d90bb3ed8bb9228db79dc2445b8aa422097d46f4
[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  */
32
33 #define _IP_VHL
34
35 #include "opt_ipdn.h"
36 #include "opt_ipdivert.h"
37 #include "opt_ipsec.h"
38 #include "opt_mbuf_stress_test.h"
39 #include "opt_mpls.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/proc.h>
50 #include <sys/priv.h>
51 #include <sys/sysctl.h>
52 #include <sys/in_cksum.h>
53 #include <sys/lock.h>
54
55 #include <sys/thread2.h>
56 #include <sys/mplock2.h>
57 #include <sys/msgport2.h>
58
59 #include <net/if.h>
60 #include <net/netisr.h>
61 #include <net/pfil.h>
62 #include <net/route.h>
63
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip_var.h>
70
71 #include <netproto/mpls/mpls_var.h>
72
73 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
74
75 #ifdef IPSEC
76 #include <netinet6/ipsec.h>
77 #include <netproto/key/key.h>
78 #ifdef IPSEC_DEBUG
79 #include <netproto/key/key_debug.h>
80 #else
81 #define KEYDEBUG(lev,arg)
82 #endif
83 #endif /*IPSEC*/
84
85 #ifdef FAST_IPSEC
86 #include <netproto/ipsec/ipsec.h>
87 #include <netproto/ipsec/xform.h>
88 #include <netproto/ipsec/key.h>
89 #endif /*FAST_IPSEC*/
90
91 #include <net/ipfw/ip_fw.h>
92 #include <net/dummynet/ip_dummynet.h>
93
94 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
95                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
96                                   (ntohl(a.s_addr)>>16)&0xFF,\
97                                   (ntohl(a.s_addr)>>8)&0xFF,\
98                                   (ntohl(a.s_addr))&0xFF, y);
99
100 u_short ip_id;
101
102 #ifdef MBUF_STRESS_TEST
103 int mbuf_frag_size = 0;
104 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
105         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
106 #endif
107
108 static int ip_do_rfc6864 = 1;
109 SYSCTL_INT(_net_inet_ip, OID_AUTO, rfc6864, CTLFLAG_RW, &ip_do_rfc6864, 0,
110     "Don't generate IP ID for DF IP datagrams");
111
112 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
113 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
114 static void     ip_mloopback
115         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
116 static int      ip_getmoptions
117         (struct sockopt *, struct ip_moptions *);
118 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
119 static int      ip_setmoptions
120         (struct sockopt *, struct ip_moptions **);
121
122 int     ip_optcopy(struct ip *, struct ip *);
123
124 extern  struct protosw inetsw[];
125
126 static int
127 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
128 {
129         struct in_ifaddr_container *iac;
130
131         /*
132          * We need to figure out if we have been forwarded to a local
133          * socket.  If so, then we should somehow "loop back" to
134          * ip_input(), and get directed to the PCB as if we had received
135          * this packet.  This is because it may be difficult to identify
136          * the packets you want to forward until they are being output
137          * and have selected an interface (e.g. locally initiated
138          * packets).  If we used the loopback inteface, we would not be
139          * able to control what happens as the packet runs through
140          * ip_input() as it is done through a ISR.
141          */
142         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
143                 /*
144                  * If the addr to forward to is one of ours, we pretend
145                  * to be the destination for this packet.
146                  */
147                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
148                         break;
149         }
150         if (iac != NULL) {
151                 struct ip *ip;
152
153                 if (m->m_pkthdr.rcvif == NULL)
154                         m->m_pkthdr.rcvif = loif;
155                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
156                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
157                                                   CSUM_PSEUDO_HDR;
158                         m->m_pkthdr.csum_data = 0xffff;
159                 }
160                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
161
162                 /*
163                  * Make sure that the IP header is in one mbuf,
164                  * required by ip_input
165                  */
166                 if (m->m_len < hlen) {
167                         m = m_pullup(m, hlen);
168                         if (m == NULL) {
169                                 /* The packet was freed; we are done */
170                                 return 1;
171                         }
172                 }
173                 ip = mtod(m, struct ip *);
174
175                 ip->ip_len = htons(ip->ip_len);
176                 ip->ip_off = htons(ip->ip_off);
177                 ip_input(m);
178
179                 return 1; /* The packet gets forwarded locally */
180         }
181         return 0;
182 }
183
184 /*
185  * IP output.  The packet in mbuf chain m contains a skeletal IP
186  * header (with len, off, ttl, proto, tos, src, dst).
187  * The mbuf chain containing the packet will be freed.
188  * The mbuf opt, if present, will not be freed.
189  */
190 int
191 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
192           int flags, struct ip_moptions *imo, struct inpcb *inp)
193 {
194         struct ip *ip;
195         struct ifnet *ifp = NULL;       /* keep compiler happy */
196         struct mbuf *m;
197         int hlen = sizeof(struct ip);
198         int len, error = 0;
199         struct sockaddr_in *dst = NULL; /* keep compiler happy */
200         struct in_ifaddr *ia = NULL;
201         int isbroadcast, sw_csum;
202         struct in_addr pkt_dst;
203         struct route iproute;
204         struct m_tag *mtag;
205 #ifdef IPSEC
206         struct secpolicy *sp = NULL;
207         struct socket *so = inp ? inp->inp_socket : NULL;
208 #endif
209 #ifdef FAST_IPSEC
210         struct secpolicy *sp = NULL;
211         struct tdb_ident *tdbi;
212 #endif /* FAST_IPSEC */
213         struct sockaddr_in *next_hop = NULL;
214         int src_was_INADDR_ANY = 0;     /* as the name says... */
215
216         ASSERT_NETISR_NCPUS(mycpuid);
217
218         m = m0;
219         M_ASSERTPKTHDR(m);
220
221         if (ro == NULL) {
222                 ro = &iproute;
223                 bzero(ro, sizeof *ro);
224         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
225                 if (flags & IP_DEBUGROUTE) {
226                         panic("ip_output: rt rt_cpuid %d accessed on cpu %d\n",
227                             ro->ro_rt->rt_cpuid, mycpuid);
228                 }
229
230                 /*
231                  * XXX
232                  * If the cached rtentry's owner CPU is not the current CPU,
233                  * then don't touch the cached rtentry (remote free is too
234                  * expensive in this context); just relocate the route.
235                  */
236                 ro = &iproute;
237                 bzero(ro, sizeof *ro);
238         }
239
240         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
241                 /* Next hop */
242                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
243                 KKASSERT(mtag != NULL);
244                 next_hop = m_tag_data(mtag);
245         }
246
247         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
248                 struct dn_pkt *dn_pkt;
249
250                 /* Extract info from dummynet tag */
251                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
252                 KKASSERT(mtag != NULL);
253                 dn_pkt = m_tag_data(mtag);
254
255                 /*
256                  * The packet was already tagged, so part of the
257                  * processing was already done, and we need to go down.
258                  * Get the calculated parameters from the tag.
259                  */
260                 ifp = dn_pkt->ifp;
261
262                 KKASSERT(ro == &iproute);
263                 *ro = dn_pkt->ro; /* structure copy */
264                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
265
266                 dst = dn_pkt->dn_dst;
267                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
268                         /* If 'dst' points into dummynet tag, adjust it */
269                         dst = (struct sockaddr_in *)&(ro->ro_dst);
270                 }
271
272                 ip = mtod(m, struct ip *);
273                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
274                 if (ro->ro_rt)
275                         ia = ifatoia(ro->ro_rt->rt_ifa);
276                 goto sendit;
277         }
278
279         if (opt) {
280                 len = 0;
281                 m = ip_insertoptions(m, opt, &len);
282                 if (len != 0)
283                         hlen = len;
284         }
285         ip = mtod(m, struct ip *);
286
287         /*
288          * Fill in IP header.
289          */
290         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
291                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
292                 ip->ip_off &= IP_DF;
293                 if (ip_do_rfc6864 && (ip->ip_off & IP_DF))
294                         ip->ip_id = 0;
295                 else
296                         ip->ip_id = ip_newid();
297                 ipstat.ips_localout++;
298         } else {
299                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
300         }
301
302 reroute:
303         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
304
305         dst = (struct sockaddr_in *)&ro->ro_dst;
306         /*
307          * If there is a cached route,
308          * check that it is to the same destination
309          * and is still up.  If not, free it and try again.
310          * The address family should also be checked in case of sharing the
311          * cache with IPv6.
