xref: /freebsd/sys/netinet/ip_divert.c (revision 681ce946f33e75c590e97c53076e86dff1fe8f4a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_sctp.h"
38 #ifndef INET
39 #error "IPDIVERT requires INET"
40 #endif
41 
42 #include <sys/param.h>
43 #include <sys/eventhandler.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/module.h>
49 #include <sys/kernel.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
56 #include <net/vnet.h>
57 
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/netisr.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_var.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #endif
72 #if defined(SCTP) || defined(SCTP_SUPPORT)
73 #include <netinet/sctp_crc32.h>
74 #endif
75 
76 #include <security/mac/mac_framework.h>
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 /*
88  * Divert sockets work in conjunction with ipfw or other packet filters,
89  * see the divert(4) manpage for features.
90  * Packets are selected by the packet filter and tagged with an
91  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
92  * the packet filter) and information on the matching filter rule for
93  * subsequent reinjection. The divert_port is used to put the packet
94  * on the corresponding divert socket, while the rule number is passed
95  * up (at least partially) as the sin_port in the struct sockaddr.
96  *
97  * Packets written to the divert socket carry in sin_addr a
98  * destination address, and in sin_port the number of the filter rule
99  * after which to continue processing.
100  * If the destination address is INADDR_ANY, the packet is treated as
101  * as outgoing and sent to ip_output(); otherwise it is treated as
102  * incoming and sent to ip_input().
103  * Further, sin_zero carries some information on the interface,
104  * which can be used in the reinject -- see comments in the code.
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  * packet filter processing will start at the rule number after the one
109  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
110  * will apply the entire ruleset to the packet).
111  */
112 
113 /* Internal variables. */
114 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo);
115 #define	V_divcbinfo			VNET(divcbinfo)
116 
117 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
118 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
119 
120 static eventhandler_tag ip_divert_event_tag;
121 
122 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
123     struct sockaddr_in *sin);
124 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
125 
126 /*
127  * Initialize divert connection block queue.
128  */
129 static void
130 div_zone_change(void *tag)
131 {
132 
133 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
134 }
135 
136 static int
137 div_inpcb_init(void *mem, int size, int flags)
138 {
139 	struct inpcb *inp = mem;
140 
141 	INP_LOCK_INIT(inp, "inp", "divinp");
142 	return (0);
143 }
144 
145 static void
146 div_init(void)
147 {
148 
149 	/*
150 	 * XXX We don't use the hash list for divert IP, but it's easier to
151 	 * allocate one-entry hash lists than it is to check all over the
152 	 * place for hashbase == NULL.
153 	 */
154 	in_pcbinfo_init(&V_divcbinfo, "div", 1, 1, "divcb", div_inpcb_init);
155 }
156 
157 static void
158 div_destroy(void *unused __unused)
159 {
160 
161 	in_pcbinfo_destroy(&V_divcbinfo);
162 }
163 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
164     div_destroy, NULL);
165 
166 /*
167  * IPPROTO_DIVERT is not in the real IP protocol number space; this
168  * function should never be called.  Just in case, drop any packets.
169  */
170 static int
171 div_input(struct mbuf **mp, int *offp, int proto)
172 {
173 	struct mbuf *m = *mp;
174 
175 	KMOD_IPSTAT_INC(ips_noproto);
176 	m_freem(m);
177 	return (IPPROTO_DONE);
178 }
179 
180 static bool
181 div_port_match(const struct inpcb *inp, void *v)
182 {
183 	uint16_t nport = *(uint16_t *)v;
184 
185 	return (inp->inp_lport == nport);
186 }
187 
188 /*
189  * Divert a packet by passing it up to the divert socket at port 'port'.
190  *
191  * Setup generic address and protocol structures for div_input routine,
192  * then pass them along with mbuf chain.
