Merge branch 'vendor/SENDMAIL'
[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 1.115 2008/10/28 07:09:26 sephe Exp $
69  */
70
71 #define _IP_VHL
72
73 #include "opt_bootp.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipdn.h"
76 #include "opt_ipdivert.h"
77 #include "opt_ipfilter.h"
78 #include "opt_ipstealth.h"
79 #include "opt_ipsec.h"
80
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/mpipe.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/time.h>
90 #include <sys/globaldata.h>
91 #include <sys/thread.h>
92 #include <sys/kernel.h>
93 #include <sys/syslog.h>
94 #include <sys/sysctl.h>
95 #include <sys/in_cksum.h>
96 #include <sys/lock.h>
97
98 #include <sys/mplock2.h>
99
100 #include <machine/stdarg.h>
101
102 #include <net/if.h>
103 #include <net/if_types.h>
104 #include <net/if_var.h>
105 #include <net/if_dl.h>
106 #include <net/pfil.h>
107 #include <net/route.h>
108 #include <net/netisr.h>
109
110 #include <netinet/in.h>
111 #include <netinet/in_systm.h>
112 #include <netinet/in_var.h>
113 #include <netinet/ip.h>
114 #include <netinet/in_pcb.h>
115 #include <netinet/ip_var.h>
116 #include <netinet/ip_icmp.h>
117 #include <netinet/ip_divert.h>
118 #include <netinet/ip_flow.h>
119
120 #include <sys/thread2.h>
121 #include <sys/msgport2.h>
122 #include <net/netmsg2.h>
123
124 #include <sys/socketvar.h>
125
126 #include <net/ipfw/ip_fw.h>
127 #include <net/dummynet/ip_dummynet.h>
128
129 #ifdef IPSEC
130 #include <netinet6/ipsec.h>
131 #include <netproto/key/key.h>
132 #endif
133
134 #ifdef FAST_IPSEC
135 #include <netproto/ipsec/ipsec.h>
136 #include <netproto/ipsec/key.h>
137 #endif
138
139 int rsvp_on = 0;
140 static int ip_rsvp_on;
141 struct socket *ip_rsvpd;
142
143 int ip_mpsafe = 1;
144 TUNABLE_INT("net.inet.ip.mpsafe", &ip_mpsafe);
145
146 int ipforwarding = 0;
147 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
148     &ipforwarding, 0, "Enable IP forwarding between interfaces");
149
150 static int ipsendredirects = 1; /* XXX */
151 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
152     &ipsendredirects, 0, "Enable sending IP redirects");
153
154 int ip_defttl = IPDEFTTL;
155 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
156     &ip_defttl, 0, "Maximum TTL on IP packets");
157
158 static int ip_dosourceroute = 0;
159 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
160     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
161
162 static int ip_acceptsourceroute = 0;
163 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
164     CTLFLAG_RW, &ip_acceptsourceroute, 0,
165     "Enable accepting source routed IP packets");
166
167 static int ip_keepfaith = 0;
168 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
169     &ip_keepfaith, 0,
170     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
171
172 static int nipq = 0;    /* total # of reass queues */
173 static int maxnipq;
174 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
175     &maxnipq, 0,
176     "Maximum number of IPv4 fragment reassembly queue entries");
177
178 static int maxfragsperpacket;
179 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
180     &maxfragsperpacket, 0,
181     "Maximum number of IPv4 fragments allowed per packet");
182
183 static int ip_sendsourcequench = 0;
184 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
185     &ip_sendsourcequench, 0,
186     "Enable the transmission of source quench packets");
187
188 int ip_do_randomid = 1;
189 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
190     &ip_do_randomid, 0,
191     "Assign random ip_id values");      
192 /*
193  * XXX - Setting ip_checkinterface mostly implements the receive side of
194  * the Strong ES model described in RFC 1122, but since the routing table
195  * and transmit implementation do not implement the Strong ES model,
196  * setting this to 1 results in an odd hybrid.
197  *
198  * XXX - ip_checkinterface currently must be disabled if you use ipnat
199  * to translate the destination address to another local interface.
200  *
201  * XXX - ip_checkinterface must be disabled if you add IP aliases
202  * to the loopback interface instead of the interface where the
203  * packets for those addresses are received.
204  */
205 static int ip_checkinterface = 0;
206 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
207     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
208
209 #ifdef DIAGNOSTIC
210 static int ipprintfs = 0;
211 #endif
212
213 extern  int udp_mpsafe_proto;
214 extern  int tcp_mpsafe_proto;
215
216 extern  struct domain inetdomain;
217 extern  struct protosw inetsw[];
218 u_char  ip_protox[IPPROTO_MAX];
219 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
220 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
221                                                 /* inet addr hash table */
222 u_long  in_ifaddrhmask;                         /* mask for hash table */
223
224 struct ip_stats ipstats_percpu[MAXCPU];
225 #ifdef SMP
226 static int
227 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
228 {
229         int cpu, error = 0;
230
231         for (cpu = 0; cpu < ncpus; ++cpu) {
232                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
233                                         sizeof(struct ip_stats))))
234                         break;
235                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
236                                        sizeof(struct ip_stats))))
237                         break;
238         }
239
240         return (error);
241 }
242 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
243     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
244 #else
245 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
246     &ipstat, ip_stats, "IP statistics");
247 #endif
248
249 /* Packet reassembly stuff */
250 #define IPREASS_NHASH_LOG2      6
251 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
252 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
253 #define IPREASS_HASH(x,y)                                               \
254     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
255
256 static struct ipq ipq[IPREASS_NHASH];
257
258 #ifdef IPCTL_DEFMTU
259 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
260     &ip_mtu, 0, "Default MTU");
261 #endif
262
263 #ifdef IPSTEALTH
264 static int ipstealth = 0;
265 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
266 #else
267 static const int ipstealth = 0;
268 #endif
269
270 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
271
272 struct pfil_head inet_pfil_hook;
273
274 /*
275  * struct ip_srcrt_opt is used to store packet state while it travels
276  * through the stack.
277  *
278  * XXX Note that the code even makes assumptions on the size and
279  * alignment of fields inside struct ip_srcrt so e.g. adding some
280  * fields will break the code.  This needs to be fixed.
281  *
282  * We need to save the IP options in case a protocol wants to respond
283  * to an incoming packet over the same route if the packet got here
284  * using IP source routing.  This allows connection establishment and
285  * maintenance when the remote end is on a network that is not known
286  * to us.
287  */
288 struct ip_srcrt {
289         struct  in_addr dst;                    /* final destination */
290         char    nop;                            /* one NOP to align */
291         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
292         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
293 };
294
295 struct ip_srcrt_opt {
296         int             ip_nhops;
297         struct ip_srcrt ip_srcrt;
298 };
299
300 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
301 static struct malloc_pipe ipq_mpipe;
302
303 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
304 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
305 static void             ip_freef(struct ipq *);
306 static void             ip_input_handler(struct netmsg *);
307
308 /*
309  * IP initialization: fill in IP protocol switch table.
310  * All protocols not implemented in kernel go to raw IP protocol handler.
311  */
312 void
313 ip_init(void)
314 {
315         struct protosw *pr;
316         uint32_t flags;
317         int i;
318 #ifdef SMP
319         int cpu;
320 #endif
321
322         /*
323          * Make sure we can handle a reasonable number of fragments but
324          * cap it at 4000 (XXX).
