042a9b40281d0efc11aa0a12c263049748732bbc
[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 struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
109 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
110 static void     ip_mloopback
111         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
112 static int      ip_getmoptions
113         (struct sockopt *, struct ip_moptions *);
114 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
115 static int      ip_setmoptions
116         (struct sockopt *, struct ip_moptions **);
117
118 int     ip_optcopy(struct ip *, struct ip *);
119
120 extern  int route_assert_owner_access;
121
122 extern  struct protosw inetsw[];
123
124 static int
125 ip_localforward(struct mbuf *m, const struct sockaddr_in *dst, int hlen)
126 {
127         struct in_ifaddr_container *iac;
128
129         /*
130          * We need to figure out if we have been forwarded to a local
131          * socket.  If so, then we should somehow "loop back" to
132          * ip_input(), and get directed to the PCB as if we had received
133          * this packet.  This is because it may be difficult to identify
134          * the packets you want to forward until they are being output
135          * and have selected an interface (e.g. locally initiated
136          * packets).  If we used the loopback inteface, we would not be
137          * able to control what happens as the packet runs through
138          * ip_input() as it is done through a ISR.
139          */
140         LIST_FOREACH(iac, INADDR_HASH(dst->sin_addr.s_addr), ia_hash) {
141                 /*
142                  * If the addr to forward to is one of ours, we pretend
143                  * to be the destination for this packet.
144                  */
145                 if (IA_SIN(iac->ia)->sin_addr.s_addr == dst->sin_addr.s_addr)
146                         break;
147         }
148         if (iac != NULL) {
149                 struct ip *ip;
150
151                 if (m->m_pkthdr.rcvif == NULL)
152                         m->m_pkthdr.rcvif = ifunit("lo0");
153                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
154                         m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
155                                                   CSUM_PSEUDO_HDR;
156                         m->m_pkthdr.csum_data = 0xffff;
157                 }
158                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
159
160                 /*
161                  * Make sure that the IP header is in one mbuf,
162                  * required by ip_input
163                  */
164                 if (m->m_len < hlen) {
165                         m = m_pullup(m, hlen);
166                         if (m == NULL) {
167                                 /* The packet was freed; we are done */
168                                 return 1;
169                         }
170                 }
171                 ip = mtod(m, struct ip *);
172
173                 ip->ip_len = htons(ip->ip_len);
174                 ip->ip_off = htons(ip->ip_off);
175                 ip_input(m);
176
177                 return 1; /* The packet gets forwarded locally */
178         }
179         return 0;
180 }
181
182 /*
183  * IP output.  The packet in mbuf chain m contains a skeletal IP
184  * header (with len, off, ttl, proto, tos, src, dst).
185  * The mbuf chain containing the packet will be freed.
186  * The mbuf opt, if present, will not be freed.
187  */
188 int
189 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
190           int flags, struct ip_moptions *imo, struct inpcb *inp)
191 {
192         struct ip *ip;
193         struct ifnet *ifp = NULL;       /* keep compiler happy */
194         struct mbuf *m;
195         int hlen = sizeof(struct ip);
196         int len, error = 0;
197         struct sockaddr_in *dst = NULL; /* keep compiler happy */
198         struct in_ifaddr *ia = NULL;
199         int isbroadcast, sw_csum;
200         struct in_addr pkt_dst;
201         struct route iproute;
202         struct m_tag *mtag;
203 #ifdef IPSEC
204         struct secpolicy *sp = NULL;
205         struct socket *so = inp ? inp->inp_socket : NULL;
206 #endif
207 #ifdef FAST_IPSEC
208         struct secpolicy *sp = NULL;
209         struct tdb_ident *tdbi;
210 #endif /* FAST_IPSEC */
211         struct sockaddr_in *next_hop = NULL;
212         int src_was_INADDR_ANY = 0;     /* as the name says... */
213
214         m = m0;
215         M_ASSERTPKTHDR(m);
216
217         if (ro == NULL) {
218                 ro = &iproute;
219                 bzero(ro, sizeof *ro);
220         } else if (ro->ro_rt != NULL && ro->ro_rt->rt_cpuid != mycpuid) {
221                 if (flags & IP_DEBUGROUTE) {
222                         if (route_assert_owner_access) {
223                                 panic("ip_output: "
224                                       "rt rt_cpuid %d accessed on cpu %d\n",
225                                       ro->ro_rt->rt_cpuid, mycpuid);
226                         } else {
227                                 kprintf("ip_output: "
228                                         "rt rt_cpuid %d accessed on cpu %d\n",
229                                         ro->ro_rt->rt_cpuid, mycpuid);
230                                 print_backtrace(-1);
231                         }
232                 }
233
234                 /*
235                  * XXX
236                  * If the cached rtentry's owner CPU is not the current CPU,
237                  * then don't touch the cached rtentry (remote free is too
238                  * expensive in this context); just relocate the route.
239                  */
240                 ro = &iproute;
241                 bzero(ro, sizeof *ro);
242         }
243
244         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
245                 /* Next hop */
246                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
247                 KKASSERT(mtag != NULL);
248                 next_hop = m_tag_data(mtag);
249         }
250
251         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
252                 struct dn_pkt *dn_pkt;
253
254                 /* Extract info from dummynet tag */
255                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
256                 KKASSERT(mtag != NULL);
257                 dn_pkt = m_tag_data(mtag);
258
259                 /*
260                  * The packet was already tagged, so part of the
261                  * processing was already done, and we need to go down.
262                  * Get the calculated parameters from the tag.
263                  */
264                 ifp = dn_pkt->ifp;
265
266                 KKASSERT(ro == &iproute);
267                 *ro = dn_pkt->ro; /* structure copy */
268                 KKASSERT(ro->ro_rt == NULL || ro->ro_rt->rt_cpuid == mycpuid);
269
270                 dst = dn_pkt->dn_dst;
271                 if (dst == (struct sockaddr_in *)&(dn_pkt->ro.ro_dst)) {
272                         /* If 'dst' points into dummynet tag, adjust it */
273                         dst = (struct sockaddr_in *)&(ro->ro_dst);
274                 }
275
276                 ip = mtod(m, struct ip *);
277                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
278                 if (ro->ro_rt)
279                         ia = ifatoia(ro->ro_rt->rt_ifa);
280                 goto sendit;
281         }
282
283         if (opt) {
284                 len = 0;
285                 m = ip_insertoptions(m, opt, &len);
286                 if (len != 0)
287                         hlen = len;
288         }
289         ip = mtod(m, struct ip *);
290
291         /*
292          * Fill in IP header.
293          */
294         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
295                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
296                 ip->ip_off &= IP_DF;
297                 ip->ip_id = ip_newid();
298                 ipstat.ips_localout++;
299         } else {
300                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
301         }
302
303 reroute:
304         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
305
306         dst = (struct sockaddr_in *)&ro->ro_dst;
307         /*
308          * If there is a cached route,
309          * check that it is to the same destination
310          * and is still up.  If not, free it and try again.
311          * The address family should also be checked in case of sharing the
312          * cache with IPv6.
313          */
314         if (ro->ro_rt &&
315             (!(ro->ro_rt->rt_flags & RTF_UP) ||
316              dst->sin_family != AF_INET ||
317              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
318                 rtfree(ro->ro_rt);
319                 ro->ro_rt = NULL;
320         }
321         if (ro->ro_rt == NULL) {
322                 bzero(dst, sizeof *dst);
323                 dst->sin_family = AF_INET;
324                 dst->sin_len = sizeof *dst;
325                 dst->sin_addr = pkt_dst;
326         }
327         /*
328          * If routing to interface only,
329          * short circuit routing lookup.
330          */
331         if (flags & IP_ROUTETOIF) {
332                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
333                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
334                         ipstat.ips_noroute++;
335                         error = ENETUNREACH;
336                         goto bad;
337                 }
338                 ifp = ia->ia_ifp;
339                 ip->ip_ttl = 1;
340                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
341         } else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
342                    imo != NULL && imo->imo_multicast_ifp != NULL) {
343                 /*
344                  * Bypass the normal routing lookup for multicast
345                  * packets if the interface is specified.
346                  */
347                 ifp = imo->imo_multicast_ifp;
348                 ia = IFP_TO_IA(ifp);
349                 isbroadcast = 0;        /* fool gcc */
350         } else {
351                 /*
352                  * If this is the case, we probably don't want to allocate
353                  * a protocol-cloned route since we didn't get one from the
354                  * ULP.  This lets TCP do its thing, while not burdening
355                  * forwarding or ICMP with the overhead of cloning a route.
