                         req = mtod(m, caddr_t);
1540                         len = m->m_len;
1541                         optname = sopt->sopt_name;
1542                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1543                         m_freem(m);
1544                         break;
1545                 }
1546 #endif /*IPSEC*/
1547
1548                 default:
1549                         error = ENOPROTOOPT;
1550                         break;
1551                 }
1552                 break;
1553
1554         case SOPT_GET:
1555                 switch (sopt->sopt_name) {
1556                 case IP_OPTIONS:
1557                 case IP_RETOPTS:
1558                         if (inp->inp_options)
1559                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1560                                                            char *),
1561                                                 inp->inp_options->m_len);
1562                         else
1563                                 sopt->sopt_valsize = 0;
1564                         break;
1565
1566                 case IP_TOS:
1567                 case IP_TTL:
1568                 case IP_MINTTL:
1569                 case IP_RECVOPTS:
1570                 case IP_RECVRETOPTS:
1571                 case IP_RECVDSTADDR:
1572                 case IP_RECVTTL:
1573                 case IP_RECVIF:
1574                 case IP_PORTRANGE:
1575                 case IP_FAITH:
1576                         switch (sopt->sopt_name) {
1577
1578                         case IP_TOS:
1579                                 optval = inp->inp_ip_tos;
1580                                 break;
1581
1582                         case IP_TTL:
1583                                 optval = inp->inp_ip_ttl;
1584                                 break;
1585                         case IP_MINTTL:
1586                                 optval = inp->inp_ip_minttl;
1587                                 break;
1588
1589 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1590
1591                         case IP_RECVOPTS:
1592                                 optval = OPTBIT(INP_RECVOPTS);
1593                                 break;
1594
1595                         case IP_RECVRETOPTS:
1596                                 optval = OPTBIT(INP_RECVRETOPTS);
1597                                 break;
1598
1599                         case IP_RECVDSTADDR:
1600                                 optval = OPTBIT(INP_RECVDSTADDR);
1601                                 break;
1602
1603                         case IP_RECVTTL:
1604                                 optval = OPTBIT(INP_RECVTTL);
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                         soopt_from_kbuf(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 != NULL) {
1644                                 req = mtod(m, caddr_t);
1645                                 len = m->m_len;
1646                         }
1647                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1648                         if (error == 0)
1649                                 error = soopt_from_mbuf(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(int optname, struct mbuf **pcbopt, struct mbuf *m)
1672 {
1673         int cnt, optlen;
1674         u_char *cp;
1675         u_char opt;
1676
1677         /* turn off any old options */
1678         if (*pcbopt)
1679                 m_free(*pcbopt);
1680         *pcbopt = 0;
1681         if (m == NULL || m->m_len == 0) {
1682                 /*
1683                  * Only turning off any previous options.
1684                  */
1685                 if (m != NULL)
1686                         m_free(m);
1687                 return (0);
1688         }
1689
1690         if (m->m_len % sizeof(int32_t))
1691                 goto bad;
1692         /*
1693          * IP first-hop destination address will be stored before
1694          * actual options; move other options back
1695          * and clear it when none present.
1696          */
1697         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1698                 goto bad;
1699         cnt = m->m_len;
1700         m->m_len += sizeof(struct in_addr);
1701         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1702         ovbcopy(mtod(m, caddr_t), cp, cnt);
1703         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1704
1705         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1706                 opt = cp[IPOPT_OPTVAL];
1707                 if (opt == IPOPT_EOL)
1708                         break;
1709                 if (opt == IPOPT_NOP)
1710                         optlen = 1;
1711                 else {
1712                         if (cnt < IPOPT_OLEN + sizeof *cp)
1713                                 goto bad;
1714                         optlen = cp[IPOPT_OLEN];
1715                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1716                                 goto bad;
1717                 }
1718                 switch (opt) {
1719
1720                 default:
1721                         break;
1722
1723                 case IPOPT_LSRR:
1724                 case IPOPT_SSRR:
1725                         /*
1726                          * user process specifies route as:
1727                          *      ->A->B->C->D
1728                          * D must be our final destination (but we can't
1729                          * check that since we may not have connected yet).