325          */
326         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
327                     IFQ_MAXLEN, 4000, 0, NULL);
328         for (i = 0; i < ncpus; ++i) {
329                 TAILQ_INIT(&in_ifaddrheads[i]);
330                 in_ifaddrhashtbls[i] =
331                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
332         }
333         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
334         if (pr == NULL)
335                 panic("ip_init");
336         for (i = 0; i < IPPROTO_MAX; i++)
337                 ip_protox[i] = pr - inetsw;
338         for (pr = inetdomain.dom_protosw;
339              pr < inetdomain.dom_protoswNPROTOSW; pr++) {
340                 if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
341                         if (pr->pr_protocol != IPPROTO_RAW)
342                                 ip_protox[pr->pr_protocol] = pr - inetsw;
343
344                         /* XXX */
345                         switch (pr->pr_protocol) {
346                         case IPPROTO_TCP:
347                                 if (tcp_mpsafe_proto)
348                                         pr->pr_flags |= PR_MPSAFE;
349                                 break;
350
351                         case IPPROTO_UDP:
352                                 if (udp_mpsafe_proto)
353                                         pr->pr_flags |= PR_MPSAFE;
354                                 break;
355                         }
356                 }
357         }
358
359         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
360         inet_pfil_hook.ph_af = AF_INET;
361         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
362                 kprintf("%s: WARNING: unable to register pfil hook, "
363                         "error %d\n", __func__, i);
364         }
365
366         for (i = 0; i < IPREASS_NHASH; i++)
367                 ipq[i].next = ipq[i].prev = &ipq[i];
368
369         maxnipq = nmbclusters / 32;
370         maxfragsperpacket = 16;
371
372         ip_id = time_second & 0xffff;
373
374         /*
375          * Initialize IP statistics counters for each CPU.
376          *
377          */
378 #ifdef SMP
379         for (cpu = 0; cpu < ncpus; ++cpu) {
380                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
381         }
382 #else
383         bzero(&ipstat, sizeof(struct ip_stats));
384 #endif
385
386 #if defined(IPSEC) || defined(FAST_IPSEC)
387         /* XXX IPSEC is not MPSAFE yet */
388         flags = NETISR_FLAG_NOTMPSAFE;
389 #else
390         if (ip_mpsafe) {
391                 kprintf("ip: MPSAFE\n");
392                 flags = NETISR_FLAG_MPSAFE;
393         } else {
394                 flags = NETISR_FLAG_NOTMPSAFE;
395         }
396 #endif
397         netisr_register(NETISR_IP, ip_mport_in, ip_mport_pktinfo,
398                         ip_input_handler, flags);
399 }
400
401 /* Do transport protocol processing. */
402 static void
403 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
404 {
405         const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
406
407         /*
408          * Switch out to protocol's input routine.
409          */
410         PR_GET_MPLOCK(pr);
411         pr->pr_input(m, hlen, ip->ip_p);
412         PR_REL_MPLOCK(pr);
413 }
414
415 static void
416 transport_processing_handler(netmsg_t netmsg)
417 {
418         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
419         struct ip *ip;
420         int hlen;
421
422         ip = mtod(pmsg->nm_packet, struct ip *);
423         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
424
425         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
426         /* netmsg was embedded in the mbuf, do not reply! */
427 }
428
429 static void
430 ip_input_handler(struct netmsg *msg0)
431 {
432         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
433
434         ip_input(m);
435         /* msg0 was embedded in the mbuf, do not reply! */
436 }
437
438 /*
439  * IP input routine.  Checksum and byte swap header.  If fragmented
440  * try to reassemble.  Process options.  Pass to next level.
441  */
442 void
443 ip_input(struct mbuf *m)
444 {
445         struct ip *ip;
446         struct in_ifaddr *ia = NULL;
447         struct in_ifaddr_container *iac;
448         int hlen, checkif;
449         u_short sum;
450         struct in_addr pkt_dst;
451         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
452         boolean_t needredispatch = FALSE;
453         struct in_addr odst;                    /* original dst address(NAT) */
454         struct m_tag *mtag;
455         struct sockaddr_in *next_hop = NULL;
456 #ifdef FAST_IPSEC
457         struct tdb_ident *tdbi;
458         struct secpolicy *sp;
459         int error;
460 #endif
461
462         M_ASSERTPKTHDR(m);
463
464         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
465                 /* Next hop */
466                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
467                 KKASSERT(mtag != NULL);
468                 next_hop = m_tag_data(mtag);
469         }
470
471         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
472                 /* dummynet already filtered us */
473                 ip = mtod(m, struct ip *);
474                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
475                 goto iphack;
476         }
477
478         ipstat.ips_total++;
479
480         /* length checks already done in ip_mport() */
481         KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
482         ip = mtod(m, struct ip *);
483
484         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
485                 ipstat.ips_badvers++;
486                 goto bad;
487         }
488
489         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
490         /* length checks already done in ip_mport() */
491         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
492         KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
493
494         /* 127/8 must not appear on wire - RFC1122 */
495         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
496             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
497                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
498                         ipstat.ips_badaddr++;
499                         goto bad;
500                 }
501         }
502
503         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
504                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
505         } else {
506                 if (hlen == sizeof(struct ip))
507                         sum = in_cksum_hdr(ip);
508                 else
509                         sum = in_cksum(m, hlen);
510         }
511         if (sum != 0) {
512                 ipstat.ips_badsum++;
513                 goto bad;
514         }
515
516 #ifdef ALTQ
517         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
518                 /* packet is dropped by traffic conditioner */
519                 return;
520         }
521 #endif
522         /*
523          * Convert fields to host representation.
524          */
525         ip->ip_len = ntohs(ip->ip_len);
526         ip->ip_off = ntohs(ip->ip_off);
527
528         /* length checks already done in ip_mport() */
529         KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
530         KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
531
532         /*
533          * Trim mbufs if longer than the IP header would have us expect.
534          */
535         if (m->m_pkthdr.len > ip->ip_len) {
536                 if (m->m_len == m->m_pkthdr.len) {
537                         m->m_len = ip->ip_len;
538                         m->m_pkthdr.len = ip->ip_len;
539                 } else {
540                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
541                 }
542         }
543 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
544         /*
545          * Bypass packet filtering for packets from a tunnel (gif).
546          */
547         if (ipsec_gethist(m, NULL))
548                 goto pass;
549 #endif
550
551         /*
552          * IpHack's section.
553          * Right now when no processing on packet has done
554          * and it is still fresh out of network we do our black
555          * deals with it.
556          * - Firewall: deny/allow/divert
557          * - Xlate: translate packet's addr/port (NAT).
558          * - Pipe: pass pkt through dummynet.
559          * - Wrap: fake packet's addr/port <unimpl.>
560          * - Encapsulate: put it in another IP and send out. <unimp.>
561          */
562
563 iphack:
564         /*
565          * If we've been forwarded from the output side, then
566          * skip the firewall a second time
567          */
568         if (next_hop != NULL)
569                 goto ours;
570
571         /* No pfil hooks */
572         if (!pfil_has_hooks(&inet_pfil_hook)) {
573                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
574                         /*
575                          * Strip dummynet tags from stranded packets
576                          */
577                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
578                         KKASSERT(mtag != NULL);
579                         m_tag_delete(m, mtag);
580                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
581                 }
582                 goto pass;
583         }
584
585         /*
586          * Run through list of hooks for input packets.
587          *
588          * NB: Beware of the destination address changing (e.g.
589          *     by NAT rewriting). When this happens, tell
590          *     ip_forward to do the right thing.
591          */
592         odst = ip->ip_dst;
593         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
594                 return;
595         if (m == NULL)  /* consumed by filter */
596                 return;
597         ip = mtod(m, struct ip *);
598         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
599         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
600
601         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
602                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
603                 KKASSERT(mtag != NULL);
604                 next_hop = m_tag_data(mtag);
605         }
606         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
607                 ip_dn_queue(m);
608                 return;
609         }
610         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
611                 needredispatch = TRUE;
612                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
613         }
614 pass:
615         /*
616          * Process options and, if not destined for us,
617          * ship it on.  ip_dooptions returns 1 when an
618          * error was detected (causing an icmp message
619          * to be sent and the original packet to be freed).
620          */
621         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
622                 return;
623
624         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
625          * matter if it is destined to another node, or whether it is
626          * a multicast one, RSVP wants it! and prevents it from being forwarded
627          * anywhere else. Also checks if the rsvp daemon is running before
628          * grabbing the packet.
629          */
630         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
631                 goto ours;
632
633         /*
634          * Check our list of addresses, to see if the packet is for us.
635          * If we don't have any addresses, assume any unicast packet
636          * we receive might be for us (and let the upper layers deal
637          * with it).
638          */
639         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
640             !(m->m_flags & (M_MCAST | M_BCAST)))
641                 goto ours;
642
643         /*
644          * Cache the destination address of the packet; this may be
645          * changed by use of 'ipfw fwd'.
