xref: /freebsd/sys/netinet/ip_divert.c (revision 586f63035fbe5e45cfc971037fd76375661ece26)
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 #include "opt_inet6.h"
41 
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/module.h>
48 #include <sys/kernel.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/protosw.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/sysctl.h>
55 #include <net/vnet.h>
56 
57 #include <net/if.h>
58 #include <net/netisr.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip_var.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/ip6_var.h>
69 #endif
70 #ifdef SCTP
71 #include <netinet/sctp_crc32.h>
72 #endif
73 
74 #include <security/mac/mac_framework.h>
75 
76 /*
77  * Divert sockets
78  */
79 
80 /*
81  * Allocate enough space to hold a full IP packet
82  */
83 #define	DIVSNDQ		(65536 + 100)
84 #define	DIVRCVQ		(65536 + 100)
85 
86 /*
87  * Divert sockets work in conjunction with ipfw or other packet filters,
88  * see the divert(4) manpage for features.
89  * Packets are selected by the packet filter and tagged with an
90  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
91  * the packet filter) and information on the matching filter rule for
92  * subsequent reinjection. The divert_port is used to put the packet
93  * on the corresponding divert socket, while the rule number is passed
94  * up (at least partially) as the sin_port in the struct sockaddr.
95  *
96  * Packets written to the divert socket carry in sin_addr a
97  * destination address, and in sin_port the number of the filter rule
98  * after which to continue processing.
99  * If the destination address is INADDR_ANY, the packet is treated as
100  * as outgoing and sent to ip_output(); otherwise it is treated as
101  * incoming and sent to ip_input().
102  * Further, sin_zero carries some information on the interface,
103  * which can be used in the reinject -- see comments in the code.
104  *
105  * On reinjection, processing in ip_input() and ip_output()
106  * will be exactly the same as for the original packet, except that
107  * packet filter processing will start at the rule number after the one
108  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
109  * will apply the entire ruleset to the packet).
110  */
111 
112 /* Internal variables. */
113 static VNET_DEFINE(struct inpcbhead, divcb);
114 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
115 
116 #define	V_divcb				VNET(divcb)
117 #define	V_divcbinfo			VNET(divcbinfo)
118 
119 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
120 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
121 
122 static eventhandler_tag ip_divert_event_tag;
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 static void
152 div_init(void)
153 {
154 
155 	/*
156 	 * XXX We don't use the hash list for divert IP, but it's easier to
157 	 * allocate one-entry hash lists than it is to check all over the
158 	 * place for hashbase == NULL.
159 	 */
160 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
161 	    div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE,
162 	    IPI_HASHFIELDS_NONE);
163 }
164 
165 static void
166 div_destroy(void)
167 {
168 
169 	in_pcbinfo_destroy(&V_divcbinfo);
170 }
171 
172 /*
173  * IPPROTO_DIVERT is not in the real IP protocol number space; this
174  * function should never be called.  Just in case, drop any packets.
175  */
176 static void
177 div_input(struct mbuf *m, int off)
178 {
179 
180 	KMOD_IPSTAT_INC(ips_noproto);
181 	m_freem(m);
182 }
183 
184 /*
185  * Divert a packet by passing it up to the divert socket at port 'port'.
186  *
187  * Setup generic address and protocol structures for div_input routine,
188  * then pass them along with mbuf chain.
189  */
190 static void
191 divert_packet(struct mbuf *m, int incoming)
192 {
193 	struct ip *ip;
194 	struct inpcb *inp;
195 	struct socket *sa;
196 	u_int16_t nport;
197 	struct sockaddr_in divsrc;
198 	struct m_tag *mtag;
199 
200 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
201 	if (mtag == NULL) {
202 		m_freem(m);
203 		return;
204 	}
205 	/* Assure header */
206 	if (m->m_len < sizeof(struct ip) &&
207 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
208 		return;
209 	ip = mtod(m, struct ip *);
210 
211 	/* Delayed checksums are currently not compatible with divert. */
212 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
213 		ip->ip_len = ntohs(ip->ip_len);
214 		in_delayed_cksum(m);
215 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
216 		ip->ip_len = htons(ip->ip_len);
217 	}
218 #ifdef SCTP
219 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
220 		ip->ip_len = ntohs(ip->ip_len);
221 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
222 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
223 		ip->ip_len = htons(ip->ip_len);
224 	}
225 #endif
226 	bzero(&divsrc, sizeof(divsrc));
227 	divsrc.sin_len = sizeof(divsrc);
228 	divsrc.sin_family = AF_INET;
229 	/* record matching rule, in host format */
230 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
231 	/*
232 	 * Record receive interface address, if any.