312          */
313         if (ro->ro_rt &&
314             (!(ro->ro_rt->rt_flags & RTF_UP) ||
315              dst->sin_family != AF_INET ||
316              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
317                 rtfree(ro->ro_rt);
318                 ro->ro_rt = NULL;
319         }
320         if (ro->ro_rt == NULL) {
321                 bzero(dst, sizeof *dst);
322                 dst->sin_family = AF_INET;
323                 dst->sin_len = sizeof *dst;
324                 dst->sin_addr = pkt_dst;
325         }
326         /*
327          * If routing to interface only,
328          * short circuit routing lookup.
329          */
330         if (flags & IP_ROUTETOIF) {
331                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
332                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
333                         ipstat.ips_noroute++;
334                         error = ENETUNREACH;
335                         goto bad;
336                 }
337                 ifp = ia->ia_ifp;
338                 ip->ip_ttl = 1;
339                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
340         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
341                    imo != NULL && imo->imo_multicast_ifp != NULL) {
342                 /*
343                  * Bypass the normal routing lookup for multicast
344                  * packets if the interface is specified.
345                  */
346                 ifp = imo->imo_multicast_ifp;
347                 ia = IFP_TO_IA(ifp);
348                 isbroadcast = 0;        /* fool gcc */
349         } else {
350                 /*
351                  * If this is the case, we probably don't want to allocate
352                  * a protocol-cloned route since we didn't get one from the
353                  * ULP.  This lets TCP do its thing, while not burdening
354                  * forwarding or ICMP with the overhead of cloning a route.
355                  * Of course, we still want to do any cloning requested by
356                  * the link layer, as this is probably required in all cases
357                  * for correct operation (as it is for ARP).
358                  */
359                 if (ro->ro_rt == NULL)
360                         rtalloc_ign(ro, RTF_PRCLONING);
361                 if (ro->ro_rt == NULL) {
362                         ipstat.ips_noroute++;
363                         error = EHOSTUNREACH;
364                         goto bad;
365                 }
366                 ia = ifatoia(ro->ro_rt->rt_ifa);
367                 ifp = ro->ro_rt->rt_ifp;
368                 ro->ro_rt->rt_use++;
369                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
370                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
371                 if (ro->ro_rt->rt_flags & RTF_HOST)
372                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
373                 else
374                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
375         }
376         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
377                 m->m_flags |= M_MCAST;
378                 /*
379                  * IP destination address is multicast.  Make sure "dst"
380                  * still points to the address in "ro".  (It may have been
381                  * changed to point to a gateway address, above.)
382                  */
383                 dst = (struct sockaddr_in *)&ro->ro_dst;
384                 /*
385                  * See if the caller provided any multicast options
386                  */
387                 if (imo != NULL) {
388                         ip->ip_ttl = imo->imo_multicast_ttl;
389                         if (imo->imo_multicast_vif != -1) {
390                                 ip->ip_src.s_addr =
391                                     ip_mcast_src ?
392                                     ip_mcast_src(imo->imo_multicast_vif) :
393                                     INADDR_ANY;
394                         }
395                 } else {
396                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
397                 }
398                 /*
399                  * Confirm that the outgoing interface supports multicast.
400                  */
401                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
402                         if (!(ifp->if_flags & IFF_MULTICAST)) {
403                                 ipstat.ips_noroute++;
404                                 error = ENETUNREACH;
405                                 goto bad;
406                         }
407                 }
408                 /*
409                  * If source address not specified yet, use address of the
410                  * outgoing interface.  In case, keep note we did that, so
411                  * if the the firewall changes the next-hop causing the
412                  * output interface to change, we can fix that.
413                  */
414                 if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
415                         /* Interface may have no addresses. */
416                         if (ia != NULL) {
417                                 ip->ip_src = IA_SIN(ia)->sin_addr;
418                                 src_was_INADDR_ANY = 1;
419                         }
420                 }
421
422                 if (ip->ip_src.s_addr != INADDR_ANY) {
423                         struct in_multi *inm;
424
425                         inm = IN_LOOKUP_MULTI(&pkt_dst, ifp);
426                         if (inm != NULL &&
427                             (imo == NULL || imo->imo_multicast_loop)) {
428                                 /*
429                                  * If we belong to the destination multicast
430                                  * group on the outgoing interface, and the
431                                  * caller did not forbid loopback, loop back
432                                  * a copy.
433                                  */
434                                 ip_mloopback(ifp, m, dst, hlen);
435                         } else {
436                                 /*
437                                  * If we are acting as a multicast router,
438                                  * perform multicast forwarding as if the
439                                  * packet had just arrived on the interface
440                                  * to which we are about to send.  The
441                                  * multicast forwarding function recursively
442                                  * calls this function, using the IP_FORWARDING
443                                  * flag to prevent infinite recursion.
444                                  *
445                                  * Multicasts that are looped back by
446                                  * ip_mloopback(), above, will be forwarded by
447                                  * the ip_input() routine, if necessary.
448                                  */
449                                 if (ip_mrouter && !(flags & IP_FORWARDING)) {
450                                         /*
451                                          * If rsvp daemon is not running, do
452                                          * not set ip_moptions. This ensures
453                                          * that the packet is multicast and
454                                          * not just sent down one link as
455                                          * prescribed by rsvpd.
456                                          */
457                                         if (!rsvp_on)
458                                                 imo = NULL;
459                                         if (ip_mforward) {
460                                                 get_mplock();
461                                                 if (ip_mforward(ip, ifp,
462                                                     m, imo) != 0) {
463                                                         m_freem(m);
464                                                         rel_mplock();
465                                                         goto done;
466                                                 }
467                                                 rel_mplock();
468                                         }
469                                 }
470                         }
471                 }
472
473                 /*
474                  * Multicasts with a time-to-live of zero may be looped-
475                  * back, above, but must not be transmitted on a network.
476                  * Also, multicasts addressed to the loopback interface
477                  * are not sent -- the above call to ip_mloopback() will
478                  * loop back a copy if this host actually belongs to the
479                  * destination group on the loopback interface.
480                  */
481                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
482                         m_freem(m);
483                         goto done;
484                 }
485
486                 goto sendit;
487         } else {
488                 m->m_flags &= ~M_MCAST;
489         }
490
491         /*
492          * If the source address is not specified yet, use the address
493          * of the outgoing interface.  In case, keep note we did that,
494          * so if the the firewall changes the next-hop causing the output
495          * interface to change, we can fix that.
496          */
497         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
498                 /* Interface may have no addresses. */
499                 if (ia != NULL) {
500                         ip->ip_src = IA_SIN(ia)->sin_addr;
501                         src_was_INADDR_ANY = 1;
502                 }
503         }
504
505         /*
506          * Look for broadcast address and
507          * verify user is allowed to send
508          * such a packet.
509          */
510         if (isbroadcast) {
511                 if (!(ifp->if_flags & IFF_BROADCAST)) {
512                         error = EADDRNOTAVAIL;
513                         goto bad;
514                 }
515                 if (!(flags & IP_ALLOWBROADCAST)) {
516                         error = EACCES;
517                         goto bad;
518                 }
519                 /* don't allow broadcast messages to be fragmented */
520                 if (ip->ip_len > ifp->if_mtu) {
521                         error = EMSGSIZE;
522                         goto bad;
523                 }
524                 m->m_flags |= M_BCAST;
525         } else {
526                 m->m_flags &= ~M_BCAST;
527         }
528
529 sendit:
530 #ifdef IPSEC
531         /* get SP for this packet */
532         if (so == NULL)
533                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
534         else
535                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
536
537         if (sp == NULL) {
538                 ipsecstat.out_inval++;
539                 goto bad;
540         }
541
542         error = 0;
543
544         /* check policy */
545         switch (sp->policy) {
546         case IPSEC_POLICY_DISCARD:
547                 /*
548                  * This packet is just discarded.
549                  */
550                 ipsecstat.out_polvio++;
551                 goto bad;
552
553         case IPSEC_POLICY_BYPASS:
554         case IPSEC_POLICY_NONE:
555         case IPSEC_POLICY_TCP:
556                 /* no need to do IPsec. */
557                 goto skip_ipsec;
558
559         case IPSEC_POLICY_IPSEC:
560                 if (sp->req == NULL) {
561                         /* acquire a policy */
562                         error = key_spdacquire(sp);
563                         goto bad;
564                 }
565                 break;
566
567         case IPSEC_POLICY_ENTRUST:
568         default:
569                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
570         }
571     {
572         struct ipsec_output_state state;
573         bzero(&state, sizeof state);
574         state.m = m;
575         if (flags & IP_ROUTETOIF) {
576                 state.ro = &iproute;
577                 bzero(&iproute, sizeof iproute);
578         } else
579                 state.ro = ro;
580         state.dst = (struct sockaddr *)dst;
581
582         ip->ip_sum = 0;
583
584         /*
585          * XXX
586          * delayed checksums are not currently compatible with IPsec
587          */
588         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
589                 in_delayed_cksum(m);
590                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
591         }
592
593         ip->ip_len = htons(ip->ip_len);
594         ip->ip_off = htons(ip->ip_off);
595
596         error = ipsec4_output(&state, sp, flags);
597
598         m = state.m;
599         if (flags & IP_ROUTETOIF) {
600                 /*
601                  * if we have tunnel mode SA, we may need to ignore
602                  * IP_ROUTETOIF.
