xref: /freebsd/sys/netinet/ip_divert.c (revision 2f6a179eb910129fb812c1ad1bdc300da1203dc0)
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  * 4. 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 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #if !defined(KLD_MODULE)
34 #include "opt_inet.h"
35 #include "opt_ipfw.h"
36 #include "opt_mac.h"
37 #ifndef INET
38 #error "IPDIVERT requires INET."
39 #endif
40 #ifndef IPFIREWALL
41 #error "IPDIVERT requires IPFIREWALL"
42 #endif
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/module.h>
51 #include <sys/kernel.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/protosw.h>
55 #include <sys/signalvar.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/sx.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 
62 #include <vm/uma.h>
63 
64 #include <net/if.h>
65 #include <net/netisr.h>
66 #include <net/route.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_divert.h>
74 #include <netinet/ip_var.h>
75 #include <netinet/ip_fw.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 /*
80  * Divert sockets
81  */
82 
83 /*
84  * Allocate enough space to hold a full IP packet
85  */
86 #define	DIVSNDQ		(65536 + 100)
87 #define	DIVRCVQ		(65536 + 100)
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 inpcbhead divcb;
115 static struct inpcbinfo divcbinfo;
116 
117 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
118 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
119 
120 /*
121  * Initialize divert connection block queue.
122  */
123 static void
124 div_zone_change(void *tag)
125 {
126 
127 	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
128 }
129 
130 static int
131 div_inpcb_init(void *mem, int size, int flags)
132 {
133 	struct inpcb *inp = mem;
134 
135 	INP_LOCK_INIT(inp, "inp", "divinp");
136 	return (0);
137 }
138 
139 static void
140 div_inpcb_fini(void *mem, int size)
141 {
142 	struct inpcb *inp = mem;
143 
144 	INP_LOCK_DESTROY(inp);
145 }
146 
147 void
148 div_init(void)
149 {
150 
151 	INP_INFO_LOCK_INIT(&divcbinfo, "div");
152 	LIST_INIT(&divcb);
153 	divcbinfo.ipi_listhead = &divcb;
154 	/*
155 	 * XXX We don't use the hash list for divert IP, but it's easier
156 	 * to allocate a one entry hash list than it is to check all
157 	 * over the place for hashbase == NULL.
158 	 */
159 	divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &divcbinfo.ipi_hashmask);
160 	divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
161 	    &divcbinfo.ipi_porthashmask);
162 	divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
163 	    NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
164 	    UMA_ZONE_NOFREE);
165 	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
166 	EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
167 		NULL, EVENTHANDLER_PRI_ANY);
168 }
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 void
175 div_input(struct mbuf *m, int off)
176 {
177 	ipstat.ips_noproto++;
178 	m_freem(m);
179 }
180 
181 /*
182  * Divert a packet by passing it up to the divert socket at port 'port'.
183  *
184  * Setup generic address and protocol structures for div_input routine,
185  * then pass them along with mbuf chain.
186  */
187 static void
188 divert_packet(struct mbuf *m, int incoming)
189 {
190 	struct ip *ip;
191 	struct inpcb *inp;
192 	struct socket *sa;
193 	u_int16_t nport;
194 	struct sockaddr_in divsrc;
195 	struct m_tag *mtag;
196 
197 	mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
198 	if (mtag == NULL) {
199 		printf("%s: no divert tag\n", __func__);
200 		m_freem(m);
201 		return;
202 	}
203 	/* Assure header */
204 	if (m->m_len < sizeof(struct ip) &&
205 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
206 		return;
207 	ip = mtod(m, struct ip *);
208 
209 	/* Delayed checksums are currently not compatible with divert. */
210 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
211 		ip->ip_len = ntohs(ip->ip_len);
212 		in_delayed_cksum(m);
213 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
214 		ip->ip_len = htons(ip->ip_len);
215 	}
216 
217 	/*
218 	 * Record receive interface address, if any.
219 	 * But only for incoming packets.
220 	 */
221 	bzero(&divsrc, sizeof(divsrc));
222 	divsrc.sin_len = sizeof(divsrc);
223 	divsrc.sin_family = AF_INET;
224 	divsrc.sin_port = divert_cookie(mtag);	/* record matching rule */
225 	if (incoming) {
226 		struct ifaddr *ifa;
227 
228 		/* Sanity check */
229 		M_ASSERTPKTHDR(m);
230 
231 		/* Find IP address for receive interface */
232 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
233 			if (ifa->ifa_addr->sa_family != AF_INET)
234 				continue;
235 			divsrc.sin_addr =
236 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
237 			break;
238 		}
239 	}
240 	/*
241 	 * Record the incoming interface name whenever we have one.