233 	 * But only for incoming packets.
234 	 */
235 	if (incoming) {
236 		struct ifaddr *ifa;
237 		struct ifnet *ifp;
238 
239 		/* Sanity check */
240 		M_ASSERTPKTHDR(m);
241 
242 		/* Find IP address for receive interface */
243 		ifp = m->m_pkthdr.rcvif;
244 		if_addr_rlock(ifp);
245 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
246 			if (ifa->ifa_addr->sa_family != AF_INET)
247 				continue;
248 			divsrc.sin_addr =
249 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
250 			break;
251 		}
252 		if_addr_runlock(ifp);
253 	}
254 	/*
255 	 * Record the incoming interface name whenever we have one.
256 	 */
257 	if (m->m_pkthdr.rcvif) {
258 		/*
259 		 * Hide the actual interface name in there in the
260 		 * sin_zero array. XXX This needs to be moved to a
261 		 * different sockaddr type for divert, e.g.
262 		 * sockaddr_div with multiple fields like
263 		 * sockaddr_dl. Presently we have only 7 bytes
264 		 * but that will do for now as most interfaces
265 		 * are 4 or less + 2 or less bytes for unit.
266 		 * There is probably a faster way of doing this,
267 		 * possibly taking it from the sockaddr_dl on the iface.
268 		 * This solves the problem of a P2P link and a LAN interface
269 		 * having the same address, which can result in the wrong
270 		 * interface being assigned to the packet when fed back
271 		 * into the divert socket. Theoretically if the daemon saves
272 		 * and re-uses the sockaddr_in as suggested in the man pages,
273 		 * this iface name will come along for the ride.
274 		 * (see div_output for the other half of this.)
275 		 */
276 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
277 		    sizeof(divsrc.sin_zero));
278 	}
279 
280 	/* Put packet on socket queue, if any */
281 	sa = NULL;
282 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
283 	INP_INFO_RLOCK(&V_divcbinfo);
284 	LIST_FOREACH(inp, &V_divcb, inp_list) {
285 		/* XXX why does only one socket match? */
286 		if (inp->inp_lport == nport) {
287 			INP_RLOCK(inp);
288 			sa = inp->inp_socket;
289 			SOCKBUF_LOCK(&sa->so_rcv);
290 			if (sbappendaddr_locked(&sa->so_rcv,
291 			    (struct sockaddr *)&divsrc, m,
292 			    (struct mbuf *)0) == 0) {
293 				SOCKBUF_UNLOCK(&sa->so_rcv);
294 				sa = NULL;	/* force mbuf reclaim below */
295 			} else
296 				sorwakeup_locked(sa);
297 			INP_RUNLOCK(inp);
298 			break;
299 		}
300 	}
301 	INP_INFO_RUNLOCK(&V_divcbinfo);
302 	if (sa == NULL) {
303 		m_freem(m);
304 		KMOD_IPSTAT_INC(ips_noproto);
305 		KMOD_IPSTAT_DEC(ips_delivered);
306         }
307 }
308 
309 /*
310  * Deliver packet back into the IP processing machinery.
311  *
312  * If no address specified, or address is 0.0.0.0, send to ip_output();
313  * otherwise, send to ip_input() and mark as having been received on
314  * the interface with that address.
315  */
316 static int
317 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
318     struct mbuf *control)
319 {
320 	struct ip *const ip = mtod(m, struct ip *);
321 	struct m_tag *mtag;
322 	struct ipfw_rule_ref *dt;
323 	int error = 0;
324 
325 	/*
326 	 * An mbuf may hasn't come from userland, but we pretend
327 	 * that it has.
