bca6d49b7ad3c1fa579997cfa4219edcc6dc3626
[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 1.90 2008/08/23 04:12:23 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
115 #include <sys/thread2.h>
116 #include <sys/msgport2.h>
117 #include <net/netmsg2.h>
118
119 #include <sys/socketvar.h>
120
121 #include <net/ipfw/ip_fw.h>
122 #include <net/dummynet/ip_dummynet.h>
123
124 #ifdef IPSEC
125 #include <netinet6/ipsec.h>
126 #include <netproto/key/key.h>
127 #endif
128
129 #ifdef FAST_IPSEC
130 #include <netproto/ipsec/ipsec.h>
131 #include <netproto/ipsec/key.h>
132 #endif
133
134 int rsvp_on = 0;
135 static int ip_rsvp_on;
136 struct socket *ip_rsvpd;
137
138 int ipforwarding = 0;
139 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
140     &ipforwarding, 0, "Enable IP forwarding between interfaces");
141
142 static int ipsendredirects = 1; /* XXX */
143 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
144     &ipsendredirects, 0, "Enable sending IP redirects");
145
146 int ip_defttl = IPDEFTTL;
147 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
148     &ip_defttl, 0, "Maximum TTL on IP packets");
149
150 static int ip_dosourceroute = 0;
151 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
152     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
153
154 static int ip_acceptsourceroute = 0;
155 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
156     CTLFLAG_RW, &ip_acceptsourceroute, 0,
157     "Enable accepting source routed IP packets");
158
159 static int ip_keepfaith = 0;
160 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
161     &ip_keepfaith, 0,
162     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
163
164 static int nipq = 0;    /* total # of reass queues */
165 static int maxnipq;
166 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
167     &maxnipq, 0,
168     "Maximum number of IPv4 fragment reassembly queue entries");
169
170 static int maxfragsperpacket;
171 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
172     &maxfragsperpacket, 0,
173     "Maximum number of IPv4 fragments allowed per packet");
174
175 static int ip_sendsourcequench = 0;
176 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
177     &ip_sendsourcequench, 0,
178     "Enable the transmission of source quench packets");
179
180 int ip_do_randomid = 1;
181 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
182     &ip_do_randomid, 0,
183     "Assign random ip_id values");      
184 /*
185  * XXX - Setting ip_checkinterface mostly implements the receive side of
186  * the Strong ES model described in RFC 1122, but since the routing table
187  * and transmit implementation do not implement the Strong ES model,
188  * setting this to 1 results in an odd hybrid.
189  *
190  * XXX - ip_checkinterface currently must be disabled if you use ipnat
191  * to translate the destination address to another local interface.
192  *
193  * XXX - ip_checkinterface must be disabled if you add IP aliases
194  * to the loopback interface instead of the interface where the
195  * packets for those addresses are received.
196  */
197 static int ip_checkinterface = 0;
198 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
199     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
200
201 #ifdef DIAGNOSTIC
202 static int ipprintfs = 0;
203 #endif
204
205 extern  struct domain inetdomain;
206 extern  struct protosw inetsw[];
207 u_char  ip_protox[IPPROTO_MAX];
208 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
209 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
210                                                 /* inet addr hash table */
211 u_long  in_ifaddrhmask;                         /* mask for hash table */
212
213 struct ip_stats ipstats_percpu[MAXCPU];
214 #ifdef SMP
215 static int
216 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
217 {
218         int cpu, error = 0;
219
220         for (cpu = 0; cpu < ncpus; ++cpu) {
221                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
222                                         sizeof(struct ip_stats))))
223                         break;
224                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
225                                        sizeof(struct ip_stats))))
226                         break;
227         }
228
229         return (error);
230 }
231 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
232     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
233 #else
234 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
235     &ipstat, ip_stats, "IP statistics");
236 #endif
237
238 /* Packet reassembly stuff */
239 #define IPREASS_NHASH_LOG2      6
240 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
241 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
242 #define IPREASS_HASH(x,y)                                               \
243     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
244
245 static struct ipq ipq[IPREASS_NHASH];
246
247 #ifdef IPCTL_DEFMTU
248 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
249     &ip_mtu, 0, "Default MTU");
250 #endif
251
252 #ifdef IPSTEALTH
253 static int ipstealth = 0;
254 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
255 #else
256 static const int ipstealth = 0;
257 #endif
258
259
260 /* Firewall hooks */
261 ip_fw_chk_t *ip_fw_chk_ptr;
262 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
263 int ip_fw_loaded;
264 int fw_enable = 1;
265 int fw_one_pass = 1;
266
267 struct pfil_head inet_pfil_hook;
268
269 /*
270  * XXX this is ugly -- the following two global variables are
271  * used to store packet state while it travels through the stack.
272  * Note that the code even makes assumptions on the size and
273  * alignment of fields inside struct ip_srcrt so e.g. adding some
274  * fields will break the code. This needs to be fixed.
275  *
276  * We need to save the IP options in case a protocol wants to respond
277  * to an incoming packet over the same route if the packet got here
278  * using IP source routing.  This allows connection establishment and
279  * maintenance when the remote end is on a network that is not known
280  * to us.
281  */
282 static int ip_nhops = 0;
283
284 static  struct ip_srcrt {
285         struct  in_addr dst;                    /* final destination */
286         char    nop;                            /* one NOP to align */
287         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
288         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
289 } ip_srcrt;
290
291 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
292 static struct malloc_pipe ipq_mpipe;
293
294 static void             save_rte(u_char *, struct in_addr);
295 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
296 static void             ip_freef(struct ipq *);
297 static void             ip_input_handler(struct netmsg *);
298 static struct mbuf      *ip_reass(struct mbuf *, u_int32_t *);
299
300 /*
301  * IP initialization: fill in IP protocol switch table.
302  * All protocols not implemented in kernel go to raw IP protocol handler.
303  */
304 void
305 ip_init(void)
306 {
307         struct protosw *pr;
308         int i;
309 #ifdef SMP
310         int cpu;
311 #endif
312
313         /*
314          * Make sure we can handle a reasonable number of fragments but
315          * cap it at 4000 (XXX).
316          */
317         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
318                     IFQ_MAXLEN, 4000, 0, NULL);
319         for (i = 0; i < ncpus; ++i) {
320                 TAILQ_INIT(&in_ifaddrheads[i]);
321                 in_ifaddrhashtbls[i] =
322                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
323         }
324         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
325         if (pr == NULL)
326                 panic("ip_init");
327         for (i = 0; i < IPPROTO_MAX; i++)
328                 ip_protox[i] = pr - inetsw;
329         for (pr = inetdomain.dom_protosw;
330              pr < inetdomain.dom_protoswNPROTOSW; pr++)
331                 if (pr->pr_domain->dom_family == PF_INET &&
332                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
333                         ip_protox[pr->pr_protocol] = pr - inetsw;
334
335         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
336         inet_pfil_hook.ph_af = AF_INET;
337         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
338                 kprintf("%s: WARNING: unable to register pfil hook, "
339                         "error %d\n", __func__, i);
340         }
341
342         for (i = 0; i < IPREASS_NHASH; i++)
343             ipq[i].next = ipq[i].prev = &ipq[i];
344
345         maxnipq = nmbclusters / 32;
346         maxfragsperpacket = 16;
347
348         ip_id = time_second & 0xffff;
349
350         /*
351          * Initialize IP statistics counters for each CPU.
352          *
353          */
354 #ifdef SMP
355         for (cpu = 0; cpu < ncpus; ++cpu) {
356                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
357         }
358 #else
359         bzero(&ipstat, sizeof(struct ip_stats));
360 #endif
361
362         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
363 }
364
365 /*
366  * XXX watch out this one. It is perhaps used as a cache for
367  * the most recently used route ? it is cleared in in_addroute()
368  * when a new route is successfully created.
369  */
370 struct route ipforward_rt[MAXCPU];
371
372 /* Do transport protocol processing. */
373 static void
374 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
375 {
376         /*
377          * Switch out to protocol's input routine.
