If there are no pfil hooks, just remove the dummynet tag (for packets that
[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 1.107 2008/09/18 11:19:42 sephe Exp $
69  */
70
71 #define _IP_VHL
72
73 #include "opt_bootp.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipdn.h"
76 #include "opt_ipdivert.h"
77 #include "opt_ipfilter.h"
78 #include "opt_ipstealth.h"
79 #include "opt_ipsec.h"
80
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/mpipe.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/time.h>
90 #include <sys/globaldata.h>
91 #include <sys/thread.h>
92 #include <sys/kernel.h>
93 #include <sys/syslog.h>
94 #include <sys/sysctl.h>
95 #include <sys/in_cksum.h>
96
97 #include <machine/stdarg.h>
98
99 #include <net/if.h>
100 #include <net/if_types.h>
101 #include <net/if_var.h>
102 #include <net/if_dl.h>
103 #include <net/pfil.h>
104 #include <net/route.h>
105 #include <net/netisr.h>
106
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/in_var.h>
110 #include <netinet/ip.h>
111 #include <netinet/in_pcb.h>
112 #include <netinet/ip_var.h>
113 #include <netinet/ip_icmp.h>
114 #include <netinet/ip_divert.h>
115
116 #include <sys/thread2.h>
117 #include <sys/msgport2.h>
118 #include <net/netmsg2.h>
119
120 #include <sys/socketvar.h>
121
122 #include <net/ipfw/ip_fw.h>
123 #include <net/dummynet/ip_dummynet.h>
124
125 #ifdef IPSEC
126 #include <netinet6/ipsec.h>
127 #include <netproto/key/key.h>
128 #endif
129
130 #ifdef FAST_IPSEC
131 #include <netproto/ipsec/ipsec.h>
132 #include <netproto/ipsec/key.h>
133 #endif
134
135 int rsvp_on = 0;
136 static int ip_rsvp_on;
137 struct socket *ip_rsvpd;
138
139 int ipforwarding = 0;
140 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
141     &ipforwarding, 0, "Enable IP forwarding between interfaces");
142
143 static int ipsendredirects = 1; /* XXX */
144 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
145     &ipsendredirects, 0, "Enable sending IP redirects");
146
147 int ip_defttl = IPDEFTTL;
148 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
149     &ip_defttl, 0, "Maximum TTL on IP packets");
150
151 static int ip_dosourceroute = 0;
152 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
153     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
154
155 static int ip_acceptsourceroute = 0;
156 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
157     CTLFLAG_RW, &ip_acceptsourceroute, 0,
158     "Enable accepting source routed IP packets");
159
160 static int ip_keepfaith = 0;
161 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
162     &ip_keepfaith, 0,
163     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
164
165 static int nipq = 0;    /* total # of reass queues */
166 static int maxnipq;
167 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
168     &maxnipq, 0,
169     "Maximum number of IPv4 fragment reassembly queue entries");
170
171 static int maxfragsperpacket;
172 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
173     &maxfragsperpacket, 0,
174     "Maximum number of IPv4 fragments allowed per packet");
175
176 static int ip_sendsourcequench = 0;
177 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
178     &ip_sendsourcequench, 0,
179     "Enable the transmission of source quench packets");
180
181 int ip_do_randomid = 1;
182 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
183     &ip_do_randomid, 0,
184     "Assign random ip_id values");      
185 /*
186  * XXX - Setting ip_checkinterface mostly implements the receive side of
187  * the Strong ES model described in RFC 1122, but since the routing table
188  * and transmit implementation do not implement the Strong ES model,
189  * setting this to 1 results in an odd hybrid.
190  *
191  * XXX - ip_checkinterface currently must be disabled if you use ipnat
192  * to translate the destination address to another local interface.
193  *
194  * XXX - ip_checkinterface must be disabled if you add IP aliases
195  * to the loopback interface instead of the interface where the
196  * packets for those addresses are received.
197  */
198 static int ip_checkinterface = 0;
199 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
200     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
201
202 #ifdef DIAGNOSTIC
203 static int ipprintfs = 0;
204 #endif
205
206 extern  struct domain inetdomain;
207 extern  struct protosw inetsw[];
208 u_char  ip_protox[IPPROTO_MAX];
209 struct  in_ifaddrhead in_ifaddrheads[MAXCPU];   /* first inet address */
210 struct  in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
211                                                 /* inet addr hash table */
212 u_long  in_ifaddrhmask;                         /* mask for hash table */
213
214 struct ip_stats ipstats_percpu[MAXCPU];
215 #ifdef SMP
216 static int
217 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
218 {
219         int cpu, error = 0;
220
221         for (cpu = 0; cpu < ncpus; ++cpu) {
222                 if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
223                                         sizeof(struct ip_stats))))
224                         break;
225                 if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
226                                        sizeof(struct ip_stats))))
227                         break;
228         }
229
230         return (error);
231 }
232 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
233     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
234 #else
235 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
236     &ipstat, ip_stats, "IP statistics");
237 #endif
238
239 /* Packet reassembly stuff */
240 #define IPREASS_NHASH_LOG2      6
241 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
242 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
243 #define IPREASS_HASH(x,y)                                               \
244     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
245
246 static struct ipq ipq[IPREASS_NHASH];
247
248 #ifdef IPCTL_DEFMTU
249 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
250     &ip_mtu, 0, "Default MTU");
251 #endif
252
253 #ifdef IPSTEALTH
254 static int ipstealth = 0;
255 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
256 #else
257 static const int ipstealth = 0;
258 #endif
259
260 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
261
262 struct pfil_head inet_pfil_hook;
263
264 /*
265  * struct ip_srcrt_opt is used to store packet state while it travels
266  * through the stack.
267  *
268  * XXX Note that the code even makes assumptions on the size and
269  * alignment of fields inside struct ip_srcrt so e.g. adding some
270  * fields will break the code.  This needs to be fixed.
271  *
272  * We need to save the IP options in case a protocol wants to respond
273  * to an incoming packet over the same route if the packet got here
274  * using IP source routing.  This allows connection establishment and
275  * maintenance when the remote end is on a network that is not known
276  * to us.
277  */
278 struct ip_srcrt {
279         struct  in_addr dst;                    /* final destination */
280         char    nop;                            /* one NOP to align */
281         char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
282         struct  in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
283 };
284
285 struct ip_srcrt_opt {
286         int             ip_nhops;
287         struct ip_srcrt ip_srcrt;
288 };
289
290 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
291 static struct malloc_pipe ipq_mpipe;
292
293 static void             save_rte(struct mbuf *, u_char *, struct in_addr);
294 static int              ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
295 static void             ip_freef(struct ipq *);
296 static void             ip_input_handler(struct netmsg *);
297
298 /*
299  * IP initialization: fill in IP protocol switch table.
300  * All protocols not implemented in kernel go to raw IP protocol handler.
301  */
302 void
303 ip_init(void)
304 {
305         struct protosw *pr;
306         int i;
307 #ifdef SMP
308         int cpu;
309 #endif
310
311         /*
312          * Make sure we can handle a reasonable number of fragments but
313          * cap it at 4000 (XXX).
314          */
315         mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
316                     IFQ_MAXLEN, 4000, 0, NULL);
317         for (i = 0; i < ncpus; ++i) {
318                 TAILQ_INIT(&in_ifaddrheads[i]);
319                 in_ifaddrhashtbls[i] =
320                         hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
321         }
322         pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
323         if (pr == NULL)
324                 panic("ip_init");
325         for (i = 0; i < IPPROTO_MAX; i++)
326                 ip_protox[i] = pr - inetsw;
327         for (pr = inetdomain.dom_protosw;
328              pr < inetdomain.dom_protoswNPROTOSW; pr++)
329                 if (pr->pr_domain->dom_family == PF_INET &&
330                     pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
331                         ip_protox[pr->pr_protocol] = pr - inetsw;
332
333         inet_pfil_hook.ph_type = PFIL_TYPE_AF;
334         inet_pfil_hook.ph_af = AF_INET;
335         if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
336                 kprintf("%s: WARNING: unable to register pfil hook, "
337                         "error %d\n", __func__, i);
338         }
339
340         for (i = 0; i < IPREASS_NHASH; i++)
341                 ipq[i].next = ipq[i].prev = &ipq[i];
342
343         maxnipq = nmbclusters / 32;
344         maxfragsperpacket = 16;
345
346         ip_id = time_second & 0xffff;
347
348         /*
349          * Initialize IP statistics counters for each CPU.