242 	 */
243 	if (m->m_pkthdr.rcvif) {
244 		/*
245 		 * Hide the actual interface name in there in the
246 		 * sin_zero array. XXX This needs to be moved to a
247 		 * different sockaddr type for divert, e.g.
248 		 * sockaddr_div with multiple fields like
249 		 * sockaddr_dl. Presently we have only 7 bytes
250 		 * but that will do for now as most interfaces
251 		 * are 4 or less + 2 or less bytes for unit.
252 		 * There is probably a faster way of doing this,
253 		 * possibly taking it from the sockaddr_dl on the iface.
254 		 * This solves the problem of a P2P link and a LAN interface
255 		 * having the same address, which can result in the wrong
256 		 * interface being assigned to the packet when fed back
257 		 * into the divert socket. Theoretically if the daemon saves
258 		 * and re-uses the sockaddr_in as suggested in the man pages,
259 		 * this iface name will come along for the ride.
260 		 * (see div_output for the other half of this.)
261 		 */
262 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
263 		    sizeof(divsrc.sin_zero));
264 	}
265 
266 	/* Put packet on socket queue, if any */
267 	sa = NULL;
268 	nport = htons((u_int16_t)divert_info(mtag));
269 	INP_INFO_RLOCK(&divcbinfo);
270 	LIST_FOREACH(inp, &divcb, inp_list) {
271 		/* XXX why does only one socket match? */
272 		if (inp->inp_lport == nport) {
273 			INP_RLOCK(inp);
274 			sa = inp->inp_socket;
275 			SOCKBUF_LOCK(&sa->so_rcv);
276 			if (sbappendaddr_locked(&sa->so_rcv,
277 			    (struct sockaddr *)&divsrc, m,
278 			    (struct mbuf *)0) == 0) {
279 				SOCKBUF_UNLOCK(&sa->so_rcv);
280 				sa = NULL;	/* force mbuf reclaim below */
281 			} else
282 				sorwakeup_locked(sa);
283 			INP_RUNLOCK(inp);
284 			break;
285 		}
286 	}
287 	INP_INFO_RUNLOCK(&divcbinfo);
288 	if (sa == NULL) {
289 		m_freem(m);
290 		ipstat.ips_noproto++;
291 		ipstat.ips_delivered--;
292         }
293 }
294 
295 /*
296  * Deliver packet back into the IP processing machinery.
297  *
298  * If no address specified, or address is 0.0.0.0, send to ip_output();
299  * otherwise, send to ip_input() and mark as having been received on
300  * the interface with that address.
301  */
302 static int
303 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
304     struct mbuf *control)
305 {
306 	struct m_tag *mtag;
307 	struct divert_tag *dt;
308 	int error = 0;
309 	struct mbuf *options;
310 
311 	/*
312 	 * An mbuf may hasn't come from userland, but we pretend
313 	 * that it has.
314 	 */
315 	m->m_pkthdr.rcvif = NULL;
316 	m->m_nextpkt = NULL;
317 
318 	if (control)
319 		m_freem(control);		/* XXX */
320 
321 	if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
322 		mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
323 		    M_NOWAIT | M_ZERO);
324 		if (mtag == NULL) {
325 			error = ENOBUFS;
326 			goto cantsend;
327 		}
328 		dt = (struct divert_tag *)(mtag+1);
329 		m_tag_prepend(m, mtag);
330 	} else
331 		dt = (struct divert_tag *)(mtag+1);
332 
333 	/* Loopback avoidance and state recovery */
334 	if (sin) {
335 		int i;
336 
337 		dt->cookie = sin->sin_port;
338 		/*
339 		 * Find receive interface with the given name, stuffed
340 		 * (if it exists) in the sin_zero[] field.
341 		 * The name is user supplied data so don't trust its size
342 		 * or that it is zero terminated.
