xref: /dragonfly/sys/netinet/ip_divert.c (revision a68e0df0)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
34  * $DragonFly: src/sys/netinet/ip_divert.c,v 1.40 2008/10/21 13:51:01 sephe Exp $
35  */
36 
37 #define	_IP_VHL
38 
39 #include "opt_inet.h"
40 #include "opt_ipfw.h"
41 #include "opt_ipdivert.h"
42 #include "opt_ipsec.h"
43 
44 #ifndef INET
45 #error "IPDIVERT requires INET."
46 #endif
47 
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/socket.h>
53 #include <sys/protosw.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
56 #include <sys/systm.h>
57 #include <sys/proc.h>
58 #include <sys/priv.h>
59 #include <sys/in_cksum.h>
60 #include <sys/lock.h>
61 #ifdef SMP
62 #include <sys/msgport.h>
63 #endif
64 
65 #include <net/if.h>
66 #include <net/route.h>
67 
68 #ifdef SMP
69 #include <net/netmsg2.h>
70 #endif
71 #include <sys/thread2.h>
72 #include <sys/mplock2.h>
73 
74 #include <netinet/in.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/ip.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip_var.h>
80 #include <netinet/ip_divert.h>
81 
82 /*
83  * Divert sockets
84  */
85 
86 /*
87  * Allocate enough space to hold a full IP packet
88  */
89 #define	DIVSNDQ		(65536 + 100)
90 #define	DIVRCVQ		(65536 + 100)
91 
92 #define DIV_IS_OUTPUT(sin)	((sin) == NULL || (sin)->sin_addr.s_addr == 0)
93 
94 #define DIV_OUTPUT	0x10000
95 #define DIV_INPUT	0x20000
96 
97 /*
98  * Divert sockets work in conjunction with ipfw, see the divert(4)
99  * manpage for features.
100  * Internally, packets selected by ipfw in ip_input() or ip_output(),
101  * and never diverted before, are passed to the input queue of the
102  * divert socket with a given 'divert_port' number (as specified in
103  * the matching ipfw rule), and they are tagged with a 16 bit cookie
104  * (representing the rule number of the matching ipfw rule), which
105  * is passed to process reading from the socket.
106  *
107  * Packets written to the divert socket are again tagged with a cookie
108  * (usually the same as above) and a destination address.
109  * If the destination address is INADDR_ANY then the packet is
110  * treated as outgoing and sent to ip_output(), otherwise it is
111  * treated as incoming and sent to ip_input().
112  * In both cases, the packet is tagged with the cookie.
113  *
114  * On reinjection, processing in ip_input() and ip_output()
115  * will be exactly the same as for the original packet, except that
116  * ipfw processing will start at the rule number after the one
117  * written in the cookie (so, tagging a packet with a cookie of 0
118  * will cause it to be effectively considered as a standard packet).
119  */
120 
121 /* Internal variables */
122 static struct inpcbinfo divcbinfo;
123 
124 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
125 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
126 
127 static struct mbuf *ip_divert(struct mbuf *, int, int);
128 
129 /*
130  * Initialize divert connection block queue.
131  */
132 void
133 div_init(void)
134 {
135 	in_pcbinfo_init(&divcbinfo);
136 	/*
137 	 * XXX We don't use the hash list for divert IP, but it's easier
138 	 * to allocate a one entry hash list than it is to check all
139 	 * over the place for hashbase == NULL.
140 	 */
141 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
142 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
143 	divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
144 					      &divcbinfo.wildcardhashmask);
145 	divcbinfo.ipi_size = sizeof(struct inpcb);
146 	ip_divert_p = ip_divert;
147 }
148 
149 /*
150  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
151  * with that protocol number to enter the system from the outside.
152  */
153 void
154 div_input(struct mbuf *m, ...)
155 {
156 	ipstat.ips_noproto++;
157 	m_freem(m);
158 }
159 
160 struct lwkt_port *
161 div_soport(struct socket *so, struct sockaddr *nam, struct mbuf **mptr)
162 {
163 	struct sockaddr_in *sin;
164 	struct mbuf *m;
165 	int dir;
166 
167 	sin = (struct sockaddr_in *)nam;
168 	m = *mptr;
169 	M_ASSERTPKTHDR(m);
170 
171 	m->m_pkthdr.rcvif = NULL;
172 	dir = DIV_IS_OUTPUT(sin) ? IP_MPORT_OUT : IP_MPORT_IN;
173 
174 	if (sin != NULL) {
175 		int i;
176 
177 		/*
178 		 * Try locating the interface, if we originally had one.
