xref: /dragonfly/sys/netinet/ip_divert.c (revision a32bc35d)
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  */
35 
36 #define	_IP_VHL
37 
38 #include "opt_inet.h"
39 #include "opt_ipdivert.h"
40 #include "opt_ipsec.h"
41 
42 #ifndef INET
43 #error "IPDIVERT requires INET."
44 #endif
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/protosw.h>
52 #include <sys/socketvar.h>
53 #include <sys/socketvar2.h>
54 #include <sys/sysctl.h>
55 #include <sys/systm.h>
56 #include <sys/proc.h>
57 #include <sys/priv.h>
58 #include <sys/in_cksum.h>
59 #include <sys/lock.h>
60 #include <sys/msgport.h>
61 
62 #include <net/if.h>
63 #include <net/route.h>
64 
65 #include <net/netmsg2.h>
66 #include <sys/thread2.h>
67 #include <sys/mplock2.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/ip.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip_var.h>
75 #include <netinet/ip_divert.h>
76 
77 /*
78  * Divert sockets
79  */
80 
81 /*
82  * Allocate enough space to hold a full IP packet
83  */
84 #define	DIVSNDQ		(65536 + 100)
85 #define	DIVRCVQ		(65536 + 100)
86 
87 #define DIV_IS_OUTPUT(sin)	((sin) == NULL || (sin)->sin_addr.s_addr == 0)
88 
89 #define DIV_OUTPUT	0x10000
90 #define DIV_INPUT	0x20000
91 
92 /*
93  * Divert sockets work in conjunction with ipfw, see the divert(4)
94  * manpage for features.
95  * Internally, packets selected by ipfw in ip_input() or ip_output(),
96  * and never diverted before, are passed to the input queue of the
97  * divert socket with a given 'divert_port' number (as specified in
98  * the matching ipfw rule), and they are tagged with a 16 bit cookie
99  * (representing the rule number of the matching ipfw rule), which
100  * is passed to process reading from the socket.
101  *
102  * Packets written to the divert socket are again tagged with a cookie
103  * (usually the same as above) and a destination address.
104  * If the destination address is INADDR_ANY then the packet is
105  * treated as outgoing and sent to ip_output(), otherwise it is
106  * treated as incoming and sent to ip_input().
107  * In both cases, the packet is tagged with the cookie.
108  *
109  * On reinjection, processing in ip_input() and ip_output()
110  * will be exactly the same as for the original packet, except that
111  * ipfw processing will start at the rule number after the one
112  * written in the cookie (so, tagging a packet with a cookie of 0
113  * will cause it to be effectively considered as a standard packet).
114  */
115 
116 /* Internal variables */
117 static struct inpcbinfo divcbinfo;
118 
119 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
120 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
121 
122 static struct mbuf *ip_divert(struct mbuf *, int, int);
123 
124 static struct lwkt_token div_token = LWKT_TOKEN_INITIALIZER(div_token);
125 
126 /*
127  * Initialize divert connection block queue.
128  */
129 void
130 div_init(void)
131 {
132 	in_pcbinfo_init(&divcbinfo);
133 	/*
134 	 * XXX We don't use the hash list for divert IP, but it's easier
135 	 * to allocate a one entry hash list than it is to check all
136 	 * over the place for hashbase == NULL.
137 	 */
138 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
139 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
140 	divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
141 					      &divcbinfo.wildcardhashmask);
142 	divcbinfo.ipi_size = sizeof(struct inpcb);
143 	ip_divert_p = ip_divert;
144 }
145 
146 /*
147  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
148  * with that protocol number to enter the system from the outside.
149  */
150 int
151 div_input(struct mbuf **mp, int *offp, int proto)
152 {
153 	struct mbuf *m = *mp;
154 
155 	ipstat.ips_noproto++;
156 	m_freem(m);
157 	return(IPPROTO_DONE);
158 }
159 
160 /*
161  * Divert a packet by passing it up to the divert socket at port 'port'.
162  *
163  * Setup generic address and protocol structures for div_input routine,
164  * then pass them along with mbuf chain.
