Another round of typo fixes (mostly in messages).
[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 1.79 2008/04/20 13:44:25 swildner 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_ifaddrhead;            /* first inet address */
209 struct  in_ifaddrhashhead *in_ifaddrhashtbl;    /* inet addr hash table */
210 u_long  in_ifaddrhmask;                         /* mask for hash table */
211
212 struct ip_stats ipstats_percpu[MAXCPU];
213 #ifdef SMP
214 static int
215 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
216 {
217         int cpu, error = 0;
218
219         for (cpu = 0; cpu < ncpus; ++cpu) {
220                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
221                                         sizeof(struct ip_stats))))
222                         break;
223                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
224                                        sizeof(struct ip_stats))))
225                         break;
226         }
227
228         return (error);
229 }
230 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
231     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
232 #else
233 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
234     &ipstat, ip_stats, "IP statistics");
235 #endif
236
237 /* Packet reassembly stuff */
238 #define IPREASS_NHASH_LOG2      6
239 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
240 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
241 #define IPREASS_HASH(x,y)                                               \
242     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
243
244 static struct ipq ipq[IPREASS_NHASH];
245
246 #ifdef IPCTL_DEFMTU
247 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
248     &ip_mtu, 0, "Default MTU");
249 #endif
250
251 #ifdef IPSTEALTH
252 static int ipstealth = 0;
253 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
254 #else
255 static const int ipstealth = 0;
256 #endif
257
258
259 /* Firewall hooks */
260 ip_fw_chk_t *ip_fw_chk_ptr;
261 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
262 int fw_enable = 1;
263 int fw_one_pass = 1;
264
265 struct pfil_head inet_pfil_hook;
266
267 /*
268  * XXX this is ugly -- the following two global variables are
269  * used to store packet state while it travels through the stack.
270  * Note that the code even makes assumptions on the size and
271  * alignment of fields inside struct ip_srcrt so e.g. adding some
272  * fields will break the code. This needs to be fixed.
273  *
274  * We need to save the IP options in case a protocol wants to respond
275  * to an incoming packet over the same route if the packet got here
276  * using IP source routing.  This allows connection establishment and
277  * maintenance when the remote end is on a network that is not known
278  * to us.
279  */
280 static int ip_nhops = 0;
281
282 static  struct ip_srcrt {
283         struct  in_addr dst;                    /* final destination */
284         char    nop;                            /* one NOP to align */
285         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
286         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
287 } ip_srcrt;
288
289 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
290 static struct malloc_pipe ipq_mpipe;
291
292 static void             save_rte (u_char *, struct in_addr);
293 static int              ip_dooptions (struct mbuf *m, int,
294                                         struct sockaddr_in *next_hop);
295 static void             ip_forward (struct mbuf *m, boolean_t using_srcrt,
296                                         struct sockaddr_in *next_hop);
297 static void             ip_freef (struct ipq *);
298 static void             ip_input_handler (struct netmsg *);
299 static struct mbuf      *ip_reass (struct mbuf *, struct ipq *,
300                                         struct ipq *, u_int32_t *);
301
302 /*
303  * IP initialization: fill in IP protocol switch table.
304  * All protocols not implemented in kernel go to raw IP protocol handler.
305  */
306 void
307 ip_init(void)
308 {
309         struct protosw *pr;
310         int i;
311 #ifdef SMP
312         int cpu;
313 #endif
314
315         /*
316          * Make sure we can handle a reasonable number of fragments but
317          * cap it at 4000 (XXX).
318          */
319         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
320                     IFQ_MAXLEN, 4000, 0, NULL);
321         TAILQ_INIT(&in_ifaddrhead);
322         in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
323         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
324         if (pr == NULL)
325                 panic("ip_init");
326         for (i = 0; i < IPPROTO_MAX; i++)
327                 ip_protox[i] = pr - inetsw;
328         for (pr = inetdomain.dom_protosw;
329              pr < inetdomain.dom_protoswNPROTOSW; pr++)
330                 if (pr->pr_domain->dom_family == PF_INET &&
331                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
332                         ip_protox[pr->pr_protocol] = pr - inetsw;
333
334         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
335         inet_pfil_hook.ph_af = AF_INET;
336         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
337                 kprintf("%s: WARNING: unable to register pfil hook, "
338                         "error %d\n", __func__, i);
339         }
340
341         for (i = 0; i < IPREASS_NHASH; i++)
342             ipq[i].next = ipq[i].prev = &ipq[i];
343
344         maxnipq = nmbclusters / 32;
345         maxfragsperpacket = 16;
346
347         ip_id = time_second & 0xffff;
348
349         /*
350          * Initialize IP statistics counters for each CPU.
351          *
352          */
353 #ifdef SMP
354         for (cpu = 0; cpu < ncpus; ++cpu) {
355                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
356         }
357 #else
358         bzero(&ipstat, sizeof(struct ip_stats));
359 #endif
360
361         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
362 }
363
364 /*
365  * XXX watch out this one. It is perhaps used as a cache for
366  * the most recently used route ? it is cleared in in_addroute()
367  * when a new route is successfully created.
368  */
369 struct route ipforward_rt[MAXCPU];
370
371 /* Do transport protocol processing. */
372 static void
373 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip,
374                            struct sockaddr_in *nexthop)
375 {
376         /*
377          * Switch out to protocol's input routine.
378          */
379         if (nexthop && ip->ip_p == IPPROTO_TCP) {
380                 /* TCP needs IPFORWARD info if available */
381                 struct m_hdr tag;
382
383                 tag.mh_type = MT_TAG;
384                 tag.mh_flags = PACKET_TAG_IPFORWARD;
385                 tag.mh_data = (caddr_t)nexthop;
386                 tag.mh_next = m;
387
388                 (*inetsw[ip_protox[ip->ip_p]].pr_input)
389                     ((struct mbuf *)&tag, hlen, ip->ip_p);
390         } else {
391                 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
392         }
393 }
394
395 struct netmsg_transport_packet {
396         struct netmsg           nm_netmsg;
397         struct mbuf             *nm_mbuf;
398         int                     nm_hlen;
399         boolean_t               nm_hasnexthop;
400         struct sockaddr_in      nm_nexthop;
401 };
402
403 static void
404 transport_processing_handler(netmsg_t netmsg)
405 {
406         struct netmsg_transport_packet *msg = (void *)netmsg;
407         struct sockaddr_in *nexthop;
408         struct ip *ip;
409
410         ip = mtod(msg->nm_mbuf, struct ip *);
411         nexthop = msg->nm_hasnexthop ? &msg->nm_nexthop : NULL;
412         transport_processing_oncpu(msg->nm_mbuf, msg->nm_hlen, ip, nexthop);
413         lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
414 }
415
416 static void
417 ip_input_handler(struct netmsg *msg0)
418 {
419         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
420
421         ip_input(m);
422         /* msg0 was embedded in the mbuf, do not reply! */
423 }
424
425 /*
426  * IP input routine.  Checksum and byte swap header.  If fragmented
427  * try to reassemble.  Process options.  Pass to next level.