646          */
647         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
648
649         /*
650          * Enable a consistency check between the destination address
651          * and the arrival interface for a unicast packet (the RFC 1122
652          * strong ES model) if IP forwarding is disabled and the packet
653          * is not locally generated and the packet is not subject to
654          * 'ipfw fwd'.
655          *
656          * XXX - Checking also should be disabled if the destination
657          * address is ipnat'ed to a different interface.
658          *
659          * XXX - Checking is incompatible with IP aliases added
660          * to the loopback interface instead of the interface where
661          * the packets are received.
662          */
663         checkif = ip_checkinterface &&
664                   !ipforwarding &&
665                   m->m_pkthdr.rcvif != NULL &&
666                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
667                   next_hop == NULL;
668
669         /*
670          * Check for exact addresses in the hash bucket.
671          */
672         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
673                 ia = iac->ia;
674
675                 /*
676                  * If the address matches, verify that the packet
677                  * arrived via the correct interface if checking is
678                  * enabled.
679                  */
680                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
681                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
682                         goto ours;
683         }
684         ia = NULL;
685
686         /*
687          * Check for broadcast addresses.
688          *
689          * Only accept broadcast packets that arrive via the matching
690          * interface.  Reception of forwarded directed broadcasts would
691          * be handled via ip_forward() and ether_output() with the loopback
692          * into the stack for SIMPLEX interfaces handled by ether_output().
693          */
694         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
695                 struct ifaddr_container *ifac;
696
697                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
698                               ifa_link) {
699                         struct ifaddr *ifa = ifac->ifa;
700
701                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
702                                 continue;
703                         if (ifa->ifa_addr->sa_family != AF_INET)
704                                 continue;
705                         ia = ifatoia(ifa);
706                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
707                                                                 pkt_dst.s_addr)
708                                 goto ours;
709                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
710                                 goto ours;
711 #ifdef BOOTP_COMPAT
712                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
713                                 goto ours;
714 #endif
715                 }
716         }
717         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
718                 struct in_multi *inm;
719
720                 /* XXX Multicast is not MPSAFE yet */
721                 get_mplock();
722
723                 if (ip_mrouter != NULL) {
724                         /*
725                          * If we are acting as a multicast router, all
726                          * incoming multicast packets are passed to the
727                          * kernel-level multicast forwarding function.
728                          * The packet is returned (relatively) intact; if
729                          * ip_mforward() returns a non-zero value, the packet
730                          * must be discarded, else it may be accepted below.
731                          */
732                         if (ip_mforward != NULL &&
733                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
734                                 rel_mplock();
735                                 ipstat.ips_cantforward++;
736                                 m_freem(m);
737                                 return;
738                         }
739
740                         /*
741                          * The process-level routing daemon needs to receive
742                          * all multicast IGMP packets, whether or not this
743                          * host belongs to their destination groups.
744                          */
745                         if (ip->ip_p == IPPROTO_IGMP) {
746                                 rel_mplock();
747                                 goto ours;
748                         }
749                         ipstat.ips_forward++;
750                 }
751                 /*
752                  * See if we belong to the destination multicast group on the
753                  * arrival interface.
754                  */
755                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
756                 if (inm == NULL) {
757                         rel_mplock();
758                         ipstat.ips_notmember++;
759                         m_freem(m);
760                         return;
761                 }
762
763                 rel_mplock();
764                 goto ours;
765         }
766         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
767                 goto ours;
768         if (ip->ip_dst.s_addr == INADDR_ANY)
769                 goto ours;
770
771         /*
772          * FAITH(Firewall Aided Internet Translator)
773          */
774         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
775                 if (ip_keepfaith) {
776                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
777                                 goto ours;
778                 }
779                 m_freem(m);
780                 return;
781         }
782
783         /*
784          * Not for us; forward if possible and desirable.
785          */
786         if (!ipforwarding) {
787                 ipstat.ips_cantforward++;
788                 m_freem(m);
789         } else {
790 #ifdef IPSEC
791                 /*
792                  * Enforce inbound IPsec SPD.
793                  */
794                 if (ipsec4_in_reject(m, NULL)) {
795                         ipsecstat.in_polvio++;
796                         goto bad;
797                 }
798 #endif
799 #ifdef FAST_IPSEC
800                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
801                 crit_enter();
802                 if (mtag != NULL) {
803                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
804                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
805                 } else {
806                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
807                                                    IP_FORWARDING, &error);
808                 }
809                 if (sp == NULL) {       /* NB: can happen if error */
810                         crit_exit();
811                         /*XXX error stat???*/
812                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
813                         goto bad;
814                 }
815
816                 /*
817                  * Check security policy against packet attributes.
818                  */
819                 error = ipsec_in_reject(sp, m);
820                 KEY_FREESP(&sp);
821                 crit_exit();
822                 if (error) {
823                         ipstat.ips_cantforward++;
824                         goto bad;
825                 }
826 #endif
827                 ip_forward(m, using_srcrt, next_hop);
828         }
829         return;
830
831 ours:
832
833         /*
834          * IPSTEALTH: Process non-routing options only
835          * if the packet is destined for us.
836          */
837         if (ipstealth &&
838             hlen > sizeof(struct ip) &&
839             ip_dooptions(m, 1, next_hop))
840                 return;
841
842         /* Count the packet in the ip address stats */
843         if (ia != NULL) {
844                 ia->ia_ifa.if_ipackets++;
845                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
846         }
847
848         /*
849          * If offset or IP_MF are set, must reassemble.
850          * Otherwise, nothing need be done.
851          * (We could look in the reassembly queue to see
852          * if the packet was previously fragmented,
853          * but it's not worth the time; just let them time out.)
854          */
855         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
856                 /*
857                  * Attempt reassembly; if it succeeds, proceed.
858                  * ip_reass() will return a different mbuf.
859                  */
860                 m = ip_reass(m);
861                 if (m == NULL)
862                         return;
863                 ip = mtod(m, struct ip *);
864
865                 /* Get the header length of the reassembled packet */
866                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
867
868                 needredispatch = TRUE;
869         } else {
870                 ip->ip_len -= hlen;
871         }
872
873 #ifdef IPSEC
874         /*
875          * enforce IPsec policy checking if we are seeing last header.
876          * note that we do not visit this with protocols with pcb layer
877          * code - like udp/tcp/raw ip.
878          */
879         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
880             ipsec4_in_reject(m, NULL)) {
881                 ipsecstat.in_polvio++;
882                 goto bad;
883         }
884 #endif
885 #if FAST_IPSEC
886         /*
887          * enforce IPsec policy checking if we are seeing last header.
888          * note that we do not visit this with protocols with pcb layer
889          * code - like udp/tcp/raw ip.
890          */
891         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
892                 /*
893                  * Check if the packet has already had IPsec processing
894                  * done.  If so, then just pass it along.  This tag gets
895                  * set during AH, ESP, etc. input handling, before the
896                  * packet is returned to the ip input queue for delivery.
897                  */
898                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
899                 crit_enter();
900                 if (mtag != NULL) {
901                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
902                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
903                 } else {
904                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
905                                                    IP_FORWARDING, &error);
906                 }
907                 if (sp != NULL) {
908                         /*
909                          * Check security policy against packet attributes.
910                          */
911                         error = ipsec_in_reject(sp, m);
912                         KEY_FREESP(&sp);
913                 } else {
914                         /* XXX error stat??? */
915                         error = EINVAL;
916 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
917                         goto bad;
918                 }
919                 crit_exit();
920                 if (error)
921                         goto bad;
922         }
923 #endif /* FAST_IPSEC */
924
925         /*
926          * NOTE: ip_len in host form and adjusted down by hlen for
927          *       protocol processing.
