xref: /freebsd/sys/netinet/ip_divert.c (revision 52267f7411adcc76ede961420e08c0e42f42d415)
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 #include <sys/vimage.h>
62 
63 #include <vm/uma.h>
64 
65 #include <net/if.h>
66 #include <net/netisr.h>
67 #include <net/route.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_pcb.h>
71 #include <netinet/in_systm.h>
72 #include <netinet/in_var.h>
73 #include <netinet/ip.h>
74 #include <netinet/ip_divert.h>
75 #include <netinet/ip_var.h>
76 #include <netinet/ip_fw.h>
77 #include <netinet/vinet.h>
78 
79 #include <security/mac/mac_framework.h>
80 
81 /*
82  * Divert sockets
83  */
84 
85 /*
86  * Allocate enough space to hold a full IP packet
87  */
88 #define	DIVSNDQ		(65536 + 100)
89 #define	DIVRCVQ		(65536 + 100)
90 
91 /*
92  * Divert sockets work in conjunction with ipfw, see the divert(4)
93  * manpage for features.
94  * Internally, packets selected by ipfw in ip_input() or ip_output(),
95  * and never diverted before, are passed to the input queue of the
96  * divert socket with a given 'divert_port' number (as specified in
97  * the matching ipfw rule), and they are tagged with a 16 bit cookie
98  * (representing the rule number of the matching ipfw rule), which
99  * is passed to process reading from the socket.
100  *
101  * Packets written to the divert socket are again tagged with a cookie
102  * (usually the same as above) and a destination address.
103  * If the destination address is INADDR_ANY then the packet is
104  * treated as outgoing and sent to ip_output(), otherwise it is
105  * treated as incoming and sent to ip_input().
106  * In both cases, the packet is tagged with the cookie.
107  *
108  * On reinjection, processing in ip_input() and ip_output()
109  * will be exactly the same as for the original packet, except that
110  * ipfw processing will start at the rule number after the one
111  * written in the cookie (so, tagging a packet with a cookie of 0
112  * will cause it to be effectively considered as a standard packet).
113  */
114 
115 /* Internal variables. */
116 #ifdef VIMAGE_GLOBALS
117 static struct inpcbhead divcb;
118 static struct inpcbinfo divcbinfo;
119 #endif
120 
121 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
122 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
123 
124 /*
125  * Initialize divert connection block queue.
126  */
127 static void
128 div_zone_change(void *tag)
129 {
130 
131 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
132 }
133 
134 static int
135 div_inpcb_init(void *mem, int size, int flags)
136 {
137 	struct inpcb *inp = mem;
138 
139 	INP_LOCK_INIT(inp, "inp", "divinp");
140 	return (0);
141 }
142 
143 static void
144 div_inpcb_fini(void *mem, int size)
145 {
146 	struct inpcb *inp = mem;
147 
148 	INP_LOCK_DESTROY(inp);
149 }
150 
151 void
152 div_init(void)
153 {
154 	INIT_VNET_INET(curvnet);
155 
156 	INP_INFO_LOCK_INIT(&V_divcbinfo, "div");
157 	LIST_INIT(&V_divcb);
158 	V_divcbinfo.ipi_listhead = &V_divcb;
159 	/*
160 	 * XXX We don't use the hash list for divert IP, but it's easier
161 	 * to allocate a one entry hash list than it is to check all
162 	 * over the place for hashbase == NULL.
163 	 */
164 	V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask);
165 	V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
166 	    &V_divcbinfo.ipi_porthashmask);
167 	V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
168 	    NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
169 	    UMA_ZONE_NOFREE);
170 	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
171 	EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
172 		NULL, EVENTHANDLER_PRI_ANY);
173 }
174 
175 /*
176  * IPPROTO_DIVERT is not in the real IP protocol number space; this
177  * function should never be called.  Just in case, drop any packets.