603                  */
604                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
605                         flags &= ~IP_ROUTETOIF;
606                         ro = state.ro;
607                 }
608         } else
609                 ro = state.ro;
610         dst = (struct sockaddr_in *)state.dst;
611         if (error) {
612                 /* mbuf is already reclaimed in ipsec4_output. */
613                 m0 = NULL;
614                 switch (error) {
615                 case EHOSTUNREACH:
616                 case ENETUNREACH:
617                 case EMSGSIZE:
618                 case ENOBUFS:
619                 case ENOMEM:
620                         break;
621                 default:
622                         kprintf("ip4_output (ipsec): error code %d\n", error);
623                         /*fall through*/
624                 case ENOENT:
625                         /* don't show these error codes to the user */
626                         error = 0;
627                         break;
628                 }
629                 goto bad;
630         }
631     }
632
633         /* be sure to update variables that are affected by ipsec4_output() */
634         ip = mtod(m, struct ip *);
635 #ifdef _IP_VHL
636         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
637 #else
638         hlen = ip->ip_hl << 2;
639 #endif
640         if (ro->ro_rt == NULL) {
641                 if (!(flags & IP_ROUTETOIF)) {
642                         kprintf("ip_output: "
643                                 "can't update route after IPsec processing\n");
644                         error = EHOSTUNREACH;   /*XXX*/
645                         goto bad;
646                 }
647         } else {
648                 ia = ifatoia(ro->ro_rt->rt_ifa);
649                 ifp = ro->ro_rt->rt_ifp;
650         }
651
652         /* make it flipped, again. */
653         ip->ip_len = ntohs(ip->ip_len);
654         ip->ip_off = ntohs(ip->ip_off);
655 skip_ipsec:
656 #endif /*IPSEC*/
657 #ifdef FAST_IPSEC
658         /*
659          * Check the security policy (SP) for the packet and, if
660          * required, do IPsec-related processing.  There are two
661          * cases here; the first time a packet is sent through
662          * it will be untagged and handled by ipsec4_checkpolicy.
663          * If the packet is resubmitted to ip_output (e.g. after
664          * AH, ESP, etc. processing), there will be a tag to bypass
665          * the lookup and related policy checking.
666          */
667         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
668         crit_enter();
669         if (mtag != NULL) {
670                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
671                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
672                 if (sp == NULL)
673                         error = -EINVAL;        /* force silent drop */
674                 m_tag_delete(m, mtag);
675         } else {
676                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
677                                         &error, inp);
678         }
679         /*
680          * There are four return cases:
681          *    sp != NULL                    apply IPsec policy
682          *    sp == NULL, error == 0        no IPsec handling needed
683          *    sp == NULL, error == -EINVAL  discard packet w/o error
684          *    sp == NULL, error != 0        discard packet, report error
685          */
686         if (sp != NULL) {
687                 /* Loop detection, check if ipsec processing already done */
688                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
689                 for (mtag = m_tag_first(m); mtag != NULL;
690                      mtag = m_tag_next(m, mtag)) {
691                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
692                                 continue;
693                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
694                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
695                                 continue;
696                         /*
697                          * Check if policy has an SA associated with it.
698                          * This can happen when an SP has yet to acquire
699                          * an SA; e.g. on first reference.  If it occurs,
700                          * then we let ipsec4_process_packet do its thing.
701                          */
702                         if (sp->req->sav == NULL)
703                                 break;
704                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
705                         if (tdbi->spi == sp->req->sav->spi &&
706                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
707                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
708                                  sizeof(union sockaddr_union)) == 0) {
709                                 /*
710                                  * No IPsec processing is needed, free
711                                  * reference to SP.
712                                  *
713                                  * NB: null pointer to avoid free at
714                                  *     done: below.
715                                  */
716                                 KEY_FREESP(&sp), sp = NULL;
717                                 crit_exit();
718                                 goto spd_done;
719                         }
720                 }
721
722                 /*
723                  * Do delayed checksums now because we send before
724                  * this is done in the normal processing path.
725                  */
726                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
727                         in_delayed_cksum(m);
728                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
729                 }
730
731                 ip->ip_len = htons(ip->ip_len);
732                 ip->ip_off = htons(ip->ip_off);
733
734                 /* NB: callee frees mbuf */
735                 error = ipsec4_process_packet(m, sp->req, flags, 0);
736                 /*
737                  * Preserve KAME behaviour: ENOENT can be returned
738                  * when an SA acquire is in progress.  Don't propagate
739                  * this to user-level; it confuses applications.
740                  *
741                  * XXX this will go away when the SADB is redone.
742                  */
743                 if (error == ENOENT)
744                         error = 0;
745                 crit_exit();
746                 goto done;
747         } else {
748                 crit_exit();
749
750                 if (error != 0) {
751                         /*
752                          * Hack: -EINVAL is used to signal that a packet
753                          * should be silently discarded.  This is typically
754                          * because we asked key management for an SA and
755                          * it was delayed (e.g. kicked up to IKE).
756                          */
757                         if (error == -EINVAL)
758                                 error = 0;
759                         goto bad;
760                 } else {
761                         /* No IPsec processing for this packet. */
762                 }
763 #ifdef notyet
764                 /*
765                  * If deferred crypto processing is needed, check that
766                  * the interface supports it.
767                  */
768                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
769                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
770                         /* notify IPsec to do its own crypto */
771                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
772                         error = EHOSTUNREACH;
773                         goto bad;
774                 }
775 #endif
776         }
777 spd_done:
778 #endif /* FAST_IPSEC */
779
780         /* We are already being fwd'd from a firewall. */
781         if (next_hop != NULL)
782                 goto pass;
783
784         /* No pfil hooks */
785         if (!pfil_has_hooks(&inet_pfil_hook)) {
786                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
787                         /*
788                          * Strip dummynet tags from stranded packets
789                          */
790                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
791                         KKASSERT(mtag != NULL);
792                         m_tag_delete(m, mtag);
793                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
794                 }
795                 goto pass;
796         }
797
798         /*
799          * IpHack's section.
800          * - Xlate: translate packet's addr/port (NAT).
801          * - Firewall: deny/allow/etc.
802          * - Wrap: fake packet's addr/port <unimpl.>
803          * - Encapsulate: put it in another IP and send out. <unimp.>
804          */
805
806         /*
807          * Run through list of hooks for output packets.
808          */
809         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
810         if (error != 0 || m == NULL)
811                 goto done;
812         ip = mtod(m, struct ip *);
813
814         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
815                 /*
816                  * Check dst to make sure it is directly reachable on the
817                  * interface we previously thought it was.
818                  * If it isn't (which may be likely in some situations) we have
819                  * to re-route it (ie, find a route for the next-hop and the
820                  * associated interface) and set them here. This is nested
821                  * forwarding which in most cases is undesirable, except where
822                  * such control is nigh impossible. So we do it here.
823                  * And I'm babbling.
824                  */
825                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
826                 KKASSERT(mtag != NULL);
827                 next_hop = m_tag_data(mtag);
828
829                 /*
830                  * Try local forwarding first
831                  */
832                 if (ip_localforward(m, next_hop, hlen))
833                         goto done;
834
835                 /*
836                  * Relocate the route based on next_hop.
837                  * If the current route is inp's cache, keep it untouched.
838                  */
839                 if (ro == &iproute && ro->ro_rt != NULL) {
840                         RTFREE(ro->ro_rt);
841                         ro->ro_rt = NULL;
842                 }
843                 ro = &iproute;
844                 bzero(ro, sizeof *ro);
845
846                 /*
847                  * Forwarding to broadcast address is not allowed.
848                  * XXX Should we follow IP_ROUTETOIF?
