- Move IPDIVERT related bits into netinet/ip_divert.c
[dragonfly.git] / sys / netinet / ip_input.c
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *      The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *      This product includes software developed by the University of
49  *      California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *      @(#)ip_input.c  8.2 (Berkeley) 1/4/94
67  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
68  * $DragonFly: src/sys/netinet/ip_input.c,v 1.103 2008/09/13 08:48:42 sephe Exp $
69  */
70
71 #define _IP_VHL
72
73 #include "opt_bootp.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipdn.h"
76 #include "opt_ipdivert.h"
77 #include "opt_ipfilter.h"
78 #include "opt_ipstealth.h"
79 #include "opt_ipsec.h"
80
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/mpipe.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/time.h>
90 #include <sys/globaldata.h>
91 #include <sys/thread.h>
92 #include <sys/kernel.h>
93 #include <sys/syslog.h>
94 #include <sys/sysctl.h>
95 #include <sys/in_cksum.h>
96
97 #include <machine/stdarg.h>
98
99 #include <net/if.h>
100 #include <net/if_types.h>
101 #include <net/if_var.h>
102 #include <net/if_dl.h>
103 #include <net/pfil.h>
104 #include <net/route.h>
105 #include <net/netisr.h>
106
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/in_var.h>
110 #include <netinet/ip.h>
111 #include <netinet/in_pcb.h>
112 #include <netinet/ip_var.h>
113 #include <netinet/ip_icmp.h>
114 #include <netinet/ip_divert.h>
115
116 #include <sys/thread2.h>
117 #include <sys/msgport2.h>
118 #include <net/netmsg2.h>
119
120 #include <sys/socketvar.h>
121
122 #include <net/ipfw/ip_fw.h>
123 #include <net/dummynet/ip_dummynet.h>
124
125 #ifdef IPSEC
126 #include <netinet6/ipsec.h>
127 #include <netproto/key/key.h>
128 #endif
129
130 #ifdef FAST_IPSEC
131 #include <netproto/ipsec/ipsec.h>
132 #include <netproto/ipsec/key.h>
133 #endif
134
135 int rsvp_on = 0;
136 static int ip_rsvp_on;
137 struct socket *ip_rsvpd;
138
139 int ipforwarding = 0;
140 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
141     &ipforwarding, 0, "Enable IP forwarding between interfaces");
142
143 static int ipsendredirects = 1; /* XXX */
144 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
145     &ipsendredirects, 0, "Enable sending IP redirects");
146
147 int ip_defttl = IPDEFTTL;
148 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
149     &ip_defttl, 0, "Maximum TTL on IP packets");
150
151 static int ip_dosourceroute = 0;
152 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
153     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
154
155 static int ip_acceptsourceroute = 0;
156 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
157     CTLFLAG_RW, &ip_acceptsourceroute, 0,
158     "Enable accepting source routed IP packets");
159
160 static int ip_keepfaith = 0;
161 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
162     &ip_keepfaith, 0,
163     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
164
165 static int nipq = 0;    /* total # of reass queues */
166 static int maxnipq;
167 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
168     &maxnipq, 0,
169     "Maximum number of IPv4 fragment reassembly queue entries");
170
171 static int maxfragsperpacket;
172 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
173     &maxfragsperpacket, 0,
174     "Maximum number of IPv4 fragments allowed per packet");
175
176 static int ip_sendsourcequench = 0;
177 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
178     &ip_sendsourcequench, 0,
179     "Enable the transmission of source quench packets");
180
181 int ip_do_randomid = 1;
182 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
183     &ip_do_randomid, 0,
184     "Assign random ip_id values");      
185 /*
186  * XXX - Setting ip_checkinterface mostly implements the receive side of
187  * the Strong ES model described in RFC 1122, but since the routing table
188  * and transmit implementation do not implement the Strong ES model,
189  * setting this to 1 results in an odd hybrid.
190  *
191  * XXX - ip_checkinterface currently must be disabled if you use ipnat
192  * to translate the destination address to another local interface.
193  *
194  * XXX - ip_checkinterface must be disabled if you add IP aliases
195  * to the loopback interface instead of the interface where the
196  * packets for those addresses are received.
197  */
198 static int ip_checkinterface = 0;
199 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
200     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
201
202 #ifdef DIAGNOSTIC
203 static int ipprintfs = 0;
204 #endif
205
206 extern  struct domain inetdomain;
207 extern  struct protosw inetsw[];
208 u_char  ip_protox[IPPROTO_MAX];
209 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
210 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
211                                                 /* inet addr hash table */
212 u_long  in_ifaddrhmask;                         /* mask for hash table */
213
214 struct ip_stats ipstats_percpu[MAXCPU];
215 #ifdef SMP
216 static int
217 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
218 {
219         int cpu, error = 0;
220
221         for (cpu = 0; cpu < ncpus; ++cpu) {
222                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
223                                         sizeof(struct ip_stats))))
224                         break;
225                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
226                                        sizeof(struct ip_stats))))
227                         break;
228         }
229
230         return (error);
231 }
232 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
233     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
234 #else
235 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
236     &ipstat, ip_stats, "IP statistics");
237 #endif
238
239 /* Packet reassembly stuff */
240 #define IPREASS_NHASH_LOG2      6
241 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
242 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
243 #define IPREASS_HASH(x,y)                                               \
244     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
245
246 static struct ipq ipq[IPREASS_NHASH];
247
248 #ifdef IPCTL_DEFMTU
249 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
250     &ip_mtu, 0, "Default MTU");
251 #endif
252
253 #ifdef IPSTEALTH
254 static int ipstealth = 0;
255 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
256 #else
257 static const int ipstealth = 0;
258 #endif
259
260
261 /* Firewall hooks */
262 ip_fw_chk_t *ip_fw_chk_ptr;
263 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
264 int ip_fw_loaded;
265 int fw_enable = 1;
266 int fw_one_pass = 1;
267
268 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
269
270 struct pfil_head inet_pfil_hook;
271
272 /*
273  * struct ip_srcrt_opt is used to store packet state while it travels
274  * through the stack.
275  *
276  * XXX Note that the code even makes assumptions on the size and
277  * alignment of fields inside struct ip_srcrt so e.g. adding some
278  * fields will break the code.  This needs to be fixed.
279  *
280  * We need to save the IP options in case a protocol wants to respond
281  * to an incoming packet over the same route if the packet got here
282  * using IP source routing.  This allows connection establishment and
283  * maintenance when the remote end is on a network that is not known
284  * to us.
285  */
286 struct ip_srcrt {
287         struct  in_addr dst;                    /* final destination */
288         char    nop;                            /* one NOP to align */
289         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
290         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
291 };
292
293 struct ip_srcrt_opt {
294         int             ip_nhops;
295         struct ip_srcrt ip_srcrt;
296 };
297
298 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
299 static struct malloc_pipe ipq_mpipe;
300
301 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
302 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
303 static void             ip_freef(struct ipq *);
304 static void             ip_input_handler(struct netmsg *);
305
306 /*
307  * IP initialization: fill in IP protocol switch table.
308  * All protocols not implemented in kernel go to raw IP protocol handler.
309  */
310 void
311 ip_init(void)
312 {
313         struct protosw *pr;
314         int i;
315 #ifdef SMP
316         int cpu;
317 #endif
318
319         /*
320          * Make sure we can handle a reasonable number of fragments but
321          * cap it at 4000 (XXX).
322          */
323         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
324                     IFQ_MAXLEN, 4000, 0, NULL);
325         for (i = 0; i < ncpus; ++i) {
326                 TAILQ_INIT(&in_ifaddrheads[i]);
327                 in_ifaddrhashtbls[i] =
328                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
329         }
330         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
331         if (pr == NULL)
332                 panic("ip_init");
333         for (i = 0; i < IPPROTO_MAX; i++)
334                 ip_protox[i] = pr - inetsw;
335         for (pr = inetdomain.dom_protosw;
336              pr < inetdomain.dom_protoswNPROTOSW; pr++)
337                 if (pr->pr_domain->dom_family == PF_INET &&
338                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
339                         ip_protox[pr->pr_protocol] = pr - inetsw;
340
341         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
342         inet_pfil_hook.ph_af = AF_INET;
343         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
344                 kprintf("%s: WARNING: unable to register pfil hook, "
345                         "error %d\n", __func__, i);
346         }
347
348         for (i = 0; i < IPREASS_NHASH; i++)
349                 ipq[i].next = ipq[i].prev = &ipq[i];
350
351         maxnipq = nmbclusters / 32;
352         maxfragsperpacket = 16;
353
354         ip_id = time_second & 0xffff;
355
356         /*
357          * Initialize IP statistics counters for each CPU.