178  */
179 void
180 div_input(struct mbuf *m, int off)
181 {
182 	INIT_VNET_INET(curvnet);
183 
184 	V_ipstat.ips_noproto++;
185 	m_freem(m);
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, int incoming)
196 {
197 	INIT_VNET_INET(curvnet);
198 	struct ip *ip;
199 	struct inpcb *inp;
200 	struct socket *sa;
201 	u_int16_t nport;
202 	struct sockaddr_in divsrc;
203 	struct m_tag *mtag;
204 
205 	mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
206 	if (mtag == NULL) {
207 		printf("%s: no divert tag\n", __func__);
208 		m_freem(m);
209 		return;
210 	}
211 	/* Assure header */
212 	if (m->m_len < sizeof(struct ip) &&
213 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
214 		return;
215 	ip = mtod(m, struct ip *);
216 
217 	/* Delayed checksums are currently not compatible with divert. */
218 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
219 		ip->ip_len = ntohs(ip->ip_len);
220 		in_delayed_cksum(m);
221 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
222 		ip->ip_len = htons(ip->ip_len);
223 	}
224 
225 	/*
226 	 * Record receive interface address, if any.
227 	 * But only for incoming packets.
228 	 */
229 	bzero(&divsrc, sizeof(divsrc));
230 	divsrc.sin_len = sizeof(divsrc);
231 	divsrc.sin_family = AF_INET;
232 	divsrc.sin_port = divert_cookie(mtag);	/* record matching rule */
233 	if (incoming) {
234 		struct ifaddr *ifa;
235 
236 		/* Sanity check */
237 		M_ASSERTPKTHDR(m);
238 
239 		/* Find IP address for receive interface */
240 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
241 			if (ifa->ifa_addr->sa_family != AF_INET)
242 				continue;
243 			divsrc.sin_addr =
244 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
245 			break;
246 		}
247 	}
248 	/*
249 	 * Record the incoming interface name whenever we have one.
250 	 */
251 	if (m->m_pkthdr.rcvif) {
252 		/*
253 		 * Hide the actual interface name in there in the
254 		 * sin_zero array. XXX This needs to be moved to a
255 		 * different sockaddr type for divert, e.g.
256 		 * sockaddr_div with multiple fields like
257 		 * sockaddr_dl. Presently we have only 7 bytes
258 		 * but that will do for now as most interfaces
259 		 * are 4 or less + 2 or less bytes for unit.
260 		 * There is probably a faster way of doing this,
261 		 * possibly taking it from the sockaddr_dl on the iface.
262 		 * This solves the problem of a P2P link and a LAN interface
263 		 * having the same address, which can result in the wrong
264 		 * interface being assigned to the packet when fed back
265 		 * into the divert socket. Theoretically if the daemon saves
266 		 * and re-uses the sockaddr_in as suggested in the man pages,
267 		 * this iface name will come along for the ride.
268 		 * (see div_output for the other half of this.)
269 		 */
270 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
271 		    sizeof(divsrc.sin_zero));
272 	}
273 
274 	/* Put packet on socket queue, if any */
275 	sa = NULL;
276 	nport = htons((u_int16_t)divert_info(mtag));
277 	INP_INFO_RLOCK(&V_divcbinfo);
278 	LIST_FOREACH(inp, &V_divcb, inp_list) {
279 		/* XXX why does only one socket match? */
280 		if (inp->inp_lport == nport) {
281 			INP_RLOCK(inp);
282 			sa = inp->inp_socket;
283 			SOCKBUF_LOCK(&sa->so_rcv);
284 			if (sbappendaddr_locked(&sa->so_rcv,
285 			    (struct sockaddr *)&divsrc, m,
286 			    (struct mbuf *)0) == 0) {
287 				SOCKBUF_UNLOCK(&sa->so_rcv);
288 				sa = NULL;	/* force mbuf reclaim below */
289 			} else
290 				sorwakeup_locked(sa);
291 			INP_RUNLOCK(inp);
292 			break;
293 		}
294 	}
295 	INP_INFO_RUNLOCK(&V_divcbinfo);
296 	if (sa == NULL) {
297 		m_freem(m);
298 		V_ipstat.ips_noproto++;
299 		V_ipstat.ips_delivered--;
300         }
301 }
302 
303 /*
304  * Deliver packet back into the IP processing machinery.