928          */
929         ipstat.ips_delivered++;
930         if (needredispatch) {
931                 struct netmsg_packet *pmsg;
932                 lwkt_port_t port;
933
934                 ip->ip_off = htons(ip->ip_off);
935                 ip->ip_len = htons(ip->ip_len + hlen);
936                 port = ip_mport_in(&m);
937                 if (port == NULL)
938                         return;
939
940                 pmsg = &m->m_hdr.mh_netmsg;
941                 netmsg_init(&pmsg->nm_netmsg, NULL, &netisr_apanic_rport,
942                             MSGF_MPSAFE, transport_processing_handler);
943                 pmsg->nm_packet = m;
944                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
945
946                 ip = mtod(m, struct ip *);
947                 ip->ip_len = ntohs(ip->ip_len) - hlen;
948                 ip->ip_off = ntohs(ip->ip_off);
949                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
950         } else {
951                 transport_processing_oncpu(m, hlen, ip);
952         }
953         return;
954
955 bad:
956         m_freem(m);
957 }
958
959 /*
960  * Take incoming datagram fragment and try to reassemble it into
961  * whole datagram.  If a chain for reassembly of this datagram already
962  * exists, then it is given as fp; otherwise have to make a chain.
963  */
964 struct mbuf *
965 ip_reass(struct mbuf *m)
966 {
967         struct ip *ip = mtod(m, struct ip *);
968         struct mbuf *p = NULL, *q, *nq;
969         struct mbuf *n;
970         struct ipq *fp = NULL;
971         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
972         int i, next;
973         u_short sum;
974
975         /* If maxnipq is 0, never accept fragments. */
976         if (maxnipq == 0) {
977                 ipstat.ips_fragments++;
978                 ipstat.ips_fragdropped++;
979                 m_freem(m);
980                 return NULL;
981         }
982
983         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
984         /*
985          * Look for queue of fragments of this datagram.
986          */
987         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
988                 if (ip->ip_id == fp->ipq_id &&
989                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
990                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
991                     ip->ip_p == fp->ipq_p)
992                         goto found;
993
994         fp = NULL;
995
996         /*
997          * Enforce upper bound on number of fragmented packets
998          * for which we attempt reassembly;
999          * If maxnipq is -1, accept all fragments without limitation.
1000          */
1001         if (nipq > maxnipq && maxnipq > 0) {
1002                 /*
1003                  * drop something from the tail of the current queue
1004                  * before proceeding further
1005                  */
1006                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1007                         for (i = 0; i < IPREASS_NHASH; i++) {
1008                                 if (ipq[i].prev != &ipq[i]) {
1009                                         ipstat.ips_fragtimeout +=
1010                                             ipq[i].prev->ipq_nfrags;
1011                                         ip_freef(ipq[i].prev);
1012                                         break;
1013                                 }
1014                         }
1015                 } else {
1016                         ipstat.ips_fragtimeout +=
1017                             ipq[sum].prev->ipq_nfrags;
1018                         ip_freef(ipq[sum].prev);
1019                 }
1020         }
1021 found:
1022         /*
1023          * Adjust ip_len to not reflect header,
1024          * convert offset of this to bytes.
1025          */
1026         ip->ip_len -= hlen;
1027         if (ip->ip_off & IP_MF) {
1028                 /*
1029                  * Make sure that fragments have a data length
1030                  * that's a non-zero multiple of 8 bytes.
1031                  */
1032                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1033                         ipstat.ips_toosmall++; /* XXX */
1034                         m_freem(m);
1035                         return NULL;
1036                 }
1037                 m->m_flags |= M_FRAG;
1038         } else
1039                 m->m_flags &= ~M_FRAG;
1040         ip->ip_off <<= 3;
1041
1042         ipstat.ips_fragments++;
1043         m->m_pkthdr.header = ip;
1044
1045         /*
1046          * If the hardware has not done csum over this fragment
1047          * then csum_data is not valid at all.
1048          */
1049         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1050             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1051                 m->m_pkthdr.csum_data = 0;
1052                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1053         }
1054
1055         /*
1056          * Presence of header sizes in mbufs
1057          * would confuse code below.
1058          */
1059         m->m_data += hlen;
1060         m->m_len -= hlen;
1061
1062         /*
1063          * If first fragment to arrive, create a reassembly queue.
1064          */
1065         if (fp == NULL) {
1066                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1067                         goto dropfrag;
1068                 insque(fp, &ipq[sum]);
1069                 nipq++;
1070                 fp->ipq_nfrags = 1;
1071                 fp->ipq_ttl = IPFRAGTTL;
1072                 fp->ipq_p = ip->ip_p;
1073                 fp->ipq_id = ip->ip_id;
1074                 fp->ipq_src = ip->ip_src;
1075                 fp->ipq_dst = ip->ip_dst;
1076                 fp->ipq_frags = m;
1077                 m->m_nextpkt = NULL;
1078                 goto inserted;
1079         } else {
1080                 fp->ipq_nfrags++;
1081         }
1082
1083 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1084
1085         /*
1086          * Find a segment which begins after this one does.
1087          */
1088         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1089                 if (GETIP(q)->ip_off > ip->ip_off)
1090                         break;
1091
1092         /*
1093          * If there is a preceding segment, it may provide some of
1094          * our data already.  If so, drop the data from the incoming
1095          * segment.  If it provides all of our data, drop us, otherwise
1096          * stick new segment in the proper place.
1097          *
1098          * If some of the data is dropped from the the preceding
1099          * segment, then it's checksum is invalidated.
1100          */
1101         if (p) {
1102                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1103                 if (i > 0) {
1104                         if (i >= ip->ip_len)
1105                                 goto dropfrag;
1106                         m_adj(m, i);
1107                         m->m_pkthdr.csum_flags = 0;
1108                         ip->ip_off += i;
1109                         ip->ip_len -= i;
1110                 }
1111                 m->m_nextpkt = p->m_nextpkt;
1112                 p->m_nextpkt = m;
1113         } else {
1114                 m->m_nextpkt = fp->ipq_frags;
1115                 fp->ipq_frags = m;
1116         }
1117
1118         /*
1119          * While we overlap succeeding segments trim them or,
1120          * if they are completely covered, dequeue them.
1121          */
1122         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1123              q = nq) {
1124                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1125                 if (i < GETIP(q)->ip_len) {
1126                         GETIP(q)->ip_len -= i;
1127                         GETIP(q)->ip_off += i;
1128                         m_adj(q, i);
1129                         q->m_pkthdr.csum_flags = 0;
1130                         break;
1131                 }
1132                 nq = q->m_nextpkt;
1133                 m->m_nextpkt = nq;
1134                 ipstat.ips_fragdropped++;
1135                 fp->ipq_nfrags--;
1136                 q->m_nextpkt = NULL;
1137                 m_freem(q);
1138         }
1139
1140 inserted:
1141         /*
1142          * Check for complete reassembly and perform frag per packet
1143          * limiting.
1144          *
1145          * Frag limiting is performed here so that the nth frag has
1146          * a chance to complete the packet before we drop the packet.
1147          * As a result, n+1 frags are actually allowed per packet, but
1148          * only n will ever be stored. (n = maxfragsperpacket.)
1149          *
1150          */
1151         next = 0;
1152         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1153                 if (GETIP(q)->ip_off != next) {
1154                         if (fp->ipq_nfrags > maxfragsperpacket) {
1155                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1156                                 ip_freef(fp);
1157                         }
1158                         return (NULL);
1159                 }
1160                 next += GETIP(q)->ip_len;
1161         }
1162         /* Make sure the last packet didn't have the IP_MF flag */
1163         if (p->m_flags & M_FRAG) {
1164                 if (fp->ipq_nfrags > maxfragsperpacket) {
1165                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1166                         ip_freef(fp);
1167                 }
1168                 return (NULL);
1169         }
1170
1171         /*
1172          * Reassembly is complete.  Make sure the packet is a sane size.
1173          */
1174         q = fp->ipq_frags;
1175         ip = GETIP(q);
1176         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1177                 ipstat.ips_toolong++;
1178                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1179                 ip_freef(fp);
1180                 return (NULL);
1181         }
1182
1183         /*
1184          * Concatenate fragments.
1185          */
1186         m = q;
1187         n = m->m_next;
1188         m->m_next = NULL;
1189         m_cat(m, n);
1190         nq = q->m_nextpkt;
1191         q->m_nextpkt = NULL;
1192         for (q = nq; q != NULL; q = nq) {
1193                 nq = q->m_nextpkt;
1194                 q->m_nextpkt = NULL;
1195                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1196                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1197                 m_cat(m, q);
1198         }
1199
1200         /*
1201          * Clean up the 1's complement checksum.  Carry over 16 bits must
1202          * be added back.  This assumes no more then 65535 packet fragments
1203          * were reassembled.  A second carry can also occur (but not a third).