358          *
359          */
360 #ifdef SMP
361         for (cpu = 0; cpu < ncpus; ++cpu) {
362                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
363         }
364 #else
365         bzero(&ipstat, sizeof(struct ip_stats));
366 #endif
367
368         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
369 }
370
371 /*
372  * XXX watch out this one. It is perhaps used as a cache for
373  * the most recently used route ? it is cleared in in_addroute()
374  * when a new route is successfully created.
375  */
376 struct route ipforward_rt[MAXCPU];
377
378 /* Do transport protocol processing. */
379 static void
380 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
381 {
382         /*
383          * Switch out to protocol's input routine.
384          */
385         (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
386 }
387
388 static void
389 transport_processing_handler(netmsg_t netmsg)
390 {
391         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
392         struct ip *ip;
393         int hlen;
394
395         ip = mtod(pmsg->nm_packet, struct ip *);
396         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
397
398         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
399         /* netmsg was embedded in the mbuf, do not reply! */
400 }
401
402 static void
403 ip_input_handler(struct netmsg *msg0)
404 {
405         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
406
407         ip_input(m);
408         /* msg0 was embedded in the mbuf, do not reply! */
409 }
410
411 /*
412  * IP input routine.  Checksum and byte swap header.  If fragmented
413  * try to reassemble.  Process options.  Pass to next level.
414  */
415 void
416 ip_input(struct mbuf *m)
417 {
418         struct ip *ip;
419         struct in_ifaddr *ia = NULL;
420         struct in_ifaddr_container *iac;
421         int i, hlen, checkif;
422         u_short sum;
423         struct in_addr pkt_dst;
424         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
425         boolean_t needredispatch = FALSE;
426         struct in_addr odst;                    /* original dst address(NAT) */
427         struct m_tag *mtag;
428         struct sockaddr_in *next_hop = NULL;
429 #ifdef FAST_IPSEC
430         struct tdb_ident *tdbi;
431         struct secpolicy *sp;
432         int error;
433 #endif
434
435         M_ASSERTPKTHDR(m);
436
437         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
438                 /* Next hop */
439                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
440                 KKASSERT(mtag != NULL);
441                 next_hop = m_tag_data(mtag);
442         }
443
444         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
445                 /* dummynet already filtered us */
446                 ip = mtod(m, struct ip *);
447                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
448                 goto iphack;
449         }
450
451         ipstat.ips_total++;
452
453         /* length checks already done in ip_demux() */
454         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
455
456         ip = mtod(m, struct ip *);
457
458         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
459                 ipstat.ips_badvers++;
460                 goto bad;
461         }
462
463         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
464         /* length checks already done in ip_demux() */
465         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
466         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
467
468         /* 127/8 must not appear on wire - RFC1122 */
469         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
470             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
471                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
472                         ipstat.ips_badaddr++;
473                         goto bad;
474                 }
475         }
476
477         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
478                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
479         } else {
480                 if (hlen == sizeof(struct ip)) {
481                         sum = in_cksum_hdr(ip);
482                 } else {
483                         sum = in_cksum(m, hlen);
484                 }
485         }
486         if (sum != 0) {
487                 ipstat.ips_badsum++;
488                 goto bad;
489         }
490
491 #ifdef ALTQ
492         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
493                 /* packet is dropped by traffic conditioner */
494                 return;
495         }
496 #endif
497         /*
498          * Convert fields to host representation.
499          */
500         ip->ip_len = ntohs(ip->ip_len);
501         if (ip->ip_len < hlen) {
502                 ipstat.ips_badlen++;
503                 goto bad;
504         }
505         ip->ip_off = ntohs(ip->ip_off);
506
507         /*
508          * Check that the amount of data in the buffers
509          * is as at least much as the IP header would have us expect.
510          * Trim mbufs if longer than we expect.
511          * Drop packet if shorter than we expect.
512          */
513         if (m->m_pkthdr.len < ip->ip_len) {
514                 ipstat.ips_tooshort++;
515                 goto bad;
516         }
517         if (m->m_pkthdr.len > ip->ip_len) {
518                 if (m->m_len == m->m_pkthdr.len) {
519                         m->m_len = ip->ip_len;
520                         m->m_pkthdr.len = ip->ip_len;
521                 } else
522                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
523         }
524 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
525         /*
526          * Bypass packet filtering for packets from a tunnel (gif).
527          */
528         if (ipsec_gethist(m, NULL))
529                 goto pass;
530 #endif
531
532         /*
533          * IpHack's section.
534          * Right now when no processing on packet has done
535          * and it is still fresh out of network we do our black
536          * deals with it.
537          * - Firewall: deny/allow/divert
538          * - Xlate: translate packet's addr/port (NAT).
539          * - Pipe: pass pkt through dummynet.
540          * - Wrap: fake packet's addr/port <unimpl.>
541          * - Encapsulate: put it in another IP and send out. <unimp.>
542          */
543
544 iphack:
545         /*
546          * If we've been forwarded from the output side, then
547          * skip the firewall a second time
548          */
549         if (next_hop != NULL)
550                 goto ours;
551
552         /*
553          * Run through list of hooks for input packets.
554          *
555          * NB: Beware of the destination address changing (e.g.
556          *     by NAT rewriting). When this happens, tell
557          *     ip_forward to do the right thing.
558          */
559         if (pfil_has_hooks(&inet_pfil_hook)) {
560                 odst = ip->ip_dst;
561                 if (pfil_run_hooks(&inet_pfil_hook, &m,
562                     m->m_pkthdr.rcvif, PFIL_IN)) {
563                         return;
564                 }
565                 if (m == NULL)                  /* consumed by filter */
566                         return;
567                 ip = mtod(m, struct ip *);
568                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
569         }
570
571         if (fw_enable && IPFW_LOADED) {
572                 struct ip_fw_args args;
573                 int tee = 0;
574
575                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
576                         /* Extract info from dummynet tag */
577                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
578                         KKASSERT(mtag != NULL);
579                         args.rule =
580                         ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
581                         KKASSERT(args.rule != NULL);
582
583                         m_tag_delete(m, mtag);
584                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
585                 } else {
586                         args.rule = NULL;
587                 }
588
589                 args.eh = NULL;
590                 args.oif = NULL;
591                 args.m = m;
592                 i = ip_fw_chk_ptr(&args);
593                 m = args.m;
594
595                 if (m == NULL)
596                         return;
597                 ip = mtod(m, struct ip *);      /* just in case m changed */
598
599                 switch (i) {
600                 case IP_FW_PASS:
601                         goto pass;
602
603                 case IP_FW_DENY:
604                         m_freem(m);
605                         return;
606
607                 case IP_FW_DUMMYNET:
608                         /* Send packet to the appropriate pipe */
609                         ip_fw_dn_io_ptr(m, args.cookie, DN_TO_IP_IN, &args);
610                         goto pass;
611
612                 case IP_FW_TEE:
613                         tee = 1;
614                         /* FALL THROUGH */
615
616                 case IP_FW_DIVERT:
617                         if (ip_divert_p != NULL) {
618                                 m = ip_divert_p(m, tee, 1);
619                                 if (m == NULL)
620                                         return;
621                                 ip = mtod(m, struct ip *);
622                                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
623                                 goto pass;
624                         } else {
625                                 m_freem(m);
626                                 return;
627                         }
628
629                 default:
630                         panic("unknown ipfw return value: %d\n", i);
631                 }
632         }
633 pass:
634         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
635                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
636                 KKASSERT(mtag != NULL);
637                 next_hop = m_tag_data(mtag);
638         }
639         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
640                 ip_dn_queue(m);
641                 return;
642         }
643         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
644                 needredispatch = TRUE;
645                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
646         }
647
648         /*
649          * Process options and, if not destined for us,
650          * ship it on.  ip_dooptions returns 1 when an
651          * error was detected (causing an icmp message
652          * to be sent and the original packet to be freed).