305  *
306  * If no address specified, or address is 0.0.0.0, send to ip_output();
307  * otherwise, send to ip_input() and mark as having been received on
308  * the interface with that address.
309  */
310 static int
311 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
312     struct mbuf *control)
313 {
314 	INIT_VNET_INET(curvnet);
315 	struct m_tag *mtag;
316 	struct divert_tag *dt;
317 	int error = 0;
318 	struct mbuf *options;
319 
320 	/*
321 	 * An mbuf may hasn't come from userland, but we pretend
322 	 * that it has.
323 	 */
324 	m->m_pkthdr.rcvif = NULL;
325 	m->m_nextpkt = NULL;
326 	M_SETFIB(m, so->so_fibnum);
327 
328 	if (control)
329 		m_freem(control);		/* XXX */
330 
331 	if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
332 		mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
333 		    M_NOWAIT | M_ZERO);
334 		if (mtag == NULL) {
335 			error = ENOBUFS;
336 			goto cantsend;
337 		}
338 		dt = (struct divert_tag *)(mtag+1);
339 		m_tag_prepend(m, mtag);
340 	} else
341 		dt = (struct divert_tag *)(mtag+1);
342 
343 	/* Loopback avoidance and state recovery */
344 	if (sin) {
345 		int i;
346 
347 		dt->cookie = sin->sin_port;
348 		/*
349 		 * Find receive interface with the given name, stuffed
350 		 * (if it exists) in the sin_zero[] field.
351 		 * The name is user supplied data so don't trust its size
352 		 * or that it is zero terminated.
353 		 */
354 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
355 			;
356 		if ( i > 0 && i < sizeof(sin->sin_zero))
357 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
358 	}
359 
360 	/* Reinject packet into the system as incoming or outgoing */
361 	if (!sin || sin->sin_addr.s_addr == 0) {
362 		struct ip *const ip = mtod(m, struct ip *);
363 		struct inpcb *inp;
364 
365 		dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
366 		INP_INFO_WLOCK(&V_divcbinfo);
367 		inp = sotoinpcb(so);
368 		INP_RLOCK(inp);
369 		/*
370 		 * Don't allow both user specified and setsockopt options,
371 		 * and don't allow packet length sizes that will crash
372 		 */
373 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
374 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
375 			error = EINVAL;
376 			INP_RUNLOCK(inp);
377 			INP_INFO_WUNLOCK(&V_divcbinfo);
378 			m_freem(m);
379 		} else {
380 			/* Convert fields to host order for ip_output() */
381 			ip->ip_len = ntohs(ip->ip_len);
382 			ip->ip_off = ntohs(ip->ip_off);
383 
384 			/* Send packet to output processing */
385 			V_ipstat.ips_rawout++;			/* XXX */
386 
387 #ifdef MAC
388 			mac_inpcb_create_mbuf(inp, m);
389 #endif
390 			/*
391 			 * Get ready to inject the packet into ip_output().
392 			 * Just in case socket options were specified on the
393 			 * divert socket, we duplicate them.  This is done
394 			 * to avoid having to hold the PCB locks over the call
395 			 * to ip_output(), as doing this results in a number of
396 			 * lock ordering complexities.
397 			 *
398 			 * Note that we set the multicast options argument for
399 			 * ip_output() to NULL since it should be invariant that
400 			 * they are not present.
401 			 */
402 			KASSERT(inp->inp_moptions == NULL,
403 			    ("multicast options set on a divert socket"));
404 			options = NULL;
405 			/*
406 			 * XXXCSJP: It is unclear to me whether or not it makes
407 			 * sense for divert sockets to have options.  However,
408 			 * for now we will duplicate them with the INP locks
409 			 * held so we can use them in ip_output() without
410 			 * requring a reference to the pcb.