1204          */
1205         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1206                                 (m->m_pkthdr.csum_data >> 16);
1207         if (m->m_pkthdr.csum_data > 0xFFFF)
1208                 m->m_pkthdr.csum_data -= 0xFFFF;
1209
1210         /*
1211          * Create header for new ip packet by
1212          * modifying header of first packet;
1213          * dequeue and discard fragment reassembly header.
1214          * Make header visible.
1215          */
1216         ip->ip_len = next;
1217         ip->ip_src = fp->ipq_src;
1218         ip->ip_dst = fp->ipq_dst;
1219         remque(fp);
1220         nipq--;
1221         mpipe_free(&ipq_mpipe, fp);
1222         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1223         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1224         /* some debugging cruft by sklower, below, will go away soon */
1225         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1226                 int plen = 0;
1227
1228                 for (n = m; n; n = n->m_next)
1229                         plen += n->m_len;
1230                 m->m_pkthdr.len = plen;
1231         }
1232
1233         ipstat.ips_reassembled++;
1234         return (m);
1235
1236 dropfrag:
1237         ipstat.ips_fragdropped++;
1238         if (fp != NULL)
1239                 fp->ipq_nfrags--;
1240         m_freem(m);
1241         return (NULL);
1242
1243 #undef GETIP
1244 }
1245
1246 /*
1247  * Free a fragment reassembly header and all
1248  * associated datagrams.
1249  */
1250 static void
1251 ip_freef(struct ipq *fp)
1252 {
1253         struct mbuf *q;
1254
1255         while (fp->ipq_frags) {
1256                 q = fp->ipq_frags;
1257                 fp->ipq_frags = q->m_nextpkt;
1258                 q->m_nextpkt = NULL;
1259                 m_freem(q);
1260         }
1261         remque(fp);
1262         mpipe_free(&ipq_mpipe, fp);
1263         nipq--;
1264 }
1265
1266 /*
1267  * IP timer processing;
1268  * if a timer expires on a reassembly
1269  * queue, discard it.
1270  */
1271 void
1272 ip_slowtimo(void)
1273 {
1274         struct ipq *fp;
1275         int i;
1276
1277         crit_enter();
1278         for (i = 0; i < IPREASS_NHASH; i++) {
1279                 fp = ipq[i].next;
1280                 if (fp == NULL)
1281                         continue;
1282                 while (fp != &ipq[i]) {
1283                         --fp->ipq_ttl;
1284                         fp = fp->next;
1285                         if (fp->prev->ipq_ttl == 0) {
1286                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1287                                 ip_freef(fp->prev);
1288                         }
1289                 }
1290         }
1291         /*
1292          * If we are over the maximum number of fragments
1293          * (due to the limit being lowered), drain off
1294          * enough to get down to the new limit.
1295          */
1296         if (maxnipq >= 0 && nipq > maxnipq) {
1297                 for (i = 0; i < IPREASS_NHASH; i++) {
1298                         while (nipq > maxnipq &&
1299                                 (ipq[i].next != &ipq[i])) {
1300                                 ipstat.ips_fragdropped +=
1301                                     ipq[i].next->ipq_nfrags;
1302                                 ip_freef(ipq[i].next);
1303                         }
1304                 }
1305         }
1306         ipflow_slowtimo();
1307         crit_exit();
1308 }
1309
1310 /*
1311  * Drain off all datagram fragments.
1312  */
1313 void
1314 ip_drain(void)
1315 {
1316         int i;
1317
1318         for (i = 0; i < IPREASS_NHASH; i++) {
1319                 while (ipq[i].next != &ipq[i]) {
1320                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1321                         ip_freef(ipq[i].next);
1322                 }
1323         }
1324         in_rtqdrain();
1325 }
1326
1327 /*
1328  * Do option processing on a datagram,
1329  * possibly discarding it if bad options are encountered,
1330  * or forwarding it if source-routed.
1331  * The pass argument is used when operating in the IPSTEALTH
1332  * mode to tell what options to process:
1333  * [LS]SRR (pass 0) or the others (pass 1).
1334  * The reason for as many as two passes is that when doing IPSTEALTH,
1335  * non-routing options should be processed only if the packet is for us.
1336  * Returns 1 if packet has been forwarded/freed,
1337  * 0 if the packet should be processed further.
1338  */
1339 static int
1340 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1341 {
1342         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1343         struct ip *ip = mtod(m, struct ip *);
1344         u_char *cp;
1345         struct in_ifaddr *ia;
1346         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1347         boolean_t forward = FALSE;
1348         struct in_addr *sin, dst;
1349         n_time ntime;
1350
1351         dst = ip->ip_dst;
1352         cp = (u_char *)(ip + 1);
1353         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1354         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1355                 opt = cp[IPOPT_OPTVAL];
1356                 if (opt == IPOPT_EOL)
1357                         break;
1358                 if (opt == IPOPT_NOP)
1359                         optlen = 1;
1360                 else {
1361                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1362                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1363                                 goto bad;
1364                         }
1365                         optlen = cp[IPOPT_OLEN];
1366                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1367                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1368                                 goto bad;
1369                         }
1370                 }
1371                 switch (opt) {
1372
1373                 default:
1374                         break;
1375
1376                 /*
1377                  * Source routing with record.
1378                  * Find interface with current destination address.
1379                  * If none on this machine then drop if strictly routed,
1380                  * or do nothing if loosely routed.
1381                  * Record interface address and bring up next address
1382                  * component.  If strictly routed make sure next
1383                  * address is on directly accessible net.
1384                  */
1385                 case IPOPT_LSRR:
1386                 case IPOPT_SSRR:
1387                         if (ipstealth && pass > 0)
1388                                 break;
1389                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1390                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1391                                 goto bad;
1392                         }
1393                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1394                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1395                                 goto bad;
1396                         }
1397                         ipaddr.sin_addr = ip->ip_dst;
1398                         ia = (struct in_ifaddr *)
1399                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1400                         if (ia == NULL) {
1401                                 if (opt == IPOPT_SSRR) {
1402                                         type = ICMP_UNREACH;
1403                                         code = ICMP_UNREACH_SRCFAIL;
1404                                         goto bad;
1405                                 }
1406                                 if (!ip_dosourceroute)
1407                                         goto nosourcerouting;
1408                                 /*
1409                                  * Loose routing, and not at next destination
1410                                  * yet; nothing to do except forward.
1411                                  */
1412                                 break;
1413                         }
1414                         off--;                  /* 0 origin */
1415                         if (off > optlen - (int)sizeof(struct in_addr)) {
1416                                 /*
1417                                  * End of source route.  Should be for us.
1418                                  */
1419                                 if (!ip_acceptsourceroute)
1420                                         goto nosourcerouting;
1421                                 save_rte(m, cp, ip->ip_src);
1422                                 break;
1423                         }
1424                         if (ipstealth)
1425                                 goto dropit;
1426                         if (!ip_dosourceroute) {
1427                                 if (ipforwarding) {
1428                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1429
1430                                         /*
1431                                          * Acting as a router, so generate ICMP
1432                                          */
1433 nosourcerouting:
1434                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1435                                         log(LOG_WARNING,
1436                                             "attempted source route from %s to %s\n",
1437                                             inet_ntoa(ip->ip_src), buf);
1438                                         type = ICMP_UNREACH;
1439                                         code = ICMP_UNREACH_SRCFAIL;
1440                                         goto bad;
1441                                 } else {
1442                                         /*
1443                                          * Not acting as a router,
1444                                          * so silently drop.
