xref: /freebsd/sys/netinet/ip_divert.c (revision 4ed925457ab06e83238a5db33e89ccc94b99a713)
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_sctp.h"
36 #ifndef INET
37 #error "IPDIVERT requires INET."
38 #endif
39 #endif
40 
41 #include <sys/param.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/module.h>
47 #include <sys/kernel.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
54 #include <net/vnet.h>
55 
56 #include <net/if.h>
57 #include <net/netisr.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_var.h>
65 #ifdef SCTP
66 #include <netinet/sctp_crc32.h>
67 #endif
68 
69 #include <security/mac/mac_framework.h>
70 
71 /*
72  * Divert sockets
73  */
74 
75 /*
76  * Allocate enough space to hold a full IP packet
77  */
78 #define	DIVSNDQ		(65536 + 100)
79 #define	DIVRCVQ		(65536 + 100)
80 
81 /*
82  * Divert sockets work in conjunction with ipfw or other packet filters,
83  * see the divert(4) manpage for features.
84  * Packets are selected by the packet filter and tagged with an
85  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
86  * the packet filter) and information on the matching filter rule for
87  * subsequent reinjection. The divert_port is used to put the packet
88  * on the corresponding divert socket, while the rule number is passed
89  * up (at least partially) as the sin_port in the struct sockaddr.
90  *
91  * Packets written to the divert socket carry in sin_addr a
92  * destination address, and in sin_port the number of the filter rule
93  * after which to continue processing.
94  * If the destination address is INADDR_ANY, the packet is treated as
95  * as outgoing and sent to ip_output(); otherwise it is treated as
96  * incoming and sent to ip_input().
97  * Further, sin_zero carries some information on the interface,
98  * which can be used in the reinject -- see comments in the code.
99  *
100  * On reinjection, processing in ip_input() and ip_output()
101  * will be exactly the same as for the original packet, except that
102  * packet filter processing will start at the rule number after the one
103  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
104  * will apply the entire ruleset to the packet).
105  */
106 
107 /* Internal variables. */
108 static VNET_DEFINE(struct inpcbhead, divcb);
109 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
110 
111 #define	V_divcb				VNET(divcb)
112 #define	V_divcbinfo			VNET(divcbinfo)
113 
114 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
115 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
116 
117 static eventhandler_tag ip_divert_event_tag;
118 
119 /*
120  * Initialize divert connection block queue.
121  */
122 static void
123 div_zone_change(void *tag)
124 {
125 
126 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
127 }
128 
129 static int
130 div_inpcb_init(void *mem, int size, int flags)
131 {
132 	struct inpcb *inp = mem;
133 
134 	INP_LOCK_INIT(inp, "inp", "divinp");
135 	return (0);
136 }
137 
138 static void
139 div_inpcb_fini(void *mem, int size)
140 {
141 	struct inpcb *inp = mem;
142 
143 	INP_LOCK_DESTROY(inp);
144 }
145 
146 static void
147 div_init(void)
148 {
149 
150 	INP_INFO_LOCK_INIT(&V_divcbinfo, "div");
151 	LIST_INIT(&V_divcb);
152 	V_divcbinfo.ipi_listhead = &V_divcb;
153 #ifdef VIMAGE
154 	V_divcbinfo.ipi_vnet = curvnet;
155 #endif
156 	/*
157 	 * XXX We don't use the hash list for divert IP, but it's easier
158 	 * to allocate a one entry hash list than it is to check all
159 	 * over the place for hashbase == NULL.
160 	 */
161 	V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask);
162 	V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
163 	    &V_divcbinfo.ipi_porthashmask);
164 	V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
165 	    NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
166 	    UMA_ZONE_NOFREE);
167 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
168 }
169 
170 static void
171 div_destroy(void)
172 {
173 
174 	INP_INFO_LOCK_DESTROY(&V_divcbinfo);
175 	uma_zdestroy(V_divcbinfo.ipi_zone);
176 	hashdestroy(V_divcbinfo.ipi_hashbase, M_PCB, V_divcbinfo.ipi_hashmask);
177 	hashdestroy(V_divcbinfo.ipi_porthashbase, M_PCB,
178 	    V_divcbinfo.ipi_porthashmask);
179 }
180 
181 /*
182  * IPPROTO_DIVERT is not in the real IP protocol number space; this
183  * function should never be called.  Just in case, drop any packets.