356                  * Of course, we still want to do any cloning requested by
357                  * the link layer, as this is probably required in all cases
358                  * for correct operation (as it is for ARP).
359                  */
360                 if (ro->ro_rt == NULL)
361                         rtalloc_ign(ro, RTF_PRCLONING);
362                 if (ro->ro_rt == NULL) {
363                         ipstat.ips_noroute++;
364                         error = EHOSTUNREACH;
365                         goto bad;
366                 }
367                 ia = ifatoia(ro->ro_rt->rt_ifa);
368                 ifp = ro->ro_rt->rt_ifp;
369                 ro->ro_rt->rt_use++;
370                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
371                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
372                 if (ro->ro_rt->rt_flags & RTF_HOST)
373                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
374                 else
375                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
376         }
377         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
378                 struct in_multi *inm;
379
380                 m->m_flags |= M_MCAST;
381                 /*
382                  * IP destination address is multicast.  Make sure "dst"
383                  * still points to the address in "ro".  (It may have been
384                  * changed to point to a gateway address, above.)
385                  */
386                 dst = (struct sockaddr_in *)&ro->ro_dst;
387                 /*
388                  * See if the caller provided any multicast options
389                  */
390                 if (imo != NULL) {
391                         ip->ip_ttl = imo->imo_multicast_ttl;
392                         if (imo->imo_multicast_vif != -1) {
393                                 ip->ip_src.s_addr =
394                                     ip_mcast_src ?
395                                     ip_mcast_src(imo->imo_multicast_vif) :
396                                     INADDR_ANY;
397                         }
398                 } else {
399                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
400                 }
401                 /*
402                  * Confirm that the outgoing interface supports multicast.
403                  */
404                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
405                         if (!(ifp->if_flags & IFF_MULTICAST)) {
406                                 ipstat.ips_noroute++;
407                                 error = ENETUNREACH;
408                                 goto bad;
409                         }
410                 }
411                 /*
412                  * If source address not specified yet, use address
413                  * of outgoing interface.
414                  */
415                 if (ip->ip_src.s_addr == INADDR_ANY) {
416                         /* Interface may have no addresses. */
417                         if (ia != NULL)
418                                 ip->ip_src = IA_SIN(ia)->sin_addr;
419                 }
420
421                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
422                 if (inm != NULL &&
423                     (imo == NULL || imo->imo_multicast_loop)) {
424                         /*
425                          * If we belong to the destination multicast group
426                          * on the outgoing interface, and the caller did not
427                          * forbid loopback, loop back a copy.
428                          */
429                         ip_mloopback(ifp, m, dst, hlen);
430                 } else {
431                         /*
432                          * If we are acting as a multicast router, perform
433                          * multicast forwarding as if the packet had just
434                          * arrived on the interface to which we are about
435                          * to send.  The multicast forwarding function
436                          * recursively calls this function, using the
437                          * IP_FORWARDING flag to prevent infinite recursion.
438                          *
439                          * Multicasts that are looped back by ip_mloopback(),
440                          * above, will be forwarded by the ip_input() routine,
441                          * if necessary.
442                          */
443                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
444                                 /*
445                                  * If rsvp daemon is not running, do not
446                                  * set ip_moptions. This ensures that the packet
447                                  * is multicast and not just sent down one link
448                                  * as prescribed by rsvpd.
449                                  */
450                                 if (!rsvp_on)
451                                         imo = NULL;
452                                 if (ip_mforward) {
453                                         get_mplock();
454                                         if (ip_mforward(ip, ifp, m, imo) != 0) {
455                                                 m_freem(m);
456                                                 rel_mplock();
457                                                 goto done;
458                                         }
459                                         rel_mplock();
460                                 }
461                         }
462                 }
463
464                 /*
465                  * Multicasts with a time-to-live of zero may be looped-
466                  * back, above, but must not be transmitted on a network.
467                  * Also, multicasts addressed to the loopback interface
468                  * are not sent -- the above call to ip_mloopback() will
469                  * loop back a copy if this host actually belongs to the
470                  * destination group on the loopback interface.
471                  */
472                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
473                         m_freem(m);
474                         goto done;
475                 }
476
477                 goto sendit;
478         } else {
479                 m->m_flags &= ~M_MCAST;
480         }
481
482         /*
483          * If the source address is not specified yet, use the address
484          * of the outoing interface. In case, keep note we did that, so
485          * if the the firewall changes the next-hop causing the output
486          * interface to change, we can fix that.
487          */
488         if (ip->ip_src.s_addr == INADDR_ANY || src_was_INADDR_ANY) {
489                 /* Interface may have no addresses. */
490                 if (ia != NULL) {
491                         ip->ip_src = IA_SIN(ia)->sin_addr;
492                         src_was_INADDR_ANY = 1;
493                 }
494         }
495
496 #ifdef ALTQ
497         /*
498          * Disable packet drop hack.
499          * Packetdrop should be done by queueing.
500          */
501 #else /* !ALTQ */
502         /*
503          * Verify that we have any chance at all of being able to queue
504          *      the packet or packet fragments
505          */
506         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
507             ifp->if_snd.ifq_maxlen) {
508                 error = ENOBUFS;
509                 ipstat.ips_odropped++;
510                 goto bad;
511         }
512 #endif /* !ALTQ */
513
514         /*
515          * Look for broadcast address and
516          * verify user is allowed to send
517          * such a packet.
518          */
519         if (isbroadcast) {
520                 if (!(ifp->if_flags & IFF_BROADCAST)) {
521                         error = EADDRNOTAVAIL;
522                         goto bad;
523                 }
524                 if (!(flags & IP_ALLOWBROADCAST)) {
525                         error = EACCES;
526                         goto bad;
527                 }
528                 /* don't allow broadcast messages to be fragmented */
529                 if (ip->ip_len > ifp->if_mtu) {
530                         error = EMSGSIZE;
531                         goto bad;
532                 }
533                 m->m_flags |= M_BCAST;
534         } else {
535                 m->m_flags &= ~M_BCAST;
536         }
537
538 sendit:
539 #ifdef IPSEC
540         /* get SP for this packet */
541         if (so == NULL)
542                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
543         else
544                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
545
546         if (sp == NULL) {
547                 ipsecstat.out_inval++;
548                 goto bad;
549         }
550
551         error = 0;
552
553         /* check policy */
554         switch (sp->policy) {
555         case IPSEC_POLICY_DISCARD:
556                 /*
557                  * This packet is just discarded.
558                  */
559                 ipsecstat.out_polvio++;
560                 goto bad;
561
562         case IPSEC_POLICY_BYPASS:
563         case IPSEC_POLICY_NONE:
564         case IPSEC_POLICY_TCP:
565                 /* no need to do IPsec. */
566                 goto skip_ipsec;
567
568         case IPSEC_POLICY_IPSEC:
569                 if (sp->req == NULL) {
570                         /* acquire a policy */
571                         error = key_spdacquire(sp);
572                         goto bad;
573                 }
574                 break;
575
576         case IPSEC_POLICY_ENTRUST:
577         default:
578                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
579         }
580     {
581         struct ipsec_output_state state;
582         bzero(&state, sizeof state);
583         state.m = m;
584         if (flags & IP_ROUTETOIF) {
585                 state.ro = &iproute;
586                 bzero(&iproute, sizeof iproute);
587         } else
588                 state.ro = ro;
589         state.dst = (struct sockaddr *)dst;
590
591         ip->ip_sum = 0;
592
593         /*
594          * XXX
595          * delayed checksums are not currently compatible with IPsec
596          */
597         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
598                 in_delayed_cksum(m);
599                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
600         }
601
602         ip->ip_len = htons(ip->ip_len);
603         ip->ip_off = htons(ip->ip_off);
604
605         error = ipsec4_output(&state, sp, flags);
606
607         m = state.m;
608         if (flags & IP_ROUTETOIF) {
609                 /*
610                  * if we have tunnel mode SA, we may need to ignore
611                  * IP_ROUTETOIF.