428  */
429 void
430 ip_input(struct mbuf *m)
431 {
432         struct ip *ip;
433         struct ipq *fp;
434         struct in_ifaddr *ia = NULL;
435         int i, hlen, checkif;
436         u_short sum;
437         struct in_addr pkt_dst;
438         u_int32_t divert_info = 0;              /* packet divert/tee info */
439         struct ip_fw_args args;
440         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
441         boolean_t needredispatch = FALSE;
442         struct in_addr odst;                    /* original dst address(NAT) */
443         struct m_tag *mtag;
444 #ifdef FAST_IPSEC
445         struct tdb_ident *tdbi;
446         struct secpolicy *sp;
447         int error;
448 #endif
449
450         args.eh = NULL;
451         args.oif = NULL;
452         args.rule = NULL;
453         args.next_hop = NULL;
454
455         /* Grab info from MT_TAG mbufs prepended to the chain. */
456         while (m != NULL && m->m_type == MT_TAG) {
457                 switch(m->_m_tag_id) {
458                 case PACKET_TAG_IPFORWARD:
459                         args.next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
460                         break;
461                 default:
462                         kprintf("ip_input: unrecognised MT_TAG tag %d\n",
463                             m->_m_tag_id);
464                         break;
465                 }
466                 m = m->m_next;
467         }
468         M_ASSERTPKTHDR(m);
469
470         /* Extract info from dummynet tag */
471         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
472         if (mtag != NULL) {
473                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
474
475                 m_tag_delete(m, mtag);
476                 mtag = NULL;
477         }
478
479         if (args.rule != NULL) {        /* dummynet already filtered us */
480                 ip = mtod(m, struct ip *);
481                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
482                 goto iphack;
483         }
484
485         ipstat.ips_total++;
486
487         /* length checks already done in ip_demux() */
488         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
489
490         ip = mtod(m, struct ip *);
491
492         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
493                 ipstat.ips_badvers++;
494                 goto bad;
495         }
496
497         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
498         /* length checks already done in ip_demux() */
499         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
500         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
501
502         /* 127/8 must not appear on wire - RFC1122 */
503         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
504             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
505                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
506                         ipstat.ips_badaddr++;
507                         goto bad;
508                 }
509         }
510
511         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
512                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
513         } else {
514                 if (hlen == sizeof(struct ip)) {
515                         sum = in_cksum_hdr(ip);
516                 } else {
517                         sum = in_cksum(m, hlen);
518                 }
519         }
520         if (sum != 0) {
521                 ipstat.ips_badsum++;
522                 goto bad;
523         }
524
525 #ifdef ALTQ
526         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
527                 /* packet is dropped by traffic conditioner */
528                 return;
529         }
530 #endif
531         /*
532          * Convert fields to host representation.
533          */
534         ip->ip_len = ntohs(ip->ip_len);
535         if (ip->ip_len < hlen) {
536                 ipstat.ips_badlen++;
537                 goto bad;
538         }
539         ip->ip_off = ntohs(ip->ip_off);
540
541         /*
542          * Check that the amount of data in the buffers
543          * is as at least much as the IP header would have us expect.
544          * Trim mbufs if longer than we expect.
545          * Drop packet if shorter than we expect.
546          */
547         if (m->m_pkthdr.len < ip->ip_len) {
548                 ipstat.ips_tooshort++;
549                 goto bad;
550         }
551         if (m->m_pkthdr.len > ip->ip_len) {
552                 if (m->m_len == m->m_pkthdr.len) {
553                         m->m_len = ip->ip_len;
554                         m->m_pkthdr.len = ip->ip_len;
555                 } else
556                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
557         }
558 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
559         /*
560          * Bypass packet filtering for packets from a tunnel (gif).
561          */
562         if (ipsec_gethist(m, NULL))
563                 goto pass;
564 #endif
565
566         /*
567          * IpHack's section.
568          * Right now when no processing on packet has done
569          * and it is still fresh out of network we do our black
570          * deals with it.
571          * - Firewall: deny/allow/divert
572          * - Xlate: translate packet's addr/port (NAT).
573          * - Pipe: pass pkt through dummynet.
574          * - Wrap: fake packet's addr/port <unimpl.>
575          * - Encapsulate: put it in another IP and send out. <unimp.>
576          */
577
578 iphack:
579
580         /*
581          * Run through list of hooks for input packets.
582          *
583          * NB: Beware of the destination address changing (e.g.
584          *     by NAT rewriting). When this happens, tell
585          *     ip_forward to do the right thing.
586          */
587         if (pfil_has_hooks(&inet_pfil_hook)) {
588                 odst = ip->ip_dst;
589                 if (pfil_run_hooks(&inet_pfil_hook, &m,
590                     m->m_pkthdr.rcvif, PFIL_IN)) {
591                         return;
592                 }
593                 if (m == NULL)                  /* consumed by filter */
594                         return;
595                 ip = mtod(m, struct ip *);
596                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
597         }
598
599         if (fw_enable && IPFW_LOADED) {
600                 /*
601                  * If we've been forwarded from the output side, then
602                  * skip the firewall a second time
603                  */
604                 if (args.next_hop != NULL)
605                         goto ours;
606
607                 args.m = m;
608                 i = ip_fw_chk_ptr(&args);
609                 m = args.m;
610
611                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
612                         if (m != NULL)
613                                 m_freem(m);
614                         return;
615                 }
616                 ip = mtod(m, struct ip *);      /* just in case m changed */
617                 if (i == 0 && args.next_hop == NULL)    /* common case */
618                         goto pass;
619                 if (i & IP_FW_PORT_DYNT_FLAG) {
620                         /* Send packet to the appropriate pipe */
621                         ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
622                         return;
623                 }
624 #ifdef IPDIVERT
625                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
626                         /* Divert or tee packet */
627                         divert_info = i;
628                         goto ours;
629                 }
630 #endif
631                 if (i == 0 && args.next_hop != NULL)
632                         goto pass;
633                 /*
634                  * if we get here, the packet must be dropped
635                  */
636                 m_freem(m);
637                 return;
638         }
639 pass:
640
641         /*
642          * Process options and, if not destined for us,
643          * ship it on.  ip_dooptions returns 1 when an
644          * error was detected (causing an icmp message
645          * to be sent and the original packet to be freed).
646          */
647         ip_nhops = 0;           /* for source routed packets */
648         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, args.next_hop))
649                 return;
650
651         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
652          * matter if it is destined to another node, or whether it is
653          * a multicast one, RSVP wants it! and prevents it from being forwarded
654          * anywhere else. Also checks if the rsvp daemon is running before
655          * grabbing the packet.
656          */
657         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
658                 goto ours;
659
660         /*
661          * Check our list of addresses, to see if the packet is for us.
662          * If we don't have any addresses, assume any unicast packet
663          * we receive might be for us (and let the upper layers deal
664          * with it).
665          */
666         if (TAILQ_EMPTY(&in_ifaddrhead) && !(m->m_flags & (M_MCAST | M_BCAST)))
667                 goto ours;
668
669         /*
670          * Cache the destination address of the packet; this may be
671          * changed by use of 'ipfw fwd'.
672          */
673         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
674
675         /*
676          * Enable a consistency check between the destination address
677          * and the arrival interface for a unicast packet (the RFC 1122
678          * strong ES model) if IP forwarding is disabled and the packet
679          * is not locally generated and the packet is not subject to
680          * 'ipfw fwd'.
681          *
682          * XXX - Checking also should be disabled if the destination
683          * address is ipnat'ed to a different interface.
684          *
685          * XXX - Checking is incompatible with IP aliases added
686          * to the loopback interface instead of the interface where
687          * the packets are received.
688          */
689         checkif = ip_checkinterface &&
690                   !ipforwarding &&
691                   m->m_pkthdr.rcvif != NULL &&
692                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
693                   (args.next_hop == NULL);
694
695         /*
696          * Check for exact addresses in the hash bucket.
697          */
698         LIST_FOREACH(ia, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
699                 /*
700                  * If the address matches, verify that the packet
701                  * arrived via the correct interface if checking is
702                  * enabled.
703                  */
704                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
705                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
706                         goto ours;
707         }
708         /*
709          * Check for broadcast addresses.
710          *
711          * Only accept broadcast packets that arrive via the matching
712          * interface.  Reception of forwarded directed broadcasts would
713          * be handled via ip_forward() and ether_output() with the loopback
714          * into the stack for SIMPLEX interfaces handled by ether_output().
715          */
716         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
717                 struct ifaddr_container *ifac;
718
719                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
720                               ifa_link) {
721                         struct ifaddr *ifa = ifac->ifa;
722
723                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
724                                 continue;
725                         if (ifa->ifa_addr->sa_family != AF_INET)
726                                 continue;
727                         ia = ifatoia(ifa);
728                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
729                                                                 pkt_dst.s_addr)
730                                 goto ours;
731                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
732                                 goto ours;
733 #ifdef BOOTP_COMPAT
734                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
735                                 goto ours;
736 #endif
737                 }
738         }
739         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
740                 struct in_multi *inm;
741
742                 if (ip_mrouter != NULL) {
743                         /*
744                          * If we are acting as a multicast router, all
745                          * incoming multicast packets are passed to the
746                          * kernel-level multicast forwarding function.
747                          * The packet is returned (relatively) intact; if
748                          * ip_mforward() returns a non-zero value, the packet
749                          * must be discarded, else it may be accepted below.
750                          */
751                         if (ip_mforward != NULL &&
752                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
753                                 ipstat.ips_cantforward++;
754                                 m_freem(m);
755                                 return;
756                         }
757
758                         /*
759                          * The process-level routing daemon needs to receive
760                          * all multicast IGMP packets, whether or not this
761                          * host belongs to their destination groups.