653          */
654         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
655                 return;
656
657         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
658          * matter if it is destined to another node, or whether it is
659          * a multicast one, RSVP wants it! and prevents it from being forwarded
660          * anywhere else. Also checks if the rsvp daemon is running before
661          * grabbing the packet.
662          */
663         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
664                 goto ours;
665
666         /*
667          * Check our list of addresses, to see if the packet is for us.
668          * If we don't have any addresses, assume any unicast packet
669          * we receive might be for us (and let the upper layers deal
670          * with it).
671          */
672         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
673             !(m->m_flags & (M_MCAST | M_BCAST)))
674                 goto ours;
675
676         /*
677          * Cache the destination address of the packet; this may be
678          * changed by use of 'ipfw fwd'.
679          */
680         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
681
682         /*
683          * Enable a consistency check between the destination address
684          * and the arrival interface for a unicast packet (the RFC 1122
685          * strong ES model) if IP forwarding is disabled and the packet
686          * is not locally generated and the packet is not subject to
687          * 'ipfw fwd'.
688          *
689          * XXX - Checking also should be disabled if the destination
690          * address is ipnat'ed to a different interface.
691          *
692          * XXX - Checking is incompatible with IP aliases added
693          * to the loopback interface instead of the interface where
694          * the packets are received.
695          */
696         checkif = ip_checkinterface &&
697                   !ipforwarding &&
698                   m->m_pkthdr.rcvif != NULL &&
699                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
700                   next_hop == NULL;
701
702         /*
703          * Check for exact addresses in the hash bucket.
704          */
705         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
706                 ia = iac->ia;
707
708                 /*
709                  * If the address matches, verify that the packet
710                  * arrived via the correct interface if checking is
711                  * enabled.
712                  */
713                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
714                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
715                         goto ours;
716         }
717         ia = NULL;
718
719         /*
720          * Check for broadcast addresses.
721          *
722          * Only accept broadcast packets that arrive via the matching
723          * interface.  Reception of forwarded directed broadcasts would
724          * be handled via ip_forward() and ether_output() with the loopback
725          * into the stack for SIMPLEX interfaces handled by ether_output().
726          */
727         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
728                 struct ifaddr_container *ifac;
729
730                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
731                               ifa_link) {
732                         struct ifaddr *ifa = ifac->ifa;
733
734                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
735                                 continue;
736                         if (ifa->ifa_addr->sa_family != AF_INET)
737                                 continue;
738                         ia = ifatoia(ifa);
739                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
740                                                                 pkt_dst.s_addr)
741                                 goto ours;
742                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
743                                 goto ours;
744 #ifdef BOOTP_COMPAT
745                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
746                                 goto ours;
747 #endif
748                 }
749         }
750         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
751                 struct in_multi *inm;
752
753                 if (ip_mrouter != NULL) {
754                         /*
755                          * If we are acting as a multicast router, all
756                          * incoming multicast packets are passed to the
757                          * kernel-level multicast forwarding function.
758                          * The packet is returned (relatively) intact; if
759                          * ip_mforward() returns a non-zero value, the packet
760                          * must be discarded, else it may be accepted below.
761                          */
762                         if (ip_mforward != NULL &&
763                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
764                                 ipstat.ips_cantforward++;
765                                 m_freem(m);
766                                 return;
767                         }
768
769                         /*
770                          * The process-level routing daemon needs to receive
771                          * all multicast IGMP packets, whether or not this
772                          * host belongs to their destination groups.
773                          */
774                         if (ip->ip_p == IPPROTO_IGMP)
775                                 goto ours;
776                         ipstat.ips_forward++;
777                 }
778                 /*
779                  * See if we belong to the destination multicast group on the
780                  * arrival interface.
781                  */
782                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
783                 if (inm == NULL) {
784                         ipstat.ips_notmember++;
785                         m_freem(m);
786                         return;
787                 }
788                 goto ours;
789         }
790         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
791                 goto ours;
792         if (ip->ip_dst.s_addr == INADDR_ANY)
793                 goto ours;
794
795         /*
796          * FAITH(Firewall Aided Internet Translator)
797          */
798         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
799                 if (ip_keepfaith) {
800                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
801                                 goto ours;
802                 }
803                 m_freem(m);
804                 return;
805         }
806
807         /*
808          * Not for us; forward if possible and desirable.
809          */
810         if (!ipforwarding) {
811                 ipstat.ips_cantforward++;
812                 m_freem(m);
813         } else {
814 #ifdef IPSEC
815                 /*
816                  * Enforce inbound IPsec SPD.
817                  */
818                 if (ipsec4_in_reject(m, NULL)) {
819                         ipsecstat.in_polvio++;
820                         goto bad;
821                 }
822 #endif
823 #ifdef FAST_IPSEC
824                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
825                 crit_enter();
826                 if (mtag != NULL) {
827                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
828                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
829                 } else {
830                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
831                                                    IP_FORWARDING, &error);
832                 }
833                 if (sp == NULL) {       /* NB: can happen if error */
834                         crit_exit();
835                         /*XXX error stat???*/
836                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
837                         goto bad;
838                 }
839
840                 /*
841                  * Check security policy against packet attributes.
842                  */
843                 error = ipsec_in_reject(sp, m);
844                 KEY_FREESP(&sp);
845                 crit_exit();
846                 if (error) {
847                         ipstat.ips_cantforward++;
848                         goto bad;
849                 }
850 #endif
851                 ip_forward(m, using_srcrt, next_hop);
852         }
853         return;
854
855 ours:
856
857         /*
858          * IPSTEALTH: Process non-routing options only
859          * if the packet is destined for us.
860          */
861         if (ipstealth &&
862             hlen > sizeof(struct ip) &&
863             ip_dooptions(m, 1, next_hop))
864                 return;
865
866         /* Count the packet in the ip address stats */
867         if (ia != NULL) {
868                 ia->ia_ifa.if_ipackets++;
869                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
870         }
871
872         /*
873          * If offset or IP_MF are set, must reassemble.
874          * Otherwise, nothing need be done.
875          * (We could look in the reassembly queue to see
876          * if the packet was previously fragmented,
877          * but it's not worth the time; just let them time out.)
878          */
879         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
880                 /*
881                  * Attempt reassembly; if it succeeds, proceed.
882                  * ip_reass() will return a different mbuf.
883                  */
884                 m = ip_reass(m);
885                 if (m == NULL)
886                         return;
887                 ip = mtod(m, struct ip *);
888
889                 /* Get the header length of the reassembled packet */
890                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
891
892                 needredispatch = TRUE;
893         } else {
894                 ip->ip_len -= hlen;
895         }
896
897 #ifdef IPSEC
898         /*
899          * enforce IPsec policy checking if we are seeing last header.
900          * note that we do not visit this with protocols with pcb layer
901          * code - like udp/tcp/raw ip.
902          */
903         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
904             ipsec4_in_reject(m, NULL)) {
905                 ipsecstat.in_polvio++;
906                 goto bad;
907         }
908 #endif
909 #if FAST_IPSEC
910         /*
911          * enforce IPsec policy checking if we are seeing last header.
912          * note that we do not visit this with protocols with pcb layer
913          * code - like udp/tcp/raw ip.
914          */
915         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
916                 /*
917                  * Check if the packet has already had IPsec processing
918                  * done.  If so, then just pass it along.  This tag gets
919                  * set during AH, ESP, etc. input handling, before the
920                  * packet is returned to the ip input queue for delivery.
921                  */
922                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
923                 crit_enter();
924                 if (mtag != NULL) {
925                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
926                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
927                 } else {
928                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
929                                                    IP_FORWARDING, &error);
930                 }
931                 if (sp != NULL) {
932                         /*
933                          * Check security policy against packet attributes.