179 		 * This is done even for outgoing packets, since for a
180 		 * forwarded packet, there must be an interface attached.
181 		 *
182 		 * Find receive interface with the given name, stuffed
183 		 * (if it exists) in the sin_zero[] field.
184 		 * The name is user supplied data so don't trust its size
185 		 * or that it is zero terminated.
186 		 */
187 		for (i = 0; sin->sin_zero[i] && i < sizeof(sin->sin_zero); i++)
188 			;
189 		if (i > 0 && i < sizeof(sin->sin_zero))
190 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
191 	}
192 
193 	if (dir == IP_MPORT_IN && m->m_pkthdr.rcvif == NULL) {
194 		/*
195 		 * No luck with the name, check by IP address.
196 		 * Clear the port and the ifname to make sure
197 		 * there are no distractions for ifa_ifwithaddr.
198 		 *
199 		 * Be careful not to trash sin->sin_port; it will
200 		 * be used later in div_output().
201 		 */
202 		struct ifaddr *ifa;
203 		u_short sin_port;
204 
205 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
206 		sin_port = sin->sin_port; /* save */
207 		sin->sin_port = 0;
208 		ifa = ifa_ifwithaddr((struct sockaddr *)sin);
209 		if (ifa == NULL) {
210 			m_freem(m);
211 			*mptr = NULL;
212 			return NULL;
213 		}
214 		sin->sin_port = sin_port; /* restore */
215 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
216 	}
217 
218 	return ip_mport(mptr, dir);
219 }
220 
221 /*
222  * Divert a packet by passing it up to the divert socket at port 'port'.
223  *
224  * Setup generic address and protocol structures for div_input routine,
225  * then pass them along with mbuf chain.
226  */
227 static void
228 div_packet(struct mbuf *m, int incoming, int port)
229 {
230 	struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
231 	struct inpcb *inp;
232 	struct socket *sa;
233 	struct m_tag *mtag;
234 	struct divert_info *divinfo;
235 	u_int16_t nport;
236 
237 	/* Locate the divert info */
238 	mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
239 	divinfo = m_tag_data(mtag);
240 	divsrc.sin_port = divinfo->skipto;
241 
242 	/*
243 	 * Record receive interface address, if any.
244 	 * But only for incoming packets.
245 	 */
246 	divsrc.sin_addr.s_addr = 0;
247 	if (incoming) {
248 		struct ifaddr_container *ifac;
249 
250 		/* Find IP address for receive interface */
251 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
252 			      ifa_link) {
253 			struct ifaddr *ifa = ifac->ifa;
254 
255 			if (ifa->ifa_addr == NULL)
256 				continue;
257 			if (ifa->ifa_addr->sa_family != AF_INET)
258 				continue;
259 			divsrc.sin_addr =
260 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
261 			break;
262 		}
263 	}
264 	/*
265 	 * Record the incoming interface name whenever we have one.
266 	 */
267 	if (m->m_pkthdr.rcvif) {
268 		/*
269 		 * Hide the actual interface name in there in the
270 		 * sin_zero array. XXX This needs to be moved to a
271 		 * different sockaddr type for divert, e.g.
272 		 * sockaddr_div with multiple fields like
273 		 * sockaddr_dl. Presently we have only 7 bytes
274 		 * but that will do for now as most interfaces
275 		 * are 4 or less + 2 or less bytes for unit.
276 		 * There is probably a faster way of doing this,
277 		 * possibly taking it from the sockaddr_dl on the iface.
278 		 * This solves the problem of a P2P link and a LAN interface
279 		 * having the same address, which can result in the wrong
280 		 * interface being assigned to the packet when fed back
281 		 * into the divert socket. Theoretically if the daemon saves
282 		 * and re-uses the sockaddr_in as suggested in the man pages,
283 		 * this iface name will come along for the ride.
284 		 * (see div_output for the other half of this.)
285 		 */
286 		ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero,
287 			  m->m_pkthdr.rcvif->if_xname);
288 	}
289 
290 	/* Put packet on socket queue, if any */
291 	sa = NULL;
292 	nport = htons((u_int16_t)port);
293 
294 	/*
295 	 * XXX
296 	 * Following loop to locate the inpcb is MPSAFE since the inpcb
297 	 * insertion/removal happens on the same CPU (CPU0), however,
298 	 * saving/testing the socket pointer is not MPSAFE.  So we still
299 	 * need to hold BGL here.
