xref: /freebsd/sys/netinet/ip_divert.c (revision 5def4c47d4bd90b209b9b4a4ba9faec15846d8fd)
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 inpcbhead, divcb);
115 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo);
116 
117 #define	V_divcb				VNET(divcb)
118 #define	V_divcbinfo			VNET(divcbinfo)
119 
120 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
121 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
122 
123 static eventhandler_tag ip_divert_event_tag;
124 
125 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
126     struct sockaddr_in *sin);
127 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
128 
129 /*
130  * Initialize divert connection block queue.
131  */
132 static void
133 div_zone_change(void *tag)
134 {
135 
136 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
137 }
138 
139 static int
140 div_inpcb_init(void *mem, int size, int flags)
141 {
142 	struct inpcb *inp = mem;
143 
144 	INP_LOCK_INIT(inp, "inp", "divinp");
145 	return (0);
146 }
147 
148 static void
149 div_init(void)
150 {
151 
152 	/*
153 	 * XXX We don't use the hash list for divert IP, but it's easier to
154 	 * allocate one-entry hash lists than it is to check all over the
155 	 * place for hashbase == NULL.
156 	 */
157 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
158 	    div_inpcb_init, IPI_HASHFIELDS_NONE);
159 }
160 
161 static void
162 div_destroy(void *unused __unused)
163 {
164 
165 	in_pcbinfo_destroy(&V_divcbinfo);
166 }
167 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
168     div_destroy, NULL);
169 
170 /*
171  * IPPROTO_DIVERT is not in the real IP protocol number space; this
172  * function should never be called.  Just in case, drop any packets.
173  */
174 static int
175 div_input(struct mbuf **mp, int *offp, int proto)
176 {
177 	struct mbuf *m = *mp;
178 
179 	KMOD_IPSTAT_INC(ips_noproto);
180 	m_freem(m);
181 	return (IPPROTO_DONE);
182 }
183 
184 /*
185  * Divert a packet by passing it up to the divert socket at port 'port'.
186  *
187  * Setup generic address and protocol structures for div_input routine,
188  * then pass them along with mbuf chain.
189  */
190 static void
191 divert_packet(struct mbuf *m, bool incoming)
192 {
193 	struct ip *ip;
194 	struct inpcb *inp;
195 	struct socket *sa;
196 	u_int16_t nport;
197 	struct sockaddr_in divsrc;
198 	struct m_tag *mtag;
199 
200 	NET_EPOCH_ASSERT();
201 
202 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
203 	if (mtag == NULL) {
204 		m_freem(m);
205 		return;
206 	}
207 	/* Assure header */
208 	if (m->m_len < sizeof(struct ip) &&
209 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
210 		return;
211 	ip = mtod(m, struct ip *);
212 
213 	/* Delayed checksums are currently not compatible with divert. */
214 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
215 		m = mb_unmapped_to_ext(m);
216 		if (m == NULL)
217 			return;
218 		in_delayed_cksum(m);
219 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
220 	}
221 #if defined(SCTP) || defined(SCTP_SUPPORT)
222 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
223 		m = mb_unmapped_to_ext(m);
224 		if (m == NULL)
225 			return;
226 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
227 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
228 	}
229 #endif
230 	bzero(&divsrc, sizeof(divsrc));
231 	divsrc.sin_len = sizeof(divsrc);
232 	divsrc.sin_family = AF_INET;
233 	/* record matching rule, in host format */
234 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
235 	/*
236 	 * Record receive interface address, if any.
237 	 * But only for incoming packets.
238 	 */
239 	if (incoming) {
240 		struct ifaddr *ifa;
241 		struct ifnet *ifp;
242 
243 		/* Sanity check */
244 		M_ASSERTPKTHDR(m);
245 
246 		/* Find IP address for receive interface */
247 		ifp = m->m_pkthdr.rcvif;
248 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
249 			if (ifa->ifa_addr->sa_family != AF_INET)
250 				continue;
251 			divsrc.sin_addr =
252 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
253 			break;
254 		}
255 	}
256 	/*
257 	 * Record the incoming interface name whenever we have one.
258 	 */
259 	if (m->m_pkthdr.rcvif) {
260 		/*
261 		 * Hide the actual interface name in there in the
262 		 * sin_zero array. XXX This needs to be moved to a
263 		 * different sockaddr type for divert, e.g.
264 		 * sockaddr_div with multiple fields like
265 		 * sockaddr_dl. Presently we have only 7 bytes
266 		 * but that will do for now as most interfaces
267 		 * are 4 or less + 2 or less bytes for unit.