849                  */
850                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
851
852                 /* We are doing forwarding now */
853                 flags |= IP_FORWARDING;
854
855                 goto reroute;
856         }
857
858         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
859                 struct dn_pkt *dn_pkt;
860
861                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
862                 KKASSERT(mtag != NULL);
863                 dn_pkt = m_tag_data(mtag);
864
865                 /*
866                  * Under certain cases it is not possible to recalculate
867                  * 'ro' and 'dst', let alone 'flags', so just save them in
868                  * dummynet tag and avoid the possible wrong reculcalation
869                  * when we come back to ip_output() again.
870                  *
871                  * All other parameters have been already used and so they
872                  * are not needed anymore.
873                  * XXX if the ifp is deleted while a pkt is in dummynet,
874                  * we are in trouble! (TODO use ifnet_detach_event)
875                  *
876                  * We need to copy *ro because for ICMP pkts (and maybe
877                  * others) the caller passed a pointer into the stack;
878                  * dst might also be a pointer into *ro so it needs to
879                  * be updated.
880                  */
881                 dn_pkt->ro = *ro;
882                 if (ro->ro_rt)
883                         ro->ro_rt->rt_refcnt++;
884                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
885                         /* 'dst' points into 'ro' */
886                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
887                 }
888                 dn_pkt->dn_dst = dst;
889                 dn_pkt->flags = flags;
890
891                 ip_dn_queue(m);
892                 goto done;
893         }
894
895         if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) {
896                 /* ipfw was disabled/unloaded. */
897                 m_freem(m);
898                 goto done;
899         }
900 pass:
901         /* 127/8 must not appear on wire - RFC1122. */
902         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
903             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
904                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
905                         ipstat.ips_badaddr++;
906                         error = EADDRNOTAVAIL;
907                         goto bad;
908                 }
909         }
910         if (ip->ip_src.s_addr == INADDR_ANY ||
911             IN_MULTICAST(ntohl(ip->ip_src.s_addr))) {
912                 ipstat.ips_badaddr++;
913                 error = EADDRNOTAVAIL;
914                 goto bad;
915         }
916
917         if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
918                 m->m_pkthdr.csum_flags |= CSUM_IP;
919                 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
920                 if (sw_csum & CSUM_DELAY_DATA) {
921                         in_delayed_cksum(m);
922                         sw_csum &= ~CSUM_DELAY_DATA;
923                 }
924                 m->m_pkthdr.csum_flags &= ifp->if_hwassist;
925         } else {
926                 sw_csum = 0;
927         }
928         m->m_pkthdr.csum_iphlen = hlen;
929
930         /*
931          * If small enough for interface, or the interface will take
932          * care of the fragmentation or segmentation for us, can just
933          * send directly.
934          */
935         if (ip->ip_len <= ifp->if_mtu ||
936             ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
937             (m->m_pkthdr.csum_flags & CSUM_TSO)) {
938                 ip->ip_len = htons(ip->ip_len);
939                 ip->ip_off = htons(ip->ip_off);
940                 ip->ip_sum = 0;
941                 if (sw_csum & CSUM_DELAY_IP) {
942                         if (ip->ip_vhl == IP_VHL_BORING)
943                                 ip->ip_sum = in_cksum_hdr(ip);
944                         else
945                                 ip->ip_sum = in_cksum(m, hlen);
946                 }
947
948                 /* Record statistics for this interface address. */
949                 if (!(flags & IP_FORWARDING) && ia) {
950                         IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
951                         IFA_STAT_INC(&ia->ia_ifa, obytes, m->m_pkthdr.len);
952                 }
953
954 #ifdef IPSEC
955                 /* clean ipsec history once it goes out of the node */
956                 ipsec_delaux(m);
957 #endif
958
959 #ifdef MBUF_STRESS_TEST
960                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
961                         struct mbuf *m1, *m2;
962                         int length, tmp;
963
964                         tmp = length = m->m_pkthdr.len;
965
966                         while ((length -= mbuf_frag_size) >= 1) {
967                                 m1 = m_split(m, length, M_NOWAIT);
968                                 if (m1 == NULL)
969                                         break;
970                                 m2 = m;
971                                 while (m2->m_next != NULL)
972                                         m2 = m2->m_next;
973                                 m2->m_next = m1;
974                         }
975                         m->m_pkthdr.len = tmp;
976                 }
977 #endif
978
979 #ifdef MPLS
980                 if (!mpls_output_process(m, ro->ro_rt))
981                         goto done;
982 #endif
983                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
984                                        ro->ro_rt);
985                 goto done;
986         }
987
988         if (ip->ip_off & IP_DF) {
989                 error = EMSGSIZE;
990                 /*
991                  * This case can happen if the user changed the MTU
992                  * of an interface after enabling IP on it.  Because
993                  * most netifs don't keep track of routes pointing to
994                  * them, there is no way for one to update all its
995                  * routes when the MTU is changed.
996                  */
997                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
998                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
999                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1000                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1001                 }
1002                 ipstat.ips_cantfrag++;
1003                 goto bad;
1004         }
1005
1006         /*
1007          * Too large for interface; fragment if possible. If successful,
1008          * on return, m will point to a list of packets to be sent.
1009          */
1010         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1011         if (error)
1012                 goto bad;
1013         for (; m; m = m0) {
1014                 m0 = m->m_nextpkt;
1015                 m->m_nextpkt = NULL;
1016 #ifdef IPSEC
1017                 /* clean ipsec history once it goes out of the node */
1018                 ipsec_delaux(m);
1019 #endif
1020                 if (error == 0) {
1021                         /* Record statistics for this interface address. */
1022                         if (ia != NULL) {
1023                                 IFA_STAT_INC(&ia->ia_ifa, opackets, 1);
1024                                 IFA_STAT_INC(&ia->ia_ifa, obytes,
1025                                     m->m_pkthdr.len);
1026                         }
1027 #ifdef MPLS
1028                         if (!mpls_output_process(m, ro->ro_rt))
1029                                 continue;
1030 #endif
1031                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1032                                                ro->ro_rt);
1033                 } else {
1034                         m_freem(m);
1035                 }
1036         }
1037
1038         if (error == 0)
1039                 ipstat.ips_fragmented++;
1040
1041 done:
1042         if (ro == &iproute && ro->ro_rt != NULL) {
1043                 RTFREE(ro->ro_rt);
1044                 ro->ro_rt = NULL;
1045         }
1046 #ifdef IPSEC
1047         if (sp != NULL) {
1048                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1049                         kprintf("DP ip_output call free SP:%p\n", sp));
1050                 key_freesp(sp);
1051         }
1052 #endif
1053 #ifdef FAST_IPSEC
1054         if (sp != NULL)
1055                 KEY_FREESP(&sp);
1056 #endif
1057         return (error);
1058 bad:
1059         m_freem(m);
1060         goto done;
1061 }
1062
1063 /*
1064  * Create a chain of fragments which fit the given mtu. m_frag points to the
1065  * mbuf to be fragmented; on return it points to the chain with the fragments.
1066  * Return 0 if no error. If error, m_frag may contain a partially built
1067  * chain of fragments that should be freed by the caller.
1068  *
1069  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1070  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1071  */
1072 int
1073 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1074             u_long if_hwassist_flags, int sw_csum)
1075 {
1076         int error = 0;
1077         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1078         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1079         int off;
1080         struct mbuf *m0 = *m_frag;      /* the original packet          */
1081         int firstlen;
1082         struct mbuf **mnext;
1083         int nfrags;
1084
1085         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1086                 ipstat.ips_cantfrag++;
1087                 return EMSGSIZE;
1088         }
1089
1090         /*
1091          * Must be able to put at least 8 bytes per fragment.
1092          */
1093         if (len < 8)
1094                 return EMSGSIZE;
1095
1096         /*
1097          * If the interface will not calculate checksums on
1098          * fragmented packets, then do it here.
1099          */
1100         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1101             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1102                 in_delayed_cksum(m0);
1103                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1104         }
1105
1106         if (len > PAGE_SIZE) {
1107                 /*
1108                  * Fragment large datagrams such that each segment
1109                  * contains a multiple of PAGE_SIZE amount of data,
1110                  * plus headers. This enables a receiver to perform
1111                  * page-flipping zero-copy optimizations.
1112                  *
1113                  * XXX When does this help given that sender and receiver
1114                  * could have different page sizes, and also mtu could
1115                  * be less than the receiver's page size ?