300 	 */
301 	get_mplock();
302 	LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
303 		if (inp->inp_flags & INP_PLACEMARKER)
304 			continue;
305 		if (inp->inp_lport == nport)
306 			sa = inp->inp_socket;
307 	}
308 	if (sa) {
309 		if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m,
310 				 NULL) == 0)
311 			m_freem(m);
312 		else
313 			sorwakeup(sa);
314 		rel_mplock();
315 	} else {
316 		rel_mplock();
317 		m_freem(m);
318 		ipstat.ips_noproto++;
319 		ipstat.ips_delivered--;
320 	}
321 }
322 
323 #ifdef SMP
324 static void
325 div_packet_handler(struct netmsg *nmsg)
326 {
327 	struct netmsg_packet *nmp;
328 	struct lwkt_msg *msg;
329 	struct mbuf *m;
330 	int port, incoming = 0;
331 
332 	nmp = (struct netmsg_packet *)nmsg;
333 	m = nmp->nm_packet;
334 
335 	msg = &nmsg->nm_lmsg;
336 	port = msg->u.ms_result32 & 0xffff;
337 	if (msg->u.ms_result32 & DIV_INPUT)
338 		incoming = 1;
339 
340 	div_packet(m, incoming, port);
341 }
342 #endif	/* SMP */
343 
344 static void
345 divert_packet(struct mbuf *m, int incoming)
346 {
347 	struct m_tag *mtag;
348 	struct divert_info *divinfo;
349 	int port;
350 
351 	M_ASSERTPKTHDR(m);
352 
353 	/* Assure header */
354 	if (m->m_len < sizeof(struct ip) &&
355 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
356 		return;
357 
358 	mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
359 	KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
360 	divinfo = m_tag_data(mtag);
361 
362 	port = divinfo->port;
363 	KASSERT(port != 0, ("%s: port=0", __func__));
364 
365 #ifdef SMP
366 	if (mycpuid != 0) {
367 		struct netmsg_packet *nmp;
368 		struct lwkt_msg *msg;
369 
370 		nmp = &m->m_hdr.mh_netmsg;
371 		netmsg_init(&nmp->nm_netmsg, NULL, &netisr_apanic_rport,
372 		    MSGF_MPSAFE, div_packet_handler);
373 		nmp->nm_packet = m;
374 
375 		msg = &nmp->nm_netmsg.nm_lmsg;
376 		msg->u.ms_result32 = port; /* port is 16bits */
377 		if (incoming)
378 			msg->u.ms_result32 |= DIV_INPUT;
379 		else
380 			msg->u.ms_result32 |= DIV_OUTPUT;
381 
382 		lwkt_sendmsg(cpu_portfn(0), &nmp->nm_netmsg.nm_lmsg);
383 	} else
384 #endif
385 	div_packet(m, incoming, port);
386 }
387 
388 /*
389  * Deliver packet back into the IP processing machinery.
390  *
391  * If no address specified, or address is 0.0.0.0, send to ip_output();
392  * otherwise, send to ip_input() and mark as having been received on
393  * the interface with that address.
394  */
395 static int
396 div_output(struct socket *so, struct mbuf *m,
397 	struct sockaddr_in *sin, struct mbuf *control)
398 {
399 	int error = 0;
400 	struct m_tag *mtag;
401 	struct divert_info *divinfo;
402 
403 	if (control)
404 		m_freem(control);		/* XXX */
405 
406 	/*
407 	 * Prepare the tag for divert info. Note that a packet
408 	 * with a 0 tag in mh_data is effectively untagged,
409 	 * so we could optimize that case.