350          *
351          */
352 #ifdef SMP
353         for (cpu = 0; cpu < ncpus; ++cpu) {
354                 bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
355         }
356 #else
357         bzero(&ipstat, sizeof(struct ip_stats));
358 #endif
359
360         netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
361 }
362
363 /*
364  * XXX watch out this one. It is perhaps used as a cache for
365  * the most recently used route ? it is cleared in in_addroute()
366  * when a new route is successfully created.
367  */
368 struct route ipforward_rt[MAXCPU];
369
370 /* Do transport protocol processing. */
371 static void
372 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
373 {
374         /*
375          * Switch out to protocol's input routine.
376          */
377         (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
378 }
379
380 static void
381 transport_processing_handler(netmsg_t netmsg)
382 {
383         struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
384         struct ip *ip;
385         int hlen;
386
387         ip = mtod(pmsg->nm_packet, struct ip *);
388         hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
389
390         transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
391         /* netmsg was embedded in the mbuf, do not reply! */
392 }
393
394 static void
395 ip_input_handler(struct netmsg *msg0)
396 {
397         struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
398
399         ip_input(m);
400         /* msg0 was embedded in the mbuf, do not reply! */
401 }
402
403 /*
404  * IP input routine.  Checksum and byte swap header.  If fragmented
405  * try to reassemble.  Process options.  Pass to next level.
406  */
407 void
408 ip_input(struct mbuf *m)
409 {
410         struct ip *ip;
411         struct in_ifaddr *ia = NULL;
412         struct in_ifaddr_container *iac;
413         int hlen, checkif;
414         u_short sum;
415         struct in_addr pkt_dst;
416         boolean_t using_srcrt = FALSE;          /* forward (by PFIL_HOOKS) */
417         boolean_t needredispatch = FALSE;
418         struct in_addr odst;                    /* original dst address(NAT) */
419         struct m_tag *mtag;
420         struct sockaddr_in *next_hop = NULL;
421 #ifdef FAST_IPSEC
422         struct tdb_ident *tdbi;
423         struct secpolicy *sp;
424         int error;
425 #endif
426
427         M_ASSERTPKTHDR(m);
428
429         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
430                 /* Next hop */
431                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
432                 KKASSERT(mtag != NULL);
433                 next_hop = m_tag_data(mtag);
434         }
435
436         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
437                 /* dummynet already filtered us */
438                 ip = mtod(m, struct ip *);
439                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
440                 goto iphack;
441         }
442
443         ipstat.ips_total++;
444
445         /* length checks already done in ip_demux() */
446         KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
447
448         ip = mtod(m, struct ip *);
449
450         if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
451                 ipstat.ips_badvers++;
452                 goto bad;
453         }
454
455         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
456         /* length checks already done in ip_demux() */
457         KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
458         KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
459
460         /* 127/8 must not appear on wire - RFC1122 */
461         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
462             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
463                 if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
464                         ipstat.ips_badaddr++;
465                         goto bad;
466                 }
467         }
468
469         if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
470                 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
471         } else {
472                 if (hlen == sizeof(struct ip)) {
473                         sum = in_cksum_hdr(ip);
474                 } else {
475                         sum = in_cksum(m, hlen);
476                 }
477         }
478         if (sum != 0) {
479                 ipstat.ips_badsum++;
480                 goto bad;
481         }
482
483 #ifdef ALTQ
484         if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
485                 /* packet is dropped by traffic conditioner */
486                 return;
487         }
488 #endif
489         /*
490          * Convert fields to host representation.
491          */
492         ip->ip_len = ntohs(ip->ip_len);
493         if (ip->ip_len < hlen) {
494                 ipstat.ips_badlen++;
495                 goto bad;
496         }
497         ip->ip_off = ntohs(ip->ip_off);
498
499         /*
500          * Check that the amount of data in the buffers
501          * is as at least much as the IP header would have us expect.
502          * Trim mbufs if longer than we expect.
503          * Drop packet if shorter than we expect.
504          */
505         if (m->m_pkthdr.len < ip->ip_len) {
506                 ipstat.ips_tooshort++;
507                 goto bad;
508         }
509         if (m->m_pkthdr.len > ip->ip_len) {
510                 if (m->m_len == m->m_pkthdr.len) {
511                         m->m_len = ip->ip_len;
512                         m->m_pkthdr.len = ip->ip_len;
513                 } else
514                         m_adj(m, ip->ip_len - m->m_pkthdr.len);
515         }
516 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
517         /*
518          * Bypass packet filtering for packets from a tunnel (gif).
519          */
520         if (ipsec_gethist(m, NULL))
521                 goto pass;
522 #endif
523
524         /*
525          * IpHack's section.
526          * Right now when no processing on packet has done
527          * and it is still fresh out of network we do our black
528          * deals with it.
529          * - Firewall: deny/allow/divert
530          * - Xlate: translate packet's addr/port (NAT).
531          * - Pipe: pass pkt through dummynet.
532          * - Wrap: fake packet's addr/port <unimpl.>
533          * - Encapsulate: put it in another IP and send out. <unimp.>
534          */
535
536 iphack:
537         /*
538          * If we've been forwarded from the output side, then
539          * skip the firewall a second time
540          */
541         if (next_hop != NULL)
542                 goto ours;
543
544         /* No pfil hooks */
545         if (!pfil_has_hooks(&inet_pfil_hook)) {
546                 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
547                         /*
548                          * Strip dummynet tags from stranded packets
549                          */
550                         mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
551                         KKASSERT(mtag != NULL);
552                         m_tag_delete(m, mtag);
553                         m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
554                 }
555                 goto pass;
556         }
557
558         /*
559          * Run through list of hooks for input packets.
560          *
561          * NB: Beware of the destination address changing (e.g.
562          *     by NAT rewriting). When this happens, tell
563          *     ip_forward to do the right thing.
564          */
565         odst = ip->ip_dst;
566         if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
567                 return;
568         if (m == NULL)  /* consumed by filter */
569                 return;
570         ip = mtod(m, struct ip *);
571         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
572         using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
573
574         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
575                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
576                 KKASSERT(mtag != NULL);
577                 next_hop = m_tag_data(mtag);
578         }
579         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
580                 ip_dn_queue(m);
581                 return;
582         }
583         if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
584                 needredispatch = TRUE;
585                 m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
586         }
587 pass:
588         /*
589          * Process options and, if not destined for us,
590          * ship it on.  ip_dooptions returns 1 when an
591          * error was detected (causing an icmp message
592          * to be sent and the original packet to be freed).
593          */
594         if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
595                 return;
596
597         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
598          * matter if it is destined to another node, or whether it is
599          * a multicast one, RSVP wants it! and prevents it from being forwarded
600          * anywhere else. Also checks if the rsvp daemon is running before
601          * grabbing the packet.
602          */
603         if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
604                 goto ours;
605
606         /*
607          * Check our list of addresses, to see if the packet is for us.
608          * If we don't have any addresses, assume any unicast packet
609          * we receive might be for us (and let the upper layers deal
610          * with it).
611          */
612         if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
613             !(m->m_flags & (M_MCAST | M_BCAST)))
614                 goto ours;
615
616         /*
617          * Cache the destination address of the packet; this may be
618          * changed by use of 'ipfw fwd'.
619          */
620         pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
621
622         /*
623          * Enable a consistency check between the destination address
624          * and the arrival interface for a unicast packet (the RFC 1122
625          * strong ES model) if IP forwarding is disabled and the packet
626          * is not locally generated and the packet is not subject to
627          * 'ipfw fwd'.
628          *
629          * XXX - Checking also should be disabled if the destination
630          * address is ipnat'ed to a different interface.
631          *
632          * XXX - Checking is incompatible with IP aliases added
633          * to the loopback interface instead of the interface where
634          * the packets are received.
635          */
636         checkif = ip_checkinterface &&
637                   !ipforwarding &&
638                   m->m_pkthdr.rcvif != NULL &&
639                   !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
640                   next_hop == NULL;
641
642         /*
643          * Check for exact addresses in the hash bucket.