184  */
185 static void
186 div_input(struct mbuf *m, int off)
187 {
188 
189 	KMOD_IPSTAT_INC(ips_noproto);
190 	m_freem(m);
191 }
192 
193 /*
194  * Divert a packet by passing it up to the divert socket at port 'port'.
195  *
196  * Setup generic address and protocol structures for div_input routine,
197  * then pass them along with mbuf chain.
198  */
199 static void
200 divert_packet(struct mbuf *m, int incoming)
201 {
202 	struct ip *ip;
203 	struct inpcb *inp;
204 	struct socket *sa;
205 	u_int16_t nport;
206 	struct sockaddr_in divsrc;
207 	struct m_tag *mtag;
208 
209 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
210 	if (mtag == NULL) {
211 		m_freem(m);
212 		return;
213 	}
214 	/* Assure header */
215 	if (m->m_len < sizeof(struct ip) &&
216 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
217 		return;
218 	ip = mtod(m, struct ip *);
219 
220 	/* Delayed checksums are currently not compatible with divert. */
221 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
222 		ip->ip_len = ntohs(ip->ip_len);
223 		in_delayed_cksum(m);
224 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
225 		ip->ip_len = htons(ip->ip_len);
226 	}
227 #ifdef SCTP
228 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
229 		ip->ip_len = ntohs(ip->ip_len);
230 		sctp_delayed_cksum(m);
231 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
232 		ip->ip_len = htons(ip->ip_len);
233 	}
234 #endif
235 	bzero(&divsrc, sizeof(divsrc));
236 	divsrc.sin_len = sizeof(divsrc);
237 	divsrc.sin_family = AF_INET;
238 	/* record matching rule, in host format */
239 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
240 	/*
241 	 * Record receive interface address, if any.
242 	 * But only for incoming packets.
243 	 */
244 	if (incoming) {
245 		struct ifaddr *ifa;
246 		struct ifnet *ifp;
247 
248 		/* Sanity check */
249 		M_ASSERTPKTHDR(m);
250 
251 		/* Find IP address for receive interface */
252 		ifp = m->m_pkthdr.rcvif;
253 		if_addr_rlock(ifp);
254 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
255 			if (ifa->ifa_addr->sa_family != AF_INET)
256 				continue;
257 			divsrc.sin_addr =
258 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
259 			break;
260 		}
261 		if_addr_runlock(ifp);
262 	}
263 	/*
264 	 * Record the incoming interface name whenever we have one.
265 	 */
266 	if (m->m_pkthdr.rcvif) {
267 		/*
268 		 * Hide the actual interface name in there in the
269 		 * sin_zero array. XXX This needs to be moved to a
270 		 * different sockaddr type for divert, e.g.
271 		 * sockaddr_div with multiple fields like
272 		 * sockaddr_dl. Presently we have only 7 bytes
273 		 * but that will do for now as most interfaces
274 		 * are 4 or less + 2 or less bytes for unit.
275 		 * There is probably a faster way of doing this,
276 		 * possibly taking it from the sockaddr_dl on the iface.
277 		 * This solves the problem of a P2P link and a LAN interface
278 		 * having the same address, which can result in the wrong
279 		 * interface being assigned to the packet when fed back
280 		 * into the divert socket. Theoretically if the daemon saves
281 		 * and re-uses the sockaddr_in as suggested in the man pages,
282 		 * this iface name will come along for the ride.
283 		 * (see div_output for the other half of this.)