762                          */
763                         if (ip->ip_p == IPPROTO_IGMP)
764                                 goto ours;
765                         ipstat.ips_forward++;
766                 }
767                 /*
768                  * See if we belong to the destination multicast group on the
769                  * arrival interface.
770                  */
771                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
772                 if (inm == NULL) {
773                         ipstat.ips_notmember++;
774                         m_freem(m);
775                         return;
776                 }
777                 goto ours;
778         }
779         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
780                 goto ours;
781         if (ip->ip_dst.s_addr == INADDR_ANY)
782                 goto ours;
783
784         /*
785          * FAITH(Firewall Aided Internet Translator)
786          */
787         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
788                 if (ip_keepfaith) {
789                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
790                                 goto ours;
791                 }
792                 m_freem(m);
793                 return;
794         }
795
796         /*
797          * Not for us; forward if possible and desirable.
798          */
799         if (!ipforwarding) {
800                 ipstat.ips_cantforward++;
801                 m_freem(m);
802         } else {
803 #ifdef IPSEC
804                 /*
805                  * Enforce inbound IPsec SPD.
806                  */
807                 if (ipsec4_in_reject(m, NULL)) {
808                         ipsecstat.in_polvio++;
809                         goto bad;
810                 }
811 #endif
812 #ifdef FAST_IPSEC
813                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
814                 crit_enter();
815                 if (mtag != NULL) {
816                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
817                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
818                 } else {
819                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
820                                                    IP_FORWARDING, &error);
821                 }
822                 if (sp == NULL) {       /* NB: can happen if error */
823                         crit_exit();
824                         /*XXX error stat???*/
825                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
826                         goto bad;
827                 }
828
829                 /*
830                  * Check security policy against packet attributes.
831                  */
832                 error = ipsec_in_reject(sp, m);
833                 KEY_FREESP(&sp);
834                 crit_exit();
835                 if (error) {
836                         ipstat.ips_cantforward++;
837                         goto bad;
838                 }
839 #endif
840                 ip_forward(m, using_srcrt, args.next_hop);
841         }
842         return;
843
844 ours:
845
846         /*
847          * IPSTEALTH: Process non-routing options only
848          * if the packet is destined for us.
849          */
850         if (ipstealth &&
851             hlen > sizeof(struct ip) &&
852             ip_dooptions(m, 1, args.next_hop))
853                 return;
854
855         /* Count the packet in the ip address stats */
856         if (ia != NULL) {
857                 ia->ia_ifa.if_ipackets++;
858                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
859         }
860
861         /*
862          * If offset or IP_MF are set, must reassemble.
863          * Otherwise, nothing need be done.
864          * (We could look in the reassembly queue to see
865          * if the packet was previously fragmented,
866          * but it's not worth the time; just let them time out.)
867          */
868         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
869
870                 /* If maxnipq is 0, never accept fragments. */
871                 if (maxnipq == 0) {
872                         ipstat.ips_fragments++;
873                         ipstat.ips_fragdropped++;
874                         goto bad;
875                 }
876
877                 sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
878                 /*
879                  * Look for queue of fragments
880                  * of this datagram.
881                  */
882                 for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
883                         if (ip->ip_id == fp->ipq_id &&
884                             ip->ip_src.s_addr == fp->ipq_src.s_addr &&
885                             ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
886                             ip->ip_p == fp->ipq_p)
887                                 goto found;
888
889                 fp = NULL;
890
891                 /*
892                  * Enforce upper bound on number of fragmented packets
893                  * for which we attempt reassembly;
894                  * If maxnipq is -1, accept all fragments without limitation.
895                  */
896                 if ((nipq > maxnipq) && (maxnipq > 0)) {
897                         /*
898                          * drop something from the tail of the current queue
899                          * before proceeding further
900                          */
901                         if (ipq[sum].prev == &ipq[sum]) {   /* gak */
902                                 for (i = 0; i < IPREASS_NHASH; i++) {
903                                         if (ipq[i].prev != &ipq[i]) {
904                                                 ipstat.ips_fragtimeout +=
905                                                     ipq[i].prev->ipq_nfrags;
906                                                 ip_freef(ipq[i].prev);
907                                                 break;
908                                         }
909                                 }
910                         } else {
911                                 ipstat.ips_fragtimeout +=
912                                     ipq[sum].prev->ipq_nfrags;
913                                 ip_freef(ipq[sum].prev);
914                         }
915                 }
916 found:
917                 /*
918                  * Adjust ip_len to not reflect header,
919                  * convert offset of this to bytes.
920                  */
921                 ip->ip_len -= hlen;
922                 if (ip->ip_off & IP_MF) {
923                         /*
924                          * Make sure that fragments have a data length
925                          * that's a non-zero multiple of 8 bytes.
926                          */
927                         if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
928                                 ipstat.ips_toosmall++; /* XXX */
929                                 goto bad;
930                         }
931                         m->m_flags |= M_FRAG;
932                 } else
933                         m->m_flags &= ~M_FRAG;
934                 ip->ip_off <<= 3;
935
936                 /*
937                  * Attempt reassembly; if it succeeds, proceed.
938                  * ip_reass() will return a different mbuf, and update
939                  * the divert info in divert_info.
940                  */
941                 ipstat.ips_fragments++;
942                 m->m_pkthdr.header = ip;
943                 m = ip_reass(m, fp, &ipq[sum], &divert_info);
944                 if (m == NULL)
945                         return;
946                 ipstat.ips_reassembled++;
947                 needredispatch = TRUE;
948                 ip = mtod(m, struct ip *);
949                 /* Get the header length of the reassembled packet */
950                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
951 #ifdef IPDIVERT
952                 /* Restore original checksum before diverting packet */
953                 if (divert_info != 0) {
954                         ip->ip_len += hlen;
955                         ip->ip_len = htons(ip->ip_len);
956                         ip->ip_off = htons(ip->ip_off);
957                         ip->ip_sum = 0;
958                         if (hlen == sizeof(struct ip))
959                                 ip->ip_sum = in_cksum_hdr(ip);
960                         else
961                                 ip->ip_sum = in_cksum(m, hlen);
962                         ip->ip_off = ntohs(ip->ip_off);
963                         ip->ip_len = ntohs(ip->ip_len);
964                         ip->ip_len -= hlen;
965                 }
966 #endif
967         } else {
968                 ip->ip_len -= hlen;
969         }
970
971 #ifdef IPDIVERT
972         /*
973          * Divert or tee packet to the divert protocol if required.
974          */
975         if (divert_info != 0) {
976                 struct mbuf *clone = NULL;
977
978                 /* Clone packet if we're doing a 'tee' */
979                 if ((divert_info & IP_FW_PORT_TEE_FLAG) != 0)
980                         clone = m_dup(m, MB_DONTWAIT);
981
982                 /* Restore packet header fields to original values */
983                 ip->ip_len += hlen;
984                 ip->ip_len = htons(ip->ip_len);
985                 ip->ip_off = htons(ip->ip_off);
986
987                 /* Deliver packet to divert input routine */
988                 divert_packet(m, 1, divert_info & 0xffff);
989                 ipstat.ips_delivered++;
990
991                 /* If 'tee', continue with original packet */
992                 if (clone == NULL)
993                         return;
994                 m = clone;
995                 ip = mtod(m, struct ip *);
996                 ip->ip_len += hlen;
997                 /*
998                  * Jump backwards to complete processing of the
999                  * packet. But first clear divert_info to avoid
1000                  * entering this block again.
1001                  * We do not need to clear args.divert_rule
1002                  * or args.next_hop as they will not be used.
1003                  *
1004                  * XXX Better safe than sorry, remove the DIVERT tag.
1005                  */
1006                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1007                 if (mtag != NULL)
1008                         m_tag_delete(m, mtag);
1009                 
1010                 divert_info = 0;
1011                 goto pass;
1012         }
1013 #endif
1014
1015 #ifdef IPSEC
1016         /*
1017          * enforce IPsec policy checking if we are seeing last header.
1018          * note that we do not visit this with protocols with pcb layer
1019          * code - like udp/tcp/raw ip.