612                  */
613                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
614                         flags &= ~IP_ROUTETOIF;
615                         ro = state.ro;
616                 }
617         } else
618                 ro = state.ro;
619         dst = (struct sockaddr_in *)state.dst;
620         if (error) {
621                 /* mbuf is already reclaimed in ipsec4_output. */
622                 m0 = NULL;
623                 switch (error) {
624                 case EHOSTUNREACH:
625                 case ENETUNREACH:
626                 case EMSGSIZE:
627                 case ENOBUFS:
628                 case ENOMEM:
629                         break;
630                 default:
631                         kprintf("ip4_output (ipsec): error code %d\n", error);
632                         /*fall through*/
633                 case ENOENT:
634                         /* don't show these error codes to the user */
635                         error = 0;
636                         break;
637                 }
638                 goto bad;
639         }
640     }
641
642         /* be sure to update variables that are affected by ipsec4_output() */
643         ip = mtod(m, struct ip *);
644 #ifdef _IP_VHL
645         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
646 #else
647         hlen = ip->ip_hl << 2;
648 #endif
649         if (ro->ro_rt == NULL) {
650                 if (!(flags & IP_ROUTETOIF)) {
651                         kprintf("ip_output: "
652                                 "can't update route after IPsec processing\n");
653                         error = EHOSTUNREACH;   /*XXX*/
654                         goto bad;
655                 }
656         } else {
657                 ia = ifatoia(ro->ro_rt->rt_ifa);
658                 ifp = ro->ro_rt->rt_ifp;
659         }
660
661         /* make it flipped, again. */
662         ip->ip_len = ntohs(ip->ip_len);
663         ip->ip_off = ntohs(ip->ip_off);
664 skip_ipsec:
665 #endif /*IPSEC*/
666 #ifdef FAST_IPSEC
667         /*
668          * Check the security policy (SP) for the packet and, if
669          * required, do IPsec-related processing.  There are two
670          * cases here; the first time a packet is sent through
671          * it will be untagged and handled by ipsec4_checkpolicy.
672          * If the packet is resubmitted to ip_output (e.g. after
673          * AH, ESP, etc. processing), there will be a tag to bypass
674          * the lookup and related policy checking.
675          */
676         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
677         crit_enter();
678         if (mtag != NULL) {
679                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
680                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
681                 if (sp == NULL)
682                         error = -EINVAL;        /* force silent drop */
683                 m_tag_delete(m, mtag);
684         } else {
685                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
686                                         &error, inp);
687         }
688         /*
689          * There are four return cases:
690          *    sp != NULL                    apply IPsec policy
691          *    sp == NULL, error == 0        no IPsec handling needed
692          *    sp == NULL, error == -EINVAL  discard packet w/o error
693          *    sp == NULL, error != 0        discard packet, report error
694          */
695         if (sp != NULL) {
696                 /* Loop detection, check if ipsec processing already done */
697                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
698                 for (mtag = m_tag_first(m); mtag != NULL;
699                      mtag = m_tag_next(m, mtag)) {
700                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
701                                 continue;
702                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
703                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
704                                 continue;
705                         /*
706                          * Check if policy has an SA associated with it.
707                          * This can happen when an SP has yet to acquire
708                          * an SA; e.g. on first reference.  If it occurs,
709                          * then we let ipsec4_process_packet do its thing.
710                          */
711                         if (sp->req->sav == NULL)
712                                 break;
713                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
714                         if (tdbi->spi == sp->req->sav->spi &&
715                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
716                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
717                                  sizeof(union sockaddr_union)) == 0) {
718                                 /*
719                                  * No IPsec processing is needed, free
720                                  * reference to SP.
721                                  *
722                                  * NB: null pointer to avoid free at
723                                  *     done: below.
724                                  */
725                                 KEY_FREESP(&sp), sp = NULL;
726                                 crit_exit();
727                                 goto spd_done;
728                         }
729                 }
730
731                 /*
732                  * Do delayed checksums now because we send before
733                  * this is done in the normal processing path.
734                  */
735                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
736                         in_delayed_cksum(m);
737                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
738                 }
739
740                 ip->ip_len = htons(ip->ip_len);
741                 ip->ip_off = htons(ip->ip_off);
742
743                 /* NB: callee frees mbuf */
744                 error = ipsec4_process_packet(m, sp->req, flags, 0);
745                 /*
746                  * Preserve KAME behaviour: ENOENT can be returned
747                  * when an SA acquire is in progress.  Don't propagate
748                  * this to user-level; it confuses applications.
749                  *
750                  * XXX this will go away when the SADB is redone.
751                  */
752                 if (error == ENOENT)
753                         error = 0;
754                 crit_exit();
755                 goto done;
756         } else {
757                 crit_exit();
758
759                 if (error != 0) {
760                         /*
761                          * Hack: -EINVAL is used to signal that a packet
762                          * should be silently discarded.  This is typically
763                          * because we asked key management for an SA and
764                          * it was delayed (e.g. kicked up to IKE).
765                          */
766                         if (error == -EINVAL)
767                                 error = 0;
768                         goto bad;
769                 } else {
770                         /* No IPsec processing for this packet. */
771                 }
772 #ifdef notyet
773                 /*
774                  * If deferred crypto processing is needed, check that
775                  * the interface supports it.
776                  */
777                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
778                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
779                         /* notify IPsec to do its own crypto */
780                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
781                         error = EHOSTUNREACH;
782                         goto bad;
783                 }
784 #endif
785         }
786 spd_done:
787 #endif /* FAST_IPSEC */
788
789         /* We are already being fwd'd from a firewall. */
790         if (next_hop != NULL)
791                 goto pass;
792
793         /* No pfil hooks */
794         if (!pfil_has_hooks(&inet_pfil_hook)) {
795                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
796                         /*
797                          * Strip dummynet tags from stranded packets
798                          */
799                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
800                         KKASSERT(mtag != NULL);
801                         m_tag_delete(m, mtag);
802                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
803                 }
804                 goto pass;
805         }
806
807         /*
808          * IpHack's section.
809          * - Xlate: translate packet's addr/port (NAT).
810          * - Firewall: deny/allow/etc.
811          * - Wrap: fake packet's addr/port <unimpl.>
812          * - Encapsulate: put it in another IP and send out. <unimp.>
813          */
814
815         /*
816          * Run through list of hooks for output packets.
817          */
818         error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
819         if (error != 0 || m == NULL)
820                 goto done;
821         ip = mtod(m, struct ip *);
822
823         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
824                 /*
825                  * Check dst to make sure it is directly reachable on the
826                  * interface we previously thought it was.
827                  * If it isn't (which may be likely in some situations) we have
828                  * to re-route it (ie, find a route for the next-hop and the
829                  * associated interface) and set them here. This is nested
830                  * forwarding which in most cases is undesirable, except where
831                  * such control is nigh impossible. So we do it here.
832                  * And I'm babbling.
833                  */
834                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
835                 KKASSERT(mtag != NULL);
836                 next_hop = m_tag_data(mtag);
837
838                 /*
839                  * Try local forwarding first
840                  */
841                 if (ip_localforward(m, next_hop, hlen))
842                         goto done;
843
844                 /*
845                  * Relocate the route based on next_hop.
846                  * If the current route is inp's cache, keep it untouched.
847                  */
848                 if (ro == &iproute && ro->ro_rt != NULL) {
849                         RTFREE(ro->ro_rt);
850                         ro->ro_rt = NULL;
851                 }
852                 ro = &iproute;
853                 bzero(ro, sizeof *ro);
854
855                 /*
856                  * Forwarding to broadcast address is not allowed.
857                  * XXX Should we follow IP_ROUTETOIF?
858                  */
859                 flags &= ~(IP_ALLOWBROADCAST | IP_ROUTETOIF);
860
861                 /* We are doing forwarding now */
862                 flags |= IP_FORWARDING;
863
864                 goto reroute;
865         }
866
867         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
868                 struct dn_pkt *dn_pkt;
869
870                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
871                 KKASSERT(mtag != NULL);
872                 dn_pkt = m_tag_data(mtag);
873
874                 /*
875                  * Under certain cases it is not possible to recalculate
876                  * 'ro' and 'dst', let alone 'flags', so just save them in
877                  * dummynet tag and avoid the possible wrong reculcalation
878                  * when we come back to ip_output() again.
879                  *
880                  * All other parameters have been already used and so they
881                  * are not needed anymore.
882                  * XXX if the ifp is deleted while a pkt is in dummynet,
883                  * we are in trouble! (TODO use ifnet_detach_event)
884                  *
885                  * We need to copy *ro because for ICMP pkts (and maybe
886                  * others) the caller passed a pointer into the stack;
887                  * dst might also be a pointer into *ro so it needs to
888                  * be updated.