378          */
379         (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
380 }
381
382 static void
383 transport_processing_handler(netmsg_t netmsg)
384 {
385         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
386         struct ip *ip;
387         int hlen;
388
389         ip = mtod(pmsg->nm_packet, struct ip *);
390         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
391
392         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
393         /* netmsg was embedded in the mbuf, do not reply! */
394 }
395
396 static void
397 ip_input_handler(struct netmsg *msg0)
398 {
399         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
400
401         ip_input(m);
402         /* msg0 was embedded in the mbuf, do not reply! */
403 }
404
405 /*
406  * IP input routine.  Checksum and byte swap header.  If fragmented
407  * try to reassemble.  Process options.  Pass to next level.
408  */
409 void
410 ip_input(struct mbuf *m)
411 {
412         struct ip *ip;
413         struct in_ifaddr *ia = NULL;
414         struct in_ifaddr_container *iac;
415         int i, hlen, checkif;
416         u_short sum;
417         struct in_addr pkt_dst;
418         u_int32_t divert_info = 0;              /* packet divert/tee info */
419         struct ip_fw_args args;
420         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
421         boolean_t needredispatch = FALSE;
422         struct in_addr odst;                    /* original dst address(NAT) */
423         struct m_tag *mtag;
424         struct sockaddr_in *next_hop = NULL;
425 #ifdef FAST_IPSEC
426         struct tdb_ident *tdbi;
427         struct secpolicy *sp;
428         int error;
429 #endif
430
431         args.eh = NULL;
432         args.oif = NULL;
433         args.rule = NULL;
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                 /* Extract info from dummynet tag */
446                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
447                 KKASSERT(mtag != NULL);
448                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
449                 KKASSERT(args.rule != NULL);
450
451                 m_tag_delete(m, mtag);
452                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
453
454                 /* dummynet already filtered us */
455                 ip = mtod(m, struct ip *);
456                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
457                 goto iphack;
458         }
459
460         ipstat.ips_total++;
461
462         /* length checks already done in ip_demux() */
463         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
464
465         ip = mtod(m, struct ip *);
466
467         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
468                 ipstat.ips_badvers++;
469                 goto bad;
470         }
471
472         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
473         /* length checks already done in ip_demux() */
474         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
475         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
476
477         /* 127/8 must not appear on wire - RFC1122 */
478         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
479             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
480                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
481                         ipstat.ips_badaddr++;
482                         goto bad;
483                 }
484         }
485
486         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
487                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
488         } else {
489                 if (hlen == sizeof(struct ip)) {
490                         sum = in_cksum_hdr(ip);
491                 } else {
492                         sum = in_cksum(m, hlen);
493                 }
494         }
495         if (sum != 0) {
496                 ipstat.ips_badsum++;
497                 goto bad;
498         }
499
500 #ifdef ALTQ
501         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
502                 /* packet is dropped by traffic conditioner */
503                 return;
504         }
505 #endif
506         /*
507          * Convert fields to host representation.
508          */
509         ip->ip_len = ntohs(ip->ip_len);
510         if (ip->ip_len < hlen) {
511                 ipstat.ips_badlen++;
512                 goto bad;
513         }
514         ip->ip_off = ntohs(ip->ip_off);
515
516         /*
517          * Check that the amount of data in the buffers
518          * is as at least much as the IP header would have us expect.
519          * Trim mbufs if longer than we expect.
520          * Drop packet if shorter than we expect.
521          */
522         if (m->m_pkthdr.len < ip->ip_len) {
523                 ipstat.ips_tooshort++;
524                 goto bad;
525         }
526         if (m->m_pkthdr.len > ip->ip_len) {
527                 if (m->m_len == m->m_pkthdr.len) {
528                         m->m_len = ip->ip_len;
529                         m->m_pkthdr.len = ip->ip_len;
530                 } else
531                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
532         }
533 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
534         /*
535          * Bypass packet filtering for packets from a tunnel (gif).
536          */
537         if (ipsec_gethist(m, NULL))
538                 goto pass;
539 #endif
540
541         /*
542          * IpHack's section.
543          * Right now when no processing on packet has done
544          * and it is still fresh out of network we do our black
545          * deals with it.
546          * - Firewall: deny/allow/divert
547          * - Xlate: translate packet's addr/port (NAT).
548          * - Pipe: pass pkt through dummynet.
549          * - Wrap: fake packet's addr/port <unimpl.>
550          * - Encapsulate: put it in another IP and send out. <unimp.>
551          */
552
553 iphack:
554
555         /*
556          * Run through list of hooks for input packets.
557          *
558          * NB: Beware of the destination address changing (e.g.
559          *     by NAT rewriting). When this happens, tell
560          *     ip_forward to do the right thing.
561          */
562         if (pfil_has_hooks(&inet_pfil_hook)) {
563                 odst = ip->ip_dst;
564                 if (pfil_run_hooks(&inet_pfil_hook, &m,
565                     m->m_pkthdr.rcvif, PFIL_IN)) {
566                         return;
567                 }
568                 if (m == NULL)                  /* consumed by filter */
569                         return;
570                 ip = mtod(m, struct ip *);
571                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
572         }
573
574         if (fw_enable && IPFW_LOADED) {
575                 /*
576                  * If we've been forwarded from the output side, then
577                  * skip the firewall a second time
578                  */
579                 if (next_hop != NULL)
580                         goto ours;
581
582                 args.m = m;
583                 i = ip_fw_chk_ptr(&args);
584                 m = args.m;
585
586                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
587                         if (m != NULL)
588                                 m_freem(m);
589                         return;
590                 }
591                 ip = mtod(m, struct ip *);      /* just in case m changed */
592
593                 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
594                         mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
595                         KKASSERT(mtag != NULL);
596                         next_hop = m_tag_data(mtag);
597                 }
598
599                 if (i == 0 && next_hop == NULL) /* common case */
600                         goto pass;
601                 if (i & IP_FW_PORT_DYNT_FLAG) {
602                         /* Send packet to the appropriate pipe */
603                         ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
604                         return;
605                 }
606 #ifdef IPDIVERT
607                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
608                         /* Divert or tee packet */
609                         divert_info = i;
610                         goto ours;
611                 }
612 #endif
613                 if (i == 0 && next_hop != NULL)
614                         goto pass;
615                 /*
616                  * if we get here, the packet must be dropped
617                  */
618                 m_freem(m);
619                 return;
620         }
621 pass:
622
623         /*
624          * Process options and, if not destined for us,
625          * ship it on.  ip_dooptions returns 1 when an
626          * error was detected (causing an icmp message
627          * to be sent and the original packet to be freed).
628          */
629         ip_nhops = 0;           /* for source routed packets */
630         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
631                 return;
632
633         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
634          * matter if it is destined to another node, or whether it is
635          * a multicast one, RSVP wants it! and prevents it from being forwarded
636          * anywhere else. Also checks if the rsvp daemon is running before
637          * grabbing the packet.
638          */
639         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
640                 goto ours;
641
642         /*
643          * Check our list of addresses, to see if the packet is for us.
644          * If we don't have any addresses, assume any unicast packet
645          * we receive might be for us (and let the upper layers deal
646          * with it).
647          */
648         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
649             !(m->m_flags & (M_MCAST | M_BCAST)))
650                 goto ours;
651
652         /*
653          * Cache the destination address of the packet; this may be
654          * changed by use of 'ipfw fwd'.
655          */
656         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
657
658         /*
659          * Enable a consistency check between the destination address
660          * and the arrival interface for a unicast packet (the RFC 1122
661          * strong ES model) if IP forwarding is disabled and the packet
662          * is not locally generated and the packet is not subject to
663          * 'ipfw fwd'.
664          *
665          * XXX - Checking also should be disabled if the destination
666          * address is ipnat'ed to a different interface.
667          *
668          * XXX - Checking is incompatible with IP aliases added
669          * to the loopback interface instead of the interface where
670          * the packets are received.
671          */
672         checkif = ip_checkinterface &&
673                   !ipforwarding &&
674                   m->m_pkthdr.rcvif != NULL &&
675                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
676                   next_hop == NULL;
677
678         /*
679          * Check for exact addresses in the hash bucket.
680          */
681         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
682                 ia = iac->ia;
683
684                 /*
685                  * If the address matches, verify that the packet
686                  * arrived via the correct interface if checking is
687                  * enabled.
688                  */
689                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
690                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
691                         goto ours;
692         }
693         ia = NULL;
694
695         /*
696          * Check for broadcast addresses.
697          *
698          * Only accept broadcast packets that arrive via the matching
699          * interface.  Reception of forwarded directed broadcasts would
700          * be handled via ip_forward() and ether_output() with the loopback
701          * into the stack for SIMPLEX interfaces handled by ether_output().