165  */
166 static void
167 div_packet(struct mbuf *m, int incoming, int port)
168 {
169 	struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
170 	struct inpcb *inp;
171 	struct socket *sa;
172 	struct m_tag *mtag;
173 	struct divert_info *divinfo;
174 	u_int16_t nport;
175 
176 	/* Locate the divert info */
177 	mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
178 	divinfo = m_tag_data(mtag);
179 	divsrc.sin_port = divinfo->skipto;
180 
181 	/*
182 	 * Record receive interface address, if any.
183 	 * But only for incoming packets.
184 	 */
185 	divsrc.sin_addr.s_addr = 0;
186 	if (incoming) {
187 		struct ifaddr_container *ifac;
188 
189 		/* Find IP address for receive interface */
190 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
191 			      ifa_link) {
192 			struct ifaddr *ifa = ifac->ifa;
193 
194 			if (ifa->ifa_addr == NULL)
195 				continue;
196 			if (ifa->ifa_addr->sa_family != AF_INET)
197 				continue;
198 			divsrc.sin_addr =
199 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
200 			break;
201 		}
202 	}
203 	/*
204 	 * Record the incoming interface name whenever we have one.
205 	 */
206 	if (m->m_pkthdr.rcvif) {
207 		/*
208 		 * Hide the actual interface name in there in the
209 		 * sin_zero array. XXX This needs to be moved to a
210 		 * different sockaddr type for divert, e.g.
211 		 * sockaddr_div with multiple fields like
212 		 * sockaddr_dl. Presently we have only 7 bytes
213 		 * but that will do for now as most interfaces
214 		 * are 4 or less + 2 or less bytes for unit.
215 		 * There is probably a faster way of doing this,
216 		 * possibly taking it from the sockaddr_dl on the iface.
217 		 * This solves the problem of a P2P link and a LAN interface
218 		 * having the same address, which can result in the wrong
219 		 * interface being assigned to the packet when fed back
220 		 * into the divert socket. Theoretically if the daemon saves
221 		 * and re-uses the sockaddr_in as suggested in the man pages,
222 		 * this iface name will come along for the ride.
223 		 * (see div_output for the other half of this.)
224 		 */
225 		ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero, "%s",
226 			  m->m_pkthdr.rcvif->if_xname);
227 	}
228 
229 	/* Put packet on socket queue, if any */
230 	sa = NULL;
231 	nport = htons((u_int16_t)port);
232 
233 	/*
234 	 * XXX
235 	 * Following loop to locate the inpcb is MPSAFE since the inpcb
236 	 * insertion/removal happens on the same CPU (CPU0), however,
237 	 * saving/testing the socket pointer is not MPSAFE.  So we still
238 	 * need to hold BGL here.
239 	 */
240 	lwkt_gettoken(&div_token);
241 	LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
242 		if (inp->inp_flags & INP_PLACEMARKER)
243 			continue;
244 		if (inp->inp_lport == nport)
245 			sa = inp->inp_socket;
246 	}
247 	if (sa) {
248 		lwkt_gettoken(&sa->so_rcv.ssb_token);
249 		if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
250 			m_freem(m);
251 		else
252 			sorwakeup(sa);
253 		lwkt_reltoken(&sa->so_rcv.ssb_token);
254 	} else {
255 		m_freem(m);
256 		ipstat.ips_noproto++;
257 		ipstat.ips_delivered--;
258 	}
259 	lwkt_reltoken(&div_token);
260 }
261 
262 #ifdef SMP
263 
264 static void
265 div_packet_handler(netmsg_t msg)
266 {
267 	struct mbuf *m;
268 	int port, incoming = 0;
269 
270 	m = msg->packet.nm_packet;
271 
272 	port = msg->lmsg.u.ms_result32 & 0xffff;
273 	if (msg->lmsg.u.ms_result32 & DIV_INPUT)
274 		incoming = 1;
275 	div_packet(m, incoming, port);
276 	/* no reply, msg embedded in mbuf */
277 }
278 
279 #endif	/* SMP */
280 
281 static void
282 divert_packet(struct mbuf *m, int incoming)
283 {
284 	struct m_tag *mtag;
285 	struct divert_info *divinfo;
286 	int port;
287 
288 	M_ASSERTPKTHDR(m);
289 
290 	/* Assure header */
291 	if (m->m_len < sizeof(struct ip) &&
292 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
293 		return;
294 
295 	mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
296 	KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
297 	divinfo = m_tag_data(mtag);
298 
299 	port = divinfo->port;
300 	KASSERT(port != 0, ("%s: port=0", __func__));
301 
302 #ifdef SMP
303 	if (mycpuid != 0) {
304 		struct netmsg_packet *nmp;
305 
306 		nmp = &m->m_hdr.mh_netmsg;
307 		netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
308 			    0, div_packet_handler);
309 		nmp->nm_packet = m;
310 
311 		nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
312 		if (incoming)
313 			nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
314 		else
315 			nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
316 
317 		lwkt_sendmsg(cpu_portfn(0), &nmp->base.lmsg);
318 	} else {
319 		div_packet(m, incoming, port);
320 	}
321 #else
322 	div_packet(m, incoming, port);
323 #endif
324 }
325 
326 /*
327  * Deliver packet back into the IP processing machinery.