1116                  */
1117                 int newlen;
1118                 struct mbuf *m;
1119
1120                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1121                         off += m->m_len;
1122
1123                 /*
1124                  * firstlen (off - hlen) must be aligned on an
1125                  * 8-byte boundary
1126                  */
1127                 if (off < hlen)
1128                         goto smart_frag_failure;
1129                 off = ((off - hlen) & ~7) + hlen;
1130                 newlen = (~PAGE_MASK) & mtu;
1131                 if ((newlen + sizeof(struct ip)) > mtu) {
1132                         /* we failed, go back the default */
1133 smart_frag_failure:
1134                         newlen = len;
1135                         off = hlen + len;
1136                 }
1137                 len = newlen;
1138
1139         } else {
1140                 off = hlen + len;
1141         }
1142
1143         firstlen = off - hlen;
1144         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1145
1146         /*
1147          * Loop through length of segment after first fragment,
1148          * make new header and copy data of each part and link onto chain.
1149          * Here, m0 is the original packet, m is the fragment being created.
1150          * The fragments are linked off the m_nextpkt of the original
1151          * packet, which after processing serves as the first fragment.
1152          */
1153         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1154                 struct ip *mhip;        /* ip header on the fragment */
1155                 struct mbuf *m;
1156                 int mhlen = sizeof(struct ip);
1157
1158                 MGETHDR(m, M_NOWAIT, MT_HEADER);
1159                 if (m == NULL) {
1160                         error = ENOBUFS;
1161                         ipstat.ips_odropped++;
1162                         goto done;
1163                 }
1164                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1165                 /*
1166                  * In the first mbuf, leave room for the link header, then
1167                  * copy the original IP header including options. The payload
1168                  * goes into an additional mbuf chain returned by m_copy().
1169                  */
1170                 m->m_data += max_linkhdr;
1171                 mhip = mtod(m, struct ip *);
1172                 *mhip = *ip;
1173                 if (hlen > sizeof(struct ip)) {
1174                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1175                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1176                 }
1177                 m->m_len = mhlen;
1178                 /* XXX do we need to add ip->ip_off below ? */
1179                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1180                 if (off + len >= ip->ip_len) {  /* last fragment */
1181                         len = ip->ip_len - off;
1182                         m->m_flags |= M_LASTFRAG;
1183                 } else
1184                         mhip->ip_off |= IP_MF;
1185                 mhip->ip_len = htons((u_short)(len + mhlen));
1186                 m->m_next = m_copy(m0, off, len);
1187                 if (m->m_next == NULL) {                /* copy failed */
1188                         m_free(m);
1189                         error = ENOBUFS;        /* ??? */
1190                         ipstat.ips_odropped++;
1191                         goto done;
1192                 }
1193                 m->m_pkthdr.len = mhlen + len;
1194                 m->m_pkthdr.rcvif = NULL;
1195                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1196                 m->m_pkthdr.csum_iphlen = mhlen;
1197                 mhip->ip_off = htons(mhip->ip_off);
1198                 mhip->ip_sum = 0;
1199                 if (sw_csum & CSUM_DELAY_IP)
1200                         mhip->ip_sum = in_cksum(m, mhlen);
1201                 *mnext = m;
1202                 mnext = &m->m_nextpkt;
1203         }
1204         ipstat.ips_ofragments += nfrags;
1205
1206         /* set first marker for fragment chain */
1207         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1208         m0->m_pkthdr.csum_data = nfrags;
1209
1210         /*
1211          * Update first fragment by trimming what's been copied out
1212          * and updating header.
1213          */
1214         m_adj(m0, hlen + firstlen - ip->ip_len);
1215         m0->m_pkthdr.len = hlen + firstlen;
1216         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1217         ip->ip_off |= IP_MF;
1218         ip->ip_off = htons(ip->ip_off);
1219         ip->ip_sum = 0;
1220         if (sw_csum & CSUM_DELAY_IP)
1221                 ip->ip_sum = in_cksum(m0, hlen);
1222
1223 done:
1224         *m_frag = m0;
1225         return error;
1226 }
1227
1228 void
1229 in_delayed_cksum(struct mbuf *m)
1230 {
1231         struct ip *ip;
1232         u_short csum, offset;
1233
1234         ip = mtod(m, struct ip *);
1235         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1236         csum = in_cksum_skip(m, ip->ip_len, offset);
1237         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1238                 csum = 0xffff;
1239         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1240
1241         if (offset + sizeof(u_short) > m->m_len) {
1242                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1243                     m->m_len, offset, ip->ip_p);
1244                 /*
1245                  * XXX
1246                  * this shouldn't happen, but if it does, the
1247                  * correct behavior may be to insert the checksum
1248                  * in the existing chain instead of rearranging it.
1249                  */
1250                 m = m_pullup(m, offset + sizeof(u_short));
1251         }
1252         *(u_short *)(m->m_data + offset) = csum;
1253 }
1254
1255 /*
1256  * Insert IP options into preformed packet.
1257  * Adjust IP destination as required for IP source routing,
1258  * as indicated by a non-zero in_addr at the start of the options.
1259  *
1260  * XXX This routine assumes that the packet has no options in place.
1261  */
1262 static struct mbuf *
1263 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1264 {
1265         struct ipoption *p = mtod(opt, struct ipoption *);
1266         struct mbuf *n;
1267         struct ip *ip = mtod(m, struct ip *);
1268         unsigned optlen;
1269
1270         optlen = opt->m_len - sizeof p->ipopt_dst;
1271         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1272                 *phlen = 0;
1273                 return (m);             /* XXX should fail */
1274         }
1275         if (p->ipopt_dst.s_addr)
1276                 ip->ip_dst = p->ipopt_dst;
1277         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1278                 MGETHDR(n, M_NOWAIT, MT_HEADER);
1279                 if (n == NULL) {
1280                         *phlen = 0;
1281                         return (m);
1282                 }
1283                 n->m_pkthdr.rcvif = NULL;
1284                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1285                 m->m_len -= sizeof(struct ip);
1286                 m->m_data += sizeof(struct ip);
1287                 n->m_next = m;
1288                 m = n;
1289                 m->m_len = optlen + sizeof(struct ip);
1290                 m->m_data += max_linkhdr;
1291                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1292         } else {
1293                 m->m_data -= optlen;
1294                 m->m_len += optlen;
1295                 m->m_pkthdr.len += optlen;
1296                 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1297         }
1298         ip = mtod(m, struct ip *);
1299         bcopy(p->ipopt_list, ip + 1, optlen);
1300         *phlen = sizeof(struct ip) + optlen;
1301         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1302         ip->ip_len += optlen;
1303         return (m);
1304 }
1305
1306 /*
1307  * Copy options from ip to jp,
1308  * omitting those not copied during fragmentation.
1309  */
1310 int
1311 ip_optcopy(struct ip *ip, struct ip *jp)
1312 {
1313         u_char *cp, *dp;
1314         int opt, optlen, cnt;
1315
1316         cp = (u_char *)(ip + 1);
1317         dp = (u_char *)(jp + 1);
1318         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1319         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1320                 opt = cp[0];
1321                 if (opt == IPOPT_EOL)
1322                         break;
1323                 if (opt == IPOPT_NOP) {
1324                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1325                         *dp++ = IPOPT_NOP;
1326                         optlen = 1;
1327                         continue;
1328                 }
1329
1330                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1331                     ("ip_optcopy: malformed ipv4 option"));
1332                 optlen = cp[IPOPT_OLEN];
1333                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1334                     ("ip_optcopy: malformed ipv4 option"));
1335
1336                 /* bogus lengths should have been caught by ip_dooptions */
1337                 if (optlen > cnt)
1338                         optlen = cnt;
1339                 if (IPOPT_COPIED(opt)) {
1340                         bcopy(cp, dp, optlen);
1341                         dp += optlen;
1342                 }
1343         }
1344         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1345                 *dp++ = IPOPT_EOL;
1346         return (optlen);
1347 }
1348
1349 /*
1350  * IP socket option processing.
