56e5878b92637dee669b4f8166db3b49062e96a4
[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 1.106 2008/09/13 12:57:07 sephe Exp $
69  */
70
71 #define _IP_VHL
72
73 #include "opt_bootp.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipdn.h"
76 #include "opt_ipdivert.h"
77 #include "opt_ipfilter.h"
78 #include "opt_ipstealth.h"
79 #include "opt_ipsec.h"
80
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/mpipe.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/time.h>
90 #include <sys/globaldata.h>
91 #include <sys/thread.h>
92 #include <sys/kernel.h>
93 #include <sys/syslog.h>
94 #include <sys/sysctl.h>
95 #include <sys/in_cksum.h>
96
97 #include <machine/stdarg.h>
98
99 #include <net/if.h>
100 #include <net/if_types.h>
101 #include <net/if_var.h>
102 #include <net/if_dl.h>
103 #include <net/pfil.h>
104 #include <net/route.h>
105 #include <net/netisr.h>
106
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/in_var.h>
110 #include <netinet/ip.h>
111 #include <netinet/in_pcb.h>
112 #include <netinet/ip_var.h>
113 #include <netinet/ip_icmp.h>
114 #include <netinet/ip_divert.h>
115
116 #include <sys/thread2.h>
117 #include <sys/msgport2.h>
118 #include <net/netmsg2.h>
119
120 #include <sys/socketvar.h>
121
122 #include <net/ipfw/ip_fw.h>
123 #include <net/dummynet/ip_dummynet.h>
124
125 #ifdef IPSEC
126 #include <netinet6/ipsec.h>
127 #include <netproto/key/key.h>
128 #endif
129
130 #ifdef FAST_IPSEC
131 #include <netproto/ipsec/ipsec.h>
132 #include <netproto/ipsec/key.h>
133 #endif
134
135 int rsvp_on = 0;
136 static int ip_rsvp_on;
137 struct socket *ip_rsvpd;
138
139 int ipforwarding = 0;
140 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
141     &ipforwarding, 0, "Enable IP forwarding between interfaces");
142
143 static int ipsendredirects = 1; /* XXX */
144 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
145     &ipsendredirects, 0, "Enable sending IP redirects");
146
147 int ip_defttl = IPDEFTTL;
148 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
149     &ip_defttl, 0, "Maximum TTL on IP packets");
150
151 static int ip_dosourceroute = 0;
152 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
153     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
154
155 static int ip_acceptsourceroute = 0;
156 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
157     CTLFLAG_RW, &ip_acceptsourceroute, 0,
158     "Enable accepting source routed IP packets");
159
160 static int ip_keepfaith = 0;
161 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
162     &ip_keepfaith, 0,
163     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
164
165 static int nipq = 0;    /* total # of reass queues */
166 static int maxnipq;
167 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
168     &maxnipq, 0,
169     "Maximum number of IPv4 fragment reassembly queue entries");
170
171 static int maxfragsperpacket;
172 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
173     &maxfragsperpacket, 0,
174     "Maximum number of IPv4 fragments allowed per packet");
175
176 static int ip_sendsourcequench = 0;
177 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
178     &ip_sendsourcequench, 0,
179     "Enable the transmission of source quench packets");
180
181 int ip_do_randomid = 1;
182 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
183     &ip_do_randomid, 0,
184     "Assign random ip_id values");      
185 /*
186  * XXX - Setting ip_checkinterface mostly implements the receive side of
187  * the Strong ES model described in RFC 1122, but since the routing table
188  * and transmit implementation do not implement the Strong ES model,
189  * setting this to 1 results in an odd hybrid.
190  *
191  * XXX - ip_checkinterface currently must be disabled if you use ipnat
192  * to translate the destination address to another local interface.
193  *
194  * XXX - ip_checkinterface must be disabled if you add IP aliases
195  * to the loopback interface instead of the interface where the
196  * packets for those addresses are received.
197  */
198 static int ip_checkinterface = 0;
199 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
200     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
201
202 #ifdef DIAGNOSTIC
203 static int ipprintfs = 0;
204 #endif
205
206 extern  struct domain inetdomain;
207 extern  struct protosw inetsw[];
208 u_char  ip_protox[IPPROTO_MAX];
209 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
210 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
211                                                 /* inet addr hash table */
212 u_long  in_ifaddrhmask;                         /* mask for hash table */
213
214 struct ip_stats ipstats_percpu[MAXCPU];
215 #ifdef SMP
216 static int
217 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
218 {
219         int cpu, error = 0;
220
221         for (cpu = 0; cpu < ncpus; ++cpu) {
222                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
223                                         sizeof(struct ip_stats))))
224                         break;
225                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
226                                        sizeof(struct ip_stats))))
227                         break;
228         }
229
230         return (error);
231 }
232 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
233     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
234 #else
235 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
236     &ipstat, ip_stats, "IP statistics");
237 #endif
238
239 /* Packet reassembly stuff */
240 #define IPREASS_NHASH_LOG2      6
241 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
242 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
243 #define IPREASS_HASH(x,y)                                               \
244     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
245
246 static struct ipq ipq[IPREASS_NHASH];
247
248 #ifdef IPCTL_DEFMTU
249 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
250     &ip_mtu, 0, "Default MTU");
251 #endif
252
253 #ifdef IPSTEALTH
254 static int ipstealth = 0;
255 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
256 #else
257 static const int ipstealth = 0;
258 #endif
259
260 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
261
262 struct pfil_head inet_pfil_hook;
263
264 /*
265  * struct ip_srcrt_opt is used to store packet state while it travels
266  * through the stack.
267  *
268  * XXX Note that the code even makes assumptions on the size and
269  * alignment of fields inside struct ip_srcrt so e.g. adding some
270  * fields will break the code.  This needs to be fixed.
271  *
272  * We need to save the IP options in case a protocol wants to respond
273  * to an incoming packet over the same route if the packet got here
274  * using IP source routing.  This allows connection establishment and
275  * maintenance when the remote end is on a network that is not known
276  * to us.
277  */
278 struct ip_srcrt {
279         struct  in_addr dst;                    /* final destination */
280         char    nop;                            /* one NOP to align */
281         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
282         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
283 };
284
285 struct ip_srcrt_opt {
286         int             ip_nhops;
287         struct ip_srcrt ip_srcrt;
288 };
289
290 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
291 static struct malloc_pipe ipq_mpipe;
292
293 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
294 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
295 static void             ip_freef(struct ipq *);
296 static void             ip_input_handler(struct netmsg *);
297
298 /*
299  * IP initialization: fill in IP protocol switch table.
300  * All protocols not implemented in kernel go to raw IP protocol handler.
301  */
302 void
303 ip_init(void)
304 {
305         struct protosw *pr;
306         int i;
307 #ifdef SMP
308         int cpu;
309 #endif
310
311         /*
312          * Make sure we can handle a reasonable number of fragments but
313          * cap it at 4000 (XXX).
314          */
315         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
316                     IFQ_MAXLEN, 4000, 0, NULL);
317         for (i = 0; i < ncpus; ++i) {
318                 TAILQ_INIT(&in_ifaddrheads[i]);
319                 in_ifaddrhashtbls[i] =
320                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
321         }
322         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
323         if (pr == NULL)
324                 panic("ip_init");
325         for (i = 0; i < IPPROTO_MAX; i++)
326                 ip_protox[i] = pr - inetsw;
327         for (pr = inetdomain.dom_protosw;
328              pr < inetdomain.dom_protoswNPROTOSW; pr++)
329                 if (pr->pr_domain->dom_family == PF_INET &&
330                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
331                         ip_protox[pr->pr_protocol] = pr - inetsw;
332
333         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
334         inet_pfil_hook.ph_af = AF_INET;
335         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
336                 kprintf("%s: WARNING: unable to register pfil hook, "
337                         "error %d\n", __func__, i);
338         }
339
340         for (i = 0; i < IPREASS_NHASH; i++)
341                 ipq[i].next = ipq[i].prev = &ipq[i];
342
343         maxnipq = nmbclusters / 32;
344         maxfragsperpacket = 16;
345
346         ip_id = time_second & 0xffff;
347
348         /*
349          * Initialize IP statistics counters for each CPU.