284 		 */
285 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
286 		    sizeof(divsrc.sin_zero));
287 	}
288 
289 	/* Put packet on socket queue, if any */
290 	sa = NULL;
291 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
292 	INP_INFO_RLOCK(&V_divcbinfo);
293 	LIST_FOREACH(inp, &V_divcb, inp_list) {
294 		/* XXX why does only one socket match? */
295 		if (inp->inp_lport == nport) {
296 			INP_RLOCK(inp);
297 			sa = inp->inp_socket;
298 			SOCKBUF_LOCK(&sa->so_rcv);
299 			if (sbappendaddr_locked(&sa->so_rcv,
300 			    (struct sockaddr *)&divsrc, m,
301 			    (struct mbuf *)0) == 0) {
302 				SOCKBUF_UNLOCK(&sa->so_rcv);
303 				sa = NULL;	/* force mbuf reclaim below */
304 			} else
305 				sorwakeup_locked(sa);
306 			INP_RUNLOCK(inp);
307 			break;
308 		}
309 	}
310 	INP_INFO_RUNLOCK(&V_divcbinfo);
311 	if (sa == NULL) {
312 		m_freem(m);
313 		KMOD_IPSTAT_INC(ips_noproto);
314 		KMOD_IPSTAT_DEC(ips_delivered);
315         }
316 }
317 
318 /*
319  * Deliver packet back into the IP processing machinery.
320  *
321  * If no address specified, or address is 0.0.0.0, send to ip_output();
322  * otherwise, send to ip_input() and mark as having been received on
323  * the interface with that address.
324  */
325 static int
326 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
327     struct mbuf *control)
328 {
329 	struct m_tag *mtag;
330 	struct ipfw_rule_ref *dt;
331 	int error = 0;
332 	struct mbuf *options;
333 
334 	/*
335 	 * An mbuf may hasn't come from userland, but we pretend
336 	 * that it has.
337 	 */
338 	m->m_pkthdr.rcvif = NULL;
339 	m->m_nextpkt = NULL;
340 	M_SETFIB(m, so->so_fibnum);
341 
342 	if (control)
343 		m_freem(control);		/* XXX */
344 
345 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
346 	if (mtag == NULL) {
347 		/* this should be normal */
348 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
349 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
350 		if (mtag == NULL) {
351 			error = ENOBUFS;
352 			goto cantsend;
353 		}
354 		m_tag_prepend(m, mtag);
355 	}
356 	dt = (struct ipfw_rule_ref *)(mtag+1);
357 
358 	/* Loopback avoidance and state recovery */
359 	if (sin) {
360 		int i;
361 
362 		/* set the starting point. We provide a non-zero slot,
363 		 * but a non_matching chain_id to skip that info and use
364 		 * the rulenum/rule_id.
365 		 */
366 		dt->slot = 1; /* dummy, chain_id is invalid */
367 		dt->chain_id = 0;
368 		dt->rulenum = sin->sin_port+1; /* host format ? */
369 		dt->rule_id = 0;
370 		/*
371 		 * Find receive interface with the given name, stuffed
372 		 * (if it exists) in the sin_zero[] field.
373 		 * The name is user supplied data so don't trust its size
374 		 * or that it is zero terminated.
375 		 */
376 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
377 			;
378 		if ( i > 0 && i < sizeof(sin->sin_zero))
379 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
380 	}
381 
382 	/* Reinject packet into the system as incoming or outgoing */
383 	if (!sin || sin->sin_addr.s_addr == 0) {
384 		struct ip *const ip = mtod(m, struct ip *);
385 		struct inpcb *inp;
386 
387 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
388 		INP_INFO_WLOCK(&V_divcbinfo);
389 		inp = sotoinpcb(so);
390 		INP_RLOCK(inp);
391 		/*
392 		 * Don't allow both user specified and setsockopt options,
393 		 * and don't allow packet length sizes that will crash
394 		 */
395 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
396 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
397 			error = EINVAL;
398 			INP_RUNLOCK(inp);
399 			INP_INFO_WUNLOCK(&V_divcbinfo);
400 			m_freem(m);
401 		} else {
402 			/* Convert fields to host order for ip_output() */
403 			ip->ip_len = ntohs(ip->ip_len);
404 			ip->ip_off = ntohs(ip->ip_off);
405 
406 			/* Send packet to output processing */
407 			KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
408 
409 #ifdef MAC
410 			mac_inpcb_create_mbuf(inp, m);
411 #endif
412 			/*
413 			 * Get ready to inject the packet into ip_output().
414 			 * Just in case socket options were specified on the
415 			 * divert socket, we duplicate them.  This is done
416 			 * to avoid having to hold the PCB locks over the call
417 			 * to ip_output(), as doing this results in a number of
418 			 * lock ordering complexities.