343 		 */
344 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
345 			;
346 		if ( i > 0 && i < sizeof(sin->sin_zero))
347 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
348 	}
349 
350 	/* Reinject packet into the system as incoming or outgoing */
351 	if (!sin || sin->sin_addr.s_addr == 0) {
352 		struct ip *const ip = mtod(m, struct ip *);
353 		struct inpcb *inp;
354 
355 		dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
356 		INP_INFO_WLOCK(&divcbinfo);
357 		inp = sotoinpcb(so);
358 		INP_RLOCK(inp);
359 		/*
360 		 * Don't allow both user specified and setsockopt options,
361 		 * and don't allow packet length sizes that will crash
362 		 */
363 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
364 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
365 			error = EINVAL;
366 			INP_RUNLOCK(inp);
367 			INP_INFO_WUNLOCK(&divcbinfo);
368 			m_freem(m);
369 		} else {
370 			/* Convert fields to host order for ip_output() */
371 			ip->ip_len = ntohs(ip->ip_len);
372 			ip->ip_off = ntohs(ip->ip_off);
373 
374 			/* Send packet to output processing */
375 			ipstat.ips_rawout++;			/* XXX */
376 
377 #ifdef MAC
378 			mac_inpcb_create_mbuf(inp, m);
379 #endif
380 			/*
381 			 * Get ready to inject the packet into ip_output().
382 			 * Just in case socket options were specified on the
383 			 * divert socket, we duplicate them.  This is done
384 			 * to avoid having to hold the PCB locks over the call
385 			 * to ip_output(), as doing this results in a number of
386 			 * lock ordering complexities.
387 			 *
388 			 * Note that we set the multicast options argument for
389 			 * ip_output() to NULL since it should be invariant that
390 			 * they are not present.
391 			 */
392 			KASSERT(inp->inp_moptions == NULL,
393 			    ("multicast options set on a divert socket"));
394 			options = NULL;
395 			/*
396 			 * XXXCSJP: It is unclear to me whether or not it makes
397 			 * sense for divert sockets to have options.  However,
398 			 * for now we will duplicate them with the INP locks
399 			 * held so we can use them in ip_output() without
400 			 * requring a reference to the pcb.
401 			 */
402 			if (inp->inp_options != NULL) {
403 				options = m_dup(inp->inp_options, M_DONTWAIT);
404 				if (options == NULL)
405 					error = ENOBUFS;
406 			}
407 			INP_RUNLOCK(inp);
408 			INP_INFO_WUNLOCK(&divcbinfo);
409 			if (error == ENOBUFS) {
410 				m_freem(m);
411 				return (error);
412 			}
413 			error = ip_output(m, options, NULL,
414 			    ((so->so_options & SO_DONTROUTE) ?
415 			    IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
416 			    IP_RAWOUTPUT, NULL, NULL);
417 			if (options != NULL)
418 				m_freem(options);
419 		}
420 	} else {
421 		dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
422 		if (m->m_pkthdr.rcvif == NULL) {
423 			/*
424 			 * No luck with the name, check by IP address.
425 			 * Clear the port and the ifname to make sure
426 			 * there are no distractions for ifa_ifwithaddr.
427 			 */
428 			struct	ifaddr *ifa;
429 
430 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
431 			sin->sin_port = 0;
432 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
433 			if (ifa == NULL) {
434 				error = EADDRNOTAVAIL;
435 				goto cantsend;
436 			}
437 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
438 		}
439 #ifdef MAC
440 		SOCK_LOCK(so);
441 		mac_socket_create_mbuf(so, m);
442 		SOCK_UNLOCK(so);
443 #endif
444 		/* Send packet to input processing via netisr */
445 		netisr_queue(NETISR_IP, m);
446 	}
447 
448 	return error;
449 
450 cantsend:
451 	m_freem(m);
452 	return error;
453 }
454 
455 static int
456 div_attach(struct socket *so, int proto, struct thread *td)
457 {
458 	struct inpcb *inp;
459 	int error;
460 
461 	inp  = sotoinpcb(so);
462 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
463 	if (td != NULL) {
464 		error = priv_check(td, PRIV_NETINET_DIVERT);
465 		if (error)
466 			return (error);
467 	}
468 	error = soreserve(so, div_sendspace, div_recvspace);
469 	if (error)
470 		return error;
471 	INP_INFO_WLOCK(&divcbinfo);
472 	error = in_pcballoc(so, &divcbinfo);
473 	if (error) {
474 		INP_INFO_WUNLOCK(&divcbinfo);
475 		return error;
476 	}
477 	inp = (struct inpcb *)so->so_pcb;
478 	INP_INFO_WUNLOCK(&divcbinfo);
479 	inp->inp_ip_p = proto;
480 	inp->inp_vflag |= INP_IPV4;
481 	inp->inp_flags |= INP_HDRINCL;
482 	INP_WUNLOCK(inp);
483 	return 0;
484 }
485 
486 static void
487 div_detach(struct socket *so)
488 {
489 	struct inpcb *inp;
490 
491 	inp = sotoinpcb(so);
492 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
493 	INP_INFO_WLOCK(&divcbinfo);
494 	INP_WLOCK(inp);
495 	in_pcbdetach(inp);
496 	in_pcbfree(inp);
497 	INP_INFO_WUNLOCK(&divcbinfo);
498 }
499 
500 static int
501 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
502 {
503 	struct inpcb *inp;
504 	int error;
505 
506 	inp = sotoinpcb(so);
507 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
508 	/* in_pcbbind assumes that nam is a sockaddr_in
509 	 * and in_pcbbind requires a valid address. Since divert
510 	 * sockets don't we need to make sure the address is
511 	 * filled in properly.