934                          */
935                         error = ipsec_in_reject(sp, m);
936                         KEY_FREESP(&sp);
937                 } else {
938                         /* XXX error stat??? */
939                         error = EINVAL;
940 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
941                         goto bad;
942                 }
943                 crit_exit();
944                 if (error)
945                         goto bad;
946         }
947 #endif /* FAST_IPSEC */
948
949         ipstat.ips_delivered++;
950         if (needredispatch) {
951                 struct netmsg_packet *pmsg;
952                 lwkt_port_t port;
953
954                 ip->ip_off = htons(ip->ip_off);
955                 ip->ip_len = htons(ip->ip_len);
956                 port = ip_mport_in(&m);
957                 if (port == NULL)
958                         return;
959
960                 pmsg = &m->m_hdr.mh_netmsg;
961                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
962                             transport_processing_handler);
963                 pmsg->nm_packet = m;
964                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
965
966                 ip = mtod(m, struct ip *);
967                 ip->ip_len = ntohs(ip->ip_len);
968                 ip->ip_off = ntohs(ip->ip_off);
969                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
970         } else {
971                 transport_processing_oncpu(m, hlen, ip);
972         }
973         return;
974
975 bad:
976         m_freem(m);
977 }
978
979 /*
980  * Take incoming datagram fragment and try to reassemble it into
981  * whole datagram.  If a chain for reassembly of this datagram already
982  * exists, then it is given as fp; otherwise have to make a chain.
983  */
984 struct mbuf *
985 ip_reass(struct mbuf *m)
986 {
987         struct ip *ip = mtod(m, struct ip *);
988         struct mbuf *p = NULL, *q, *nq;
989         struct mbuf *n;
990         struct ipq *fp = NULL;
991         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
992         int i, next;
993         u_short sum;
994
995         /* If maxnipq is 0, never accept fragments. */
996         if (maxnipq == 0) {
997                 ipstat.ips_fragments++;
998                 ipstat.ips_fragdropped++;
999                 m_freem(m);
1000                 return NULL;
1001         }
1002
1003         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1004         /*
1005          * Look for queue of fragments of this datagram.
1006          */
1007         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
1008                 if (ip->ip_id == fp->ipq_id &&
1009                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1010                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1011                     ip->ip_p == fp->ipq_p)
1012                         goto found;
1013
1014         fp = NULL;
1015
1016         /*
1017          * Enforce upper bound on number of fragmented packets
1018          * for which we attempt reassembly;
1019          * If maxnipq is -1, accept all fragments without limitation.
1020          */
1021         if (nipq > maxnipq && maxnipq > 0) {
1022                 /*
1023                  * drop something from the tail of the current queue
1024                  * before proceeding further
1025                  */
1026                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1027                         for (i = 0; i < IPREASS_NHASH; i++) {
1028                                 if (ipq[i].prev != &ipq[i]) {
1029                                         ipstat.ips_fragtimeout +=
1030                                             ipq[i].prev->ipq_nfrags;
1031                                         ip_freef(ipq[i].prev);
1032                                         break;
1033                                 }
1034                         }
1035                 } else {
1036                         ipstat.ips_fragtimeout +=
1037                             ipq[sum].prev->ipq_nfrags;
1038                         ip_freef(ipq[sum].prev);
1039                 }
1040         }
1041 found:
1042         /*
1043          * Adjust ip_len to not reflect header,
1044          * convert offset of this to bytes.
1045          */
1046         ip->ip_len -= hlen;
1047         if (ip->ip_off & IP_MF) {
1048                 /*
1049                  * Make sure that fragments have a data length
1050                  * that's a non-zero multiple of 8 bytes.
1051                  */
1052                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1053                         ipstat.ips_toosmall++; /* XXX */
1054                         m_freem(m);
1055                         return NULL;
1056                 }
1057                 m->m_flags |= M_FRAG;
1058         } else
1059                 m->m_flags &= ~M_FRAG;
1060         ip->ip_off <<= 3;
1061
1062         ipstat.ips_fragments++;
1063         m->m_pkthdr.header = ip;
1064
1065         /*
1066          * If the hardware has not done csum over this fragment
1067          * then csum_data is not valid at all.
1068          */
1069         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1070             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1071                 m->m_pkthdr.csum_data = 0;
1072                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1073         }
1074
1075         /*
1076          * Presence of header sizes in mbufs
1077          * would confuse code below.
1078          */
1079         m->m_data += hlen;
1080         m->m_len -= hlen;
1081
1082         /*
1083          * If first fragment to arrive, create a reassembly queue.
1084          */
1085         if (fp == NULL) {
1086                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1087                         goto dropfrag;
1088                 insque(fp, &ipq[sum]);
1089                 nipq++;
1090                 fp->ipq_nfrags = 1;
1091                 fp->ipq_ttl = IPFRAGTTL;
1092                 fp->ipq_p = ip->ip_p;
1093                 fp->ipq_id = ip->ip_id;
1094                 fp->ipq_src = ip->ip_src;
1095                 fp->ipq_dst = ip->ip_dst;
1096                 fp->ipq_frags = m;
1097                 m->m_nextpkt = NULL;
1098                 goto inserted;
1099         } else {
1100                 fp->ipq_nfrags++;
1101         }
1102
1103 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1104
1105         /*
1106          * Find a segment which begins after this one does.
1107          */
1108         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1109                 if (GETIP(q)->ip_off > ip->ip_off)
1110                         break;
1111
1112         /*
1113          * If there is a preceding segment, it may provide some of
1114          * our data already.  If so, drop the data from the incoming
1115          * segment.  If it provides all of our data, drop us, otherwise
1116          * stick new segment in the proper place.
1117          *
1118          * If some of the data is dropped from the the preceding
1119          * segment, then it's checksum is invalidated.
1120          */
1121         if (p) {
1122                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1123                 if (i > 0) {
1124                         if (i >= ip->ip_len)
1125                                 goto dropfrag;
1126                         m_adj(m, i);
1127                         m->m_pkthdr.csum_flags = 0;
1128                         ip->ip_off += i;
1129                         ip->ip_len -= i;
1130                 }
1131                 m->m_nextpkt = p->m_nextpkt;
1132                 p->m_nextpkt = m;
1133         } else {
1134                 m->m_nextpkt = fp->ipq_frags;
1135                 fp->ipq_frags = m;
1136         }
1137
1138         /*
1139          * While we overlap succeeding segments trim them or,
1140          * if they are completely covered, dequeue them.
1141          */
1142         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1143              q = nq) {
1144                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1145                 if (i < GETIP(q)->ip_len) {
1146                         GETIP(q)->ip_len -= i;
1147                         GETIP(q)->ip_off += i;
1148                         m_adj(q, i);
1149                         q->m_pkthdr.csum_flags = 0;
1150                         break;
1151                 }
1152                 nq = q->m_nextpkt;
1153                 m->m_nextpkt = nq;
1154                 ipstat.ips_fragdropped++;
1155                 fp->ipq_nfrags--;
1156                 q->m_nextpkt = NULL;
1157                 m_freem(q);
1158         }
1159
1160 inserted:
1161         /*
1162          * Check for complete reassembly and perform frag per packet
1163          * limiting.
1164          *
1165          * Frag limiting is performed here so that the nth frag has
1166          * a chance to complete the packet before we drop the packet.
1167          * As a result, n+1 frags are actually allowed per packet, but
1168          * only n will ever be stored. (n = maxfragsperpacket.)
1169          *
1170          */
1171         next = 0;
1172         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1173                 if (GETIP(q)->ip_off != next) {
1174                         if (fp->ipq_nfrags > maxfragsperpacket) {
1175                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1176                                 ip_freef(fp);
1177                         }
1178                         return (NULL);
1179                 }
1180                 next += GETIP(q)->ip_len;
1181         }
1182         /* Make sure the last packet didn't have the IP_MF flag */
1183         if (p->m_flags & M_FRAG) {
1184                 if (fp->ipq_nfrags > maxfragsperpacket) {
1185                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1186                         ip_freef(fp);
1187                 }
1188                 return (NULL);
1189         }
1190
1191         /*
1192          * Reassembly is complete.  Make sure the packet is a sane size.
1193          */
1194         q = fp->ipq_frags;
1195         ip = GETIP(q);
1196         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1197                 ipstat.ips_toolong++;
1198                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1199                 ip_freef(fp);
1200                 return (NULL);
1201         }
1202
1203         /*
1204          * Concatenate fragments.