350          *
351          */
352 #ifdef SMP
353         for (cpu = 0; cpu < ncpus; ++cpu) {
354                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
355         }
356 #else
357         bzero(&ipstat, sizeof(struct ip_stats));
358 #endif
359
360         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
361 }
362
363 /*
364  * XXX watch out this one. It is perhaps used as a cache for
365  * the most recently used route ? it is cleared in in_addroute()
366  * when a new route is successfully created.
367  */
368 struct route ipforward_rt[MAXCPU];
369
370 /* Do transport protocol processing. */
371 static void
372 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
373 {
374         /*
375          * Switch out to protocol's input routine.
376          */
377         (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
378 }
379
380 static void
381 transport_processing_handler(netmsg_t netmsg)
382 {
383         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
384         struct ip *ip;
385         int hlen;
386
387         ip = mtod(pmsg->nm_packet, struct ip *);
388         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
389
390         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
391         /* netmsg was embedded in the mbuf, do not reply! */
392 }
393
394 static void
395 ip_input_handler(struct netmsg *msg0)
396 {
397         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
398
399         ip_input(m);
400         /* msg0 was embedded in the mbuf, do not reply! */
401 }
402
403 /*
404  * IP input routine.  Checksum and byte swap header.  If fragmented
405  * try to reassemble.  Process options.  Pass to next level.
406  */
407 void
408 ip_input(struct mbuf *m)
409 {
410         struct ip *ip;
411         struct in_ifaddr *ia = NULL;
412         struct in_ifaddr_container *iac;
413         int hlen, checkif;
414         u_short sum;
415         struct in_addr pkt_dst;
416         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
417         boolean_t needredispatch = FALSE;
418         struct in_addr odst;                    /* original dst address(NAT) */
419         struct m_tag *mtag;
420         struct sockaddr_in *next_hop = NULL;
421 #ifdef FAST_IPSEC
422         struct tdb_ident *tdbi;
423         struct secpolicy *sp;
424         int error;
425 #endif
426
427         M_ASSERTPKTHDR(m);
428
429         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
430                 /* Next hop */
431                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
432                 KKASSERT(mtag != NULL);
433                 next_hop = m_tag_data(mtag);
434         }
435
436         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
437                 /* dummynet already filtered us */
438                 ip = mtod(m, struct ip *);
439                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
440                 goto iphack;
441         }
442
443         ipstat.ips_total++;
444
445         /* length checks already done in ip_demux() */
446         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
447
448         ip = mtod(m, struct ip *);
449
450         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
451                 ipstat.ips_badvers++;
452                 goto bad;
453         }
454
455         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
456         /* length checks already done in ip_demux() */
457         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
458         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
459
460         /* 127/8 must not appear on wire - RFC1122 */
461         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
462             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
463                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
464                         ipstat.ips_badaddr++;
465                         goto bad;
466                 }
467         }
468
469         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
470                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
471         } else {
472                 if (hlen == sizeof(struct ip)) {
473                         sum = in_cksum_hdr(ip);
474                 } else {
475                         sum = in_cksum(m, hlen);
476                 }
477         }
478         if (sum != 0) {
479                 ipstat.ips_badsum++;
480                 goto bad;
481         }
482
483 #ifdef ALTQ
484         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
485                 /* packet is dropped by traffic conditioner */
486                 return;
487         }
488 #endif
489         /*
490          * Convert fields to host representation.
491          */
492         ip->ip_len = ntohs(ip->ip_len);
493         if (ip->ip_len < hlen) {
494                 ipstat.ips_badlen++;
495                 goto bad;
496         }
497         ip->ip_off = ntohs(ip->ip_off);
498
499         /*
500          * Check that the amount of data in the buffers
501          * is as at least much as the IP header would have us expect.
502          * Trim mbufs if longer than we expect.
503          * Drop packet if shorter than we expect.
504          */
505         if (m->m_pkthdr.len < ip->ip_len) {
506                 ipstat.ips_tooshort++;
507                 goto bad;
508         }
509         if (m->m_pkthdr.len > ip->ip_len) {
510                 if (m->m_len == m->m_pkthdr.len) {
511                         m->m_len = ip->ip_len;
512                         m->m_pkthdr.len = ip->ip_len;
513                 } else
514                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
515         }
516 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
517         /*
518          * Bypass packet filtering for packets from a tunnel (gif).
519          */
520         if (ipsec_gethist(m, NULL))
521                 goto pass;
522 #endif
523
524         /*
525          * IpHack's section.
526          * Right now when no processing on packet has done
527          * and it is still fresh out of network we do our black
528          * deals with it.
529          * - Firewall: deny/allow/divert
530          * - Xlate: translate packet's addr/port (NAT).
531          * - Pipe: pass pkt through dummynet.
532          * - Wrap: fake packet's addr/port <unimpl.>
533          * - Encapsulate: put it in another IP and send out. <unimp.>
534          */
535
536 iphack:
537         /*
538          * If we've been forwarded from the output side, then
539          * skip the firewall a second time
540          */
541         if (next_hop != NULL)
542                 goto ours;
543
544         /*
545          * Run through list of hooks for input packets.
546          *
547          * NB: Beware of the destination address changing (e.g.
548          *     by NAT rewriting). When this happens, tell
549          *     ip_forward to do the right thing.
550          */
551         if (pfil_has_hooks(&inet_pfil_hook)) {
552                 odst = ip->ip_dst;
553                 if (pfil_run_hooks(&inet_pfil_hook, &m,
554                     m->m_pkthdr.rcvif, PFIL_IN)) {
555                         return;
556                 }
557                 if (m == NULL)                  /* consumed by filter */
558                         return;
559                 ip = mtod(m, struct ip *);
560                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
561                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
562         }
563
564         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
565                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
566                 KKASSERT(mtag != NULL);
567                 next_hop = m_tag_data(mtag);
568         }
569         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
570                 ip_dn_queue(m);
571                 return;
572         }
573         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
574                 needredispatch = TRUE;
575                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
576         }
577
578 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
579 pass:
580 #endif
581         /*
582          * Process options and, if not destined for us,
583          * ship it on.  ip_dooptions returns 1 when an
584          * error was detected (causing an icmp message
585          * to be sent and the original packet to be freed).
586          */
587         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
588                 return;
589
590         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
591          * matter if it is destined to another node, or whether it is
592          * a multicast one, RSVP wants it! and prevents it from being forwarded
593          * anywhere else. Also checks if the rsvp daemon is running before
594          * grabbing the packet.
595          */
596         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
597                 goto ours;
598
599         /*
600          * Check our list of addresses, to see if the packet is for us.
601          * If we don't have any addresses, assume any unicast packet
602          * we receive might be for us (and let the upper layers deal
603          * with it).
604          */
605         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
606             !(m->m_flags & (M_MCAST | M_BCAST)))
607                 goto ours;
608
609         /*
610          * Cache the destination address of the packet; this may be
611          * changed by use of 'ipfw fwd'.
612          */
613         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
614
615         /*
616          * Enable a consistency check between the destination address
617          * and the arrival interface for a unicast packet (the RFC 1122
618          * strong ES model) if IP forwarding is disabled and the packet
619          * is not locally generated and the packet is not subject to
620          * 'ipfw fwd'.
621          *
622          * XXX - Checking also should be disabled if the destination
623          * address is ipnat'ed to a different interface.
624          *
625          * XXX - Checking is incompatible with IP aliases added
626          * to the loopback interface instead of the interface where
627          * the packets are received.
628          */
629         checkif = ip_checkinterface &&
630                   !ipforwarding &&
631                   m->m_pkthdr.rcvif != NULL &&
632                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
633                   next_hop == NULL;
634
635         /*
636          * Check for exact addresses in the hash bucket.