193  */
194 static void
195 divert_packet(struct mbuf *m, bool incoming)
196 {
197 	struct ip *ip;
198 	struct inpcb *inp;
199 	struct socket *sa;
200 	u_int16_t nport;
201 	struct sockaddr_in divsrc;
202 	struct inpcb_iterator inpi = INP_ITERATOR(&V_divcbinfo,
203 	    INPLOOKUP_RLOCKPCB, div_port_match, &nport);
204 	struct m_tag *mtag;
205 
206 	NET_EPOCH_ASSERT();
207 
208 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
209 	if (mtag == NULL) {
210 		m_freem(m);
211 		return;
212 	}
213 	/* Assure header */
214 	if (m->m_len < sizeof(struct ip) &&
215 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
216 		return;
217 	ip = mtod(m, struct ip *);
218 
219 	/* Delayed checksums are currently not compatible with divert. */
220 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
221 		in_delayed_cksum(m);
222 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
223 	}
224 #if defined(SCTP) || defined(SCTP_SUPPORT)
225 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
226 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
227 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
228 	}
229 #endif
230 #ifdef INET6
231 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
232 		in6_delayed_cksum(m, m->m_pkthdr.len -
233 		    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
234 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
235 	}
236 #if defined(SCTP) || defined(SCTP_SUPPORT)
237 	if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
238 		sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
239 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
240 	}
241 #endif
242 #endif /* INET6 */
243 	bzero(&divsrc, sizeof(divsrc));
244 	divsrc.sin_len = sizeof(divsrc);
245 	divsrc.sin_family = AF_INET;
246 	/* record matching rule, in host format */
247 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
248 	/*
249 	 * Record receive interface address, if any.
250 	 * But only for incoming packets.
251 	 */
252 	if (incoming) {
253 		struct ifaddr *ifa;
254 		struct ifnet *ifp;
255 
256 		/* Sanity check */
257 		M_ASSERTPKTHDR(m);
258 
259 		/* Find IP address for receive interface */
260 		ifp = m->m_pkthdr.rcvif;
261 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
262 			if (ifa->ifa_addr->sa_family != AF_INET)
263 				continue;
264 			divsrc.sin_addr =
265 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
266 			break;
267 		}
268 	}
269 	/*
270 	 * Record the incoming interface name whenever we have one.
271 	 */
272 	if (m->m_pkthdr.rcvif) {
273 		/*
274 		 * Hide the actual interface name in there in the
275 		 * sin_zero array. XXX This needs to be moved to a
276 		 * different sockaddr type for divert, e.g.
277 		 * sockaddr_div with multiple fields like
278 		 * sockaddr_dl. Presently we have only 7 bytes
279 		 * but that will do for now as most interfaces
280 		 * are 4 or less + 2 or less bytes for unit.
281 		 * There is probably a faster way of doing this,
282 		 * possibly taking it from the sockaddr_dl on the iface.
283 		 * This solves the problem of a P2P link and a LAN interface
284 		 * having the same address, which can result in the wrong
285 		 * interface being assigned to the packet when fed back
286 		 * into the divert socket. Theoretically if the daemon saves
287 		 * and re-uses the sockaddr_in as suggested in the man pages,
288 		 * this iface name will come along for the ride.
289 		 * (see div_output for the other half of this.)
290 		 */
291 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
292 		    sizeof(divsrc.sin_zero));
293 	}
294 
295 	/* Put packet on socket queue, if any */
296 	sa = NULL;
297 	/* nport is inp_next's context. */
298 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
299 	while ((inp = inp_next(&inpi)) != NULL) {
300 		sa = inp->inp_socket;
301 		SOCKBUF_LOCK(&sa->so_rcv);
302 		if (sbappendaddr_locked(&sa->so_rcv,
303 		    (struct sockaddr *)&divsrc, m, NULL) == 0) {
304 			soroverflow_locked(sa);
305 			sa = NULL;	/* force mbuf reclaim below */
306 		} else
307 			sorwakeup_locked(sa);
308 		/* XXX why does only one socket match? */
309 		INP_RUNLOCK(inp);
310 		break;
311 	}
312 	if (sa == NULL) {
313 		m_freem(m);
314 		KMOD_IPSTAT_INC(ips_noproto);
315 		KMOD_IPSTAT_DEC(ips_delivered);
316         }
317 }
318 
319 /*
320  * Deliver packet back into the IP processing machinery.