1351  */
1352 void
1353 ip_ctloutput(netmsg_t msg)
1354 {
1355         struct socket *so = msg->base.nm_so;
1356         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1357         struct  inpcb *inp = so->so_pcb;
1358         int     error, optval;
1359
1360         error = optval = 0;
1361
1362         /* Get socket's owner cpuid hint */
1363         if (sopt->sopt_level == SOL_SOCKET &&
1364             sopt->sopt_dir == SOPT_GET &&
1365             sopt->sopt_name == SO_CPUHINT) {
1366                 optval = mycpuid;
1367                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1368                 goto done;
1369         }
1370
1371         if (sopt->sopt_level != IPPROTO_IP) {
1372                 error = EINVAL;
1373                 goto done;
1374         }
1375
1376         switch (sopt->sopt_name) {
1377         case IP_MULTICAST_IF:
1378         case IP_MULTICAST_VIF:
1379         case IP_MULTICAST_TTL:
1380         case IP_MULTICAST_LOOP:
1381         case IP_ADD_MEMBERSHIP:
1382         case IP_DROP_MEMBERSHIP:
1383                 /*
1384                  * Handle multicast options in netisr0
1385                  */
1386                 if (&curthread->td_msgport != netisr_cpuport(0)) {
1387                         /* NOTE: so_port MUST NOT be checked in netisr0 */
1388                         msg->lmsg.ms_flags |= MSGF_IGNSOPORT;
1389                         lwkt_forwardmsg(netisr_cpuport(0), &msg->lmsg);
1390                         return;
1391                 }
1392                 break;
1393         }
1394
1395         switch (sopt->sopt_dir) {
1396         case SOPT_SET:
1397                 switch (sopt->sopt_name) {
1398                 case IP_OPTIONS:
1399 #ifdef notyet
1400                 case IP_RETOPTS:
1401 #endif
1402                 {
1403                         struct mbuf *m;
1404                         if (sopt->sopt_valsize > MLEN) {
1405                                 error = EMSGSIZE;
1406                                 break;
1407                         }
1408                         MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_HEADER);
1409                         if (m == NULL) {
1410                                 error = ENOBUFS;
1411                                 break;
1412                         }
1413                         m->m_len = sopt->sopt_valsize;
1414                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1415                                               m->m_len);
1416                         error = ip_pcbopts(sopt->sopt_name,
1417                                            &inp->inp_options, m);
1418                         goto done;
1419                 }
1420
1421                 case IP_TOS:
1422                 case IP_TTL:
1423                 case IP_MINTTL:
1424                 case IP_RECVOPTS:
1425                 case IP_RECVRETOPTS:
1426                 case IP_RECVDSTADDR:
1427                 case IP_RECVIF:
1428                 case IP_RECVTTL:
1429                 case IP_FAITH:
1430                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1431                                              sizeof optval);
1432                         if (error)
1433                                 break;
1434                         switch (sopt->sopt_name) {
1435                         case IP_TOS:
1436                                 inp->inp_ip_tos = optval;
1437                                 break;
1438
1439                         case IP_TTL:
1440                                 inp->inp_ip_ttl = optval;
1441                                 break;
1442                         case IP_MINTTL:
1443                                 if (optval >= 0 && optval <= MAXTTL)
1444                                         inp->inp_ip_minttl = optval;
1445                                 else
1446                                         error = EINVAL;
1447                                 break;
1448 #define OPTSET(bit) \
1449         if (optval) \
1450                 inp->inp_flags |= bit; \
1451         else \
1452                 inp->inp_flags &= ~bit;
1453
1454                         case IP_RECVOPTS:
1455                                 OPTSET(INP_RECVOPTS);
1456                                 break;
1457
1458                         case IP_RECVRETOPTS:
1459                                 OPTSET(INP_RECVRETOPTS);
1460                                 break;
1461
1462                         case IP_RECVDSTADDR:
1463                                 OPTSET(INP_RECVDSTADDR);
1464                                 break;
1465
1466                         case IP_RECVIF:
1467                                 OPTSET(INP_RECVIF);
1468                                 break;
1469
1470                         case IP_RECVTTL:
1471                                 OPTSET(INP_RECVTTL);
1472                                 break;
1473
1474                         case IP_FAITH:
1475                                 OPTSET(INP_FAITH);
1476                                 break;
1477                         }
1478                         break;
1479 #undef OPTSET
1480
1481                 case IP_MULTICAST_IF:
1482                 case IP_MULTICAST_VIF:
1483                 case IP_MULTICAST_TTL:
1484                 case IP_MULTICAST_LOOP:
1485                 case IP_ADD_MEMBERSHIP:
1486                 case IP_DROP_MEMBERSHIP:
1487                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1488                         break;
1489
1490                 case IP_PORTRANGE:
1491                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1492                                             sizeof optval);
1493                         if (error)
1494                                 break;
1495
1496                         switch (optval) {
1497                         case IP_PORTRANGE_DEFAULT:
1498                                 inp->inp_flags &= ~(INP_LOWPORT);
1499                                 inp->inp_flags &= ~(INP_HIGHPORT);
1500                                 break;
1501
1502                         case IP_PORTRANGE_HIGH:
1503                                 inp->inp_flags &= ~(INP_LOWPORT);
1504                                 inp->inp_flags |= INP_HIGHPORT;
1505                                 break;
1506
1507                         case IP_PORTRANGE_LOW:
1508                                 inp->inp_flags &= ~(INP_HIGHPORT);
1509                                 inp->inp_flags |= INP_LOWPORT;
1510                                 break;
1511
1512                         default:
1513                                 error = EINVAL;
1514                                 break;
1515                         }
1516                         break;
1517
1518 #if defined(IPSEC) || defined(FAST_IPSEC)
1519                 case IP_IPSEC_POLICY:
1520                 {
1521                         caddr_t req;
1522                         size_t len = 0;
1523                         int priv;
1524                         struct mbuf *m;
1525                         int optname;
1526
1527                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1528                                 break;
1529                         soopt_to_mbuf(sopt, m);
1530                         priv = (sopt->sopt_td != NULL &&
1531                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1532                         req = mtod(m, caddr_t);
1533                         len = m->m_len;
1534                         optname = sopt->sopt_name;
1535                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1536                         m_freem(m);
1537                         break;
1538                 }
1539 #endif /*IPSEC*/
1540
1541                 default:
1542                         error = ENOPROTOOPT;
1543                         break;
1544                 }
1545                 break;
1546
1547         case SOPT_GET:
1548                 switch (sopt->sopt_name) {
1549                 case IP_OPTIONS:
1550                 case IP_RETOPTS:
1551                         if (inp->inp_options)
1552                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1553                                                            char *),
1554                                                 inp->inp_options->m_len);
1555                         else
1556                                 sopt->sopt_valsize = 0;
1557                         break;
1558
1559                 case IP_TOS:
1560                 case IP_TTL:
1561                 case IP_MINTTL:
1562                 case IP_RECVOPTS:
1563                 case IP_RECVRETOPTS:
1564                 case IP_RECVDSTADDR:
1565                 case IP_RECVTTL:
1566                 case IP_RECVIF:
1567                 case IP_PORTRANGE:
1568                 case IP_FAITH:
1569                         switch (sopt->sopt_name) {
1570
1571                         case IP_TOS:
1572                                 optval = inp->inp_ip_tos;
1573                                 break;
1574
1575                         case IP_TTL:
1576                                 optval = inp->inp_ip_ttl;
1577                                 break;
1578                         case IP_MINTTL:
1579                                 optval = inp->inp_ip_minttl;
1580                                 break;
1581
1582 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1583
1584                         case IP_RECVOPTS:
1585                                 optval = OPTBIT(INP_RECVOPTS);
1586                                 break;
1587
1588                         case IP_RECVRETOPTS:
1589                                 optval = OPTBIT(INP_RECVRETOPTS);
1590                                 break;
1591
1592                         case IP_RECVDSTADDR:
1593                                 optval = OPTBIT(INP_RECVDSTADDR);
1594                                 break;
1595
1596                         case IP_RECVTTL:
1597                                 optval = OPTBIT(INP_RECVTTL);
1598                                 break;
1599
1600                         case IP_RECVIF:
1601                                 optval = OPTBIT(INP_RECVIF);
1602                                 break;
1603
1604                         case IP_PORTRANGE:
1605                                 if (inp->inp_flags & INP_HIGHPORT)
1606                                         optval = IP_PORTRANGE_HIGH;
1607                                 else if (inp->inp_flags & INP_LOWPORT)
1608                                         optval = IP_PORTRANGE_LOW;
1609                                 else
1610                                         optval = 0;
1611                                 break;
1612
1613                         case IP_FAITH:
1614                                 optval = OPTBIT(INP_FAITH);
1615                                 break;
1616                         }
1617                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1618                         break;
1619
1620                 case IP_MULTICAST_IF:
1621                 case IP_MULTICAST_VIF:
1622                 case IP_MULTICAST_TTL:
1623                 case IP_MULTICAST_LOOP:
1624                 case IP_ADD_MEMBERSHIP:
1625                 case IP_DROP_MEMBERSHIP:
1626                         error = ip_getmoptions(sopt, inp->inp_moptions);
1627                         break;
1628
1629 #if defined(IPSEC) || defined(FAST_IPSEC)
1630                 case IP_IPSEC_POLICY:
1631                 {
1632                         struct mbuf *m = NULL;
1633                         caddr_t req = NULL;
1634                         size_t len = 0;
1635
1636                         if (m != NULL) {
1637                                 req = mtod(m, caddr_t);
1638                                 len = m->m_len;
1639                         }
1640                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1641                         if (error == 0)
1642                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1643                         if (error == 0)
1644                                 m_freem(m);
1645                         break;
1646                 }
1647 #endif /*IPSEC*/
1648
1649                 default:
1650                         error = ENOPROTOOPT;
1651                         break;
1652                 }
1653                 break;
1654         }
1655 done:
1656         lwkt_replymsg(&msg->lmsg, error);
1657 }
1658
1659 /*
1660  * Set up IP options in pcb for insertion in output packets.