410 	 */
411 	mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
412 	if (mtag == NULL) {
413 		error = ENOBUFS;
414 		goto cantsend;
415 	}
416 	m_tag_prepend(m, mtag);
417 
418 	/* Loopback avoidance and state recovery */
419 	divinfo = m_tag_data(mtag);
420 	if (sin)
421 		divinfo->skipto = sin->sin_port;
422 	else
423 		divinfo->skipto = 0;
424 
425 	/* Reinject packet into the system as incoming or outgoing */
426 	if (DIV_IS_OUTPUT(sin)) {
427 		struct ip *const ip = mtod(m, struct ip *);
428 
429 		/* Don't allow packet length sizes that will crash */
430 		if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
431 			error = EINVAL;
432 			goto cantsend;
433 		}
434 
435 		/* Convert fields to host order for ip_output() */
436 		ip->ip_len = ntohs(ip->ip_len);
437 		ip->ip_off = ntohs(ip->ip_off);
438 
439 		/* Send packet to output processing */
440 		ipstat.ips_rawout++;			/* XXX */
441 		error = ip_output(m, NULL, NULL,
442 			    (so->so_options & SO_DONTROUTE) |
443 			    IP_ALLOWBROADCAST | IP_RAWOUTPUT,
444 			    NULL, NULL);
445 	} else {
446 		ip_input(m);
447 	}
448 	return error;
449 
450 cantsend:
451 	m_freem(m);
452 	return error;
453 }
454 
455 static int
456 div_attach(struct socket *so, int proto, struct pru_attach_info *ai)
457 {
458 	struct inpcb *inp;
459 	int error;
460 
461 	inp  = so->so_pcb;
462 	if (inp)
463 		panic("div_attach");
464 	if ((error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY)) != 0)
465 		return error;
466 
467 	error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
468 	if (error)
469 		return error;
470 	error = in_pcballoc(so, &divcbinfo);
471 	if (error)
472 		return error;
473 	inp = (struct inpcb *)so->so_pcb;
474 	inp->inp_ip_p = proto;
475 	inp->inp_vflag |= INP_IPV4;
476 	inp->inp_flags |= INP_HDRINCL;
477 	/*
478 	 * The socket is always "connected" because
479 	 * we always know "where" to send the packet.
480 	 */
481 	so->so_port = cpu0_soport(so, NULL, NULL);
482 	so->so_state |= SS_ISCONNECTED;
483 	return 0;
484 }
485 
486 static int
487 div_detach(struct socket *so)
488 {
489 	struct inpcb *inp;
490 
491 	inp = so->so_pcb;
492 	if (inp == NULL)
493 		panic("div_detach");
494 	in_pcbdetach(inp);
495 	return 0;
496 }
497 
498 static int
499 div_abort(struct socket *so)
500 {
501 	soisdisconnected(so);
502 	return div_detach(so);
503 }
504 
505 static int
506 div_disconnect(struct socket *so)
507 {
508 	if (!(so->so_state & SS_ISCONNECTED))
509 		return ENOTCONN;
510 	return div_abort(so);
511 }
512 
513 static int
514 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
515 {
516 	int error;
517 
518 	/*
519 	 * in_pcbbind assumes that nam is a sockaddr_in
520 	 * and in_pcbbind requires a valid address. Since divert
521 	 * sockets don't we need to make sure the address is
522 	 * filled in properly.
523 	 * XXX -- divert should not be abusing in_pcbind
524 	 * and should probably have its own family.
525 	 */
526 	if (nam->sa_family != AF_INET) {
527 		error = EAFNOSUPPORT;
528 	} else {
529 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
530 		error = in_pcbbind(so->so_pcb, nam, td);
531 	}
532 	return error;
533 }
534 
535 static int
536 div_shutdown(struct socket *so)
537 {
538 	socantsendmore(so);
539 	return 0;
540 }
541 
542 static int
543 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
544 	 struct mbuf *control, struct thread *td)
545 {
546 	/* Length check already done in ip_mport() */
547 	KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
548 
549 	/* Send packet */
550 	return div_output(so, m, (struct sockaddr_in *)nam, control);
551 }
552 
553 SYSCTL_DECL(_net_inet_divert);
554 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
555 	    in_pcblist_global, "S,xinpcb", "List of active divert sockets");
556 
557 struct pr_usrreqs div_usrreqs = {
558 	.pru_abort = div_abort,
559 	.pru_accept = pru_accept_notsupp,
560 	.pru_attach = div_attach,
561 	.pru_bind = div_bind,
562 	.pru_connect = pru_connect_notsupp,
563 	.pru_connect2 = pru_connect2_notsupp,
564 	.pru_control = in_control,
565 	.pru_detach = div_detach,
566 	.pru_disconnect = div_disconnect,
567 	.pru_listen = pru_listen_notsupp,
568 	.pru_peeraddr = in_setpeeraddr,
569 	.pru_rcvd = pru_rcvd_notsupp,
570 	.pru_rcvoob = pru_rcvoob_notsupp,
571 	.pru_send = div_send,
572 	.pru_sense = pru_sense_null,
573 	.pru_shutdown = div_shutdown,
574 	.pru_sockaddr = in_setsockaddr,
575 	.pru_sosend = sosend,
576 	.pru_soreceive = soreceive,
577 	.pru_sopoll = sopoll
578 };
579 
580 static struct mbuf *
581 ip_divert_out(struct mbuf *m, int tee)
582 {
583 	struct mbuf *clone = NULL;
584 	struct ip *ip = mtod(m, struct ip *);
585 
586 	/* Clone packet if we're doing a 'tee' */
587 	if (tee)
588 		clone = m_dup(m, MB_DONTWAIT);
589 
590 	/*
591 	 * XXX
592 	 * delayed checksums are not currently compatible
593 	 * with divert sockets.