644          */
645         LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
646                 ia = iac->ia;
647
648                 /*
649                  * If the address matches, verify that the packet
650                  * arrived via the correct interface if checking is
651                  * enabled.
652                  */
653                 if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
654                     (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
655                         goto ours;
656         }
657         ia = NULL;
658
659         /*
660          * Check for broadcast addresses.
661          *
662          * Only accept broadcast packets that arrive via the matching
663          * interface.  Reception of forwarded directed broadcasts would
664          * be handled via ip_forward() and ether_output() with the loopback
665          * into the stack for SIMPLEX interfaces handled by ether_output().
666          */
667         if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
668                 struct ifaddr_container *ifac;
669
670                 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
671                               ifa_link) {
672                         struct ifaddr *ifa = ifac->ifa;
673
674                         if (ifa->ifa_addr == NULL) /* shutdown/startup race */
675                                 continue;
676                         if (ifa->ifa_addr->sa_family != AF_INET)
677                                 continue;
678                         ia = ifatoia(ifa);
679                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
680                                                                 pkt_dst.s_addr)
681                                 goto ours;
682                         if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
683                                 goto ours;
684 #ifdef BOOTP_COMPAT
685                         if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
686                                 goto ours;
687 #endif
688                 }
689         }
690         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
691                 struct in_multi *inm;
692
693                 if (ip_mrouter != NULL) {
694                         /*
695                          * If we are acting as a multicast router, all
696                          * incoming multicast packets are passed to the
697                          * kernel-level multicast forwarding function.
698                          * The packet is returned (relatively) intact; if
699                          * ip_mforward() returns a non-zero value, the packet
700                          * must be discarded, else it may be accepted below.
701                          */
702                         if (ip_mforward != NULL &&
703                             ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
704                                 ipstat.ips_cantforward++;
705                                 m_freem(m);
706                                 return;
707                         }
708
709                         /*
710                          * The process-level routing daemon needs to receive
711                          * all multicast IGMP packets, whether or not this
712                          * host belongs to their destination groups.
713                          */
714                         if (ip->ip_p == IPPROTO_IGMP)
715                                 goto ours;
716                         ipstat.ips_forward++;
717                 }
718                 /*
719                  * See if we belong to the destination multicast group on the
720                  * arrival interface.
721                  */
722                 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
723                 if (inm == NULL) {
724                         ipstat.ips_notmember++;
725                         m_freem(m);
726                         return;
727                 }
728                 goto ours;
729         }
730         if (ip->ip_dst.s_addr == INADDR_BROADCAST)
731                 goto ours;
732         if (ip->ip_dst.s_addr == INADDR_ANY)
733                 goto ours;
734
735         /*
736          * FAITH(Firewall Aided Internet Translator)
737          */
738         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
739                 if (ip_keepfaith) {
740                         if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
741                                 goto ours;
742                 }
743                 m_freem(m);
744                 return;
745         }
746
747         /*
748          * Not for us; forward if possible and desirable.
749          */
750         if (!ipforwarding) {
751                 ipstat.ips_cantforward++;
752                 m_freem(m);
753         } else {
754 #ifdef IPSEC
755                 /*
756                  * Enforce inbound IPsec SPD.
757                  */
758                 if (ipsec4_in_reject(m, NULL)) {
759                         ipsecstat.in_polvio++;
760                         goto bad;
761                 }
762 #endif
763 #ifdef FAST_IPSEC
764                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
765                 crit_enter();
766                 if (mtag != NULL) {
767                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
768                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
769                 } else {
770                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
771                                                    IP_FORWARDING, &error);
772                 }
773                 if (sp == NULL) {       /* NB: can happen if error */
774                         crit_exit();
775                         /*XXX error stat???*/
776                         DPRINTF(("ip_input: no SP for forwarding\n"));  /*XXX*/
777                         goto bad;
778                 }
779
780                 /*
781                  * Check security policy against packet attributes.
782                  */
783                 error = ipsec_in_reject(sp, m);
784                 KEY_FREESP(&sp);
785                 crit_exit();
786                 if (error) {
787                         ipstat.ips_cantforward++;
788                         goto bad;
789                 }
790 #endif
791                 ip_forward(m, using_srcrt, next_hop);
792         }
793         return;
794
795 ours:
796
797         /*
798          * IPSTEALTH: Process non-routing options only
799          * if the packet is destined for us.
800          */
801         if (ipstealth &&
802             hlen > sizeof(struct ip) &&
803             ip_dooptions(m, 1, next_hop))
804                 return;
805
806         /* Count the packet in the ip address stats */
807         if (ia != NULL) {
808                 ia->ia_ifa.if_ipackets++;
809                 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
810         }
811
812         /*
813          * If offset or IP_MF are set, must reassemble.
814          * Otherwise, nothing need be done.
815          * (We could look in the reassembly queue to see
816          * if the packet was previously fragmented,
817          * but it's not worth the time; just let them time out.)
818          */
819         if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
820                 /*
821                  * Attempt reassembly; if it succeeds, proceed.
822                  * ip_reass() will return a different mbuf.
823                  */
824                 m = ip_reass(m);
825                 if (m == NULL)
826                         return;
827                 ip = mtod(m, struct ip *);
828
829                 /* Get the header length of the reassembled packet */
830                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
831
832                 needredispatch = TRUE;
833         } else {
834                 ip->ip_len -= hlen;
835         }
836
837 #ifdef IPSEC
838         /*
839          * enforce IPsec policy checking if we are seeing last header.
840          * note that we do not visit this with protocols with pcb layer
841          * code - like udp/tcp/raw ip.
842          */
843         if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
844             ipsec4_in_reject(m, NULL)) {
845                 ipsecstat.in_polvio++;
846                 goto bad;
847         }
848 #endif
849 #if FAST_IPSEC
850         /*
851          * enforce IPsec policy checking if we are seeing last header.
852          * note that we do not visit this with protocols with pcb layer
853          * code - like udp/tcp/raw ip.
854          */
855         if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
856                 /*
857                  * Check if the packet has already had IPsec processing
858                  * done.  If so, then just pass it along.  This tag gets
859                  * set during AH, ESP, etc. input handling, before the
860                  * packet is returned to the ip input queue for delivery.
861                  */
862                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
863                 crit_enter();
864                 if (mtag != NULL) {
865                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
866                         sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
867                 } else {
868                         sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
869                                                    IP_FORWARDING, &error);
870                 }
871                 if (sp != NULL) {
872                         /*
873                          * Check security policy against packet attributes.
874                          */
875                         error = ipsec_in_reject(sp, m);
876                         KEY_FREESP(&sp);
877                 } else {
878                         /* XXX error stat??? */
879                         error = EINVAL;
880 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
881                         goto bad;
882                 }
883                 crit_exit();
884                 if (error)
885                         goto bad;
886         }
887 #endif /* FAST_IPSEC */
888
889         ipstat.ips_delivered++;
890         if (needredispatch) {
891                 struct netmsg_packet *pmsg;
892                 lwkt_port_t port;
893
894                 ip->ip_off = htons(ip->ip_off);
895                 ip->ip_len = htons(ip->ip_len);
896                 port = ip_mport_in(&m);
897                 if (port == NULL)
898                         return;
899
900                 pmsg = &m->m_hdr.mh_netmsg;
901                 netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
902                             transport_processing_handler);
903                 pmsg->nm_packet = m;
904                 pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
905
906                 ip = mtod(m, struct ip *);
907                 ip->ip_len = ntohs(ip->ip_len);
908                 ip->ip_off = ntohs(ip->ip_off);
909                 lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
910         } else {
911                 transport_processing_oncpu(m, hlen, ip);
912         }
913         return;
914
915 bad:
916         m_freem(m);
917 }
918
919 /*
920  * Take incoming datagram fragment and try to reassemble it into
921  * whole datagram.  If a chain for reassembly of this datagram already
922  * exists, then it is given as fp; otherwise have to make a chain.