1020          */
1021         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
1022             ipsec4_in_reject(m, NULL)) {
1023                 ipsecstat.in_polvio++;
1024                 goto bad;
1025         }
1026 #endif
1027 #if FAST_IPSEC
1028         /*
1029          * enforce IPsec policy checking if we are seeing last header.
1030          * note that we do not visit this with protocols with pcb layer
1031          * code - like udp/tcp/raw ip.
1032          */
1033         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1034                 /*
1035                  * Check if the packet has already had IPsec processing
1036                  * done.  If so, then just pass it along.  This tag gets
1037                  * set during AH, ESP, etc. input handling, before the
1038                  * packet is returned to the ip input queue for delivery.
1039                  */
1040                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1041                 crit_enter();
1042                 if (mtag != NULL) {
1043                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
1044                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1045                 } else {
1046                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1047                                                    IP_FORWARDING, &error);
1048                 }
1049                 if (sp != NULL) {
1050                         /*
1051                          * Check security policy against packet attributes.
1052                          */
1053                         error = ipsec_in_reject(sp, m);
1054                         KEY_FREESP(&sp);
1055                 } else {
1056                         /* XXX error stat??? */
1057                         error = EINVAL;
1058 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1059                         goto bad;
1060                 }
1061                 crit_exit();
1062                 if (error)
1063                         goto bad;
1064         }
1065 #endif /* FAST_IPSEC */
1066
1067         ipstat.ips_delivered++;
1068         if (needredispatch) {
1069                 struct netmsg_transport_packet *msg;
1070                 lwkt_port_t port;
1071
1072                 ip->ip_off = htons(ip->ip_off);
1073                 ip->ip_len = htons(ip->ip_len);
1074                 port = ip_mport_in(&m);
1075                 if (port == NULL)
1076                         return;
1077
1078                 msg = kmalloc(sizeof(struct netmsg_transport_packet), M_LWKTMSG,
1079                              M_INTWAIT | M_NULLOK);
1080                 if (msg == NULL)
1081                         goto bad;
1082
1083                 netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0,
1084                             transport_processing_handler);
1085                 msg->nm_hlen = hlen;
1086                 msg->nm_hasnexthop = (args.next_hop != NULL);
1087                 if (msg->nm_hasnexthop)
1088                         msg->nm_nexthop = *args.next_hop;  /* structure copy */
1089
1090                 msg->nm_mbuf = m;
1091                 ip = mtod(m, struct ip *);
1092                 ip->ip_len = ntohs(ip->ip_len);
1093                 ip->ip_off = ntohs(ip->ip_off);
1094                 lwkt_sendmsg(port, &msg->nm_netmsg.nm_lmsg);
1095         } else {
1096                 transport_processing_oncpu(m, hlen, ip, args.next_hop);
1097         }
1098         return;
1099
1100 bad:
1101         m_freem(m);
1102 }
1103
1104 /*
1105  * Take incoming datagram fragment and try to reassemble it into
1106  * whole datagram.  If a chain for reassembly of this datagram already
1107  * exists, then it is given as fp; otherwise have to make a chain.
1108  *
1109  * When IPDIVERT enabled, keep additional state with each packet that
1110  * tells us if we need to divert or tee the packet we're building.
1111  * In particular, *divinfo includes the port and TEE flag.
1112  */
1113
1114 static struct mbuf *
1115 ip_reass(struct mbuf *m, struct ipq *fp, struct ipq *where,
1116          u_int32_t *divinfo)
1117 {
1118         struct ip *ip = mtod(m, struct ip *);
1119         struct mbuf *p = NULL, *q, *nq;
1120         struct mbuf *n;
1121         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1122         int i, next;
1123 #ifdef IPDIVERT
1124         struct m_tag *mtag;
1125 #endif
1126
1127         /*
1128          * If the hardware has not done csum over this fragment
1129          * then csum_data is not valid at all.
1130          */
1131         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1132             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1133                 m->m_pkthdr.csum_data = 0;
1134                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1135         }
1136
1137         /*
1138          * Presence of header sizes in mbufs
1139          * would confuse code below.
1140          */
1141         m->m_data += hlen;
1142         m->m_len -= hlen;
1143
1144         /*
1145          * If first fragment to arrive, create a reassembly queue.
1146          */
1147         if (fp == NULL) {
1148                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1149                         goto dropfrag;
1150                 insque(fp, where);
1151                 nipq++;
1152                 fp->ipq_nfrags = 1;
1153                 fp->ipq_ttl = IPFRAGTTL;
1154                 fp->ipq_p = ip->ip_p;
1155                 fp->ipq_id = ip->ip_id;
1156                 fp->ipq_src = ip->ip_src;
1157                 fp->ipq_dst = ip->ip_dst;
1158                 fp->ipq_frags = m;
1159                 m->m_nextpkt = NULL;
1160 #ifdef IPDIVERT
1161                 fp->ipq_div_info = 0;
1162 #endif
1163                 goto inserted;
1164         } else {
1165                 fp->ipq_nfrags++;
1166         }
1167
1168 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1169
1170         /*
1171          * Find a segment which begins after this one does.
1172          */
1173         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1174                 if (GETIP(q)->ip_off > ip->ip_off)
1175                         break;
1176
1177         /*
1178          * If there is a preceding segment, it may provide some of
1179          * our data already.  If so, drop the data from the incoming
1180          * segment.  If it provides all of our data, drop us, otherwise
1181          * stick new segment in the proper place.
1182          *
1183          * If some of the data is dropped from the the preceding
1184          * segment, then it's checksum is invalidated.
1185          */
1186         if (p) {
1187                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1188                 if (i > 0) {
1189                         if (i >= ip->ip_len)
1190                                 goto dropfrag;
1191                         m_adj(m, i);
1192                         m->m_pkthdr.csum_flags = 0;
1193                         ip->ip_off += i;
1194                         ip->ip_len -= i;
1195                 }
1196                 m->m_nextpkt = p->m_nextpkt;
1197                 p->m_nextpkt = m;
1198         } else {
1199                 m->m_nextpkt = fp->ipq_frags;
1200                 fp->ipq_frags = m;
1201         }
1202
1203         /*
1204          * While we overlap succeeding segments trim them or,
1205          * if they are completely covered, dequeue them.
1206          */
1207         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1208              q = nq) {
1209                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1210                 if (i < GETIP(q)->ip_len) {
1211                         GETIP(q)->ip_len -= i;
1212                         GETIP(q)->ip_off += i;
1213                         m_adj(q, i);
1214                         q->m_pkthdr.csum_flags = 0;
1215                         break;
1216                 }
1217                 nq = q->m_nextpkt;
1218                 m->m_nextpkt = nq;
1219                 ipstat.ips_fragdropped++;
1220                 fp->ipq_nfrags--;
1221                 q->m_nextpkt = NULL;
1222                 m_freem(q);
1223         }
1224
1225 inserted:
1226
1227 #ifdef IPDIVERT
1228         /*
1229          * Transfer firewall instructions to the fragment structure.
1230          * Only trust info in the fragment at offset 0.
1231          */
1232         if (ip->ip_off == 0) {
1233                 fp->ipq_div_info = *divinfo;
1234         } else {
1235                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1236                 if (mtag != NULL)
1237                         m_tag_delete(m, mtag);
1238         }
1239         *divinfo = 0;
1240 #endif
1241
1242         /*
1243          * Check for complete reassembly and perform frag per packet
1244          * limiting.
1245          *
1246          * Frag limiting is performed here so that the nth frag has
1247          * a chance to complete the packet before we drop the packet.
1248          * As a result, n+1 frags are actually allowed per packet, but
1249          * only n will ever be stored. (n = maxfragsperpacket.)
1250          *
1251          */
1252         next = 0;
1253         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1254                 if (GETIP(q)->ip_off != next) {
1255                         if (fp->ipq_nfrags > maxfragsperpacket) {
1256                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1257                                 ip_freef(fp);
1258                         }
1259                         return (NULL);
1260                 }
1261                 next += GETIP(q)->ip_len;
1262         }
1263         /* Make sure the last packet didn't have the IP_MF flag */
1264         if (p->m_flags & M_FRAG) {
1265                 if (fp->ipq_nfrags > maxfragsperpacket) {
1266                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1267                         ip_freef(fp);
1268                 }
1269                 return (NULL);
1270         }
1271
1272         /*
1273          * Reassembly is complete.  Make sure the packet is a sane size.