328  *
329  * If no address specified, or address is 0.0.0.0, send to ip_output();
330  * otherwise, send to ip_input() and mark as having been received on
331  * the interface with that address.
332  */
333 static int
334 div_output(struct socket *so, struct mbuf *m,
335 	struct sockaddr_in *sin, struct mbuf *control)
336 {
337 	int error = 0;
338 	struct m_tag *mtag;
339 	struct divert_info *divinfo;
340 
341 	if (control)
342 		m_freem(control);		/* XXX */
343 
344 	/*
345 	 * Prepare the tag for divert info. Note that a packet
346 	 * with a 0 tag in mh_data is effectively untagged,
347 	 * so we could optimize that case.
348 	 */
349 	mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
350 	if (mtag == NULL) {
351 		error = ENOBUFS;
352 		goto cantsend;
353 	}
354 	m_tag_prepend(m, mtag);
355 
356 	/* Loopback avoidance and state recovery */
357 	divinfo = m_tag_data(mtag);
358 	if (sin)
359 		divinfo->skipto = sin->sin_port;
360 	else
361 		divinfo->skipto = 0;
362 
363 	/* Reinject packet into the system as incoming or outgoing */
364 	if (DIV_IS_OUTPUT(sin)) {
365 		struct ip *const ip = mtod(m, struct ip *);
366 
367 		/* Don't allow packet length sizes that will crash */
368 		if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
369 			error = EINVAL;
370 			goto cantsend;
371 		}
372 
373 		/* Convert fields to host order for ip_output() */
374 		ip->ip_len = ntohs(ip->ip_len);
375 		ip->ip_off = ntohs(ip->ip_off);
376 
377 		/* Send packet to output processing */
378 		ipstat.ips_rawout++;			/* XXX */
379 		error = ip_output(m, NULL, NULL,
380 			    (so->so_options & SO_DONTROUTE) |
381 			    IP_ALLOWBROADCAST | IP_RAWOUTPUT,
382 			    NULL, NULL);
383 	} else {
384 		ip_input(m);
385 	}
386 	return error;
387 
388 cantsend:
389 	m_freem(m);
390 	return error;
391 }
392 
393 static void
394 div_attach(netmsg_t msg)
395 {
396 	struct socket *so = msg->attach.base.nm_so;
397 	int proto = msg->attach.nm_proto;
398 	struct pru_attach_info *ai = msg->attach.nm_ai;
399 	struct inpcb *inp;
400 	int error;
401 
402 	inp  = so->so_pcb;
403 	if (inp)
404 		panic("div_attach");
405 	error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY);
406 	if (error)
407 		goto out;
408 
409 	error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
410 	if (error)
411 		goto out;
412 	lwkt_gettoken(&div_token);
413 	sosetport(so, cpu_portfn(0));
414 	error = in_pcballoc(so, &divcbinfo);
415 	if (error) {
416 		lwkt_reltoken(&div_token);
417 		goto out;
418 	}
419 	inp = (struct inpcb *)so->so_pcb;
420 	inp->inp_ip_p = proto;
421 	inp->inp_vflag |= INP_IPV4;
422 	inp->inp_flags |= INP_HDRINCL;
423 	/*
424 	 * The socket is always "connected" because
425 	 * we always know "where" to send the packet.