321  *
322  * If no address specified, or address is 0.0.0.0, send to ip_output();
323  * otherwise, send to ip_input() and mark as having been received on
324  * the interface with that address.
325  */
326 static int
327 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
328     struct mbuf *control)
329 {
330 	struct epoch_tracker et;
331 	const struct ip *ip;
332 	struct m_tag *mtag;
333 	struct ipfw_rule_ref *dt;
334 	int error, family;
335 
336 	if (control) {
337 		m_freem(control);		/* XXX */
338 		control = NULL;
339 	}
340 
341 	if (sin != NULL) {
342 		if (sin->sin_family != AF_INET) {
343 			m_freem(m);
344 			return (EAFNOSUPPORT);
345 		}
346 		if (sin->sin_len != sizeof(*sin)) {
347 			m_freem(m);
348 			return (EINVAL);
349 		}
350 	}
351 
352 	/*
353 	 * An mbuf may hasn't come from userland, but we pretend
354 	 * that it has.
355 	 */
356 	m->m_pkthdr.rcvif = NULL;
357 	m->m_nextpkt = NULL;
358 	M_SETFIB(m, so->so_fibnum);
359 
360 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
361 	if (mtag == NULL) {
362 		/* this should be normal */
363 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
364 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
365 		if (mtag == NULL) {
366 			m_freem(m);
367 			return (ENOBUFS);
368 		}
369 		m_tag_prepend(m, mtag);
370 	}
371 	dt = (struct ipfw_rule_ref *)(mtag+1);
372 
373 	/* Loopback avoidance and state recovery */
374 	if (sin) {
375 		int i;
376 
377 		/* set the starting point. We provide a non-zero slot,
378 		 * but a non_matching chain_id to skip that info and use
379 		 * the rulenum/rule_id.
380 		 */
381 		dt->slot = 1; /* dummy, chain_id is invalid */
382 		dt->chain_id = 0;
383 		dt->rulenum = sin->sin_port+1; /* host format ? */
384 		dt->rule_id = 0;
385 		/* XXX: broken for IPv6 */
386 		/*
387 		 * Find receive interface with the given name, stuffed
388 		 * (if it exists) in the sin_zero[] field.
389 		 * The name is user supplied data so don't trust its size
390 		 * or that it is zero terminated.
391 		 */
392 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
393 			;
394 		if ( i > 0 && i < sizeof(sin->sin_zero))
395 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
396 	}
397 
398 	ip = mtod(m, struct ip *);
399 	switch (ip->ip_v) {
400 	case IPVERSION:
401 		family = AF_INET;
402 		break;
403 #ifdef INET6
404 	case IPV6_VERSION >> 4:
405 		family = AF_INET6;
406 		break;
407 #endif
408 	default:
409 		m_freem(m);
410 		return (EAFNOSUPPORT);
411 	}
412 
413 	/* Reinject packet into the system as incoming or outgoing */
414 	NET_EPOCH_ENTER(et);
415 	if (!sin || sin->sin_addr.s_addr == 0) {
416 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
417 		error = div_output_outbound(family, so, m);
418 	} else {
419 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
420 		error = div_output_inbound(family, so, m, sin);
421 	}
422 	NET_EPOCH_EXIT(et);
423 
424 	return (error);
425 }
426 
427 /*
428  * Sends mbuf @m to the wire via ip[6]_output().
429  *
430  * Returns 0 on success or an errno value on failure.  @m is always consumed.
431  */
432 static int
433 div_output_outbound(int family, struct socket *so, struct mbuf *m)
434 {
435 	struct ip *const ip = mtod(m, struct ip *);
436 	struct mbuf *options;
437 	struct inpcb *inp;
438 	int error;
439 
440 	inp = sotoinpcb(so);
441 	INP_RLOCK(inp);
442 	switch (family) {
443 	case AF_INET:
444 		/*
445 		 * Don't allow both user specified and setsockopt
446 		 * options, and don't allow packet length sizes that
447 		 * will crash.
