xref: /dragonfly/sys/netinet/ip_divert.c (revision 61c0377f)
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. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
30  */
31 
32 #define	_IP_VHL
33 
34 #include "opt_inet.h"
35 #include "opt_ipdivert.h"
36 #include "opt_ipsec.h"
37 
38 #ifndef INET
39 #error "IPDIVERT requires INET."
40 #endif
41 
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/socket.h>
47 #include <sys/protosw.h>
48 #include <sys/socketvar.h>
49 #include <sys/socketvar2.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/proc.h>
53 #include <sys/priv.h>
54 #include <sys/in_cksum.h>
55 #include <sys/lock.h>
56 #include <sys/msgport.h>
57 
58 #include <net/if.h>
59 #include <net/route.h>
60 
61 #include <net/netmsg2.h>
62 #include <net/netisr2.h>
63 #include <sys/thread2.h>
64 #include <sys/mplock2.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/ip.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_divert.h>
73 
74 /*
75  * Divert sockets
76  */
77 
78 /*
79  * Allocate enough space to hold a full IP packet
80  */
81 #define	DIVSNDQ		(65536 + 100)
82 #define	DIVRCVQ		(65536 + 100)
83 
84 #define DIV_IS_OUTPUT(sin)	((sin) == NULL || (sin)->sin_addr.s_addr == 0)
85 
86 #define DIV_OUTPUT	0x10000
87 #define DIV_INPUT	0x20000
88 
89 /*
90  * Divert sockets work in conjunction with ipfw, see the divert(4)
91  * manpage for features.
92  * Internally, packets selected by ipfw in ip_input() or ip_output(),
93  * and never diverted before, are passed to the input queue of the
94  * divert socket with a given 'divert_port' number (as specified in
95  * the matching ipfw rule), and they are tagged with a 16 bit cookie
96  * (representing the rule number of the matching ipfw rule), which
97  * is passed to process reading from the socket.
98  *
99  * Packets written to the divert socket are again tagged with a cookie
100  * (usually the same as above) and a destination address.
101  * If the destination address is INADDR_ANY then the packet is
102  * treated as outgoing and sent to ip_output(), otherwise it is
103  * treated as incoming and sent to ip_input().
104  * In both cases, the packet is tagged with the cookie.
105  *
106  * On reinjection, processing in ip_input() and ip_output()
107  * will be exactly the same as for the original packet, except that
108  * ipfw processing will start at the rule number after the one
109  * written in the cookie (so, tagging a packet with a cookie of 0
110  * will cause it to be effectively considered as a standard packet).
111  */
112 
113 /* Internal variables */
114 static struct inpcbinfo divcbinfo;
115 static struct inpcbportinfo divcbportinfo;
116 
117 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
118 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
119 
120 static struct mbuf *ip_divert(struct mbuf *, int, int);
121 
122 static struct lwkt_token div_token = LWKT_TOKEN_INITIALIZER(div_token);
123 
124 /*
125  * Initialize divert connection block queue.
126  */
127 void
128 div_init(void)
129 {
130 	in_pcbinfo_init(&divcbinfo);
131 	in_pcbportinfo_init(&divcbportinfo, 1, FALSE, 0);
132 	/*
133 	 * XXX We don't use the hash list for divert IP, but it's easier
134 	 * to allocate a one entry hash list than it is to check all
135 	 * over the place for hashbase == NULL.
136 	 */
137 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
138 	divcbinfo.portinfo = &divcbportinfo;
139 	divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
140 					      &divcbinfo.wildcardhashmask);
141 	divcbinfo.ipi_size = sizeof(struct inpcb);
142 	ip_divert_p = ip_divert;
143 }
144 
145 /*
146  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
147  * with that protocol number to enter the system from the outside.
148  */
149 int
150 div_input(struct mbuf **mp, int *offp, int proto)
151 {
152 	struct mbuf *m = *mp;
153 
154 	ipstat.ips_noproto++;
155 	m_freem(m);
156 	return(IPPROTO_DONE);
157 }
158 
159 /*
160  * Divert a packet by passing it up to the divert socket at port 'port'.
161  *
162  * Setup generic address and protocol structures for div_input routine,
163  * then pass them along with mbuf chain.
