Parallelize in_ifaddrhashtbl
[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 1.81 2008/06/09 11:24:24 sephe Exp $
69  */
70
71 #define _IP_VHL
72
73 #include "opt_bootp.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipdn.h"
76 #include "opt_ipdivert.h"
77 #include "opt_ipfilter.h"
78 #include "opt_ipstealth.h"
79 #include "opt_ipsec.h"
80
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/mpipe.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/time.h>
90 #include <sys/globaldata.h>
91 #include <sys/thread.h>
92 #include <sys/kernel.h>
93 #include <sys/syslog.h>
94 #include <sys/sysctl.h>
95 #include <sys/in_cksum.h>
96
97 #include <machine/stdarg.h>
98
99 #include <net/if.h>
100 #include <net/if_types.h>
101 #include <net/if_var.h>
102 #include <net/if_dl.h>
103 #include <net/pfil.h>
104 #include <net/route.h>
105 #include <net/netisr.h>
106
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/in_var.h>
110 #include <netinet/ip.h>
111 #include <netinet/in_pcb.h>
112 #include <netinet/ip_var.h>
113 #include <netinet/ip_icmp.h>
114
115 #include <sys/thread2.h>
116 #include <sys/msgport2.h>
117 #include <net/netmsg2.h>
118
119 #include <sys/socketvar.h>
120
121 #include <net/ipfw/ip_fw.h>
122 #include <net/dummynet/ip_dummynet.h>
123
124 #ifdef IPSEC
125 #include <netinet6/ipsec.h>
126 #include <netproto/key/key.h>
127 #endif
128
129 #ifdef FAST_IPSEC
130 #include <netproto/ipsec/ipsec.h>
131 #include <netproto/ipsec/key.h>
132 #endif
133
134 int rsvp_on = 0;
135 static int ip_rsvp_on;
136 struct socket *ip_rsvpd;
137
138 int ipforwarding = 0;
139 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
140     &ipforwarding, 0, "Enable IP forwarding between interfaces");
141
142 static int ipsendredirects = 1; /* XXX */
143 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
144     &ipsendredirects, 0, "Enable sending IP redirects");
145
146 int ip_defttl = IPDEFTTL;
147 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
148     &ip_defttl, 0, "Maximum TTL on IP packets");
149
150 static int ip_dosourceroute = 0;
151 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
152     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
153
154 static int ip_acceptsourceroute = 0;
155 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
156     CTLFLAG_RW, &ip_acceptsourceroute, 0,
157     "Enable accepting source routed IP packets");
158
159 static int ip_keepfaith = 0;
160 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
161     &ip_keepfaith, 0,
162     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
163
164 static int nipq = 0;    /* total # of reass queues */
165 static int maxnipq;
166 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
167     &maxnipq, 0,
168     "Maximum number of IPv4 fragment reassembly queue entries");
169
170 static int maxfragsperpacket;
171 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
172     &maxfragsperpacket, 0,
173     "Maximum number of IPv4 fragments allowed per packet");
174
175 static int ip_sendsourcequench = 0;
176 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
177     &ip_sendsourcequench, 0,
178     "Enable the transmission of source quench packets");
179
180 int ip_do_randomid = 1;
181 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
182     &ip_do_randomid, 0,
183     "Assign random ip_id values");      
184 /*
185  * XXX - Setting ip_checkinterface mostly implements the receive side of
186  * the Strong ES model described in RFC 1122, but since the routing table
187  * and transmit implementation do not implement the Strong ES model,
188  * setting this to 1 results in an odd hybrid.
189  *
190  * XXX - ip_checkinterface currently must be disabled if you use ipnat
191  * to translate the destination address to another local interface.
192  *
193  * XXX - ip_checkinterface must be disabled if you add IP aliases
194  * to the loopback interface instead of the interface where the
195  * packets for those addresses are received.
196  */
197 static int ip_checkinterface = 0;
198 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
199     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
200
201 #ifdef DIAGNOSTIC
202 static int ipprintfs = 0;
203 #endif
204
205 extern  struct domain inetdomain;
206 extern  struct protosw inetsw[];
207 u_char  ip_protox[IPPROTO_MAX];
208 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
209 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
210                                                 /* inet addr hash table */
211 u_long  in_ifaddrhmask;                         /* mask for hash table */
212
213 struct ip_stats ipstats_percpu[MAXCPU];
214 #ifdef SMP
215 static int
216 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
217 {
218         int cpu, error = 0;
219
220         for (cpu = 0; cpu < ncpus; ++cpu) {
221                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
222                                         sizeof(struct ip_stats))))
223                         break;
224                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
225                                        sizeof(struct ip_stats))))
226                         break;
227         }
228
229         return (error);
230 }
231 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
232     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
233 #else
234 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
235     &ipstat, ip_stats, "IP statistics");
236 #endif
237
238 /* Packet reassembly stuff */
239 #define IPREASS_NHASH_LOG2      6
240 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
241 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
242 #define IPREASS_HASH(x,y)                                               \
243     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
244
245 static struct ipq ipq[IPREASS_NHASH];
246
247 #ifdef IPCTL_DEFMTU
248 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
249     &ip_mtu, 0, "Default MTU");
250 #endif
251
252 #ifdef IPSTEALTH
253 static int ipstealth = 0;
254 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
255 #else
256 static const int ipstealth = 0;
257 #endif
258
259
260 /* Firewall hooks */
261 ip_fw_chk_t *ip_fw_chk_ptr;
262 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
263 int fw_enable = 1;
264 int fw_one_pass = 1;
265
266 struct pfil_head inet_pfil_hook;
267
268 /*
269  * XXX this is ugly -- the following two global variables are
270  * used to store packet state while it travels through the stack.
271  * Note that the code even makes assumptions on the size and
272  * alignment of fields inside struct ip_srcrt so e.g. adding some
273  * fields will break the code. This needs to be fixed.
274  *
275  * We need to save the IP options in case a protocol wants to respond
276  * to an incoming packet over the same route if the packet got here
277  * using IP source routing.  This allows connection establishment and
278  * maintenance when the remote end is on a network that is not known
279  * to us.
280  */
281 static int ip_nhops = 0;
282
283 static  struct ip_srcrt {
284         struct  in_addr dst;                    /* final destination */
285         char    nop;                            /* one NOP to align */
286         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
287         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
288 } ip_srcrt;
289
290 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
291 static struct malloc_pipe ipq_mpipe;
292
293 static void             save_rte (u_char *, struct in_addr);
294 static int              ip_dooptions (struct mbuf *m, int,
295                                         struct sockaddr_in *next_hop);
296 static void             ip_forward (struct mbuf *m, boolean_t using_srcrt,
297                                         struct sockaddr_in *next_hop);
298 static void             ip_freef (struct ipq *);
299 static void             ip_input_handler (struct netmsg *);
300 static struct mbuf      *ip_reass (struct mbuf *, struct ipq *,
301                                         struct ipq *, u_int32_t *);
302
303 /*
304  * IP initialization: fill in IP protocol switch table.
305  * All protocols not implemented in kernel go to raw IP protocol handler.
306  */
307 void
308 ip_init(void)
309 {
310         struct protosw *pr;
311         int i;
312 #ifdef SMP
313         int cpu;
314 #endif
315
316         /*
317          * Make sure we can handle a reasonable number of fragments but
318          * cap it at 4000 (XXX).
319          */
320         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
321                     IFQ_MAXLEN, 4000, 0, NULL);
322         for (i = 0; i < ncpus; ++i) {
323                 TAILQ_INIT(&in_ifaddrheads[i]);
324                 in_ifaddrhashtbls[i] =
325                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
326         }
327         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
328         if (pr == NULL)
329                 panic("ip_init");
330         for (i = 0; i < IPPROTO_MAX; i++)
331                 ip_protox[i] = pr - inetsw;
332         for (pr = inetdomain.dom_protosw;
333              pr < inetdomain.dom_protoswNPROTOSW; pr++)
334                 if (pr->pr_domain->dom_family == PF_INET &&
335                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
336                         ip_protox[pr->pr_protocol] = pr - inetsw;
337
338         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
339         inet_pfil_hook.ph_af = AF_INET;
340         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
341                 kprintf("%s: WARNING: unable to register pfil hook, "
342                         "error %d\n", __func__, i);
343         }
344
345         for (i = 0; i < IPREASS_NHASH; i++)
346             ipq[i].next = ipq[i].prev = &ipq[i];
347
348         maxnipq = nmbclusters / 32;
349         maxfragsperpacket = 16;
350
351         ip_id = time_second & 0xffff;
352
353         /*
354          * Initialize IP statistics counters for each CPU.