702          */
703         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
704                 struct ifaddr_container *ifac;
705
706                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
707                               ifa_link) {
708                         struct ifaddr *ifa = ifac->ifa;
709
710                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
711                                 continue;
712                         if (ifa->ifa_addr->sa_family != AF_INET)
713                                 continue;
714                         ia = ifatoia(ifa);
715                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
716                                                                 pkt_dst.s_addr)
717                                 goto ours;
718                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
719                                 goto ours;
720 #ifdef BOOTP_COMPAT
721                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
722                                 goto ours;
723 #endif
724                 }
725         }
726         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
727                 struct in_multi *inm;
728
729                 if (ip_mrouter != NULL) {
730                         /*
731                          * If we are acting as a multicast router, all
732                          * incoming multicast packets are passed to the
733                          * kernel-level multicast forwarding function.
734                          * The packet is returned (relatively) intact; if
735                          * ip_mforward() returns a non-zero value, the packet
736                          * must be discarded, else it may be accepted below.
737                          */
738                         if (ip_mforward != NULL &&
739                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
740                                 ipstat.ips_cantforward++;
741                                 m_freem(m);
742                                 return;
743                         }
744
745                         /*
746                          * The process-level routing daemon needs to receive
747                          * all multicast IGMP packets, whether or not this
748                          * host belongs to their destination groups.
749                          */
750                         if (ip->ip_p == IPPROTO_IGMP)
751                                 goto ours;
752                         ipstat.ips_forward++;
753                 }
754                 /*
755                  * See if we belong to the destination multicast group on the
756                  * arrival interface.
757                  */
758                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
759                 if (inm == NULL) {
760                         ipstat.ips_notmember++;
761                         m_freem(m);
762                         return;
763                 }
764                 goto ours;
765         }
766         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
767                 goto ours;
768         if (ip->ip_dst.s_addr == INADDR_ANY)
769                 goto ours;
770
771         /*
772          * FAITH(Firewall Aided Internet Translator)
773          */
774         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
775                 if (ip_keepfaith) {
776                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
777                                 goto ours;
778                 }
779                 m_freem(m);
780                 return;
781         }
782
783         /*
784          * Not for us; forward if possible and desirable.
785          */
786         if (!ipforwarding) {
787                 ipstat.ips_cantforward++;
788                 m_freem(m);
789         } else {
790 #ifdef IPSEC
791                 /*
792                  * Enforce inbound IPsec SPD.
793                  */
794                 if (ipsec4_in_reject(m, NULL)) {
795                         ipsecstat.in_polvio++;
796                         goto bad;
797                 }
798 #endif
799 #ifdef FAST_IPSEC
800                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
801                 crit_enter();
802                 if (mtag != NULL) {
803                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
804                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
805                 } else {
806                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
807                                                    IP_FORWARDING, &error);
808                 }
809                 if (sp == NULL) {       /* NB: can happen if error */
810                         crit_exit();
811                         /*XXX error stat???*/
812                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
813                         goto bad;
814                 }
815
816                 /*
817                  * Check security policy against packet attributes.
818                  */
819                 error = ipsec_in_reject(sp, m);
820                 KEY_FREESP(&sp);
821                 crit_exit();
822                 if (error) {
823                         ipstat.ips_cantforward++;
824                         goto bad;
825                 }
826 #endif
827                 ip_forward(m, using_srcrt, next_hop);
828         }
829         return;
830
831 ours:
832
833         /*
834          * IPSTEALTH: Process non-routing options only
835          * if the packet is destined for us.
836          */
837         if (ipstealth &&
838             hlen > sizeof(struct ip) &&
839             ip_dooptions(m, 1, next_hop))
840                 return;
841
842         /* Count the packet in the ip address stats */
843         if (ia != NULL) {
844                 ia->ia_ifa.if_ipackets++;
845                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
846         }
847
848         /*
849          * If offset or IP_MF are set, must reassemble.
850          * Otherwise, nothing need be done.
851          * (We could look in the reassembly queue to see
852          * if the packet was previously fragmented,
853          * but it's not worth the time; just let them time out.)
854          */
855         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
856                 /*
857                  * Attempt reassembly; if it succeeds, proceed.
858                  * ip_reass() will return a different mbuf, and update
859                  * the divert info in divert_info.
860                  */
861                 m = ip_reass(m, &divert_info);
862                 if (m == NULL)
863                         return;
864
865                 needredispatch = TRUE;
866                 ip = mtod(m, struct ip *);
867                 /* Get the header length of the reassembled packet */
868                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
869 #ifdef IPDIVERT
870                 /* Restore original checksum before diverting packet */
871                 if (divert_info != 0) {
872                         ip->ip_len += hlen;
873                         ip->ip_len = htons(ip->ip_len);
874                         ip->ip_off = htons(ip->ip_off);
875                         ip->ip_sum = 0;
876                         if (hlen == sizeof(struct ip))
877                                 ip->ip_sum = in_cksum_hdr(ip);
878                         else
879                                 ip->ip_sum = in_cksum(m, hlen);
880                         ip->ip_off = ntohs(ip->ip_off);
881                         ip->ip_len = ntohs(ip->ip_len);
882                         ip->ip_len -= hlen;
883                 }
884 #endif
885         } else {
886                 ip->ip_len -= hlen;
887         }
888
889 #ifdef IPDIVERT
890         /*
891          * Divert or tee packet to the divert protocol if required.
892          */
893         if (divert_info != 0) {
894                 struct mbuf *clone = NULL;
895
896                 /* Clone packet if we're doing a 'tee' */
897                 if ((divert_info & IP_FW_PORT_TEE_FLAG) != 0)
898                         clone = m_dup(m, MB_DONTWAIT);
899
900                 /* Restore packet header fields to original values */
901                 ip->ip_len += hlen;
902                 ip->ip_len = htons(ip->ip_len);
903                 ip->ip_off = htons(ip->ip_off);
904
905                 /* Deliver packet to divert input routine */
906                 divert_packet(m, 1, divert_info & 0xffff);
907                 ipstat.ips_delivered++;
908
909                 /* If 'tee', continue with original packet */
910                 if (clone == NULL)
911                         return;
912                 m = clone;
913                 ip = mtod(m, struct ip *);
914                 ip->ip_len += hlen;
915                 /*
916                  * Jump backwards to complete processing of the
917                  * packet. But first clear divert_info to avoid
918                  * entering this block again.
919                  * We do not need to clear args.divert_rule as
920                  * it will not be used.
921                  *
922                  * XXX Better safe than sorry, remove the DIVERT tag.
923                  */
924                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
925                 if (mtag != NULL)
926                         m_tag_delete(m, mtag);
927                 
928                 divert_info = 0;
929                 goto pass;
930         }
931 #endif
932
933 #ifdef IPSEC
934         /*
935          * enforce IPsec policy checking if we are seeing last header.
936          * note that we do not visit this with protocols with pcb layer
937          * code - like udp/tcp/raw ip.
938          */
939         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
940             ipsec4_in_reject(m, NULL)) {
941                 ipsecstat.in_polvio++;
942                 goto bad;
943         }
944 #endif
945 #if FAST_IPSEC
946         /*
947          * enforce IPsec policy checking if we are seeing last header.
948          * note that we do not visit this with protocols with pcb layer
949          * code - like udp/tcp/raw ip.
950          */
951         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
952                 /*
953                  * Check if the packet has already had IPsec processing
954                  * done.  If so, then just pass it along.  This tag gets
955                  * set during AH, ESP, etc. input handling, before the
956                  * packet is returned to the ip input queue for delivery.
957                  */
958                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
959                 crit_enter();
960                 if (mtag != NULL) {
961                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
962                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
963                 } else {
964                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
965                                                    IP_FORWARDING, &error);
966                 }
967                 if (sp != NULL) {
968                         /*
969                          * Check security policy against packet attributes.