923  */
924 struct mbuf *
925 ip_reass(struct mbuf *m)
926 {
927         struct ip *ip = mtod(m, struct ip *);
928         struct mbuf *p = NULL, *q, *nq;
929         struct mbuf *n;
930         struct ipq *fp = NULL;
931         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
932         int i, next;
933         u_short sum;
934
935         /* If maxnipq is 0, never accept fragments. */
936         if (maxnipq == 0) {
937                 ipstat.ips_fragments++;
938                 ipstat.ips_fragdropped++;
939                 m_freem(m);
940                 return NULL;
941         }
942
943         sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
944         /*
945          * Look for queue of fragments of this datagram.
946          */
947         for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
948                 if (ip->ip_id == fp->ipq_id &&
949                     ip->ip_src.s_addr == fp->ipq_src.s_addr &&
950                     ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
951                     ip->ip_p == fp->ipq_p)
952                         goto found;
953
954         fp = NULL;
955
956         /*
957          * Enforce upper bound on number of fragmented packets
958          * for which we attempt reassembly;
959          * If maxnipq is -1, accept all fragments without limitation.
960          */
961         if (nipq > maxnipq && maxnipq > 0) {
962                 /*
963                  * drop something from the tail of the current queue
964                  * before proceeding further
965                  */
966                 if (ipq[sum].prev == &ipq[sum]) {   /* gak */
967                         for (i = 0; i < IPREASS_NHASH; i++) {
968                                 if (ipq[i].prev != &ipq[i]) {
969                                         ipstat.ips_fragtimeout +=
970                                             ipq[i].prev->ipq_nfrags;
971                                         ip_freef(ipq[i].prev);
972                                         break;
973                                 }
974                         }
975                 } else {
976                         ipstat.ips_fragtimeout +=
977                             ipq[sum].prev->ipq_nfrags;
978                         ip_freef(ipq[sum].prev);
979                 }
980         }
981 found:
982         /*
983          * Adjust ip_len to not reflect header,
984          * convert offset of this to bytes.
985          */
986         ip->ip_len -= hlen;
987         if (ip->ip_off & IP_MF) {
988                 /*
989                  * Make sure that fragments have a data length
990                  * that's a non-zero multiple of 8 bytes.
991                  */
992                 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
993                         ipstat.ips_toosmall++; /* XXX */
994                         m_freem(m);
995                         return NULL;
996                 }
997                 m->m_flags |= M_FRAG;
998         } else
999                 m->m_flags &= ~M_FRAG;
1000         ip->ip_off <<= 3;
1001
1002         ipstat.ips_fragments++;
1003         m->m_pkthdr.header = ip;
1004
1005         /*
1006          * If the hardware has not done csum over this fragment
1007          * then csum_data is not valid at all.
1008          */
1009         if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1010             == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1011                 m->m_pkthdr.csum_data = 0;
1012                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1013         }
1014
1015         /*
1016          * Presence of header sizes in mbufs
1017          * would confuse code below.
1018          */
1019         m->m_data += hlen;
1020         m->m_len -= hlen;
1021
1022         /*
1023          * If first fragment to arrive, create a reassembly queue.
1024          */
1025         if (fp == NULL) {
1026                 if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1027                         goto dropfrag;
1028                 insque(fp, &ipq[sum]);
1029                 nipq++;
1030                 fp->ipq_nfrags = 1;
1031                 fp->ipq_ttl = IPFRAGTTL;
1032                 fp->ipq_p = ip->ip_p;
1033                 fp->ipq_id = ip->ip_id;
1034                 fp->ipq_src = ip->ip_src;
1035                 fp->ipq_dst = ip->ip_dst;
1036                 fp->ipq_frags = m;
1037                 m->m_nextpkt = NULL;
1038                 goto inserted;
1039         } else {
1040                 fp->ipq_nfrags++;
1041         }
1042
1043 #define GETIP(m)        ((struct ip*)((m)->m_pkthdr.header))
1044
1045         /*
1046          * Find a segment which begins after this one does.
1047          */
1048         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1049                 if (GETIP(q)->ip_off > ip->ip_off)
1050                         break;
1051
1052         /*
1053          * If there is a preceding segment, it may provide some of
1054          * our data already.  If so, drop the data from the incoming
1055          * segment.  If it provides all of our data, drop us, otherwise
1056          * stick new segment in the proper place.
1057          *
1058          * If some of the data is dropped from the the preceding
1059          * segment, then it's checksum is invalidated.
1060          */
1061         if (p) {
1062                 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1063                 if (i > 0) {
1064                         if (i >= ip->ip_len)
1065                                 goto dropfrag;
1066                         m_adj(m, i);
1067                         m->m_pkthdr.csum_flags = 0;
1068                         ip->ip_off += i;
1069                         ip->ip_len -= i;
1070                 }
1071                 m->m_nextpkt = p->m_nextpkt;
1072                 p->m_nextpkt = m;
1073         } else {
1074                 m->m_nextpkt = fp->ipq_frags;
1075                 fp->ipq_frags = m;
1076         }
1077
1078         /*
1079          * While we overlap succeeding segments trim them or,
1080          * if they are completely covered, dequeue them.
1081          */
1082         for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1083              q = nq) {
1084                 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1085                 if (i < GETIP(q)->ip_len) {
1086                         GETIP(q)->ip_len -= i;
1087                         GETIP(q)->ip_off += i;
1088                         m_adj(q, i);
1089                         q->m_pkthdr.csum_flags = 0;
1090                         break;
1091                 }
1092                 nq = q->m_nextpkt;
1093                 m->m_nextpkt = nq;
1094                 ipstat.ips_fragdropped++;
1095                 fp->ipq_nfrags--;
1096                 q->m_nextpkt = NULL;
1097                 m_freem(q);
1098         }
1099
1100 inserted:
1101         /*
1102          * Check for complete reassembly and perform frag per packet
1103          * limiting.
1104          *
1105          * Frag limiting is performed here so that the nth frag has
1106          * a chance to complete the packet before we drop the packet.
1107          * As a result, n+1 frags are actually allowed per packet, but
1108          * only n will ever be stored. (n = maxfragsperpacket.)
1109          *
1110          */
1111         next = 0;
1112         for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1113                 if (GETIP(q)->ip_off != next) {
1114                         if (fp->ipq_nfrags > maxfragsperpacket) {
1115                                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1116                                 ip_freef(fp);
1117                         }
1118                         return (NULL);
1119                 }
1120                 next += GETIP(q)->ip_len;
1121         }
1122         /* Make sure the last packet didn't have the IP_MF flag */
1123         if (p->m_flags & M_FRAG) {
1124                 if (fp->ipq_nfrags > maxfragsperpacket) {
1125                         ipstat.ips_fragdropped += fp->ipq_nfrags;
1126                         ip_freef(fp);
1127                 }
1128                 return (NULL);
1129         }
1130
1131         /*
1132          * Reassembly is complete.  Make sure the packet is a sane size.
1133          */
1134         q = fp->ipq_frags;
1135         ip = GETIP(q);
1136         if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1137                 ipstat.ips_toolong++;
1138                 ipstat.ips_fragdropped += fp->ipq_nfrags;
1139                 ip_freef(fp);
1140                 return (NULL);
1141         }
1142
1143         /*
1144          * Concatenate fragments.
1145          */
1146         m = q;
1147         n = m->m_next;
1148         m->m_next = NULL;
1149         m_cat(m, n);
1150         nq = q->m_nextpkt;
1151         q->m_nextpkt = NULL;
1152         for (q = nq; q != NULL; q = nq) {
1153                 nq = q->m_nextpkt;
1154                 q->m_nextpkt = NULL;
1155                 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1156                 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1157                 m_cat(m, q);
1158         }
1159
1160         /*
1161          * Clean up the 1's complement checksum.  Carry over 16 bits must
1162          * be added back.  This assumes no more then 65535 packet fragments
1163          * were reassembled.  A second carry can also occur (but not a third).
1164          */
1165         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1166                                 (m->m_pkthdr.csum_data >> 16);
1167         if (m->m_pkthdr.csum_data > 0xFFFF)
1168                 m->m_pkthdr.csum_data -= 0xFFFF;
1169
1170         /*
1171          * Create header for new ip packet by
1172          * modifying header of first packet;
1173          * dequeue and discard fragment reassembly header.
1174          * Make header visible.