355          *
356          */
357 #ifdef SMP
358         for (cpu = 0; cpu < ncpus; ++cpu) {
359                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
360         }
361 #else
362         bzero(&ipstat, sizeof(struct ip_stats));
363 #endif
364
365         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
366 }
367
368 /*
369  * XXX watch out this one. It is perhaps used as a cache for
370  * the most recently used route ? it is cleared in in_addroute()
371  * when a new route is successfully created.
372  */
373 struct route ipforward_rt[MAXCPU];
374
375 /* Do transport protocol processing. */
376 static void
377 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip,
378                            struct sockaddr_in *nexthop)
379 {
380         /*
381          * Switch out to protocol's input routine.
382          */
383         if (nexthop && ip->ip_p == IPPROTO_TCP) {
384                 /* TCP needs IPFORWARD info if available */
385                 struct m_hdr tag;
386
387                 tag.mh_type = MT_TAG;
388                 tag.mh_flags = PACKET_TAG_IPFORWARD;
389                 tag.mh_data = (caddr_t)nexthop;
390                 tag.mh_next = m;
391
392                 (*inetsw[ip_protox[ip->ip_p]].pr_input)
393                     ((struct mbuf *)&tag, hlen, ip->ip_p);
394         } else {
395                 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
396         }
397 }
398
399 struct netmsg_transport_packet {
400         struct netmsg           nm_netmsg;
401         struct mbuf             *nm_mbuf;
402         int                     nm_hlen;
403         boolean_t               nm_hasnexthop;
404         struct sockaddr_in      nm_nexthop;
405 };
406
407 static void
408 transport_processing_handler(netmsg_t netmsg)
409 {
410         struct netmsg_transport_packet *msg = (void *)netmsg;
411         struct sockaddr_in *nexthop;
412         struct ip *ip;
413
414         ip = mtod(msg->nm_mbuf, struct ip *);
415         nexthop = msg->nm_hasnexthop ? &msg->nm_nexthop : NULL;
416         transport_processing_oncpu(msg->nm_mbuf, msg->nm_hlen, ip, nexthop);
417         lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
418 }
419
420 static void
421 ip_input_handler(struct netmsg *msg0)
422 {
423         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
424
425         ip_input(m);
426         /* msg0 was embedded in the mbuf, do not reply! */
427 }
428
429 /*
430  * IP input routine.  Checksum and byte swap header.  If fragmented
431  * try to reassemble.  Process options.  Pass to next level.
432  */
433 void
434 ip_input(struct mbuf *m)
435 {
436         struct ip *ip;
437         struct ipq *fp;
438         struct in_ifaddr *ia = NULL;
439         struct in_ifaddr_container *iac;
440         int i, hlen, checkif;
441         u_short sum;
442         struct in_addr pkt_dst;
443         u_int32_t divert_info = 0;              /* packet divert/tee info */
444         struct ip_fw_args args;
445         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
446         boolean_t needredispatch = FALSE;
447         struct in_addr odst;                    /* original dst address(NAT) */
448         struct m_tag *mtag;
449 #ifdef FAST_IPSEC
450         struct tdb_ident *tdbi;
451         struct secpolicy *sp;
452         int error;
453 #endif
454
455         args.eh = NULL;
456         args.oif = NULL;
457         args.rule = NULL;
458         args.next_hop = NULL;
459
460         /* Grab info from MT_TAG mbufs prepended to the chain. */
461         while (m != NULL && m->m_type == MT_TAG) {
462                 switch(m->_m_tag_id) {
463                 case PACKET_TAG_IPFORWARD:
464                         args.next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
465                         break;
466                 default:
467                         kprintf("ip_input: unrecognised MT_TAG tag %d\n",
468                             m->_m_tag_id);
469                         break;
470                 }
471                 m = m->m_next;
472         }
473         M_ASSERTPKTHDR(m);
474
475         /* Extract info from dummynet tag */
476         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
477         if (mtag != NULL) {
478                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
479
480                 m_tag_delete(m, mtag);
481                 mtag = NULL;
482         }
483
484         if (args.rule != NULL) {        /* dummynet already filtered us */
485                 ip = mtod(m, struct ip *);
486                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
487                 goto iphack;
488         }
489
490         ipstat.ips_total++;
491
492         /* length checks already done in ip_demux() */
493         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
494
495         ip = mtod(m, struct ip *);
496
497         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
498                 ipstat.ips_badvers++;
499                 goto bad;
500         }
501
502         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
503         /* length checks already done in ip_demux() */
504         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
505         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
506
507         /* 127/8 must not appear on wire - RFC1122 */
508         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
509             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
510                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
511                         ipstat.ips_badaddr++;
512                         goto bad;
513                 }
514         }
515
516         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
517                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
518         } else {
519                 if (hlen == sizeof(struct ip)) {
520                         sum = in_cksum_hdr(ip);
521                 } else {
522                         sum = in_cksum(m, hlen);
523                 }
524         }
525         if (sum != 0) {
526                 ipstat.ips_badsum++;
527                 goto bad;
528         }
529
530 #ifdef ALTQ
531         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
532                 /* packet is dropped by traffic conditioner */
533                 return;
534         }
535 #endif
536         /*
537          * Convert fields to host representation.
538          */
539         ip->ip_len = ntohs(ip->ip_len);
540         if (ip->ip_len < hlen) {
541                 ipstat.ips_badlen++;
542                 goto bad;
543         }
544         ip->ip_off = ntohs(ip->ip_off);
545
546         /*
547          * Check that the amount of data in the buffers
548          * is as at least much as the IP header would have us expect.
549          * Trim mbufs if longer than we expect.
550          * Drop packet if shorter than we expect.
551          */
552         if (m->m_pkthdr.len < ip->ip_len) {
553                 ipstat.ips_tooshort++;
554                 goto bad;
555         }
556         if (m->m_pkthdr.len > ip->ip_len) {
557                 if (m->m_len == m->m_pkthdr.len) {
558                         m->m_len = ip->ip_len;
559                         m->m_pkthdr.len = ip->ip_len;
560                 } else
561                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
562         }
563 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
564         /*
565          * Bypass packet filtering for packets from a tunnel (gif).
566          */
567         if (ipsec_gethist(m, NULL))
568                 goto pass;
569 #endif
570
571         /*
572          * IpHack's section.
573          * Right now when no processing on packet has done
574          * and it is still fresh out of network we do our black
575          * deals with it.
576          * - Firewall: deny/allow/divert
577          * - Xlate: translate packet's addr/port (NAT).
578          * - Pipe: pass pkt through dummynet.
579          * - Wrap: fake packet's addr/port <unimpl.>
580          * - Encapsulate: put it in another IP and send out. <unimp.>
581          */
582
583 iphack:
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         if (pfil_has_hooks(&inet_pfil_hook)) {
593                 odst = ip->ip_dst;
594                 if (pfil_run_hooks(&inet_pfil_hook, &m,
595                     m->m_pkthdr.rcvif, PFIL_IN)) {
596                         return;
597                 }
598                 if (m == NULL)                  /* consumed by filter */
599                         return;
600                 ip = mtod(m, struct ip *);
601                 using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
602         }
603
604         if (fw_enable && IPFW_LOADED) {
605                 /*
606                  * If we've been forwarded from the output side, then
607                  * skip the firewall a second time
608                  */
609                 if (args.next_hop != NULL)
610                         goto ours;
611
612                 args.m = m;
613                 i = ip_fw_chk_ptr(&args);
614                 m = args.m;
615
616                 if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {  /* drop */
617                         if (m != NULL)
618                                 m_freem(m);
619                         return;
620                 }
621                 ip = mtod(m, struct ip *);      /* just in case m changed */
622                 if (i == 0 && args.next_hop == NULL)    /* common case */
623                         goto pass;
624                 if (i & IP_FW_PORT_DYNT_FLAG) {
625                         /* Send packet to the appropriate pipe */
626                         ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
627                         return;
628                 }
629 #ifdef IPDIVERT
630                 if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
631                         /* Divert or tee packet */
632                         divert_info = i;
633                         goto ours;
634                 }
635 #endif
636                 if (i == 0 && args.next_hop != NULL)
637                         goto pass;
638                 /*
639                  * if we get here, the packet must be dropped
640                  */
641                 m_freem(m);
642                 return;
643         }
644 pass:
645
646         /*
647          * Process options and, if not destined for us,
648          * ship it on.  ip_dooptions returns 1 when an
649          * error was detected (causing an icmp message
650          * to be sent and the original packet to be freed).
651          */
652         ip_nhops = 0;           /* for source routed packets */
653         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, args.next_hop))
654                 return;
655
656         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
657          * matter if it is destined to another node, or whether it is
658          * a multicast one, RSVP wants it! and prevents it from being forwarded
659          * anywhere else. Also checks if the rsvp daemon is running before
660          * grabbing the packet.
661          */
662         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
663                 goto ours;
664
665         /*
666          * Check our list of addresses, to see if the packet is for us.