448 		 */
449 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
450 		    inp->inp_options != NULL) ||
451 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
452 			INP_RUNLOCK(inp);
453 			m_freem(m);
454 			return (EINVAL);
455 		}
456 		break;
457 #ifdef INET6
458 	case AF_INET6:
459 	    {
460 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
461 
462 		/* Don't allow packet length sizes that will crash */
463 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
464 			INP_RUNLOCK(inp);
465 			m_freem(m);
466 			return (EINVAL);
467 		}
468 		break;
469 	    }
470 #endif
471 	}
472 
473 	/* Send packet to output processing */
474 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
475 
476 #ifdef MAC
477 	mac_inpcb_create_mbuf(inp, m);
478 #endif
479 	/*
480 	 * Get ready to inject the packet into ip_output().
481 	 * Just in case socket options were specified on the
482 	 * divert socket, we duplicate them.  This is done
483 	 * to avoid having to hold the PCB locks over the call
484 	 * to ip_output(), as doing this results in a number of
485 	 * lock ordering complexities.
486 	 *
487 	 * Note that we set the multicast options argument for
488 	 * ip_output() to NULL since it should be invariant that
489 	 * they are not present.
490 	 */
491 	KASSERT(inp->inp_moptions == NULL,
492 	    ("multicast options set on a divert socket"));
493 	/*
494 	 * XXXCSJP: It is unclear to me whether or not it makes
495 	 * sense for divert sockets to have options.  However,
496 	 * for now we will duplicate them with the INP locks
497 	 * held so we can use them in ip_output() without
498 	 * requring a reference to the pcb.
499 	 */
500 	options = NULL;
501 	if (inp->inp_options != NULL) {
502 		options = m_dup(inp->inp_options, M_NOWAIT);
503 		if (options == NULL) {
504 			INP_RUNLOCK(inp);
505 			m_freem(m);
506 			return (ENOBUFS);
507 		}
508 	}
509 	INP_RUNLOCK(inp);
510 
511 	error = 0;
512 	switch (family) {
513 	case AF_INET:
514 		error = ip_output(m, options, NULL,
515 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
516 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
517 		break;
518 #ifdef INET6
519 	case AF_INET6:
520 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
521 		break;
522 #endif
523 	}
524 	if (options != NULL)
525 		m_freem(options);
526 
527 	return (error);
528 }
529 
530 /*
531  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
532  *
533  * Returns 0 on success or an errno value on failure.  @m is always consumed.
534  */
535 static int
536 div_output_inbound(int family, struct socket *so, struct mbuf *m,
537     struct sockaddr_in *sin)
538 {
539 	const struct ip *ip;
540 	struct ifaddr *ifa;
541 
542 	if (m->m_pkthdr.rcvif == NULL) {
543 		/*
544 		 * No luck with the name, check by IP address.
545 		 * Clear the port and the ifname to make sure
546 		 * there are no distractions for ifa_ifwithaddr.
547 		 */
548 
549 		/* XXX: broken for IPv6 */
550 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
551 		sin->sin_port = 0;
552 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
553 		if (ifa == NULL) {
554 			m_freem(m);
555 			return (EADDRNOTAVAIL);
556 		}
557 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
558 	}
559 #ifdef MAC
560 	mac_socket_create_mbuf(so, m);
561 #endif
562 	/* Send packet to input processing via netisr */
563 	switch (family) {
564 	case AF_INET:
565 		ip = mtod(m, struct ip *);
566 		/*
567 		 * Restore M_BCAST flag when destination address is
568 		 * broadcast. It is expected by ip_tryforward().