889                  */
890                 dn_pkt->ro = *ro;
891                 if (ro->ro_rt)
892                         ro->ro_rt->rt_refcnt++;
893                 if (dst == (struct sockaddr_in *)&ro->ro_dst) {
894                         /* 'dst' points into 'ro' */
895                         dst = (struct sockaddr_in *)&(dn_pkt->ro.ro_dst);
896                 }
897                 dn_pkt->dn_dst = dst;
898                 dn_pkt->flags = flags;
899
900                 ip_dn_queue(m);
901                 goto done;
902         }
903 pass:
904         /* 127/8 must not appear on wire - RFC1122. */
905         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
906             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
907                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
908                         ipstat.ips_badaddr++;
909                         error = EADDRNOTAVAIL;
910                         goto bad;
911                 }
912         }
913
914         if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
915                 m->m_pkthdr.csum_flags |= CSUM_IP;
916                 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
917                 if (sw_csum & CSUM_DELAY_DATA) {
918                         in_delayed_cksum(m);
919                         sw_csum &= ~CSUM_DELAY_DATA;
920                 }
921                 m->m_pkthdr.csum_flags &= ifp->if_hwassist;
922         } else {
923                 sw_csum = 0;
924         }
925
926         /*
927          * If small enough for interface, or the interface will take
928          * care of the fragmentation or segmentation for us, can just
929          * send directly.
930          */
931         if (ip->ip_len <= ifp->if_mtu ||
932             ((ifp->if_hwassist & CSUM_FRAGMENT) && !(ip->ip_off & IP_DF)) ||
933             (m->m_pkthdr.csum_flags & CSUM_TSO)) {
934                 ip->ip_len = htons(ip->ip_len);
935                 ip->ip_off = htons(ip->ip_off);
936                 ip->ip_sum = 0;
937                 if (sw_csum & CSUM_DELAY_IP) {
938                         if (ip->ip_vhl == IP_VHL_BORING)
939                                 ip->ip_sum = in_cksum_hdr(ip);
940                         else
941                                 ip->ip_sum = in_cksum(m, hlen);
942                 }
943
944                 /* Record statistics for this interface address. */
945                 if (!(flags & IP_FORWARDING) && ia) {
946                         ia->ia_ifa.if_opackets++;
947                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
948                 }
949
950 #ifdef IPSEC
951                 /* clean ipsec history once it goes out of the node */
952                 ipsec_delaux(m);
953 #endif
954
955 #ifdef MBUF_STRESS_TEST
956                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
957                         struct mbuf *m1, *m2;
958                         int length, tmp;
959
960                         tmp = length = m->m_pkthdr.len;
961
962                         while ((length -= mbuf_frag_size) >= 1) {
963                                 m1 = m_split(m, length, MB_DONTWAIT);
964                                 if (m1 == NULL)
965                                         break;
966                                 m2 = m;
967                                 while (m2->m_next != NULL)
968                                         m2 = m2->m_next;
969                                 m2->m_next = m1;
970                         }
971                         m->m_pkthdr.len = tmp;
972                 }
973 #endif
974
975 #ifdef MPLS
976                 if (!mpls_output_process(m, ro->ro_rt))
977                         goto done;
978 #endif
979                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
980                                        ro->ro_rt);
981                 goto done;
982         }
983
984         if (ip->ip_off & IP_DF) {
985                 error = EMSGSIZE;
986                 /*
987                  * This case can happen if the user changed the MTU
988                  * of an interface after enabling IP on it.  Because
989                  * most netifs don't keep track of routes pointing to
990                  * them, there is no way for one to update all its
991                  * routes when the MTU is changed.
992                  */
993                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
994                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
995                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
996                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
997                 }
998                 ipstat.ips_cantfrag++;
999                 goto bad;
1000         }
1001
1002         /*
1003          * Too large for interface; fragment if possible. If successful,
1004          * on return, m will point to a list of packets to be sent.
1005          */
1006         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1007         if (error)
1008                 goto bad;
1009         for (; m; m = m0) {
1010                 m0 = m->m_nextpkt;
1011                 m->m_nextpkt = NULL;
1012 #ifdef IPSEC
1013                 /* clean ipsec history once it goes out of the node */
1014                 ipsec_delaux(m);
1015 #endif
1016                 if (error == 0) {
1017                         /* Record statistics for this interface address. */
1018                         if (ia != NULL) {
1019                                 ia->ia_ifa.if_opackets++;
1020                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1021                         }
1022 #ifdef MPLS
1023                         if (!mpls_output_process(m, ro->ro_rt))
1024                                 continue;
1025 #endif
1026                         error = ifp->if_output(ifp, m, (struct sockaddr *)dst,
1027                                                ro->ro_rt);
1028                 } else {
1029                         m_freem(m);
1030                 }
1031         }
1032
1033         if (error == 0)
1034                 ipstat.ips_fragmented++;
1035
1036 done:
1037         if (ro == &iproute && ro->ro_rt != NULL) {
1038                 RTFREE(ro->ro_rt);
1039                 ro->ro_rt = NULL;
1040         }
1041 #ifdef IPSEC
1042         if (sp != NULL) {
1043                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1044                         kprintf("DP ip_output call free SP:%p\n", sp));
1045                 key_freesp(sp);
1046         }
1047 #endif
1048 #ifdef FAST_IPSEC
1049         if (sp != NULL)
1050                 KEY_FREESP(&sp);
1051 #endif
1052         return (error);
1053 bad:
1054         m_freem(m);
1055         goto done;
1056 }
1057
1058 /*
1059  * Create a chain of fragments which fit the given mtu. m_frag points to the
1060  * mbuf to be fragmented; on return it points to the chain with the fragments.
1061  * Return 0 if no error. If error, m_frag may contain a partially built
1062  * chain of fragments that should be freed by the caller.
1063  *
1064  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1065  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1066  */
1067 int
1068 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1069             u_long if_hwassist_flags, int sw_csum)
1070 {
1071         int error = 0;
1072         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1073         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1074         int off;
1075         struct mbuf *m0 = *m_frag;      /* the original packet          */
1076         int firstlen;
1077         struct mbuf **mnext;
1078         int nfrags;
1079
1080         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1081                 ipstat.ips_cantfrag++;
1082                 return EMSGSIZE;
1083         }
1084
1085         /*
1086          * Must be able to put at least 8 bytes per fragment.
1087          */
1088         if (len < 8)
1089                 return EMSGSIZE;
1090
1091         /*
1092          * If the interface will not calculate checksums on
1093          * fragmented packets, then do it here.
1094          */
1095         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1096             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1097                 in_delayed_cksum(m0);
1098                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1099         }
1100
1101         if (len > PAGE_SIZE) {
1102                 /*
1103                  * Fragment large datagrams such that each segment
1104                  * contains a multiple of PAGE_SIZE amount of data,
1105                  * plus headers. This enables a receiver to perform
1106                  * page-flipping zero-copy optimizations.
1107                  *
1108                  * XXX When does this help given that sender and receiver
1109                  * could have different page sizes, and also mtu could
1110                  * be less than the receiver's page size ?
1111                  */
1112                 int newlen;
1113                 struct mbuf *m;
1114
1115                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1116                         off += m->m_len;
1117
1118                 /*
1119                  * firstlen (off - hlen) must be aligned on an
1120                  * 8-byte boundary
1121                  */
1122                 if (off < hlen)
1123                         goto smart_frag_failure;
1124                 off = ((off - hlen) & ~7) + hlen;
1125                 newlen = (~PAGE_MASK) & mtu;
1126                 if ((newlen + sizeof(struct ip)) > mtu) {
1127                         /* we failed, go back the default */
1128 smart_frag_failure:
1129                         newlen = len;
1130                         off = hlen + len;
1131                 }
1132                 len = newlen;
1133
1134         } else {
1135                 off = hlen + len;
1136         }
1137
1138         firstlen = off - hlen;
1139         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1140
1141         /*
1142          * Loop through length of segment after first fragment,
1143          * make new header and copy data of each part and link onto chain.
1144          * Here, m0 is the original packet, m is the fragment being created.
1145          * The fragments are linked off the m_nextpkt of the original
1146          * packet, which after processing serves as the first fragment.