667          * If we don't have any addresses, assume any unicast packet
668          * we receive might be for us (and let the upper layers deal
669          * with it).
670          */
671         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
672             !(m->m_flags & (M_MCAST | M_BCAST)))
673                 goto ours;
674
675         /*
676          * Cache the destination address of the packet; this may be
677          * changed by use of 'ipfw fwd'.
678          */
679         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
680
681         /*
682          * Enable a consistency check between the destination address
683          * and the arrival interface for a unicast packet (the RFC 1122
684          * strong ES model) if IP forwarding is disabled and the packet
685          * is not locally generated and the packet is not subject to
686          * 'ipfw fwd'.
687          *
688          * XXX - Checking also should be disabled if the destination
689          * address is ipnat'ed to a different interface.
690          *
691          * XXX - Checking is incompatible with IP aliases added
692          * to the loopback interface instead of the interface where
693          * the packets are received.
694          */
695         checkif = ip_checkinterface &&
696                   !ipforwarding &&
697                   m->m_pkthdr.rcvif != NULL &&
698                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
699                   (args.next_hop == NULL);
700
701         /*
702          * Check for exact addresses in the hash bucket.
703          */
704         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
705                 ia = iac->ia;
706
707                 /*
708                  * If the address matches, verify that the packet
709                  * arrived via the correct interface if checking is
710                  * enabled.
711                  */
712                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
713                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
714                         goto ours;
715         }
716         ia = NULL;
717
718         /*
719          * Check for broadcast addresses.
720          *
721          * Only accept broadcast packets that arrive via the matching
722          * interface.  Reception of forwarded directed broadcasts would
723          * be handled via ip_forward() and ether_output() with the loopback
724          * into the stack for SIMPLEX interfaces handled by ether_output().
725          */
726         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
727                 struct ifaddr_container *ifac;
728
729                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
730                               ifa_link) {
731                         struct ifaddr *ifa = ifac->ifa;
732
733                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
734                                 continue;
735                         if (ifa->ifa_addr->sa_family != AF_INET)
736                                 continue;
737                         ia = ifatoia(ifa);
738                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
739                                                                 pkt_dst.s_addr)
740                                 goto ours;
741                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
742                                 goto ours;
743 #ifdef BOOTP_COMPAT
744                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
745                                 goto ours;
746 #endif
747                 }
748         }
749         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
750                 struct in_multi *inm;
751
752                 if (ip_mrouter != NULL) {
753                         /*
754                          * If we are acting as a multicast router, all
755                          * incoming multicast packets are passed to the
756                          * kernel-level multicast forwarding function.
757                          * The packet is returned (relatively) intact; if
758                          * ip_mforward() returns a non-zero value, the packet
759                          * must be discarded, else it may be accepted below.
760                          */
761                         if (ip_mforward != NULL &&
762                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
763                                 ipstat.ips_cantforward++;
764                                 m_freem(m);
765                                 return;
766                         }
767
768                         /*
769                          * The process-level routing daemon needs to receive
770                          * all multicast IGMP packets, whether or not this
771                          * host belongs to their destination groups.
772                          */
773                         if (ip->ip_p == IPPROTO_IGMP)
774                                 goto ours;
775                         ipstat.ips_forward++;
776                 }
777                 /*
778                  * See if we belong to the destination multicast group on the
779                  * arrival interface.
780                  */
781                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
782                 if (inm == NULL) {
783                         ipstat.ips_notmember++;
784                         m_freem(m);
785                         return;
786                 }
787                 goto ours;
788         }
789         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
790                 goto ours;
791         if (ip->ip_dst.s_addr == INADDR_ANY)
792                 goto ours;
793
794         /*
795          * FAITH(Firewall Aided Internet Translator)
796          */
797         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
798                 if (ip_keepfaith) {
799                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
800                                 goto ours;
801                 }
802                 m_freem(m);
803                 return;
804         }
805
806         /*
807          * Not for us; forward if possible and desirable.
808          */
809         if (!ipforwarding) {
810                 ipstat.ips_cantforward++;
811                 m_freem(m);
812         } else {
813 #ifdef IPSEC
814                 /*
815                  * Enforce inbound IPsec SPD.
816                  */
817                 if (ipsec4_in_reject(m, NULL)) {
818                         ipsecstat.in_polvio++;
819                         goto bad;
820                 }
821 #endif
822 #ifdef FAST_IPSEC
823                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
824                 crit_enter();
825                 if (mtag != NULL) {
826                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
827                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
828                 } else {
829                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
830                                                    IP_FORWARDING, &error);
831                 }
832                 if (sp == NULL) {       /* NB: can happen if error */
833                         crit_exit();
834                         /*XXX error stat???*/
835                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
836                         goto bad;
837                 }
838
839                 /*
840                  * Check security policy against packet attributes.
841                  */
842                 error = ipsec_in_reject(sp, m);
843                 KEY_FREESP(&sp);
844                 crit_exit();
845                 if (error) {
846                         ipstat.ips_cantforward++;
847                         goto bad;
848                 }
849 #endif
850                 ip_forward(m, using_srcrt, args.next_hop);
851         }
852         return;
853
854 ours:
855
856         /*
857          * IPSTEALTH: Process non-routing options only
858          * if the packet is destined for us.
859          */
860         if (ipstealth &&
861             hlen > sizeof(struct ip) &&
862             ip_dooptions(m, 1, args.next_hop))
863                 return;
864
865         /* Count the packet in the ip address stats */
866         if (ia != NULL) {
867                 ia->ia_ifa.if_ipackets++;
868                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
869         }
870
871         /*
872          * If offset or IP_MF are set, must reassemble.
873          * Otherwise, nothing need be done.
874          * (We could look in the reassembly queue to see
875          * if the packet was previously fragmented,
876          * but it's not worth the time; just let them time out.)
877          */
878         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
879
880                 /* If maxnipq is 0, never accept fragments. */
881                 if (maxnipq == 0) {
882                         ipstat.ips_fragments++;
883                         ipstat.ips_fragdropped++;
884                         goto bad;
885                 }
886
887                 sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
888                 /*
889                  * Look for queue of fragments
890                  * of this datagram.
891                  */
892                 for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
893                         if (ip->ip_id == fp->ipq_id &&
894                             ip->ip_src.s_addr == fp->ipq_src.s_addr &&
895                             ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
896                             ip->ip_p == fp->ipq_p)
897                                 goto found;
898
899                 fp = NULL;
900
901                 /*
902                  * Enforce upper bound on number of fragmented packets
903                  * for which we attempt reassembly;
904                  * If maxnipq is -1, accept all fragments without limitation.
905                  */
906                 if ((nipq > maxnipq) && (maxnipq > 0)) {
907                         /*
908                          * drop something from the tail of the current queue
909                          * before proceeding further
910                          */
911                         if (ipq[sum].prev == &ipq[sum]) {   /* gak */
912                                 for (i = 0; i < IPREASS_NHASH; i++) {
913                                         if (ipq[i].prev != &ipq[i]) {
914                                                 ipstat.ips_fragtimeout +=
915                                                     ipq[i].prev->ipq_nfrags;
916                                                 ip_freef(ipq[i].prev);
917                                                 break;
918                                         }
919                                 }
920                         } else {
921                                 ipstat.ips_fragtimeout +=
922                                     ipq[sum].prev->ipq_nfrags;
923                                 ip_freef(ipq[sum].prev);
924                         }
925                 }
926 found:
927                 /*
928                  * Adjust ip_len to not reflect header,
929                  * convert offset of this to bytes.
930                  */
931                 ip->ip_len -= hlen;
932                 if (ip->ip_off & IP_MF) {
933                         /*
934                          * Make sure that fragments have a data length
935                          * that's a non-zero multiple of 8 bytes.
936                          */
937                         if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
938                                 ipstat.ips_toosmall++; /* XXX */
939                                 goto bad;
940                         }
941                         m->m_flags |= M_FRAG;
942                 } else
943                         m->m_flags &= ~M_FRAG;
944                 ip->ip_off <<= 3;
945
946                 /*
947                  * Attempt reassembly; if it succeeds, proceed.
948                  * ip_reass() will return a different mbuf, and update
949                  * the divert info in divert_info.
950                  */
951                 ipstat.ips_fragments++;
952                 m->m_pkthdr.header = ip;
953                 m = ip_reass(m, fp, &ipq[sum], &divert_info);
954                 if (m == NULL)
955                         return;
956                 ipstat.ips_reassembled++;
957                 needredispatch = TRUE;
958                 ip = mtod(m, struct ip *);
959                 /* Get the header length of the reassembled packet */
960                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
961 #ifdef IPDIVERT
962                 /* Restore original checksum before diverting packet */
963                 if (divert_info != 0) {
964                         ip->ip_len += hlen;
965                         ip->ip_len = htons(ip->ip_len);
966                         ip->ip_off = htons(ip->ip_off);
967                         ip->ip_sum = 0;
968                         if (hlen == sizeof(struct ip))
969                                 ip->ip_sum = in_cksum_hdr(ip);
970                         else
971                                 ip->ip_sum = in_cksum(m, hlen);
972                         ip->ip_off = ntohs(ip->ip_off);
973                         ip->ip_len = ntohs(ip->ip_len);
974                         ip->ip_len -= hlen;
975                 }
976 #endif
977         } else {
978                 ip->ip_len -= hlen;
979         }
980
981 #ifdef IPDIVERT
982         /*
983          * Divert or tee packet to the divert protocol if required.