411 			 */
412 			if (inp->inp_options != NULL) {
413 				options = m_dup(inp->inp_options, M_DONTWAIT);
414 				if (options == NULL)
415 					error = ENOBUFS;
416 			}
417 			INP_RUNLOCK(inp);
418 			INP_INFO_WUNLOCK(&V_divcbinfo);
419 			if (error == ENOBUFS) {
420 				m_freem(m);
421 				return (error);
422 			}
423 			error = ip_output(m, options, NULL,
424 			    ((so->so_options & SO_DONTROUTE) ?
425 			    IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
426 			    IP_RAWOUTPUT, NULL, NULL);
427 			if (options != NULL)
428 				m_freem(options);
429 		}
430 	} else {
431 		dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
432 		if (m->m_pkthdr.rcvif == NULL) {
433 			/*
434 			 * No luck with the name, check by IP address.
435 			 * Clear the port and the ifname to make sure
436 			 * there are no distractions for ifa_ifwithaddr.
437 			 */
438 			struct	ifaddr *ifa;
439 
440 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
441 			sin->sin_port = 0;
442 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
443 			if (ifa == NULL) {
444 				error = EADDRNOTAVAIL;
445 				goto cantsend;
446 			}
447 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
448 		}
449 #ifdef MAC
450 		SOCK_LOCK(so);
451 		mac_socket_create_mbuf(so, m);
452 		SOCK_UNLOCK(so);
453 #endif
454 		/* Send packet to input processing via netisr */
455 		netisr_queue(NETISR_IP, m);
456 	}
457 
458 	return error;
459 
460 cantsend:
461 	m_freem(m);
462 	return error;
463 }
464 
465 static int
466 div_attach(struct socket *so, int proto, struct thread *td)
467 {
468 	INIT_VNET_INET(so->so_vnet);
469 	struct inpcb *inp;
470 	int error;
471 
472 	inp  = sotoinpcb(so);
473 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
474 	if (td != NULL) {
475 		error = priv_check(td, PRIV_NETINET_DIVERT);
476 		if (error)
477 			return (error);
478 	}
479 	error = soreserve(so, div_sendspace, div_recvspace);
480 	if (error)
481 		return error;
482 	INP_INFO_WLOCK(&V_divcbinfo);
483 	error = in_pcballoc(so, &V_divcbinfo);
484 	if (error) {
485 		INP_INFO_WUNLOCK(&V_divcbinfo);
486 		return error;
487 	}
488 	inp = (struct inpcb *)so->so_pcb;
489 	INP_INFO_WUNLOCK(&V_divcbinfo);
490 	inp->inp_ip_p = proto;
491 	inp->inp_vflag |= INP_IPV4;
492 	inp->inp_flags |= INP_HDRINCL;
493 	INP_WUNLOCK(inp);
494 	return 0;
495 }
496 
497 static void
498 div_detach(struct socket *so)
499 {
500 	INIT_VNET_INET(so->so_vnet);
501 	struct inpcb *inp;
502 
503 	inp = sotoinpcb(so);
504 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
505 	INP_INFO_WLOCK(&V_divcbinfo);
506 	INP_WLOCK(inp);
507 	in_pcbdetach(inp);
508 	in_pcbfree(inp);
509 	INP_INFO_WUNLOCK(&V_divcbinfo);
510 }
511 
512 static int
513 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
514 {
515 	INIT_VNET_INET(so->so_vnet);
516 	struct inpcb *inp;
517 	int error;
518 
519 	inp = sotoinpcb(so);
520 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
521 	/* in_pcbbind assumes that nam is a sockaddr_in
522 	 * and in_pcbbind requires a valid address. Since divert
523 	 * sockets don't we need to make sure the address is
524 	 * filled in properly.
525 	 * XXX -- divert should not be abusing in_pcbind
526 	 * and should probably have its own family.