426 	 */
427 	sosetstate(so, SS_ISCONNECTED);
428 	lwkt_reltoken(&div_token);
429 	error = 0;
430 out:
431 	lwkt_replymsg(&msg->attach.base.lmsg, error);
432 }
433 
434 static void
435 div_detach(netmsg_t msg)
436 {
437 	struct socket *so = msg->detach.base.nm_so;
438 	struct inpcb *inp;
439 
440 	inp = so->so_pcb;
441 	if (inp == NULL)
442 		panic("div_detach");
443 	in_pcbdetach(inp);
444 	lwkt_replymsg(&msg->detach.base.lmsg, 0);
445 }
446 
447 /*
448  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
449  *	 will sofree() it when we return.
450  */
451 static void
452 div_abort(netmsg_t msg)
453 {
454 	struct socket *so = msg->abort.base.nm_so;
455 
456 	soisdisconnected(so);
457 	div_detach(msg);
458 	/* msg invalid now */
459 }
460 
461 static void
462 div_disconnect(netmsg_t msg)
463 {
464 	struct socket *so = msg->disconnect.base.nm_so;
465 	int error;
466 
467 	if (so->so_state & SS_ISCONNECTED) {
468 		soreference(so);
469 		div_abort(msg);
470 		/* msg invalid now */
471 		sofree(so);
472 		return;
473 	}
474 	error = ENOTCONN;
475 	lwkt_replymsg(&msg->disconnect.base.lmsg, error);
476 }
477 
478 static void
479 div_bind(netmsg_t msg)
480 {
481 	struct socket *so = msg->bind.base.nm_so;
482 	struct sockaddr *nam = msg->bind.nm_nam;
483 	int error;
484 
485 	/*
486 	 * in_pcbbind assumes that nam is a sockaddr_in
487 	 * and in_pcbbind requires a valid address. Since divert
488 	 * sockets don't we need to make sure the address is
489 	 * filled in properly.
490 	 * XXX -- divert should not be abusing in_pcbind
491 	 * and should probably have its own family.
492 	 */
493 	if (nam->sa_family != AF_INET) {
494 		error = EAFNOSUPPORT;
495 	} else {
496 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
497 		error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
498 	}
499 	lwkt_replymsg(&msg->bind.base.lmsg, error);
500 }
501 
502 static void
503 div_shutdown(netmsg_t msg)
504 {
505 	struct socket *so = msg->shutdown.base.nm_so;
506 
507 	socantsendmore(so);
508 
509 	lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
510 }
511 
512 static void
513 div_send(netmsg_t msg)
514 {
515 	struct socket *so = msg->send.base.nm_so;
516 	struct mbuf *m = msg->send.nm_m;
517 	struct sockaddr *nam = msg->send.nm_addr;
518 	struct mbuf *control = msg->send.nm_control;
519 	int error;
520 
521 	/* Length check already done in ip_cpufn() */
522 	KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
523 
524 	/* Send packet */
525 	error = div_output(so, m, (struct sockaddr_in *)nam, control);
526 	lwkt_replymsg(&msg->send.base.lmsg, error);
527 }
528 
529 SYSCTL_DECL(_net_inet_divert);
530 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
531 	    in_pcblist_global, "S,xinpcb", "List of active divert sockets");
532 
533 struct pr_usrreqs div_usrreqs = {
534 	.pru_abort = div_abort,
535 	.pru_accept = pr_generic_notsupp,
536 	.pru_attach = div_attach,
537 	.pru_bind = div_bind,
538 	.pru_connect = pr_generic_notsupp,
539 	.pru_connect2 = pr_generic_notsupp,
540 	.pru_control = in_control_dispatch,
541 	.pru_detach = div_detach,
542 	.pru_disconnect = div_disconnect,
543 	.pru_listen = pr_generic_notsupp,
544 	.pru_peeraddr = in_setpeeraddr_dispatch,
545 	.pru_rcvd = pr_generic_notsupp,
546 	.pru_rcvoob = pr_generic_notsupp,
547 	.pru_send = div_send,
548 	.pru_sense = pru_sense_null,
549 	.pru_shutdown = div_shutdown,
550 	.pru_sockaddr = in_setsockaddr_dispatch,
551 	.pru_sosend = sosend,
552 	.pru_soreceive = soreceive
553 };
554 
555 static struct mbuf *
556 ip_divert_out(struct mbuf *m, int tee)
557 {
558 	struct mbuf *clone = NULL;
559 	struct ip *ip = mtod(m, struct ip *);
560 
561 	/* Clone packet if we're doing a 'tee' */
562 	if (tee)
563 		clone = m_dup(m, MB_DONTWAIT);
564 
565 	/*
566 	 * XXX
567 	 * delayed checksums are not currently compatible
568 	 * with divert sockets.