984          */
985         if (divert_info != 0) {
986                 struct mbuf *clone = NULL;
987
988                 /* Clone packet if we're doing a 'tee' */
989                 if ((divert_info & IP_FW_PORT_TEE_FLAG) != 0)
990                         clone = m_dup(m, MB_DONTWAIT);
991
992                 /* Restore packet header fields to original values */
993                 ip->ip_len += hlen;
994                 ip->ip_len = htons(ip->ip_len);
995                 ip->ip_off = htons(ip->ip_off);
996
997                 /* Deliver packet to divert input routine */
998                 divert_packet(m, 1, divert_info & 0xffff);
999                 ipstat.ips_delivered++;
1000
1001                 /* If 'tee', continue with original packet */
1002                 if (clone == NULL)
1003                         return;
1004                 m = clone;
1005                 ip = mtod(m, struct ip *);
1006                 ip->ip_len += hlen;
1007                 /*
1008                  * Jump backwards to complete processing of the
1009                  * packet. But first clear divert_info to avoid
1010                  * entering this block again.
1011                  * We do not need to clear args.divert_rule
1012                  * or args.next_hop as they will not be used.
1013                  *
1014                  * XXX Better safe than sorry, remove the DIVERT tag.
1015                  */
1016                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1017                 if (mtag != NULL)
1018                         m_tag_delete(m, mtag);
1019                 
1020                 divert_info = 0;
1021                 goto pass;
1022         }
1023 #endif
1024
1025 #ifdef IPSEC
1026         /*
1027          * enforce IPsec policy checking if we are seeing last header.
1028          * note that we do not visit this with protocols with pcb layer
1029          * code - like udp/tcp/raw ip.
1030          */
1031         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
1032             ipsec4_in_reject(m, NULL)) {
1033                 ipsecstat.in_polvio++;
1034                 goto bad;
1035         }
1036 #endif
1037 #if FAST_IPSEC
1038         /*
1039          * enforce IPsec policy checking if we are seeing last header.
1040          * note that we do not visit this with protocols with pcb layer
1041          * code - like udp/tcp/raw ip.
1042          */
1043         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1044                 /*
1045                  * Check if the packet has already had IPsec processing
1046                  * done.  If so, then just pass it along.  This tag gets
1047                  * set during AH, ESP, etc. input handling, before the
1048                  * packet is returned to the ip input queue for delivery.
1049                  */
1050                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1051                 crit_enter();
1052                 if (mtag != NULL) {
1053                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
1054                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1055                 } else {
1056                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1057                                                    IP_FORWARDING, &error);
1058                 }
1059                 if (sp != NULL) {
1060                         /*
1061                          * Check security policy against packet attributes.
1062                          */
1063                         error = ipsec_in_reject(sp, m);
1064                         KEY_FREESP(&sp);
1065                 } else {
1066                         /* XXX error stat??? */
1067                         error = EINVAL;
1068 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1069                         goto bad;
1070                 }
1071                 crit_exit();
1072                 if (error)
1073                         goto bad;
1074         }
1075 #endif /* FAST_IPSEC */
1076
1077         ipstat.ips_delivered++;
1078         if (needredispatch) {
1079                 struct netmsg_transport_packet *msg;
1080                 lwkt_port_t port;
1081
1082                 ip->ip_off = htons(ip->ip_off);
1083                 ip->ip_len = htons(ip->ip_len);
1084                 port = ip_mport_in(&m);
1085                 if (port == NULL)
1086                         return;
1087
1088                 msg = kmalloc(sizeof(struct netmsg_transport_packet), M_LWKTMSG,
1089                              M_INTWAIT | M_NULLOK);
1090                 if (msg == NULL)
1091                         goto bad;
1092
1093                 netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0,
1094                             transport_processing_handler);
1095                 msg->nm_hlen = hlen;
1096                 msg->nm_hasnexthop = (args.next_hop != NULL);
1097                 if (msg->nm_hasnexthop)
1098                         msg->nm_nexthop = *args.next_hop;  /* structure copy */
1099
1100                 msg->nm_mbuf = m;
1101                 ip = mtod(m, struct ip *);
1102                 ip->ip_len = ntohs(ip->ip_len);
1103                 ip->ip_off = ntohs(ip->ip_off);
1104                 lwkt_sendmsg(port, &msg->nm_netmsg.nm_lmsg);
1105         } else {
1106                 transport_processing_oncpu(m, hlen, ip, args.next_hop);
1107         }
1108         return;
1109
1110 bad:
1111         m_freem(m);
1112 }
1113
1114 /*
1115  * Take incoming datagram fragment and try to reassemble it into
1116  * whole datagram.  If a chain for reassembly of this datagram already
1117  * exists, then it is given as fp; otherwise have to make a chain.
1118  *
1119  * When IPDIVERT enabled, keep additional state with each packet that
1120  * tells us if we need to divert or tee the packet we're building.
1121  * In particular, *divinfo includes the port and TEE flag.
1122  */
1123
1124 static struct mbuf *
1125 ip_reass(struct mbuf *m, struct ipq *fp, struct ipq *where,
1126          u_int32_t *divinfo)
1127 {
1128         struct ip *ip = mtod(m, struct ip *);
1129         struct mbuf *p = NULL, *q, *nq;
1130         struct mbuf *n;
1131         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1132         int i, next;
1133 #ifdef IPDIVERT
1134         struct m_tag *mtag;
1135 #endif
1136
1137         /*
1138          * If the hardware has not done csum over this fragment
1139          * then csum_data is not valid at all.
1140          */
1141         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1142             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1143                 m->m_pkthdr.csum_data = 0;
1144                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1145         }
1146
1147         /*
1148          * Presence of header sizes in mbufs
1149          * would confuse code below.
1150          */
1151         m->m_data += hlen;
1152         m->m_len -= hlen;
1153
1154         /*
1155          * If first fragment to arrive, create a reassembly queue.
1156          */
1157         if (fp == NULL) {
1158                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1159                         goto dropfrag;
1160                 insque(fp, where);
1161                 nipq++;
1162                 fp->ipq_nfrags = 1;
1163                 fp->ipq_ttl = IPFRAGTTL;
1164                 fp->ipq_p = ip->ip_p;
1165                 fp->ipq_id = ip->ip_id;
1166                 fp->ipq_src = ip->ip_src;
1167                 fp->ipq_dst = ip->ip_dst;
1168                 fp->ipq_frags = m;
1169                 m->m_nextpkt = NULL;
1170 #ifdef IPDIVERT
1171                 fp->ipq_div_info = 0;
1172 #endif
1173                 goto inserted;
1174         } else {
1175                 fp->ipq_nfrags++;
1176         }
1177
1178 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1179
1180         /*
1181          * Find a segment which begins after this one does.
1182          */
1183         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1184                 if (GETIP(q)->ip_off > ip->ip_off)
1185                         break;
1186
1187         /*
1188          * If there is a preceding segment, it may provide some of
1189          * our data already.  If so, drop the data from the incoming
1190          * segment.  If it provides all of our data, drop us, otherwise
1191          * stick new segment in the proper place.
1192          *
1193          * If some of the data is dropped from the the preceding
1194          * segment, then it's checksum is invalidated.
1195          */
1196         if (p) {
1197                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1198                 if (i > 0) {
1199                         if (i >= ip->ip_len)
1200                                 goto dropfrag;
1201                         m_adj(m, i);
1202                         m->m_pkthdr.csum_flags = 0;
1203                         ip->ip_off += i;
1204                         ip->ip_len -= i;
1205                 }
1206                 m->m_nextpkt = p->m_nextpkt;
1207                 p->m_nextpkt = m;
1208         } else {
1209                 m->m_nextpkt = fp->ipq_frags;
1210                 fp->ipq_frags = m;
1211         }
1212
1213         /*
1214          * While we overlap succeeding segments trim them or,
1215          * if they are completely covered, dequeue them.