527 	 */
528 	if (nam->sa_family != AF_INET)
529 		return EAFNOSUPPORT;
530 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
531 	INP_INFO_WLOCK(&V_divcbinfo);
532 	INP_WLOCK(inp);
533 	error = in_pcbbind(inp, nam, td->td_ucred);
534 	INP_WUNLOCK(inp);
535 	INP_INFO_WUNLOCK(&V_divcbinfo);
536 	return error;
537 }
538 
539 static int
540 div_shutdown(struct socket *so)
541 {
542 	struct inpcb *inp;
543 
544 	inp = sotoinpcb(so);
545 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
546 	INP_WLOCK(inp);
547 	socantsendmore(so);
548 	INP_WUNLOCK(inp);
549 	return 0;
550 }
551 
552 static int
553 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
554     struct mbuf *control, struct thread *td)
555 {
556 	INIT_VNET_INET(so->so_vnet);
557 
558 	/* Packet must have a header (but that's about it) */
559 	if (m->m_len < sizeof (struct ip) &&
560 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
561 		V_ipstat.ips_toosmall++;
562 		m_freem(m);
563 		return EINVAL;
564 	}
565 
566 	/* Send packet */
567 	return div_output(so, m, (struct sockaddr_in *)nam, control);
568 }
569 
570 void
571 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
572 {
573         struct in_addr faddr;
574 
575 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
576 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
577         	return;
578 	if (PRC_IS_REDIRECT(cmd))
579 		return;
580 }
581 
582 static int
583 div_pcblist(SYSCTL_HANDLER_ARGS)
584 {
585 	INIT_VNET_INET(curvnet);
586 	int error, i, n;
587 	struct inpcb *inp, **inp_list;
588 	inp_gen_t gencnt;
589 	struct xinpgen xig;
590 
591 	/*
592 	 * The process of preparing the TCB list is too time-consuming and
593 	 * resource-intensive to repeat twice on every request.
594 	 */
595 	if (req->oldptr == 0) {
596 		n = V_divcbinfo.ipi_count;
597 		req->oldidx = 2 * (sizeof xig)
598 			+ (n + n/8) * sizeof(struct xinpcb);
599 		return 0;
600 	}
601 
602 	if (req->newptr != 0)
603 		return EPERM;
604 
605 	/*
606 	 * OK, now we're committed to doing something.
607 	 */
608 	INP_INFO_RLOCK(&V_divcbinfo);
609 	gencnt = V_divcbinfo.ipi_gencnt;
610 	n = V_divcbinfo.ipi_count;
611 	INP_INFO_RUNLOCK(&V_divcbinfo);
612 
613 	error = sysctl_wire_old_buffer(req,
614 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
615 	if (error != 0)
616 		return (error);
617 
618 	xig.xig_len = sizeof xig;
619 	xig.xig_count = n;
620 	xig.xig_gen = gencnt;
621 	xig.xig_sogen = so_gencnt;
622 	error = SYSCTL_OUT(req, &xig, sizeof xig);
623 	if (error)
624 		return error;
625 
626 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
627 	if (inp_list == 0)
628 		return ENOMEM;
629 
630 	INP_INFO_RLOCK(&V_divcbinfo);
631 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
632 	     inp = LIST_NEXT(inp, inp_list)) {
633 		INP_RLOCK(inp);
634 		if (inp->inp_gencnt <= gencnt &&
635 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
636 			inp_list[i++] = inp;
637 		INP_RUNLOCK(inp);
638 	}
639 	INP_INFO_RUNLOCK(&V_divcbinfo);
640 	n = i;
641 
642 	error = 0;
643 	for (i = 0; i < n; i++) {
644 		inp = inp_list[i];
645 		INP_RLOCK(inp);
646 		if (inp->inp_gencnt <= gencnt) {
647 			struct xinpcb xi;
648 			bzero(&xi, sizeof(xi));
649 			xi.xi_len = sizeof xi;
650 			/* XXX should avoid extra copy */
651 			bcopy(inp, &xi.xi_inp, sizeof *inp);
652 			if (inp->inp_socket)
653 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
654 			INP_RUNLOCK(inp);
655 			error = SYSCTL_OUT(req, &xi, sizeof xi);
656 		} else
657 			INP_RUNLOCK(inp);
658 	}
659 	if (!error) {
660 		/*
661 		 * Give the user an updated idea of our state.