1274          */
1275         q = fp->ipq_frags;
1276         ip = GETIP(q);
1277         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1278                 ipstat.ips_toolong++;
1279                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1280                 ip_freef(fp);
1281                 return (NULL);
1282         }
1283
1284         /*
1285          * Concatenate fragments.
1286          */
1287         m = q;
1288         n = m->m_next;
1289         m->m_next = NULL;
1290         m_cat(m, n);
1291         nq = q->m_nextpkt;
1292         q->m_nextpkt = NULL;
1293         for (q = nq; q != NULL; q = nq) {
1294                 nq = q->m_nextpkt;
1295                 q->m_nextpkt = NULL;
1296                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1297                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1298                 m_cat(m, q);
1299         }
1300
1301         /*
1302          * Clean up the 1's complement checksum.  Carry over 16 bits must
1303          * be added back.  This assumes no more then 65535 packet fragments
1304          * were reassembled.  A second carry can also occur (but not a third).
1305          */
1306         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1307                                 (m->m_pkthdr.csum_data >> 16);
1308         if (m->m_pkthdr.csum_data > 0xFFFF)
1309                 m->m_pkthdr.csum_data -= 0xFFFF;
1310
1311
1312 #ifdef IPDIVERT
1313         /*
1314          * Extract firewall instructions from the fragment structure.
1315          */
1316         *divinfo = fp->ipq_div_info;
1317 #endif
1318
1319         /*
1320          * Create header for new ip packet by
1321          * modifying header of first packet;
1322          * dequeue and discard fragment reassembly header.
1323          * Make header visible.
1324          */
1325         ip->ip_len = next;
1326         ip->ip_src = fp->ipq_src;
1327         ip->ip_dst = fp->ipq_dst;
1328         remque(fp);
1329         nipq--;
1330         mpipe_free(&ipq_mpipe, fp);
1331         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1332         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1333         /* some debugging cruft by sklower, below, will go away soon */
1334         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1335                 int plen = 0;
1336
1337                 for (n = m; n; n = n->m_next)
1338                         plen += n->m_len;
1339                 m->m_pkthdr.len = plen;
1340         }
1341         return (m);
1342
1343 dropfrag:
1344 #ifdef IPDIVERT
1345         *divinfo = 0;
1346 #endif
1347         ipstat.ips_fragdropped++;
1348         if (fp != NULL)
1349                 fp->ipq_nfrags--;
1350         m_freem(m);
1351         return (NULL);
1352
1353 #undef GETIP
1354 }
1355
1356 /*
1357  * Free a fragment reassembly header and all
1358  * associated datagrams.
1359  */
1360 static void
1361 ip_freef(struct ipq *fp)
1362 {
1363         struct mbuf *q;
1364
1365         while (fp->ipq_frags) {
1366                 q = fp->ipq_frags;
1367                 fp->ipq_frags = q->m_nextpkt;
1368                 q->m_nextpkt = NULL;
1369                 m_freem(q);
1370         }
1371         remque(fp);
1372         mpipe_free(&ipq_mpipe, fp);
1373         nipq--;
1374 }
1375
1376 /*
1377  * IP timer processing;
1378  * if a timer expires on a reassembly
1379  * queue, discard it.
1380  */
1381 void
1382 ip_slowtimo(void)
1383 {
1384         struct ipq *fp;
1385         int i;
1386
1387         crit_enter();
1388         for (i = 0; i < IPREASS_NHASH; i++) {
1389                 fp = ipq[i].next;
1390                 if (fp == NULL)
1391                         continue;
1392                 while (fp != &ipq[i]) {
1393                         --fp->ipq_ttl;
1394                         fp = fp->next;
1395                         if (fp->prev->ipq_ttl == 0) {
1396                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1397                                 ip_freef(fp->prev);
1398                         }
1399                 }
1400         }
1401         /*
1402          * If we are over the maximum number of fragments
1403          * (due to the limit being lowered), drain off
1404          * enough to get down to the new limit.
1405          */
1406         if (maxnipq >= 0 && nipq > maxnipq) {
1407                 for (i = 0; i < IPREASS_NHASH; i++) {
1408                         while (nipq > maxnipq &&
1409                                 (ipq[i].next != &ipq[i])) {
1410                                 ipstat.ips_fragdropped +=
1411                                     ipq[i].next->ipq_nfrags;
1412                                 ip_freef(ipq[i].next);
1413                         }
1414                 }
1415         }
1416         ipflow_slowtimo();
1417         crit_exit();
1418 }
1419
1420 /*
1421  * Drain off all datagram fragments.
1422  */
1423 void
1424 ip_drain(void)
1425 {
1426         int i;
1427
1428         for (i = 0; i < IPREASS_NHASH; i++) {
1429                 while (ipq[i].next != &ipq[i]) {
1430                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1431                         ip_freef(ipq[i].next);
1432                 }
1433         }
1434         in_rtqdrain();
1435 }
1436
1437 /*
1438  * Do option processing on a datagram,
1439  * possibly discarding it if bad options are encountered,
1440  * or forwarding it if source-routed.
1441  * The pass argument is used when operating in the IPSTEALTH
1442  * mode to tell what options to process:
1443  * [LS]SRR (pass 0) or the others (pass 1).
1444  * The reason for as many as two passes is that when doing IPSTEALTH,
1445  * non-routing options should be processed only if the packet is for us.
1446  * Returns 1 if packet has been forwarded/freed,
1447  * 0 if the packet should be processed further.
1448  */
1449 static int
1450 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1451 {
1452         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1453         struct ip *ip = mtod(m, struct ip *);
1454         u_char *cp;
1455         struct in_ifaddr *ia;
1456         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1457         boolean_t forward = FALSE;
1458         struct in_addr *sin, dst;
1459         n_time ntime;
1460
1461         dst = ip->ip_dst;
1462         cp = (u_char *)(ip + 1);
1463         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1464         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1465                 opt = cp[IPOPT_OPTVAL];
1466                 if (opt == IPOPT_EOL)
1467                         break;
1468                 if (opt == IPOPT_NOP)
1469                         optlen = 1;
1470                 else {
1471                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1472                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1473                                 goto bad;
1474                         }
1475                         optlen = cp[IPOPT_OLEN];
1476                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1477                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1478                                 goto bad;
1479                         }
1480                 }
1481                 switch (opt) {
1482
1483                 default:
1484                         break;
1485
1486                 /*
1487                  * Source routing with record.
1488                  * Find interface with current destination address.
1489                  * If none on this machine then drop if strictly routed,
1490                  * or do nothing if loosely routed.
1491                  * Record interface address and bring up next address
1492                  * component.  If strictly routed make sure next
1493                  * address is on directly accessible net.
1494                  */
1495                 case IPOPT_LSRR:
1496                 case IPOPT_SSRR:
1497                         if (ipstealth && pass > 0)
1498                                 break;
1499                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1500                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1501                                 goto bad;
1502                         }
1503                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1504                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1505                                 goto bad;
1506                         }
1507                         ipaddr.sin_addr = ip->ip_dst;
1508                         ia = (struct in_ifaddr *)
1509                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1510                         if (ia == NULL) {
1511                                 if (opt == IPOPT_SSRR) {
1512                                         type = ICMP_UNREACH;
1513                                         code = ICMP_UNREACH_SRCFAIL;
1514                                         goto bad;
1515                                 }
1516                                 if (!ip_dosourceroute)
1517                                         goto nosourcerouting;
1518                                 /*
1519                                  * Loose routing, and not at next destination
1520                                  * yet; nothing to do except forward.
1521                                  */
1522                                 break;
1523                         }
1524                         off--;                  /* 0 origin */
1525                         if (off > optlen - (int)sizeof(struct in_addr)) {
1526                                 /*
1527                                  * End of source route.  Should be for us.
1528                                  */
1529                                 if (!ip_acceptsourceroute)
1530                                         goto nosourcerouting;
1531                                 save_rte(cp, ip->ip_src);
1532                                 break;
1533                         }
1534                         if (ipstealth)
1535                                 goto dropit;
1536                         if (!ip_dosourceroute) {
1537                                 if (ipforwarding) {
1538                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1539
1540                                         /*
1541                                          * Acting as a router, so generate ICMP
1542                                          */
1543 nosourcerouting:
1544                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1545                                         log(LOG_WARNING,
1546                                             "attempted source route from %s to %s\n",
1547                                             inet_ntoa(ip->ip_src), buf);
1548                                         type = ICMP_UNREACH;
1549                                         code = ICMP_UNREACH_SRCFAIL;
1550                                         goto bad;
1551                                 } else {
1552                                         /*
1553                                          * Not acting as a router,
1554                                          * so silently drop.