1445                                          */
1446 dropit:
1447                                         ipstat.ips_cantforward++;
1448                                         m_freem(m);
1449                                         return (1);
1450                                 }
1451                         }
1452
1453                         /*
1454                          * locate outgoing interface
1455                          */
1456                         memcpy(&ipaddr.sin_addr, cp + off,
1457                             sizeof ipaddr.sin_addr);
1458
1459                         if (opt == IPOPT_SSRR) {
1460 #define INA     struct in_ifaddr *
1461 #define SA      struct sockaddr *
1462                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1463                                                                         == NULL)
1464                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1465                         } else {
1466                                 ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1467                         }
1468                         if (ia == NULL) {
1469                                 type = ICMP_UNREACH;
1470                                 code = ICMP_UNREACH_SRCFAIL;
1471                                 goto bad;
1472                         }
1473                         ip->ip_dst = ipaddr.sin_addr;
1474                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1475                             sizeof(struct in_addr));
1476                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1477                         /*
1478                          * Let ip_intr's mcast routing check handle mcast pkts
1479                          */
1480                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1481                         break;
1482
1483                 case IPOPT_RR:
1484                         if (ipstealth && pass == 0)
1485                                 break;
1486                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1487                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1488                                 goto bad;
1489                         }
1490                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1491                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1492                                 goto bad;
1493                         }
1494                         /*
1495                          * If no space remains, ignore.
1496                          */
1497                         off--;                  /* 0 origin */
1498                         if (off > optlen - (int)sizeof(struct in_addr))
1499                                 break;
1500                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1501                             sizeof ipaddr.sin_addr);
1502                         /*
1503                          * locate outgoing interface; if we're the destination,
1504                          * use the incoming interface (should be same).
1505                          */
1506                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1507                             (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1508                                 type = ICMP_UNREACH;
1509                                 code = ICMP_UNREACH_HOST;
1510                                 goto bad;
1511                         }
1512                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1513                             sizeof(struct in_addr));
1514                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1515                         break;
1516
1517                 case IPOPT_TS:
1518                         if (ipstealth && pass == 0)
1519                                 break;
1520                         code = cp - (u_char *)ip;
1521                         if (optlen < 4 || optlen > 40) {
1522                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1523                                 goto bad;
1524                         }
1525                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1526                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1527                                 goto bad;
1528                         }
1529                         if (off > optlen - (int)sizeof(int32_t)) {
1530                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1531                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1532                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1533                                         goto bad;
1534                                 }
1535                                 break;
1536                         }
1537                         off--;                          /* 0 origin */
1538                         sin = (struct in_addr *)(cp + off);
1539                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1540
1541                         case IPOPT_TS_TSONLY:
1542                                 break;
1543
1544                         case IPOPT_TS_TSANDADDR:
1545                                 if (off + sizeof(n_time) +
1546                                     sizeof(struct in_addr) > optlen) {
1547                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1548                                         goto bad;
1549                                 }
1550                                 ipaddr.sin_addr = dst;
1551                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1552                                                             m->m_pkthdr.rcvif);
1553                                 if (ia == NULL)
1554                                         continue;
1555                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1556                                     sizeof(struct in_addr));
1557                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1558                                 off += sizeof(struct in_addr);
1559                                 break;
1560
1561                         case IPOPT_TS_PRESPEC:
1562                                 if (off + sizeof(n_time) +
1563                                     sizeof(struct in_addr) > optlen) {
1564                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1565                                         goto bad;
1566                                 }
1567                                 memcpy(&ipaddr.sin_addr, sin,
1568                                     sizeof(struct in_addr));
1569                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1570                                         continue;
1571                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1572                                 off += sizeof(struct in_addr);
1573                                 break;
1574
1575                         default:
1576                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1577                                 goto bad;
1578                         }
1579                         ntime = iptime();
1580                         memcpy(cp + off, &ntime, sizeof(n_time));
1581                         cp[IPOPT_OFFSET] += sizeof(n_time);
1582                 }
1583         }
1584         if (forward && ipforwarding) {
1585                 ip_forward(m, TRUE, next_hop);
1586                 return (1);
1587         }
1588         return (0);
1589 bad:
1590         icmp_error(m, type, code, 0, 0);
1591         ipstat.ips_badoptions++;
1592         return (1);
1593 }
1594
1595 /*
1596  * Given address of next destination (final or next hop),
1597  * return internet address info of interface to be used to get there.
1598  */
1599 struct in_ifaddr *
1600 ip_rtaddr(struct in_addr dst, struct route *ro0)
1601 {
1602         struct route sro, *ro;
1603         struct sockaddr_in *sin;
1604         struct in_ifaddr *ia;
1605
1606         if (ro0 != NULL) {
1607                 ro = ro0;
1608         } else {
1609                 bzero(&sro, sizeof(sro));
1610                 ro = &sro;
1611         }
1612
1613         sin = (struct sockaddr_in *)&ro->ro_dst;
1614
1615         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1616                 if (ro->ro_rt != NULL) {
1617                         RTFREE(ro->ro_rt);
1618                         ro->ro_rt = NULL;
1619                 }
1620                 sin->sin_family = AF_INET;
1621                 sin->sin_len = sizeof *sin;
1622                 sin->sin_addr = dst;
1623                 rtalloc_ign(ro, RTF_PRCLONING);
1624         }
1625
1626         if (ro->ro_rt == NULL)
1627                 return (NULL);
1628
1629         ia = ifatoia(ro->ro_rt->rt_ifa);
1630
1631         if (ro == &sro)
1632                 RTFREE(ro->ro_rt);
1633         return ia;
1634 }
1635
1636 /*
1637  * Save incoming source route for use in replies,
1638  * to be picked up later by ip_srcroute if the receiver is interested.
1639  */
1640 static void
1641 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1642 {
1643         struct m_tag *mtag;
1644         struct ip_srcrt_opt *opt;
1645         unsigned olen;
1646
1647         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1648         if (mtag == NULL)
1649                 return;
1650         opt = m_tag_data(mtag);
1651
1652         olen = option[IPOPT_OLEN];
1653 #ifdef DIAGNOSTIC
1654         if (ipprintfs)
1655                 kprintf("save_rte: olen %d\n", olen);
1656 #endif
1657         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1658                 m_tag_free(mtag);
1659                 return;
1660         }
1661         bcopy(option, opt->ip_srcrt.srcopt, olen);
1662         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1663         opt->ip_srcrt.dst = dst;
1664         m_tag_prepend(m, mtag);
1665 }
1666
1667 /*
1668  * Retrieve incoming source route for use in replies,
1669  * in the same form used by setsockopt.
1670  * The first hop is placed before the options, will be removed later.
1671  */
1672 struct mbuf *
1673 ip_srcroute(struct mbuf *m0)
1674 {
1675         struct in_addr *p, *q;
1676         struct mbuf *m;
1677         struct m_tag *mtag;
1678         struct ip_srcrt_opt *opt;
1679
1680         if (m0 == NULL)
1681                 return NULL;
1682
1683         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1684         if (mtag == NULL)
1685                 return NULL;
1686         opt = m_tag_data(mtag);
1687
1688         if (opt->ip_nhops == 0)
1689                 return (NULL);
1690         m = m_get(MB_DONTWAIT, MT_HEADER);
1691         if (m == NULL)
1692                 return (NULL);
1693
1694 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1695
1696         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1697         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1698                    sizeof(struct in_addr) + OPTSIZ;
1699 #ifdef DIAGNOSTIC
1700         if (ipprintfs) {
1701                 kprintf("ip_srcroute: nhops %d mlen %d",
1702                         opt->ip_nhops, m->m_len);
1703         }
1704 #endif
1705
1706         /*
1707          * First save first hop for return route
1708          */
1709         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1710         *(mtod(m, struct in_addr *)) = *p--;
1711 #ifdef DIAGNOSTIC
1712         if (ipprintfs)
1713                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1714 #endif
1715
1716         /*
1717          * Copy option fields and padding (nop) to mbuf.
1718          */
1719         opt->ip_srcrt.nop = IPOPT_NOP;
1720         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1721         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1722             OPTSIZ);
1723         q = (struct in_addr *)(mtod(m, caddr_t) +
1724             sizeof(struct in_addr) + OPTSIZ);
1725 #undef OPTSIZ
1726         /*
1727          * Record return path as an IP source route,
1728          * reversing the path (pointers are now aligned).