268 		 * There is probably a faster way of doing this,
269 		 * possibly taking it from the sockaddr_dl on the iface.
270 		 * This solves the problem of a P2P link and a LAN interface
271 		 * having the same address, which can result in the wrong
272 		 * interface being assigned to the packet when fed back
273 		 * into the divert socket. Theoretically if the daemon saves
274 		 * and re-uses the sockaddr_in as suggested in the man pages,
275 		 * this iface name will come along for the ride.
276 		 * (see div_output for the other half of this.)
277 		 */
278 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
279 		    sizeof(divsrc.sin_zero));
280 	}
281 
282 	/* Put packet on socket queue, if any */
283 	sa = NULL;
284 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
285 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
286 		/* XXX why does only one socket match? */
287 		if (inp->inp_lport == nport) {
288 			INP_RLOCK(inp);
289 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
290 				INP_RUNLOCK(inp);
291 				continue;
292 			}
293 			sa = inp->inp_socket;
294 			SOCKBUF_LOCK(&sa->so_rcv);
295 			if (sbappendaddr_locked(&sa->so_rcv,
296 			    (struct sockaddr *)&divsrc, m,
297 			    (struct mbuf *)0) == 0) {
298 				SOCKBUF_UNLOCK(&sa->so_rcv);
299 				sa = NULL;	/* force mbuf reclaim below */
300 			} else
301 				sorwakeup_locked(sa);
302 			INP_RUNLOCK(inp);
303 			break;
304 		}
305 	}
306 	if (sa == NULL) {
307 		m_freem(m);
308 		KMOD_IPSTAT_INC(ips_noproto);
309 		KMOD_IPSTAT_DEC(ips_delivered);
310         }
311 }
312 
313 /*
314  * Deliver packet back into the IP processing machinery.
315  *
316  * If no address specified, or address is 0.0.0.0, send to ip_output();
317  * otherwise, send to ip_input() and mark as having been received on
318  * the interface with that address.
319  */
320 static int
321 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
322     struct mbuf *control)
323 {
324 	struct epoch_tracker et;
325 	const struct ip *ip;
326 	struct m_tag *mtag;
327 	struct ipfw_rule_ref *dt;
328 	int error, family;
329 
330 	if (control) {
331 		m_freem(control);		/* XXX */
332 		control = NULL;
333 	}
334 
335 	if (sin != NULL) {
336 		if (sin->sin_family != AF_INET) {
337 			m_freem(m);
338 			return (EAFNOSUPPORT);
339 		}
340 		if (sin->sin_len != sizeof(*sin)) {
341 			m_freem(m);
342 			return (EINVAL);
343 		}
344 	}
345 
346 	/*
347 	 * An mbuf may hasn't come from userland, but we pretend
348 	 * that it has.
349 	 */
350 	m->m_pkthdr.rcvif = NULL;
351 	m->m_nextpkt = NULL;
352 	M_SETFIB(m, so->so_fibnum);
353 
354 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
355 	if (mtag == NULL) {
356 		/* this should be normal */
357 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
358 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
359 		if (mtag == NULL) {
360 			m_freem(m);
361 			return (ENOBUFS);
362 		}
363 		m_tag_prepend(m, mtag);
364 	}
365 	dt = (struct ipfw_rule_ref *)(mtag+1);
366 
367 	/* Loopback avoidance and state recovery */
368 	if (sin) {
369 		int i;
370 
371 		/* set the starting point. We provide a non-zero slot,
372 		 * but a non_matching chain_id to skip that info and use
373 		 * the rulenum/rule_id.
374 		 */
375 		dt->slot = 1; /* dummy, chain_id is invalid */
376 		dt->chain_id = 0;
377 		dt->rulenum = sin->sin_port+1; /* host format ? */
378 		dt->rule_id = 0;
379 		/* XXX: broken for IPv6 */
380 		/*
381 		 * Find receive interface with the given name, stuffed
382 		 * (if it exists) in the sin_zero[] field.
383 		 * The name is user supplied data so don't trust its size
384 		 * or that it is zero terminated.