594 	 */
595 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
596 		in_delayed_cksum(m);
597 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
598 	}
599 
600 	/* Restore packet header fields to original values */
601 	ip->ip_len = htons(ip->ip_len);
602 	ip->ip_off = htons(ip->ip_off);
603 
604 	/* Deliver packet to divert input routine */
605 	divert_packet(m, 0);
606 
607 	/* If 'tee', continue with original packet */
608 	return clone;
609 }
610 
611 static struct mbuf *
612 ip_divert_in(struct mbuf *m, int tee)
613 {
614 	struct mbuf *clone = NULL;
615 	struct ip *ip = mtod(m, struct ip *);
616 	struct m_tag *mtag;
617 
618 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
619 		const struct divert_info *divinfo;
620 		u_short frag_off;
621 		int hlen;
622 
623 		/*
624 		 * Only trust divert info in the fragment
625 		 * at offset 0.
626 		 */
627 		frag_off = ip->ip_off << 3;
628 		if (frag_off != 0) {
629 			mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
630 			m_tag_delete(m, mtag);
631 		}
632 
633 		/*
634 		 * Attempt reassembly; if it succeeds, proceed.
635 		 * ip_reass() will return a different mbuf.
636 		 */
637 		m = ip_reass(m);
638 		if (m == NULL)
639 			return NULL;
640 		ip = mtod(m, struct ip *);
641 
642 		/* Caller need to redispatch the packet, if it is for us */
643 		m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
644 
645 		/*
646 		 * Get the header length of the reassembled
647 		 * packet
648 		 */
649 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
650 
651 		/*
652 		 * Restore original checksum before diverting
653 		 * packet
654 		 */
655 		ip->ip_len += hlen;
656 		ip->ip_len = htons(ip->ip_len);
657 		ip->ip_off = htons(ip->ip_off);
658 		ip->ip_sum = 0;
659 		if (hlen == sizeof(struct ip))
660 			ip->ip_sum = in_cksum_hdr(ip);
661 		else
662 			ip->ip_sum = in_cksum(m, hlen);
663 		ip->ip_off = ntohs(ip->ip_off);
664 		ip->ip_len = ntohs(ip->ip_len);
665 
666 		/*
667 		 * Only use the saved divert info
668 		 */
669 		mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
670 		if (mtag == NULL) {
671 			/* Wrongly configured ipfw */
672 			kprintf("ip_input no divert info\n");
673 			m_freem(m);
674 			return NULL;
675 		}
676 		divinfo = m_tag_data(mtag);
677 		tee = divinfo->tee;
678 	}
679 
680 	/*
681 	 * Divert or tee packet to the divert protocol if
682 	 * required.
683 	 */
684 
685 	/* Clone packet if we're doing a 'tee' */
686 	if (tee)
687 		clone = m_dup(m, MB_DONTWAIT);
688 
689 	/*
690 	 * Restore packet header fields to original
691 	 * values
692 	 */
693 	ip->ip_len = htons(ip->ip_len);
694 	ip->ip_off = htons(ip->ip_off);
695 
696 	/* Deliver packet to divert input routine */
697 	divert_packet(m, 1);
698 
699 	/* Catch invalid reference */
700 	m = NULL;
701 	ip = NULL;
702 
703 	ipstat.ips_delivered++;
704 
705 	/* If 'tee', continue with original packet */
706 	if (clone != NULL) {
707 		/*
708 		 * Complete processing of the packet.
709 		 * XXX Better safe than sorry, remove the DIVERT tag.
710 		 */
711 		mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
712 		KKASSERT(mtag != NULL);
713 		m_tag_delete(clone, mtag);
714 	}
715 	return clone;
716 }
717 
718 static struct mbuf *
719 ip_divert(struct mbuf *m, int tee, int incoming)
720 {
721 	struct mbuf *ret;
722 
723 	if (incoming)
724 		ret = ip_divert_in(m, tee);
725 	else
726 		ret = ip_divert_out(m, tee);
727 	return ret;
728 }
729