970                          */
971                         error = ipsec_in_reject(sp, m);
972                         KEY_FREESP(&sp);
973                 } else {
974                         /* XXX error stat??? */
975                         error = EINVAL;
976 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
977                         goto bad;
978                 }
979                 crit_exit();
980                 if (error)
981                         goto bad;
982         }
983 #endif /* FAST_IPSEC */
984
985         ipstat.ips_delivered++;
986         if (needredispatch) {
987                 struct netmsg_packet *pmsg;
988                 lwkt_port_t port;
989
990                 ip->ip_off = htons(ip->ip_off);
991                 ip->ip_len = htons(ip->ip_len);
992                 port = ip_mport_in(&m);
993                 if (port == NULL)
994                         return;
995
996                 pmsg = &m->m_hdr.mh_netmsg;
997                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
998                             transport_processing_handler);
999                 pmsg->nm_packet = m;
1000                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
1001
1002                 ip = mtod(m, struct ip *);
1003                 ip->ip_len = ntohs(ip->ip_len);
1004                 ip->ip_off = ntohs(ip->ip_off);
1005                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
1006         } else {
1007                 transport_processing_oncpu(m, hlen, ip);
1008         }
1009         return;
1010
1011 bad:
1012         m_freem(m);
1013 }
1014
1015 /*
1016  * Take incoming datagram fragment and try to reassemble it into
1017  * whole datagram.  If a chain for reassembly of this datagram already
1018  * exists, then it is given as fp; otherwise have to make a chain.
1019  *
1020  * When IPDIVERT enabled, keep additional state with each packet that
1021  * tells us if we need to divert or tee the packet we're building.
1022  * In particular, *divinfo includes the port and TEE flag.
1023  */
1024
1025 static struct mbuf *
1026 ip_reass(struct mbuf *m, u_int32_t *divinfo)
1027 {
1028         struct ip *ip = mtod(m, struct ip *);
1029         struct mbuf *p = NULL, *q, *nq;
1030         struct mbuf *n;
1031         struct ipq *fp = NULL;
1032         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1033         int i, next;
1034         u_short sum;
1035 #ifdef IPDIVERT
1036         struct m_tag *mtag;
1037 #endif
1038
1039         /* If maxnipq is 0, never accept fragments. */
1040         if (maxnipq == 0) {
1041                 ipstat.ips_fragments++;
1042                 ipstat.ips_fragdropped++;
1043                 m_freem(m);
1044                 return NULL;
1045         }
1046
1047         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1048         /*
1049          * Look for queue of fragments of this datagram.
1050          */
1051         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
1052                 if (ip->ip_id == fp->ipq_id &&
1053                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1054                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1055                     ip->ip_p == fp->ipq_p)
1056                         goto found;
1057
1058         fp = NULL;
1059
1060         /*
1061          * Enforce upper bound on number of fragmented packets
1062          * for which we attempt reassembly;
1063          * If maxnipq is -1, accept all fragments without limitation.
1064          */
1065         if (nipq > maxnipq && maxnipq > 0) {
1066                 /*
1067                  * drop something from the tail of the current queue
1068                  * before proceeding further
1069                  */
1070                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1071                         for (i = 0; i < IPREASS_NHASH; i++) {
1072                                 if (ipq[i].prev != &ipq[i]) {
1073                                         ipstat.ips_fragtimeout +=
1074                                             ipq[i].prev->ipq_nfrags;
1075                                         ip_freef(ipq[i].prev);
1076                                         break;
1077                                 }
1078                         }
1079                 } else {
1080                         ipstat.ips_fragtimeout +=
1081                             ipq[sum].prev->ipq_nfrags;
1082                         ip_freef(ipq[sum].prev);
1083                 }
1084         }
1085 found:
1086         /*
1087          * Adjust ip_len to not reflect header,
1088          * convert offset of this to bytes.
1089          */
1090         ip->ip_len -= hlen;
1091         if (ip->ip_off & IP_MF) {
1092                 /*
1093                  * Make sure that fragments have a data length
1094                  * that's a non-zero multiple of 8 bytes.
1095                  */
1096                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1097                         ipstat.ips_toosmall++; /* XXX */
1098                         m_freem(m);
1099                         return NULL;
1100                 }
1101                 m->m_flags |= M_FRAG;
1102         } else
1103                 m->m_flags &= ~M_FRAG;
1104         ip->ip_off <<= 3;
1105
1106         ipstat.ips_fragments++;
1107         m->m_pkthdr.header = ip;
1108
1109         /*
1110          * If the hardware has not done csum over this fragment
1111          * then csum_data is not valid at all.
1112          */
1113         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1114             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1115                 m->m_pkthdr.csum_data = 0;
1116                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1117         }
1118
1119         /*
1120          * Presence of header sizes in mbufs
1121          * would confuse code below.
1122          */
1123         m->m_data += hlen;
1124         m->m_len -= hlen;
1125
1126         /*
1127          * If first fragment to arrive, create a reassembly queue.
1128          */
1129         if (fp == NULL) {
1130                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1131                         goto dropfrag;
1132                 insque(fp, &ipq[sum]);
1133                 nipq++;
1134                 fp->ipq_nfrags = 1;
1135                 fp->ipq_ttl = IPFRAGTTL;
1136                 fp->ipq_p = ip->ip_p;
1137                 fp->ipq_id = ip->ip_id;
1138                 fp->ipq_src = ip->ip_src;
1139                 fp->ipq_dst = ip->ip_dst;
1140                 fp->ipq_frags = m;
1141                 m->m_nextpkt = NULL;
1142 #ifdef IPDIVERT
1143                 fp->ipq_div_info = 0;
1144 #endif
1145                 goto inserted;
1146         } else {
1147                 fp->ipq_nfrags++;
1148         }
1149
1150 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1151
1152         /*
1153          * Find a segment which begins after this one does.
1154          */
1155         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1156                 if (GETIP(q)->ip_off > ip->ip_off)
1157                         break;
1158
1159         /*
1160          * If there is a preceding segment, it may provide some of
1161          * our data already.  If so, drop the data from the incoming
1162          * segment.  If it provides all of our data, drop us, otherwise
1163          * stick new segment in the proper place.
1164          *
1165          * If some of the data is dropped from the the preceding
1166          * segment, then it's checksum is invalidated.
1167          */
1168         if (p) {
1169                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1170                 if (i > 0) {
1171                         if (i >= ip->ip_len)
1172                                 goto dropfrag;
1173                         m_adj(m, i);
1174                         m->m_pkthdr.csum_flags = 0;
1175                         ip->ip_off += i;
1176                         ip->ip_len -= i;
1177                 }
1178                 m->m_nextpkt = p->m_nextpkt;
1179                 p->m_nextpkt = m;
1180         } else {
1181                 m->m_nextpkt = fp->ipq_frags;
1182                 fp->ipq_frags = m;
1183         }
1184
1185         /*
1186          * While we overlap succeeding segments trim them or,
1187          * if they are completely covered, dequeue them.
1188          */
1189         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1190              q = nq) {
1191                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1192                 if (i < GETIP(q)->ip_len) {
1193                         GETIP(q)->ip_len -= i;
1194                         GETIP(q)->ip_off += i;
1195                         m_adj(q, i);
1196                         q->m_pkthdr.csum_flags = 0;
1197                         break;
1198                 }
1199                 nq = q->m_nextpkt;
1200                 m->m_nextpkt = nq;
1201                 ipstat.ips_fragdropped++;
1202                 fp->ipq_nfrags--;
1203                 q->m_nextpkt = NULL;
1204                 m_freem(q);
1205         }
1206
1207 inserted:
1208
1209 #ifdef IPDIVERT
1210         /*
1211          * Transfer firewall instructions to the fragment structure.
1212          * Only trust info in the fragment at offset 0.
1213          */
1214         if (ip->ip_off == 0) {
1215                 fp->ipq_div_info = *divinfo;
1216         } else {
1217                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1218                 if (mtag != NULL)
1219                         m_tag_delete(m, mtag);
1220         }
1221         *divinfo = 0;
1222 #endif
1223
1224         /*
1225          * Check for complete reassembly and perform frag per packet
1226          * limiting.
1227          *
1228          * Frag limiting is performed here so that the nth frag has
1229          * a chance to complete the packet before we drop the packet.
1230          * As a result, n+1 frags are actually allowed per packet, but
1231          * only n will ever be stored. (n = maxfragsperpacket.)