164  */
165 static void
166 div_packet(struct mbuf *m, int incoming, int port)
167 {
168 	struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
169 	struct inpcb *inp;
170 	struct socket *sa;
171 	struct m_tag *mtag;
172 	struct divert_info *divinfo;
173 	u_int16_t nport;
174 
175 	/* Locate the divert info */
176 	mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
177 	divinfo = m_tag_data(mtag);
178 	divsrc.sin_port = divinfo->skipto;
179 
180 	/*
181 	 * Record receive interface address, if any.
182 	 * But only for incoming packets.
183 	 */
184 	divsrc.sin_addr.s_addr = 0;
185 	if (incoming) {
186 		struct ifaddr_container *ifac;
187 
188 		/* Find IP address for receive interface */
189 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
190 			      ifa_link) {
191 			struct ifaddr *ifa = ifac->ifa;
192 
193 			if (ifa->ifa_addr == NULL)
194 				continue;
195 			if (ifa->ifa_addr->sa_family != AF_INET)
196 				continue;
197 			divsrc.sin_addr =
198 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
199 			break;
200 		}
201 	}
202 	/*
203 	 * Record the incoming interface name whenever we have one.
204 	 */
205 	if (m->m_pkthdr.rcvif) {
206 		/*
207 		 * Hide the actual interface name in there in the
208 		 * sin_zero array. XXX This needs to be moved to a
209 		 * different sockaddr type for divert, e.g.
210 		 * sockaddr_div with multiple fields like
211 		 * sockaddr_dl. Presently we have only 7 bytes
212 		 * but that will do for now as most interfaces
213 		 * are 4 or less + 2 or less bytes for unit.
214 		 * There is probably a faster way of doing this,
215 		 * possibly taking it from the sockaddr_dl on the iface.
216 		 * This solves the problem of a P2P link and a LAN interface
217 		 * having the same address, which can result in the wrong
218 		 * interface being assigned to the packet when fed back
219 		 * into the divert socket. Theoretically if the daemon saves
220 		 * and re-uses the sockaddr_in as suggested in the man pages,
221 		 * this iface name will come along for the ride.
222 		 * (see div_output for the other half of this.)
223 		 */
224 		ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero, "%s",
225 			  m->m_pkthdr.rcvif->if_xname);
226 	}
227 
228 	/* Put packet on socket queue, if any */
229 	sa = NULL;
230 	nport = htons((u_int16_t)port);
231 
232 	/*
233 	 * XXX
234 	 * Following loop to locate the inpcb is MPSAFE since the inpcb
235 	 * insertion/removal happens on the same CPU (CPU0), however,
236 	 * saving/testing the socket pointer is not MPSAFE.  So we still
237 	 * need to hold BGL here.
238 	 */
239 	lwkt_gettoken(&div_token);
240 	LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
241 		if (inp->inp_flags & INP_PLACEMARKER)
242 			continue;
243 		if (inp->inp_lport == nport)
244 			sa = inp->inp_socket;
245 	}
246 	if (sa) {
247 		lwkt_gettoken(&sa->so_rcv.ssb_token);
248 		if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
249 			m_freem(m);
250 		else
251 			sorwakeup(sa);
252 		lwkt_reltoken(&sa->so_rcv.ssb_token);
253 	} else {
254 		m_freem(m);
255 		ipstat.ips_noproto++;
256 		ipstat.ips_delivered--;
257 	}
258 	lwkt_reltoken(&div_token);
259 }
260 
261 static void
262 div_packet_handler(netmsg_t msg)
263 {
264 	struct mbuf *m;
265 	int port, incoming = 0;
266 
267 	m = msg->packet.nm_packet;
268 
269 	port = msg->lmsg.u.ms_result32 & 0xffff;
270 	if (msg->lmsg.u.ms_result32 & DIV_INPUT)
271 		incoming = 1;
272 	div_packet(m, incoming, port);
273 	/* no reply, msg embedded in mbuf */
274 }
275 
276 static void
277 divert_packet(struct mbuf *m, int incoming)
278 {
279 	struct m_tag *mtag;
280 	struct divert_info *divinfo;
281 	int port;
282 
283 	M_ASSERTPKTHDR(m);
284 
285 	/* Assure header */
286 	if (m->m_len < sizeof(struct ip) &&
287 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
288 		return;
289 
290 	mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
291 	KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
292 	divinfo = m_tag_data(mtag);
293 
294 	port = divinfo->port;
295 	KASSERT(port != 0, ("%s: port=0", __func__));
296 
297 	if (mycpuid != 0) {
298 		struct netmsg_packet *nmp;
299 
300 		nmp = &m->m_hdr.mh_netmsg;
301 		netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
302 			    0, div_packet_handler);
303 		nmp->nm_packet = m;
304 
305 		nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
306 		if (incoming)
307 			nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
308 		else
309 			nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
310 
311 		lwkt_sendmsg(netisr_cpuport(0), &nmp->base.lmsg);
312 	} else {
313 		div_packet(m, incoming, port);
314 	}
315 }
316 
317 /*
318  * Deliver packet back into the IP processing machinery.