1661  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1662  * with destination address if source routed.
1663  */
1664 static int
1665 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1666 {
1667         int cnt, optlen;
1668         u_char *cp;
1669         u_char opt;
1670
1671         /* turn off any old options */
1672         if (*pcbopt)
1673                 m_free(*pcbopt);
1674         *pcbopt = NULL;
1675         if (m == NULL || m->m_len == 0) {
1676                 /*
1677                  * Only turning off any previous options.
1678                  */
1679                 if (m != NULL)
1680                         m_free(m);
1681                 return (0);
1682         }
1683
1684         if (m->m_len % sizeof(int32_t))
1685                 goto bad;
1686         /*
1687          * IP first-hop destination address will be stored before
1688          * actual options; move other options back
1689          * and clear it when none present.
1690          */
1691         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1692                 goto bad;
1693         cnt = m->m_len;
1694         m->m_len += sizeof(struct in_addr);
1695         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1696         bcopy(mtod(m, caddr_t), cp, cnt);
1697         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1698
1699         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1700                 opt = cp[IPOPT_OPTVAL];
1701                 if (opt == IPOPT_EOL)
1702                         break;
1703                 if (opt == IPOPT_NOP)
1704                         optlen = 1;
1705                 else {
1706                         if (cnt < IPOPT_OLEN + sizeof *cp)
1707                                 goto bad;
1708                         optlen = cp[IPOPT_OLEN];
1709                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1710                                 goto bad;
1711                 }
1712                 switch (opt) {
1713
1714                 default:
1715                         break;
1716
1717                 case IPOPT_LSRR:
1718                 case IPOPT_SSRR:
1719                         /*
1720                          * user process specifies route as:
1721                          *      ->A->B->C->D
1722                          * D must be our final destination (but we can't
1723                          * check that since we may not have connected yet).
1724                          * A is first hop destination, which doesn't appear in
1725                          * actual IP option, but is stored before the options.
1726                          */
1727                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1728                                 goto bad;
1729                         m->m_len -= sizeof(struct in_addr);
1730                         cnt -= sizeof(struct in_addr);
1731                         optlen -= sizeof(struct in_addr);
1732                         cp[IPOPT_OLEN] = optlen;
1733                         /*
1734                          * Move first hop before start of options.
1735                          */
1736                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1737                               sizeof(struct in_addr));
1738                         /*
1739                          * Then copy rest of options back
1740                          * to close up the deleted entry.
1741                          */
1742                         bcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1743                               &cp[IPOPT_OFFSET+1],
1744                               cnt - (IPOPT_MINOFF - 1));
1745                         break;
1746                 }
1747         }
1748         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1749                 goto bad;
1750         *pcbopt = m;
1751         return (0);
1752
1753 bad:
1754         m_free(m);
1755         return (EINVAL);
1756 }
1757
1758 /*
1759  * XXX
1760  * The whole multicast option thing needs to be re-thought.
1761  * Several of these options are equally applicable to non-multicast
1762  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1763  * standard option (IP_TTL).
1764  */
1765
1766 /*
1767  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1768  */
1769 static struct ifnet *
1770 ip_multicast_if(struct in_addr *a, int *ifindexp)
1771 {
1772         int ifindex;
1773         struct ifnet *ifp;
1774
1775         if (ifindexp)
1776                 *ifindexp = 0;
1777         if (ntohl(a->s_addr) >> 24 == 0) {
1778                 ifindex = ntohl(a->s_addr) & 0xffffff;
1779                 if (ifindex < 0 || if_index < ifindex)
1780                         return NULL;
1781                 ifp = ifindex2ifnet[ifindex];
1782                 if (ifindexp)
1783                         *ifindexp = ifindex;
1784         } else {
1785                 ifp = INADDR_TO_IFP(a);
1786         }
1787         return ifp;
1788 }
1789
1790 /*
1791  * Set the IP multicast options in response to user setsockopt().
1792  */
1793 static int
1794 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1795 {
1796         int error = 0;
1797         int i;
1798         struct in_addr addr;
1799         struct ip_mreq mreq;
1800         struct ifnet *ifp;
1801         struct ip_moptions *imo = *imop;
1802         int ifindex;
1803
1804         if (imo == NULL) {
1805                 /*
1806                  * No multicast option buffer attached to the pcb;
1807                  * allocate one and initialize to default values.
1808                  */
1809                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1810
1811                 imo->imo_multicast_ifp = NULL;
1812                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1813                 imo->imo_multicast_vif = -1;
1814                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1815                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1816                 imo->imo_num_memberships = 0;
1817                 /* Assign imo to imop after all fields are setup */
1818                 cpu_sfence();
1819                 *imop = imo;
1820         }
1821         switch (sopt->sopt_name) {
1822         /* store an index number for the vif you wanna use in the send */
1823         case IP_MULTICAST_VIF:
1824                 if (legal_vif_num == 0) {
1825                         error = EOPNOTSUPP;
1826                         break;
1827                 }
1828                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1829                 if (error)
1830                         break;
1831                 if (!legal_vif_num(i) && (i != -1)) {
1832                         error = EINVAL;
1833                         break;
1834                 }
1835                 imo->imo_multicast_vif = i;
1836                 break;
1837
1838         case IP_MULTICAST_IF:
1839                 /*
1840                  * Select the interface for outgoing multicast packets.
1841                  */
1842                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1843                 if (error)
1844                         break;
1845
1846                 /*
1847                  * INADDR_ANY is used to remove a previous selection.
1848                  * When no interface is selected, a default one is
1849                  * chosen every time a multicast packet is sent.
1850                  */
1851                 if (addr.s_addr == INADDR_ANY) {
1852                         imo->imo_multicast_ifp = NULL;
1853                         break;
1854                 }
1855                 /*
1856                  * The selected interface is identified by its local
1857                  * IP address.  Find the interface and confirm that
1858                  * it supports multicasting.
1859                  */
1860                 crit_enter();
1861                 ifp = ip_multicast_if(&addr, &ifindex);
1862                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1863                         crit_exit();
1864                         error = EADDRNOTAVAIL;
1865                         break;
1866                 }
1867                 imo->imo_multicast_ifp = ifp;
1868                 if (ifindex)
1869                         imo->imo_multicast_addr = addr;
1870                 else
1871                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1872                 crit_exit();
1873                 break;
1874
1875         case IP_MULTICAST_TTL:
1876                 /*
1877                  * Set the IP time-to-live for outgoing multicast packets.
1878                  * The original multicast API required a char argument,
1879                  * which is inconsistent with the rest of the socket API.
1880                  * We allow either a char or an int.
1881                  */
1882                 if (sopt->sopt_valsize == 1) {
1883                         u_char ttl;
1884                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1885                         if (error)
1886                                 break;
1887                         imo->imo_multicast_ttl = ttl;
1888                 } else {
1889                         u_int ttl;
1890                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1891                         if (error)
1892                                 break;
1893                         if (ttl > 255)
1894                                 error = EINVAL;
1895                         else
1896                                 imo->imo_multicast_ttl = ttl;
1897                 }
1898                 break;
1899
1900         case IP_MULTICAST_LOOP:
1901                 /*
1902                  * Set the loopback flag for outgoing multicast packets.
1903                  * Must be zero or one.  The original multicast API required a
1904                  * char argument, which is inconsistent with the rest
1905                  * of the socket API.  We allow either a char or an int.
1906                  */
1907                 if (sopt->sopt_valsize == 1) {
1908                         u_char loop;
1909
1910                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1911                         if (error)
1912                                 break;
1913                         imo->imo_multicast_loop = !!loop;
1914                 } else {
1915                         u_int loop;
1916
1917                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1918                                             sizeof loop);
1919                         if (error)
1920                                 break;
1921                         imo->imo_multicast_loop = !!loop;
1922                 }
1923                 break;
1924
1925         case IP_ADD_MEMBERSHIP:
1926                 /*
1927                  * Add a multicast group membership.
1928                  * Group must be a valid IP multicast address.