512 	 * XXX -- divert should not be abusing in_pcbind
513 	 * and should probably have its own family.
514 	 */
515 	if (nam->sa_family != AF_INET)
516 		return EAFNOSUPPORT;
517 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
518 	INP_INFO_WLOCK(&divcbinfo);
519 	INP_WLOCK(inp);
520 	error = in_pcbbind(inp, nam, td->td_ucred);
521 	INP_WUNLOCK(inp);
522 	INP_INFO_WUNLOCK(&divcbinfo);
523 	return error;
524 }
525 
526 static int
527 div_shutdown(struct socket *so)
528 {
529 	struct inpcb *inp;
530 
531 	inp = sotoinpcb(so);
532 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
533 	INP_WLOCK(inp);
534 	socantsendmore(so);
535 	INP_WUNLOCK(inp);
536 	return 0;
537 }
538 
539 static int
540 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
541     struct mbuf *control, struct thread *td)
542 {
543 	/* Packet must have a header (but that's about it) */
544 	if (m->m_len < sizeof (struct ip) &&
545 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
546 		ipstat.ips_toosmall++;
547 		m_freem(m);
548 		return EINVAL;
549 	}
550 
551 	/* Send packet */
552 	return div_output(so, m, (struct sockaddr_in *)nam, control);
553 }
554 
555 void
556 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
557 {
558         struct in_addr faddr;
559 
560 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
561 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
562         	return;
563 	if (PRC_IS_REDIRECT(cmd))
564 		return;
565 }
566 
567 static int
568 div_pcblist(SYSCTL_HANDLER_ARGS)
569 {
570 	int error, i, n;
571 	struct inpcb *inp, **inp_list;
572 	inp_gen_t gencnt;
573 	struct xinpgen xig;
574 
575 	/*
576 	 * The process of preparing the TCB list is too time-consuming and
577 	 * resource-intensive to repeat twice on every request.
578 	 */
579 	if (req->oldptr == 0) {
580 		n = divcbinfo.ipi_count;
581 		req->oldidx = 2 * (sizeof xig)
582 			+ (n + n/8) * sizeof(struct xinpcb);
583 		return 0;
584 	}
585 
586 	if (req->newptr != 0)
587 		return EPERM;
588 
589 	/*
590 	 * OK, now we're committed to doing something.
591 	 */
592 	INP_INFO_RLOCK(&divcbinfo);
593 	gencnt = divcbinfo.ipi_gencnt;
594 	n = divcbinfo.ipi_count;
595 	INP_INFO_RUNLOCK(&divcbinfo);
596 
597 	error = sysctl_wire_old_buffer(req,
598 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
599 	if (error != 0)
600 		return (error);
601 
602 	xig.xig_len = sizeof xig;
603 	xig.xig_count = n;
604 	xig.xig_gen = gencnt;
605 	xig.xig_sogen = so_gencnt;
606 	error = SYSCTL_OUT(req, &xig, sizeof xig);
607 	if (error)
608 		return error;
609 
610 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
611 	if (inp_list == 0)
612 		return ENOMEM;
613 
614 	INP_INFO_RLOCK(&divcbinfo);
615 	for (inp = LIST_FIRST(divcbinfo.ipi_listhead), i = 0; inp && i < n;
616 	     inp = LIST_NEXT(inp, inp_list)) {
617 		INP_RLOCK(inp);
618 		if (inp->inp_gencnt <= gencnt &&
619 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
620 			inp_list[i++] = inp;
621 		INP_RUNLOCK(inp);
622 	}
623 	INP_INFO_RUNLOCK(&divcbinfo);
624 	n = i;
625 
626 	error = 0;
627 	for (i = 0; i < n; i++) {
628 		inp = inp_list[i];
629 		INP_RLOCK(inp);
630 		if (inp->inp_gencnt <= gencnt) {
631 			struct xinpcb xi;
632 			bzero(&xi, sizeof(xi));
633 			xi.xi_len = sizeof xi;
634 			/* XXX should avoid extra copy */
635 			bcopy(inp, &xi.xi_inp, sizeof *inp);
636 			if (inp->inp_socket)
637 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
638 			INP_RUNLOCK(inp);
639 			error = SYSCTL_OUT(req, &xi, sizeof xi);
640 		} else
641 			INP_RUNLOCK(inp);
642 	}
643 	if (!error) {
644 		/*
645 		 * Give the user an updated idea of our state.