328 	 */
329 	m->m_pkthdr.rcvif = NULL;
330 	m->m_nextpkt = NULL;
331 	M_SETFIB(m, so->so_fibnum);
332 
333 	if (control)
334 		m_freem(control);		/* XXX */
335 
336 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
337 	if (mtag == NULL) {
338 		/* this should be normal */
339 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
340 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
341 		if (mtag == NULL) {
342 			error = ENOBUFS;
343 			goto cantsend;
344 		}
345 		m_tag_prepend(m, mtag);
346 	}
347 	dt = (struct ipfw_rule_ref *)(mtag+1);
348 
349 	/* Loopback avoidance and state recovery */
350 	if (sin) {
351 		int i;
352 
353 		/* set the starting point. We provide a non-zero slot,
354 		 * but a non_matching chain_id to skip that info and use
355 		 * the rulenum/rule_id.
356 		 */
357 		dt->slot = 1; /* dummy, chain_id is invalid */
358 		dt->chain_id = 0;
359 		dt->rulenum = sin->sin_port+1; /* host format ? */
360 		dt->rule_id = 0;
361 		/*
362 		 * Find receive interface with the given name, stuffed
363 		 * (if it exists) in the sin_zero[] field.
364 		 * The name is user supplied data so don't trust its size
365 		 * or that it is zero terminated.
366 		 */
367 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
368 			;
369 		if ( i > 0 && i < sizeof(sin->sin_zero))
370 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
371 	}
372 
373 	/* Reinject packet into the system as incoming or outgoing */
374 	if (!sin || sin->sin_addr.s_addr == 0) {
375 		struct mbuf *options = NULL;
376 		struct inpcb *inp;
377 
378 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
379 		inp = sotoinpcb(so);
380 		INP_RLOCK(inp);
381 		switch (ip->ip_v) {
382 		case IPVERSION:
383 			/*
384 			 * Don't allow both user specified and setsockopt
385 			 * options, and don't allow packet length sizes that
386 			 * will crash.
387 			 */
388 			if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
389 			    inp->inp_options != NULL) ||
390 			    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
391 				error = EINVAL;
392 				INP_RUNLOCK(inp);
393 				goto cantsend;
394 			}
395 
396 			/* Convert fields to host order for ip_output() */
397 			ip->ip_len = ntohs(ip->ip_len);
398 			ip->ip_off = ntohs(ip->ip_off);
399 			break;
400 #ifdef INET6
401 		case IPV6_VERSION >> 4:
402 		    {
403 			struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
404 
405 			/* Don't allow packet length sizes that will crash */
406 			if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
407 				error = EINVAL;
408 				INP_RUNLOCK(inp);
409 				goto cantsend;
410 			}
411 
412 			ip6->ip6_plen = ntohs(ip6->ip6_plen);
413 			break;
414 		    }
415 #endif
416 		default:
417 			error = EINVAL;
418 			INP_RUNLOCK(inp);
419 			goto cantsend;
420 		}
421 
422 		/* Send packet to output processing */
423 		KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
424 
425 #ifdef MAC
426 		mac_inpcb_create_mbuf(inp, m);
427 #endif
428 		/*
429 		 * Get ready to inject the packet into ip_output().
430 		 * Just in case socket options were specified on the
431 		 * divert socket, we duplicate them.  This is done
432 		 * to avoid having to hold the PCB locks over the call
433 		 * to ip_output(), as doing this results in a number of
434 		 * lock ordering complexities.
435 		 *
436 		 * Note that we set the multicast options argument for
437 		 * ip_output() to NULL since it should be invariant that
438 		 * they are not present.
439 		 */
440 		KASSERT(inp->inp_moptions == NULL,
441 		    ("multicast options set on a divert socket"));
442 		/*
443 		 * XXXCSJP: It is unclear to me whether or not it makes
444 		 * sense for divert sockets to have options.  However,
445 		 * for now we will duplicate them with the INP locks
446 		 * held so we can use them in ip_output() without
447 		 * requring a reference to the pcb.