385 		 */
386 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
387 			;
388 		if ( i > 0 && i < sizeof(sin->sin_zero))
389 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
390 	}
391 
392 	ip = mtod(m, struct ip *);
393 	switch (ip->ip_v) {
394 	case IPVERSION:
395 		family = AF_INET;
396 		break;
397 #ifdef INET6
398 	case IPV6_VERSION >> 4:
399 		family = AF_INET6;
400 		break;
401 #endif
402 	default:
403 		m_freem(m);
404 		return (EAFNOSUPPORT);
405 	}
406 
407 	/* Reinject packet into the system as incoming or outgoing */
408 	NET_EPOCH_ENTER(et);
409 	if (!sin || sin->sin_addr.s_addr == 0) {
410 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
411 		error = div_output_outbound(family, so, m);
412 	} else {
413 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
414 		error = div_output_inbound(family, so, m, sin);
415 	}
416 	NET_EPOCH_EXIT(et);
417 
418 	return (error);
419 }
420 
421 /*
422  * Sends mbuf @m to the wire via ip[6]_output().
423  *
424  * Returns 0 on success or an errno value on failure.  @m is always consumed.
425  */
426 static int
427 div_output_outbound(int family, struct socket *so, struct mbuf *m)
428 {
429 	struct ip *const ip = mtod(m, struct ip *);
430 	struct mbuf *options;
431 	struct inpcb *inp;
432 	int error;
433 
434 	inp = sotoinpcb(so);
435 	INP_RLOCK(inp);
436 	switch (family) {
437 	case AF_INET:
438 		/*
439 		 * Don't allow both user specified and setsockopt
440 		 * options, and don't allow packet length sizes that
441 		 * will crash.
442 		 */
443 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
444 		    inp->inp_options != NULL) ||
445 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
446 			INP_RUNLOCK(inp);
447 			m_freem(m);
448 			return (EINVAL);
449 		}
450 		break;
451 #ifdef INET6
452 	case AF_INET6:
453 	    {
454 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
455 
456 		/* Don't allow packet length sizes that will crash */
457 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
458 			INP_RUNLOCK(inp);
459 			m_freem(m);
460 			return (EINVAL);
461 		}
462 		break;
463 	    }
464 #endif
465 	}
466 
467 	/* Send packet to output processing */
468 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
469 
470 #ifdef MAC
471 	mac_inpcb_create_mbuf(inp, m);
472 #endif
473 	/*
474 	 * Get ready to inject the packet into ip_output().
475 	 * Just in case socket options were specified on the
476 	 * divert socket, we duplicate them.  This is done
477 	 * to avoid having to hold the PCB locks over the call
478 	 * to ip_output(), as doing this results in a number of
479 	 * lock ordering complexities.
480 	 *
481 	 * Note that we set the multicast options argument for
482 	 * ip_output() to NULL since it should be invariant that
483 	 * they are not present.
484 	 */
485 	KASSERT(inp->inp_moptions == NULL,
486 	    ("multicast options set on a divert socket"));
487 	/*
488 	 * XXXCSJP: It is unclear to me whether or not it makes
489 	 * sense for divert sockets to have options.  However,
490 	 * for now we will duplicate them with the INP locks
491 	 * held so we can use them in ip_output() without
492 	 * requring a reference to the pcb.
493 	 */
494 	options = NULL;
495 	if (inp->inp_options != NULL) {
496 		options = m_dup(inp->inp_options, M_NOWAIT);
497 		if (options == NULL) {
498 			INP_RUNLOCK(inp);
499 			m_freem(m);
500 			return (ENOBUFS);
501 		}
502 	}
503 	INP_RUNLOCK(inp);
504 
505 	error = 0;
506 	switch (family) {
507 	case AF_INET:
508 		error = ip_output(m, options, NULL,
509 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
510 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
511 		break;
512 #ifdef INET6
513 	case AF_INET6:
514 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
515 		break;
516 #endif
517 	}
518 	if (options != NULL)
519 		m_freem(options);
520 
521 	return (error);
522 }
523 
524 /*
525  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
526  *
527  * Returns 0 on success or an errno value on failure.  @m is always consumed.
528  */
529 static int
530 div_output_inbound(int family, struct socket *so, struct mbuf *m,
531     struct sockaddr_in *sin)
532 {
533 	const struct ip *ip;
534 	struct ifaddr *ifa;
535 
536 	if (m->m_pkthdr.rcvif == NULL) {
537 		/*
538 		 * No luck with the name, check by IP address.
539 		 * Clear the port and the ifname to make sure
540 		 * there are no distractions for ifa_ifwithaddr.
541 		 */
542 
543 		/* XXX: broken for IPv6 */
544 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
545 		sin->sin_port = 0;
546 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
547 		if (ifa == NULL) {
548 			m_freem(m);
549 			return (EADDRNOTAVAIL);
550 		}
551 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
552 	}
553 #ifdef MAC
554 	mac_socket_create_mbuf(so, m);
555 #endif
556 	/* Send packet to input processing via netisr */
557 	switch (family) {
558 	case AF_INET:
559 		ip = mtod(m, struct ip *);
560 		/*
561 		 * Restore M_BCAST flag when destination address is
562 		 * broadcast. It is expected by ip_tryforward().