1147          */
1148         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1149                 struct ip *mhip;        /* ip header on the fragment */
1150                 struct mbuf *m;
1151                 int mhlen = sizeof(struct ip);
1152
1153                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1154                 if (m == NULL) {
1155                         error = ENOBUFS;
1156                         ipstat.ips_odropped++;
1157                         goto done;
1158                 }
1159                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1160                 /*
1161                  * In the first mbuf, leave room for the link header, then
1162                  * copy the original IP header including options. The payload
1163                  * goes into an additional mbuf chain returned by m_copy().
1164                  */
1165                 m->m_data += max_linkhdr;
1166                 mhip = mtod(m, struct ip *);
1167                 *mhip = *ip;
1168                 if (hlen > sizeof(struct ip)) {
1169                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1170                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1171                 }
1172                 m->m_len = mhlen;
1173                 /* XXX do we need to add ip->ip_off below ? */
1174                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1175                 if (off + len >= ip->ip_len) {  /* last fragment */
1176                         len = ip->ip_len - off;
1177                         m->m_flags |= M_LASTFRAG;
1178                 } else
1179                         mhip->ip_off |= IP_MF;
1180                 mhip->ip_len = htons((u_short)(len + mhlen));
1181                 m->m_next = m_copy(m0, off, len);
1182                 if (m->m_next == NULL) {                /* copy failed */
1183                         m_free(m);
1184                         error = ENOBUFS;        /* ??? */
1185                         ipstat.ips_odropped++;
1186                         goto done;
1187                 }
1188                 m->m_pkthdr.len = mhlen + len;
1189                 m->m_pkthdr.rcvif = NULL;
1190                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1191                 mhip->ip_off = htons(mhip->ip_off);
1192                 mhip->ip_sum = 0;
1193                 if (sw_csum & CSUM_DELAY_IP)
1194                         mhip->ip_sum = in_cksum(m, mhlen);
1195                 *mnext = m;
1196                 mnext = &m->m_nextpkt;
1197         }
1198         ipstat.ips_ofragments += nfrags;
1199
1200         /* set first marker for fragment chain */
1201         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1202         m0->m_pkthdr.csum_data = nfrags;
1203
1204         /*
1205          * Update first fragment by trimming what's been copied out
1206          * and updating header.
1207          */
1208         m_adj(m0, hlen + firstlen - ip->ip_len);
1209         m0->m_pkthdr.len = hlen + firstlen;
1210         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1211         ip->ip_off |= IP_MF;
1212         ip->ip_off = htons(ip->ip_off);
1213         ip->ip_sum = 0;
1214         if (sw_csum & CSUM_DELAY_IP)
1215                 ip->ip_sum = in_cksum(m0, hlen);
1216
1217 done:
1218         *m_frag = m0;
1219         return error;
1220 }
1221
1222 void
1223 in_delayed_cksum(struct mbuf *m)
1224 {
1225         struct ip *ip;
1226         u_short csum, offset;
1227
1228         ip = mtod(m, struct ip *);
1229         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1230         csum = in_cksum_skip(m, ip->ip_len, offset);
1231         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1232                 csum = 0xffff;
1233         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1234
1235         if (offset + sizeof(u_short) > m->m_len) {
1236                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1237                     m->m_len, offset, ip->ip_p);
1238                 /*
1239                  * XXX
1240                  * this shouldn't happen, but if it does, the
1241                  * correct behavior may be to insert the checksum
1242                  * in the existing chain instead of rearranging it.
1243                  */
1244                 m = m_pullup(m, offset + sizeof(u_short));
1245         }
1246         *(u_short *)(m->m_data + offset) = csum;
1247 }
1248
1249 /*
1250  * Insert IP options into preformed packet.
1251  * Adjust IP destination as required for IP source routing,
1252  * as indicated by a non-zero in_addr at the start of the options.
1253  *
1254  * XXX This routine assumes that the packet has no options in place.
1255  */
1256 static struct mbuf *
1257 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1258 {
1259         struct ipoption *p = mtod(opt, struct ipoption *);
1260         struct mbuf *n;
1261         struct ip *ip = mtod(m, struct ip *);
1262         unsigned optlen;
1263
1264         optlen = opt->m_len - sizeof p->ipopt_dst;
1265         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1266                 *phlen = 0;
1267                 return (m);             /* XXX should fail */
1268         }
1269         if (p->ipopt_dst.s_addr)
1270                 ip->ip_dst = p->ipopt_dst;
1271         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1272                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1273                 if (n == NULL) {
1274                         *phlen = 0;
1275                         return (m);
1276                 }
1277                 n->m_pkthdr.rcvif = NULL;
1278                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1279                 m->m_len -= sizeof(struct ip);
1280                 m->m_data += sizeof(struct ip);
1281                 n->m_next = m;
1282                 m = n;
1283                 m->m_len = optlen + sizeof(struct ip);
1284                 m->m_data += max_linkhdr;
1285                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1286         } else {
1287                 m->m_data -= optlen;
1288                 m->m_len += optlen;
1289                 m->m_pkthdr.len += optlen;
1290                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1291         }
1292         ip = mtod(m, struct ip *);
1293         bcopy(p->ipopt_list, ip + 1, optlen);
1294         *phlen = sizeof(struct ip) + optlen;
1295         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1296         ip->ip_len += optlen;
1297         return (m);
1298 }
1299
1300 /*
1301  * Copy options from ip to jp,
1302  * omitting those not copied during fragmentation.
1303  */
1304 int
1305 ip_optcopy(struct ip *ip, struct ip *jp)
1306 {
1307         u_char *cp, *dp;
1308         int opt, optlen, cnt;
1309
1310         cp = (u_char *)(ip + 1);
1311         dp = (u_char *)(jp + 1);
1312         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1313         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1314                 opt = cp[0];
1315                 if (opt == IPOPT_EOL)
1316                         break;
1317                 if (opt == IPOPT_NOP) {
1318                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1319                         *dp++ = IPOPT_NOP;
1320                         optlen = 1;
1321                         continue;
1322                 }
1323
1324                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1325                     ("ip_optcopy: malformed ipv4 option"));
1326                 optlen = cp[IPOPT_OLEN];
1327                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1328                     ("ip_optcopy: malformed ipv4 option"));
1329
1330                 /* bogus lengths should have been caught by ip_dooptions */
1331                 if (optlen > cnt)
1332                         optlen = cnt;
1333                 if (IPOPT_COPIED(opt)) {
1334                         bcopy(cp, dp, optlen);
1335                         dp += optlen;
1336                 }
1337         }
1338         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1339                 *dp++ = IPOPT_EOL;
1340         return (optlen);
1341 }
1342
1343 /*
1344  * IP socket option processing.