646 		 * If the generation differs from what we told
647 		 * her before, she knows that something happened
648 		 * while we were processing this request, and it
649 		 * might be necessary to retry.
650 		 */
651 		INP_INFO_RLOCK(&divcbinfo);
652 		xig.xig_gen = divcbinfo.ipi_gencnt;
653 		xig.xig_sogen = so_gencnt;
654 		xig.xig_count = divcbinfo.ipi_count;
655 		INP_INFO_RUNLOCK(&divcbinfo);
656 		error = SYSCTL_OUT(req, &xig, sizeof xig);
657 	}
658 	free(inp_list, M_TEMP);
659 	return error;
660 }
661 
662 #ifdef SYSCTL_NODE
663 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
664 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
665 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
666 #endif
667 
668 struct pr_usrreqs div_usrreqs = {
669 	.pru_attach =		div_attach,
670 	.pru_bind =		div_bind,
671 	.pru_control =		in_control,
672 	.pru_detach =		div_detach,
673 	.pru_peeraddr =		in_getpeeraddr,
674 	.pru_send =		div_send,
675 	.pru_shutdown =		div_shutdown,
676 	.pru_sockaddr =		in_getsockaddr,
677 	.pru_sosetlabel =	in_pcbsosetlabel
678 };
679 
680 struct protosw div_protosw = {
681 	.pr_type =		SOCK_RAW,
682 	.pr_protocol =		IPPROTO_DIVERT,
683 	.pr_flags =		PR_ATOMIC|PR_ADDR,
684 	.pr_input =		div_input,
685 	.pr_ctlinput =		div_ctlinput,
686 	.pr_ctloutput =		ip_ctloutput,
687 	.pr_init =		div_init,
688 	.pr_usrreqs =		&div_usrreqs
689 };
690 
691 static int
692 div_modevent(module_t mod, int type, void *unused)
693 {
694 	int err = 0;
695 	int n;
696 
697 	switch (type) {
698 	case MOD_LOAD:
699 		/*
700 		 * Protocol will be initialized by pf_proto_register().
701 		 * We don't have to register ip_protox because we are not
702 		 * a true IP protocol that goes over the wire.
703 		 */
704 		err = pf_proto_register(PF_INET, &div_protosw);
705 		ip_divert_ptr = divert_packet;
706 		break;
707 	case MOD_QUIESCE:
708 		/*
709 		 * IPDIVERT may normally not be unloaded because of the
710 		 * potential race conditions.  Tell kldunload we can't be
711 		 * unloaded unless the unload is forced.
712 		 */
713 		err = EPERM;
714 		break;
715 	case MOD_UNLOAD:
716 		/*
717 		 * Forced unload.
718 		 *
719 		 * Module ipdivert can only be unloaded if no sockets are
720 		 * connected.  Maybe this can be changed later to forcefully
721 		 * disconnect any open sockets.
722 		 *
723 		 * XXXRW: Note that there is a slight race here, as a new
724 		 * socket open request could be spinning on the lock and then
725 		 * we destroy the lock.
726 		 */
727 		INP_INFO_WLOCK(&divcbinfo);
728 		n = divcbinfo.ipi_count;
729 		if (n != 0) {
730 			err = EBUSY;
731 			INP_INFO_WUNLOCK(&divcbinfo);
732 			break;
733 		}
734 		ip_divert_ptr = NULL;
735 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
736 		INP_INFO_WUNLOCK(&divcbinfo);
737 		INP_INFO_LOCK_DESTROY(&divcbinfo);
738 		uma_zdestroy(divcbinfo.ipi_zone);
739 		break;
740 	default:
741 		err = EOPNOTSUPP;
742 		break;
743 	}
744 	return err;
745 }
746 
747 static moduledata_t ipdivertmod = {
748         "ipdivert",
749         div_modevent,
750         0
751 };
752 
753 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
754 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
755 MODULE_VERSION(ipdivert, 1);
756