419 			 *
420 			 * Note that we set the multicast options argument for
421 			 * ip_output() to NULL since it should be invariant that
422 			 * they are not present.
423 			 */
424 			KASSERT(inp->inp_moptions == NULL,
425 			    ("multicast options set on a divert socket"));
426 			options = NULL;
427 			/*
428 			 * XXXCSJP: It is unclear to me whether or not it makes
429 			 * sense for divert sockets to have options.  However,
430 			 * for now we will duplicate them with the INP locks
431 			 * held so we can use them in ip_output() without
432 			 * requring a reference to the pcb.
433 			 */
434 			if (inp->inp_options != NULL) {
435 				options = m_dup(inp->inp_options, M_DONTWAIT);
436 				if (options == NULL)
437 					error = ENOBUFS;
438 			}
439 			INP_RUNLOCK(inp);
440 			INP_INFO_WUNLOCK(&V_divcbinfo);
441 			if (error == ENOBUFS) {
442 				m_freem(m);
443 				return (error);
444 			}
445 			error = ip_output(m, options, NULL,
446 			    ((so->so_options & SO_DONTROUTE) ?
447 			    IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
448 			    IP_RAWOUTPUT, NULL, NULL);
449 			if (options != NULL)
450 				m_freem(options);
451 		}
452 	} else {
453 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
454 		if (m->m_pkthdr.rcvif == NULL) {
455 			/*
456 			 * No luck with the name, check by IP address.
457 			 * Clear the port and the ifname to make sure
458 			 * there are no distractions for ifa_ifwithaddr.
459 			 */
460 			struct	ifaddr *ifa;
461 
462 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
463 			sin->sin_port = 0;
464 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
465 			if (ifa == NULL) {
466 				error = EADDRNOTAVAIL;
467 				goto cantsend;
468 			}
469 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
470 			ifa_free(ifa);
471 		}
472 #ifdef MAC
473 		mac_socket_create_mbuf(so, m);
474 #endif
475 		/* Send packet to input processing via netisr */
476 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
477 	}
478 
479 	return error;
480 
481 cantsend:
482 	m_freem(m);
483 	return error;
484 }
485 
486 static int
487 div_attach(struct socket *so, int proto, struct thread *td)
488 {
489 	struct inpcb *inp;
490 	int error;
491 
492 	inp  = sotoinpcb(so);
493 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
494 	if (td != NULL) {
495 		error = priv_check(td, PRIV_NETINET_DIVERT);
496 		if (error)
497 			return (error);
498 	}
499 	error = soreserve(so, div_sendspace, div_recvspace);
500 	if (error)
501 		return error;
502 	INP_INFO_WLOCK(&V_divcbinfo);
503 	error = in_pcballoc(so, &V_divcbinfo);
504 	if (error) {
505 		INP_INFO_WUNLOCK(&V_divcbinfo);
506 		return error;
507 	}
508 	inp = (struct inpcb *)so->so_pcb;
509 	INP_INFO_WUNLOCK(&V_divcbinfo);
510 	inp->inp_ip_p = proto;
511 	inp->inp_vflag |= INP_IPV4;
512 	inp->inp_flags |= INP_HDRINCL;
513 	INP_WUNLOCK(inp);
514 	return 0;
515 }
516 
517 static void
518 div_detach(struct socket *so)
519 {
520 	struct inpcb *inp;
521 
522 	inp = sotoinpcb(so);
523 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
524 	INP_INFO_WLOCK(&V_divcbinfo);
525 	INP_WLOCK(inp);
526 	in_pcbdetach(inp);
527 	in_pcbfree(inp);
528 	INP_INFO_WUNLOCK(&V_divcbinfo);
529 }
530 
531 static int
532 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
533 {
534 	struct inpcb *inp;
535 	int error;
536 
537 	inp = sotoinpcb(so);
538 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
539 	/* in_pcbbind assumes that nam is a sockaddr_in
540 	 * and in_pcbbind requires a valid address. Since divert
541 	 * sockets don't we need to make sure the address is
542 	 * filled in properly.
543 	 * XXX -- divert should not be abusing in_pcbind
544 	 * and should probably have its own family.