569 		 */
570 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
571 			m->m_flags |= M_MCAST;
572 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
573 			m->m_flags |= M_BCAST;
574 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
575 		break;
576 #ifdef INET6
577 	case AF_INET6:
578 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
579 		break;
580 #endif
581 	default:
582 		m_freem(m);
583 		return (EINVAL);
584 	}
585 
586 	return (0);
587 }
588 
589 static int
590 div_attach(struct socket *so, int proto, struct thread *td)
591 {
592 	struct inpcb *inp;
593 	int error;
594 
595 	inp  = sotoinpcb(so);
596 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
597 	if (td != NULL) {
598 		error = priv_check(td, PRIV_NETINET_DIVERT);
599 		if (error)
600 			return (error);
601 	}
602 	error = soreserve(so, div_sendspace, div_recvspace);
603 	if (error)
604 		return error;
605 	error = in_pcballoc(so, &V_divcbinfo);
606 	if (error)
607 		return error;
608 	inp = (struct inpcb *)so->so_pcb;
609 	inp->inp_ip_p = proto;
610 	inp->inp_vflag |= INP_IPV4;
611 	inp->inp_flags |= INP_HDRINCL;
612 	INP_WUNLOCK(inp);
613 	return 0;
614 }
615 
616 static void
617 div_detach(struct socket *so)
618 {
619 	struct inpcb *inp;
620 
621 	inp = sotoinpcb(so);
622 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
623 	INP_WLOCK(inp);
624 	in_pcbdetach(inp);
625 	in_pcbfree(inp);
626 }
627 
628 static int
629 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
630 {
631 	struct inpcb *inp;
632 	int error;
633 
634 	inp = sotoinpcb(so);
635 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
636 	/* in_pcbbind assumes that nam is a sockaddr_in
637 	 * and in_pcbbind requires a valid address. Since divert
638 	 * sockets don't we need to make sure the address is
639 	 * filled in properly.
640 	 * XXX -- divert should not be abusing in_pcbind
641 	 * and should probably have its own family.
642 	 */
643 	if (nam->sa_family != AF_INET)
644 		return EAFNOSUPPORT;
645 	if (nam->sa_len != sizeof(struct sockaddr_in))
646 		return EINVAL;
647 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
648 	INP_WLOCK(inp);
649 	INP_HASH_WLOCK(&V_divcbinfo);
650 	error = in_pcbbind(inp, nam, td->td_ucred);
651 	INP_HASH_WUNLOCK(&V_divcbinfo);
652 	INP_WUNLOCK(inp);
653 	return error;
654 }
655 
656 static int
657 div_shutdown(struct socket *so)
658 {
659 	struct inpcb *inp;
660 
661 	inp = sotoinpcb(so);
662 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
663 	INP_WLOCK(inp);
664 	socantsendmore(so);
665 	INP_WUNLOCK(inp);
666 	return 0;
667 }
668 
669 static int
670 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
671     struct mbuf *control, struct thread *td)
672 {
673 
674 	/* Packet must have a header (but that's about it) */
675 	if (m->m_len < sizeof (struct ip) &&
676 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
677 		KMOD_IPSTAT_INC(ips_toosmall);
678 		if (control != NULL)
679 			m_freem(control);
680 		m_freem(m);
681 		return EINVAL;
682 	}
683 
684 	/* Send packet */
685 	return div_output(so, m, (struct sockaddr_in *)nam, control);
686 }
687 
688 static int
689 div_pcblist(SYSCTL_HANDLER_ARGS)
690 {
691 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_divcbinfo,
692 	    INPLOOKUP_RLOCKPCB);
693 	struct xinpgen xig;
694 	struct inpcb *inp;
695 	int error;
696 
697 	if (req->newptr != 0)
698 		return EPERM;
699 
700 	if (req->oldptr == 0) {
701 		int n;
702 
703 		n = V_divcbinfo.ipi_count;
704 		n += imax(n / 8, 10);
705 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
706 		return 0;
707 	}
708 
709 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
710 		return (error);
711 
712 	bzero(&xig, sizeof(xig));
713 	xig.xig_len = sizeof xig;
714 	xig.xig_count = V_divcbinfo.ipi_count;
715 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
716 	xig.xig_sogen = so_gencnt;
717 	error = SYSCTL_OUT(req, &xig, sizeof xig);
718 	if (error)
719 		return error;
720 
721 	while ((inp = inp_next(&inpi)) != NULL) {
722 		if (inp->inp_gencnt <= xig.xig_gen) {
723 			struct xinpcb xi;
724 
725 			in_pcbtoxinpcb(inp, &xi);
726 			error = SYSCTL_OUT(req, &xi, sizeof xi);
727 			if (error) {
728 				INP_RUNLOCK(inp);
729 				break;
730 			}
731 		}
732 	}
733 
734 	if (!error) {
735 		/*
736 		 * Give the user an updated idea of our state.