448 		 */
449 		if (inp->inp_options != NULL) {
450 			options = m_dup(inp->inp_options, M_NOWAIT);
451 			if (options == NULL) {
452 				INP_RUNLOCK(inp);
453 				error = ENOBUFS;
454 				goto cantsend;
455 			}
456 		}
457 		INP_RUNLOCK(inp);
458 
459 		switch (ip->ip_v) {
460 		case IPVERSION:
461 			error = ip_output(m, options, NULL,
462 			    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
463 			    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
464 			break;
465 #ifdef INET6
466 		case IPV6_VERSION >> 4:
467 			error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
468 			break;
469 #endif
470 		}
471 		if (options != NULL)
472 			m_freem(options);
473 	} else {
474 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
475 		if (m->m_pkthdr.rcvif == NULL) {
476 			/*
477 			 * No luck with the name, check by IP address.
478 			 * Clear the port and the ifname to make sure
479 			 * there are no distractions for ifa_ifwithaddr.
480 			 */
481 			struct	ifaddr *ifa;
482 
483 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
484 			sin->sin_port = 0;
485 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
486 			if (ifa == NULL) {
487 				error = EADDRNOTAVAIL;
488 				goto cantsend;
489 			}
490 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
491 			ifa_free(ifa);
492 		}
493 #ifdef MAC
494 		mac_socket_create_mbuf(so, m);
495 #endif
496 		/* Send packet to input processing via netisr */
497 		switch (ip->ip_v) {
498 		case IPVERSION:
499 			netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
500 			break;
501 #ifdef INET6
502 		case IPV6_VERSION >> 4:
503 			netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
504 			break;
505 #endif
506 		default:
507 			error = EINVAL;
508 			goto cantsend;
509 		}
510 	}
511 
512 	return (error);
513 
514 cantsend:
515 	m_freem(m);
516 	return (error);
517 }
518 
519 static int
520 div_attach(struct socket *so, int proto, struct thread *td)
521 {
522 	struct inpcb *inp;
523 	int error;
524 
525 	inp  = sotoinpcb(so);
526 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
527 	if (td != NULL) {
528 		error = priv_check(td, PRIV_NETINET_DIVERT);
529 		if (error)
530 			return (error);
531 	}
532 	error = soreserve(so, div_sendspace, div_recvspace);
533 	if (error)
534 		return error;
535 	INP_INFO_WLOCK(&V_divcbinfo);
536 	error = in_pcballoc(so, &V_divcbinfo);
537 	if (error) {
538 		INP_INFO_WUNLOCK(&V_divcbinfo);
539 		return error;
540 	}
541 	inp = (struct inpcb *)so->so_pcb;
542 	INP_INFO_WUNLOCK(&V_divcbinfo);
543 	inp->inp_ip_p = proto;
544 	inp->inp_vflag |= INP_IPV4;
545 	inp->inp_flags |= INP_HDRINCL;
546 	INP_WUNLOCK(inp);
547 	return 0;
548 }
549 
550 static void
551 div_detach(struct socket *so)
552 {
553 	struct inpcb *inp;
554 
555 	inp = sotoinpcb(so);
556 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
557 	INP_INFO_WLOCK(&V_divcbinfo);
558 	INP_WLOCK(inp);
559 	in_pcbdetach(inp);
560 	in_pcbfree(inp);
561 	INP_INFO_WUNLOCK(&V_divcbinfo);
562 }
563 
564 static int
565 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
566 {
567 	struct inpcb *inp;
568 	int error;
569 
570 	inp = sotoinpcb(so);
571 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
572 	/* in_pcbbind assumes that nam is a sockaddr_in
573 	 * and in_pcbbind requires a valid address. Since divert
574 	 * sockets don't we need to make sure the address is
575 	 * filled in properly.
576 	 * XXX -- divert should not be abusing in_pcbind
577 	 * and should probably have its own family.
