xref: /freebsd/sys/netinet/ip_divert.c (revision cbd30a72ca196976c1c700400ecd424baa1b9c16)
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  * 3. 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 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_sctp.h"
36 #ifndef INET
37 #error "IPDIVERT requires INET"
38 #endif
39 
40 #include <sys/param.h>
41 #include <sys/eventhandler.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/if_var.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, 0, IPI_HASHFIELDS_NONE);
162 }
163 
164 static void
165 div_destroy(void *unused __unused)
166 {
167 
168 	in_pcbinfo_destroy(&V_divcbinfo);
169 }
170 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
171     div_destroy, NULL);
172 
173 /*
174  * IPPROTO_DIVERT is not in the real IP protocol number space; this
175  * function should never be called.  Just in case, drop any packets.
176  */
177 static int
178 div_input(struct mbuf **mp, int *offp, int proto)
179 {
180 	struct mbuf *m = *mp;
181 
182 	KMOD_IPSTAT_INC(ips_noproto);
183 	m_freem(m);
184 	return (IPPROTO_DONE);
185 }
186 
187 /*
188  * Divert a packet by passing it up to the divert socket at port 'port'.
189  *
190  * Setup generic address and protocol structures for div_input routine,
191  * then pass them along with mbuf chain.
192  */
193 static void
194 divert_packet(struct mbuf *m, int incoming)
195 {
196 	struct ip *ip;
197 	struct inpcb *inp;
198 	struct socket *sa;
199 	u_int16_t nport;
200 	struct sockaddr_in divsrc;
201 	struct m_tag *mtag;
202 
203 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
204 	if (mtag == NULL) {
205 		m_freem(m);
206 		return;
207 	}
208 	/* Assure header */
209 	if (m->m_len < sizeof(struct ip) &&
210 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
211 		return;
212 	ip = mtod(m, struct ip *);
213 
214 	/* Delayed checksums are currently not compatible with divert. */
215 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
216 		in_delayed_cksum(m);
217 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
218 	}
219 #ifdef SCTP
220 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
221 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
222 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
223 	}
224 #endif
225 	bzero(&divsrc, sizeof(divsrc));
226 	divsrc.sin_len = sizeof(divsrc);
227 	divsrc.sin_family = AF_INET;
228 	/* record matching rule, in host format */
229 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
230 	/*
231 	 * Record receive interface address, if any.
232 	 * But only for incoming packets.
233 	 */
234 	if (incoming) {
235 		struct ifaddr *ifa;
236 		struct ifnet *ifp;
237 
238 		/* Sanity check */
239 		M_ASSERTPKTHDR(m);
240 
241 		/* Find IP address for receive interface */
242 		ifp = m->m_pkthdr.rcvif;
243 		if_addr_rlock(ifp);
244 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
245 			if (ifa->ifa_addr->sa_family != AF_INET)
246 				continue;
247 			divsrc.sin_addr =
248 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
249 			break;
250 		}
251 		if_addr_runlock(ifp);
252 	}
253 	/*
254 	 * Record the incoming interface name whenever we have one.
255 	 */
256 	if (m->m_pkthdr.rcvif) {
257 		/*
258 		 * Hide the actual interface name in there in the
259 		 * sin_zero array. XXX This needs to be moved to a
260 		 * different sockaddr type for divert, e.g.
261 		 * sockaddr_div with multiple fields like
262 		 * sockaddr_dl. Presently we have only 7 bytes
263 		 * but that will do for now as most interfaces
264 		 * are 4 or less + 2 or less bytes for unit.
265 		 * There is probably a faster way of doing this,
266 		 * possibly taking it from the sockaddr_dl on the iface.
267 		 * This solves the problem of a P2P link and a LAN interface
268 		 * having the same address, which can result in the wrong
269 		 * interface being assigned to the packet when fed back
270 		 * into the divert socket. Theoretically if the daemon saves
271 		 * and re-uses the sockaddr_in as suggested in the man pages,
272 		 * this iface name will come along for the ride.