1555                                          */
1556 dropit:
1557                                         ipstat.ips_cantforward++;
1558                                         m_freem(m);
1559                                         return (1);
1560                                 }
1561                         }
1562
1563                         /*
1564                          * locate outgoing interface
1565                          */
1566                         memcpy(&ipaddr.sin_addr, cp + off,
1567                             sizeof ipaddr.sin_addr);
1568
1569                         if (opt == IPOPT_SSRR) {
1570 #define INA     struct in_ifaddr *
1571 #define SA      struct sockaddr *
1572                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1573                                                                         == NULL)
1574                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1575                         } else
1576                                 ia = ip_rtaddr(ipaddr.sin_addr,
1577                                                &ipforward_rt[mycpuid]);
1578                         if (ia == NULL) {
1579                                 type = ICMP_UNREACH;
1580                                 code = ICMP_UNREACH_SRCFAIL;
1581                                 goto bad;
1582                         }
1583                         ip->ip_dst = ipaddr.sin_addr;
1584                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1585                             sizeof(struct in_addr));
1586                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1587                         /*
1588                          * Let ip_intr's mcast routing check handle mcast pkts
1589                          */
1590                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1591                         break;
1592
1593                 case IPOPT_RR:
1594                         if (ipstealth && pass == 0)
1595                                 break;
1596                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1597                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1598                                 goto bad;
1599                         }
1600                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1601                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1602                                 goto bad;
1603                         }
1604                         /*
1605                          * If no space remains, ignore.
1606                          */
1607                         off--;                  /* 0 origin */
1608                         if (off > optlen - (int)sizeof(struct in_addr))
1609                                 break;
1610                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1611                             sizeof ipaddr.sin_addr);
1612                         /*
1613                          * locate outgoing interface; if we're the destination,
1614                          * use the incoming interface (should be same).
1615                          */
1616                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1617                             (ia = ip_rtaddr(ipaddr.sin_addr,
1618                                             &ipforward_rt[mycpuid]))
1619                                                                      == NULL) {
1620                                 type = ICMP_UNREACH;
1621                                 code = ICMP_UNREACH_HOST;
1622                                 goto bad;
1623                         }
1624                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1625                             sizeof(struct in_addr));
1626                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1627                         break;
1628
1629                 case IPOPT_TS:
1630                         if (ipstealth && pass == 0)
1631                                 break;
1632                         code = cp - (u_char *)ip;
1633                         if (optlen < 4 || optlen > 40) {
1634                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1635                                 goto bad;
1636                         }
1637                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1638                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1639                                 goto bad;
1640                         }
1641                         if (off > optlen - (int)sizeof(int32_t)) {
1642                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1643                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1644                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1645                                         goto bad;
1646                                 }
1647                                 break;
1648                         }
1649                         off--;                          /* 0 origin */
1650                         sin = (struct in_addr *)(cp + off);
1651                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1652
1653                         case IPOPT_TS_TSONLY:
1654                                 break;
1655
1656                         case IPOPT_TS_TSANDADDR:
1657                                 if (off + sizeof(n_time) +
1658                                     sizeof(struct in_addr) > optlen) {
1659                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1660                                         goto bad;
1661                                 }
1662                                 ipaddr.sin_addr = dst;
1663                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1664                                                             m->m_pkthdr.rcvif);
1665                                 if (ia == NULL)
1666                                         continue;
1667                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1668                                     sizeof(struct in_addr));
1669                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1670                                 off += sizeof(struct in_addr);
1671                                 break;
1672
1673                         case IPOPT_TS_PRESPEC:
1674                                 if (off + sizeof(n_time) +
1675                                     sizeof(struct in_addr) > optlen) {
1676                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1677                                         goto bad;
1678                                 }
1679                                 memcpy(&ipaddr.sin_addr, sin,
1680                                     sizeof(struct in_addr));
1681                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1682                                         continue;
1683                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1684                                 off += sizeof(struct in_addr);
1685                                 break;
1686
1687                         default:
1688                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1689                                 goto bad;
1690                         }
1691                         ntime = iptime();
1692                         memcpy(cp + off, &ntime, sizeof(n_time));
1693                         cp[IPOPT_OFFSET] += sizeof(n_time);
1694                 }
1695         }
1696         if (forward && ipforwarding) {
1697                 ip_forward(m, TRUE, next_hop);
1698                 return (1);
1699         }
1700         return (0);
1701 bad:
1702         icmp_error(m, type, code, 0, 0);
1703         ipstat.ips_badoptions++;
1704         return (1);
1705 }
1706
1707 /*
1708  * Given address of next destination (final or next hop),
1709  * return internet address info of interface to be used to get there.
1710  */
1711 struct in_ifaddr *
1712 ip_rtaddr(struct in_addr dst, struct route *ro)
1713 {
1714         struct sockaddr_in *sin;
1715
1716         sin = (struct sockaddr_in *)&ro->ro_dst;
1717
1718         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1719                 if (ro->ro_rt != NULL) {
1720                         RTFREE(ro->ro_rt);
1721                         ro->ro_rt = NULL;
1722                 }
1723                 sin->sin_family = AF_INET;
1724                 sin->sin_len = sizeof *sin;
1725                 sin->sin_addr = dst;
1726                 rtalloc_ign(ro, RTF_PRCLONING);
1727         }
1728
1729         if (ro->ro_rt == NULL)
1730                 return (NULL);
1731
1732         return (ifatoia(ro->ro_rt->rt_ifa));
1733 }
1734
1735 /*
1736  * Save incoming source route for use in replies,
1737  * to be picked up later by ip_srcroute if the receiver is interested.
1738  */
1739 void
1740 save_rte(u_char *option, struct in_addr dst)
1741 {
1742         unsigned olen;
1743
1744         olen = option[IPOPT_OLEN];
1745 #ifdef DIAGNOSTIC
1746         if (ipprintfs)
1747                 kprintf("save_rte: olen %d\n", olen);
1748 #endif
1749         if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst)))
1750                 return;
1751         bcopy(option, ip_srcrt.srcopt, olen);
1752         ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1753         ip_srcrt.dst = dst;
1754 }
1755
1756 /*
1757  * Retrieve incoming source route for use in replies,
1758  * in the same form used by setsockopt.
1759  * The first hop is placed before the options, will be removed later.
1760  */
1761 struct mbuf *
1762 ip_srcroute(void)
1763 {
1764         struct in_addr *p, *q;
1765         struct mbuf *m;
1766
1767         if (ip_nhops == 0)
1768                 return (NULL);
1769         m = m_get(MB_DONTWAIT, MT_HEADER);
1770         if (m == NULL)
1771                 return (NULL);
1772
1773 #define OPTSIZ  (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt))
1774
1775         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1776         m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) +
1777             OPTSIZ;
1778 #ifdef DIAGNOSTIC
1779         if (ipprintfs)
1780                 kprintf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
1781 #endif
1782
1783         /*
1784          * First save first hop for return route
1785          */
1786         p = &ip_srcrt.route[ip_nhops - 1];
1787         *(mtod(m, struct in_addr *)) = *p--;
1788 #ifdef DIAGNOSTIC
1789         if (ipprintfs)
1790                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1791 #endif
1792
1793         /*
1794          * Copy option fields and padding (nop) to mbuf.
1795          */
1796         ip_srcrt.nop = IPOPT_NOP;
1797         ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1798         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &ip_srcrt.nop,
1799             OPTSIZ);
1800         q = (struct in_addr *)(mtod(m, caddr_t) +
1801             sizeof(struct in_addr) + OPTSIZ);
1802 #undef OPTSIZ
1803         /*
1804          * Record return path as an IP source route,
1805          * reversing the path (pointers are now aligned).
1806          */
1807         while (p >= ip_srcrt.route) {
1808 #ifdef DIAGNOSTIC
1809                 if (ipprintfs)
1810                         kprintf(" %x", ntohl(q->s_addr));
1811 #endif
1812                 *q++ = *p--;
1813         }
1814         /*
1815          * Last hop goes to final destination.