637          */
638         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
639                 ia = iac->ia;
640
641                 /*
642                  * If the address matches, verify that the packet
643                  * arrived via the correct interface if checking is
644                  * enabled.
645                  */
646                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
647                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
648                         goto ours;
649         }
650         ia = NULL;
651
652         /*
653          * Check for broadcast addresses.
654          *
655          * Only accept broadcast packets that arrive via the matching
656          * interface.  Reception of forwarded directed broadcasts would
657          * be handled via ip_forward() and ether_output() with the loopback
658          * into the stack for SIMPLEX interfaces handled by ether_output().
659          */
660         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
661                 struct ifaddr_container *ifac;
662
663                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
664                               ifa_link) {
665                         struct ifaddr *ifa = ifac->ifa;
666
667                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
668                                 continue;
669                         if (ifa->ifa_addr->sa_family != AF_INET)
670                                 continue;
671                         ia = ifatoia(ifa);
672                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
673                                                                 pkt_dst.s_addr)
674                                 goto ours;
675                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
676                                 goto ours;
677 #ifdef BOOTP_COMPAT
678                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
679                                 goto ours;
680 #endif
681                 }
682         }
683         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
684                 struct in_multi *inm;
685
686                 if (ip_mrouter != NULL) {
687                         /*
688                          * If we are acting as a multicast router, all
689                          * incoming multicast packets are passed to the
690                          * kernel-level multicast forwarding function.
691                          * The packet is returned (relatively) intact; if
692                          * ip_mforward() returns a non-zero value, the packet
693                          * must be discarded, else it may be accepted below.
694                          */
695                         if (ip_mforward != NULL &&
696                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
697                                 ipstat.ips_cantforward++;
698                                 m_freem(m);
699                                 return;
700                         }
701
702                         /*
703                          * The process-level routing daemon needs to receive
704                          * all multicast IGMP packets, whether or not this
705                          * host belongs to their destination groups.
706                          */
707                         if (ip->ip_p == IPPROTO_IGMP)
708                                 goto ours;
709                         ipstat.ips_forward++;
710                 }
711                 /*
712                  * See if we belong to the destination multicast group on the
713                  * arrival interface.
714                  */
715                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
716                 if (inm == NULL) {
717                         ipstat.ips_notmember++;
718                         m_freem(m);
719                         return;
720                 }
721                 goto ours;
722         }
723         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
724                 goto ours;
725         if (ip->ip_dst.s_addr == INADDR_ANY)
726                 goto ours;
727
728         /*
729          * FAITH(Firewall Aided Internet Translator)
730          */
731         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
732                 if (ip_keepfaith) {
733                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
734                                 goto ours;
735                 }
736                 m_freem(m);
737                 return;
738         }
739
740         /*
741          * Not for us; forward if possible and desirable.
742          */
743         if (!ipforwarding) {
744                 ipstat.ips_cantforward++;
745                 m_freem(m);
746         } else {
747 #ifdef IPSEC
748                 /*
749                  * Enforce inbound IPsec SPD.
750                  */
751                 if (ipsec4_in_reject(m, NULL)) {
752                         ipsecstat.in_polvio++;
753                         goto bad;
754                 }
755 #endif
756 #ifdef FAST_IPSEC
757                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
758                 crit_enter();
759                 if (mtag != NULL) {
760                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
761                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
762                 } else {
763                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
764                                                    IP_FORWARDING, &error);
765                 }
766                 if (sp == NULL) {       /* NB: can happen if error */
767                         crit_exit();
768                         /*XXX error stat???*/
769                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
770                         goto bad;
771                 }
772
773                 /*
774                  * Check security policy against packet attributes.
775                  */
776                 error = ipsec_in_reject(sp, m);
777                 KEY_FREESP(&sp);
778                 crit_exit();
779                 if (error) {
780                         ipstat.ips_cantforward++;
781                         goto bad;
782                 }
783 #endif
784                 ip_forward(m, using_srcrt, next_hop);
785         }
786         return;
787
788 ours:
789
790         /*
791          * IPSTEALTH: Process non-routing options only
792          * if the packet is destined for us.
793          */
794         if (ipstealth &&
795             hlen > sizeof(struct ip) &&
796             ip_dooptions(m, 1, next_hop))
797                 return;
798
799         /* Count the packet in the ip address stats */
800         if (ia != NULL) {
801                 ia->ia_ifa.if_ipackets++;
802                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
803         }
804
805         /*
806          * If offset or IP_MF are set, must reassemble.
807          * Otherwise, nothing need be done.
808          * (We could look in the reassembly queue to see
809          * if the packet was previously fragmented,
810          * but it's not worth the time; just let them time out.)
811          */
812         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
813                 /*
814                  * Attempt reassembly; if it succeeds, proceed.
815                  * ip_reass() will return a different mbuf.
816                  */
817                 m = ip_reass(m);
818                 if (m == NULL)
819                         return;
820                 ip = mtod(m, struct ip *);
821
822                 /* Get the header length of the reassembled packet */
823                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
824
825                 needredispatch = TRUE;
826         } else {
827                 ip->ip_len -= hlen;
828         }
829
830 #ifdef IPSEC
831         /*
832          * enforce IPsec policy checking if we are seeing last header.
833          * note that we do not visit this with protocols with pcb layer
834          * code - like udp/tcp/raw ip.
835          */
836         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
837             ipsec4_in_reject(m, NULL)) {
838                 ipsecstat.in_polvio++;
839                 goto bad;
840         }
841 #endif
842 #if FAST_IPSEC
843         /*
844          * enforce IPsec policy checking if we are seeing last header.
845          * note that we do not visit this with protocols with pcb layer
846          * code - like udp/tcp/raw ip.
847          */
848         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
849                 /*
850                  * Check if the packet has already had IPsec processing
851                  * done.  If so, then just pass it along.  This tag gets
852                  * set during AH, ESP, etc. input handling, before the
853                  * packet is returned to the ip input queue for delivery.
854                  */
855                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
856                 crit_enter();
857                 if (mtag != NULL) {
858                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
859                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
860                 } else {
861                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
862                                                    IP_FORWARDING, &error);
863                 }
864                 if (sp != NULL) {
865                         /*
866                          * Check security policy against packet attributes.
867                          */
868                         error = ipsec_in_reject(sp, m);
869                         KEY_FREESP(&sp);
870                 } else {
871                         /* XXX error stat??? */
872                         error = EINVAL;
873 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
874                         goto bad;
875                 }
876                 crit_exit();
877                 if (error)
878                         goto bad;
879         }
880 #endif /* FAST_IPSEC */
881
882         ipstat.ips_delivered++;
883         if (needredispatch) {
884                 struct netmsg_packet *pmsg;
885                 lwkt_port_t port;
886
887                 ip->ip_off = htons(ip->ip_off);
888                 ip->ip_len = htons(ip->ip_len);
889                 port = ip_mport_in(&m);
890                 if (port == NULL)
891                         return;
892
893                 pmsg = &m->m_hdr.mh_netmsg;
894                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
895                             transport_processing_handler);
896                 pmsg->nm_packet = m;
897                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
898
899                 ip = mtod(m, struct ip *);
900                 ip->ip_len = ntohs(ip->ip_len);
901                 ip->ip_off = ntohs(ip->ip_off);
902                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
903         } else {
904                 transport_processing_oncpu(m, hlen, ip);
905         }
906         return;
907
908 bad:
909         m_freem(m);
910 }
911
912 /*
913  * Take incoming datagram fragment and try to reassemble it into
914  * whole datagram.  If a chain for reassembly of this datagram already
915  * exists, then it is given as fp; otherwise have to make a chain.