1175          */
1176         ip->ip_len = next;
1177         ip->ip_src = fp->ipq_src;
1178         ip->ip_dst = fp->ipq_dst;
1179         remque(fp);
1180         nipq--;
1181         mpipe_free(&ipq_mpipe, fp);
1182         m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1183         m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1184         /* some debugging cruft by sklower, below, will go away soon */
1185         if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1186                 int plen = 0;
1187
1188                 for (n = m; n; n = n->m_next)
1189                         plen += n->m_len;
1190                 m->m_pkthdr.len = plen;
1191         }
1192
1193         ipstat.ips_reassembled++;
1194         return (m);
1195
1196 dropfrag:
1197         ipstat.ips_fragdropped++;
1198         if (fp != NULL)
1199                 fp->ipq_nfrags--;
1200         m_freem(m);
1201         return (NULL);
1202
1203 #undef GETIP
1204 }
1205
1206 /*
1207  * Free a fragment reassembly header and all
1208  * associated datagrams.
1209  */
1210 static void
1211 ip_freef(struct ipq *fp)
1212 {
1213         struct mbuf *q;
1214
1215         while (fp->ipq_frags) {
1216                 q = fp->ipq_frags;
1217                 fp->ipq_frags = q->m_nextpkt;
1218                 q->m_nextpkt = NULL;
1219                 m_freem(q);
1220         }
1221         remque(fp);
1222         mpipe_free(&ipq_mpipe, fp);
1223         nipq--;
1224 }
1225
1226 /*
1227  * IP timer processing;
1228  * if a timer expires on a reassembly
1229  * queue, discard it.
1230  */
1231 void
1232 ip_slowtimo(void)
1233 {
1234         struct ipq *fp;
1235         int i;
1236
1237         crit_enter();
1238         for (i = 0; i < IPREASS_NHASH; i++) {
1239                 fp = ipq[i].next;
1240                 if (fp == NULL)
1241                         continue;
1242                 while (fp != &ipq[i]) {
1243                         --fp->ipq_ttl;
1244                         fp = fp->next;
1245                         if (fp->prev->ipq_ttl == 0) {
1246                                 ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1247                                 ip_freef(fp->prev);
1248                         }
1249                 }
1250         }
1251         /*
1252          * If we are over the maximum number of fragments
1253          * (due to the limit being lowered), drain off
1254          * enough to get down to the new limit.
1255          */
1256         if (maxnipq >= 0 && nipq > maxnipq) {
1257                 for (i = 0; i < IPREASS_NHASH; i++) {
1258                         while (nipq > maxnipq &&
1259                                 (ipq[i].next != &ipq[i])) {
1260                                 ipstat.ips_fragdropped +=
1261                                     ipq[i].next->ipq_nfrags;
1262                                 ip_freef(ipq[i].next);
1263                         }
1264                 }
1265         }
1266         ipflow_slowtimo();
1267         crit_exit();
1268 }
1269
1270 /*
1271  * Drain off all datagram fragments.
1272  */
1273 void
1274 ip_drain(void)
1275 {
1276         int i;
1277
1278         for (i = 0; i < IPREASS_NHASH; i++) {
1279                 while (ipq[i].next != &ipq[i]) {
1280                         ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1281                         ip_freef(ipq[i].next);
1282                 }
1283         }
1284         in_rtqdrain();
1285 }
1286
1287 /*
1288  * Do option processing on a datagram,
1289  * possibly discarding it if bad options are encountered,
1290  * or forwarding it if source-routed.
1291  * The pass argument is used when operating in the IPSTEALTH
1292  * mode to tell what options to process:
1293  * [LS]SRR (pass 0) or the others (pass 1).
1294  * The reason for as many as two passes is that when doing IPSTEALTH,
1295  * non-routing options should be processed only if the packet is for us.
1296  * Returns 1 if packet has been forwarded/freed,
1297  * 0 if the packet should be processed further.
1298  */
1299 static int
1300 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1301 {
1302         struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1303         struct ip *ip = mtod(m, struct ip *);
1304         u_char *cp;
1305         struct in_ifaddr *ia;
1306         int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1307         boolean_t forward = FALSE;
1308         struct in_addr *sin, dst;
1309         n_time ntime;
1310
1311         dst = ip->ip_dst;
1312         cp = (u_char *)(ip + 1);
1313         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1314         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1315                 opt = cp[IPOPT_OPTVAL];
1316                 if (opt == IPOPT_EOL)
1317                         break;
1318                 if (opt == IPOPT_NOP)
1319                         optlen = 1;
1320                 else {
1321                         if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1322                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1323                                 goto bad;
1324                         }
1325                         optlen = cp[IPOPT_OLEN];
1326                         if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1327                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1328                                 goto bad;
1329                         }
1330                 }
1331                 switch (opt) {
1332
1333                 default:
1334                         break;
1335
1336                 /*
1337                  * Source routing with record.
1338                  * Find interface with current destination address.
1339                  * If none on this machine then drop if strictly routed,
1340                  * or do nothing if loosely routed.
1341                  * Record interface address and bring up next address
1342                  * component.  If strictly routed make sure next
1343                  * address is on directly accessible net.
1344                  */
1345                 case IPOPT_LSRR:
1346                 case IPOPT_SSRR:
1347                         if (ipstealth && pass > 0)
1348                                 break;
1349                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1350                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1351                                 goto bad;
1352                         }
1353                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1354                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1355                                 goto bad;
1356                         }
1357                         ipaddr.sin_addr = ip->ip_dst;
1358                         ia = (struct in_ifaddr *)
1359                                 ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1360                         if (ia == NULL) {
1361                                 if (opt == IPOPT_SSRR) {
1362                                         type = ICMP_UNREACH;
1363                                         code = ICMP_UNREACH_SRCFAIL;
1364                                         goto bad;
1365                                 }
1366                                 if (!ip_dosourceroute)
1367                                         goto nosourcerouting;
1368                                 /*
1369                                  * Loose routing, and not at next destination
1370                                  * yet; nothing to do except forward.
1371                                  */
1372                                 break;
1373                         }
1374                         off--;                  /* 0 origin */
1375                         if (off > optlen - (int)sizeof(struct in_addr)) {
1376                                 /*
1377                                  * End of source route.  Should be for us.
1378                                  */
1379                                 if (!ip_acceptsourceroute)
1380                                         goto nosourcerouting;
1381                                 save_rte(m, cp, ip->ip_src);
1382                                 break;
1383                         }
1384                         if (ipstealth)
1385                                 goto dropit;
1386                         if (!ip_dosourceroute) {
1387                                 if (ipforwarding) {
1388                                         char buf[sizeof "aaa.bbb.ccc.ddd"];
1389
1390                                         /*
1391                                          * Acting as a router, so generate ICMP
1392                                          */
1393 nosourcerouting:
1394                                         strcpy(buf, inet_ntoa(ip->ip_dst));
1395                                         log(LOG_WARNING,
1396                                             "attempted source route from %s to %s\n",
1397                                             inet_ntoa(ip->ip_src), buf);
1398                                         type = ICMP_UNREACH;
1399                                         code = ICMP_UNREACH_SRCFAIL;
1400                                         goto bad;
1401                                 } else {
1402                                         /*
1403                                          * Not acting as a router,
1404                                          * so silently drop.
1405                                          */
1406 dropit:
1407                                         ipstat.ips_cantforward++;
1408                                         m_freem(m);
1409                                         return (1);
1410                                 }
1411                         }
1412
1413                         /*
1414                          * locate outgoing interface
1415                          */
1416                         memcpy(&ipaddr.sin_addr, cp + off,
1417                             sizeof ipaddr.sin_addr);
1418
1419                         if (opt == IPOPT_SSRR) {
1420 #define INA     struct in_ifaddr *
1421 #define SA      struct sockaddr *
1422                                 if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1423                                                                         == NULL)
1424                                         ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1425                         } else
1426                                 ia = ip_rtaddr(ipaddr.sin_addr,
1427                                                &ipforward_rt[mycpuid]);
1428                         if (ia == NULL) {
1429                                 type = ICMP_UNREACH;
1430                                 code = ICMP_UNREACH_SRCFAIL;
1431                                 goto bad;
1432                         }
1433                         ip->ip_dst = ipaddr.sin_addr;
1434                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1435                             sizeof(struct in_addr));
1436                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1437                         /*
1438                          * Let ip_intr's mcast routing check handle mcast pkts
1439                          */
1440                         forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1441                         break;
1442
1443                 case IPOPT_RR:
1444                         if (ipstealth && pass == 0)
1445                                 break;
1446                         if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1447                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1448                                 goto bad;
1449                         }
1450                         if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1451                                 code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1452                                 goto bad;
1453                         }
1454                         /*
1455                          * If no space remains, ignore.