1232          *
1233          */
1234         next = 0;
1235         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1236                 if (GETIP(q)->ip_off != next) {
1237                         if (fp->ipq_nfrags > maxfragsperpacket) {
1238                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1239                                 ip_freef(fp);
1240                         }
1241                         return (NULL);
1242                 }
1243                 next += GETIP(q)->ip_len;
1244         }
1245         /* Make sure the last packet didn't have the IP_MF flag */
1246         if (p->m_flags & M_FRAG) {
1247                 if (fp->ipq_nfrags > maxfragsperpacket) {
1248                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1249                         ip_freef(fp);
1250                 }
1251                 return (NULL);
1252         }
1253
1254         /*
1255          * Reassembly is complete.  Make sure the packet is a sane size.
1256          */
1257         q = fp->ipq_frags;
1258         ip = GETIP(q);
1259         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1260                 ipstat.ips_toolong++;
1261                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1262                 ip_freef(fp);
1263                 return (NULL);
1264         }
1265
1266         /*
1267          * Concatenate fragments.
1268          */
1269         m = q;
1270         n = m->m_next;
1271         m->m_next = NULL;
1272         m_cat(m, n);
1273         nq = q->m_nextpkt;
1274         q->m_nextpkt = NULL;
1275         for (q = nq; q != NULL; q = nq) {
1276                 nq = q->m_nextpkt;
1277                 q->m_nextpkt = NULL;
1278                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1279                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1280                 m_cat(m, q);
1281         }
1282
1283         /*
1284          * Clean up the 1's complement checksum.  Carry over 16 bits must
1285          * be added back.  This assumes no more then 65535 packet fragments
1286          * were reassembled.  A second carry can also occur (but not a third).
1287          */
1288         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1289                                 (m->m_pkthdr.csum_data >> 16);
1290         if (m->m_pkthdr.csum_data > 0xFFFF)
1291                 m->m_pkthdr.csum_data -= 0xFFFF;
1292
1293
1294 #ifdef IPDIVERT
1295         /*
1296          * Extract firewall instructions from the fragment structure.
1297          */
1298         *divinfo = fp->ipq_div_info;
1299 #endif
1300
1301         /*
1302          * Create header for new ip packet by
1303          * modifying header of first packet;
1304          * dequeue and discard fragment reassembly header.
1305          * Make header visible.
1306          */
1307         ip->ip_len = next;
1308         ip->ip_src = fp->ipq_src;
1309         ip->ip_dst = fp->ipq_dst;
1310         remque(fp);
1311         nipq--;
1312         mpipe_free(&ipq_mpipe, fp);
1313         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1314         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1315         /* some debugging cruft by sklower, below, will go away soon */
1316         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1317                 int plen = 0;
1318
1319                 for (n = m; n; n = n->m_next)
1320                         plen += n->m_len;
1321                 m->m_pkthdr.len = plen;
1322         }
1323
1324         ipstat.ips_reassembled++;
1325         return (m);
1326
1327 dropfrag:
1328 #ifdef IPDIVERT
1329         *divinfo = 0;
1330 #endif
1331         ipstat.ips_fragdropped++;
1332         if (fp != NULL)
1333                 fp->ipq_nfrags--;
1334         m_freem(m);
1335         return (NULL);
1336
1337 #undef GETIP
1338 }
1339
1340 /*
1341  * Free a fragment reassembly header and all
1342  * associated datagrams.
1343  */
1344 static void
1345 ip_freef(struct ipq *fp)
1346 {
1347         struct mbuf *q;
1348
1349         while (fp->ipq_frags) {
1350                 q = fp->ipq_frags;
1351                 fp->ipq_frags = q->m_nextpkt;
1352                 q->m_nextpkt = NULL;
1353                 m_freem(q);
1354         }
1355         remque(fp);
1356         mpipe_free(&ipq_mpipe, fp);
1357         nipq--;
1358 }
1359
1360 /*
1361  * IP timer processing;
1362  * if a timer expires on a reassembly
1363  * queue, discard it.
1364  */
1365 void
1366 ip_slowtimo(void)
1367 {
1368         struct ipq *fp;
1369         int i;
1370
1371         crit_enter();
1372         for (i = 0; i < IPREASS_NHASH; i++) {
1373                 fp = ipq[i].next;
1374                 if (fp == NULL)
1375                         continue;
1376                 while (fp != &ipq[i]) {
1377                         --fp->ipq_ttl;
1378                         fp = fp->next;
1379                         if (fp->prev->ipq_ttl == 0) {
1380                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1381                                 ip_freef(fp->prev);
1382                         }
1383                 }
1384         }
1385         /*
1386          * If we are over the maximum number of fragments
1387          * (due to the limit being lowered), drain off
1388          * enough to get down to the new limit.
1389          */
1390         if (maxnipq >= 0 && nipq > maxnipq) {
1391                 for (i = 0; i < IPREASS_NHASH; i++) {
1392                         while (nipq > maxnipq &&
1393                                 (ipq[i].next != &ipq[i])) {
1394                                 ipstat.ips_fragdropped +=
1395                                     ipq[i].next->ipq_nfrags;
1396                                 ip_freef(ipq[i].next);
1397                         }
1398                 }
1399         }
1400         ipflow_slowtimo();
1401         crit_exit();
1402 }
1403
1404 /*
1405  * Drain off all datagram fragments.
1406  */
1407 void
1408 ip_drain(void)
1409 {
1410         int i;
1411
1412         for (i = 0; i < IPREASS_NHASH; i++) {
1413                 while (ipq[i].next != &ipq[i]) {
1414                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1415                         ip_freef(ipq[i].next);
1416                 }
1417         }
1418         in_rtqdrain();
1419 }
1420
1421 /*
1422  * Do option processing on a datagram,
1423  * possibly discarding it if bad options are encountered,
1424  * or forwarding it if source-routed.
1425  * The pass argument is used when operating in the IPSTEALTH
1426  * mode to tell what options to process:
1427  * [LS]SRR (pass 0) or the others (pass 1).
1428  * The reason for as many as two passes is that when doing IPSTEALTH,
1429  * non-routing options should be processed only if the packet is for us.
1430  * Returns 1 if packet has been forwarded/freed,
1431  * 0 if the packet should be processed further.
1432  */
1433 static int
1434 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1435 {
1436         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1437         struct ip *ip = mtod(m, struct ip *);
1438         u_char *cp;
1439         struct in_ifaddr *ia;
1440         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1441         boolean_t forward = FALSE;
1442         struct in_addr *sin, dst;
1443         n_time ntime;
1444
1445         dst = ip->ip_dst;
1446         cp = (u_char *)(ip + 1);
1447         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1448         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1449                 opt = cp[IPOPT_OPTVAL];
1450                 if (opt == IPOPT_EOL)
1451                         break;
1452                 if (opt == IPOPT_NOP)
1453                         optlen = 1;
1454                 else {
1455                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1456                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1457                                 goto bad;
1458                         }
1459                         optlen = cp[IPOPT_OLEN];
1460                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1461                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1462                                 goto bad;
1463                         }
1464                 }
1465                 switch (opt) {
1466
1467                 default:
1468                         break;
1469
1470                 /*
1471                  * Source routing with record.
1472                  * Find interface with current destination address.
1473                  * If none on this machine then drop if strictly routed,
1474                  * or do nothing if loosely routed.
1475                  * Record interface address and bring up next address
1476                  * component.  If strictly routed make sure next
1477                  * address is on directly accessible net.
1478                  */
1479                 case IPOPT_LSRR:
1480                 case IPOPT_SSRR:
1481                         if (ipstealth && pass > 0)
1482                                 break;
1483                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1484                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1485                                 goto bad;
1486                         }
1487                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1488                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1489                                 goto bad;
1490                         }
1491                         ipaddr.sin_addr = ip->ip_dst;
1492                         ia = (struct in_ifaddr *)
1493                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1494                         if (ia == NULL) {
1495                                 if (opt == IPOPT_SSRR) {
1496                                         type = ICMP_UNREACH;
1497                                         code = ICMP_UNREACH_SRCFAIL;
1498                                         goto bad;
1499                                 }
1500                                 if (!ip_dosourceroute)
1501                                         goto nosourcerouting;
1502                                 /*
1503                                  * Loose routing, and not at next destination
1504                                  * yet; nothing to do except forward.
1505                                  */
1506                                 break;
1507                         }
1508                         off--;                  /* 0 origin */
1509                         if (off > optlen - (int)sizeof(struct in_addr)) {
1510                                 /*
1511                                  * End of source route.  Should be for us.