1345  */
1346 void
1347 ip_ctloutput(netmsg_t msg)
1348 {
1349         struct socket *so = msg->base.nm_so;
1350         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1351         struct  inpcb *inp = so->so_pcb;
1352         int     error, optval;
1353
1354         error = optval = 0;
1355         if (sopt->sopt_level != IPPROTO_IP) {
1356                 error = EINVAL;
1357                 goto done;
1358         }
1359
1360         switch (sopt->sopt_dir) {
1361         case SOPT_SET:
1362                 switch (sopt->sopt_name) {
1363                 case IP_OPTIONS:
1364 #ifdef notyet
1365                 case IP_RETOPTS:
1366 #endif
1367                 {
1368                         struct mbuf *m;
1369                         if (sopt->sopt_valsize > MLEN) {
1370                                 error = EMSGSIZE;
1371                                 break;
1372                         }
1373                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1374                         if (m == NULL) {
1375                                 error = ENOBUFS;
1376                                 break;
1377                         }
1378                         m->m_len = sopt->sopt_valsize;
1379                         error = soopt_to_kbuf(sopt, mtod(m, void *), m->m_len,
1380                                               m->m_len);
1381                         error = ip_pcbopts(sopt->sopt_name,
1382                                            &inp->inp_options, m);
1383                         goto done;
1384                 }
1385
1386                 case IP_TOS:
1387                 case IP_TTL:
1388                 case IP_MINTTL:
1389                 case IP_RECVOPTS:
1390                 case IP_RECVRETOPTS:
1391                 case IP_RECVDSTADDR:
1392                 case IP_RECVIF:
1393                 case IP_RECVTTL:
1394                 case IP_FAITH:
1395                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1396                                              sizeof optval);
1397                         if (error)
1398                                 break;
1399                         switch (sopt->sopt_name) {
1400                         case IP_TOS:
1401                                 inp->inp_ip_tos = optval;
1402                                 break;
1403
1404                         case IP_TTL:
1405                                 inp->inp_ip_ttl = optval;
1406                                 break;
1407                         case IP_MINTTL:
1408                                 if (optval >= 0 && optval <= MAXTTL)
1409                                         inp->inp_ip_minttl = optval;
1410                                 else
1411                                         error = EINVAL;
1412                                 break;
1413 #define OPTSET(bit) \
1414         if (optval) \
1415                 inp->inp_flags |= bit; \
1416         else \
1417                 inp->inp_flags &= ~bit;
1418
1419                         case IP_RECVOPTS:
1420                                 OPTSET(INP_RECVOPTS);
1421                                 break;
1422
1423                         case IP_RECVRETOPTS:
1424                                 OPTSET(INP_RECVRETOPTS);
1425                                 break;
1426
1427                         case IP_RECVDSTADDR:
1428                                 OPTSET(INP_RECVDSTADDR);
1429                                 break;
1430
1431                         case IP_RECVIF:
1432                                 OPTSET(INP_RECVIF);
1433                                 break;
1434
1435                         case IP_RECVTTL:
1436                                 OPTSET(INP_RECVTTL);
1437                                 break;
1438
1439                         case IP_FAITH:
1440                                 OPTSET(INP_FAITH);
1441                                 break;
1442                         }
1443                         break;
1444 #undef OPTSET
1445
1446                 case IP_MULTICAST_IF:
1447                 case IP_MULTICAST_VIF:
1448                 case IP_MULTICAST_TTL:
1449                 case IP_MULTICAST_LOOP:
1450                 case IP_ADD_MEMBERSHIP:
1451                 case IP_DROP_MEMBERSHIP:
1452                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1453                         break;
1454
1455                 case IP_PORTRANGE:
1456                         error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1457                                             sizeof optval);
1458                         if (error)
1459                                 break;
1460
1461                         switch (optval) {
1462                         case IP_PORTRANGE_DEFAULT:
1463                                 inp->inp_flags &= ~(INP_LOWPORT);
1464                                 inp->inp_flags &= ~(INP_HIGHPORT);
1465                                 break;
1466
1467                         case IP_PORTRANGE_HIGH:
1468                                 inp->inp_flags &= ~(INP_LOWPORT);
1469                                 inp->inp_flags |= INP_HIGHPORT;
1470                                 break;
1471
1472                         case IP_PORTRANGE_LOW:
1473                                 inp->inp_flags &= ~(INP_HIGHPORT);
1474                                 inp->inp_flags |= INP_LOWPORT;
1475                                 break;
1476
1477                         default:
1478                                 error = EINVAL;
1479                                 break;
1480                         }
1481                         break;
1482
1483 #if defined(IPSEC) || defined(FAST_IPSEC)
1484                 case IP_IPSEC_POLICY:
1485                 {
1486                         caddr_t req;
1487                         size_t len = 0;
1488                         int priv;
1489                         struct mbuf *m;
1490                         int optname;
1491
1492                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1493                                 break;
1494                         soopt_to_mbuf(sopt, m);
1495                         priv = (sopt->sopt_td != NULL &&
1496                                 priv_check(sopt->sopt_td, PRIV_ROOT) != 0) ? 0 : 1;
1497                         req = mtod(m, caddr_t);
1498                         len = m->m_len;
1499                         optname = sopt->sopt_name;
1500                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1501                         m_freem(m);
1502                         break;
1503                 }
1504 #endif /*IPSEC*/
1505
1506                 default:
1507                         error = ENOPROTOOPT;
1508                         break;
1509                 }
1510                 break;
1511
1512         case SOPT_GET:
1513                 switch (sopt->sopt_name) {
1514                 case IP_OPTIONS:
1515                 case IP_RETOPTS:
1516                         if (inp->inp_options)
1517                                 soopt_from_kbuf(sopt, mtod(inp->inp_options,
1518                                                            char *),
1519                                                 inp->inp_options->m_len);
1520                         else
1521                                 sopt->sopt_valsize = 0;
1522                         break;
1523
1524                 case IP_TOS:
1525                 case IP_TTL:
1526                 case IP_MINTTL:
1527                 case IP_RECVOPTS:
1528                 case IP_RECVRETOPTS:
1529                 case IP_RECVDSTADDR:
1530                 case IP_RECVTTL:
1531                 case IP_RECVIF:
1532                 case IP_PORTRANGE:
1533                 case IP_FAITH:
1534                         switch (sopt->sopt_name) {
1535
1536                         case IP_TOS:
1537                                 optval = inp->inp_ip_tos;
1538                                 break;
1539
1540                         case IP_TTL:
1541                                 optval = inp->inp_ip_ttl;
1542                                 break;
1543                         case IP_MINTTL:
1544                                 optval = inp->inp_ip_minttl;
1545                                 break;
1546
1547 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1548
1549                         case IP_RECVOPTS:
1550                                 optval = OPTBIT(INP_RECVOPTS);
1551                                 break;
1552
1553                         case IP_RECVRETOPTS:
1554                                 optval = OPTBIT(INP_RECVRETOPTS);
1555                                 break;
1556
1557                         case IP_RECVDSTADDR:
1558                                 optval = OPTBIT(INP_RECVDSTADDR);
1559                                 break;
1560
1561                         case IP_RECVTTL:
1562                                 optval = OPTBIT(INP_RECVTTL);
1563                                 break;
1564
1565                         case IP_RECVIF:
1566                                 optval = OPTBIT(INP_RECVIF);
1567                                 break;
1568
1569                         case IP_PORTRANGE:
1570                                 if (inp->inp_flags & INP_HIGHPORT)
1571                                         optval = IP_PORTRANGE_HIGH;
1572                                 else if (inp->inp_flags & INP_LOWPORT)
1573                                         optval = IP_PORTRANGE_LOW;
1574                                 else
1575                                         optval = 0;
1576                                 break;
1577
1578                         case IP_FAITH:
1579                                 optval = OPTBIT(INP_FAITH);
1580                                 break;
1581                         }
1582                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1583                         break;
1584
1585                 case IP_MULTICAST_IF:
1586                 case IP_MULTICAST_VIF:
1587                 case IP_MULTICAST_TTL:
1588                 case IP_MULTICAST_LOOP:
1589                 case IP_ADD_MEMBERSHIP:
1590                 case IP_DROP_MEMBERSHIP:
1591                         error = ip_getmoptions(sopt, inp->inp_moptions);
1592                         break;
1593
1594 #if defined(IPSEC) || defined(FAST_IPSEC)
1595                 case IP_IPSEC_POLICY:
1596                 {
1597                         struct mbuf *m = NULL;
1598                         caddr_t req = NULL;
1599                         size_t len = 0;
1600
1601                         if (m != NULL) {
1602                                 req = mtod(m, caddr_t);
1603                                 len = m->m_len;
1604                         }
1605                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1606                         if (error == 0)
1607                                 error = soopt_from_mbuf(sopt, m); /* XXX */
1608                         if (error == 0)
1609                                 m_freem(m);
1610                         break;
1611                 }
1612 #endif /*IPSEC*/
1613
1614                 default:
1615                         error = ENOPROTOOPT;
1616                         break;
1617                 }
1618                 break;
1619         }
1620 done:
1621         lwkt_replymsg(&msg->lmsg, error);
1622 }
1623
1624 /*
1625  * Set up IP options in pcb for insertion in output packets.
1626  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1627  * with destination address if source routed.
1628  */
1629 static int
1630 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1631 {
1632         int cnt, optlen;
1633         u_char *cp;
1634         u_char opt;
1635
1636         /* turn off any old options */
1637         if (*pcbopt)
1638                 m_free(*pcbopt);
1639         *pcbopt = NULL;
1640         if (m == NULL || m->m_len == 0) {
1641                 /*
1642                  * Only turning off any previous options.
1643                  */
1644                 if (m != NULL)
1645                         m_free(m);
1646                 return (0);
1647         }
1648
1649         if (m->m_len % sizeof(int32_t))
1650                 goto bad;
1651         /*
1652          * IP first-hop destination address will be stored before
1653          * actual options; move other options back
1654          * and clear it when none present.