569 	 */
570 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
571 		in_delayed_cksum(m);
572 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
573 	}
574 
575 	/* Restore packet header fields to original values */
576 	ip->ip_len = htons(ip->ip_len);
577 	ip->ip_off = htons(ip->ip_off);
578 
579 	/* Deliver packet to divert input routine */
580 	divert_packet(m, 0);
581 
582 	/* If 'tee', continue with original packet */
583 	return clone;
584 }
585 
586 static struct mbuf *
587 ip_divert_in(struct mbuf *m, int tee)
588 {
589 	struct mbuf *clone = NULL;
590 	struct ip *ip = mtod(m, struct ip *);
591 	struct m_tag *mtag;
592 
593 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
594 		const struct divert_info *divinfo;
595 		u_short frag_off;
596 		int hlen;
597 
598 		/*
599 		 * Only trust divert info in the fragment
600 		 * at offset 0.
601 		 */
602 		frag_off = ip->ip_off << 3;
603 		if (frag_off != 0) {
604 			mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
605 			m_tag_delete(m, mtag);
606 		}
607 
608 		/*
609 		 * Attempt reassembly; if it succeeds, proceed.
610 		 * ip_reass() will return a different mbuf.
611 		 */
612 		m = ip_reass(m);
613 		if (m == NULL)
614 			return NULL;
615 		ip = mtod(m, struct ip *);
616 
617 		/* Caller need to redispatch the packet, if it is for us */
618 		m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
619 
620 		/*
621 		 * Get the header length of the reassembled
622 		 * packet
623 		 */
624 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
625 
626 		/*
627 		 * Restore original checksum before diverting
628 		 * packet
629 		 */
630 		ip->ip_len += hlen;
631 		ip->ip_len = htons(ip->ip_len);
632 		ip->ip_off = htons(ip->ip_off);
633 		ip->ip_sum = 0;
634 		if (hlen == sizeof(struct ip))
635 			ip->ip_sum = in_cksum_hdr(ip);
636 		else
637 			ip->ip_sum = in_cksum(m, hlen);
638 		ip->ip_off = ntohs(ip->ip_off);
639 		ip->ip_len = ntohs(ip->ip_len);
640 
641 		/*
642 		 * Only use the saved divert info
643 		 */
644 		mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
645 		if (mtag == NULL) {
646 			/* Wrongly configured ipfw */
647 			kprintf("ip_input no divert info\n");
648 			m_freem(m);
649 			return NULL;
650 		}
651 		divinfo = m_tag_data(mtag);
652 		tee = divinfo->tee;
653 	}
654 
655 	/*
656 	 * Divert or tee packet to the divert protocol if
657 	 * required.
658 	 */
659 
660 	/* Clone packet if we're doing a 'tee' */
661 	if (tee)
662 		clone = m_dup(m, MB_DONTWAIT);
663 
664 	/*
665 	 * Restore packet header fields to original
666 	 * values
667 	 */
668 	ip->ip_len = htons(ip->ip_len);
669 	ip->ip_off = htons(ip->ip_off);
670 
671 	/* Deliver packet to divert input routine */
672 	divert_packet(m, 1);
673 
674 	/* Catch invalid reference */
675 	m = NULL;
676 	ip = NULL;
677 
678 	ipstat.ips_delivered++;
679 
680 	/* If 'tee', continue with original packet */
681 	if (clone != NULL) {
682 		/*
683 		 * Complete processing of the packet.
684 		 * XXX Better safe than sorry, remove the DIVERT tag.
685 		 */
686 		mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
687 		KKASSERT(mtag != NULL);
688 		m_tag_delete(clone, mtag);
689 	}
690 	return clone;
691 }
692 
693 static struct mbuf *
694 ip_divert(struct mbuf *m, int tee, int incoming)
695 {
696 	struct mbuf *ret;
697 
698 	if (incoming)
699 		ret = ip_divert_in(m, tee);
700 	else
701 		ret = ip_divert_out(m, tee);
702 	return ret;
703 }
704