1512                                  */
1513                                 if (!ip_acceptsourceroute)
1514                                         goto nosourcerouting;
1515                                 save_rte(cp, ip->ip_src);
1516                                 break;
1517                         }
1518                         if (ipstealth)
1519                                 goto dropit;
1520                         if (!ip_dosourceroute) {
1521                                 if (ipforwarding) {
1522                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1523
1524                                         /*
1525                                          * Acting as a router, so generate ICMP
1526                                          */
1527 nosourcerouting:
1528                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1529                                         log(LOG_WARNING,
1530                                             "attempted source route from %s to %s\n",
1531                                             inet_ntoa(ip->ip_src), buf);
1532                                         type = ICMP_UNREACH;
1533                                         code = ICMP_UNREACH_SRCFAIL;
1534                                         goto bad;
1535                                 } else {
1536                                         /*
1537                                          * Not acting as a router,
1538                                          * so silently drop.
1539                                          */
1540 dropit:
1541                                         ipstat.ips_cantforward++;
1542                                         m_freem(m);
1543                                         return (1);
1544                                 }
1545                         }
1546
1547                         /*
1548                          * locate outgoing interface
1549                          */
1550                         memcpy(&ipaddr.sin_addr, cp + off,
1551                             sizeof ipaddr.sin_addr);
1552
1553                         if (opt == IPOPT_SSRR) {
1554 #define INA     struct in_ifaddr *
1555 #define SA      struct sockaddr *
1556                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1557                                                                         == NULL)
1558                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1559                         } else
1560                                 ia = ip_rtaddr(ipaddr.sin_addr,
1561                                                &ipforward_rt[mycpuid]);
1562                         if (ia == NULL) {
1563                                 type = ICMP_UNREACH;
1564                                 code = ICMP_UNREACH_SRCFAIL;
1565                                 goto bad;
1566                         }
1567                         ip->ip_dst = ipaddr.sin_addr;
1568                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1569                             sizeof(struct in_addr));
1570                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1571                         /*
1572                          * Let ip_intr's mcast routing check handle mcast pkts
1573                          */
1574                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1575                         break;
1576
1577                 case IPOPT_RR:
1578                         if (ipstealth && pass == 0)
1579                                 break;
1580                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1581                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1582                                 goto bad;
1583                         }
1584                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1585                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1586                                 goto bad;
1587                         }
1588                         /*
1589                          * If no space remains, ignore.
1590                          */
1591                         off--;                  /* 0 origin */
1592                         if (off > optlen - (int)sizeof(struct in_addr))
1593                                 break;
1594                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1595                             sizeof ipaddr.sin_addr);
1596                         /*
1597                          * locate outgoing interface; if we're the destination,
1598                          * use the incoming interface (should be same).
1599                          */
1600                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1601                             (ia = ip_rtaddr(ipaddr.sin_addr,
1602                                             &ipforward_rt[mycpuid]))
1603                                                                      == NULL) {
1604                                 type = ICMP_UNREACH;
1605                                 code = ICMP_UNREACH_HOST;
1606                                 goto bad;
1607                         }
1608                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1609                             sizeof(struct in_addr));
1610                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1611                         break;
1612
1613                 case IPOPT_TS:
1614                         if (ipstealth && pass == 0)
1615                                 break;
1616                         code = cp - (u_char *)ip;
1617                         if (optlen < 4 || optlen > 40) {
1618                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1619                                 goto bad;
1620                         }
1621                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1622                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1623                                 goto bad;
1624                         }
1625                         if (off > optlen - (int)sizeof(int32_t)) {
1626                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1627                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1628                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1629                                         goto bad;
1630                                 }
1631                                 break;
1632                         }
1633                         off--;                          /* 0 origin */
1634                         sin = (struct in_addr *)(cp + off);
1635                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1636
1637                         case IPOPT_TS_TSONLY:
1638                                 break;
1639
1640                         case IPOPT_TS_TSANDADDR:
1641                                 if (off + sizeof(n_time) +
1642                                     sizeof(struct in_addr) > optlen) {
1643                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1644                                         goto bad;
1645                                 }
1646                                 ipaddr.sin_addr = dst;
1647                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1648                                                             m->m_pkthdr.rcvif);
1649                                 if (ia == NULL)
1650                                         continue;
1651                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1652                                     sizeof(struct in_addr));
1653                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1654                                 off += sizeof(struct in_addr);
1655                                 break;
1656
1657                         case IPOPT_TS_PRESPEC:
1658                                 if (off + sizeof(n_time) +
1659                                     sizeof(struct in_addr) > optlen) {
1660                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1661                                         goto bad;
1662                                 }
1663                                 memcpy(&ipaddr.sin_addr, sin,
1664                                     sizeof(struct in_addr));
1665                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1666                                         continue;
1667                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1668                                 off += sizeof(struct in_addr);
1669                                 break;
1670
1671                         default:
1672                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1673                                 goto bad;
1674                         }
1675                         ntime = iptime();
1676                         memcpy(cp + off, &ntime, sizeof(n_time));
1677                         cp[IPOPT_OFFSET] += sizeof(n_time);
1678                 }
1679         }
1680         if (forward && ipforwarding) {
1681                 ip_forward(m, TRUE, next_hop);
1682                 return (1);
1683         }
1684         return (0);
1685 bad:
1686         icmp_error(m, type, code, 0, 0);
1687         ipstat.ips_badoptions++;
1688         return (1);
1689 }
1690
1691 /*
1692  * Given address of next destination (final or next hop),
1693  * return internet address info of interface to be used to get there.
1694  */
1695 struct in_ifaddr *
1696 ip_rtaddr(struct in_addr dst, struct route *ro)
1697 {
1698         struct sockaddr_in *sin;
1699
1700         sin = (struct sockaddr_in *)&ro->ro_dst;
1701
1702         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1703                 if (ro->ro_rt != NULL) {
1704                         RTFREE(ro->ro_rt);
1705                         ro->ro_rt = NULL;
1706                 }
1707                 sin->sin_family = AF_INET;
1708                 sin->sin_len = sizeof *sin;
1709                 sin->sin_addr = dst;
1710                 rtalloc_ign(ro, RTF_PRCLONING);
1711         }
1712
1713         if (ro->ro_rt == NULL)
1714                 return (NULL);
1715
1716         return (ifatoia(ro->ro_rt->rt_ifa));
1717 }
1718
1719 /*
1720  * Save incoming source route for use in replies,
1721  * to be picked up later by ip_srcroute if the receiver is interested.
1722  */
1723 void
1724 save_rte(u_char *option, struct in_addr dst)
1725 {
1726         unsigned olen;
1727
1728         olen = option[IPOPT_OLEN];
1729 #ifdef DIAGNOSTIC
1730         if (ipprintfs)
1731                 kprintf("save_rte: olen %d\n", olen);
1732 #endif
1733         if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst)))
1734                 return;
1735         bcopy(option, ip_srcrt.srcopt, olen);
1736         ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1737         ip_srcrt.dst = dst;
1738 }
1739
1740 /*
1741  * Retrieve incoming source route for use in replies,
1742  * in the same form used by setsockopt.
1743  * The first hop is placed before the options, will be removed later.
1744  */
1745 struct mbuf *
1746 ip_srcroute(void)
1747 {
1748         struct in_addr *p, *q;
1749         struct mbuf *m;
1750
1751         if (ip_nhops == 0)
1752                 return (NULL);
1753         m = m_get(MB_DONTWAIT, MT_HEADER);
1754         if (m == NULL)
1755                 return (NULL);
1756
1757 #define OPTSIZ  (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt))
1758
1759         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1760         m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) +
1761             OPTSIZ;
1762 #ifdef DIAGNOSTIC
1763         if (ipprintfs)
1764                 kprintf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
1765 #endif
1766
1767         /*
1768          * First save first hop for return route
1769          */
1770         p = &ip_srcrt.route[ip_nhops - 1];
1771         *(mtod(m, struct in_addr *)) = *p--;
1772 #ifdef DIAGNOSTIC
1773         if (ipprintfs)
1774                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1775 #endif
1776
1777         /*
1778          * Copy option fields and padding (nop) to mbuf.
1779          */
1780         ip_srcrt.nop = IPOPT_NOP;
1781         ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1782         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &ip_srcrt.nop,
1783             OPTSIZ);
1784         q = (struct in_addr *)(mtod(m, caddr_t) +
1785             sizeof(struct in_addr) + OPTSIZ);
1786 #undef OPTSIZ
1787         /*
1788          * Record return path as an IP source route,
1789          * reversing the path (pointers are now aligned).