1730                          * A is first hop destination, which doesn't appear in
1731                          * actual IP option, but is stored before the options.
1732                          */
1733                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1734                                 goto bad;
1735                         m->m_len -= sizeof(struct in_addr);
1736                         cnt -= sizeof(struct in_addr);
1737                         optlen -= sizeof(struct in_addr);
1738                         cp[IPOPT_OLEN] = optlen;
1739                         /*
1740                          * Move first hop before start of options.
1741                          */
1742                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1743                               sizeof(struct in_addr));
1744                         /*
1745                          * Then copy rest of options back
1746                          * to close up the deleted entry.
1747                          */
1748                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1749                                 &cp[IPOPT_OFFSET+1],
1750                                 cnt - (IPOPT_MINOFF - 1));
1751                         break;
1752                 }
1753         }
1754         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1755                 goto bad;
1756         *pcbopt = m;
1757         return (0);
1758
1759 bad:
1760         m_free(m);
1761         return (EINVAL);
1762 }
1763
1764 /*
1765  * XXX
1766  * The whole multicast option thing needs to be re-thought.
1767  * Several of these options are equally applicable to non-multicast
1768  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1769  * standard option (IP_TTL).
1770  */
1771
1772 /*
1773  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1774  */
1775 static struct ifnet *
1776 ip_multicast_if(struct in_addr *a, int *ifindexp)
1777 {
1778         int ifindex;
1779         struct ifnet *ifp;
1780
1781         if (ifindexp)
1782                 *ifindexp = 0;
1783         if (ntohl(a->s_addr) >> 24 == 0) {
1784                 ifindex = ntohl(a->s_addr) & 0xffffff;
1785                 if (ifindex < 0 || if_index < ifindex)
1786                         return NULL;
1787                 ifp = ifindex2ifnet[ifindex];
1788                 if (ifindexp)
1789                         *ifindexp = ifindex;
1790         } else {
1791                 ifp = INADDR_TO_IFP(a);
1792         }
1793         return ifp;
1794 }
1795
1796 /*
1797  * Set the IP multicast options in response to user setsockopt().
1798  */
1799 static int
1800 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1801 {
1802         int error = 0;
1803         int i;
1804         struct in_addr addr;
1805         struct ip_mreq mreq;
1806         struct ifnet *ifp;
1807         struct ip_moptions *imo = *imop;
1808         int ifindex;
1809
1810         if (imo == NULL) {
1811                 /*
1812                  * No multicast option buffer attached to the pcb;
1813                  * allocate one and initialize to default values.
1814                  */
1815                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1816
1817                 *imop = imo;
1818                 imo->imo_multicast_ifp = NULL;
1819                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1820                 imo->imo_multicast_vif = -1;
1821                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1822                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1823                 imo->imo_num_memberships = 0;
1824         }
1825         switch (sopt->sopt_name) {
1826         /* store an index number for the vif you wanna use in the send */
1827         case IP_MULTICAST_VIF:
1828                 if (legal_vif_num == 0) {
1829                         error = EOPNOTSUPP;
1830                         break;
1831                 }
1832                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1833                 if (error)
1834                         break;
1835                 if (!legal_vif_num(i) && (i != -1)) {
1836                         error = EINVAL;
1837                         break;
1838                 }
1839                 imo->imo_multicast_vif = i;
1840                 break;
1841
1842         case IP_MULTICAST_IF:
1843                 /*
1844                  * Select the interface for outgoing multicast packets.
1845                  */
1846                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1847                 if (error)
1848                         break;
1849
1850                 /*
1851                  * INADDR_ANY is used to remove a previous selection.
1852                  * When no interface is selected, a default one is
1853                  * chosen every time a multicast packet is sent.