1929                  */
1930                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1931                 if (error)
1932                         break;
1933
1934                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1935                         error = EINVAL;
1936                         break;
1937                 }
1938                 crit_enter();
1939                 /*
1940                  * If no interface address was provided, use the interface of
1941                  * the route to the given multicast address.
1942                  */
1943                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1944                         struct sockaddr_in dst;
1945                         struct rtentry *rt;
1946
1947                         bzero(&dst, sizeof(struct sockaddr_in));
1948                         dst.sin_len = sizeof(struct sockaddr_in);
1949                         dst.sin_family = AF_INET;
1950                         dst.sin_addr = mreq.imr_multiaddr;
1951                         rt = rtlookup((struct sockaddr *)&dst);
1952                         if (rt == NULL) {
1953                                 error = EADDRNOTAVAIL;
1954                                 crit_exit();
1955                                 break;
1956                         }
1957                         --rt->rt_refcnt;
1958                         ifp = rt->rt_ifp;
1959                 } else {
1960                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1961                 }
1962
1963                 /*
1964                  * See if we found an interface, and confirm that it
1965                  * supports multicast.
1966                  */
1967                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1968                         error = EADDRNOTAVAIL;
1969                         crit_exit();
1970                         break;
1971                 }
1972                 /*
1973                  * See if the membership already exists or if all the
1974                  * membership slots are full.
1975                  */
1976                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1977                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1978                             imo->imo_membership[i]->inm_addr.s_addr
1979                                                 == mreq.imr_multiaddr.s_addr)
1980                                 break;
1981                 }
1982                 if (i < imo->imo_num_memberships) {
1983                         error = EADDRINUSE;
1984                         crit_exit();
1985                         break;
1986                 }
1987                 if (i == IP_MAX_MEMBERSHIPS) {
1988                         error = ETOOMANYREFS;
1989                         crit_exit();
1990                         break;
1991                 }
1992                 /*
1993                  * Everything looks good; add a new record to the multicast
1994                  * address list for the given interface.
1995                  */
1996                 if ((imo->imo_membership[i] =
1997                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1998                         error = ENOBUFS;
1999                         crit_exit();
2000                         break;
2001                 }
2002                 ++imo->imo_num_memberships;
2003                 crit_exit();
2004                 break;
2005
2006         case IP_DROP_MEMBERSHIP:
2007                 /*
2008                  * Drop a multicast group membership.
2009                  * Group must be a valid IP multicast address.
2010                  */
2011                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
2012                 if (error)
2013                         break;
2014
2015                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2016                         error = EINVAL;
2017                         break;
2018                 }
2019
2020                 crit_enter();
2021                 /*
2022                  * If an interface address was specified, get a pointer
2023                  * to its ifnet structure.
2024                  */
2025                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2026                         ifp = NULL;
2027                 else {
2028                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2029                         if (ifp == NULL) {
2030                                 error = EADDRNOTAVAIL;
2031                                 crit_exit();
2032                                 break;
2033                         }
2034                 }
2035                 /*
2036                  * Find the membership in the membership array.
2037                  */
2038                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2039                         if ((ifp == NULL ||
2040                              imo->imo_membership[i]->inm_ifp == ifp) &&
2041                             imo->imo_membership[i]->inm_addr.s_addr ==
2042                             mreq.imr_multiaddr.s_addr)
2043                                 break;
2044                 }
2045                 if (i == imo->imo_num_memberships) {
2046                         error = EADDRNOTAVAIL;
2047                         crit_exit();
2048                         break;
2049                 }
2050                 /*
2051                  * Give up the multicast address record to which the
2052                  * membership points.
2053                  */
2054                 in_delmulti(imo->imo_membership[i]);
2055                 /*
2056                  * Remove the gap in the membership array.
2057                  */
2058                 for (++i; i < imo->imo_num_memberships; ++i)
2059                         imo->imo_membership[i-1] = imo->imo_membership[i];
2060                 --imo->imo_num_memberships;
2061                 crit_exit();
2062                 break;
2063
2064         default:
2065                 error = EOPNOTSUPP;
2066                 break;
2067         }
2068
2069         return (error);
2070 }
2071
2072 /*
2073  * Return the IP multicast options in response to user getsockopt().
2074  */
2075 static int
2076 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2077 {
2078         struct in_addr addr;
2079         struct in_ifaddr *ia;
2080         int error, optval;
2081         u_char coptval;
2082
2083         error = 0;
2084         switch (sopt->sopt_name) {
2085         case IP_MULTICAST_VIF:
2086                 if (imo != NULL)
2087                         optval = imo->imo_multicast_vif;
2088                 else
2089                         optval = -1;
2090                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2091                 break;
2092
2093         case IP_MULTICAST_IF:
2094                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2095                         addr.s_addr = INADDR_ANY;
2096                 else if (imo->imo_multicast_addr.s_addr) {
2097                         /* return the value user has set */
2098                         addr = imo->imo_multicast_addr;
2099                 } else {
2100                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2101                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2102                                 : IA_SIN(ia)->sin_addr.s_addr;
2103                 }
2104                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2105                 break;
2106
2107         case IP_MULTICAST_TTL:
2108                 if (imo == NULL)
2109                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2110                 else
2111                         optval = coptval = imo->imo_multicast_ttl;
2112                 if (sopt->sopt_valsize == 1)
2113                         soopt_from_kbuf(sopt, &coptval, 1);
2114                 else
2115                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2116                 break;
2117
2118         case IP_MULTICAST_LOOP:
2119                 if (imo == NULL)
2120                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2121                 else
2122                         optval = coptval = imo->imo_multicast_loop;
2123                 if (sopt->sopt_valsize == 1)
2124                         soopt_from_kbuf(sopt, &coptval, 1);
2125                 else
2126                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2127                 break;
2128
2129         default:
2130                 error = ENOPROTOOPT;
2131                 break;
2132         }
2133         return (error);
2134 }
2135
2136 /*
2137  * Discard the IP multicast options.
2138  */
2139 void
2140 ip_freemoptions(struct ip_moptions *imo)
2141 {
2142         int i;
2143
2144         if (imo != NULL) {
2145                 for (i = 0; i < imo->imo_num_memberships; ++i)
2146                         in_delmulti(imo->imo_membership[i]);
2147                 kfree(imo, M_IPMOPTS);
2148         }
2149 }
2150
2151 /*
2152  * Routine called from ip_output() to loop back a copy of an IP multicast
2153  * packet to the input queue of a specified interface.  Note that this
2154  * calls the output routine of the loopback "driver", but with an interface
2155  * pointer that might NOT be a loopback interface -- evil, but easier than
2156  * replicating that code here.
2157  */
2158 static void
2159 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2160              int hlen)
2161 {
2162         struct ip *ip;
2163         struct mbuf *copym;
2164
2165         copym = m_copypacket(m, M_NOWAIT);
2166         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2167                 copym = m_pullup(copym, hlen);
2168         if (copym != NULL) {
2169                 /*
2170                  * if the checksum hasn't been computed, mark it as valid
2171                  */
2172                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2173                         in_delayed_cksum(copym);
2174                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2175                         copym->m_pkthdr.csum_flags |=
2176                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2177                         copym->m_pkthdr.csum_data = 0xffff;
2178                 }
2179                 /*
2180                  * We don't bother to fragment if the IP length is greater
2181                  * than the interface's MTU.  Can this possibly matter?
2182                  */
2183                 ip = mtod(copym, struct ip *);
2184                 ip->ip_len = htons(ip->ip_len);
2185                 ip->ip_off = htons(ip->ip_off);
2186                 ip->ip_sum = 0;
2187                 if (ip->ip_vhl == IP_VHL_BORING) {
2188                         ip->ip_sum = in_cksum_hdr(ip);
2189                 } else {
2190                         ip->ip_sum = in_cksum(copym, hlen);
2191                 }
2192                 /*
2193                  * NB:
2194                  * It's not clear whether there are any lingering
2195                  * reentrancy problems in other areas which might
2196                  * be exposed by using ip_input directly (in
2197                  * particular, everything which modifies the packet
2198                  * in-place).  Yet another option is using the
2199                  * protosw directly to deliver the looped back
2200                  * packet.  For the moment, we'll err on the side
2201                  * of safety by using if_simloop().
2202                  */
2203 #if 1 /* XXX */
2204                 if (dst->sin_family != AF_INET) {
2205                         kprintf("ip_mloopback: bad address family %d\n",
2206                                                 dst->sin_family);
2207                         dst->sin_family = AF_INET;
2208                 }
2209 #endif
2210                 if_simloop(ifp, copym, dst->sin_family, 0);
2211         }
2212 }