xref: /freebsd/sys/netinet/ip_divert.c (revision 54ebdd631db8c0bba2baab0155f603a8b5cf014a)
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/rwlock.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sx.h>
60 #include <sys/sysctl.h>
61 #include <sys/systm.h>
62 #include <sys/vimage.h>
63 
64 #include <vm/uma.h>
65 
66 #include <net/if.h>
67 #include <net/netisr.h>
68 #include <net/route.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip.h>
75 #include <netinet/ip_divert.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_fw.h>
78 #include <netinet/vinet.h>
79 
80 #include <security/mac/mac_framework.h>
81 
82 /*
83  * Divert sockets
84  */
85 
86 /*
87  * Allocate enough space to hold a full IP packet
88  */
89 #define	DIVSNDQ		(65536 + 100)
90 #define	DIVRCVQ		(65536 + 100)
91 
92 /*
93  * Divert sockets work in conjunction with ipfw, see the divert(4)
94  * manpage for features.
95  * Internally, packets selected by ipfw in ip_input() or ip_output(),
96  * and never diverted before, are passed to the input queue of the
97  * divert socket with a given 'divert_port' number (as specified in
98  * the matching ipfw rule), and they are tagged with a 16 bit cookie
99  * (representing the rule number of the matching ipfw rule), which
100  * is passed to process reading from the socket.
101  *
102  * Packets written to the divert socket are again tagged with a cookie
103  * (usually the same as above) and a destination address.
104  * If the destination address is INADDR_ANY then the packet is
105  * treated as outgoing and sent to ip_output(), otherwise it is
106  * treated as incoming and sent to ip_input().
107  * In both cases, the packet is tagged with the cookie.
108  *
109  * On reinjection, processing in ip_input() and ip_output()
110  * will be exactly the same as for the original packet, except that
111  * ipfw processing will start at the rule number after the one
112  * written in the cookie (so, tagging a packet with a cookie of 0
113  * will cause it to be effectively considered as a standard packet).
114  */
115 
116 /* Internal variables. */
117 #ifdef VIMAGE_GLOBALS
118 static struct inpcbhead divcb;
119 static struct inpcbinfo divcbinfo;
120 #endif
121 
122 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
123 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
124 
125 /*
126  * Initialize divert connection block queue.
127  */
128 static void
129 div_zone_change(void *tag)
130 {
131 
132 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
133 }
134 
135 static int
136 div_inpcb_init(void *mem, int size, int flags)
137 {
138 	struct inpcb *inp = mem;
139 
140 	INP_LOCK_INIT(inp, "inp", "divinp");
141 	return (0);
142 }
143 
144 static void
145 div_inpcb_fini(void *mem, int size)
146 {
147 	struct inpcb *inp = mem;
148 
149 	INP_LOCK_DESTROY(inp);
150 }
151 
152 void
153 div_init(void)
154 {
155 	INIT_VNET_INET(curvnet);
156 
157 	INP_INFO_LOCK_INIT(&V_divcbinfo, "div");
158 	LIST_INIT(&V_divcb);
159 	V_divcbinfo.ipi_listhead = &V_divcb;
160 	/*
161 	 * XXX We don't use the hash list for divert IP, but it's easier
162 	 * to allocate a one entry hash list than it is to check all
163 	 * over the place for hashbase == NULL.
164 	 */
165 	V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask);
166 	V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
167 	    &V_divcbinfo.ipi_porthashmask);
168 	V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
169 	    NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
170 	    UMA_ZONE_NOFREE);
171 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
172 	EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
173 		NULL, EVENTHANDLER_PRI_ANY);
174 }
175 
176 /*
177  * IPPROTO_DIVERT is not in the real IP protocol number space; this
178  * function should never be called.  Just in case, drop any packets.
179  */
180 void
181 div_input(struct mbuf *m, int off)
182 {
183 	INIT_VNET_INET(curvnet);
184 
185 	V_ipstat.ips_noproto++;
186 	m_freem(m);
187 }
188 
189 /*
190  * Divert a packet by passing it up to the divert socket at port 'port'.