916  */
917 struct mbuf *
918 ip_reass(struct mbuf *m)
919 {
920         struct ip *ip = mtod(m, struct ip *);
921         struct mbuf *p = NULL, *q, *nq;
922         struct mbuf *n;
923         struct ipq *fp = NULL;
924         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
925         int i, next;
926         u_short sum;
927
928         /* If maxnipq is 0, never accept fragments. */
929         if (maxnipq == 0) {
930                 ipstat.ips_fragments++;
931                 ipstat.ips_fragdropped++;
932                 m_freem(m);
933                 return NULL;
934         }
935
936         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
937         /*
938          * Look for queue of fragments of this datagram.
939          */
940         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
941                 if (ip->ip_id == fp->ipq_id &&
942                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
943                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
944                     ip->ip_p == fp->ipq_p)
945                         goto found;
946
947         fp = NULL;
948
949         /*
950          * Enforce upper bound on number of fragmented packets
951          * for which we attempt reassembly;
952          * If maxnipq is -1, accept all fragments without limitation.
953          */
954         if (nipq > maxnipq && maxnipq > 0) {
955                 /*
956                  * drop something from the tail of the current queue
957                  * before proceeding further
958                  */
959                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
960                         for (i = 0; i < IPREASS_NHASH; i++) {
961                                 if (ipq[i].prev != &ipq[i]) {
962                                         ipstat.ips_fragtimeout +=
963                                             ipq[i].prev->ipq_nfrags;
964                                         ip_freef(ipq[i].prev);
965                                         break;
966                                 }
967                         }
968                 } else {
969                         ipstat.ips_fragtimeout +=
970                             ipq[sum].prev->ipq_nfrags;
971                         ip_freef(ipq[sum].prev);
972                 }
973         }
974 found:
975         /*
976          * Adjust ip_len to not reflect header,
977          * convert offset of this to bytes.
978          */
979         ip->ip_len -= hlen;
980         if (ip->ip_off & IP_MF) {
981                 /*
982                  * Make sure that fragments have a data length
983                  * that's a non-zero multiple of 8 bytes.
984                  */
985                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
986                         ipstat.ips_toosmall++; /* XXX */
987                         m_freem(m);
988                         return NULL;
989                 }
990                 m->m_flags |= M_FRAG;
991         } else
992                 m->m_flags &= ~M_FRAG;
993         ip->ip_off <<= 3;
994
995         ipstat.ips_fragments++;
996         m->m_pkthdr.header = ip;
997
998         /*
999          * If the hardware has not done csum over this fragment
1000          * then csum_data is not valid at all.
1001          */
1002         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1003             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1004                 m->m_pkthdr.csum_data = 0;
1005                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1006         }
1007
1008         /*
1009          * Presence of header sizes in mbufs
1010          * would confuse code below.
1011          */
1012         m->m_data += hlen;
1013         m->m_len -= hlen;
1014
1015         /*
1016          * If first fragment to arrive, create a reassembly queue.
1017          */
1018         if (fp == NULL) {
1019                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1020                         goto dropfrag;
1021                 insque(fp, &ipq[sum]);
1022                 nipq++;
1023                 fp->ipq_nfrags = 1;
1024                 fp->ipq_ttl = IPFRAGTTL;
1025                 fp->ipq_p = ip->ip_p;
1026                 fp->ipq_id = ip->ip_id;
1027                 fp->ipq_src = ip->ip_src;
1028                 fp->ipq_dst = ip->ip_dst;
1029                 fp->ipq_frags = m;
1030                 m->m_nextpkt = NULL;
1031                 goto inserted;
1032         } else {
1033                 fp->ipq_nfrags++;
1034         }
1035
1036 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1037
1038         /*
1039          * Find a segment which begins after this one does.
1040          */
1041         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1042                 if (GETIP(q)->ip_off > ip->ip_off)
1043                         break;
1044
1045         /*
1046          * If there is a preceding segment, it may provide some of
1047          * our data already.  If so, drop the data from the incoming
1048          * segment.  If it provides all of our data, drop us, otherwise
1049          * stick new segment in the proper place.
1050          *
1051          * If some of the data is dropped from the the preceding
1052          * segment, then it's checksum is invalidated.
1053          */
1054         if (p) {
1055                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1056                 if (i > 0) {
1057                         if (i >= ip->ip_len)
1058                                 goto dropfrag;
1059                         m_adj(m, i);
1060                         m->m_pkthdr.csum_flags = 0;
1061                         ip->ip_off += i;
1062                         ip->ip_len -= i;
1063                 }
1064                 m->m_nextpkt = p->m_nextpkt;
1065                 p->m_nextpkt = m;
1066         } else {
1067                 m->m_nextpkt = fp->ipq_frags;
1068                 fp->ipq_frags = m;
1069         }
1070
1071         /*
1072          * While we overlap succeeding segments trim them or,
1073          * if they are completely covered, dequeue them.
1074          */
1075         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1076              q = nq) {
1077                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1078                 if (i < GETIP(q)->ip_len) {
1079                         GETIP(q)->ip_len -= i;
1080                         GETIP(q)->ip_off += i;
1081                         m_adj(q, i);
1082                         q->m_pkthdr.csum_flags = 0;
1083                         break;
1084                 }
1085                 nq = q->m_nextpkt;
1086                 m->m_nextpkt = nq;
1087                 ipstat.ips_fragdropped++;
1088                 fp->ipq_nfrags--;
1089                 q->m_nextpkt = NULL;
1090                 m_freem(q);
1091         }
1092
1093 inserted:
1094         /*
1095          * Check for complete reassembly and perform frag per packet
1096          * limiting.
1097          *
1098          * Frag limiting is performed here so that the nth frag has
1099          * a chance to complete the packet before we drop the packet.
1100          * As a result, n+1 frags are actually allowed per packet, but
1101          * only n will ever be stored. (n = maxfragsperpacket.)
1102          *
1103          */
1104         next = 0;
1105         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1106                 if (GETIP(q)->ip_off != next) {
1107                         if (fp->ipq_nfrags > maxfragsperpacket) {
1108                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1109                                 ip_freef(fp);
1110                         }
1111                         return (NULL);
1112                 }
1113                 next += GETIP(q)->ip_len;
1114         }
1115         /* Make sure the last packet didn't have the IP_MF flag */
1116         if (p->m_flags & M_FRAG) {
1117                 if (fp->ipq_nfrags > maxfragsperpacket) {
1118                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1119                         ip_freef(fp);
1120                 }
1121                 return (NULL);
1122         }
1123
1124         /*
1125          * Reassembly is complete.  Make sure the packet is a sane size.
1126          */
1127         q = fp->ipq_frags;
1128         ip = GETIP(q);
1129         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1130                 ipstat.ips_toolong++;
1131                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1132                 ip_freef(fp);
1133                 return (NULL);
1134         }
1135
1136         /*
1137          * Concatenate fragments.
1138          */
1139         m = q;
1140         n = m->m_next;
1141         m->m_next = NULL;
1142         m_cat(m, n);
1143         nq = q->m_nextpkt;
1144         q->m_nextpkt = NULL;
1145         for (q = nq; q != NULL; q = nq) {
1146                 nq = q->m_nextpkt;
1147                 q->m_nextpkt = NULL;
1148                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1149                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1150                 m_cat(m, q);
1151         }
1152
1153         /*
1154          * Clean up the 1's complement checksum.  Carry over 16 bits must
1155          * be added back.  This assumes no more then 65535 packet fragments
1156          * were reassembled.  A second carry can also occur (but not a third).
1157          */
1158         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1159                                 (m->m_pkthdr.csum_data >> 16);
1160         if (m->m_pkthdr.csum_data > 0xFFFF)
1161                 m->m_pkthdr.csum_data -= 0xFFFF;
1162
1163         /*
1164          * Create header for new ip packet by
1165          * modifying header of first packet;
1166          * dequeue and discard fragment reassembly header.
1167          * Make header visible.