1854                  */
1855                 if (addr.s_addr == INADDR_ANY) {
1856                         imo->imo_multicast_ifp = NULL;
1857                         break;
1858                 }
1859                 /*
1860                  * The selected interface is identified by its local
1861                  * IP address.  Find the interface and confirm that
1862                  * it supports multicasting.
1863                  */
1864                 crit_enter();
1865                 ifp = ip_multicast_if(&addr, &ifindex);
1866                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1867                         crit_exit();
1868                         error = EADDRNOTAVAIL;
1869                         break;
1870                 }
1871                 imo->imo_multicast_ifp = ifp;
1872                 if (ifindex)
1873                         imo->imo_multicast_addr = addr;
1874                 else
1875                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1876                 crit_exit();
1877                 break;
1878
1879         case IP_MULTICAST_TTL:
1880                 /*
1881                  * Set the IP time-to-live for outgoing multicast packets.
1882                  * The original multicast API required a char argument,
1883                  * which is inconsistent with the rest of the socket API.
1884                  * We allow either a char or an int.
1885                  */
1886                 if (sopt->sopt_valsize == 1) {
1887                         u_char ttl;
1888                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1889                         if (error)
1890                                 break;
1891                         imo->imo_multicast_ttl = ttl;
1892                 } else {
1893                         u_int ttl;
1894                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1895                         if (error)
1896                                 break;
1897                         if (ttl > 255)
1898                                 error = EINVAL;
1899                         else
1900                                 imo->imo_multicast_ttl = ttl;
1901                 }
1902                 break;
1903
1904         case IP_MULTICAST_LOOP:
1905                 /*
1906                  * Set the loopback flag for outgoing multicast packets.
1907                  * Must be zero or one.  The original multicast API required a
1908                  * char argument, which is inconsistent with the rest
1909                  * of the socket API.  We allow either a char or an int.
1910                  */
1911                 if (sopt->sopt_valsize == 1) {
1912                         u_char loop;
1913
1914                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1915                         if (error)
1916                                 break;
1917                         imo->imo_multicast_loop = !!loop;
1918                 } else {
1919                         u_int loop;
1920
1921                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1922                                             sizeof loop);
1923                         if (error)
1924                                 break;
1925                         imo->imo_multicast_loop = !!loop;
1926                 }
1927                 break;
1928
1929         case IP_ADD_MEMBERSHIP:
1930                 /*
1931                  * Add a multicast group membership.
1932                  * Group must be a valid IP multicast address.
1933                  */
1934                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1935                 if (error)
1936                         break;
1937
1938                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1939                         error = EINVAL;
1940                         break;
1941                 }
1942                 crit_enter();
1943                 /*
1944                  * If no interface address was provided, use the interface of
1945                  * the route to the given multicast address.
1946                  */
1947                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1948                         struct sockaddr_in dst;
1949                         struct rtentry *rt;
1950
1951                         bzero(&dst, sizeof(struct sockaddr_in));
1952                         dst.sin_len = sizeof(struct sockaddr_in);
1953                         dst.sin_family = AF_INET;
1954                         dst.sin_addr = mreq.imr_multiaddr;
1955                         rt = rtlookup((struct sockaddr *)&dst);
1956                         if (rt == NULL) {
1957                                 error = EADDRNOTAVAIL;
1958                                 crit_exit();
1959                                 break;
1960                         }
1961                         --rt->rt_refcnt;
1962                         ifp = rt->rt_ifp;
1963                 } else {
1964                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1965                 }
1966
1967                 /*
1968                  * See if we found an interface, and confirm that it
1969                  * supports multicast.
1970                  */
1971                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1972                         error = EADDRNOTAVAIL;
1973                         crit_exit();
1974                         break;
1975                 }
1976                 /*
1977                  * See if the membership already exists or if all the
1978                  * membership slots are full.