1216          */
1217         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1218              q = nq) {
1219                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1220                 if (i < GETIP(q)->ip_len) {
1221                         GETIP(q)->ip_len -= i;
1222                         GETIP(q)->ip_off += i;
1223                         m_adj(q, i);
1224                         q->m_pkthdr.csum_flags = 0;
1225                         break;
1226                 }
1227                 nq = q->m_nextpkt;
1228                 m->m_nextpkt = nq;
1229                 ipstat.ips_fragdropped++;
1230                 fp->ipq_nfrags--;
1231                 q->m_nextpkt = NULL;
1232                 m_freem(q);
1233         }
1234
1235 inserted:
1236
1237 #ifdef IPDIVERT
1238         /*
1239          * Transfer firewall instructions to the fragment structure.
1240          * Only trust info in the fragment at offset 0.
1241          */
1242         if (ip->ip_off == 0) {
1243                 fp->ipq_div_info = *divinfo;
1244         } else {
1245                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1246                 if (mtag != NULL)
1247                         m_tag_delete(m, mtag);
1248         }
1249         *divinfo = 0;
1250 #endif
1251
1252         /*
1253          * Check for complete reassembly and perform frag per packet
1254          * limiting.
1255          *
1256          * Frag limiting is performed here so that the nth frag has
1257          * a chance to complete the packet before we drop the packet.
1258          * As a result, n+1 frags are actually allowed per packet, but
1259          * only n will ever be stored. (n = maxfragsperpacket.)
1260          *
1261          */
1262         next = 0;
1263         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1264                 if (GETIP(q)->ip_off != next) {
1265                         if (fp->ipq_nfrags > maxfragsperpacket) {
1266                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1267                                 ip_freef(fp);
1268                         }
1269                         return (NULL);
1270                 }
1271                 next += GETIP(q)->ip_len;
1272         }
1273         /* Make sure the last packet didn't have the IP_MF flag */
1274         if (p->m_flags & M_FRAG) {
1275                 if (fp->ipq_nfrags > maxfragsperpacket) {
1276                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1277                         ip_freef(fp);
1278                 }
1279                 return (NULL);
1280         }
1281
1282         /*
1283          * Reassembly is complete.  Make sure the packet is a sane size.
1284          */
1285         q = fp->ipq_frags;
1286         ip = GETIP(q);
1287         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1288                 ipstat.ips_toolong++;
1289                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1290                 ip_freef(fp);
1291                 return (NULL);
1292         }
1293
1294         /*
1295          * Concatenate fragments.
1296          */
1297         m = q;
1298         n = m->m_next;
1299         m->m_next = NULL;
1300         m_cat(m, n);
1301         nq = q->m_nextpkt;
1302         q->m_nextpkt = NULL;
1303         for (q = nq; q != NULL; q = nq) {
1304                 nq = q->m_nextpkt;
1305                 q->m_nextpkt = NULL;
1306                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1307                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1308                 m_cat(m, q);
1309         }
1310
1311         /*
1312          * Clean up the 1's complement checksum.  Carry over 16 bits must
1313          * be added back.  This assumes no more then 65535 packet fragments
1314          * were reassembled.  A second carry can also occur (but not a third).
1315          */
1316         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1317                                 (m->m_pkthdr.csum_data >> 16);
1318         if (m->m_pkthdr.csum_data > 0xFFFF)
1319                 m->m_pkthdr.csum_data -= 0xFFFF;
1320
1321
1322 #ifdef IPDIVERT
1323         /*
1324          * Extract firewall instructions from the fragment structure.
1325          */
1326         *divinfo = fp->ipq_div_info;
1327 #endif
1328
1329         /*
1330          * Create header for new ip packet by
1331          * modifying header of first packet;
1332          * dequeue and discard fragment reassembly header.
1333          * Make header visible.
1334          */
1335         ip->ip_len = next;
1336         ip->ip_src = fp->ipq_src;
1337         ip->ip_dst = fp->ipq_dst;
1338         remque(fp);
1339         nipq--;
1340         mpipe_free(&ipq_mpipe, fp);
1341         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1342         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1343         /* some debugging cruft by sklower, below, will go away soon */
1344         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1345                 int plen = 0;
1346
1347                 for (n = m; n; n = n->m_next)
1348                         plen += n->m_len;
1349                 m->m_pkthdr.len = plen;
1350         }
1351         return (m);
1352
1353 dropfrag:
1354 #ifdef IPDIVERT
1355         *divinfo = 0;
1356 #endif
1357         ipstat.ips_fragdropped++;
1358         if (fp != NULL)
1359                 fp->ipq_nfrags--;
1360         m_freem(m);
1361         return (NULL);
1362
1363 #undef GETIP
1364 }
1365
1366 /*
1367  * Free a fragment reassembly header and all
1368  * associated datagrams.
1369  */
1370 static void
1371 ip_freef(struct ipq *fp)
1372 {
1373         struct mbuf *q;
1374
1375         while (fp->ipq_frags) {
1376                 q = fp->ipq_frags;
1377                 fp->ipq_frags = q->m_nextpkt;
1378                 q->m_nextpkt = NULL;
1379                 m_freem(q);
1380         }
1381         remque(fp);
1382         mpipe_free(&ipq_mpipe, fp);
1383         nipq--;
1384 }
1385
1386 /*
1387  * IP timer processing;
1388  * if a timer expires on a reassembly
1389  * queue, discard it.
1390  */
1391 void
1392 ip_slowtimo(void)
1393 {
1394         struct ipq *fp;
1395         int i;
1396
1397         crit_enter();
1398         for (i = 0; i < IPREASS_NHASH; i++) {
1399                 fp = ipq[i].next;
1400                 if (fp == NULL)
1401                         continue;
1402                 while (fp != &ipq[i]) {
1403                         --fp->ipq_ttl;
1404                         fp = fp->next;
1405                         if (fp->prev->ipq_ttl == 0) {
1406                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1407                                 ip_freef(fp->prev);
1408                         }
1409                 }
1410         }
1411         /*
1412          * If we are over the maximum number of fragments
1413          * (due to the limit being lowered), drain off
1414          * enough to get down to the new limit.
1415          */
1416         if (maxnipq >= 0 && nipq > maxnipq) {
1417                 for (i = 0; i < IPREASS_NHASH; i++) {
1418                         while (nipq > maxnipq &&
1419                                 (ipq[i].next != &ipq[i])) {
1420                                 ipstat.ips_fragdropped +=
1421                                     ipq[i].next->ipq_nfrags;
1422                                 ip_freef(ipq[i].next);
1423                         }
1424                 }
1425         }
1426         ipflow_slowtimo();
1427         crit_exit();
1428 }
1429
1430 /*
1431  * Drain off all datagram fragments.
1432  */
1433 void
1434 ip_drain(void)
1435 {
1436         int i;
1437
1438         for (i = 0; i < IPREASS_NHASH; i++) {
1439                 while (ipq[i].next != &ipq[i]) {
1440                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1441                         ip_freef(ipq[i].next);
1442                 }
1443         }
1444         in_rtqdrain();
1445 }
1446
1447 /*
1448  * Do option processing on a datagram,
1449  * possibly discarding it if bad options are encountered,
1450  * or forwarding it if source-routed.
1451  * The pass argument is used when operating in the IPSTEALTH
1452  * mode to tell what options to process:
1453  * [LS]SRR (pass 0) or the others (pass 1).
1454  * The reason for as many as two passes is that when doing IPSTEALTH,
1455  * non-routing options should be processed only if the packet is for us.
1456  * Returns 1 if packet has been forwarded/freed,
1457  * 0 if the packet should be processed further.
1458  */
1459 static int
1460 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1461 {
1462         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1463         struct ip *ip = mtod(m, struct ip *);
1464         u_char *cp;
1465         struct in_ifaddr *ia;
1466         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1467         boolean_t forward = FALSE;
1468         struct in_addr *sin, dst;
1469         n_time ntime;
1470
1471         dst = ip->ip_dst;
1472         cp = (u_char *)(ip + 1);
1473         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1474         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1475                 opt = cp[IPOPT_OPTVAL];
1476                 if (opt == IPOPT_EOL)
1477                         break;
1478                 if (opt == IPOPT_NOP)
1479                         optlen = 1;
1480                 else {
1481                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1482                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1483                                 goto bad;
1484                         }
1485                         optlen = cp[IPOPT_OLEN];
1486                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1487                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1488                                 goto bad;
1489                         }
1490                 }
1491                 switch (opt) {
1492
1493                 default:
1494                         break;
1495
1496                 /*
1497                  * Source routing with record.
1498                  * Find interface with current destination address.
1499                  * If none on this machine then drop if strictly routed,
1500                  * or do nothing if loosely routed.
1501                  * Record interface address and bring up next address
1502                  * component.  If strictly routed make sure next
1503                  * address is on directly accessible net.