1790          */
1791         while (p >= ip_srcrt.route) {
1792 #ifdef DIAGNOSTIC
1793                 if (ipprintfs)
1794                         kprintf(" %x", ntohl(q->s_addr));
1795 #endif
1796                 *q++ = *p--;
1797         }
1798         /*
1799          * Last hop goes to final destination.
1800          */
1801         *q = ip_srcrt.dst;
1802 #ifdef DIAGNOSTIC
1803         if (ipprintfs)
1804                 kprintf(" %x\n", ntohl(q->s_addr));
1805 #endif
1806         return (m);
1807 }
1808
1809 /*
1810  * Strip out IP options.
1811  */
1812 void
1813 ip_stripoptions(struct mbuf *m)
1814 {
1815         int datalen;
1816         struct ip *ip = mtod(m, struct ip *);
1817         caddr_t opts;
1818         int optlen;
1819
1820         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1821         opts = (caddr_t)(ip + 1);
1822         datalen = m->m_len - (sizeof(struct ip) + optlen);
1823         bcopy(opts + optlen, opts, datalen);
1824         m->m_len -= optlen;
1825         if (m->m_flags & M_PKTHDR)
1826                 m->m_pkthdr.len -= optlen;
1827         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1828 }
1829
1830 u_char inetctlerrmap[PRC_NCMDS] = {
1831         0,              0,              0,              0,
1832         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1833         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1834         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1835         0,              0,              0,              0,
1836         ENOPROTOOPT,    ECONNREFUSED
1837 };
1838
1839 /*
1840  * Forward a packet.  If some error occurs return the sender
1841  * an icmp packet.  Note we can't always generate a meaningful
1842  * icmp message because icmp doesn't have a large enough repertoire
1843  * of codes and types.
1844  *
1845  * If not forwarding, just drop the packet.  This could be confusing
1846  * if ipforwarding was zero but some routing protocol was advancing
1847  * us as a gateway to somewhere.  However, we must let the routing
1848  * protocol deal with that.
1849  *
1850  * The using_srcrt parameter indicates whether the packet is being forwarded
1851  * via a source route.
1852  */
1853 void
1854 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1855 {
1856         struct ip *ip = mtod(m, struct ip *);
1857         struct sockaddr_in *ipforward_rtaddr;
1858         struct rtentry *rt;
1859         int error, type = 0, code = 0, destmtu = 0;
1860         struct mbuf *mcopy;
1861         n_long dest;
1862         struct in_addr pkt_dst;
1863         struct route *cache_rt = &ipforward_rt[mycpuid];
1864
1865         dest = INADDR_ANY;
1866         /*
1867          * Cache the destination address of the packet; this may be
1868          * changed by use of 'ipfw fwd'.
1869          */
1870         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1871
1872 #ifdef DIAGNOSTIC
1873         if (ipprintfs)
1874                 kprintf("forward: src %x dst %x ttl %x\n",
1875                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1876 #endif
1877
1878         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1879                 ipstat.ips_cantforward++;
1880                 m_freem(m);
1881                 return;
1882         }
1883         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1884                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1885                 return;
1886         }
1887
1888         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1889         if (cache_rt->ro_rt == NULL ||
1890             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1891                 if (cache_rt->ro_rt != NULL) {
1892                         RTFREE(cache_rt->ro_rt);
1893                         cache_rt->ro_rt = NULL;
1894                 }
1895                 ipforward_rtaddr->sin_family = AF_INET;
1896                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1897                 ipforward_rtaddr->sin_addr = pkt_dst;
1898                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1899                 if (cache_rt->ro_rt == NULL) {
1900                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1901                         return;
1902                 }
1903         }
1904         rt = cache_rt->ro_rt;
1905
1906         /*
1907          * Save the IP header and at most 8 bytes of the payload,
1908          * in case we need to generate an ICMP message to the src.
1909          *
1910          * XXX this can be optimized a lot by saving the data in a local
1911          * buffer on the stack (72 bytes at most), and only allocating the
1912          * mbuf if really necessary. The vast majority of the packets
1913          * are forwarded without having to send an ICMP back (either
1914          * because unnecessary, or because rate limited), so we are
1915          * really we are wasting a lot of work here.
1916          *
1917          * We don't use m_copy() because it might return a reference
1918          * to a shared cluster. Both this function and ip_output()
1919          * assume exclusive access to the IP header in `m', so any
1920          * data in a cluster may change before we reach icmp_error().
1921          */
1922         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1923         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1924                 /*
1925                  * It's probably ok if the pkthdr dup fails (because
1926                  * the deep copy of the tag chain failed), but for now
1927                  * be conservative and just discard the copy since
1928                  * code below may some day want the tags.
1929                  */
1930                 m_free(mcopy);
1931                 mcopy = NULL;
1932         }
1933         if (mcopy != NULL) {
1934                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1935                     (int)ip->ip_len);
1936                 mcopy->m_pkthdr.len = mcopy->m_len;
1937                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1938         }
1939
1940         if (!ipstealth)
1941                 ip->ip_ttl -= IPTTLDEC;
1942
1943         /*
1944          * If forwarding packet using same interface that it came in on,
1945          * perhaps should send a redirect to sender to shortcut a hop.
1946          * Only send redirect if source is sending directly to us,
1947          * and if packet was not source routed (or has any options).
1948          * Also, don't send redirect if forwarding using a default route
1949          * or a route modified by a redirect.
1950          */
1951         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1952             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1953             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1954             ipsendredirects && !using_srcrt && next_hop == NULL) {
1955                 u_long src = ntohl(ip->ip_src.s_addr);
1956                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1957
1958                 if (rt_ifa != NULL &&
1959                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1960                         if (rt->rt_flags & RTF_GATEWAY)
1961                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1962                         else
1963                                 dest = pkt_dst.s_addr;
1964                         /*
1965                          * Router requirements says to only send
1966                          * host redirects.
1967                          */
1968                         type = ICMP_REDIRECT;
1969                         code = ICMP_REDIRECT_HOST;
1970 #ifdef DIAGNOSTIC
1971                         if (ipprintfs)
1972                                 kprintf("redirect (%d) to %x\n", code, dest);
1973 #endif
1974                 }
1975         }
1976
1977         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1978         if (error == 0) {
1979                 ipstat.ips_forward++;
1980                 if (type == 0) {
1981                         if (mcopy) {
1982                                 ipflow_create(cache_rt, mcopy);
1983                                 m_freem(mcopy);
1984                         }
1985                         return;         /* most common case */
1986                 } else {
1987                         ipstat.ips_redirectsent++;
1988                 }
1989         } else {
1990                 ipstat.ips_cantforward++;
1991         }
1992
1993         if (mcopy == NULL)
1994                 return;
1995
1996         /*
1997          * Send ICMP message.
1998          */
1999
2000         switch (error) {
2001
2002         case 0:                         /* forwarded, but need redirect */
2003                 /* type, code set above */
2004                 break;
2005
2006         case ENETUNREACH:               /* shouldn't happen, checked above */
2007         case EHOSTUNREACH:
2008         case ENETDOWN:
2009         case EHOSTDOWN:
2010         default:
2011                 type = ICMP_UNREACH;
2012                 code = ICMP_UNREACH_HOST;
2013                 break;
2014
2015         case EMSGSIZE:
2016                 type = ICMP_UNREACH;
2017                 code = ICMP_UNREACH_NEEDFRAG;
2018 #ifdef IPSEC
2019                 /*
2020                  * If the packet is routed over IPsec tunnel, tell the
2021                  * originator the tunnel MTU.
2022                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2023                  * XXX quickhack!!!
2024                  */
2025                 if (cache_rt->ro_rt != NULL) {
2026                         struct secpolicy *sp = NULL;
2027                         int ipsecerror;
2028                         int ipsechdr;
2029                         struct route *ro;
2030
2031                         sp = ipsec4_getpolicybyaddr(mcopy,
2032                                                     IPSEC_DIR_OUTBOUND,
2033                                                     IP_FORWARDING,
2034                                                     &ipsecerror);
2035
2036                         if (sp == NULL)
2037                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2038                         else {
2039                                 /* count IPsec header size */
2040                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2041                                                          IPSEC_DIR_OUTBOUND,
2042                                                          NULL);
2043
2044                                 /*
2045                                  * find the correct route for outer IPv4
2046                                  * header, compute tunnel MTU.