191  *
192  * Setup generic address and protocol structures for div_input routine,
193  * then pass them along with mbuf chain.
194  */
195 static void
196 divert_packet(struct mbuf *m, int incoming)
197 {
198 	INIT_VNET_INET(curvnet);
199 	struct ip *ip;
200 	struct inpcb *inp;
201 	struct socket *sa;
202 	u_int16_t nport;
203 	struct sockaddr_in divsrc;
204 	struct m_tag *mtag;
205 
206 	mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
207 	if (mtag == NULL) {
208 		printf("%s: no divert tag\n", __func__);
209 		m_freem(m);
210 		return;
211 	}
212 	/* Assure header */
213 	if (m->m_len < sizeof(struct ip) &&
214 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
215 		return;
216 	ip = mtod(m, struct ip *);
217 
218 	/* Delayed checksums are currently not compatible with divert. */
219 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
220 		ip->ip_len = ntohs(ip->ip_len);
221 		in_delayed_cksum(m);
222 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
223 		ip->ip_len = htons(ip->ip_len);
224 	}
225 
226 	/*
227 	 * Record receive interface address, if any.
228 	 * But only for incoming packets.
229 	 */
230 	bzero(&divsrc, sizeof(divsrc));
231 	divsrc.sin_len = sizeof(divsrc);
232 	divsrc.sin_family = AF_INET;
233 	divsrc.sin_port = divert_cookie(mtag);	/* record matching rule */
234 	if (incoming) {
235 		struct ifaddr *ifa;
236 
237 		/* Sanity check */
238 		M_ASSERTPKTHDR(m);
239 
240 		/* Find IP address for receive interface */
241 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
242 			if (ifa->ifa_addr->sa_family != AF_INET)
243 				continue;
244 			divsrc.sin_addr =
245 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
246 			break;
247 		}
248 	}
249 	/*
250 	 * Record the incoming interface name whenever we have one.
251 	 */
252 	if (m->m_pkthdr.rcvif) {
253 		/*
254 		 * Hide the actual interface name in there in the
255 		 * sin_zero array. XXX This needs to be moved to a
256 		 * different sockaddr type for divert, e.g.
257 		 * sockaddr_div with multiple fields like
258 		 * sockaddr_dl. Presently we have only 7 bytes
259 		 * but that will do for now as most interfaces
260 		 * are 4 or less + 2 or less bytes for unit.
261 		 * There is probably a faster way of doing this,
262 		 * possibly taking it from the sockaddr_dl on the iface.
263 		 * This solves the problem of a P2P link and a LAN interface
264 		 * having the same address, which can result in the wrong
265 		 * interface being assigned to the packet when fed back
266 		 * into the divert socket. Theoretically if the daemon saves
267 		 * and re-uses the sockaddr_in as suggested in the man pages,
268 		 * this iface name will come along for the ride.
269 		 * (see div_output for the other half of this.)
270 		 */
271 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
272 		    sizeof(divsrc.sin_zero));
273 	}
274 
275 	/* Put packet on socket queue, if any */
276 	sa = NULL;
277 	nport = htons((u_int16_t)divert_info(mtag));
278 	INP_INFO_RLOCK(&V_divcbinfo);
279 	LIST_FOREACH(inp, &V_divcb, inp_list) {
280 		/* XXX why does only one socket match? */
281 		if (inp->inp_lport == nport) {
282 			INP_RLOCK(inp);
283 			sa = inp->inp_socket;
284 			SOCKBUF_LOCK(&sa->so_rcv);
285 			if (sbappendaddr_locked(&sa->so_rcv,
286 			    (struct sockaddr *)&divsrc, m,
287 			    (struct mbuf *)0) == 0) {
288 				SOCKBUF_UNLOCK(&sa->so_rcv);
289 				sa = NULL;	/* force mbuf reclaim below */
290 			} else
291 				sorwakeup_locked(sa);
292 			INP_RUNLOCK(inp);
293 			break;
294 		}
295 	}
296 	INP_INFO_RUNLOCK(&V_divcbinfo);
297 	if (sa == NULL) {
298 		m_freem(m);
299 		V_ipstat.ips_noproto++;
300 		V_ipstat.ips_delivered--;
301         }
302 }
303 
304 /*
305  * Deliver packet back into the IP processing machinery.