1729          */
1730         while (p >= opt->ip_srcrt.route) {
1731 #ifdef DIAGNOSTIC
1732                 if (ipprintfs)
1733                         kprintf(" %x", ntohl(q->s_addr));
1734 #endif
1735                 *q++ = *p--;
1736         }
1737         /*
1738          * Last hop goes to final destination.
1739          */
1740         *q = opt->ip_srcrt.dst;
1741         m_tag_delete(m0, mtag);
1742 #ifdef DIAGNOSTIC
1743         if (ipprintfs)
1744                 kprintf(" %x\n", ntohl(q->s_addr));
1745 #endif
1746         return (m);
1747 }
1748
1749 /*
1750  * Strip out IP options.
1751  */
1752 void
1753 ip_stripoptions(struct mbuf *m)
1754 {
1755         int datalen;
1756         struct ip *ip = mtod(m, struct ip *);
1757         caddr_t opts;
1758         int optlen;
1759
1760         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1761         opts = (caddr_t)(ip + 1);
1762         datalen = m->m_len - (sizeof(struct ip) + optlen);
1763         bcopy(opts + optlen, opts, datalen);
1764         m->m_len -= optlen;
1765         if (m->m_flags & M_PKTHDR)
1766                 m->m_pkthdr.len -= optlen;
1767         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1768 }
1769
1770 u_char inetctlerrmap[PRC_NCMDS] = {
1771         0,              0,              0,              0,
1772         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1773         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1774         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1775         0,              0,              0,              0,
1776         ENOPROTOOPT,    ECONNREFUSED
1777 };
1778
1779 /*
1780  * Forward a packet.  If some error occurs return the sender
1781  * an icmp packet.  Note we can't always generate a meaningful
1782  * icmp message because icmp doesn't have a large enough repertoire
1783  * of codes and types.
1784  *
1785  * If not forwarding, just drop the packet.  This could be confusing
1786  * if ipforwarding was zero but some routing protocol was advancing
1787  * us as a gateway to somewhere.  However, we must let the routing
1788  * protocol deal with that.
1789  *
1790  * The using_srcrt parameter indicates whether the packet is being forwarded
1791  * via a source route.
1792  */
1793 void
1794 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1795 {
1796         struct ip *ip = mtod(m, struct ip *);
1797         struct rtentry *rt;
1798         struct route fwd_ro;
1799         int error, type = 0, code = 0, destmtu = 0;
1800         struct mbuf *mcopy;
1801         n_long dest;
1802         struct in_addr pkt_dst;
1803
1804         dest = INADDR_ANY;
1805         /*
1806          * Cache the destination address of the packet; this may be
1807          * changed by use of 'ipfw fwd'.
1808          */
1809         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1810
1811 #ifdef DIAGNOSTIC
1812         if (ipprintfs)
1813                 kprintf("forward: src %x dst %x ttl %x\n",
1814                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1815 #endif
1816
1817         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1818                 ipstat.ips_cantforward++;
1819                 m_freem(m);
1820                 return;
1821         }
1822         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1823                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1824                 return;
1825         }
1826
1827         bzero(&fwd_ro, sizeof(fwd_ro));
1828         ip_rtaddr(pkt_dst, &fwd_ro);
1829         if (fwd_ro.ro_rt == NULL) {
1830                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1831                 return;
1832         }
1833         rt = fwd_ro.ro_rt;
1834
1835         /*
1836          * Save the IP header and at most 8 bytes of the payload,
1837          * in case we need to generate an ICMP message to the src.
1838          *
1839          * XXX this can be optimized a lot by saving the data in a local
1840          * buffer on the stack (72 bytes at most), and only allocating the
1841          * mbuf if really necessary. The vast majority of the packets
1842          * are forwarded without having to send an ICMP back (either
1843          * because unnecessary, or because rate limited), so we are
1844          * really we are wasting a lot of work here.
1845          *
1846          * We don't use m_copy() because it might return a reference
1847          * to a shared cluster. Both this function and ip_output()
1848          * assume exclusive access to the IP header in `m', so any
1849          * data in a cluster may change before we reach icmp_error().
1850          */
1851         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1852         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1853                 /*
1854                  * It's probably ok if the pkthdr dup fails (because
1855                  * the deep copy of the tag chain failed), but for now
1856                  * be conservative and just discard the copy since
1857                  * code below may some day want the tags.
1858                  */
1859                 m_free(mcopy);
1860                 mcopy = NULL;
1861         }
1862         if (mcopy != NULL) {
1863                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1864                     (int)ip->ip_len);
1865                 mcopy->m_pkthdr.len = mcopy->m_len;
1866                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1867         }
1868
1869         if (!ipstealth)
1870                 ip->ip_ttl -= IPTTLDEC;
1871
1872         /*
1873          * If forwarding packet using same interface that it came in on,
1874          * perhaps should send a redirect to sender to shortcut a hop.
1875          * Only send redirect if source is sending directly to us,
1876          * and if packet was not source routed (or has any options).
1877          * Also, don't send redirect if forwarding using a default route
1878          * or a route modified by a redirect.
1879          */
1880         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1881             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1882             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1883             ipsendredirects && !using_srcrt && next_hop == NULL) {
1884                 u_long src = ntohl(ip->ip_src.s_addr);
1885                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1886
1887                 if (rt_ifa != NULL &&
1888                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1889                         if (rt->rt_flags & RTF_GATEWAY)
1890                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1891                         else
1892                                 dest = pkt_dst.s_addr;
1893                         /*
1894                          * Router requirements says to only send
1895                          * host redirects.
1896                          */
1897                         type = ICMP_REDIRECT;
1898                         code = ICMP_REDIRECT_HOST;
1899 #ifdef DIAGNOSTIC
1900                         if (ipprintfs)
1901                                 kprintf("redirect (%d) to %x\n", code, dest);
1902 #endif
1903                 }
1904         }
1905
1906         error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
1907         if (error == 0) {
1908                 ipstat.ips_forward++;
1909                 if (type == 0) {
1910                         if (mcopy) {
1911                                 ipflow_create(&fwd_ro, mcopy);
1912                                 m_freem(mcopy);
1913                         }
1914                         goto done;
1915                 } else {
1916                         ipstat.ips_redirectsent++;
1917                 }
1918         } else {
1919                 ipstat.ips_cantforward++;
1920         }
1921
1922         if (mcopy == NULL)
1923                 goto done;
1924
1925         /*
1926          * Send ICMP message.
1927          */
1928
1929         switch (error) {
1930
1931         case 0:                         /* forwarded, but need redirect */
1932                 /* type, code set above */
1933                 break;
1934
1935         case ENETUNREACH:               /* shouldn't happen, checked above */
1936         case EHOSTUNREACH:
1937         case ENETDOWN:
1938         case EHOSTDOWN:
1939         default:
1940                 type = ICMP_UNREACH;
1941                 code = ICMP_UNREACH_HOST;
1942                 break;
1943
1944         case EMSGSIZE:
1945                 type = ICMP_UNREACH;
1946                 code = ICMP_UNREACH_NEEDFRAG;
1947 #ifdef IPSEC
1948                 /*
1949                  * If the packet is routed over IPsec tunnel, tell the
1950                  * originator the tunnel MTU.
1951                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1952                  * XXX quickhack!!!
1953                  */
1954                 if (fwd_ro.ro_rt != NULL) {
1955                         struct secpolicy *sp = NULL;
1956                         int ipsecerror;
1957                         int ipsechdr;
1958                         struct route *ro;
1959
1960                         sp = ipsec4_getpolicybyaddr(mcopy,
1961                                                     IPSEC_DIR_OUTBOUND,
1962                                                     IP_FORWARDING,
1963                                                     &ipsecerror);
1964
1965                         if (sp == NULL)
1966                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
1967                         else {
1968                                 /* count IPsec header size */
1969                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1970                                                          IPSEC_DIR_OUTBOUND,
1971                                                          NULL);
1972
1973                                 /*
1974                                  * find the correct route for outer IPv4
1975                                  * header, compute tunnel MTU.