1979                  */
1980                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1981                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1982                             imo->imo_membership[i]->inm_addr.s_addr
1983                                                 == mreq.imr_multiaddr.s_addr)
1984                                 break;
1985                 }
1986                 if (i < imo->imo_num_memberships) {
1987                         error = EADDRINUSE;
1988                         crit_exit();
1989                         break;
1990                 }
1991                 if (i == IP_MAX_MEMBERSHIPS) {
1992                         error = ETOOMANYREFS;
1993                         crit_exit();
1994                         break;
1995                 }
1996                 /*
1997                  * Everything looks good; add a new record to the multicast
1998                  * address list for the given interface.
1999                  */
2000                 if ((imo->imo_membership[i] =
2001                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2002                         error = ENOBUFS;
2003                         crit_exit();
2004                         break;
2005                 }
2006                 ++imo->imo_num_memberships;
2007                 crit_exit();
2008                 break;
2009
2010         case IP_DROP_MEMBERSHIP:
2011                 /*
2012                  * Drop a multicast group membership.
2013                  * Group must be a valid IP multicast address.
2014                  */
2015                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2016                 if (error)
2017                         break;
2018
2019                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2020                         error = EINVAL;
2021                         break;
2022                 }
2023
2024                 crit_enter();
2025                 /*
2026                  * If an interface address was specified, get a pointer
2027                  * to its ifnet structure.
2028                  */
2029                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2030                         ifp = NULL;
2031                 else {
2032                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2033                         if (ifp == NULL) {
2034                                 error = EADDRNOTAVAIL;
2035                                 crit_exit();
2036                                 break;
2037                         }
2038                 }
2039                 /*
2040                  * Find the membership in the membership array.
2041                  */
2042                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2043                         if ((ifp == NULL ||
2044                              imo->imo_membership[i]->inm_ifp == ifp) &&
2045                             imo->imo_membership[i]->inm_addr.s_addr ==
2046                             mreq.imr_multiaddr.s_addr)
2047                                 break;
2048                 }
2049                 if (i == imo->imo_num_memberships) {
2050                         error = EADDRNOTAVAIL;
2051                         crit_exit();
2052                         break;
2053                 }
2054                 /*
2055                  * Give up the multicast address record to which the
2056                  * membership points.
2057                  */
2058                 in_delmulti(imo->imo_membership[i]);
2059                 /*
2060                  * Remove the gap in the membership array.
2061                  */
2062                 for (++i; i < imo->imo_num_memberships; ++i)
2063                         imo->imo_membership[i-1] = imo->imo_membership[i];
2064                 --imo->imo_num_memberships;
2065                 crit_exit();
2066                 break;
2067
2068         default:
2069                 error = EOPNOTSUPP;
2070                 break;
2071         }
2072
2073         /*
2074          * If all options have default values, no need to keep the mbuf.
2075          */
2076         if (imo->imo_multicast_ifp == NULL &&
2077             imo->imo_multicast_vif == -1 &&
2078             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2079             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2080             imo->imo_num_memberships == 0) {
2081                 kfree(*imop, M_IPMOPTS);
2082                 *imop = NULL;
2083         }
2084
2085         return (error);
2086 }
2087
2088 /*
2089  * Return the IP multicast options in response to user getsockopt().