306  *
307  * If no address specified, or address is 0.0.0.0, send to ip_output();
308  * otherwise, send to ip_input() and mark as having been received on
309  * the interface with that address.
310  */
311 static int
312 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
313     struct mbuf *control)
314 {
315 	INIT_VNET_INET(curvnet);
316 	struct m_tag *mtag;
317 	struct divert_tag *dt;
318 	int error = 0;
319 	struct mbuf *options;
320 
321 	/*
322 	 * An mbuf may hasn't come from userland, but we pretend
323 	 * that it has.
324 	 */
325 	m->m_pkthdr.rcvif = NULL;
326 	m->m_nextpkt = NULL;
327 	M_SETFIB(m, so->so_fibnum);
328 
329 	if (control)
330 		m_freem(control);		/* XXX */
331 
332 	if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
333 		mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
334 		    M_NOWAIT | M_ZERO);
335 		if (mtag == NULL) {
336 			error = ENOBUFS;
337 			goto cantsend;
338 		}
339 		dt = (struct divert_tag *)(mtag+1);
340 		m_tag_prepend(m, mtag);
341 	} else
342 		dt = (struct divert_tag *)(mtag+1);
343 
344 	/* Loopback avoidance and state recovery */
345 	if (sin) {
346 		int i;
347 
348 		dt->cookie = sin->sin_port;
349 		/*
350 		 * Find receive interface with the given name, stuffed
351 		 * (if it exists) in the sin_zero[] field.
352 		 * The name is user supplied data so don't trust its size
353 		 * or that it is zero terminated.
354 		 */
355 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
356 			;
357 		if ( i > 0 && i < sizeof(sin->sin_zero))
358 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
359 	}
360 
361 	/* Reinject packet into the system as incoming or outgoing */
362 	if (!sin || sin->sin_addr.s_addr == 0) {
363 		struct ip *const ip = mtod(m, struct ip *);
364 		struct inpcb *inp;
365 
366 		dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
367 		INP_INFO_WLOCK(&V_divcbinfo);
368 		inp = sotoinpcb(so);
369 		INP_RLOCK(inp);
370 		/*
371 		 * Don't allow both user specified and setsockopt options,
372 		 * and don't allow packet length sizes that will crash
373 		 */
374 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
375 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
376 			error = EINVAL;
377 			INP_RUNLOCK(inp);
378 			INP_INFO_WUNLOCK(&V_divcbinfo);
379 			m_freem(m);
380 		} else {
381 			/* Convert fields to host order for ip_output() */
382 			ip->ip_len = ntohs(ip->ip_len);
383 			ip->ip_off = ntohs(ip->ip_off);
384 
385 			/* Send packet to output processing */
386 			V_ipstat.ips_rawout++;			/* XXX */
387 
388 #ifdef MAC
389 			mac_inpcb_create_mbuf(inp, m);
390 #endif
391 			/*
392 			 * Get ready to inject the packet into ip_output().
393 			 * Just in case socket options were specified on the
394 			 * divert socket, we duplicate them.  This is done
395 			 * to avoid having to hold the PCB locks over the call
396 			 * to ip_output(), as doing this results in a number of
397 			 * lock ordering complexities.
398 			 *
399 			 * Note that we set the multicast options argument for
400 			 * ip_output() to NULL since it should be invariant that
401 			 * they are not present.
402 			 */
403 			KASSERT(inp->inp_moptions == NULL,
404 			    ("multicast options set on a divert socket"));
405 			options = NULL;
406 			/*
407 			 * XXXCSJP: It is unclear to me whether or not it makes
408 			 * sense for divert sockets to have options.  However,
409 			 * for now we will duplicate them with the INP locks
410 			 * held so we can use them in ip_output() without
411 			 * requring a reference to the pcb.