1976                                  *
1977                                  */
1978                                 if (sp->req != NULL && sp->req->sav != NULL &&
1979                                     sp->req->sav->sah != NULL) {
1980                                         ro = &sp->req->sav->sah->sa_route;
1981                                         if (ro->ro_rt != NULL &&
1982                                             ro->ro_rt->rt_ifp != NULL) {
1983                                                 destmtu =
1984                                                     ro->ro_rt->rt_ifp->if_mtu;
1985                                                 destmtu -= ipsechdr;
1986                                         }
1987                                 }
1988
1989                                 key_freesp(sp);
1990                         }
1991                 }
1992 #elif FAST_IPSEC
1993                 /*
1994                  * If the packet is routed over IPsec tunnel, tell the
1995                  * originator the tunnel MTU.
1996                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1997                  * XXX quickhack!!!
1998                  */
1999                 if (fwd_ro.ro_rt != NULL) {
2000                         struct secpolicy *sp = NULL;
2001                         int ipsecerror;
2002                         int ipsechdr;
2003                         struct route *ro;
2004
2005                         sp = ipsec_getpolicybyaddr(mcopy,
2006                                                    IPSEC_DIR_OUTBOUND,
2007                                                    IP_FORWARDING,
2008                                                    &ipsecerror);
2009
2010                         if (sp == NULL)
2011                                 destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2012                         else {
2013                                 /* count IPsec header size */
2014                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2015                                                          IPSEC_DIR_OUTBOUND,
2016                                                          NULL);
2017
2018                                 /*
2019                                  * find the correct route for outer IPv4
2020                                  * header, compute tunnel MTU.
2021                                  */
2022
2023                                 if (sp->req != NULL &&
2024                                     sp->req->sav != NULL &&
2025                                     sp->req->sav->sah != NULL) {
2026                                         ro = &sp->req->sav->sah->sa_route;
2027                                         if (ro->ro_rt != NULL &&
2028                                             ro->ro_rt->rt_ifp != NULL) {
2029                                                 destmtu =
2030                                                     ro->ro_rt->rt_ifp->if_mtu;
2031                                                 destmtu -= ipsechdr;
2032                                         }
2033                                 }
2034
2035                                 KEY_FREESP(&sp);
2036                         }
2037                 }
2038 #else /* !IPSEC && !FAST_IPSEC */
2039                 if (fwd_ro.ro_rt != NULL)
2040                         destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2041 #endif /*IPSEC*/
2042                 ipstat.ips_cantfrag++;
2043                 break;
2044
2045         case ENOBUFS:
2046                 /*
2047                  * A router should not generate ICMP_SOURCEQUENCH as
2048                  * required in RFC1812 Requirements for IP Version 4 Routers.
2049                  * Source quench could be a big problem under DoS attacks,
2050                  * or if the underlying interface is rate-limited.
2051                  * Those who need source quench packets may re-enable them
2052                  * via the net.inet.ip.sendsourcequench sysctl.
2053                  */
2054                 if (!ip_sendsourcequench) {
2055                         m_freem(mcopy);
2056                         goto done;
2057                 } else {
2058                         type = ICMP_SOURCEQUENCH;
2059                         code = 0;
2060                 }
2061                 break;
2062
2063         case EACCES:                    /* ipfw denied packet */
2064                 m_freem(mcopy);
2065                 goto done;
2066         }
2067         icmp_error(mcopy, type, code, dest, destmtu);
2068 done:
2069         if (fwd_ro.ro_rt != NULL)
2070                 RTFREE(fwd_ro.ro_rt);
2071 }
2072
2073 void
2074 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2075                struct mbuf *m)
2076 {
2077         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2078                 struct timeval tv;
2079
2080                 microtime(&tv);
2081                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2082                     SCM_TIMESTAMP, SOL_SOCKET);
2083                 if (*mp)
2084                         mp = &(*mp)->m_next;
2085         }
2086         if (inp->inp_flags & INP_RECVDSTADDR) {
2087                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2088                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2089                 if (*mp)
2090                         mp = &(*mp)->m_next;
2091         }
2092         if (inp->inp_flags & INP_RECVTTL) {
2093                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2094                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2095                 if (*mp)
2096                         mp = &(*mp)->m_next;
2097         }
2098 #ifdef notyet
2099         /* XXX
2100          * Moving these out of udp_input() made them even more broken
2101          * than they already were.
2102          */
2103         /* options were tossed already */
2104         if (inp->inp_flags & INP_RECVOPTS) {
2105                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2106                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2107                 if (*mp)
2108                         mp = &(*mp)->m_next;
2109         }
2110         /* ip_srcroute doesn't do what we want here, need to fix */
2111         if (inp->inp_flags & INP_RECVRETOPTS) {
2112                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2113                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2114                 if (*mp)
2115                         mp = &(*mp)->m_next;
2116         }
2117 #endif
2118         if (inp->inp_flags & INP_RECVIF) {
2119                 struct ifnet *ifp;
2120                 struct sdlbuf {
2121                         struct sockaddr_dl sdl;
2122                         u_char  pad[32];
2123                 } sdlbuf;
2124                 struct sockaddr_dl *sdp;
2125                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2126
2127                 if (((ifp = m->m_pkthdr.rcvif)) &&
2128                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2129                         sdp = IF_LLSOCKADDR(ifp);
2130                         /*
2131                          * Change our mind and don't try copy.
2132                          */
2133                         if ((sdp->sdl_family != AF_LINK) ||
2134                             (sdp->sdl_len > sizeof(sdlbuf))) {
2135                                 goto makedummy;
2136                         }
2137                         bcopy(sdp, sdl2, sdp->sdl_len);
2138                 } else {
2139 makedummy:
2140                         sdl2->sdl_len =
2141                             offsetof(struct sockaddr_dl, sdl_data[0]);
2142                         sdl2->sdl_family = AF_LINK;
2143                         sdl2->sdl_index = 0;
2144                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2145                 }
2146                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2147                         IP_RECVIF, IPPROTO_IP);
2148                 if (*mp)
2149                         mp = &(*mp)->m_next;
2150         }
2151 }
2152
2153 /*
2154  * XXX these routines are called from the upper part of the kernel.
2155  *
2156  * They could also be moved to ip_mroute.c, since all the RSVP
2157  *  handling is done there already.
2158  */
2159 int
2160 ip_rsvp_init(struct socket *so)
2161 {
2162         if (so->so_type != SOCK_RAW ||
2163             so->so_proto->pr_protocol != IPPROTO_RSVP)
2164                 return EOPNOTSUPP;
2165
2166         if (ip_rsvpd != NULL)
2167                 return EADDRINUSE;
2168
2169         ip_rsvpd = so;
2170         /*
2171          * This may seem silly, but we need to be sure we don't over-increment
2172          * the RSVP counter, in case something slips up.
2173          */
2174         if (!ip_rsvp_on) {
2175                 ip_rsvp_on = 1;
2176                 rsvp_on++;
2177         }
2178
2179         return 0;
2180 }
2181
2182 int
2183 ip_rsvp_done(void)
2184 {
2185         ip_rsvpd = NULL;
2186         /*
2187          * This may seem silly, but we need to be sure we don't over-decrement
2188          * the RSVP counter, in case something slips up.
2189          */
2190         if (ip_rsvp_on) {
2191                 ip_rsvp_on = 0;
2192                 rsvp_on--;
2193         }
2194         return 0;
2195 }
2196
2197 void
2198 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2199 {
2200         int off, proto;
2201         __va_list ap;
2202
2203         __va_start(ap, m);
2204         off = __va_arg(ap, int);
2205         proto = __va_arg(ap, int);
2206         __va_end(ap);
2207
2208         if (rsvp_input_p) { /* call the real one if loaded */
2209                 rsvp_input_p(m, off, proto);
2210                 return;
2211         }
2212
2213         /* Can still get packets with rsvp_on = 0 if there is a local member
2214          * of the group to which the RSVP packet is addressed.  But in this
2215          * case we want to throw the packet away.
2216          */
2217
2218         if (!rsvp_on) {
2219                 m_freem(m);
2220                 return;
2221         }
2222
2223         if (ip_rsvpd != NULL) {
2224                 rip_input(m, off, proto);
2225                 return;
2226         }
2227         /* Drop the packet */
2228         m_freem(m);
2229 }