1168          */
1169         ip->ip_len = next;
1170         ip->ip_src = fp->ipq_src;
1171         ip->ip_dst = fp->ipq_dst;
1172         remque(fp);
1173         nipq--;
1174         mpipe_free(&ipq_mpipe, fp);
1175         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1176         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1177         /* some debugging cruft by sklower, below, will go away soon */
1178         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1179                 int plen = 0;
1180
1181                 for (n = m; n; n = n->m_next)
1182                         plen += n->m_len;
1183                 m->m_pkthdr.len = plen;
1184         }
1185
1186         ipstat.ips_reassembled++;
1187         return (m);
1188
1189 dropfrag:
1190         ipstat.ips_fragdropped++;
1191         if (fp != NULL)
1192                 fp->ipq_nfrags--;
1193         m_freem(m);
1194         return (NULL);
1195
1196 #undef GETIP
1197 }
1198
1199 /*
1200  * Free a fragment reassembly header and all
1201  * associated datagrams.
1202  */
1203 static void
1204 ip_freef(struct ipq *fp)
1205 {
1206         struct mbuf *q;
1207
1208         while (fp->ipq_frags) {
1209                 q = fp->ipq_frags;
1210                 fp->ipq_frags = q->m_nextpkt;
1211                 q->m_nextpkt = NULL;
1212                 m_freem(q);
1213         }
1214         remque(fp);
1215         mpipe_free(&ipq_mpipe, fp);
1216         nipq--;
1217 }
1218
1219 /*
1220  * IP timer processing;
1221  * if a timer expires on a reassembly
1222  * queue, discard it.
1223  */
1224 void
1225 ip_slowtimo(void)
1226 {
1227         struct ipq *fp;
1228         int i;
1229
1230         crit_enter();
1231         for (i = 0; i < IPREASS_NHASH; i++) {
1232                 fp = ipq[i].next;
1233                 if (fp == NULL)
1234                         continue;
1235                 while (fp != &ipq[i]) {
1236                         --fp->ipq_ttl;
1237                         fp = fp->next;
1238                         if (fp->prev->ipq_ttl == 0) {
1239                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1240                                 ip_freef(fp->prev);
1241                         }
1242                 }
1243         }
1244         /*
1245          * If we are over the maximum number of fragments
1246          * (due to the limit being lowered), drain off
1247          * enough to get down to the new limit.
1248          */
1249         if (maxnipq >= 0 && nipq > maxnipq) {
1250                 for (i = 0; i < IPREASS_NHASH; i++) {
1251                         while (nipq > maxnipq &&
1252                                 (ipq[i].next != &ipq[i])) {
1253                                 ipstat.ips_fragdropped +=
1254                                     ipq[i].next->ipq_nfrags;
1255                                 ip_freef(ipq[i].next);
1256                         }
1257                 }
1258         }
1259         ipflow_slowtimo();
1260         crit_exit();
1261 }
1262
1263 /*
1264  * Drain off all datagram fragments.
1265  */
1266 void
1267 ip_drain(void)
1268 {
1269         int i;
1270
1271         for (i = 0; i < IPREASS_NHASH; i++) {
1272                 while (ipq[i].next != &ipq[i]) {
1273                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1274                         ip_freef(ipq[i].next);
1275                 }
1276         }
1277         in_rtqdrain();
1278 }
1279
1280 /*
1281  * Do option processing on a datagram,
1282  * possibly discarding it if bad options are encountered,
1283  * or forwarding it if source-routed.
1284  * The pass argument is used when operating in the IPSTEALTH
1285  * mode to tell what options to process:
1286  * [LS]SRR (pass 0) or the others (pass 1).
1287  * The reason for as many as two passes is that when doing IPSTEALTH,
1288  * non-routing options should be processed only if the packet is for us.
1289  * Returns 1 if packet has been forwarded/freed,
1290  * 0 if the packet should be processed further.
1291  */
1292 static int
1293 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1294 {
1295         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1296         struct ip *ip = mtod(m, struct ip *);
1297         u_char *cp;
1298         struct in_ifaddr *ia;
1299         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1300         boolean_t forward = FALSE;
1301         struct in_addr *sin, dst;
1302         n_time ntime;
1303
1304         dst = ip->ip_dst;
1305         cp = (u_char *)(ip + 1);
1306         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1307         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1308                 opt = cp[IPOPT_OPTVAL];
1309                 if (opt == IPOPT_EOL)
1310                         break;
1311                 if (opt == IPOPT_NOP)
1312                         optlen = 1;
1313                 else {
1314                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1315                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1316                                 goto bad;
1317                         }
1318                         optlen = cp[IPOPT_OLEN];
1319                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1320                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1321                                 goto bad;
1322                         }
1323                 }
1324                 switch (opt) {
1325
1326                 default:
1327                         break;
1328
1329                 /*
1330                  * Source routing with record.
1331                  * Find interface with current destination address.
1332                  * If none on this machine then drop if strictly routed,
1333                  * or do nothing if loosely routed.
1334                  * Record interface address and bring up next address
1335                  * component.  If strictly routed make sure next
1336                  * address is on directly accessible net.
1337                  */
1338                 case IPOPT_LSRR:
1339                 case IPOPT_SSRR:
1340                         if (ipstealth && pass > 0)
1341                                 break;
1342                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1343                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1344                                 goto bad;
1345                         }
1346                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1347                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1348                                 goto bad;
1349                         }
1350                         ipaddr.sin_addr = ip->ip_dst;
1351                         ia = (struct in_ifaddr *)
1352                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1353                         if (ia == NULL) {
1354                                 if (opt == IPOPT_SSRR) {
1355                                         type = ICMP_UNREACH;
1356                                         code = ICMP_UNREACH_SRCFAIL;
1357                                         goto bad;
1358                                 }
1359                                 if (!ip_dosourceroute)
1360                                         goto nosourcerouting;
1361                                 /*
1362                                  * Loose routing, and not at next destination
1363                                  * yet; nothing to do except forward.
1364                                  */
1365                                 break;
1366                         }
1367                         off--;                  /* 0 origin */
1368                         if (off > optlen - (int)sizeof(struct in_addr)) {
1369                                 /*
1370                                  * End of source route.  Should be for us.
1371                                  */
1372                                 if (!ip_acceptsourceroute)
1373                                         goto nosourcerouting;
1374                                 save_rte(m, cp, ip->ip_src);
1375                                 break;
1376                         }
1377                         if (ipstealth)
1378                                 goto dropit;
1379                         if (!ip_dosourceroute) {
1380                                 if (ipforwarding) {
1381                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1382
1383                                         /*
1384                                          * Acting as a router, so generate ICMP
1385                                          */
1386 nosourcerouting:
1387                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1388                                         log(LOG_WARNING,
1389                                             "attempted source route from %s to %s\n",
1390                                             inet_ntoa(ip->ip_src), buf);
1391                                         type = ICMP_UNREACH;
1392                                         code = ICMP_UNREACH_SRCFAIL;
1393                                         goto bad;
1394                                 } else {
1395                                         /*
1396                                          * Not acting as a router,
1397                                          * so silently drop.
1398                                          */
1399 dropit:
1400                                         ipstat.ips_cantforward++;
1401                                         m_freem(m);
1402                                         return (1);
1403                                 }
1404                         }
1405
1406                         /*
1407                          * locate outgoing interface
1408                          */
1409                         memcpy(&ipaddr.sin_addr, cp + off,
1410                             sizeof ipaddr.sin_addr);
1411
1412                         if (opt == IPOPT_SSRR) {
1413 #define INA     struct in_ifaddr *
1414 #define SA      struct sockaddr *
1415                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1416                                                                         == NULL)
1417                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1418                         } else
1419                                 ia = ip_rtaddr(ipaddr.sin_addr,
1420                                                &ipforward_rt[mycpuid]);
1421                         if (ia == NULL) {
1422                                 type = ICMP_UNREACH;
1423                                 code = ICMP_UNREACH_SRCFAIL;
1424                                 goto bad;
1425                         }
1426                         ip->ip_dst = ipaddr.sin_addr;
1427                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1428                             sizeof(struct in_addr));
1429                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1430                         /*
1431                          * Let ip_intr's mcast routing check handle mcast pkts
1432                          */
1433                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1434                         break;
1435
1436                 case IPOPT_RR:
1437                         if (ipstealth && pass == 0)
1438                                 break;
1439                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1440                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1441                                 goto bad;
1442                         }
1443                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1444                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1445                                 goto bad;
1446                         }
1447                         /*
1448                          * If no space remains, ignore.