1456                          */
1457                         off--;                  /* 0 origin */
1458                         if (off > optlen - (int)sizeof(struct in_addr))
1459                                 break;
1460                         memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1461                             sizeof ipaddr.sin_addr);
1462                         /*
1463                          * locate outgoing interface; if we're the destination,
1464                          * use the incoming interface (should be same).
1465                          */
1466                         if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1467                             (ia = ip_rtaddr(ipaddr.sin_addr,
1468                                             &ipforward_rt[mycpuid]))
1469                                                                      == NULL) {
1470                                 type = ICMP_UNREACH;
1471                                 code = ICMP_UNREACH_HOST;
1472                                 goto bad;
1473                         }
1474                         memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1475                             sizeof(struct in_addr));
1476                         cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1477                         break;
1478
1479                 case IPOPT_TS:
1480                         if (ipstealth && pass == 0)
1481                                 break;
1482                         code = cp - (u_char *)ip;
1483                         if (optlen < 4 || optlen > 40) {
1484                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1485                                 goto bad;
1486                         }
1487                         if ((off = cp[IPOPT_OFFSET]) < 5) {
1488                                 code = &cp[IPOPT_OLEN] - (u_char *)ip;
1489                                 goto bad;
1490                         }
1491                         if (off > optlen - (int)sizeof(int32_t)) {
1492                                 cp[IPOPT_OFFSET + 1] += (1 << 4);
1493                                 if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1494                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1495                                         goto bad;
1496                                 }
1497                                 break;
1498                         }
1499                         off--;                          /* 0 origin */
1500                         sin = (struct in_addr *)(cp + off);
1501                         switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1502
1503                         case IPOPT_TS_TSONLY:
1504                                 break;
1505
1506                         case IPOPT_TS_TSANDADDR:
1507                                 if (off + sizeof(n_time) +
1508                                     sizeof(struct in_addr) > optlen) {
1509                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1510                                         goto bad;
1511                                 }
1512                                 ipaddr.sin_addr = dst;
1513                                 ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1514                                                             m->m_pkthdr.rcvif);
1515                                 if (ia == NULL)
1516                                         continue;
1517                                 memcpy(sin, &IA_SIN(ia)->sin_addr,
1518                                     sizeof(struct in_addr));
1519                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1520                                 off += sizeof(struct in_addr);
1521                                 break;
1522
1523                         case IPOPT_TS_PRESPEC:
1524                                 if (off + sizeof(n_time) +
1525                                     sizeof(struct in_addr) > optlen) {
1526                                         code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1527                                         goto bad;
1528                                 }
1529                                 memcpy(&ipaddr.sin_addr, sin,
1530                                     sizeof(struct in_addr));
1531                                 if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1532                                         continue;
1533                                 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1534                                 off += sizeof(struct in_addr);
1535                                 break;
1536
1537                         default:
1538                                 code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1539                                 goto bad;
1540                         }
1541                         ntime = iptime();
1542                         memcpy(cp + off, &ntime, sizeof(n_time));
1543                         cp[IPOPT_OFFSET] += sizeof(n_time);
1544                 }
1545         }
1546         if (forward && ipforwarding) {
1547                 ip_forward(m, TRUE, next_hop);
1548                 return (1);
1549         }
1550         return (0);
1551 bad:
1552         icmp_error(m, type, code, 0, 0);
1553         ipstat.ips_badoptions++;
1554         return (1);
1555 }
1556
1557 /*
1558  * Given address of next destination (final or next hop),
1559  * return internet address info of interface to be used to get there.
1560  */
1561 struct in_ifaddr *
1562 ip_rtaddr(struct in_addr dst, struct route *ro)
1563 {
1564         struct sockaddr_in *sin;
1565
1566         sin = (struct sockaddr_in *)&ro->ro_dst;
1567
1568         if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1569                 if (ro->ro_rt != NULL) {
1570                         RTFREE(ro->ro_rt);
1571                         ro->ro_rt = NULL;
1572                 }
1573                 sin->sin_family = AF_INET;
1574                 sin->sin_len = sizeof *sin;
1575                 sin->sin_addr = dst;
1576                 rtalloc_ign(ro, RTF_PRCLONING);
1577         }
1578
1579         if (ro->ro_rt == NULL)
1580                 return (NULL);
1581
1582         return (ifatoia(ro->ro_rt->rt_ifa));
1583 }
1584
1585 /*
1586  * Save incoming source route for use in replies,
1587  * to be picked up later by ip_srcroute if the receiver is interested.
1588  */
1589 static void
1590 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1591 {
1592         struct m_tag *mtag;
1593         struct ip_srcrt_opt *opt;
1594         unsigned olen;
1595
1596         mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1597         if (mtag == NULL)
1598                 return;
1599         opt = m_tag_data(mtag);
1600
1601         olen = option[IPOPT_OLEN];
1602 #ifdef DIAGNOSTIC
1603         if (ipprintfs)
1604                 kprintf("save_rte: olen %d\n", olen);
1605 #endif
1606         if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1607                 m_tag_free(mtag);
1608                 return;
1609         }
1610         bcopy(option, opt->ip_srcrt.srcopt, olen);
1611         opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1612         opt->ip_srcrt.dst = dst;
1613         m_tag_prepend(m, mtag);
1614 }
1615
1616 /*
1617  * Retrieve incoming source route for use in replies,
1618  * in the same form used by setsockopt.
1619  * The first hop is placed before the options, will be removed later.
1620  */
1621 struct mbuf *
1622 ip_srcroute(struct mbuf *m0)
1623 {
1624         struct in_addr *p, *q;
1625         struct mbuf *m;
1626         struct m_tag *mtag;
1627         struct ip_srcrt_opt *opt;
1628
1629         if (m0 == NULL)
1630                 return NULL;
1631
1632         mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1633         if (mtag == NULL)
1634                 return NULL;
1635         opt = m_tag_data(mtag);
1636
1637         if (opt->ip_nhops == 0)
1638                 return (NULL);
1639         m = m_get(MB_DONTWAIT, MT_HEADER);
1640         if (m == NULL)
1641                 return (NULL);
1642
1643 #define OPTSIZ  (sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1644
1645         /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1646         m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1647                    sizeof(struct in_addr) + OPTSIZ;
1648 #ifdef DIAGNOSTIC
1649         if (ipprintfs) {
1650                 kprintf("ip_srcroute: nhops %d mlen %d",
1651                         opt->ip_nhops, m->m_len);
1652         }
1653 #endif
1654
1655         /*
1656          * First save first hop for return route
1657          */
1658         p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1659         *(mtod(m, struct in_addr *)) = *p--;
1660 #ifdef DIAGNOSTIC
1661         if (ipprintfs)
1662                 kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1663 #endif
1664
1665         /*
1666          * Copy option fields and padding (nop) to mbuf.
1667          */
1668         opt->ip_srcrt.nop = IPOPT_NOP;
1669         opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1670         memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1671             OPTSIZ);
1672         q = (struct in_addr *)(mtod(m, caddr_t) +
1673             sizeof(struct in_addr) + OPTSIZ);
1674 #undef OPTSIZ
1675         /*
1676          * Record return path as an IP source route,
1677          * reversing the path (pointers are now aligned).
1678          */
1679         while (p >= opt->ip_srcrt.route) {
1680 #ifdef DIAGNOSTIC
1681                 if (ipprintfs)
1682                         kprintf(" %x", ntohl(q->s_addr));
1683 #endif
1684                 *q++ = *p--;
1685         }
1686         /*
1687          * Last hop goes to final destination.
1688          */
1689         *q = opt->ip_srcrt.dst;
1690         m_tag_delete(m0, mtag);
1691 #ifdef DIAGNOSTIC
1692         if (ipprintfs)
1693                 kprintf(" %x\n", ntohl(q->s_addr));
1694 #endif
1695         return (m);
1696 }
1697
1698 /*
1699  * Strip out IP options.