1205          */
1206         m = q;
1207         n = m->m_next;
1208         m->m_next = NULL;
1209         m_cat(m, n);
1210         nq = q->m_nextpkt;
1211         q->m_nextpkt = NULL;
1212         for (q = nq; q != NULL; q = nq) {
1213                 nq = q->m_nextpkt;
1214                 q->m_nextpkt = NULL;
1215                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1216                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1217                 m_cat(m, q);
1218         }
1219
1220         /*
1221          * Clean up the 1's complement checksum.  Carry over 16 bits must
1222          * be added back.  This assumes no more then 65535 packet fragments
1223          * were reassembled.  A second carry can also occur (but not a third).
1224          */
1225         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1226                                 (m->m_pkthdr.csum_data >> 16);
1227         if (m->m_pkthdr.csum_data > 0xFFFF)
1228                 m->m_pkthdr.csum_data -= 0xFFFF;
1229
1230         /*
1231          * Create header for new ip packet by
1232          * modifying header of first packet;
1233          * dequeue and discard fragment reassembly header.
1234          * Make header visible.
1235          */
1236         ip->ip_len = next;
1237         ip->ip_src = fp->ipq_src;
1238         ip->ip_dst = fp->ipq_dst;
1239         remque(fp);
1240         nipq--;
1241         mpipe_free(&ipq_mpipe, fp);
1242         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1243         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1244         /* some debugging cruft by sklower, below, will go away soon */
1245         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1246                 int plen = 0;
1247
1248                 for (n = m; n; n = n->m_next)
1249                         plen += n->m_len;
1250                 m->m_pkthdr.len = plen;
1251         }
1252
1253         ipstat.ips_reassembled++;
1254         return (m);
1255
1256 dropfrag:
1257         ipstat.ips_fragdropped++;
1258         if (fp != NULL)
1259                 fp->ipq_nfrags--;
1260         m_freem(m);
1261         return (NULL);
1262
1263 #undef GETIP
1264 }
1265
1266 /*
1267  * Free a fragment reassembly header and all
1268  * associated datagrams.
1269  */
1270 static void
1271 ip_freef(struct ipq *fp)
1272 {
1273         struct mbuf *q;
1274
1275         while (fp->ipq_frags) {
1276                 q = fp->ipq_frags;
1277                 fp->ipq_frags = q->m_nextpkt;
1278                 q->m_nextpkt = NULL;
1279                 m_freem(q);
1280         }
1281         remque(fp);
1282         mpipe_free(&ipq_mpipe, fp);
1283         nipq--;
1284 }
1285
1286 /*
1287  * IP timer processing;
1288  * if a timer expires on a reassembly
1289  * queue, discard it.
1290  */
1291 void
1292 ip_slowtimo(void)
1293 {
1294         struct ipq *fp;
1295         int i;
1296
1297         crit_enter();
1298         for (i = 0; i < IPREASS_NHASH; i++) {
1299                 fp = ipq[i].next;
1300                 if (fp == NULL)
1301                         continue;
1302                 while (fp != &ipq[i]) {
1303                         --fp->ipq_ttl;
1304                         fp = fp->next;
1305                         if (fp->prev->ipq_ttl == 0) {
1306                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1307                                 ip_freef(fp->prev);
1308                         }
1309                 }
1310         }
1311         /*
1312          * If we are over the maximum number of fragments
1313          * (due to the limit being lowered), drain off
1314          * enough to get down to the new limit.
1315          */
1316         if (maxnipq >= 0 && nipq > maxnipq) {
1317                 for (i = 0; i < IPREASS_NHASH; i++) {
1318                         while (nipq > maxnipq &&
1319                                 (ipq[i].next != &ipq[i])) {
1320                                 ipstat.ips_fragdropped +=
1321                                     ipq[i].next->ipq_nfrags;
1322                                 ip_freef(ipq[i].next);
1323                         }
1324                 }
1325         }
1326         ipflow_slowtimo();
1327         crit_exit();
1328 }
1329
1330 /*
1331  * Drain off all datagram fragments.
1332  */
1333 void
1334 ip_drain(void)
1335 {
1336         int i;
1337
1338         for (i = 0; i < IPREASS_NHASH; i++) {
1339                 while (ipq[i].next != &ipq[i]) {
1340                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1341                         ip_freef(ipq[i].next);
1342                 }
1343         }
1344         in_rtqdrain();
1345 }
1346
1347 /*
1348  * Do option processing on a datagram,
1349  * possibly discarding it if bad options are encountered,
1350  * or forwarding it if source-routed.
1351  * The pass argument is used when operating in the IPSTEALTH
1352  * mode to tell what options to process:
1353  * [LS]SRR (pass 0) or the others (pass 1).
1354  * The reason for as many as two passes is that when doing IPSTEALTH,
1355  * non-routing options should be processed only if the packet is for us.
1356  * Returns 1 if packet has been forwarded/freed,
1357  * 0 if the packet should be processed further.
1358  */
1359 static int
1360 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1361 {
1362         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1363         struct ip *ip = mtod(m, struct ip *);
1364         u_char *cp;
1365         struct in_ifaddr *ia;
1366         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1367         boolean_t forward = FALSE;
1368         struct in_addr *sin, dst;
1369         n_time ntime;
1370
1371         dst = ip->ip_dst;
1372         cp = (u_char *)(ip + 1);
1373         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1374         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1375                 opt = cp[IPOPT_OPTVAL];
1376                 if (opt == IPOPT_EOL)
1377                         break;
1378                 if (opt == IPOPT_NOP)
1379                         optlen = 1;
1380                 else {
1381                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1382                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1383                                 goto bad;
1384                         }
1385                         optlen = cp[IPOPT_OLEN];
1386                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1387                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1388                                 goto bad;
1389                         }
1390                 }
1391                 switch (opt) {
1392
1393                 default:
1394                         break;
1395
1396                 /*
1397                  * Source routing with record.
1398                  * Find interface with current destination address.
1399                  * If none on this machine then drop if strictly routed,
1400                  * or do nothing if loosely routed.
1401                  * Record interface address and bring up next address
1402                  * component.  If strictly routed make sure next
1403                  * address is on directly accessible net.
1404                  */
1405                 case IPOPT_LSRR:
1406                 case IPOPT_SSRR:
1407                         if (ipstealth && pass > 0)
1408                                 break;
1409                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1410                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1411                                 goto bad;
1412                         }
1413                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1414                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1415                                 goto bad;
1416                         }
1417                         ipaddr.sin_addr = ip->ip_dst;
1418                         ia = (struct in_ifaddr *)
1419                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1420                         if (ia == NULL) {
1421                                 if (opt == IPOPT_SSRR) {
1422                                         type = ICMP_UNREACH;
1423                                         code = ICMP_UNREACH_SRCFAIL;
1424                                         goto bad;
1425                                 }
1426                                 if (!ip_dosourceroute)
1427                                         goto nosourcerouting;
1428                                 /*
1429                                  * Loose routing, and not at next destination
1430                                  * yet; nothing to do except forward.
1431                                  */
1432                                 break;
1433                         }
1434                         off--;                  /* 0 origin */
1435                         if (off > optlen - (int)sizeof(struct in_addr)) {
1436                                 /*
1437                                  * End of source route.  Should be for us.
1438                                  */
1439                                 if (!ip_acceptsourceroute)
1440                                         goto nosourcerouting;
1441                                 save_rte(m, cp, ip->ip_src);
1442                                 break;
1443                         }
1444                         if (ipstealth)
1445                                 goto dropit;
1446                         if (!ip_dosourceroute) {
1447                                 if (ipforwarding) {
1448                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1449
1450                                         /*
1451                                          * Acting as a router, so generate ICMP
1452                                          */
1453 nosourcerouting:
1454                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1455                                         log(LOG_WARNING,
1456                                             "attempted source route from %s to %s\n",
1457                                             inet_ntoa(ip->ip_src), buf);
1458                                         type = ICMP_UNREACH;
1459                                         code = ICMP_UNREACH_SRCFAIL;
1460                                         goto bad;
1461                                 } else {
1462                                         /*
1463                                          * Not acting as a router,
1464                                          * so silently drop.