412 			 */
413 			if (inp->inp_options != NULL) {
414 				options = m_dup(inp->inp_options, M_DONTWAIT);
415 				if (options == NULL)
416 					error = ENOBUFS;
417 			}
418 			INP_RUNLOCK(inp);
419 			INP_INFO_WUNLOCK(&V_divcbinfo);
420 			if (error == ENOBUFS) {
421 				m_freem(m);
422 				return (error);
423 			}
424 			error = ip_output(m, options, NULL,
425 			    ((so->so_options & SO_DONTROUTE) ?
426 			    IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
427 			    IP_RAWOUTPUT, NULL, NULL);
428 			if (options != NULL)
429 				m_freem(options);
430 		}
431 	} else {
432 		dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
433 		if (m->m_pkthdr.rcvif == NULL) {
434 			/*
435 			 * No luck with the name, check by IP address.
436 			 * Clear the port and the ifname to make sure
437 			 * there are no distractions for ifa_ifwithaddr.
438 			 */
439 			struct	ifaddr *ifa;
440 
441 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
442 			sin->sin_port = 0;
443 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
444 			if (ifa == NULL) {
445 				error = EADDRNOTAVAIL;
446 				goto cantsend;
447 			}
448 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
449 		}
450 #ifdef MAC
451 		SOCK_LOCK(so);
452 		mac_socket_create_mbuf(so, m);
453 		SOCK_UNLOCK(so);
454 #endif
455 		/* Send packet to input processing via netisr */
456 		netisr_queue(NETISR_IP, m);
457 	}
458 
459 	return error;
460 
461 cantsend:
462 	m_freem(m);
463 	return error;
464 }
465 
466 static int
467 div_attach(struct socket *so, int proto, struct thread *td)
468 {
469 	INIT_VNET_INET(so->so_vnet);
470 	struct inpcb *inp;
471 	int error;
472 
473 	inp  = sotoinpcb(so);
474 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
475 	if (td != NULL) {
476 		error = priv_check(td, PRIV_NETINET_DIVERT);
477 		if (error)
478 			return (error);
479 	}
480 	error = soreserve(so, div_sendspace, div_recvspace);
481 	if (error)
482 		return error;
483 	INP_INFO_WLOCK(&V_divcbinfo);
484 	error = in_pcballoc(so, &V_divcbinfo);
485 	if (error) {
486 		INP_INFO_WUNLOCK(&V_divcbinfo);
487 		return error;
488 	}
489 	inp = (struct inpcb *)so->so_pcb;
490 	INP_INFO_WUNLOCK(&V_divcbinfo);
491 	inp->inp_ip_p = proto;
492 	inp->inp_vflag |= INP_IPV4;
493 	inp->inp_flags |= INP_HDRINCL;
494 	INP_WUNLOCK(inp);
495 	return 0;
496 }
497 
498 static void
499 div_detach(struct socket *so)
500 {
501 	INIT_VNET_INET(so->so_vnet);
502 	struct inpcb *inp;
503 
504 	inp = sotoinpcb(so);
505 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
506 	INP_INFO_WLOCK(&V_divcbinfo);
507 	INP_WLOCK(inp);
508 	in_pcbdetach(inp);
509 	in_pcbfree(inp);
510 	INP_INFO_WUNLOCK(&V_divcbinfo);
511 }
512 
513 static int
514 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
515 {
516 	INIT_VNET_INET(so->so_vnet);
517 	struct inpcb *inp;
518 	int error;
519 
520 	inp = sotoinpcb(so);
521 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
522 	/* in_pcbbind assumes that nam is a sockaddr_in
523 	 * and in_pcbbind requires a valid address. Since divert
524 	 * sockets don't we need to make sure the address is
525 	 * filled in properly.
526 	 * XXX -- divert should not be abusing in_pcbind
527 	 * and should probably have its own family.