319  *
320  * If no address specified, or address is 0.0.0.0, send to ip_output();
321  * otherwise, send to ip_input() and mark as having been received on
322  * the interface with that address.
323  */
324 static int
325 div_output(struct socket *so, struct mbuf *m,
326 	struct sockaddr_in *sin, struct mbuf *control)
327 {
328 	int error = 0;
329 	struct m_tag *mtag;
330 	struct divert_info *divinfo;
331 
332 	if (control)
333 		m_freem(control);		/* XXX */
334 
335 	/*
336 	 * Prepare the tag for divert info. Note that a packet
337 	 * with a 0 tag in mh_data is effectively untagged,
338 	 * so we could optimize that case.
339 	 */
340 	mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
341 	if (mtag == NULL) {
342 		error = ENOBUFS;
343 		goto cantsend;
344 	}
345 	m_tag_prepend(m, mtag);
346 
347 	/* Loopback avoidance and state recovery */
348 	divinfo = m_tag_data(mtag);
349 	if (sin)
350 		divinfo->skipto = sin->sin_port;
351 	else
352 		divinfo->skipto = 0;
353 
354 	/* Reinject packet into the system as incoming or outgoing */
355 	if (DIV_IS_OUTPUT(sin)) {
356 		struct ip *const ip = mtod(m, struct ip *);
357 
358 		/* Don't allow packet length sizes that will crash */
359 		if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
360 			error = EINVAL;
361 			goto cantsend;
362 		}
363 
364 		/* Convert fields to host order for ip_output() */
365 		ip->ip_len = ntohs(ip->ip_len);
366 		ip->ip_off = ntohs(ip->ip_off);
367 
368 		/* Send packet to output processing */
369 		ipstat.ips_rawout++;			/* XXX */
370 		error = ip_output(m, NULL, NULL,
371 			    (so->so_options & SO_DONTROUTE) |
372 			    IP_ALLOWBROADCAST | IP_RAWOUTPUT,
373 			    NULL, NULL);
374 	} else {
375 		ip_input(m);
376 	}
377 	return error;
378 
379 cantsend:
380 	m_freem(m);
381 	return error;
382 }
383 
384 static void
385 div_attach(netmsg_t msg)
386 {
387 	struct socket *so = msg->attach.base.nm_so;
388 	int proto = msg->attach.nm_proto;
389 	struct pru_attach_info *ai = msg->attach.nm_ai;
390 	struct inpcb *inp;
391 	int error;
392 
393 	inp  = so->so_pcb;
394 	if (inp)
395 		panic("div_attach");
396 	error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY);
397 	if (error)
398 		goto out;
399 
400 	error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
401 	if (error)
402 		goto out;
403 	lwkt_gettoken(&div_token);
404 	sosetport(so, netisr_cpuport(0));
405 	error = in_pcballoc(so, &divcbinfo);
406 	if (error) {
407 		lwkt_reltoken(&div_token);
408 		goto out;
409 	}
410 	inp = (struct inpcb *)so->so_pcb;
411 	inp->inp_ip_p = proto;
412 	inp->inp_vflag |= INP_IPV4;
413 	inp->inp_flags |= INP_HDRINCL;
414 	/*
415 	 * The socket is always "connected" because
416 	 * we always know "where" to send the packet.