1504                  */
1505                 case IPOPT_LSRR:
1506                 case IPOPT_SSRR:
1507                         if (ipstealth && pass > 0)
1508                                 break;
1509                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1510                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1511                                 goto bad;
1512                         }
1513                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1514                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1515                                 goto bad;
1516                         }
1517                         ipaddr.sin_addr = ip->ip_dst;
1518                         ia = (struct in_ifaddr *)
1519                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1520                         if (ia == NULL) {
1521                                 if (opt == IPOPT_SSRR) {
1522                                         type = ICMP_UNREACH;
1523                                         code = ICMP_UNREACH_SRCFAIL;
1524                                         goto bad;
1525                                 }
1526                                 if (!ip_dosourceroute)
1527                                         goto nosourcerouting;
1528                                 /*
1529                                  * Loose routing, and not at next destination
1530                                  * yet; nothing to do except forward.
1531                                  */
1532                                 break;
1533                         }
1534                         off--;                  /* 0 origin */
1535                         if (off > optlen - (int)sizeof(struct in_addr)) {
1536                                 /*
1537                                  * End of source route.  Should be for us.
1538                                  */
1539                                 if (!ip_acceptsourceroute)
1540                                         goto nosourcerouting;
1541                                 save_rte(cp, ip->ip_src);
1542                                 break;
1543                         }
1544                         if (ipstealth)
1545                                 goto dropit;
1546                         if (!ip_dosourceroute) {
1547                                 if (ipforwarding) {
1548                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1549
1550                                         /*
1551                                          * Acting as a router, so generate ICMP
1552                                          */
1553 nosourcerouting:
1554                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1555                                         log(LOG_WARNING,
1556                                             "attempted source route from %s to %s\n",
1557                                             inet_ntoa(ip->ip_src), buf);
1558                                         type = ICMP_UNREACH;
1559                                         code = ICMP_UNREACH_SRCFAIL;
1560                                         goto bad;
1561                                 } else {
1562                                         /*
1563                                          * Not acting as a router,
1564                                          * so silently drop.
1565                                          */
1566 dropit:
1567                                         ipstat.ips_cantforward++;
1568                                         m_freem(m);
1569                                         return (1);
1570                                 }
1571                         }
1572
1573                         /*
1574                          * locate outgoing interface
1575                          */
1576                         memcpy(&ipaddr.sin_addr, cp + off,
1577                             sizeof ipaddr.sin_addr);
1578
1579                         if (opt == IPOPT_SSRR) {
1580 #define INA     struct in_ifaddr *
1581 #define SA      struct sockaddr *
1582                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1583                                                                         == NULL)
1584                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1585                         } else
1586                                 ia = ip_rtaddr(ipaddr.sin_addr,
1587                                                &ipforward_rt[mycpuid]);
1588                         if (ia == NULL) {
1589                                 type = ICMP_UNREACH;
1590                                 code = ICMP_UNREACH_SRCFAIL;
1591                                 goto bad;
1592                         }
1593                         ip->ip_dst = ipaddr.sin_addr;
1594                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1595                             sizeof(struct in_addr));
1596                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1597                         /*
1598                          * Let ip_intr's mcast routing check handle mcast pkts
1599                          */
1600                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1601                         break;
1602
1603                 case IPOPT_RR:
1604                         if (ipstealth && pass == 0)
1605                                 break;
1606                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1607                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1608                                 goto bad;
1609                         }
1610                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1611                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1612                                 goto bad;
1613                         }
1614                         /*
1615                          * If no space remains, ignore.
1616                          */
1617                         off--;                  /* 0 origin */
1618                         if (off > optlen - (int)sizeof(struct in_addr))
1619                                 break;
1620                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1621                             sizeof ipaddr.sin_addr);
1622                         /*
1623                          * locate outgoing interface; if we're the destination,
1624                          * use the incoming interface (should be same).
1625                          */
1626                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1627                             (ia = ip_rtaddr(ipaddr.sin_addr,
1628                                             &ipforward_rt[mycpuid]))
1629                                                                      == NULL) {
1630                                 type = ICMP_UNREACH;
1631                                 code = ICMP_UNREACH_HOST;
1632                                 goto bad;
1633                         }
1634                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1635                             sizeof(struct in_addr));
1636                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1637                         break;
1638
1639                 case IPOPT_TS:
1640                         if (ipstealth && pass == 0)
1641                                 break;
1642                         code = cp - (u_char *)ip;
1643                         if (optlen < 4 || optlen > 40) {
1644                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1645                                 goto bad;
1646                         }
1647                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1648                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1649                                 goto bad;
1650                         }
1651                         if (off > optlen - (int)sizeof(int32_t)) {
1652                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1653                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1654                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1655                                         goto bad;
1656                                 }
1657                                 break;
1658                         }
1659                         off--;                          /* 0 origin */
1660                         sin = (struct in_addr *)(cp + off);
1661                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1662
1663                         case IPOPT_TS_TSONLY:
1664                                 break;
1665
1666                         case IPOPT_TS_TSANDADDR:
1667                                 if (off + sizeof(n_time) +
1668                                     sizeof(struct in_addr) > optlen) {
1669                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1670                                         goto bad;
1671                                 }
1672                                 ipaddr.sin_addr = dst;
1673                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1674                                                             m->m_pkthdr.rcvif);
1675                                 if (ia == NULL)
1676                                         continue;
1677                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1678                                     sizeof(struct in_addr));
1679                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1680                                 off += sizeof(struct in_addr);
1681                                 break;
1682
1683                         case IPOPT_TS_PRESPEC:
1684                                 if (off + sizeof(n_time) +
1685                                     sizeof(struct in_addr) > optlen) {
1686                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1687                                         goto bad;
1688                                 }
1689                                 memcpy(&ipaddr.sin_addr, sin,
1690                                     sizeof(struct in_addr));
1691                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1692                                         continue;
1693                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1694                                 off += sizeof(struct in_addr);
1695                                 break;
1696
1697                         default:
1698                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1699                                 goto bad;
1700                         }
1701                         ntime = iptime();
1702                         memcpy(cp + off, &ntime, sizeof(n_time));
1703                         cp[IPOPT_OFFSET] += sizeof(n_time);
1704                 }
1705         }
1706         if (forward && ipforwarding) {
1707                 ip_forward(m, TRUE, next_hop);
1708                 return (1);
1709         }
1710         return (0);
1711 bad:
1712         icmp_error(m, type, code, 0, 0);
1713         ipstat.ips_badoptions++;
1714         return (1);
1715 }
1716
1717 /*
1718  * Given address of next destination (final or next hop),
1719  * return internet address info of interface to be used to get there.
1720  */
1721 struct in_ifaddr *
1722 ip_rtaddr(struct in_addr dst, struct route *ro)
1723 {
1724         struct sockaddr_in *sin;
1725
1726         sin = (struct sockaddr_in *)&ro->ro_dst;
1727
1728         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1729                 if (ro->ro_rt != NULL) {
1730                         RTFREE(ro->ro_rt);
1731                         ro->ro_rt = NULL;
1732                 }
1733                 sin->sin_family = AF_INET;
1734                 sin->sin_len = sizeof *sin;
1735                 sin->sin_addr = dst;
1736                 rtalloc_ign(ro, RTF_PRCLONING);
1737         }
1738
1739         if (ro->ro_rt == NULL)
1740                 return (NULL);
1741
1742         return (ifatoia(ro->ro_rt->rt_ifa));
1743 }
1744
1745 /*
1746  * Save incoming source route for use in replies,
1747  * to be picked up later by ip_srcroute if the receiver is interested.
1748  */
1749 void
1750 save_rte(u_char *option, struct in_addr dst)
1751 {
1752         unsigned olen;
1753
1754         olen = option[IPOPT_OLEN];
1755 #ifdef DIAGNOSTIC
1756         if (ipprintfs)
1757                 kprintf("save_rte: olen %d\n", olen);
1758 #endif
1759         if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst)))
1760                 return;
1761         bcopy(option, ip_srcrt.srcopt, olen);
1762         ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1763         ip_srcrt.dst = dst;
1764 }
1765
1766 /*
1767  * Retrieve incoming source route for use in replies,
1768  * in the same form used by setsockopt.
1769  * The first hop is placed before the options, will be removed later.
1770  */
1771 struct mbuf *
1772 ip_srcroute(void)
1773 {
1774         struct in_addr *p, *q;
1775         struct mbuf *m;
1776
1777         if (ip_nhops == 0)
1778                 return (NULL);
1779         m = m_get(MB_DONTWAIT, MT_HEADER);
1780         if (m == NULL)
1781                 return (NULL);
1782
1783 #define OPTSIZ  (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt))
1784
1785         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1786         m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) +
1787             OPTSIZ;
1788 #ifdef DIAGNOSTIC
1789         if (ipprintfs)
1790                 kprintf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
1791 #endif
1792
1793         /*
1794          * First save first hop for return route
1795          */
1796         p = &ip_srcrt.route[ip_nhops - 1];
1797         *(mtod(m, struct in_addr *)) = *p--;
1798 #ifdef DIAGNOSTIC
1799         if (ipprintfs)
1800                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1801 #endif
1802
1803         /*
1804          * Copy option fields and padding (nop) to mbuf.