273 		 * (see div_output for the other half of this.)
274 		 */
275 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
276 		    sizeof(divsrc.sin_zero));
277 	}
278 
279 	/* Put packet on socket queue, if any */
280 	sa = NULL;
281 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
282 	INP_INFO_RLOCK(&V_divcbinfo);
283 	LIST_FOREACH(inp, &V_divcb, inp_list) {
284 		/* XXX why does only one socket match? */
285 		if (inp->inp_lport == nport) {
286 			INP_RLOCK(inp);
287 			sa = inp->inp_socket;
288 			SOCKBUF_LOCK(&sa->so_rcv);
289 			if (sbappendaddr_locked(&sa->so_rcv,
290 			    (struct sockaddr *)&divsrc, m,
291 			    (struct mbuf *)0) == 0) {
292 				SOCKBUF_UNLOCK(&sa->so_rcv);
293 				sa = NULL;	/* force mbuf reclaim below */
294 			} else
295 				sorwakeup_locked(sa);
296 			INP_RUNLOCK(inp);
297 			break;
298 		}
299 	}
300 	INP_INFO_RUNLOCK(&V_divcbinfo);
301 	if (sa == NULL) {
302 		m_freem(m);
303 		KMOD_IPSTAT_INC(ips_noproto);
304 		KMOD_IPSTAT_DEC(ips_delivered);
305         }
306 }
307 
308 /*
309  * Deliver packet back into the IP processing machinery.
310  *
311  * If no address specified, or address is 0.0.0.0, send to ip_output();
312  * otherwise, send to ip_input() and mark as having been received on
313  * the interface with that address.
314  */
315 static int
316 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
317     struct mbuf *control)
318 {
319 	struct ip *const ip = mtod(m, struct ip *);
320 	struct m_tag *mtag;
321 	struct ipfw_rule_ref *dt;
322 	int error = 0;
323 
324 	/*
325 	 * An mbuf may hasn't come from userland, but we pretend
326 	 * that it has.
327 	 */
328 	m->m_pkthdr.rcvif = NULL;
329 	m->m_nextpkt = NULL;
330 	M_SETFIB(m, so->so_fibnum);
331 
332 	if (control)
333 		m_freem(control);		/* XXX */
334 
335 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
336 	if (mtag == NULL) {
337 		/* this should be normal */
338 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
339 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
340 		if (mtag == NULL) {
341 			error = ENOBUFS;
342 			goto cantsend;
343 		}
344 		m_tag_prepend(m, mtag);
345 	}
346 	dt = (struct ipfw_rule_ref *)(mtag+1);
347 
348 	/* Loopback avoidance and state recovery */
349 	if (sin) {
350 		int i;
351 
352 		/* set the starting point. We provide a non-zero slot,
353 		 * but a non_matching chain_id to skip that info and use
354 		 * the rulenum/rule_id.
355 		 */
356 		dt->slot = 1; /* dummy, chain_id is invalid */
357 		dt->chain_id = 0;
358 		dt->rulenum = sin->sin_port+1; /* host format ? */
359 		dt->rule_id = 0;
360 		/*
361 		 * Find receive interface with the given name, stuffed
362 		 * (if it exists) in the sin_zero[] field.
363 		 * The name is user supplied data so don't trust its size
364 		 * or that it is zero terminated.
365 		 */
366 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
367 			;
368 		if ( i > 0 && i < sizeof(sin->sin_zero))
369 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
370 	}
371 
372 	/* Reinject packet into the system as incoming or outgoing */
373 	if (!sin || sin->sin_addr.s_addr == 0) {
374 		struct mbuf *options = NULL;
375 		struct inpcb *inp;
376 
377 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
378 		inp = sotoinpcb(so);
379 		INP_RLOCK(inp);
380 		switch (ip->ip_v) {
381 		case IPVERSION:
382 			/*
383 			 * Don't allow both user specified and setsockopt
384 			 * options, and don't allow packet length sizes that
385 			 * will crash.