528 	 */
529 	if (nam->sa_family != AF_INET)
530 		return EAFNOSUPPORT;
531 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
532 	INP_INFO_WLOCK(&V_divcbinfo);
533 	INP_WLOCK(inp);
534 	error = in_pcbbind(inp, nam, td->td_ucred);
535 	INP_WUNLOCK(inp);
536 	INP_INFO_WUNLOCK(&V_divcbinfo);
537 	return error;
538 }
539 
540 static int
541 div_shutdown(struct socket *so)
542 {
543 	struct inpcb *inp;
544 
545 	inp = sotoinpcb(so);
546 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
547 	INP_WLOCK(inp);
548 	socantsendmore(so);
549 	INP_WUNLOCK(inp);
550 	return 0;
551 }
552 
553 static int
554 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
555     struct mbuf *control, struct thread *td)
556 {
557 	INIT_VNET_INET(so->so_vnet);
558 
559 	/* Packet must have a header (but that's about it) */
560 	if (m->m_len < sizeof (struct ip) &&
561 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
562 		V_ipstat.ips_toosmall++;
563 		m_freem(m);
564 		return EINVAL;
565 	}
566 
567 	/* Send packet */
568 	return div_output(so, m, (struct sockaddr_in *)nam, control);
569 }
570 
571 void
572 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
573 {
574         struct in_addr faddr;
575 
576 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
577 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
578         	return;
579 	if (PRC_IS_REDIRECT(cmd))
580 		return;
581 }
582 
583 static int
584 div_pcblist(SYSCTL_HANDLER_ARGS)
585 {
586 	INIT_VNET_INET(curvnet);
587 	int error, i, n;
588 	struct inpcb *inp, **inp_list;
589 	inp_gen_t gencnt;
590 	struct xinpgen xig;
591 
592 	/*
593 	 * The process of preparing the TCB list is too time-consuming and
594 	 * resource-intensive to repeat twice on every request.
595 	 */
596 	if (req->oldptr == 0) {
597 		n = V_divcbinfo.ipi_count;
598 		req->oldidx = 2 * (sizeof xig)
599 			+ (n + n/8) * sizeof(struct xinpcb);
600 		return 0;
601 	}
602 
603 	if (req->newptr != 0)
604 		return EPERM;
605 
606 	/*
607 	 * OK, now we're committed to doing something.
608 	 */
609 	INP_INFO_RLOCK(&V_divcbinfo);
610 	gencnt = V_divcbinfo.ipi_gencnt;
611 	n = V_divcbinfo.ipi_count;
612 	INP_INFO_RUNLOCK(&V_divcbinfo);
613 
614 	error = sysctl_wire_old_buffer(req,
615 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
616 	if (error != 0)
617 		return (error);
618 
619 	xig.xig_len = sizeof xig;
620 	xig.xig_count = n;
621 	xig.xig_gen = gencnt;
622 	xig.xig_sogen = so_gencnt;
623 	error = SYSCTL_OUT(req, &xig, sizeof xig);
624 	if (error)
625 		return error;
626 
627 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
628 	if (inp_list == 0)
629 		return ENOMEM;
630 
631 	INP_INFO_RLOCK(&V_divcbinfo);
632 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
633 	     inp = LIST_NEXT(inp, inp_list)) {
634 		INP_RLOCK(inp);
635 		if (inp->inp_gencnt <= gencnt &&
636 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
637 			inp_list[i++] = inp;
638 		INP_RUNLOCK(inp);
639 	}
640 	INP_INFO_RUNLOCK(&V_divcbinfo);
641 	n = i;
642 
643 	error = 0;
644 	for (i = 0; i < n; i++) {
645 		inp = inp_list[i];
646 		INP_RLOCK(inp);
647 		if (inp->inp_gencnt <= gencnt) {
648 			struct xinpcb xi;
649 			bzero(&xi, sizeof(xi));
650 			xi.xi_len = sizeof xi;
651 			/* XXX should avoid extra copy */
652 			bcopy(inp, &xi.xi_inp, sizeof *inp);
653 			if (inp->inp_socket)
654 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
655 			INP_RUNLOCK(inp);
656 			error = SYSCTL_OUT(req, &xi, sizeof xi);
657 		} else
658 			INP_RUNLOCK(inp);
659 	}
660 	if (!error) {
661 		/*
662 		 * Give the user an updated idea of our state.