737 		 * If the generation differs from what we told
738 		 * her before, she knows that something happened
739 		 * while we were processing this request, and it
740 		 * might be necessary to retry.
741 		 */
742 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
743 		xig.xig_sogen = so_gencnt;
744 		xig.xig_count = V_divcbinfo.ipi_count;
745 		error = SYSCTL_OUT(req, &xig, sizeof xig);
746 	}
747 
748 	return (error);
749 }
750 
751 #ifdef SYSCTL_NODE
752 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
753     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
754     "IPDIVERT");
755 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
756    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
757     NULL, 0, div_pcblist, "S,xinpcb",
758     "List of active divert sockets");
759 #endif
760 
761 struct pr_usrreqs div_usrreqs = {
762 	.pru_attach =		div_attach,
763 	.pru_bind =		div_bind,
764 	.pru_control =		in_control,
765 	.pru_detach =		div_detach,
766 	.pru_peeraddr =		in_getpeeraddr,
767 	.pru_send =		div_send,
768 	.pru_shutdown =		div_shutdown,
769 	.pru_sockaddr =		in_getsockaddr,
770 	.pru_sosetlabel =	in_pcbsosetlabel
771 };
772 
773 struct protosw div_protosw = {
774 	.pr_type =		SOCK_RAW,
775 	.pr_protocol =		IPPROTO_DIVERT,
776 	.pr_flags =		PR_ATOMIC|PR_ADDR,
777 	.pr_input =		div_input,
778 	.pr_init =		div_init,
779 	.pr_usrreqs =		&div_usrreqs
780 };
781 
782 static int
783 div_modevent(module_t mod, int type, void *unused)
784 {
785 	int err = 0;
786 
787 	switch (type) {
788 	case MOD_LOAD:
789 		/*
790 		 * Protocol will be initialized by pf_proto_register().
791 		 * We don't have to register ip_protox because we are not
792 		 * a true IP protocol that goes over the wire.
793 		 */
794 		err = pf_proto_register(PF_INET, &div_protosw);
795 		if (err != 0)
796 			return (err);
797 		ip_divert_ptr = divert_packet;
798 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
799 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
800 		break;
801 	case MOD_QUIESCE:
802 		/*
803 		 * IPDIVERT may normally not be unloaded because of the
804 		 * potential race conditions.  Tell kldunload we can't be
805 		 * unloaded unless the unload is forced.
806 		 */
807 		err = EPERM;
808 		break;
809 	case MOD_UNLOAD:
810 		/*
811 		 * Forced unload.
812 		 *
813 		 * Module ipdivert can only be unloaded if no sockets are
814 		 * connected.  Maybe this can be changed later to forcefully
815 		 * disconnect any open sockets.
816 		 *
817 		 * XXXRW: Note that there is a slight race here, as a new
818 		 * socket open request could be spinning on the lock and then
819 		 * we destroy the lock.
820 		 */
821 		INP_INFO_WLOCK(&V_divcbinfo);
822 		if (V_divcbinfo.ipi_count != 0) {
823 			err = EBUSY;
824 			INP_INFO_WUNLOCK(&V_divcbinfo);
825 			break;
826 		}
827 		ip_divert_ptr = NULL;
828 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
829 		INP_INFO_WUNLOCK(&V_divcbinfo);
830 #ifndef VIMAGE
831 		div_destroy(NULL);
832 #endif
833 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
834 		break;
835 	default:
836 		err = EOPNOTSUPP;
837 		break;
838 	}
839 	return err;
840 }
841 
842 static moduledata_t ipdivertmod = {
843         "ipdivert",
844         div_modevent,
845         0
846 };
847 
848 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
849 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
850 MODULE_VERSION(ipdivert, 1);
851