1449                          */
1450                         off--;                  /* 0 origin */
1451                         if (off > optlen - (int)sizeof(struct in_addr))
1452                                 break;
1453                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1454                             sizeof ipaddr.sin_addr);
1455                         /*
1456                          * locate outgoing interface; if we're the destination,
1457                          * use the incoming interface (should be same).
1458                          */
1459                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1460                             (ia = ip_rtaddr(ipaddr.sin_addr,
1461                                             &ipforward_rt[mycpuid]))
1462                                                                      == NULL) {
1463                                 type = ICMP_UNREACH;
1464                                 code = ICMP_UNREACH_HOST;
1465                                 goto bad;
1466                         }
1467                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1468                             sizeof(struct in_addr));
1469                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1470                         break;
1471
1472                 case IPOPT_TS:
1473                         if (ipstealth && pass == 0)
1474                                 break;
1475                         code = cp - (u_char *)ip;
1476                         if (optlen < 4 || optlen > 40) {
1477                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1478                                 goto bad;
1479                         }
1480                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1481                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1482                                 goto bad;
1483                         }
1484                         if (off > optlen - (int)sizeof(int32_t)) {
1485                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1486                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1487                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1488                                         goto bad;
1489                                 }
1490                                 break;
1491                         }
1492                         off--;                          /* 0 origin */
1493                         sin = (struct in_addr *)(cp + off);
1494                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1495
1496                         case IPOPT_TS_TSONLY:
1497                                 break;
1498
1499                         case IPOPT_TS_TSANDADDR:
1500                                 if (off + sizeof(n_time) +
1501                                     sizeof(struct in_addr) > optlen) {
1502                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1503                                         goto bad;
1504                                 }
1505                                 ipaddr.sin_addr = dst;
1506                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1507                                                             m->m_pkthdr.rcvif);
1508                                 if (ia == NULL)
1509                                         continue;
1510                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1511                                     sizeof(struct in_addr));
1512                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1513                                 off += sizeof(struct in_addr);
1514                                 break;
1515
1516                         case IPOPT_TS_PRESPEC:
1517                                 if (off + sizeof(n_time) +
1518                                     sizeof(struct in_addr) > optlen) {
1519                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1520                                         goto bad;
1521                                 }
1522                                 memcpy(&ipaddr.sin_addr, sin,
1523                                     sizeof(struct in_addr));
1524                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1525                                         continue;
1526                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1527                                 off += sizeof(struct in_addr);
1528                                 break;
1529
1530                         default:
1531                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1532                                 goto bad;
1533                         }
1534                         ntime = iptime();
1535                         memcpy(cp + off, &ntime, sizeof(n_time));
1536                         cp[IPOPT_OFFSET] += sizeof(n_time);
1537                 }
1538         }
1539         if (forward && ipforwarding) {
1540                 ip_forward(m, TRUE, next_hop);
1541                 return (1);
1542         }
1543         return (0);
1544 bad:
1545         icmp_error(m, type, code, 0, 0);
1546         ipstat.ips_badoptions++;
1547         return (1);
1548 }
1549
1550 /*
1551  * Given address of next destination (final or next hop),
1552  * return internet address info of interface to be used to get there.
1553  */
1554 struct in_ifaddr *
1555 ip_rtaddr(struct in_addr dst, struct route *ro)
1556 {
1557         struct sockaddr_in *sin;
1558
1559         sin = (struct sockaddr_in *)&ro->ro_dst;
1560
1561         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1562                 if (ro->ro_rt != NULL) {
1563                         RTFREE(ro->ro_rt);
1564                         ro->ro_rt = NULL;
1565                 }
1566                 sin->sin_family = AF_INET;
1567                 sin->sin_len = sizeof *sin;
1568                 sin->sin_addr = dst;
1569                 rtalloc_ign(ro, RTF_PRCLONING);
1570         }
1571
1572         if (ro->ro_rt == NULL)
1573                 return (NULL);
1574
1575         return (ifatoia(ro->ro_rt->rt_ifa));
1576 }
1577
1578 /*
1579  * Save incoming source route for use in replies,
1580  * to be picked up later by ip_srcroute if the receiver is interested.
1581  */
1582 static void
1583 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1584 {
1585         struct m_tag *mtag;
1586         struct ip_srcrt_opt *opt;
1587         unsigned olen;
1588
1589         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1590         if (mtag == NULL)
1591                 return;
1592         opt = m_tag_data(mtag);
1593
1594         olen = option[IPOPT_OLEN];
1595 #ifdef DIAGNOSTIC
1596         if (ipprintfs)
1597                 kprintf("save_rte: olen %d\n", olen);
1598 #endif
1599         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1600                 m_tag_free(mtag);
1601                 return;
1602         }
1603         bcopy(option, opt->ip_srcrt.srcopt, olen);
1604         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1605         opt->ip_srcrt.dst = dst;
1606         m_tag_prepend(m, mtag);
1607 }
1608
1609 /*
1610  * Retrieve incoming source route for use in replies,
1611  * in the same form used by setsockopt.
1612  * The first hop is placed before the options, will be removed later.
1613  */
1614 struct mbuf *
1615 ip_srcroute(struct mbuf *m0)
1616 {
1617         struct in_addr *p, *q;
1618         struct mbuf *m;
1619         struct m_tag *mtag;
1620         struct ip_srcrt_opt *opt;
1621
1622         if (m0 == NULL)
1623                 return NULL;
1624
1625         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1626         if (mtag == NULL)
1627                 return NULL;
1628         opt = m_tag_data(mtag);
1629
1630         if (opt->ip_nhops == 0)
1631                 return (NULL);
1632         m = m_get(MB_DONTWAIT, MT_HEADER);
1633         if (m == NULL)
1634                 return (NULL);
1635
1636 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1637
1638         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1639         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1640                    sizeof(struct in_addr) + OPTSIZ;
1641 #ifdef DIAGNOSTIC
1642         if (ipprintfs) {
1643                 kprintf("ip_srcroute: nhops %d mlen %d",
1644                         opt->ip_nhops, m->m_len);
1645         }
1646 #endif
1647
1648         /*
1649          * First save first hop for return route
1650          */
1651         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1652         *(mtod(m, struct in_addr *)) = *p--;
1653 #ifdef DIAGNOSTIC
1654         if (ipprintfs)
1655                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1656 #endif
1657
1658         /*
1659          * Copy option fields and padding (nop) to mbuf.
1660          */
1661         opt->ip_srcrt.nop = IPOPT_NOP;
1662         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1663         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1664             OPTSIZ);
1665         q = (struct in_addr *)(mtod(m, caddr_t) +
1666             sizeof(struct in_addr) + OPTSIZ);
1667 #undef OPTSIZ
1668         /*
1669          * Record return path as an IP source route,
1670          * reversing the path (pointers are now aligned).
1671          */
1672         while (p >= opt->ip_srcrt.route) {
1673 #ifdef DIAGNOSTIC
1674                 if (ipprintfs)
1675                         kprintf(" %x", ntohl(q->s_addr));
1676 #endif
1677                 *q++ = *p--;
1678         }
1679         /*
1680          * Last hop goes to final destination.
1681          */
1682         *q = opt->ip_srcrt.dst;
1683         m_tag_delete(m0, mtag);
1684 #ifdef DIAGNOSTIC
1685         if (ipprintfs)
1686                 kprintf(" %x\n", ntohl(q->s_addr));
1687 #endif
1688         return (m);
1689 }
1690
1691 /*
1692  * Strip out IP options.