1655          */
1656         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1657                 goto bad;
1658         cnt = m->m_len;
1659         m->m_len += sizeof(struct in_addr);
1660         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1661         ovbcopy(mtod(m, caddr_t), cp, cnt);
1662         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1663
1664         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1665                 opt = cp[IPOPT_OPTVAL];
1666                 if (opt == IPOPT_EOL)
1667                         break;
1668                 if (opt == IPOPT_NOP)
1669                         optlen = 1;
1670                 else {
1671                         if (cnt < IPOPT_OLEN + sizeof *cp)
1672                                 goto bad;
1673                         optlen = cp[IPOPT_OLEN];
1674                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1675                                 goto bad;
1676                 }
1677                 switch (opt) {
1678
1679                 default:
1680                         break;
1681
1682                 case IPOPT_LSRR:
1683                 case IPOPT_SSRR:
1684                         /*
1685                          * user process specifies route as:
1686                          *      ->A->B->C->D
1687                          * D must be our final destination (but we can't
1688                          * check that since we may not have connected yet).
1689                          * A is first hop destination, which doesn't appear in
1690                          * actual IP option, but is stored before the options.
1691                          */
1692                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1693                                 goto bad;
1694                         m->m_len -= sizeof(struct in_addr);
1695                         cnt -= sizeof(struct in_addr);
1696                         optlen -= sizeof(struct in_addr);
1697                         cp[IPOPT_OLEN] = optlen;
1698                         /*
1699                          * Move first hop before start of options.
1700                          */
1701                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1702                               sizeof(struct in_addr));
1703                         /*
1704                          * Then copy rest of options back
1705                          * to close up the deleted entry.
1706                          */
1707                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1708                                 &cp[IPOPT_OFFSET+1],
1709                                 cnt - (IPOPT_MINOFF - 1));
1710                         break;
1711                 }
1712         }
1713         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1714                 goto bad;
1715         *pcbopt = m;
1716         return (0);
1717
1718 bad:
1719         m_free(m);
1720         return (EINVAL);
1721 }
1722
1723 /*
1724  * XXX
1725  * The whole multicast option thing needs to be re-thought.
1726  * Several of these options are equally applicable to non-multicast
1727  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1728  * standard option (IP_TTL).
1729  */
1730
1731 /*
1732  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1733  */
1734 static struct ifnet *
1735 ip_multicast_if(struct in_addr *a, int *ifindexp)
1736 {
1737         int ifindex;
1738         struct ifnet *ifp;
1739
1740         if (ifindexp)
1741                 *ifindexp = 0;
1742         if (ntohl(a->s_addr) >> 24 == 0) {
1743                 ifindex = ntohl(a->s_addr) & 0xffffff;
1744                 if (ifindex < 0 || if_index < ifindex)
1745                         return NULL;
1746                 ifp = ifindex2ifnet[ifindex];
1747                 if (ifindexp)
1748                         *ifindexp = ifindex;
1749         } else {
1750                 ifp = INADDR_TO_IFP(a);
1751         }
1752         return ifp;
1753 }
1754
1755 /*
1756  * Set the IP multicast options in response to user setsockopt().
1757  */
1758 static int
1759 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1760 {
1761         int error = 0;
1762         int i;
1763         struct in_addr addr;
1764         struct ip_mreq mreq;
1765         struct ifnet *ifp;
1766         struct ip_moptions *imo = *imop;
1767         int ifindex;
1768
1769         if (imo == NULL) {
1770                 /*
1771                  * No multicast option buffer attached to the pcb;
1772                  * allocate one and initialize to default values.
1773                  */
1774                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1775
1776                 *imop = imo;
1777                 imo->imo_multicast_ifp = NULL;
1778                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1779                 imo->imo_multicast_vif = -1;
1780                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1781                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1782                 imo->imo_num_memberships = 0;
1783         }
1784         switch (sopt->sopt_name) {
1785         /* store an index number for the vif you wanna use in the send */
1786         case IP_MULTICAST_VIF:
1787                 if (legal_vif_num == 0) {
1788                         error = EOPNOTSUPP;
1789                         break;
1790                 }
1791                 error = soopt_to_kbuf(sopt, &i, sizeof i, sizeof i);
1792                 if (error)
1793                         break;
1794                 if (!legal_vif_num(i) && (i != -1)) {
1795                         error = EINVAL;
1796                         break;
1797                 }
1798                 imo->imo_multicast_vif = i;
1799                 break;
1800
1801         case IP_MULTICAST_IF:
1802                 /*
1803                  * Select the interface for outgoing multicast packets.
1804                  */
1805                 error = soopt_to_kbuf(sopt, &addr, sizeof addr, sizeof addr);
1806                 if (error)
1807                         break;
1808
1809                 /*
1810                  * INADDR_ANY is used to remove a previous selection.
1811                  * When no interface is selected, a default one is
1812                  * chosen every time a multicast packet is sent.
1813                  */
1814                 if (addr.s_addr == INADDR_ANY) {
1815                         imo->imo_multicast_ifp = NULL;
1816                         break;
1817                 }
1818                 /*
1819                  * The selected interface is identified by its local
1820                  * IP address.  Find the interface and confirm that
1821                  * it supports multicasting.
1822                  */
1823                 crit_enter();
1824                 ifp = ip_multicast_if(&addr, &ifindex);
1825                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1826                         crit_exit();
1827                         error = EADDRNOTAVAIL;
1828                         break;
1829                 }
1830                 imo->imo_multicast_ifp = ifp;
1831                 if (ifindex)
1832                         imo->imo_multicast_addr = addr;
1833                 else
1834                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1835                 crit_exit();
1836                 break;
1837
1838         case IP_MULTICAST_TTL:
1839                 /*
1840                  * Set the IP time-to-live for outgoing multicast packets.
1841                  * The original multicast API required a char argument,
1842                  * which is inconsistent with the rest of the socket API.
1843                  * We allow either a char or an int.
1844                  */
1845                 if (sopt->sopt_valsize == 1) {
1846                         u_char ttl;
1847                         error = soopt_to_kbuf(sopt, &ttl, 1, 1);
1848                         if (error)
1849                                 break;
1850                         imo->imo_multicast_ttl = ttl;
1851                 } else {
1852                         u_int ttl;
1853                         error = soopt_to_kbuf(sopt, &ttl, sizeof ttl, sizeof ttl);
1854                         if (error)
1855                                 break;
1856                         if (ttl > 255)
1857                                 error = EINVAL;
1858                         else
1859                                 imo->imo_multicast_ttl = ttl;
1860                 }
1861                 break;
1862
1863         case IP_MULTICAST_LOOP:
1864                 /*
1865                  * Set the loopback flag for outgoing multicast packets.
1866                  * Must be zero or one.  The original multicast API required a
1867                  * char argument, which is inconsistent with the rest
1868                  * of the socket API.  We allow either a char or an int.
1869                  */
1870                 if (sopt->sopt_valsize == 1) {
1871                         u_char loop;
1872
1873                         error = soopt_to_kbuf(sopt, &loop, 1, 1);
1874                         if (error)
1875                                 break;
1876                         imo->imo_multicast_loop = !!loop;
1877                 } else {
1878                         u_int loop;
1879
1880                         error = soopt_to_kbuf(sopt, &loop, sizeof loop,
1881                                             sizeof loop);
1882                         if (error)
1883                                 break;
1884                         imo->imo_multicast_loop = !!loop;
1885                 }
1886                 break;
1887
1888         case IP_ADD_MEMBERSHIP:
1889                 /*
1890                  * Add a multicast group membership.
1891                  * Group must be a valid IP multicast address.
1892                  */
1893                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1894                 if (error)
1895                         break;
1896
1897                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1898                         error = EINVAL;
1899                         break;
1900                 }
1901                 crit_enter();
1902                 /*
1903                  * If no interface address was provided, use the interface of
1904                  * the route to the given multicast address.
1905                  */
1906                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1907                         struct sockaddr_in dst;
1908                         struct rtentry *rt;
1909
1910                         bzero(&dst, sizeof(struct sockaddr_in));
1911                         dst.sin_len = sizeof(struct sockaddr_in);
1912                         dst.sin_family = AF_INET;
1913                         dst.sin_addr = mreq.imr_multiaddr;
1914                         rt = rtlookup((struct sockaddr *)&dst);
1915                         if (rt == NULL) {
1916                                 error = EADDRNOTAVAIL;
1917                                 crit_exit();
1918                                 break;
1919                         }
1920                         --rt->rt_refcnt;
1921                         ifp = rt->rt_ifp;
1922                 } else {
1923                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1924                 }
1925
1926                 /*
1927                  * See if we found an interface, and confirm that it
1928                  * supports multicast.