578 	 */
579 	if (nam->sa_family != AF_INET)
580 		return EAFNOSUPPORT;
581 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
582 	INP_INFO_WLOCK(&V_divcbinfo);
583 	INP_WLOCK(inp);
584 	INP_HASH_WLOCK(&V_divcbinfo);
585 	error = in_pcbbind(inp, nam, td->td_ucred);
586 	INP_HASH_WUNLOCK(&V_divcbinfo);
587 	INP_WUNLOCK(inp);
588 	INP_INFO_WUNLOCK(&V_divcbinfo);
589 	return error;
590 }
591 
592 static int
593 div_shutdown(struct socket *so)
594 {
595 	struct inpcb *inp;
596 
597 	inp = sotoinpcb(so);
598 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
599 	INP_WLOCK(inp);
600 	socantsendmore(so);
601 	INP_WUNLOCK(inp);
602 	return 0;
603 }
604 
605 static int
606 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
607     struct mbuf *control, struct thread *td)
608 {
609 
610 	/* Packet must have a header (but that's about it) */
611 	if (m->m_len < sizeof (struct ip) &&
612 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
613 		KMOD_IPSTAT_INC(ips_toosmall);
614 		m_freem(m);
615 		return EINVAL;
616 	}
617 
618 	/* Send packet */
619 	return div_output(so, m, (struct sockaddr_in *)nam, control);
620 }
621 
622 static void
623 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
624 {
625         struct in_addr faddr;
626 
627 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
628 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
629         	return;
630 	if (PRC_IS_REDIRECT(cmd))
631 		return;
632 }
633 
634 static int
635 div_pcblist(SYSCTL_HANDLER_ARGS)
636 {
637 	int error, i, n;
638 	struct inpcb *inp, **inp_list;
639 	inp_gen_t gencnt;
640 	struct xinpgen xig;
641 
642 	/*
643 	 * The process of preparing the TCB list is too time-consuming and
644 	 * resource-intensive to repeat twice on every request.
645 	 */
646 	if (req->oldptr == 0) {
647 		n = V_divcbinfo.ipi_count;
648 		n += imax(n / 8, 10);
649 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
650 		return 0;
651 	}
652 
653 	if (req->newptr != 0)
654 		return EPERM;
655 
656 	/*
657 	 * OK, now we're committed to doing something.
658 	 */
659 	INP_INFO_RLOCK(&V_divcbinfo);
660 	gencnt = V_divcbinfo.ipi_gencnt;
661 	n = V_divcbinfo.ipi_count;
662 	INP_INFO_RUNLOCK(&V_divcbinfo);
663 
664 	error = sysctl_wire_old_buffer(req,
665 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
666 	if (error != 0)
667 		return (error);
668 
669 	xig.xig_len = sizeof xig;
670 	xig.xig_count = n;
671 	xig.xig_gen = gencnt;
672 	xig.xig_sogen = so_gencnt;
673 	error = SYSCTL_OUT(req, &xig, sizeof xig);
674 	if (error)
675 		return error;
676 
677 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
678 	if (inp_list == 0)
679 		return ENOMEM;
680 
681 	INP_INFO_RLOCK(&V_divcbinfo);
682 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
683 	     inp = LIST_NEXT(inp, inp_list)) {
684 		INP_WLOCK(inp);
685 		if (inp->inp_gencnt <= gencnt &&
686 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
687 			in_pcbref(inp);
688 			inp_list[i++] = inp;
689 		}
690 		INP_WUNLOCK(inp);
691 	}
692 	INP_INFO_RUNLOCK(&V_divcbinfo);
693 	n = i;
694 
695 	error = 0;
696 	for (i = 0; i < n; i++) {
697 		inp = inp_list[i];
698 		INP_RLOCK(inp);
699 		if (inp->inp_gencnt <= gencnt) {
700 			struct xinpcb xi;
701 			bzero(&xi, sizeof(xi));
702 			xi.xi_len = sizeof xi;
703 			/* XXX should avoid extra copy */
704 			bcopy(inp, &xi.xi_inp, sizeof *inp);
705 			if (inp->inp_socket)
706 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
707 			INP_RUNLOCK(inp);
708 			error = SYSCTL_OUT(req, &xi, sizeof xi);
709 		} else
710 			INP_RUNLOCK(inp);
711 	}
712 	INP_INFO_WLOCK(&V_divcbinfo);
713 	for (i = 0; i < n; i++) {
714 		inp = inp_list[i];
715 		INP_RLOCK(inp);
716 		if (!in_pcbrele_rlocked(inp))
717 			INP_RUNLOCK(inp);
718 	}
719 	INP_INFO_WUNLOCK(&V_divcbinfo);
720 
721 	if (!error) {
722 		/*
723 		 * Give the user an updated idea of our state.