1805          */
1806         ip_srcrt.nop = IPOPT_NOP;
1807         ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1808         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &ip_srcrt.nop,
1809             OPTSIZ);
1810         q = (struct in_addr *)(mtod(m, caddr_t) +
1811             sizeof(struct in_addr) + OPTSIZ);
1812 #undef OPTSIZ
1813         /*
1814          * Record return path as an IP source route,
1815          * reversing the path (pointers are now aligned).
1816          */
1817         while (p >= ip_srcrt.route) {
1818 #ifdef DIAGNOSTIC
1819                 if (ipprintfs)
1820                         kprintf(" %x", ntohl(q->s_addr));
1821 #endif
1822                 *q++ = *p--;
1823         }
1824         /*
1825          * Last hop goes to final destination.
1826          */
1827         *q = ip_srcrt.dst;
1828 #ifdef DIAGNOSTIC
1829         if (ipprintfs)
1830                 kprintf(" %x\n", ntohl(q->s_addr));
1831 #endif
1832         return (m);
1833 }
1834
1835 /*
1836  * Strip out IP options.
1837  */
1838 void
1839 ip_stripoptions(struct mbuf *m)
1840 {
1841         int datalen;
1842         struct ip *ip = mtod(m, struct ip *);
1843         caddr_t opts;
1844         int optlen;
1845
1846         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1847         opts = (caddr_t)(ip + 1);
1848         datalen = m->m_len - (sizeof(struct ip) + optlen);
1849         bcopy(opts + optlen, opts, datalen);
1850         m->m_len -= optlen;
1851         if (m->m_flags & M_PKTHDR)
1852                 m->m_pkthdr.len -= optlen;
1853         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1854 }
1855
1856 u_char inetctlerrmap[PRC_NCMDS] = {
1857         0,              0,              0,              0,
1858         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1859         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1860         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1861         0,              0,              0,              0,
1862         ENOPROTOOPT,    ECONNREFUSED
1863 };
1864
1865 /*
1866  * Forward a packet.  If some error occurs return the sender
1867  * an icmp packet.  Note we can't always generate a meaningful
1868  * icmp message because icmp doesn't have a large enough repertoire
1869  * of codes and types.
1870  *
1871  * If not forwarding, just drop the packet.  This could be confusing
1872  * if ipforwarding was zero but some routing protocol was advancing
1873  * us as a gateway to somewhere.  However, we must let the routing
1874  * protocol deal with that.
1875  *
1876  * The using_srcrt parameter indicates whether the packet is being forwarded
1877  * via a source route.
1878  */
1879 static void
1880 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1881 {
1882         struct ip *ip = mtod(m, struct ip *);
1883         struct sockaddr_in *ipforward_rtaddr;
1884         struct rtentry *rt;
1885         int error, type = 0, code = 0, destmtu = 0;
1886         struct mbuf *mcopy;
1887         n_long dest;
1888         struct in_addr pkt_dst;
1889         struct m_hdr tag;
1890         struct route *cache_rt = &ipforward_rt[mycpuid];
1891
1892         dest = INADDR_ANY;
1893         /*
1894          * Cache the destination address of the packet; this may be
1895          * changed by use of 'ipfw fwd'.
1896          */
1897         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1898
1899 #ifdef DIAGNOSTIC
1900         if (ipprintfs)
1901                 kprintf("forward: src %x dst %x ttl %x\n",
1902                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1903 #endif
1904
1905         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1906                 ipstat.ips_cantforward++;
1907                 m_freem(m);
1908                 return;
1909         }
1910         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1911                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1912                 return;
1913         }
1914
1915         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1916         if (cache_rt->ro_rt == NULL ||
1917             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1918                 if (cache_rt->ro_rt != NULL) {
1919                         RTFREE(cache_rt->ro_rt);
1920                         cache_rt->ro_rt = NULL;
1921                 }
1922                 ipforward_rtaddr->sin_family = AF_INET;
1923                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1924                 ipforward_rtaddr->sin_addr = pkt_dst;
1925                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1926                 if (cache_rt->ro_rt == NULL) {
1927                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1928                         return;
1929                 }
1930         }
1931         rt = cache_rt->ro_rt;
1932
1933         /*
1934          * Save the IP header and at most 8 bytes of the payload,
1935          * in case we need to generate an ICMP message to the src.
1936          *
1937          * XXX this can be optimized a lot by saving the data in a local
1938          * buffer on the stack (72 bytes at most), and only allocating the
1939          * mbuf if really necessary. The vast majority of the packets
1940          * are forwarded without having to send an ICMP back (either
1941          * because unnecessary, or because rate limited), so we are
1942          * really we are wasting a lot of work here.
1943          *
1944          * We don't use m_copy() because it might return a reference
1945          * to a shared cluster. Both this function and ip_output()
1946          * assume exclusive access to the IP header in `m', so any
1947          * data in a cluster may change before we reach icmp_error().
1948          */
1949         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1950         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1951                 /*
1952                  * It's probably ok if the pkthdr dup fails (because
1953                  * the deep copy of the tag chain failed), but for now
1954                  * be conservative and just discard the copy since
1955                  * code below may some day want the tags.
1956                  */
1957                 m_free(mcopy);
1958                 mcopy = NULL;
1959         }
1960         if (mcopy != NULL) {
1961                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1962                     (int)ip->ip_len);
1963                 mcopy->m_pkthdr.len = mcopy->m_len;
1964                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1965         }
1966
1967         if (!ipstealth)
1968                 ip->ip_ttl -= IPTTLDEC;
1969
1970         /*
1971          * If forwarding packet using same interface that it came in on,
1972          * perhaps should send a redirect to sender to shortcut a hop.
1973          * Only send redirect if source is sending directly to us,
1974          * and if packet was not source routed (or has any options).
1975          * Also, don't send redirect if forwarding using a default route
1976          * or a route modified by a redirect.
1977          */
1978         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1979             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1980             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1981             ipsendredirects && !using_srcrt && next_hop == NULL) {
1982                 u_long src = ntohl(ip->ip_src.s_addr);
1983                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1984
1985                 if (rt_ifa != NULL &&
1986                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1987                         if (rt->rt_flags & RTF_GATEWAY)
1988                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1989                         else
1990                                 dest = pkt_dst.s_addr;
1991                         /*
1992                          * Router requirements says to only send
1993                          * host redirects.
1994                          */
1995                         type = ICMP_REDIRECT;
1996                         code = ICMP_REDIRECT_HOST;
1997 #ifdef DIAGNOSTIC
1998                         if (ipprintfs)
1999                                 kprintf("redirect (%d) to %x\n", code, dest);
2000 #endif
2001                 }
2002         }
2003
2004         if (next_hop != NULL) {
2005                 /* Pass IPFORWARD info if available */
2006                 tag.mh_type = MT_TAG;
2007                 tag.mh_flags = PACKET_TAG_IPFORWARD;
2008                 tag.mh_data = (caddr_t)next_hop;
2009                 tag.mh_next = m;
2010                 m = (struct mbuf *)&tag;
2011         }
2012
2013         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL,
2014                           NULL);
2015         if (error == 0) {
2016                 ipstat.ips_forward++;
2017                 if (type == 0) {
2018                         if (mcopy) {
2019                                 /* ipflow_create() will free mcopy */
2020                                 ipflow_create(cache_rt, mcopy);
2021                         }
2022                         return;         /* most common case */
2023                 } else {
2024                         ipstat.ips_redirectsent++;
2025                 }
2026         } else {
2027                 ipstat.ips_cantforward++;
2028         }
2029
2030         if (mcopy == NULL)
2031                 return;
2032
2033         /*
2034          * Send ICMP message.
2035          */
2036
2037         switch (error) {
2038
2039         case 0:                         /* forwarded, but need redirect */
2040                 /* type, code set above */
2041                 break;
2042
2043         case ENETUNREACH:               /* shouldn't happen, checked above */
2044         case EHOSTUNREACH:
2045         case ENETDOWN:
2046         case EHOSTDOWN:
2047         default:
2048                 type = ICMP_UNREACH;
2049                 code = ICMP_UNREACH_HOST;
2050                 break;
2051
2052         case EMSGSIZE:
2053                 type = ICMP_UNREACH;
2054                 code = ICMP_UNREACH_NEEDFRAG;
2055 #ifdef IPSEC
2056                 /*
2057                  * If the packet is routed over IPsec tunnel, tell the
2058                  * originator the tunnel MTU.
2059                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2060                  * XXX quickhack!!!