1465                                          */
1466 dropit:
1467                                         ipstat.ips_cantforward++;
1468                                         m_freem(m);
1469                                         return (1);
1470                                 }
1471                         }
1472
1473                         /*
1474                          * locate outgoing interface
1475                          */
1476                         memcpy(&ipaddr.sin_addr, cp + off,
1477                             sizeof ipaddr.sin_addr);
1478
1479                         if (opt == IPOPT_SSRR) {
1480 #define INA     struct in_ifaddr *
1481 #define SA      struct sockaddr *
1482                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1483                                                                         == NULL)
1484                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1485                         } else
1486                                 ia = ip_rtaddr(ipaddr.sin_addr,
1487                                                &ipforward_rt[mycpuid]);
1488                         if (ia == NULL) {
1489                                 type = ICMP_UNREACH;
1490                                 code = ICMP_UNREACH_SRCFAIL;
1491                                 goto bad;
1492                         }
1493                         ip->ip_dst = ipaddr.sin_addr;
1494                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1495                             sizeof(struct in_addr));
1496                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1497                         /*
1498                          * Let ip_intr's mcast routing check handle mcast pkts
1499                          */
1500                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1501                         break;
1502
1503                 case IPOPT_RR:
1504                         if (ipstealth && pass == 0)
1505                                 break;
1506                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1507                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1508                                 goto bad;
1509                         }
1510                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1511                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1512                                 goto bad;
1513                         }
1514                         /*
1515                          * If no space remains, ignore.
1516                          */
1517                         off--;                  /* 0 origin */
1518                         if (off > optlen - (int)sizeof(struct in_addr))
1519                                 break;
1520                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1521                             sizeof ipaddr.sin_addr);
1522                         /*
1523                          * locate outgoing interface; if we're the destination,
1524                          * use the incoming interface (should be same).
1525                          */
1526                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1527                             (ia = ip_rtaddr(ipaddr.sin_addr,
1528                                             &ipforward_rt[mycpuid]))
1529                                                                      == NULL) {
1530                                 type = ICMP_UNREACH;
1531                                 code = ICMP_UNREACH_HOST;
1532                                 goto bad;
1533                         }
1534                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1535                             sizeof(struct in_addr));
1536                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1537                         break;
1538
1539                 case IPOPT_TS:
1540                         if (ipstealth && pass == 0)
1541                                 break;
1542                         code = cp - (u_char *)ip;
1543                         if (optlen < 4 || optlen > 40) {
1544                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1545                                 goto bad;
1546                         }
1547                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1548                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1549                                 goto bad;
1550                         }
1551                         if (off > optlen - (int)sizeof(int32_t)) {
1552                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1553                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1554                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1555                                         goto bad;
1556                                 }
1557                                 break;
1558                         }
1559                         off--;                          /* 0 origin */
1560                         sin = (struct in_addr *)(cp + off);
1561                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1562
1563                         case IPOPT_TS_TSONLY:
1564                                 break;
1565
1566                         case IPOPT_TS_TSANDADDR:
1567                                 if (off + sizeof(n_time) +
1568                                     sizeof(struct in_addr) > optlen) {
1569                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1570                                         goto bad;
1571                                 }
1572                                 ipaddr.sin_addr = dst;
1573                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1574                                                             m->m_pkthdr.rcvif);
1575                                 if (ia == NULL)
1576                                         continue;
1577                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1578                                     sizeof(struct in_addr));
1579                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1580                                 off += sizeof(struct in_addr);
1581                                 break;
1582
1583                         case IPOPT_TS_PRESPEC:
1584                                 if (off + sizeof(n_time) +
1585                                     sizeof(struct in_addr) > optlen) {
1586                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1587                                         goto bad;
1588                                 }
1589                                 memcpy(&ipaddr.sin_addr, sin,
1590                                     sizeof(struct in_addr));
1591                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1592                                         continue;
1593                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1594                                 off += sizeof(struct in_addr);
1595                                 break;
1596
1597                         default:
1598                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1599                                 goto bad;
1600                         }
1601                         ntime = iptime();
1602                         memcpy(cp + off, &ntime, sizeof(n_time));
1603                         cp[IPOPT_OFFSET] += sizeof(n_time);
1604                 }
1605         }
1606         if (forward && ipforwarding) {
1607                 ip_forward(m, TRUE, next_hop);
1608                 return (1);
1609         }
1610         return (0);
1611 bad:
1612         icmp_error(m, type, code, 0, 0);
1613         ipstat.ips_badoptions++;
1614         return (1);
1615 }
1616
1617 /*
1618  * Given address of next destination (final or next hop),
1619  * return internet address info of interface to be used to get there.
1620  */
1621 struct in_ifaddr *
1622 ip_rtaddr(struct in_addr dst, struct route *ro)
1623 {
1624         struct sockaddr_in *sin;
1625
1626         sin = (struct sockaddr_in *)&ro->ro_dst;
1627
1628         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1629                 if (ro->ro_rt != NULL) {
1630                         RTFREE(ro->ro_rt);
1631                         ro->ro_rt = NULL;
1632                 }
1633                 sin->sin_family = AF_INET;
1634                 sin->sin_len = sizeof *sin;
1635                 sin->sin_addr = dst;
1636                 rtalloc_ign(ro, RTF_PRCLONING);
1637         }
1638
1639         if (ro->ro_rt == NULL)
1640                 return (NULL);
1641
1642         return (ifatoia(ro->ro_rt->rt_ifa));
1643 }
1644
1645 /*
1646  * Save incoming source route for use in replies,
1647  * to be picked up later by ip_srcroute if the receiver is interested.
1648  */
1649 static void
1650 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1651 {
1652         struct m_tag *mtag;
1653         struct ip_srcrt_opt *opt;
1654         unsigned olen;
1655
1656         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1657         if (mtag == NULL)
1658                 return;
1659         opt = m_tag_data(mtag);
1660
1661         olen = option[IPOPT_OLEN];
1662 #ifdef DIAGNOSTIC
1663         if (ipprintfs)
1664                 kprintf("save_rte: olen %d\n", olen);
1665 #endif
1666         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1667                 m_tag_free(mtag);
1668                 return;
1669         }
1670         bcopy(option, opt->ip_srcrt.srcopt, olen);
1671         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1672         opt->ip_srcrt.dst = dst;
1673         m_tag_prepend(m, mtag);
1674 }
1675
1676 /*
1677  * Retrieve incoming source route for use in replies,
1678  * in the same form used by setsockopt.
1679  * The first hop is placed before the options, will be removed later.
1680  */
1681 struct mbuf *
1682 ip_srcroute(struct mbuf *m0)
1683 {
1684         struct in_addr *p, *q;
1685         struct mbuf *m;
1686         struct m_tag *mtag;
1687         struct ip_srcrt_opt *opt;
1688
1689         if (m0 == NULL)
1690                 return NULL;
1691
1692         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1693         if (mtag == NULL)
1694                 return NULL;
1695         opt = m_tag_data(mtag);
1696
1697         if (opt->ip_nhops == 0)
1698                 return (NULL);
1699         m = m_get(MB_DONTWAIT, MT_HEADER);
1700         if (m == NULL)
1701                 return (NULL);
1702
1703 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1704
1705         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1706         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1707                    sizeof(struct in_addr) + OPTSIZ;
1708 #ifdef DIAGNOSTIC
1709         if (ipprintfs) {
1710                 kprintf("ip_srcroute: nhops %d mlen %d",
1711                         opt->ip_nhops, m->m_len);
1712         }
1713 #endif
1714
1715         /*
1716          * First save first hop for return route
1717          */
1718         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1719         *(mtod(m, struct in_addr *)) = *p--;
1720 #ifdef DIAGNOSTIC
1721         if (ipprintfs)
1722                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1723 #endif
1724
1725         /*
1726          * Copy option fields and padding (nop) to mbuf.
1727          */
1728         opt->ip_srcrt.nop = IPOPT_NOP;
1729         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1730         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1731             OPTSIZ);
1732         q = (struct in_addr *)(mtod(m, caddr_t) +
1733             sizeof(struct in_addr) + OPTSIZ);
1734 #undef OPTSIZ
1735         /*
1736          * Record return path as an IP source route,
1737          * reversing the path (pointers are now aligned).