1700  */
1701 void
1702 ip_stripoptions(struct mbuf *m)
1703 {
1704         int datalen;
1705         struct ip *ip = mtod(m, struct ip *);
1706         caddr_t opts;
1707         int optlen;
1708
1709         optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1710         opts = (caddr_t)(ip + 1);
1711         datalen = m->m_len - (sizeof(struct ip) + optlen);
1712         bcopy(opts + optlen, opts, datalen);
1713         m->m_len -= optlen;
1714         if (m->m_flags & M_PKTHDR)
1715                 m->m_pkthdr.len -= optlen;
1716         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1717 }
1718
1719 u_char inetctlerrmap[PRC_NCMDS] = {
1720         0,              0,              0,              0,
1721         0,              EMSGSIZE,       EHOSTDOWN,      EHOSTUNREACH,
1722         EHOSTUNREACH,   EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
1723         EMSGSIZE,       EHOSTUNREACH,   0,              0,
1724         0,              0,              0,              0,
1725         ENOPROTOOPT,    ECONNREFUSED
1726 };
1727
1728 /*
1729  * Forward a packet.  If some error occurs return the sender
1730  * an icmp packet.  Note we can't always generate a meaningful
1731  * icmp message because icmp doesn't have a large enough repertoire
1732  * of codes and types.
1733  *
1734  * If not forwarding, just drop the packet.  This could be confusing
1735  * if ipforwarding was zero but some routing protocol was advancing
1736  * us as a gateway to somewhere.  However, we must let the routing
1737  * protocol deal with that.
1738  *
1739  * The using_srcrt parameter indicates whether the packet is being forwarded
1740  * via a source route.
1741  */
1742 void
1743 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1744 {
1745         struct ip *ip = mtod(m, struct ip *);
1746         struct sockaddr_in *ipforward_rtaddr;
1747         struct rtentry *rt;
1748         int error, type = 0, code = 0, destmtu = 0;
1749         struct mbuf *mcopy;
1750         n_long dest;
1751         struct in_addr pkt_dst;
1752         struct route *cache_rt = &ipforward_rt[mycpuid];
1753
1754         dest = INADDR_ANY;
1755         /*
1756          * Cache the destination address of the packet; this may be
1757          * changed by use of 'ipfw fwd'.
1758          */
1759         pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1760
1761 #ifdef DIAGNOSTIC
1762         if (ipprintfs)
1763                 kprintf("forward: src %x dst %x ttl %x\n",
1764                        ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1765 #endif
1766
1767         if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1768                 ipstat.ips_cantforward++;
1769                 m_freem(m);
1770                 return;
1771         }
1772         if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1773                 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1774                 return;
1775         }
1776
1777         ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1778         if (cache_rt->ro_rt == NULL ||
1779             ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1780                 if (cache_rt->ro_rt != NULL) {
1781                         RTFREE(cache_rt->ro_rt);
1782                         cache_rt->ro_rt = NULL;
1783                 }
1784                 ipforward_rtaddr->sin_family = AF_INET;
1785                 ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1786                 ipforward_rtaddr->sin_addr = pkt_dst;
1787                 rtalloc_ign(cache_rt, RTF_PRCLONING);
1788                 if (cache_rt->ro_rt == NULL) {
1789                         icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1790                         return;
1791                 }
1792         }
1793         rt = cache_rt->ro_rt;
1794
1795         /*
1796          * Save the IP header and at most 8 bytes of the payload,
1797          * in case we need to generate an ICMP message to the src.
1798          *
1799          * XXX this can be optimized a lot by saving the data in a local
1800          * buffer on the stack (72 bytes at most), and only allocating the
1801          * mbuf if really necessary. The vast majority of the packets
1802          * are forwarded without having to send an ICMP back (either
1803          * because unnecessary, or because rate limited), so we are
1804          * really we are wasting a lot of work here.
1805          *
1806          * We don't use m_copy() because it might return a reference
1807          * to a shared cluster. Both this function and ip_output()
1808          * assume exclusive access to the IP header in `m', so any
1809          * data in a cluster may change before we reach icmp_error().
1810          */
1811         MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1812         if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1813                 /*
1814                  * It's probably ok if the pkthdr dup fails (because
1815                  * the deep copy of the tag chain failed), but for now
1816                  * be conservative and just discard the copy since
1817                  * code below may some day want the tags.
1818                  */
1819                 m_free(mcopy);
1820                 mcopy = NULL;
1821         }
1822         if (mcopy != NULL) {
1823                 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1824                     (int)ip->ip_len);
1825                 mcopy->m_pkthdr.len = mcopy->m_len;
1826                 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1827         }
1828
1829         if (!ipstealth)
1830                 ip->ip_ttl -= IPTTLDEC;
1831
1832         /*
1833          * If forwarding packet using same interface that it came in on,
1834          * perhaps should send a redirect to sender to shortcut a hop.
1835          * Only send redirect if source is sending directly to us,
1836          * and if packet was not source routed (or has any options).
1837          * Also, don't send redirect if forwarding using a default route
1838          * or a route modified by a redirect.
1839          */
1840         if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1841             !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1842             satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1843             ipsendredirects && !using_srcrt && next_hop == NULL) {
1844                 u_long src = ntohl(ip->ip_src.s_addr);
1845                 struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1846
1847                 if (rt_ifa != NULL &&
1848                     (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1849                         if (rt->rt_flags & RTF_GATEWAY)
1850                                 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1851                         else
1852                                 dest = pkt_dst.s_addr;
1853                         /*
1854                          * Router requirements says to only send
1855                          * host redirects.
1856                          */
1857                         type = ICMP_REDIRECT;
1858                         code = ICMP_REDIRECT_HOST;
1859 #ifdef DIAGNOSTIC
1860                         if (ipprintfs)
1861                                 kprintf("redirect (%d) to %x\n", code, dest);
1862 #endif
1863                 }
1864         }
1865
1866         error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
1867         if (error == 0) {
1868                 ipstat.ips_forward++;
1869                 if (type == 0) {
1870                         if (mcopy) {
1871                                 ipflow_create(cache_rt, mcopy);
1872                                 m_freem(mcopy);
1873                         }
1874                         return;         /* most common case */
1875                 } else {
1876                         ipstat.ips_redirectsent++;
1877                 }
1878         } else {
1879                 ipstat.ips_cantforward++;
1880         }
1881
1882         if (mcopy == NULL)
1883                 return;
1884
1885         /*
1886          * Send ICMP message.
1887          */
1888
1889         switch (error) {
1890
1891         case 0:                         /* forwarded, but need redirect */
1892                 /* type, code set above */
1893                 break;
1894
1895         case ENETUNREACH:               /* shouldn't happen, checked above */
1896         case EHOSTUNREACH:
1897         case ENETDOWN:
1898         case EHOSTDOWN:
1899         default:
1900                 type = ICMP_UNREACH;
1901                 code = ICMP_UNREACH_HOST;
1902                 break;
1903
1904         case EMSGSIZE:
1905                 type = ICMP_UNREACH;
1906                 code = ICMP_UNREACH_NEEDFRAG;
1907 #ifdef IPSEC
1908                 /*
1909                  * If the packet is routed over IPsec tunnel, tell the
1910                  * originator the tunnel MTU.
1911                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1912                  * XXX quickhack!!!
1913                  */
1914                 if (cache_rt->ro_rt != NULL) {
1915                         struct secpolicy *sp = NULL;
1916                         int ipsecerror;
1917                         int ipsechdr;
1918                         struct route *ro;
1919
1920                         sp = ipsec4_getpolicybyaddr(mcopy,
1921                                                     IPSEC_DIR_OUTBOUND,
1922                                                     IP_FORWARDING,
1923                                                     &ipsecerror);
1924
1925                         if (sp == NULL)
1926                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1927                         else {
1928                                 /* count IPsec header size */
1929                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1930                                                          IPSEC_DIR_OUTBOUND,
1931                                                          NULL);
1932
1933                                 /*
1934                                  * find the correct route for outer IPv4
1935                                  * header, compute tunnel MTU.