563 		 */
564 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
565 			m->m_flags |= M_MCAST;
566 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
567 			m->m_flags |= M_BCAST;
568 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
569 		break;
570 #ifdef INET6
571 	case AF_INET6:
572 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
573 		break;
574 #endif
575 	default:
576 		m_freem(m);
577 		return (EINVAL);
578 	}
579 
580 	return (0);
581 }
582 
583 static int
584 div_attach(struct socket *so, int proto, struct thread *td)
585 {
586 	struct inpcb *inp;
587 	int error;
588 
589 	inp  = sotoinpcb(so);
590 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
591 	if (td != NULL) {
592 		error = priv_check(td, PRIV_NETINET_DIVERT);
593 		if (error)
594 			return (error);
595 	}
596 	error = soreserve(so, div_sendspace, div_recvspace);
597 	if (error)
598 		return error;
599 	INP_INFO_WLOCK(&V_divcbinfo);
600 	error = in_pcballoc(so, &V_divcbinfo);
601 	if (error) {
602 		INP_INFO_WUNLOCK(&V_divcbinfo);
603 		return error;
604 	}
605 	inp = (struct inpcb *)so->so_pcb;
606 	INP_INFO_WUNLOCK(&V_divcbinfo);
607 	inp->inp_ip_p = proto;
608 	inp->inp_vflag |= INP_IPV4;
609 	inp->inp_flags |= INP_HDRINCL;
610 	INP_WUNLOCK(inp);
611 	return 0;
612 }
613 
614 static void
615 div_detach(struct socket *so)
616 {
617 	struct inpcb *inp;
618 
619 	inp = sotoinpcb(so);
620 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
621 	INP_INFO_WLOCK(&V_divcbinfo);
622 	INP_WLOCK(inp);
623 	in_pcbdetach(inp);
624 	in_pcbfree(inp);
625 	INP_INFO_WUNLOCK(&V_divcbinfo);
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_INFO_WLOCK(&V_divcbinfo);
649 	INP_WLOCK(inp);
650 	INP_HASH_WLOCK(&V_divcbinfo);
651 	error = in_pcbbind(inp, nam, td->td_ucred);
652 	INP_HASH_WUNLOCK(&V_divcbinfo);
653 	INP_WUNLOCK(inp);
654 	INP_INFO_WUNLOCK(&V_divcbinfo);
655 	return error;
656 }
657 
658 static int
659 div_shutdown(struct socket *so)
660 {
661 	struct inpcb *inp;
662 
663 	inp = sotoinpcb(so);
664 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
665 	INP_WLOCK(inp);
666 	socantsendmore(so);
667 	INP_WUNLOCK(inp);
668 	return 0;
669 }
670 
671 static int
672 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
673     struct mbuf *control, struct thread *td)
674 {
675 
676 	/* Packet must have a header (but that's about it) */
677 	if (m->m_len < sizeof (struct ip) &&
678 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
679 		KMOD_IPSTAT_INC(ips_toosmall);
680 		if (control != NULL)
681 			m_freem(control);
682 		m_freem(m);
683 		return EINVAL;
684 	}
685 
686 	/* Send packet */
687 	return div_output(so, m, (struct sockaddr_in *)nam, control);
688 }
689 
690 static void
691 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
692 {
693         struct in_addr faddr;
694 
695 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
696 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
697         	return;
698 	if (PRC_IS_REDIRECT(cmd))
699 		return;
700 }
701 
702 static int
703 div_pcblist(SYSCTL_HANDLER_ARGS)
704 {
705 	struct xinpgen xig;
706 	struct epoch_tracker et;
707 	struct inpcb *inp;
708 	int error;
709 
710 	if (req->newptr != 0)
711 		return EPERM;
712 
713 	if (req->oldptr == 0) {
714 		int n;
715 
716 		n = V_divcbinfo.ipi_count;
717 		n += imax(n / 8, 10);
718 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
719 		return 0;
720 	}
721 
722 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
723 		return (error);
724 
725 	bzero(&xig, sizeof(xig));
726 	xig.xig_len = sizeof xig;
727 	xig.xig_count = V_divcbinfo.ipi_count;
728 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
729 	xig.xig_sogen = so_gencnt;
730 	error = SYSCTL_OUT(req, &xig, sizeof xig);
731 	if (error)
732 		return error;
733 
734 	NET_EPOCH_ENTER(et);
735 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
736 	    inp != NULL;
737 	    inp = CK_LIST_NEXT(inp, inp_list)) {
738 		INP_RLOCK(inp);
739 		if (inp->inp_gencnt <= xig.xig_gen) {
740 			struct xinpcb xi;
741 
742 			in_pcbtoxinpcb(inp, &xi);
743 			INP_RUNLOCK(inp);
744 			error = SYSCTL_OUT(req, &xi, sizeof xi);
745 		} else
746 			INP_RUNLOCK(inp);
747 	}
748 	NET_EPOCH_EXIT(et);
749 
750 	if (!error) {
751 		/*
752 		 * Give the user an updated idea of our state.