1693  */
1694 void
1695 ip_stripoptions(struct mbuf *m)
1696 {
1697         int datalen;
1698         struct ip *ip = mtod(m, struct ip *);
1699         caddr_t opts;
1700         int optlen;
1701
1702         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1703         opts = (caddr_t)(ip + 1);
1704         datalen = m->m_len - (sizeof(struct ip) + optlen);
1705         bcopy(opts + optlen, opts, datalen);
1706         m->m_len -= optlen;
1707         if (m->m_flags & M_PKTHDR)
1708                 m->m_pkthdr.len -= optlen;
1709         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1710 }
1711
1712 u_char inetctlerrmap[PRC_NCMDS] = {
1713         0,              0,              0,              0,
1714         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1715         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1716         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1717         0,              0,              0,              0,
1718         ENOPROTOOPT,    ECONNREFUSED
1719 };
1720
1721 /*
1722  * Forward a packet.  If some error occurs return the sender
1723  * an icmp packet.  Note we can't always generate a meaningful
1724  * icmp message because icmp doesn't have a large enough repertoire
1725  * of codes and types.
1726  *
1727  * If not forwarding, just drop the packet.  This could be confusing
1728  * if ipforwarding was zero but some routing protocol was advancing
1729  * us as a gateway to somewhere.  However, we must let the routing
1730  * protocol deal with that.
1731  *
1732  * The using_srcrt parameter indicates whether the packet is being forwarded
1733  * via a source route.
1734  */
1735 void
1736 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1737 {
1738         struct ip *ip = mtod(m, struct ip *);
1739         struct sockaddr_in *ipforward_rtaddr;
1740         struct rtentry *rt;
1741         int error, type = 0, code = 0, destmtu = 0;
1742         struct mbuf *mcopy;
1743         n_long dest;
1744         struct in_addr pkt_dst;
1745         struct route *cache_rt = &ipforward_rt[mycpuid];
1746
1747         dest = INADDR_ANY;
1748         /*
1749          * Cache the destination address of the packet; this may be
1750          * changed by use of 'ipfw fwd'.
1751          */
1752         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1753
1754 #ifdef DIAGNOSTIC
1755         if (ipprintfs)
1756                 kprintf("forward: src %x dst %x ttl %x\n",
1757                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1758 #endif
1759
1760         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1761                 ipstat.ips_cantforward++;
1762                 m_freem(m);
1763                 return;
1764         }
1765         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1766                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1767                 return;
1768         }
1769
1770         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1771         if (cache_rt->ro_rt == NULL ||
1772             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1773                 if (cache_rt->ro_rt != NULL) {
1774                         RTFREE(cache_rt->ro_rt);
1775                         cache_rt->ro_rt = NULL;
1776                 }
1777                 ipforward_rtaddr->sin_family = AF_INET;
1778                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1779                 ipforward_rtaddr->sin_addr = pkt_dst;
1780                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1781                 if (cache_rt->ro_rt == NULL) {
1782                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1783                         return;
1784                 }
1785         }
1786         rt = cache_rt->ro_rt;
1787
1788         /*
1789          * Save the IP header and at most 8 bytes of the payload,
1790          * in case we need to generate an ICMP message to the src.
1791          *
1792          * XXX this can be optimized a lot by saving the data in a local
1793          * buffer on the stack (72 bytes at most), and only allocating the
1794          * mbuf if really necessary. The vast majority of the packets
1795          * are forwarded without having to send an ICMP back (either
1796          * because unnecessary, or because rate limited), so we are
1797          * really we are wasting a lot of work here.
1798          *
1799          * We don't use m_copy() because it might return a reference
1800          * to a shared cluster. Both this function and ip_output()
1801          * assume exclusive access to the IP header in `m', so any
1802          * data in a cluster may change before we reach icmp_error().
1803          */
1804         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1805         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1806                 /*
1807                  * It's probably ok if the pkthdr dup fails (because
1808                  * the deep copy of the tag chain failed), but for now
1809                  * be conservative and just discard the copy since
1810                  * code below may some day want the tags.
1811                  */
1812                 m_free(mcopy);
1813                 mcopy = NULL;
1814         }
1815         if (mcopy != NULL) {
1816                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1817                     (int)ip->ip_len);
1818                 mcopy->m_pkthdr.len = mcopy->m_len;
1819                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1820         }
1821
1822         if (!ipstealth)
1823                 ip->ip_ttl -= IPTTLDEC;
1824
1825         /*
1826          * If forwarding packet using same interface that it came in on,
1827          * perhaps should send a redirect to sender to shortcut a hop.
1828          * Only send redirect if source is sending directly to us,
1829          * and if packet was not source routed (or has any options).
1830          * Also, don't send redirect if forwarding using a default route
1831          * or a route modified by a redirect.
1832          */
1833         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1834             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1835             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1836             ipsendredirects && !using_srcrt && next_hop == NULL) {
1837                 u_long src = ntohl(ip->ip_src.s_addr);
1838                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1839
1840                 if (rt_ifa != NULL &&
1841                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1842                         if (rt->rt_flags & RTF_GATEWAY)
1843                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1844                         else
1845                                 dest = pkt_dst.s_addr;
1846                         /*
1847                          * Router requirements says to only send
1848                          * host redirects.
1849                          */
1850                         type = ICMP_REDIRECT;
1851                         code = ICMP_REDIRECT_HOST;
1852 #ifdef DIAGNOSTIC
1853                         if (ipprintfs)
1854                                 kprintf("redirect (%d) to %x\n", code, dest);
1855 #endif
1856                 }
1857         }
1858
1859         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1860         if (error == 0) {
1861                 ipstat.ips_forward++;
1862                 if (type == 0) {
1863                         if (mcopy) {
1864                                 ipflow_create(cache_rt, mcopy);
1865                                 m_freem(mcopy);
1866                         }
1867                         return;         /* most common case */
1868                 } else {
1869                         ipstat.ips_redirectsent++;
1870                 }
1871         } else {
1872                 ipstat.ips_cantforward++;
1873         }
1874
1875         if (mcopy == NULL)
1876                 return;
1877
1878         /*
1879          * Send ICMP message.
1880          */
1881
1882         switch (error) {
1883
1884         case 0:                         /* forwarded, but need redirect */
1885                 /* type, code set above */
1886                 break;
1887
1888         case ENETUNREACH:               /* shouldn't happen, checked above */
1889         case EHOSTUNREACH:
1890         case ENETDOWN:
1891         case EHOSTDOWN:
1892         default:
1893                 type = ICMP_UNREACH;
1894                 code = ICMP_UNREACH_HOST;
1895                 break;
1896
1897         case EMSGSIZE:
1898                 type = ICMP_UNREACH;
1899                 code = ICMP_UNREACH_NEEDFRAG;
1900 #ifdef IPSEC
1901                 /*
1902                  * If the packet is routed over IPsec tunnel, tell the
1903                  * originator the tunnel MTU.
1904                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1905                  * XXX quickhack!!!
1906                  */
1907                 if (cache_rt->ro_rt != NULL) {
1908                         struct secpolicy *sp = NULL;
1909                         int ipsecerror;
1910                         int ipsechdr;
1911                         struct route *ro;
1912
1913                         sp = ipsec4_getpolicybyaddr(mcopy,
1914                                                     IPSEC_DIR_OUTBOUND,
1915                                                     IP_FORWARDING,
1916                                                     &ipsecerror);
1917
1918                         if (sp == NULL)
1919                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1920                         else {
1921                                 /* count IPsec header size */
1922                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1923                                                          IPSEC_DIR_OUTBOUND,
1924                                                          NULL);
1925
1926                                 /*
1927                                  * find the correct route for outer IPv4
1928                                  * header, compute tunnel MTU.