2090  */
2091 static int
2092 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2093 {
2094         struct in_addr addr;
2095         struct in_ifaddr *ia;
2096         int error, optval;
2097         u_char coptval;
2098
2099         error = 0;
2100         switch (sopt->sopt_name) {
2101         case IP_MULTICAST_VIF:
2102                 if (imo != NULL)
2103                         optval = imo->imo_multicast_vif;
2104                 else
2105                         optval = -1;
2106                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2107                 break;
2108
2109         case IP_MULTICAST_IF:
2110                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2111                         addr.s_addr = INADDR_ANY;
2112                 else if (imo->imo_multicast_addr.s_addr) {
2113                         /* return the value user has set */
2114                         addr = imo->imo_multicast_addr;
2115                 } else {
2116                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2117                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2118                                 : IA_SIN(ia)->sin_addr.s_addr;
2119                 }
2120                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2121                 break;
2122
2123         case IP_MULTICAST_TTL:
2124                 if (imo == NULL)
2125                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2126                 else
2127                         optval = coptval = imo->imo_multicast_ttl;
2128                 if (sopt->sopt_valsize == 1)
2129                         soopt_from_kbuf(sopt, &coptval, 1);
2130                 else
2131                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2132                 break;
2133
2134         case IP_MULTICAST_LOOP:
2135                 if (imo == NULL)
2136                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2137                 else
2138                         optval = coptval = imo->imo_multicast_loop;
2139                 if (sopt->sopt_valsize == 1)
2140                         soopt_from_kbuf(sopt, &coptval, 1);
2141                 else
2142                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2143                 break;
2144
2145         default:
2146                 error = ENOPROTOOPT;
2147                 break;
2148         }
2149         return (error);
2150 }
2151
2152 /*
2153  * Discard the IP multicast options.
2154  */
2155 void
2156 ip_freemoptions(struct ip_moptions *imo)
2157 {
2158         int i;
2159
2160         if (imo != NULL) {
2161                 for (i = 0; i < imo->imo_num_memberships; ++i)
2162                         in_delmulti(imo->imo_membership[i]);
2163                 kfree(imo, M_IPMOPTS);
2164         }
2165 }
2166
2167 /*
2168  * Routine called from ip_output() to loop back a copy of an IP multicast
2169  * packet to the input queue of a specified interface.  Note that this
2170  * calls the output routine of the loopback "driver", but with an interface
2171  * pointer that might NOT be a loopback interface -- evil, but easier than
2172  * replicating that code here.
2173  */
2174 static void
2175 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2176              int hlen)
2177 {
2178         struct ip *ip;
2179         struct mbuf *copym;
2180
2181         copym = m_copypacket(m, MB_DONTWAIT);
2182         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2183                 copym = m_pullup(copym, hlen);
2184         if (copym != NULL) {
2185                 /*
2186                  * if the checksum hasn't been computed, mark it as valid
2187                  */
2188                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2189                         in_delayed_cksum(copym);
2190                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2191                         copym->m_pkthdr.csum_flags |=
2192                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2193                         copym->m_pkthdr.csum_data = 0xffff;
2194                 }
2195                 /*
2196                  * We don't bother to fragment if the IP length is greater
2197                  * than the interface's MTU.  Can this possibly matter?
2198                  */
2199                 ip = mtod(copym, struct ip *);
2200                 ip->ip_len = htons(ip->ip_len);
2201                 ip->ip_off = htons(ip->ip_off);
2202                 ip->ip_sum = 0;
2203                 if (ip->ip_vhl == IP_VHL_BORING) {
2204                         ip->ip_sum = in_cksum_hdr(ip);
2205                 } else {
2206                         ip->ip_sum = in_cksum(copym, hlen);
2207                 }
2208                 /*
2209                  * NB:
2210                  * It's not clear whether there are any lingering
2211                  * reentrancy problems in other areas which might
2212                  * be exposed by using ip_input directly (in
2213                  * particular, everything which modifies the packet
2214                  * in-place).  Yet another option is using the
2215                  * protosw directly to deliver the looped back
2216                  * packet.  For the moment, we'll err on the side
2217                  * of safety by using if_simloop().
2218                  */
2219 #if 1 /* XXX */
2220                 if (dst->sin_family != AF_INET) {
2221                         kprintf("ip_mloopback: bad address family %d\n",
2222                                                 dst->sin_family);
2223                         dst->sin_family = AF_INET;
2224                 }
2225 #endif
2226
2227 #ifdef notdef
2228                 copym->m_pkthdr.rcvif = ifp;
2229                 ip_input(copym);
2230 #else
2231                 if_simloop(ifp, copym, dst->sin_family, 0);
2232 #endif
2233         }
2234 }