1936                                  *
1937                                  */
1938                                 if (sp->req != NULL && sp->req->sav != NULL &&
1939                                     sp->req->sav->sah != NULL) {
1940                                         ro = &sp->req->sav->sah->sa_route;
1941                                         if (ro->ro_rt != NULL &&
1942                                             ro->ro_rt->rt_ifp != NULL) {
1943                                                 destmtu =
1944                                                     ro->ro_rt->rt_ifp->if_mtu;
1945                                                 destmtu -= ipsechdr;
1946                                         }
1947                                 }
1948
1949                                 key_freesp(sp);
1950                         }
1951                 }
1952 #elif FAST_IPSEC
1953                 /*
1954                  * If the packet is routed over IPsec tunnel, tell the
1955                  * originator the tunnel MTU.
1956                  *      tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1957                  * XXX quickhack!!!
1958                  */
1959                 if (cache_rt->ro_rt != NULL) {
1960                         struct secpolicy *sp = NULL;
1961                         int ipsecerror;
1962                         int ipsechdr;
1963                         struct route *ro;
1964
1965                         sp = ipsec_getpolicybyaddr(mcopy,
1966                                                    IPSEC_DIR_OUTBOUND,
1967                                                    IP_FORWARDING,
1968                                                    &ipsecerror);
1969
1970                         if (sp == NULL)
1971                                 destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
1972                         else {
1973                                 /* count IPsec header size */
1974                                 ipsechdr = ipsec4_hdrsiz(mcopy,
1975                                                          IPSEC_DIR_OUTBOUND,
1976                                                          NULL);
1977
1978                                 /*
1979                                  * find the correct route for outer IPv4
1980                                  * header, compute tunnel MTU.
1981                                  */
1982
1983                                 if (sp->req != NULL &&
1984                                     sp->req->sav != NULL &&
1985                                     sp->req->sav->sah != NULL) {
1986                                         ro = &sp->req->sav->sah->sa_route;
1987                                         if (ro->ro_rt != NULL &&
1988                                             ro->ro_rt->rt_ifp != NULL) {
1989                                                 destmtu =
1990                                                     ro->ro_rt->rt_ifp->if_mtu;
1991                                                 destmtu -= ipsechdr;
1992                                         }
1993                                 }
1994
1995                                 KEY_FREESP(&sp);
1996                         }
1997                 }
1998 #else /* !IPSEC && !FAST_IPSEC */
1999                 if (cache_rt->ro_rt != NULL)
2000                         destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2001 #endif /*IPSEC*/
2002                 ipstat.ips_cantfrag++;
2003                 break;
2004
2005         case ENOBUFS:
2006                 /*
2007                  * A router should not generate ICMP_SOURCEQUENCH as
2008                  * required in RFC1812 Requirements for IP Version 4 Routers.
2009                  * Source quench could be a big problem under DoS attacks,
2010                  * or if the underlying interface is rate-limited.
2011                  * Those who need source quench packets may re-enable them
2012                  * via the net.inet.ip.sendsourcequench sysctl.
2013                  */
2014                 if (!ip_sendsourcequench) {
2015                         m_freem(mcopy);
2016                         return;
2017                 } else {
2018                         type = ICMP_SOURCEQUENCH;
2019                         code = 0;
2020                 }
2021                 break;
2022
2023         case EACCES:                    /* ipfw denied packet */
2024                 m_freem(mcopy);
2025                 return;
2026         }
2027         icmp_error(mcopy, type, code, dest, destmtu);
2028 }
2029
2030 void
2031 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2032                struct mbuf *m)
2033 {
2034         if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2035                 struct timeval tv;
2036
2037                 microtime(&tv);
2038                 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2039                     SCM_TIMESTAMP, SOL_SOCKET);
2040                 if (*mp)
2041                         mp = &(*mp)->m_next;
2042         }
2043         if (inp->inp_flags & INP_RECVDSTADDR) {
2044                 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2045                     sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2046                 if (*mp)
2047                         mp = &(*mp)->m_next;
2048         }
2049         if (inp->inp_flags & INP_RECVTTL) {
2050                 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2051                     sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2052                 if (*mp)
2053                         mp = &(*mp)->m_next;
2054         }
2055 #ifdef notyet
2056         /* XXX
2057          * Moving these out of udp_input() made them even more broken
2058          * than they already were.
2059          */
2060         /* options were tossed already */
2061         if (inp->inp_flags & INP_RECVOPTS) {
2062                 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2063                     sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2064                 if (*mp)
2065                         mp = &(*mp)->m_next;
2066         }
2067         /* ip_srcroute doesn't do what we want here, need to fix */
2068         if (inp->inp_flags & INP_RECVRETOPTS) {
2069                 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2070                     sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2071                 if (*mp)
2072                         mp = &(*mp)->m_next;
2073         }
2074 #endif
2075         if (inp->inp_flags & INP_RECVIF) {
2076                 struct ifnet *ifp;
2077                 struct sdlbuf {
2078                         struct sockaddr_dl sdl;
2079                         u_char  pad[32];
2080                 } sdlbuf;
2081                 struct sockaddr_dl *sdp;
2082                 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2083
2084                 if (((ifp = m->m_pkthdr.rcvif)) &&
2085                     ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2086                         sdp = IF_LLSOCKADDR(ifp);
2087                         /*
2088                          * Change our mind and don't try copy.
2089                          */
2090                         if ((sdp->sdl_family != AF_LINK) ||
2091                             (sdp->sdl_len > sizeof(sdlbuf))) {
2092                                 goto makedummy;
2093                         }
2094                         bcopy(sdp, sdl2, sdp->sdl_len);
2095                 } else {
2096 makedummy:
2097                         sdl2->sdl_len =
2098                             offsetof(struct sockaddr_dl, sdl_data[0]);
2099                         sdl2->sdl_family = AF_LINK;
2100                         sdl2->sdl_index = 0;
2101                         sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2102                 }
2103                 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2104                         IP_RECVIF, IPPROTO_IP);
2105                 if (*mp)
2106                         mp = &(*mp)->m_next;
2107         }
2108 }
2109
2110 /*
2111  * XXX these routines are called from the upper part of the kernel.
2112  *
2113  * They could also be moved to ip_mroute.c, since all the RSVP
2114  *  handling is done there already.
2115  */
2116 int
2117 ip_rsvp_init(struct socket *so)
2118 {
2119         if (so->so_type != SOCK_RAW ||
2120             so->so_proto->pr_protocol != IPPROTO_RSVP)
2121                 return EOPNOTSUPP;
2122
2123         if (ip_rsvpd != NULL)
2124                 return EADDRINUSE;
2125
2126         ip_rsvpd = so;
2127         /*
2128          * This may seem silly, but we need to be sure we don't over-increment
2129          * the RSVP counter, in case something slips up.
2130          */
2131         if (!ip_rsvp_on) {
2132                 ip_rsvp_on = 1;
2133                 rsvp_on++;
2134         }
2135
2136         return 0;
2137 }
2138
2139 int
2140 ip_rsvp_done(void)
2141 {
2142         ip_rsvpd = NULL;
2143         /*
2144          * This may seem silly, but we need to be sure we don't over-decrement
2145          * the RSVP counter, in case something slips up.
2146          */
2147         if (ip_rsvp_on) {
2148                 ip_rsvp_on = 0;
2149                 rsvp_on--;
2150         }
2151         return 0;
2152 }
2153
2154 void
2155 rsvp_input(struct mbuf *m, ...) /* XXX must fixup manually */
2156 {
2157         int off, proto;
2158         __va_list ap;
2159
2160         __va_start(ap, m);
2161         off = __va_arg(ap, int);
2162         proto = __va_arg(ap, int);
2163         __va_end(ap);
2164
2165         if (rsvp_input_p) { /* call the real one if loaded */
2166                 rsvp_input_p(m, off, proto);
2167                 return;
2168         }
2169
2170         /* Can still get packets with rsvp_on = 0 if there is a local member
2171          * of the group to which the RSVP packet is addressed.  But in this
2172          * case we want to throw the packet away.
2173          */
2174
2175         if (!rsvp_on) {
2176                 m_freem(m);
2177                 return;
2178         }
2179
2180         if (ip_rsvpd != NULL) {
2181                 rip_input(m, off, proto);
2182                 return;
2183         }
2184         /* Drop the packet */
2185         m_freem(m);
2186 }