1929                                  *
1930                                  */
1931                                 if (sp->req != NULL && sp->req->sav != NULL &&
1932                                     sp->req->sav->sah != NULL) {
1933                                         ro = &sp->req->sav->sah->sa_route;
1934                                         if (ro->ro_rt != NULL &&
1935                                             ro->ro_rt->rt_ifp != NULL) {
1936                                                 destmtu =
1937                                                     ro->ro_rt->rt_ifp->if_mtu;
1938                                                 destmtu -= ipsechdr;
1939                                         }
1940                                 }
1941
1942                                 key_freesp(sp);
1943                         }
1944                 }
1945 #elif FAST_IPSEC
1946                 /*
1947                  * If the packet is routed over IPsec tunnel, tell the
1948                  * originator the tunnel MTU.
1949                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1950                  * XXX quickhack!!!
1951                  */
1952                 if (cache_rt->ro_rt != NULL) {
1953                         struct secpolicy *sp = NULL;
1954                         int ipsecerror;
1955                         int ipsechdr;
1956                         struct route *ro;
1957
1958                         sp = ipsec_getpolicybyaddr(mcopy,
1959                                                    IPSEC_DIR_OUTBOUND,
1960                                                    IP_FORWARDING,
1961                                                    &ipsecerror);
1962
1963                         if (sp == NULL)
1964                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1965                         else {
1966                                 /* count IPsec header size */
1967                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1968                                                          IPSEC_DIR_OUTBOUND,
1969                                                          NULL);
1970
1971                                 /*
1972                                  * find the correct route for outer IPv4
1973                                  * header, compute tunnel MTU.
1974                                  */
1975
1976                                 if (sp->req != NULL &&
1977                                     sp->req->sav != NULL &&
1978                                     sp->req->sav->sah != NULL) {
1979                                         ro = &sp->req->sav->sah->sa_route;
1980                                         if (ro->ro_rt != NULL &&
1981                                             ro->ro_rt->rt_ifp != NULL) {
1982                                                 destmtu =
1983                                                     ro->ro_rt->rt_ifp->if_mtu;
1984                                                 destmtu -= ipsechdr;
1985                                         }
1986                                 }
1987
1988                                 KEY_FREESP(&sp);
1989                         }
1990                 }
1991 #else /* !IPSEC && !FAST_IPSEC */
1992                 if (cache_rt->ro_rt != NULL)
1993                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1994 #endif /*IPSEC*/
1995                 ipstat.ips_cantfrag++;
1996                 break;
1997
1998         case ENOBUFS:
1999                 /*
2000                  * A router should not generate ICMP_SOURCEQUENCH as
2001                  * required in RFC1812 Requirements for IP Version 4 Routers.
2002                  * Source quench could be a big problem under DoS attacks,
2003                  * or if the underlying interface is rate-limited.
2004                  * Those who need source quench packets may re-enable them
2005                  * via the net.inet.ip.sendsourcequench sysctl.
2006                  */
2007                 if (!ip_sendsourcequench) {
2008                         m_freem(mcopy);
2009                         return;
2010                 } else {
2011                         type = ICMP_SOURCEQUENCH;
2012                         code = 0;
2013                 }
2014                 break;
2015
2016         case EACCES:                    /* ipfw denied packet */
2017                 m_freem(mcopy);
2018                 return;
2019         }
2020         icmp_error(mcopy, type, code, dest, destmtu);
2021 }
2022
2023 void
2024 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2025                struct mbuf *m)
2026 {
2027         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2028                 struct timeval tv;
2029
2030                 microtime(&tv);
2031                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2032                     SCM_TIMESTAMP, SOL_SOCKET);
2033                 if (*mp)
2034                         mp = &(*mp)->m_next;
2035         }
2036         if (inp->inp_flags & INP_RECVDSTADDR) {
2037                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2038                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2039                 if (*mp)
2040                         mp = &(*mp)->m_next;
2041         }
2042         if (inp->inp_flags & INP_RECVTTL) {
2043                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2044                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2045                 if (*mp)
2046                         mp = &(*mp)->m_next;
2047         }
2048 #ifdef notyet
2049         /* XXX
2050          * Moving these out of udp_input() made them even more broken
2051          * than they already were.
2052          */
2053         /* options were tossed already */
2054         if (inp->inp_flags & INP_RECVOPTS) {
2055                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2056                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2057                 if (*mp)
2058                         mp = &(*mp)->m_next;
2059         }
2060         /* ip_srcroute doesn't do what we want here, need to fix */
2061         if (inp->inp_flags & INP_RECVRETOPTS) {
2062                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2063                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2064                 if (*mp)
2065                         mp = &(*mp)->m_next;
2066         }
2067 #endif
2068         if (inp->inp_flags & INP_RECVIF) {
2069                 struct ifnet *ifp;
2070                 struct sdlbuf {
2071                         struct sockaddr_dl sdl;
2072                         u_char  pad[32];
2073                 } sdlbuf;
2074                 struct sockaddr_dl *sdp;
2075                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2076
2077                 if (((ifp = m->m_pkthdr.rcvif)) &&
2078                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2079                         sdp = IF_LLSOCKADDR(ifp);
2080                         /*
2081                          * Change our mind and don't try copy.
2082                          */
2083                         if ((sdp->sdl_family != AF_LINK) ||
2084                             (sdp->sdl_len > sizeof(sdlbuf))) {
2085                                 goto makedummy;
2086                         }
2087                         bcopy(sdp, sdl2, sdp->sdl_len);
2088                 } else {
2089 makedummy:
2090                         sdl2->sdl_len =
2091                             offsetof(struct sockaddr_dl, sdl_data[0]);
2092                         sdl2->sdl_family = AF_LINK;
2093                         sdl2->sdl_index = 0;
2094                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2095                 }
2096                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2097                         IP_RECVIF, IPPROTO_IP);
2098                 if (*mp)
2099                         mp = &(*mp)->m_next;
2100         }
2101 }
2102
2103 /*
2104  * XXX these routines are called from the upper part of the kernel.
2105  *
2106  * They could also be moved to ip_mroute.c, since all the RSVP
2107  *  handling is done there already.
2108  */
2109 int
2110 ip_rsvp_init(struct socket *so)
2111 {
2112         if (so->so_type != SOCK_RAW ||
2113             so->so_proto->pr_protocol != IPPROTO_RSVP)
2114                 return EOPNOTSUPP;
2115
2116         if (ip_rsvpd != NULL)
2117                 return EADDRINUSE;
2118
2119         ip_rsvpd = so;
2120         /*
2121          * This may seem silly, but we need to be sure we don't over-increment
2122          * the RSVP counter, in case something slips up.
2123          */
2124         if (!ip_rsvp_on) {
2125                 ip_rsvp_on = 1;
2126                 rsvp_on++;
2127         }
2128
2129         return 0;
2130 }
2131
2132 int
2133 ip_rsvp_done(void)
2134 {
2135         ip_rsvpd = NULL;
2136         /*
2137          * This may seem silly, but we need to be sure we don't over-decrement
2138          * the RSVP counter, in case something slips up.
2139          */
2140         if (ip_rsvp_on) {
2141                 ip_rsvp_on = 0;
2142                 rsvp_on--;
2143         }
2144         return 0;
2145 }
2146
2147 void
2148 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2149 {
2150         int off, proto;
2151         __va_list ap;
2152
2153         __va_start(ap, m);
2154         off = __va_arg(ap, int);
2155         proto = __va_arg(ap, int);
2156         __va_end(ap);
2157
2158         if (rsvp_input_p) { /* call the real one if loaded */
2159                 rsvp_input_p(m, off, proto);
2160                 return;
2161         }
2162
2163         /* Can still get packets with rsvp_on = 0 if there is a local member
2164          * of the group to which the RSVP packet is addressed.  But in this
2165          * case we want to throw the packet away.
2166          */
2167
2168         if (!rsvp_on) {
2169                 m_freem(m);
2170                 return;
2171         }
2172
2173         if (ip_rsvpd != NULL) {
2174                 rip_input(m, off, proto);
2175                 return;
2176         }
2177         /* Drop the packet */
2178         m_freem(m);
2179 }