663 		 * If the generation differs from what we told
664 		 * her before, she knows that something happened
665 		 * while we were processing this request, and it
666 		 * might be necessary to retry.
667 		 */
668 		INP_INFO_RLOCK(&V_divcbinfo);
669 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
670 		xig.xig_sogen = so_gencnt;
671 		xig.xig_count = V_divcbinfo.ipi_count;
672 		INP_INFO_RUNLOCK(&V_divcbinfo);
673 		error = SYSCTL_OUT(req, &xig, sizeof xig);
674 	}
675 	free(inp_list, M_TEMP);
676 	return error;
677 }
678 
679 #ifdef SYSCTL_NODE
680 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
681 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
682 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
683 #endif
684 
685 struct pr_usrreqs div_usrreqs = {
686 	.pru_attach =		div_attach,
687 	.pru_bind =		div_bind,
688 	.pru_control =		in_control,
689 	.pru_detach =		div_detach,
690 	.pru_peeraddr =		in_getpeeraddr,
691 	.pru_send =		div_send,
692 	.pru_shutdown =		div_shutdown,
693 	.pru_sockaddr =		in_getsockaddr,
694 	.pru_sosetlabel =	in_pcbsosetlabel
695 };
696 
697 struct protosw div_protosw = {
698 	.pr_type =		SOCK_RAW,
699 	.pr_protocol =		IPPROTO_DIVERT,
700 	.pr_flags =		PR_ATOMIC|PR_ADDR,
701 	.pr_input =		div_input,
702 	.pr_ctlinput =		div_ctlinput,
703 	.pr_ctloutput =		ip_ctloutput,
704 	.pr_init =		div_init,
705 	.pr_usrreqs =		&div_usrreqs
706 };
707 
708 static int
709 div_modevent(module_t mod, int type, void *unused)
710 {
711 	int err = 0;
712 	int n;
713 
714 	switch (type) {
715 	case MOD_LOAD:
716 		/*
717 		 * Protocol will be initialized by pf_proto_register().
718 		 * We don't have to register ip_protox because we are not
719 		 * a true IP protocol that goes over the wire.
720 		 */
721 		err = pf_proto_register(PF_INET, &div_protosw);
722 		ip_divert_ptr = divert_packet;
723 		break;
724 	case MOD_QUIESCE:
725 		/*
726 		 * IPDIVERT may normally not be unloaded because of the
727 		 * potential race conditions.  Tell kldunload we can't be
728 		 * unloaded unless the unload is forced.
729 		 */
730 		err = EPERM;
731 		break;
732 	case MOD_UNLOAD:
733 		/*
734 		 * Forced unload.
735 		 *
736 		 * Module ipdivert can only be unloaded if no sockets are
737 		 * connected.  Maybe this can be changed later to forcefully
738 		 * disconnect any open sockets.
739 		 *
740 		 * XXXRW: Note that there is a slight race here, as a new
741 		 * socket open request could be spinning on the lock and then
742 		 * we destroy the lock.
743 		 */
744 		INP_INFO_WLOCK(&V_divcbinfo);
745 		n = V_divcbinfo.ipi_count;
746 		if (n != 0) {
747 			err = EBUSY;
748 			INP_INFO_WUNLOCK(&V_divcbinfo);
749 			break;
750 		}
751 		ip_divert_ptr = NULL;
752 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
753 		INP_INFO_WUNLOCK(&V_divcbinfo);
754 		INP_INFO_LOCK_DESTROY(&V_divcbinfo);
755 		uma_zdestroy(V_divcbinfo.ipi_zone);
756 		break;
757 	default:
758 		err = EOPNOTSUPP;
759 		break;
760 	}
761 	return err;
762 }
763 
764 static moduledata_t ipdivertmod = {
765         "ipdivert",
766         div_modevent,
767         0
768 };
769 
770 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
771 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
772 MODULE_VERSION(ipdivert, 1);
773