753 		 * If the generation differs from what we told
754 		 * her before, she knows that something happened
755 		 * while we were processing this request, and it
756 		 * might be necessary to retry.
757 		 */
758 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
759 		xig.xig_sogen = so_gencnt;
760 		xig.xig_count = V_divcbinfo.ipi_count;
761 		error = SYSCTL_OUT(req, &xig, sizeof xig);
762 	}
763 
764 	return (error);
765 }
766 
767 #ifdef SYSCTL_NODE
768 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
769     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
770     "IPDIVERT");
771 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
772    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
773     NULL, 0, div_pcblist, "S,xinpcb",
774     "List of active divert sockets");
775 #endif
776 
777 struct pr_usrreqs div_usrreqs = {
778 	.pru_attach =		div_attach,
779 	.pru_bind =		div_bind,
780 	.pru_control =		in_control,
781 	.pru_detach =		div_detach,
782 	.pru_peeraddr =		in_getpeeraddr,
783 	.pru_send =		div_send,
784 	.pru_shutdown =		div_shutdown,
785 	.pru_sockaddr =		in_getsockaddr,
786 	.pru_sosetlabel =	in_pcbsosetlabel
787 };
788 
789 struct protosw div_protosw = {
790 	.pr_type =		SOCK_RAW,
791 	.pr_protocol =		IPPROTO_DIVERT,
792 	.pr_flags =		PR_ATOMIC|PR_ADDR,
793 	.pr_input =		div_input,
794 	.pr_ctlinput =		div_ctlinput,
795 	.pr_ctloutput =		ip_ctloutput,
796 	.pr_init =		div_init,
797 	.pr_usrreqs =		&div_usrreqs
798 };
799 
800 static int
801 div_modevent(module_t mod, int type, void *unused)
802 {
803 	int err = 0;
804 
805 	switch (type) {
806 	case MOD_LOAD:
807 		/*
808 		 * Protocol will be initialized by pf_proto_register().
809 		 * We don't have to register ip_protox because we are not
810 		 * a true IP protocol that goes over the wire.
811 		 */
812 		err = pf_proto_register(PF_INET, &div_protosw);
813 		if (err != 0)
814 			return (err);
815 		ip_divert_ptr = divert_packet;
816 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
817 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
818 		break;
819 	case MOD_QUIESCE:
820 		/*
821 		 * IPDIVERT may normally not be unloaded because of the
822 		 * potential race conditions.  Tell kldunload we can't be
823 		 * unloaded unless the unload is forced.
824 		 */
825 		err = EPERM;
826 		break;
827 	case MOD_UNLOAD:
828 		/*
829 		 * Forced unload.
830 		 *
831 		 * Module ipdivert can only be unloaded if no sockets are
832 		 * connected.  Maybe this can be changed later to forcefully
833 		 * disconnect any open sockets.
834 		 *
835 		 * XXXRW: Note that there is a slight race here, as a new
836 		 * socket open request could be spinning on the lock and then
837 		 * we destroy the lock.
838 		 */
839 		INP_INFO_WLOCK(&V_divcbinfo);
840 		if (V_divcbinfo.ipi_count != 0) {
841 			err = EBUSY;
842 			INP_INFO_WUNLOCK(&V_divcbinfo);
843 			break;
844 		}
845 		ip_divert_ptr = NULL;
846 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
847 		INP_INFO_WUNLOCK(&V_divcbinfo);
848 #ifndef VIMAGE
849 		div_destroy(NULL);
850 #endif
851 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
852 		break;
853 	default:
854 		err = EOPNOTSUPP;
855 		break;
856 	}
857 	return err;
858 }
859 
860 static moduledata_t ipdivertmod = {
861         "ipdivert",
862         div_modevent,
863         0
864 };
865 
866 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
867 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
868 MODULE_VERSION(ipdivert, 1);
869