1929                  */
1930                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1931                         error = EADDRNOTAVAIL;
1932                         crit_exit();
1933                         break;
1934                 }
1935                 /*
1936                  * See if the membership already exists or if all the
1937                  * membership slots are full.
1938                  */
1939                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1940                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1941                             imo->imo_membership[i]->inm_addr.s_addr
1942                                                 == mreq.imr_multiaddr.s_addr)
1943                                 break;
1944                 }
1945                 if (i < imo->imo_num_memberships) {
1946                         error = EADDRINUSE;
1947                         crit_exit();
1948                         break;
1949                 }
1950                 if (i == IP_MAX_MEMBERSHIPS) {
1951                         error = ETOOMANYREFS;
1952                         crit_exit();
1953                         break;
1954                 }
1955                 /*
1956                  * Everything looks good; add a new record to the multicast
1957                  * address list for the given interface.
1958                  */
1959                 if ((imo->imo_membership[i] =
1960                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
1961                         error = ENOBUFS;
1962                         crit_exit();
1963                         break;
1964                 }
1965                 ++imo->imo_num_memberships;
1966                 crit_exit();
1967                 break;
1968
1969         case IP_DROP_MEMBERSHIP:
1970                 /*
1971                  * Drop a multicast group membership.
1972                  * Group must be a valid IP multicast address.
1973                  */
1974                 error = soopt_to_kbuf(sopt, &mreq, sizeof mreq, sizeof mreq);
1975                 if (error)
1976                         break;
1977
1978                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1979                         error = EINVAL;
1980                         break;
1981                 }
1982
1983                 crit_enter();
1984                 /*
1985                  * If an interface address was specified, get a pointer
1986                  * to its ifnet structure.
1987                  */
1988                 if (mreq.imr_interface.s_addr == INADDR_ANY)
1989                         ifp = NULL;
1990                 else {
1991                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1992                         if (ifp == NULL) {
1993                                 error = EADDRNOTAVAIL;
1994                                 crit_exit();
1995                                 break;
1996                         }
1997                 }
1998                 /*
1999                  * Find the membership in the membership array.
2000                  */
2001                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2002                         if ((ifp == NULL ||
2003                              imo->imo_membership[i]->inm_ifp == ifp) &&
2004                             imo->imo_membership[i]->inm_addr.s_addr ==
2005                             mreq.imr_multiaddr.s_addr)
2006                                 break;
2007                 }
2008                 if (i == imo->imo_num_memberships) {
2009                         error = EADDRNOTAVAIL;
2010                         crit_exit();
2011                         break;
2012                 }
2013                 /*
2014                  * Give up the multicast address record to which the
2015                  * membership points.
2016                  */
2017                 in_delmulti(imo->imo_membership[i]);
2018                 /*
2019                  * Remove the gap in the membership array.
2020                  */
2021                 for (++i; i < imo->imo_num_memberships; ++i)
2022                         imo->imo_membership[i-1] = imo->imo_membership[i];
2023                 --imo->imo_num_memberships;
2024                 crit_exit();
2025                 break;
2026
2027         default:
2028                 error = EOPNOTSUPP;
2029                 break;
2030         }
2031
2032         /*
2033          * If all options have default values, no need to keep the mbuf.
2034          */
2035         if (imo->imo_multicast_ifp == NULL &&
2036             imo->imo_multicast_vif == -1 &&
2037             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2038             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2039             imo->imo_num_memberships == 0) {
2040                 kfree(*imop, M_IPMOPTS);
2041                 *imop = NULL;
2042         }
2043
2044         return (error);
2045 }
2046
2047 /*
2048  * Return the IP multicast options in response to user getsockopt().
2049  */
2050 static int
2051 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2052 {
2053         struct in_addr addr;
2054         struct in_ifaddr *ia;
2055         int error, optval;
2056         u_char coptval;
2057
2058         error = 0;
2059         switch (sopt->sopt_name) {
2060         case IP_MULTICAST_VIF:
2061                 if (imo != NULL)
2062                         optval = imo->imo_multicast_vif;
2063                 else
2064                         optval = -1;
2065                 soopt_from_kbuf(sopt, &optval, sizeof optval);
2066                 break;
2067
2068         case IP_MULTICAST_IF:
2069                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2070                         addr.s_addr = INADDR_ANY;
2071                 else if (imo->imo_multicast_addr.s_addr) {
2072                         /* return the value user has set */
2073                         addr = imo->imo_multicast_addr;
2074                 } else {
2075                         ia = IFP_TO_IA(imo->imo_multicast_ifp);
2076                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2077                                 : IA_SIN(ia)->sin_addr.s_addr;
2078                 }
2079                 soopt_from_kbuf(sopt, &addr, sizeof addr);
2080                 break;
2081
2082         case IP_MULTICAST_TTL:
2083                 if (imo == NULL)
2084                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2085                 else
2086                         optval = coptval = imo->imo_multicast_ttl;
2087                 if (sopt->sopt_valsize == 1)
2088                         soopt_from_kbuf(sopt, &coptval, 1);
2089                 else
2090                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2091                 break;
2092
2093         case IP_MULTICAST_LOOP:
2094                 if (imo == NULL)
2095                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2096                 else
2097                         optval = coptval = imo->imo_multicast_loop;
2098                 if (sopt->sopt_valsize == 1)
2099                         soopt_from_kbuf(sopt, &coptval, 1);
2100                 else
2101                         soopt_from_kbuf(sopt, &optval, sizeof optval);
2102                 break;
2103
2104         default:
2105                 error = ENOPROTOOPT;
2106                 break;
2107         }
2108         return (error);
2109 }
2110
2111 /*
2112  * Discard the IP multicast options.
2113  */
2114 void
2115 ip_freemoptions(struct ip_moptions *imo)
2116 {
2117         int i;
2118
2119         if (imo != NULL) {
2120                 for (i = 0; i < imo->imo_num_memberships; ++i)
2121                         in_delmulti(imo->imo_membership[i]);
2122                 kfree(imo, M_IPMOPTS);
2123         }
2124 }
2125
2126 /*
2127  * Routine called from ip_output() to loop back a copy of an IP multicast
2128  * packet to the input queue of a specified interface.  Note that this
2129  * calls the output routine of the loopback "driver", but with an interface
2130  * pointer that might NOT be a loopback interface -- evil, but easier than
2131  * replicating that code here.
2132  */
2133 static void
2134 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2135              int hlen)
2136 {
2137         struct ip *ip;
2138         struct mbuf *copym;
2139
2140         copym = m_copypacket(m, MB_DONTWAIT);
2141         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2142                 copym = m_pullup(copym, hlen);
2143         if (copym != NULL) {
2144                 /*
2145                  * if the checksum hasn't been computed, mark it as valid
2146                  */
2147                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2148                         in_delayed_cksum(copym);
2149                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2150                         copym->m_pkthdr.csum_flags |=
2151                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2152                         copym->m_pkthdr.csum_data = 0xffff;
2153                 }
2154                 /*
2155                  * We don't bother to fragment if the IP length is greater
2156                  * than the interface's MTU.  Can this possibly matter?
2157                  */
2158                 ip = mtod(copym, struct ip *);
2159                 ip->ip_len = htons(ip->ip_len);
2160                 ip->ip_off = htons(ip->ip_off);
2161                 ip->ip_sum = 0;
2162                 if (ip->ip_vhl == IP_VHL_BORING) {
2163                         ip->ip_sum = in_cksum_hdr(ip);
2164                 } else {
2165                         ip->ip_sum = in_cksum(copym, hlen);
2166                 }
2167                 /*
2168                  * NB:
2169                  * It's not clear whether there are any lingering
2170                  * reentrancy problems in other areas which might
2171                  * be exposed by using ip_input directly (in
2172                  * particular, everything which modifies the packet
2173                  * in-place).  Yet another option is using the
2174                  * protosw directly to deliver the looped back
2175                  * packet.  For the moment, we'll err on the side
2176                  * of safety by using if_simloop().
2177                  */
2178 #if 1 /* XXX */
2179                 if (dst->sin_family != AF_INET) {
2180                         kprintf("ip_mloopback: bad address family %d\n",
2181                                                 dst->sin_family);
2182                         dst->sin_family = AF_INET;
2183                 }
2184 #endif
2185                 get_mplock();   /* is if_simloop() mpsafe yet? */
2186                 if_simloop(ifp, copym, dst->sin_family, 0);
2187                 rel_mplock();
2188         }
2189 }