2047                                  *
2048                                  */
2049                                 if (sp->req != NULL && sp->req->sav != NULL &&
2050                                     sp->req->sav->sah != NULL) {
2051                                         ro = &sp->req->sav->sah->sa_route;
2052                                         if (ro->ro_rt != NULL &&
2053                                             ro->ro_rt->rt_ifp != NULL) {
2054                                                 destmtu =
2055                                                     ro->ro_rt->rt_ifp->if_mtu;
2056                                                 destmtu -= ipsechdr;
2057                                         }
2058                                 }
2059
2060                                 key_freesp(sp);
2061                         }
2062                 }
2063 #elif FAST_IPSEC
2064                 /*
2065                  * If the packet is routed over IPsec tunnel, tell the
2066                  * originator the tunnel MTU.
2067                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2068                  * XXX quickhack!!!
2069                  */
2070                 if (cache_rt->ro_rt != NULL) {
2071                         struct secpolicy *sp = NULL;
2072                         int ipsecerror;
2073                         int ipsechdr;
2074                         struct route *ro;
2075
2076                         sp = ipsec_getpolicybyaddr(mcopy,
2077                                                    IPSEC_DIR_OUTBOUND,
2078                                                    IP_FORWARDING,
2079                                                    &ipsecerror);
2080
2081                         if (sp == NULL)
2082                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2083                         else {
2084                                 /* count IPsec header size */
2085                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2086                                                          IPSEC_DIR_OUTBOUND,
2087                                                          NULL);
2088
2089                                 /*
2090                                  * find the correct route for outer IPv4
2091                                  * header, compute tunnel MTU.
2092                                  */
2093
2094                                 if (sp->req != NULL &&
2095                                     sp->req->sav != NULL &&
2096                                     sp->req->sav->sah != NULL) {
2097                                         ro = &sp->req->sav->sah->sa_route;
2098                                         if (ro->ro_rt != NULL &&
2099                                             ro->ro_rt->rt_ifp != NULL) {
2100                                                 destmtu =
2101                                                     ro->ro_rt->rt_ifp->if_mtu;
2102                                                 destmtu -= ipsechdr;
2103                                         }
2104                                 }
2105
2106                                 KEY_FREESP(&sp);
2107                         }
2108                 }
2109 #else /* !IPSEC && !FAST_IPSEC */
2110                 if (cache_rt->ro_rt != NULL)
2111                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2112 #endif /*IPSEC*/
2113                 ipstat.ips_cantfrag++;
2114                 break;
2115
2116         case ENOBUFS:
2117                 /*
2118                  * A router should not generate ICMP_SOURCEQUENCH as
2119                  * required in RFC1812 Requirements for IP Version 4 Routers.
2120                  * Source quench could be a big problem under DoS attacks,
2121                  * or if the underlying interface is rate-limited.
2122                  * Those who need source quench packets may re-enable them
2123                  * via the net.inet.ip.sendsourcequench sysctl.
2124                  */
2125                 if (!ip_sendsourcequench) {
2126                         m_freem(mcopy);
2127                         return;
2128                 } else {
2129                         type = ICMP_SOURCEQUENCH;
2130                         code = 0;
2131                 }
2132                 break;
2133
2134         case EACCES:                    /* ipfw denied packet */
2135                 m_freem(mcopy);
2136                 return;
2137         }
2138         icmp_error(mcopy, type, code, dest, destmtu);
2139 }
2140
2141 void
2142 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2143                struct mbuf *m)
2144 {
2145         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2146                 struct timeval tv;
2147
2148                 microtime(&tv);
2149                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2150                     SCM_TIMESTAMP, SOL_SOCKET);
2151                 if (*mp)
2152                         mp = &(*mp)->m_next;
2153         }
2154         if (inp->inp_flags & INP_RECVDSTADDR) {
2155                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2156                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2157                 if (*mp)
2158                         mp = &(*mp)->m_next;
2159         }
2160         if (inp->inp_flags & INP_RECVTTL) {
2161                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2162                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2163                 if (*mp)
2164                         mp = &(*mp)->m_next;
2165         }
2166 #ifdef notyet
2167         /* XXX
2168          * Moving these out of udp_input() made them even more broken
2169          * than they already were.
2170          */
2171         /* options were tossed already */
2172         if (inp->inp_flags & INP_RECVOPTS) {
2173                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2174                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2175                 if (*mp)
2176                         mp = &(*mp)->m_next;
2177         }
2178         /* ip_srcroute doesn't do what we want here, need to fix */
2179         if (inp->inp_flags & INP_RECVRETOPTS) {
2180                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(),
2181                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2182                 if (*mp)
2183                         mp = &(*mp)->m_next;
2184         }
2185 #endif
2186         if (inp->inp_flags & INP_RECVIF) {
2187                 struct ifnet *ifp;
2188                 struct sdlbuf {
2189                         struct sockaddr_dl sdl;
2190                         u_char  pad[32];
2191                 } sdlbuf;
2192                 struct sockaddr_dl *sdp;
2193                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2194
2195                 if (((ifp = m->m_pkthdr.rcvif)) &&
2196                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2197                         sdp = IF_LLSOCKADDR(ifp);
2198                         /*
2199                          * Change our mind and don't try copy.
2200                          */
2201                         if ((sdp->sdl_family != AF_LINK) ||
2202                             (sdp->sdl_len > sizeof(sdlbuf))) {
2203                                 goto makedummy;
2204                         }
2205                         bcopy(sdp, sdl2, sdp->sdl_len);
2206                 } else {
2207 makedummy:
2208                         sdl2->sdl_len =
2209                             offsetof(struct sockaddr_dl, sdl_data[0]);
2210                         sdl2->sdl_family = AF_LINK;
2211                         sdl2->sdl_index = 0;
2212                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2213                 }
2214                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2215                         IP_RECVIF, IPPROTO_IP);
2216                 if (*mp)
2217                         mp = &(*mp)->m_next;
2218         }
2219 }
2220
2221 /*
2222  * XXX these routines are called from the upper part of the kernel.
2223  *
2224  * They could also be moved to ip_mroute.c, since all the RSVP
2225  *  handling is done there already.
2226  */
2227 int
2228 ip_rsvp_init(struct socket *so)
2229 {
2230         if (so->so_type != SOCK_RAW ||
2231             so->so_proto->pr_protocol != IPPROTO_RSVP)
2232                 return EOPNOTSUPP;
2233
2234         if (ip_rsvpd != NULL)
2235                 return EADDRINUSE;
2236
2237         ip_rsvpd = so;
2238         /*
2239          * This may seem silly, but we need to be sure we don't over-increment
2240          * the RSVP counter, in case something slips up.
2241          */
2242         if (!ip_rsvp_on) {
2243                 ip_rsvp_on = 1;
2244                 rsvp_on++;
2245         }
2246
2247         return 0;
2248 }
2249
2250 int
2251 ip_rsvp_done(void)
2252 {
2253         ip_rsvpd = NULL;
2254         /*
2255          * This may seem silly, but we need to be sure we don't over-decrement
2256          * the RSVP counter, in case something slips up.
2257          */
2258         if (ip_rsvp_on) {
2259                 ip_rsvp_on = 0;
2260                 rsvp_on--;
2261         }
2262         return 0;
2263 }
2264
2265 void
2266 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2267 {
2268         int off, proto;
2269         __va_list ap;
2270
2271         __va_start(ap, m);
2272         off = __va_arg(ap, int);
2273         proto = __va_arg(ap, int);
2274         __va_end(ap);
2275
2276         if (rsvp_input_p) { /* call the real one if loaded */
2277                 rsvp_input_p(m, off, proto);
2278                 return;
2279         }
2280
2281         /* Can still get packets with rsvp_on = 0 if there is a local member
2282          * of the group to which the RSVP packet is addressed.  But in this
2283          * case we want to throw the packet away.
2284          */
2285
2286         if (!rsvp_on) {
2287                 m_freem(m);
2288                 return;
2289         }
2290
2291         if (ip_rsvpd != NULL) {
2292                 rip_input(m, off, proto);
2293                 return;
2294         }
2295         /* Drop the packet */
2296         m_freem(m);
2297 }