1738          */
1739         while (p >= opt->ip_srcrt.route) {
1740 #ifdef DIAGNOSTIC
1741                 if (ipprintfs)
1742                         kprintf(" %x", ntohl(q->s_addr));
1743 #endif
1744                 *q++ = *p--;
1745         }
1746         /*
1747          * Last hop goes to final destination.
1748          */
1749         *q = opt->ip_srcrt.dst;
1750         m_tag_delete(m0, mtag);
1751 #ifdef DIAGNOSTIC
1752         if (ipprintfs)
1753                 kprintf(" %x\n", ntohl(q->s_addr));
1754 #endif
1755         return (m);
1756 }
1757
1758 /*
1759  * Strip out IP options.
1760  */
1761 void
1762 ip_stripoptions(struct mbuf *m)
1763 {
1764         int datalen;
1765         struct ip *ip = mtod(m, struct ip *);
1766         caddr_t opts;
1767         int optlen;
1768
1769         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1770         opts = (caddr_t)(ip + 1);
1771         datalen = m->m_len - (sizeof(struct ip) + optlen);
1772         bcopy(opts + optlen, opts, datalen);
1773         m->m_len -= optlen;
1774         if (m->m_flags & M_PKTHDR)
1775                 m->m_pkthdr.len -= optlen;
1776         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1777 }
1778
1779 u_char inetctlerrmap[PRC_NCMDS] = {
1780         0,              0,              0,              0,
1781         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1782         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1783         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1784         0,              0,              0,              0,
1785         ENOPROTOOPT,    ECONNREFUSED
1786 };
1787
1788 /*
1789  * Forward a packet.  If some error occurs return the sender
1790  * an icmp packet.  Note we can't always generate a meaningful
1791  * icmp message because icmp doesn't have a large enough repertoire
1792  * of codes and types.
1793  *
1794  * If not forwarding, just drop the packet.  This could be confusing
1795  * if ipforwarding was zero but some routing protocol was advancing
1796  * us as a gateway to somewhere.  However, we must let the routing
1797  * protocol deal with that.
1798  *
1799  * The using_srcrt parameter indicates whether the packet is being forwarded
1800  * via a source route.
1801  */
1802 void
1803 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1804 {
1805         struct ip *ip = mtod(m, struct ip *);
1806         struct sockaddr_in *ipforward_rtaddr;
1807         struct rtentry *rt;
1808         int error, type = 0, code = 0, destmtu = 0;
1809         struct mbuf *mcopy;
1810         n_long dest;
1811         struct in_addr pkt_dst;
1812         struct route *cache_rt = &ipforward_rt[mycpuid];
1813
1814         dest = INADDR_ANY;
1815         /*
1816          * Cache the destination address of the packet; this may be
1817          * changed by use of 'ipfw fwd'.
1818          */
1819         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1820
1821 #ifdef DIAGNOSTIC
1822         if (ipprintfs)
1823                 kprintf("forward: src %x dst %x ttl %x\n",
1824                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1825 #endif
1826
1827         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1828                 ipstat.ips_cantforward++;
1829                 m_freem(m);
1830                 return;
1831         }
1832         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1833                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1834                 return;
1835         }
1836
1837         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1838         if (cache_rt->ro_rt == NULL ||
1839             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1840                 if (cache_rt->ro_rt != NULL) {
1841                         RTFREE(cache_rt->ro_rt);
1842                         cache_rt->ro_rt = NULL;
1843                 }
1844                 ipforward_rtaddr->sin_family = AF_INET;
1845                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1846                 ipforward_rtaddr->sin_addr = pkt_dst;
1847                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1848                 if (cache_rt->ro_rt == NULL) {
1849                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1850                         return;
1851                 }
1852         }
1853         rt = cache_rt->ro_rt;
1854
1855         /*
1856          * Save the IP header and at most 8 bytes of the payload,
1857          * in case we need to generate an ICMP message to the src.
1858          *
1859          * XXX this can be optimized a lot by saving the data in a local
1860          * buffer on the stack (72 bytes at most), and only allocating the
1861          * mbuf if really necessary. The vast majority of the packets
1862          * are forwarded without having to send an ICMP back (either
1863          * because unnecessary, or because rate limited), so we are
1864          * really we are wasting a lot of work here.
1865          *
1866          * We don't use m_copy() because it might return a reference
1867          * to a shared cluster. Both this function and ip_output()
1868          * assume exclusive access to the IP header in `m', so any
1869          * data in a cluster may change before we reach icmp_error().
1870          */
1871         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1872         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1873                 /*
1874                  * It's probably ok if the pkthdr dup fails (because
1875                  * the deep copy of the tag chain failed), but for now
1876                  * be conservative and just discard the copy since
1877                  * code below may some day want the tags.
1878                  */
1879                 m_free(mcopy);
1880                 mcopy = NULL;
1881         }
1882         if (mcopy != NULL) {
1883                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1884                     (int)ip->ip_len);
1885                 mcopy->m_pkthdr.len = mcopy->m_len;
1886                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1887         }
1888
1889         if (!ipstealth)
1890                 ip->ip_ttl -= IPTTLDEC;
1891
1892         /*
1893          * If forwarding packet using same interface that it came in on,
1894          * perhaps should send a redirect to sender to shortcut a hop.
1895          * Only send redirect if source is sending directly to us,
1896          * and if packet was not source routed (or has any options).
1897          * Also, don't send redirect if forwarding using a default route
1898          * or a route modified by a redirect.
1899          */
1900         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1901             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1902             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1903             ipsendredirects && !using_srcrt && next_hop == NULL) {
1904                 u_long src = ntohl(ip->ip_src.s_addr);
1905                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1906
1907                 if (rt_ifa != NULL &&
1908                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1909                         if (rt->rt_flags & RTF_GATEWAY)
1910                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1911                         else
1912                                 dest = pkt_dst.s_addr;
1913                         /*
1914                          * Router requirements says to only send
1915                          * host redirects.
1916                          */
1917                         type = ICMP_REDIRECT;
1918                         code = ICMP_REDIRECT_HOST;
1919 #ifdef DIAGNOSTIC
1920                         if (ipprintfs)
1921                                 kprintf("redirect (%d) to %x\n", code, dest);
1922 #endif
1923                 }
1924         }
1925
1926         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1927         if (error == 0) {
1928                 ipstat.ips_forward++;
1929                 if (type == 0) {
1930                         if (mcopy) {
1931                                 ipflow_create(cache_rt, mcopy);
1932                                 m_freem(mcopy);
1933                         }
1934                         return;         /* most common case */
1935                 } else {
1936                         ipstat.ips_redirectsent++;
1937                 }
1938         } else {
1939                 ipstat.ips_cantforward++;
1940         }
1941
1942         if (mcopy == NULL)
1943                 return;
1944
1945         /*
1946          * Send ICMP message.
1947          */
1948
1949         switch (error) {
1950
1951         case 0:                         /* forwarded, but need redirect */
1952                 /* type, code set above */
1953                 break;
1954
1955         case ENETUNREACH:               /* shouldn't happen, checked above */
1956         case EHOSTUNREACH:
1957         case ENETDOWN:
1958         case EHOSTDOWN:
1959         default:
1960                 type = ICMP_UNREACH;
1961                 code = ICMP_UNREACH_HOST;
1962                 break;
1963
1964         case EMSGSIZE:
1965                 type = ICMP_UNREACH;
1966                 code = ICMP_UNREACH_NEEDFRAG;
1967 #ifdef IPSEC
1968                 /*
1969                  * If the packet is routed over IPsec tunnel, tell the
1970                  * originator the tunnel MTU.
1971                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1972                  * XXX quickhack!!!
1973                  */
1974                 if (cache_rt->ro_rt != NULL) {
1975                         struct secpolicy *sp = NULL;
1976                         int ipsecerror;
1977                         int ipsechdr;
1978                         struct route *ro;
1979
1980                         sp = ipsec4_getpolicybyaddr(mcopy,
1981                                                     IPSEC_DIR_OUTBOUND,
1982                                                     IP_FORWARDING,
1983                                                     &ipsecerror);
1984
1985                         if (sp == NULL)
1986                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1987                         else {
1988                                 /* count IPsec header size */
1989                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1990                                                          IPSEC_DIR_OUTBOUND,
1991                                                          NULL);
1992
1993                                 /*
1994                                  * find the correct route for outer IPv4
1995                                  * header, compute tunnel MTU.