545 	 */
546 	if (nam->sa_family != AF_INET)
547 		return EAFNOSUPPORT;
548 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
549 	INP_INFO_WLOCK(&V_divcbinfo);
550 	INP_WLOCK(inp);
551 	error = in_pcbbind(inp, nam, td->td_ucred);
552 	INP_WUNLOCK(inp);
553 	INP_INFO_WUNLOCK(&V_divcbinfo);
554 	return error;
555 }
556 
557 static int
558 div_shutdown(struct socket *so)
559 {
560 	struct inpcb *inp;
561 
562 	inp = sotoinpcb(so);
563 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
564 	INP_WLOCK(inp);
565 	socantsendmore(so);
566 	INP_WUNLOCK(inp);
567 	return 0;
568 }
569 
570 static int
571 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
572     struct mbuf *control, struct thread *td)
573 {
574 
575 	/* Packet must have a header (but that's about it) */
576 	if (m->m_len < sizeof (struct ip) &&
577 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
578 		KMOD_IPSTAT_INC(ips_toosmall);
579 		m_freem(m);
580 		return EINVAL;
581 	}
582 
583 	/* Send packet */
584 	return div_output(so, m, (struct sockaddr_in *)nam, control);
585 }
586 
587 static void
588 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
589 {
590         struct in_addr faddr;
591 
592 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
593 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
594         	return;
595 	if (PRC_IS_REDIRECT(cmd))
596 		return;
597 }
598 
599 static int
600 div_pcblist(SYSCTL_HANDLER_ARGS)
601 {
602 	int error, i, n;
603 	struct inpcb *inp, **inp_list;
604 	inp_gen_t gencnt;
605 	struct xinpgen xig;
606 
607 	/*
608 	 * The process of preparing the TCB list is too time-consuming and
609 	 * resource-intensive to repeat twice on every request.
610 	 */
611 	if (req->oldptr == 0) {
612 		n = V_divcbinfo.ipi_count;
613 		req->oldidx = 2 * (sizeof xig)
614 			+ (n + n/8) * sizeof(struct xinpcb);
615 		return 0;
616 	}
617 
618 	if (req->newptr != 0)
619 		return EPERM;
620 
621 	/*
622 	 * OK, now we're committed to doing something.
623 	 */
624 	INP_INFO_RLOCK(&V_divcbinfo);
625 	gencnt = V_divcbinfo.ipi_gencnt;
626 	n = V_divcbinfo.ipi_count;
627 	INP_INFO_RUNLOCK(&V_divcbinfo);
628 
629 	error = sysctl_wire_old_buffer(req,
630 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
631 	if (error != 0)
632 		return (error);
633 
634 	xig.xig_len = sizeof xig;
635 	xig.xig_count = n;
636 	xig.xig_gen = gencnt;
637 	xig.xig_sogen = so_gencnt;
638 	error = SYSCTL_OUT(req, &xig, sizeof xig);
639 	if (error)
640 		return error;
641 
642 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
643 	if (inp_list == 0)
644 		return ENOMEM;
645 
646 	INP_INFO_RLOCK(&V_divcbinfo);
647 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
648 	     inp = LIST_NEXT(inp, inp_list)) {
649 		INP_RLOCK(inp);
650 		if (inp->inp_gencnt <= gencnt &&
651 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
652 			inp_list[i++] = inp;
653 		INP_RUNLOCK(inp);
654 	}
655 	INP_INFO_RUNLOCK(&V_divcbinfo);
656 	n = i;
657 
658 	error = 0;
659 	for (i = 0; i < n; i++) {
660 		inp = inp_list[i];
661 		INP_RLOCK(inp);
662 		if (inp->inp_gencnt <= gencnt) {
663 			struct xinpcb xi;
664 			bzero(&xi, sizeof(xi));
665 			xi.xi_len = sizeof xi;
666 			/* XXX should avoid extra copy */
667 			bcopy(inp, &xi.xi_inp, sizeof *inp);
668 			if (inp->inp_socket)
669 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
670 			INP_RUNLOCK(inp);
671 			error = SYSCTL_OUT(req, &xi, sizeof xi);
672 		} else
673 			INP_RUNLOCK(inp);
674 	}
675 	if (!error) {
676 		/*
677 		 * Give the user an updated idea of our state.