417 	 */
418 	sosetstate(so, SS_ISCONNECTED);
419 	lwkt_reltoken(&div_token);
420 	error = 0;
421 out:
422 	lwkt_replymsg(&msg->attach.base.lmsg, error);
423 }
424 
425 static void
426 div_detach(netmsg_t msg)
427 {
428 	struct socket *so = msg->detach.base.nm_so;
429 	struct inpcb *inp;
430 
431 	inp = so->so_pcb;
432 	if (inp == NULL)
433 		panic("div_detach");
434 	in_pcbdetach(inp);
435 	lwkt_replymsg(&msg->detach.base.lmsg, 0);
436 }
437 
438 /*
439  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
440  *	 will sofree() it when we return.
441  */
442 static void
443 div_abort(netmsg_t msg)
444 {
445 	struct socket *so = msg->abort.base.nm_so;
446 
447 	soisdisconnected(so);
448 	div_detach(msg);
449 	/* msg invalid now */
450 }
451 
452 static void
453 div_disconnect(netmsg_t msg)
454 {
455 	struct socket *so = msg->disconnect.base.nm_so;
456 	int error;
457 
458 	if (so->so_state & SS_ISCONNECTED) {
459 		soreference(so);
460 		div_abort(msg);
461 		/* msg invalid now */
462 		sofree(so);
463 		return;
464 	}
465 	error = ENOTCONN;
466 	lwkt_replymsg(&msg->disconnect.base.lmsg, error);
467 }
468 
469 static void
470 div_bind(netmsg_t msg)
471 {
472 	struct socket *so = msg->bind.base.nm_so;
473 	struct sockaddr *nam = msg->bind.nm_nam;
474 	int error;
475 
476 	/*
477 	 * in_pcbbind assumes that nam is a sockaddr_in
478 	 * and in_pcbbind requires a valid address. Since divert
479 	 * sockets don't we need to make sure the address is
480 	 * filled in properly.
481 	 * XXX -- divert should not be abusing in_pcbind
482 	 * and should probably have its own family.
483 	 */
484 	if (nam->sa_family != AF_INET) {
485 		error = EAFNOSUPPORT;
486 	} else {
487 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
488 		error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
489 	}
490 	lwkt_replymsg(&msg->bind.base.lmsg, error);
491 }
492 
493 static void
494 div_shutdown(netmsg_t msg)
495 {
496 	struct socket *so = msg->shutdown.base.nm_so;
497 
498 	socantsendmore(so);
499 
500 	lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
501 }
502 
503 static void
504 div_send(netmsg_t msg)
505 {
506 	struct socket *so = msg->send.base.nm_so;
507 	struct mbuf *m = msg->send.nm_m;
508 	struct sockaddr *nam = msg->send.nm_addr;
509 	struct mbuf *control = msg->send.nm_control;
510 	int error;
511 
512 	/* Length check already done in ip_hashfn() */
513 	KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
514 
515 	/* Send packet */
516 	error = div_output(so, m, (struct sockaddr_in *)nam, control);
517 	lwkt_replymsg(&msg->send.base.lmsg, error);
518 }
519 
520 SYSCTL_DECL(_net_inet_divert);
521 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
522 	    in_pcblist_global, "S,xinpcb", "List of active divert sockets");
523 
524 struct pr_usrreqs div_usrreqs = {
525 	.pru_abort = div_abort,
526 	.pru_accept = pr_generic_notsupp,
527 	.pru_attach = div_attach,
528 	.pru_bind = div_bind,
529 	.pru_connect = pr_generic_notsupp,
530 	.pru_connect2 = pr_generic_notsupp,
531 	.pru_control = in_control_dispatch,
532 	.pru_detach = div_detach,
533 	.pru_disconnect = div_disconnect,
534 	.pru_listen = pr_generic_notsupp,
535 	.pru_peeraddr = in_setpeeraddr_dispatch,
536 	.pru_rcvd = pr_generic_notsupp,
537 	.pru_rcvoob = pr_generic_notsupp,
538 	.pru_send = div_send,
539 	.pru_sense = pru_sense_null,
540 	.pru_shutdown = div_shutdown,
541 	.pru_sockaddr = in_setsockaddr_dispatch,
542 	.pru_sosend = sosend,
543 	.pru_soreceive = soreceive
544 };
545 
546 static struct mbuf *
547 ip_divert_out(struct mbuf *m, int tee)
548 {
549 	struct mbuf *clone = NULL;
550 	struct ip *ip = mtod(m, struct ip *);
551 
552 	/* Clone packet if we're doing a 'tee' */
553 	if (tee)
554 		clone = m_dup(m, MB_DONTWAIT);
555 
556 	/*
557 	 * XXX
558 	 * delayed checksums are not currently compatible
559 	 * with divert sockets.