2061                  */
2062                 if (cache_rt->ro_rt != NULL) {
2063                         struct secpolicy *sp = NULL;
2064                         int ipsecerror;
2065                         int ipsechdr;
2066                         struct route *ro;
2067
2068                         sp = ipsec4_getpolicybyaddr(mcopy,
2069                                                     IPSEC_DIR_OUTBOUND,
2070                                                     IP_FORWARDING,
2071                                                     &ipsecerror);
2072
2073                         if (sp == NULL)
2074                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2075                         else {
2076                                 /* count IPsec header size */
2077                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2078                                                          IPSEC_DIR_OUTBOUND,
2079                                                          NULL);
2080
2081                                 /*
2082                                  * find the correct route for outer IPv4
2083                                  * header, compute tunnel MTU.
2084                                  *
2085                                  */
2086                                 if (sp->req != NULL && sp->req->sav != NULL &&
2087                                     sp->req->sav->sah != NULL) {
2088                                         ro = &sp->req->sav->sah->sa_route;
2089                                         if (ro->ro_rt != NULL &&
2090                                             ro->ro_rt->rt_ifp != NULL) {
2091                                                 destmtu =
2092                                                     ro->ro_rt->rt_ifp->if_mtu;
2093                                                 destmtu -= ipsechdr;
2094                                         }
2095                                 }
2096
2097                                 key_freesp(sp);
2098                         }
2099                 }
2100 #elif FAST_IPSEC
2101                 /*
2102                  * If the packet is routed over IPsec tunnel, tell the
2103                  * originator the tunnel MTU.
2104                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2105                  * XXX quickhack!!!
2106                  */
2107                 if (cache_rt->ro_rt != NULL) {
2108                         struct secpolicy *sp = NULL;
2109                         int ipsecerror;
2110                         int ipsechdr;
2111                         struct route *ro;
2112
2113                         sp = ipsec_getpolicybyaddr(mcopy,
2114                                                    IPSEC_DIR_OUTBOUND,
2115                                                    IP_FORWARDING,
2116                                                    &ipsecerror);
2117
2118                         if (sp == NULL)
2119                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2120                         else {
2121                                 /* count IPsec header size */
2122                                 ipsechdr = ipsec4_hdrsiz(mcopy,
2123                                                          IPSEC_DIR_OUTBOUND,
2124                                                          NULL);
2125
2126                                 /*
2127                                  * find the correct route for outer IPv4
2128                                  * header, compute tunnel MTU.
2129                                  */
2130
2131                                 if (sp->req != NULL &&
2132                                     sp->req->sav != NULL &&
2133                                     sp->req->sav->sah != NULL) {
2134                                         ro = &sp->req->sav->sah->sa_route;
2135                                         if (ro->ro_rt != NULL &&
2136                                             ro->ro_rt->rt_ifp != NULL) {
2137                                                 destmtu =
2138                                                     ro->ro_rt->rt_ifp->if_mtu;
2139                                                 destmtu -= ipsechdr;
2140                                         }
2141                                 }
2142
2143                                 KEY_FREESP(&sp);
2144                         }
2145                 }
2146 #else /* !IPSEC && !FAST_IPSEC */
2147                 if (cache_rt->ro_rt != NULL)
2148                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2149 #endif /*IPSEC*/
2150                 ipstat.ips_cantfrag++;
2151                 break;
2152
2153         case ENOBUFS:
2154                 /*
2155                  * A router should not generate ICMP_SOURCEQUENCH as
2156                  * required in RFC1812 Requirements for IP Version 4 Routers.
2157                  * Source quench could be a big problem under DoS attacks,
2158                  * or if the underlying interface is rate-limited.
2159                  * Those who need source quench packets may re-enable them
2160                  * via the net.inet.ip.sendsourcequench sysctl.
2161                  */
2162                 if (!ip_sendsourcequench) {
2163                         m_freem(mcopy);
2164                         return;
2165                 } else {
2166                         type = ICMP_SOURCEQUENCH;
2167                         code = 0;
2168                 }
2169                 break;
2170
2171         case EACCES:                    /* ipfw denied packet */
2172                 m_freem(mcopy);
2173                 return;
2174         }
2175         icmp_error(mcopy, type, code, dest, destmtu);
2176 }
2177
2178 void
2179 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2180                struct mbuf *m)
2181 {
2182         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2183                 struct timeval tv;
2184
2185                 microtime(&tv);
2186                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2187                     SCM_TIMESTAMP, SOL_SOCKET);
2188                 if (*mp)
2189                         mp = &(*mp)->m_next;
2190         }
2191         if (inp->inp_flags & INP_RECVDSTADDR) {
2192                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2193                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2194                 if (*mp)
2195                         mp = &(*mp)->m_next;
2196         }
2197         if (inp->inp_flags & INP_RECVTTL) {
2198                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2199                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2200                 if (*mp)
2201                         mp = &(*mp)->m_next;
2202         }
2203 #ifdef notyet
2204         /* XXX
2205          * Moving these out of udp_input() made them even more broken
2206          * than they already were.
2207          */
2208         /* options were tossed already */
2209         if (inp->inp_flags & INP_RECVOPTS) {
2210                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2211                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2212                 if (*mp)
2213                         mp = &(*mp)->m_next;
2214         }
2215         /* ip_srcroute doesn't do what we want here, need to fix */
2216         if (inp->inp_flags & INP_RECVRETOPTS) {
2217                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(),
2218                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2219                 if (*mp)
2220                         mp = &(*mp)->m_next;
2221         }
2222 #endif
2223         if (inp->inp_flags & INP_RECVIF) {
2224                 struct ifnet *ifp;
2225                 struct sdlbuf {
2226                         struct sockaddr_dl sdl;
2227                         u_char  pad[32];
2228                 } sdlbuf;
2229                 struct sockaddr_dl *sdp;
2230                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2231
2232                 if (((ifp = m->m_pkthdr.rcvif)) &&
2233                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2234                         sdp = IF_LLSOCKADDR(ifp);
2235                         /*
2236                          * Change our mind and don't try copy.
2237                          */
2238                         if ((sdp->sdl_family != AF_LINK) ||
2239                             (sdp->sdl_len > sizeof(sdlbuf))) {
2240                                 goto makedummy;
2241                         }
2242                         bcopy(sdp, sdl2, sdp->sdl_len);
2243                 } else {
2244 makedummy:
2245                         sdl2->sdl_len =
2246                             offsetof(struct sockaddr_dl, sdl_data[0]);
2247                         sdl2->sdl_family = AF_LINK;
2248                         sdl2->sdl_index = 0;
2249                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2250                 }
2251                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2252                         IP_RECVIF, IPPROTO_IP);
2253                 if (*mp)
2254                         mp = &(*mp)->m_next;
2255         }
2256 }
2257
2258 /*
2259  * XXX these routines are called from the upper part of the kernel.
2260  *
2261  * They could also be moved to ip_mroute.c, since all the RSVP
2262  *  handling is done there already.
2263  */
2264 int
2265 ip_rsvp_init(struct socket *so)
2266 {
2267         if (so->so_type != SOCK_RAW ||
2268             so->so_proto->pr_protocol != IPPROTO_RSVP)
2269                 return EOPNOTSUPP;
2270
2271         if (ip_rsvpd != NULL)
2272                 return EADDRINUSE;
2273
2274         ip_rsvpd = so;
2275         /*
2276          * This may seem silly, but we need to be sure we don't over-increment
2277          * the RSVP counter, in case something slips up.
2278          */
2279         if (!ip_rsvp_on) {
2280                 ip_rsvp_on = 1;
2281                 rsvp_on++;
2282         }
2283
2284         return 0;
2285 }
2286
2287 int
2288 ip_rsvp_done(void)
2289 {
2290         ip_rsvpd = NULL;
2291         /*
2292          * This may seem silly, but we need to be sure we don't over-decrement
2293          * the RSVP counter, in case something slips up.
2294          */
2295         if (ip_rsvp_on) {
2296                 ip_rsvp_on = 0;
2297                 rsvp_on--;
2298         }
2299         return 0;
2300 }
2301
2302 void
2303 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2304 {
2305         int off, proto;
2306         __va_list ap;
2307
2308         __va_start(ap, m);
2309         off = __va_arg(ap, int);
2310         proto = __va_arg(ap, int);
2311         __va_end(ap);
2312
2313         if (rsvp_input_p) { /* call the real one if loaded */
2314                 rsvp_input_p(m, off, proto);
2315                 return;
2316         }
2317
2318         /* Can still get packets with rsvp_on = 0 if there is a local member
2319          * of the group to which the RSVP packet is addressed.  But in this
2320          * case we want to throw the packet away.
2321          */
2322
2323         if (!rsvp_on) {
2324                 m_freem(m);
2325                 return;
2326         }
2327
2328         if (ip_rsvpd != NULL) {
2329                 rip_input(m, off, proto);
2330                 return;
2331         }
2332         /* Drop the packet */
2333         m_freem(m);
2334 }