1816          */
1817         *q = ip_srcrt.dst;
1818 #ifdef DIAGNOSTIC
1819         if (ipprintfs)
1820                 kprintf(" %x\n", ntohl(q->s_addr));
1821 #endif
1822         return (m);
1823 }
1824
1825 /*
1826  * Strip out IP options.
1827  */
1828 void
1829 ip_stripoptions(struct mbuf *m)
1830 {
1831         int datalen;
1832         struct ip *ip = mtod(m, struct ip *);
1833         caddr_t opts;
1834         int optlen;
1835
1836         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1837         opts = (caddr_t)(ip + 1);
1838         datalen = m->m_len - (sizeof(struct ip) + optlen);
1839         bcopy(opts + optlen, opts, datalen);
1840         m->m_len -= optlen;
1841         if (m->m_flags & M_PKTHDR)
1842                 m->m_pkthdr.len -= optlen;
1843         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1844 }
1845
1846 u_char inetctlerrmap[PRC_NCMDS] = {
1847         0,              0,              0,              0,
1848         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1849         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1850         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1851         0,              0,              0,              0,
1852         ENOPROTOOPT,    ECONNREFUSED
1853 };
1854
1855 /*
1856  * Forward a packet.  If some error occurs return the sender
1857  * an icmp packet.  Note we can't always generate a meaningful
1858  * icmp message because icmp doesn't have a large enough repertoire
1859  * of codes and types.
1860  *
1861  * If not forwarding, just drop the packet.  This could be confusing
1862  * if ipforwarding was zero but some routing protocol was advancing
1863  * us as a gateway to somewhere.  However, we must let the routing
1864  * protocol deal with that.
1865  *
1866  * The using_srcrt parameter indicates whether the packet is being forwarded
1867  * via a source route.
1868  */
1869 static void
1870 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1871 {
1872         struct ip *ip = mtod(m, struct ip *);
1873         struct sockaddr_in *ipforward_rtaddr;
1874         struct rtentry *rt;
1875         int error, type = 0, code = 0, destmtu = 0;
1876         struct mbuf *mcopy;
1877         n_long dest;
1878         struct in_addr pkt_dst;
1879         struct m_hdr tag;
1880         struct route *cache_rt = &ipforward_rt[mycpuid];
1881
1882         dest = INADDR_ANY;
1883         /*
1884          * Cache the destination address of the packet; this may be
1885          * changed by use of 'ipfw fwd'.
1886          */
1887         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1888
1889 #ifdef DIAGNOSTIC
1890         if (ipprintfs)
1891                 kprintf("forward: src %x dst %x ttl %x\n",
1892                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1893 #endif
1894
1895         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1896                 ipstat.ips_cantforward++;
1897                 m_freem(m);
1898                 return;
1899         }
1900         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1901                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1902                 return;
1903         }
1904
1905         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1906         if (cache_rt->ro_rt == NULL ||
1907             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1908                 if (cache_rt->ro_rt != NULL) {
1909                         RTFREE(cache_rt->ro_rt);
1910                         cache_rt->ro_rt = NULL;
1911                 }
1912                 ipforward_rtaddr->sin_family = AF_INET;
1913                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1914                 ipforward_rtaddr->sin_addr = pkt_dst;
1915                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1916                 if (cache_rt->ro_rt == NULL) {
1917                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1918                         return;
1919                 }
1920         }
1921         rt = cache_rt->ro_rt;
1922
1923         /*
1924          * Save the IP header and at most 8 bytes of the payload,
1925          * in case we need to generate an ICMP message to the src.
1926          *
1927          * XXX this can be optimized a lot by saving the data in a local
1928          * buffer on the stack (72 bytes at most), and only allocating the
1929          * mbuf if really necessary. The vast majority of the packets
1930          * are forwarded without having to send an ICMP back (either
1931          * because unnecessary, or because rate limited), so we are
1932          * really we are wasting a lot of work here.
1933          *
1934          * We don't use m_copy() because it might return a reference
1935          * to a shared cluster. Both this function and ip_output()
1936          * assume exclusive access to the IP header in `m', so any
1937          * data in a cluster may change before we reach icmp_error().
1938          */
1939         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1940         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1941                 /*
1942                  * It's probably ok if the pkthdr dup fails (because
1943                  * the deep copy of the tag chain failed), but for now
1944                  * be conservative and just discard the copy since
1945                  * code below may some day want the tags.
1946                  */
1947                 m_free(mcopy);
1948                 mcopy = NULL;
1949         }
1950         if (mcopy != NULL) {
1951                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1952                     (int)ip->ip_len);
1953                 mcopy->m_pkthdr.len = mcopy->m_len;
1954                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1955         }
1956
1957         if (!ipstealth)
1958                 ip->ip_ttl -= IPTTLDEC;
1959
1960         /*
1961          * If forwarding packet using same interface that it came in on,
1962          * perhaps should send a redirect to sender to shortcut a hop.
1963          * Only send redirect if source is sending directly to us,
1964          * and if packet was not source routed (or has any options).
1965          * Also, don't send redirect if forwarding using a default route
1966          * or a route modified by a redirect.
1967          */
1968         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1969             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1970             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1971             ipsendredirects && !using_srcrt && next_hop == NULL) {
1972                 u_long src = ntohl(ip->ip_src.s_addr);
1973                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1974
1975                 if (rt_ifa != NULL &&
1976                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1977                         if (rt->rt_flags & RTF_GATEWAY)
1978                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1979                         else
1980                                 dest = pkt_dst.s_addr;
1981                         /*
1982                          * Router requirements says to only send
1983                          * host redirects.
1984                          */
1985                         type = ICMP_REDIRECT;
1986                         code = ICMP_REDIRECT_HOST;
1987 #ifdef DIAGNOSTIC
1988                         if (ipprintfs)
1989                                 kprintf("redirect (%d) to %x\n", code, dest);
1990 #endif
1991                 }
1992         }
1993
1994         if (next_hop != NULL) {
1995                 /* Pass IPFORWARD info if available */
1996                 tag.mh_type = MT_TAG;
1997                 tag.mh_flags = PACKET_TAG_IPFORWARD;
1998                 tag.mh_data = (caddr_t)next_hop;
1999                 tag.mh_next = m;
2000                 m = (struct mbuf *)&tag;
2001         }
2002
2003         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL,
2004                           NULL);
2005         if (error == 0) {
2006                 ipstat.ips_forward++;
2007                 if (type == 0) {
2008                         if (mcopy) {
2009                                 /* ipflow_create() will free mcopy */
2010                                 ipflow_create(cache_rt, mcopy);
2011                         }
2012                         return;         /* most common case */
2013                 } else {
2014                         ipstat.ips_redirectsent++;
2015                 }
2016         } else {
2017                 ipstat.ips_cantforward++;
2018         }
2019
2020         if (mcopy == NULL)
2021                 return;
2022
2023         /*
2024          * Send ICMP message.
2025          */
2026
2027         switch (error) {
2028
2029         case 0:                         /* forwarded, but need redirect */
2030                 /* type, code set above */
2031                 break;
2032
2033         case ENETUNREACH:               /* shouldn't happen, checked above */
2034         case EHOSTUNREACH:
2035         case ENETDOWN:
2036         case EHOSTDOWN:
2037         default:
2038                 type = ICMP_UNREACH;
2039                 code = ICMP_UNREACH_HOST;
2040                 break;
2041
2042         case EMSGSIZE:
2043                 type = ICMP_UNREACH;
2044                 code = ICMP_UNREACH_NEEDFRAG;
2045 #ifdef IPSEC
2046                 /*
2047                  * If the packet is routed over IPsec tunnel, tell the
2048                  * originator the tunnel MTU.
2049                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2050                  * XXX quickhack!!!
2051                  */
2052                 if (cache_rt->ro_rt != NULL) {
2053                         struct secpolicy *sp = NULL;
2054                         int ipsecerror;
2055                         int ipsechdr;
2056                         struct route *ro;
2057
2058                         sp = ipsec4_getpolicybyaddr(mcopy,
2059                                                     IPSEC_DIR_OUTBOUND,
2060                                                     IP_FORWARDING,
2061                                                     &ipsecerror);
2062
2063                         if (sp == NULL)
2064                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2065                         else {
2066                                 /* count IPsec header size */
2067                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2068                                                          IPSEC_DIR_OUTBOUND,
2069                                                          NULL);
2070
2071                                 /*
2072                                  * find the correct route for outer IPv4
2073                                  * header, compute tunnel MTU.