678 		 * If the generation differs from what we told
679 		 * her before, she knows that something happened
680 		 * while we were processing this request, and it
681 		 * might be necessary to retry.
682 		 */
683 		INP_INFO_RLOCK(&V_divcbinfo);
684 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
685 		xig.xig_sogen = so_gencnt;
686 		xig.xig_count = V_divcbinfo.ipi_count;
687 		INP_INFO_RUNLOCK(&V_divcbinfo);
688 		error = SYSCTL_OUT(req, &xig, sizeof xig);
689 	}
690 	free(inp_list, M_TEMP);
691 	return error;
692 }
693 
694 #ifdef SYSCTL_NODE
695 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
696 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
697 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
698 #endif
699 
700 struct pr_usrreqs div_usrreqs = {
701 	.pru_attach =		div_attach,
702 	.pru_bind =		div_bind,
703 	.pru_control =		in_control,
704 	.pru_detach =		div_detach,
705 	.pru_peeraddr =		in_getpeeraddr,
706 	.pru_send =		div_send,
707 	.pru_shutdown =		div_shutdown,
708 	.pru_sockaddr =		in_getsockaddr,
709 	.pru_sosetlabel =	in_pcbsosetlabel
710 };
711 
712 struct protosw div_protosw = {
713 	.pr_type =		SOCK_RAW,
714 	.pr_protocol =		IPPROTO_DIVERT,
715 	.pr_flags =		PR_ATOMIC|PR_ADDR,
716 	.pr_input =		div_input,
717 	.pr_ctlinput =		div_ctlinput,
718 	.pr_ctloutput =		ip_ctloutput,
719 	.pr_init =		div_init,
720 #ifdef VIMAGE
721 	.pr_destroy =		div_destroy,
722 #endif
723 	.pr_usrreqs =		&div_usrreqs
724 };
725 
726 static int
727 div_modevent(module_t mod, int type, void *unused)
728 {
729 	int err = 0;
730 #ifndef VIMAGE
731 	int n;
732 #endif
733 
734 	switch (type) {
735 	case MOD_LOAD:
736 		/*
737 		 * Protocol will be initialized by pf_proto_register().
738 		 * We don't have to register ip_protox because we are not
739 		 * a true IP protocol that goes over the wire.
740 		 */
741 		err = pf_proto_register(PF_INET, &div_protosw);
742 		if (err != 0)
743 			return (err);
744 		ip_divert_ptr = divert_packet;
745 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
746 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
747 		break;
748 	case MOD_QUIESCE:
749 		/*
750 		 * IPDIVERT may normally not be unloaded because of the
751 		 * potential race conditions.  Tell kldunload we can't be
752 		 * unloaded unless the unload is forced.
753 		 */
754 		err = EPERM;
755 		break;
756 	case MOD_UNLOAD:
757 #ifdef VIMAGE
758 		err = EPERM;
759 		break;
760 #else
761 		/*
762 		 * Forced unload.
763 		 *
764 		 * Module ipdivert can only be unloaded if no sockets are
765 		 * connected.  Maybe this can be changed later to forcefully
766 		 * disconnect any open sockets.
767 		 *
768 		 * XXXRW: Note that there is a slight race here, as a new
769 		 * socket open request could be spinning on the lock and then
770 		 * we destroy the lock.
771 		 */
772 		INP_INFO_WLOCK(&V_divcbinfo);
773 		n = V_divcbinfo.ipi_count;
774 		if (n != 0) {
775 			err = EBUSY;
776 			INP_INFO_WUNLOCK(&V_divcbinfo);
777 			break;
778 		}
779 		ip_divert_ptr = NULL;
780 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
781 		INP_INFO_WUNLOCK(&V_divcbinfo);
782 		div_destroy();
783 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
784 		break;
785 #endif /* !VIMAGE */
786 	default:
787 		err = EOPNOTSUPP;
788 		break;
789 	}
790 	return err;
791 }
792 
793 static moduledata_t ipdivertmod = {
794         "ipdivert",
795         div_modevent,
796         0
797 };
798 
799 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
800 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
801 MODULE_VERSION(ipdivert, 1);
802