1996                                  *
1997                                  */
1998                                 if (sp->req != NULL && sp->req->sav != NULL &&
1999                                     sp->req->sav->sah != NULL) {
2000                                         ro = &sp->req->sav->sah->sa_route;
2001                                         if (ro->ro_rt != NULL &&
2002                                             ro->ro_rt->rt_ifp != NULL) {
2003                                                 destmtu =
2004                                                     ro->ro_rt->rt_ifp->if_mtu;
2005                                                 destmtu -= ipsechdr;
2006                                         }
2007                                 }
2008
2009                                 key_freesp(sp);
2010                         }
2011                 }
2012 #elif FAST_IPSEC
2013                 /*
2014                  * If the packet is routed over IPsec tunnel, tell the
2015                  * originator the tunnel MTU.
2016                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2017                  * XXX quickhack!!!
2018                  */
2019                 if (cache_rt->ro_rt != NULL) {
2020                         struct secpolicy *sp = NULL;
2021                         int ipsecerror;
2022                         int ipsechdr;
2023                         struct route *ro;
2024
2025                         sp = ipsec_getpolicybyaddr(mcopy,
2026                                                    IPSEC_DIR_OUTBOUND,
2027                                                    IP_FORWARDING,
2028                                                    &ipsecerror);
2029
2030                         if (sp == NULL)
2031                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2032                         else {
2033                                 /* count IPsec header size */
2034                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2035                                                          IPSEC_DIR_OUTBOUND,
2036                                                          NULL);
2037
2038                                 /*
2039                                  * find the correct route for outer IPv4
2040                                  * header, compute tunnel MTU.
2041                                  */
2042
2043                                 if (sp->req != NULL &&
2044                                     sp->req->sav != NULL &&
2045                                     sp->req->sav->sah != NULL) {
2046                                         ro = &sp->req->sav->sah->sa_route;
2047                                         if (ro->ro_rt != NULL &&
2048                                             ro->ro_rt->rt_ifp != NULL) {
2049                                                 destmtu =
2050                                                     ro->ro_rt->rt_ifp->if_mtu;
2051                                                 destmtu -= ipsechdr;
2052                                         }
2053                                 }
2054
2055                                 KEY_FREESP(&sp);
2056                         }
2057                 }
2058 #else /* !IPSEC && !FAST_IPSEC */
2059                 if (cache_rt->ro_rt != NULL)
2060                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2061 #endif /*IPSEC*/
2062                 ipstat.ips_cantfrag++;
2063                 break;
2064
2065         case ENOBUFS:
2066                 /*
2067                  * A router should not generate ICMP_SOURCEQUENCH as
2068                  * required in RFC1812 Requirements for IP Version 4 Routers.
2069                  * Source quench could be a big problem under DoS attacks,
2070                  * or if the underlying interface is rate-limited.
2071                  * Those who need source quench packets may re-enable them
2072                  * via the net.inet.ip.sendsourcequench sysctl.
2073                  */
2074                 if (!ip_sendsourcequench) {
2075                         m_freem(mcopy);
2076                         return;
2077                 } else {
2078                         type = ICMP_SOURCEQUENCH;
2079                         code = 0;
2080                 }
2081                 break;
2082
2083         case EACCES:                    /* ipfw denied packet */
2084                 m_freem(mcopy);
2085                 return;
2086         }
2087         icmp_error(mcopy, type, code, dest, destmtu);
2088 }
2089
2090 void
2091 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2092                struct mbuf *m)
2093 {
2094         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2095                 struct timeval tv;
2096
2097                 microtime(&tv);
2098                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2099                     SCM_TIMESTAMP, SOL_SOCKET);
2100                 if (*mp)
2101                         mp = &(*mp)->m_next;
2102         }
2103         if (inp->inp_flags & INP_RECVDSTADDR) {
2104                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2105                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2106                 if (*mp)
2107                         mp = &(*mp)->m_next;
2108         }
2109         if (inp->inp_flags & INP_RECVTTL) {
2110                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2111                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2112                 if (*mp)
2113                         mp = &(*mp)->m_next;
2114         }
2115 #ifdef notyet
2116         /* XXX
2117          * Moving these out of udp_input() made them even more broken
2118          * than they already were.
2119          */
2120         /* options were tossed already */
2121         if (inp->inp_flags & INP_RECVOPTS) {
2122                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2123                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2124                 if (*mp)
2125                         mp = &(*mp)->m_next;
2126         }
2127         /* ip_srcroute doesn't do what we want here, need to fix */
2128         if (inp->inp_flags & INP_RECVRETOPTS) {
2129                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2130                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2131                 if (*mp)
2132                         mp = &(*mp)->m_next;
2133         }
2134 #endif
2135         if (inp->inp_flags & INP_RECVIF) {
2136                 struct ifnet *ifp;
2137                 struct sdlbuf {
2138                         struct sockaddr_dl sdl;
2139                         u_char  pad[32];
2140                 } sdlbuf;
2141                 struct sockaddr_dl *sdp;
2142                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2143
2144                 if (((ifp = m->m_pkthdr.rcvif)) &&
2145                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2146                         sdp = IF_LLSOCKADDR(ifp);
2147                         /*
2148                          * Change our mind and don't try copy.
2149                          */
2150                         if ((sdp->sdl_family != AF_LINK) ||
2151                             (sdp->sdl_len > sizeof(sdlbuf))) {
2152                                 goto makedummy;
2153                         }
2154                         bcopy(sdp, sdl2, sdp->sdl_len);
2155                 } else {
2156 makedummy:
2157                         sdl2->sdl_len =
2158                             offsetof(struct sockaddr_dl, sdl_data[0]);
2159                         sdl2->sdl_family = AF_LINK;
2160                         sdl2->sdl_index = 0;
2161                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2162                 }
2163                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2164                         IP_RECVIF, IPPROTO_IP);
2165                 if (*mp)
2166                         mp = &(*mp)->m_next;
2167         }
2168 }
2169
2170 /*
2171  * XXX these routines are called from the upper part of the kernel.
2172  *
2173  * They could also be moved to ip_mroute.c, since all the RSVP
2174  *  handling is done there already.
2175  */
2176 int
2177 ip_rsvp_init(struct socket *so)
2178 {
2179         if (so->so_type != SOCK_RAW ||
2180             so->so_proto->pr_protocol != IPPROTO_RSVP)
2181                 return EOPNOTSUPP;
2182
2183         if (ip_rsvpd != NULL)
2184                 return EADDRINUSE;
2185
2186         ip_rsvpd = so;
2187         /*
2188          * This may seem silly, but we need to be sure we don't over-increment
2189          * the RSVP counter, in case something slips up.
2190          */
2191         if (!ip_rsvp_on) {
2192                 ip_rsvp_on = 1;
2193                 rsvp_on++;
2194         }
2195
2196         return 0;
2197 }
2198
2199 int
2200 ip_rsvp_done(void)
2201 {
2202         ip_rsvpd = NULL;
2203         /*
2204          * This may seem silly, but we need to be sure we don't over-decrement
2205          * the RSVP counter, in case something slips up.
2206          */
2207         if (ip_rsvp_on) {
2208                 ip_rsvp_on = 0;
2209                 rsvp_on--;
2210         }
2211         return 0;
2212 }
2213
2214 void
2215 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2216 {
2217         int off, proto;
2218         __va_list ap;
2219
2220         __va_start(ap, m);
2221         off = __va_arg(ap, int);
2222         proto = __va_arg(ap, int);
2223         __va_end(ap);
2224
2225         if (rsvp_input_p) { /* call the real one if loaded */
2226                 rsvp_input_p(m, off, proto);
2227                 return;
2228         }
2229
2230         /* Can still get packets with rsvp_on = 0 if there is a local member
2231          * of the group to which the RSVP packet is addressed.  But in this
2232          * case we want to throw the packet away.
2233          */
2234
2235         if (!rsvp_on) {
2236                 m_freem(m);
2237                 return;
2238         }
2239
2240         if (ip_rsvpd != NULL) {
2241                 rip_input(m, off, proto);
2242                 return;
2243         }
2244         /* Drop the packet */
2245         m_freem(m);
2246 }