2074                                  *
2075                                  */
2076                                 if (sp->req != NULL && sp->req->sav != NULL &&
2077                                     sp->req->sav->sah != NULL) {
2078                                         ro = &sp->req->sav->sah->sa_route;
2079                                         if (ro->ro_rt != NULL &&
2080                                             ro->ro_rt->rt_ifp != NULL) {
2081                                                 destmtu =
2082                                                     ro->ro_rt->rt_ifp->if_mtu;
2083                                                 destmtu -= ipsechdr;
2084                                         }
2085                                 }
2086
2087                                 key_freesp(sp);
2088                         }
2089                 }
2090 #elif FAST_IPSEC
2091                 /*
2092                  * If the packet is routed over IPsec tunnel, tell the
2093                  * originator the tunnel MTU.
2094                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2095                  * XXX quickhack!!!
2096                  */
2097                 if (cache_rt->ro_rt != NULL) {
2098                         struct secpolicy *sp = NULL;
2099                         int ipsecerror;
2100                         int ipsechdr;
2101                         struct route *ro;
2102
2103                         sp = ipsec_getpolicybyaddr(mcopy,
2104                                                    IPSEC_DIR_OUTBOUND,
2105                                                    IP_FORWARDING,
2106                                                    &ipsecerror);
2107
2108                         if (sp == NULL)
2109                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2110                         else {
2111                                 /* count IPsec header size */
2112                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2113                                                          IPSEC_DIR_OUTBOUND,
2114                                                          NULL);
2115
2116                                 /*
2117                                  * find the correct route for outer IPv4
2118                                  * header, compute tunnel MTU.
2119                                  */
2120
2121                                 if (sp->req != NULL &&
2122                                     sp->req->sav != NULL &&
2123                                     sp->req->sav->sah != NULL) {
2124                                         ro = &sp->req->sav->sah->sa_route;
2125                                         if (ro->ro_rt != NULL &&
2126                                             ro->ro_rt->rt_ifp != NULL) {
2127                                                 destmtu =
2128                                                     ro->ro_rt->rt_ifp->if_mtu;
2129                                                 destmtu -= ipsechdr;
2130                                         }
2131                                 }
2132
2133                                 KEY_FREESP(&sp);
2134                         }
2135                 }
2136 #else /* !IPSEC && !FAST_IPSEC */
2137                 if (cache_rt->ro_rt != NULL)
2138                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2139 #endif /*IPSEC*/
2140                 ipstat.ips_cantfrag++;
2141                 break;
2142
2143         case ENOBUFS:
2144                 /*
2145                  * A router should not generate ICMP_SOURCEQUENCH as
2146                  * required in RFC1812 Requirements for IP Version 4 Routers.
2147                  * Source quench could be a big problem under DoS attacks,
2148                  * or if the underlying interface is rate-limited.
2149                  * Those who need source quench packets may re-enable them
2150                  * via the net.inet.ip.sendsourcequench sysctl.
2151                  */
2152                 if (!ip_sendsourcequench) {
2153                         m_freem(mcopy);
2154                         return;
2155                 } else {
2156                         type = ICMP_SOURCEQUENCH;
2157                         code = 0;
2158                 }
2159                 break;
2160
2161         case EACCES:                    /* ipfw denied packet */
2162                 m_freem(mcopy);
2163                 return;
2164         }
2165         icmp_error(mcopy, type, code, dest, destmtu);
2166 }
2167
2168 void
2169 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2170                struct mbuf *m)
2171 {
2172         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2173                 struct timeval tv;
2174
2175                 microtime(&tv);
2176                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2177                     SCM_TIMESTAMP, SOL_SOCKET);
2178                 if (*mp)
2179                         mp = &(*mp)->m_next;
2180         }
2181         if (inp->inp_flags & INP_RECVDSTADDR) {
2182                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2183                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2184                 if (*mp)
2185                         mp = &(*mp)->m_next;
2186         }
2187         if (inp->inp_flags & INP_RECVTTL) {
2188                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2189                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2190                 if (*mp)
2191                         mp = &(*mp)->m_next;
2192         }
2193 #ifdef notyet
2194         /* XXX
2195          * Moving these out of udp_input() made them even more broken
2196          * than they already were.
2197          */
2198         /* options were tossed already */
2199         if (inp->inp_flags & INP_RECVOPTS) {
2200                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2201                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2202                 if (*mp)
2203                         mp = &(*mp)->m_next;
2204         }
2205         /* ip_srcroute doesn't do what we want here, need to fix */
2206         if (inp->inp_flags & INP_RECVRETOPTS) {
2207                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(),
2208                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2209                 if (*mp)
2210                         mp = &(*mp)->m_next;
2211         }
2212 #endif
2213         if (inp->inp_flags & INP_RECVIF) {
2214                 struct ifnet *ifp;
2215                 struct sdlbuf {
2216                         struct sockaddr_dl sdl;
2217                         u_char  pad[32];
2218                 } sdlbuf;
2219                 struct sockaddr_dl *sdp;
2220                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2221
2222                 if (((ifp = m->m_pkthdr.rcvif)) &&
2223                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2224                         sdp = IF_LLSOCKADDR(ifp);
2225                         /*
2226                          * Change our mind and don't try copy.
2227                          */
2228                         if ((sdp->sdl_family != AF_LINK) ||
2229                             (sdp->sdl_len > sizeof(sdlbuf))) {
2230                                 goto makedummy;
2231                         }
2232                         bcopy(sdp, sdl2, sdp->sdl_len);
2233                 } else {
2234 makedummy:
2235                         sdl2->sdl_len =
2236                             offsetof(struct sockaddr_dl, sdl_data[0]);
2237                         sdl2->sdl_family = AF_LINK;
2238                         sdl2->sdl_index = 0;
2239                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2240                 }
2241                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2242                         IP_RECVIF, IPPROTO_IP);
2243                 if (*mp)
2244                         mp = &(*mp)->m_next;
2245         }
2246 }
2247
2248 /*
2249  * XXX these routines are called from the upper part of the kernel.
2250  *
2251  * They could also be moved to ip_mroute.c, since all the RSVP
2252  *  handling is done there already.
2253  */
2254 int
2255 ip_rsvp_init(struct socket *so)
2256 {
2257         if (so->so_type != SOCK_RAW ||
2258             so->so_proto->pr_protocol != IPPROTO_RSVP)
2259                 return EOPNOTSUPP;
2260
2261         if (ip_rsvpd != NULL)
2262                 return EADDRINUSE;
2263
2264         ip_rsvpd = so;
2265         /*
2266          * This may seem silly, but we need to be sure we don't over-increment
2267          * the RSVP counter, in case something slips up.
2268          */
2269         if (!ip_rsvp_on) {
2270                 ip_rsvp_on = 1;
2271                 rsvp_on++;
2272         }
2273
2274         return 0;
2275 }
2276
2277 int
2278 ip_rsvp_done(void)
2279 {
2280         ip_rsvpd = NULL;
2281         /*
2282          * This may seem silly, but we need to be sure we don't over-decrement
2283          * the RSVP counter, in case something slips up.
2284          */
2285         if (ip_rsvp_on) {
2286                 ip_rsvp_on = 0;
2287                 rsvp_on--;
2288         }
2289         return 0;
2290 }
2291
2292 void
2293 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2294 {
2295         int off, proto;
2296         __va_list ap;
2297
2298         __va_start(ap, m);
2299         off = __va_arg(ap, int);
2300         proto = __va_arg(ap, int);
2301         __va_end(ap);
2302
2303         if (rsvp_input_p) { /* call the real one if loaded */
2304                 rsvp_input_p(m, off, proto);
2305                 return;
2306         }
2307
2308         /* Can still get packets with rsvp_on = 0 if there is a local member
2309          * of the group to which the RSVP packet is addressed.  But in this
2310          * case we want to throw the packet away.
2311          */
2312
2313         if (!rsvp_on) {
2314                 m_freem(m);
2315                 return;
2316         }
2317
2318         if (ip_rsvpd != NULL) {
2319                 rip_input(m, off, proto);
2320                 return;
2321         }
2322         /* Drop the packet */
2323         m_freem(m);
2324 }