662 		 * If the generation differs from what we told
663 		 * her before, she knows that something happened
664 		 * while we were processing this request, and it
665 		 * might be necessary to retry.
666 		 */
667 		INP_INFO_RLOCK(&V_divcbinfo);
668 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
669 		xig.xig_sogen = so_gencnt;
670 		xig.xig_count = V_divcbinfo.ipi_count;
671 		INP_INFO_RUNLOCK(&V_divcbinfo);
672 		error = SYSCTL_OUT(req, &xig, sizeof xig);
673 	}
674 	free(inp_list, M_TEMP);
675 	return error;
676 }
677 
678 #ifdef SYSCTL_NODE
679 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
680 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
681 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
682 #endif
683 
684 struct pr_usrreqs div_usrreqs = {
685 	.pru_attach =		div_attach,
686 	.pru_bind =		div_bind,
687 	.pru_control =		in_control,
688 	.pru_detach =		div_detach,
689 	.pru_peeraddr =		in_getpeeraddr,
690 	.pru_send =		div_send,
691 	.pru_shutdown =		div_shutdown,
692 	.pru_sockaddr =		in_getsockaddr,
693 	.pru_sosetlabel =	in_pcbsosetlabel
694 };
695 
696 struct protosw div_protosw = {
697 	.pr_type =		SOCK_RAW,
698 	.pr_protocol =		IPPROTO_DIVERT,
699 	.pr_flags =		PR_ATOMIC|PR_ADDR,
700 	.pr_input =		div_input,
701 	.pr_ctlinput =		div_ctlinput,
702 	.pr_ctloutput =		ip_ctloutput,
703 	.pr_init =		div_init,
704 	.pr_usrreqs =		&div_usrreqs
705 };
706 
707 static int
708 div_modevent(module_t mod, int type, void *unused)
709 {
710 	int err = 0;
711 	int n;
712 
713 	switch (type) {
714 	case MOD_LOAD:
715 		/*
716 		 * Protocol will be initialized by pf_proto_register().
717 		 * We don't have to register ip_protox because we are not
718 		 * a true IP protocol that goes over the wire.
719 		 */
720 		err = pf_proto_register(PF_INET, &div_protosw);
721 		ip_divert_ptr = divert_packet;
722 		break;
723 	case MOD_QUIESCE:
724 		/*
725 		 * IPDIVERT may normally not be unloaded because of the
726 		 * potential race conditions.  Tell kldunload we can't be
727 		 * unloaded unless the unload is forced.
728 		 */
729 		err = EPERM;
730 		break;
731 	case MOD_UNLOAD:
732 		/*
733 		 * Forced unload.
734 		 *
735 		 * Module ipdivert can only be unloaded if no sockets are
736 		 * connected.  Maybe this can be changed later to forcefully
737 		 * disconnect any open sockets.
738 		 *
739 		 * XXXRW: Note that there is a slight race here, as a new
740 		 * socket open request could be spinning on the lock and then
741 		 * we destroy the lock.
742 		 */
743 		INP_INFO_WLOCK(&V_divcbinfo);
744 		n = V_divcbinfo.ipi_count;
745 		if (n != 0) {
746 			err = EBUSY;
747 			INP_INFO_WUNLOCK(&V_divcbinfo);
748 			break;
749 		}
750 		ip_divert_ptr = NULL;
751 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
752 		INP_INFO_WUNLOCK(&V_divcbinfo);
753 		INP_INFO_LOCK_DESTROY(&V_divcbinfo);
754 		uma_zdestroy(V_divcbinfo.ipi_zone);
755 		break;
756 	default:
757 		err = EOPNOTSUPP;
758 		break;
759 	}
760 	return err;
761 }
762 
763 static moduledata_t ipdivertmod = {
764         "ipdivert",
765         div_modevent,
766         0
767 };
768 
769 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
770 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
771 MODULE_VERSION(ipdivert, 1);
772