560 	 */
561 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
562 		in_delayed_cksum(m);
563 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
564 	}
565 
566 	/* Restore packet header fields to original values */
567 	ip->ip_len = htons(ip->ip_len);
568 	ip->ip_off = htons(ip->ip_off);
569 
570 	/* Deliver packet to divert input routine */
571 	divert_packet(m, 0);
572 
573 	/* If 'tee', continue with original packet */
574 	return clone;
575 }
576 
577 static struct mbuf *
578 ip_divert_in(struct mbuf *m, int tee)
579 {
580 	struct mbuf *clone = NULL;
581 	struct ip *ip = mtod(m, struct ip *);
582 	struct m_tag *mtag;
583 
584 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
585 		const struct divert_info *divinfo;
586 		u_short frag_off;
587 		int hlen;
588 
589 		/*
590 		 * Only trust divert info in the fragment
591 		 * at offset 0.
592 		 */
593 		frag_off = ip->ip_off << 3;
594 		if (frag_off != 0) {
595 			mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
596 			m_tag_delete(m, mtag);
597 		}
598 
599 		/*
600 		 * Attempt reassembly; if it succeeds, proceed.
601 		 * ip_reass() will return a different mbuf.
602 		 */
603 		m = ip_reass(m);
604 		if (m == NULL)
605 			return NULL;
606 		ip = mtod(m, struct ip *);
607 
608 		/* Caller need to redispatch the packet, if it is for us */
609 		m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
610 
611 		/*
612 		 * Get the header length of the reassembled
613 		 * packet
614 		 */
615 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
616 
617 		/*
618 		 * Restore original checksum before diverting
619 		 * packet
620 		 */
621 		ip->ip_len += hlen;
622 		ip->ip_len = htons(ip->ip_len);
623 		ip->ip_off = htons(ip->ip_off);
624 		ip->ip_sum = 0;
625 		if (hlen == sizeof(struct ip))
626 			ip->ip_sum = in_cksum_hdr(ip);
627 		else
628 			ip->ip_sum = in_cksum(m, hlen);
629 		ip->ip_off = ntohs(ip->ip_off);
630 		ip->ip_len = ntohs(ip->ip_len);
631 
632 		/*
633 		 * Only use the saved divert info
634 		 */
635 		mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
636 		if (mtag == NULL) {
637 			/* Wrongly configured ipfw */
638 			kprintf("ip_input no divert info\n");
639 			m_freem(m);
640 			return NULL;
641 		}
642 		divinfo = m_tag_data(mtag);
643 		tee = divinfo->tee;
644 	}
645 
646 	/*
647 	 * Divert or tee packet to the divert protocol if
648 	 * required.
649 	 */
650 
651 	/* Clone packet if we're doing a 'tee' */
652 	if (tee)
653 		clone = m_dup(m, MB_DONTWAIT);
654 
655 	/*
656 	 * Restore packet header fields to original
657 	 * values
658 	 */
659 	ip->ip_len = htons(ip->ip_len);
660 	ip->ip_off = htons(ip->ip_off);
661 
662 	/* Deliver packet to divert input routine */
663 	divert_packet(m, 1);
664 
665 	/* Catch invalid reference */
666 	m = NULL;
667 	ip = NULL;
668 
669 	ipstat.ips_delivered++;
670 
671 	/* If 'tee', continue with original packet */
672 	if (clone != NULL) {
673 		/*
674 		 * Complete processing of the packet.
675 		 * XXX Better safe than sorry, remove the DIVERT tag.
676 		 */
677 		mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
678 		KKASSERT(mtag != NULL);
679 		m_tag_delete(clone, mtag);
680 	}
681 	return clone;
682 }
683 
684 static struct mbuf *
685 ip_divert(struct mbuf *m, int tee, int incoming)
686 {
687 	struct mbuf *ret;
688 
689 	if (incoming)
690 		ret = ip_divert_in(m, tee);
691 	else
692 		ret = ip_divert_out(m, tee);
693 	return ret;
694 }
695