724 		 * If the generation differs from what we told
725 		 * her before, she knows that something happened
726 		 * while we were processing this request, and it
727 		 * might be necessary to retry.
728 		 */
729 		INP_INFO_RLOCK(&V_divcbinfo);
730 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
731 		xig.xig_sogen = so_gencnt;
732 		xig.xig_count = V_divcbinfo.ipi_count;
733 		INP_INFO_RUNLOCK(&V_divcbinfo);
734 		error = SYSCTL_OUT(req, &xig, sizeof xig);
735 	}
736 	free(inp_list, M_TEMP);
737 	return error;
738 }
739 
740 #ifdef SYSCTL_NODE
741 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
742     "IPDIVERT");
743 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
744     NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
745 #endif
746 
747 struct pr_usrreqs div_usrreqs = {
748 	.pru_attach =		div_attach,
749 	.pru_bind =		div_bind,
750 	.pru_control =		in_control,
751 	.pru_detach =		div_detach,
752 	.pru_peeraddr =		in_getpeeraddr,
753 	.pru_send =		div_send,
754 	.pru_shutdown =		div_shutdown,
755 	.pru_sockaddr =		in_getsockaddr,
756 	.pru_sosetlabel =	in_pcbsosetlabel
757 };
758 
759 struct protosw div_protosw = {
760 	.pr_type =		SOCK_RAW,
761 	.pr_protocol =		IPPROTO_DIVERT,
762 	.pr_flags =		PR_ATOMIC|PR_ADDR,
763 	.pr_input =		div_input,
764 	.pr_ctlinput =		div_ctlinput,
765 	.pr_ctloutput =		ip_ctloutput,
766 	.pr_init =		div_init,
767 #ifdef VIMAGE
768 	.pr_destroy =		div_destroy,
769 #endif
770 	.pr_usrreqs =		&div_usrreqs
771 };
772 
773 static int
774 div_modevent(module_t mod, int type, void *unused)
775 {
776 	int err = 0;
777 #ifndef VIMAGE
778 	int n;
779 #endif
780 
781 	switch (type) {
782 	case MOD_LOAD:
783 		/*
784 		 * Protocol will be initialized by pf_proto_register().
785 		 * We don't have to register ip_protox because we are not
786 		 * a true IP protocol that goes over the wire.
787 		 */
788 		err = pf_proto_register(PF_INET, &div_protosw);
789 		if (err != 0)
790 			return (err);
791 		ip_divert_ptr = divert_packet;
792 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
793 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
794 		break;
795 	case MOD_QUIESCE:
796 		/*
797 		 * IPDIVERT may normally not be unloaded because of the
798 		 * potential race conditions.  Tell kldunload we can't be
799 		 * unloaded unless the unload is forced.
800 		 */
801 		err = EPERM;
802 		break;
803 	case MOD_UNLOAD:
804 #ifdef VIMAGE
805 		err = EPERM;
806 		break;
807 #else
808 		/*
809 		 * Forced unload.
810 		 *
811 		 * Module ipdivert can only be unloaded if no sockets are
812 		 * connected.  Maybe this can be changed later to forcefully
813 		 * disconnect any open sockets.
814 		 *
815 		 * XXXRW: Note that there is a slight race here, as a new
816 		 * socket open request could be spinning on the lock and then
817 		 * we destroy the lock.
818 		 */
819 		INP_INFO_WLOCK(&V_divcbinfo);
820 		n = V_divcbinfo.ipi_count;
821 		if (n != 0) {
822 			err = EBUSY;
823 			INP_INFO_WUNLOCK(&V_divcbinfo);
824 			break;
825 		}
826 		ip_divert_ptr = NULL;
827 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
828 		INP_INFO_WUNLOCK(&V_divcbinfo);
829 		div_destroy();
830 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
831 		break;
832 #endif /* !VIMAGE */
833 	default:
834 		err = EOPNOTSUPP;
835 		break;
836 	}
837 	return err;
838 }
839 
840 static moduledata_t ipdivertmod = {
841         "ipdivert",
842         div_modevent,
843         0
844 };
845 
846 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
847 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
848 MODULE_VERSION(ipdivert, 1);
849