386 			 */
387 			if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
388 			    inp->inp_options != NULL) ||
389 			    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
390 				error = EINVAL;
391 				INP_RUNLOCK(inp);
392 				goto cantsend;
393 			}
394 			break;
395 #ifdef INET6
396 		case IPV6_VERSION >> 4:
397 		    {
398 			struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
399 
400 			/* Don't allow packet length sizes that will crash */
401 			if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
402 				error = EINVAL;
403 				INP_RUNLOCK(inp);
404 				goto cantsend;
405 			}
406 			break;
407 		    }
408 #endif
409 		default:
410 			error = EINVAL;
411 			INP_RUNLOCK(inp);
412 			goto cantsend;
413 		}
414 
415 		/* Send packet to output processing */
416 		KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
417 
418 #ifdef MAC
419 		mac_inpcb_create_mbuf(inp, m);
420 #endif
421 		/*
422 		 * Get ready to inject the packet into ip_output().
423 		 * Just in case socket options were specified on the
424 		 * divert socket, we duplicate them.  This is done
425 		 * to avoid having to hold the PCB locks over the call
426 		 * to ip_output(), as doing this results in a number of
427 		 * lock ordering complexities.
428 		 *
429 		 * Note that we set the multicast options argument for
430 		 * ip_output() to NULL since it should be invariant that
431 		 * they are not present.
432 		 */
433 		KASSERT(inp->inp_moptions == NULL,
434 		    ("multicast options set on a divert socket"));
435 		/*
436 		 * XXXCSJP: It is unclear to me whether or not it makes
437 		 * sense for divert sockets to have options.  However,
438 		 * for now we will duplicate them with the INP locks
439 		 * held so we can use them in ip_output() without
440 		 * requring a reference to the pcb.
441 		 */
442 		if (inp->inp_options != NULL) {
443 			options = m_dup(inp->inp_options, M_NOWAIT);
444 			if (options == NULL) {
445 				INP_RUNLOCK(inp);
446 				error = ENOBUFS;
447 				goto cantsend;
448 			}
449 		}
450 		INP_RUNLOCK(inp);
451 
452 		switch (ip->ip_v) {
453 		case IPVERSION:
454 			error = ip_output(m, options, NULL,
455 			    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
456 			    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
457 			break;
458 #ifdef INET6
459 		case IPV6_VERSION >> 4:
460 			error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
461 			break;
462 #endif
463 		}
464 		if (options != NULL)
465 			m_freem(options);
466 	} else {
467 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
468 		if (m->m_pkthdr.rcvif == NULL) {
469 			/*
470 			 * No luck with the name, check by IP address.
471 			 * Clear the port and the ifname to make sure
472 			 * there are no distractions for ifa_ifwithaddr.
473 			 */
474 			struct	ifaddr *ifa;
475 
476 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
477 			sin->sin_port = 0;
478 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
479 			if (ifa == NULL) {
480 				error = EADDRNOTAVAIL;
481 				goto cantsend;
482 			}
483 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
484 			ifa_free(ifa);
485 		}
486 #ifdef MAC
487 		mac_socket_create_mbuf(so, m);
488 #endif
489 		/* Send packet to input processing via netisr */
490 		switch (ip->ip_v) {
491 		case IPVERSION:
492 			netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
493 			break;
494 #ifdef INET6
495 		case IPV6_VERSION >> 4:
496 			netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
497 			break;
498 #endif
499 		default:
500 			error = EINVAL;
501 			goto cantsend;
502 		}
503 	}
504 
505 	return (error);
506 
507 cantsend:
508 	m_freem(m);
509 	return (error);
510 }
511 
512 static int
513 div_attach(struct socket *so, int proto, struct thread *td)
514 {
515 	struct inpcb *inp;
516 	int error;
517 
518 	inp  = sotoinpcb(so);
519 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
520 	if (td != NULL) {
521 		error = priv_check(td, PRIV_NETINET_DIVERT);
522 		if (error)
523 			return (error);
524 	}
525 	error = soreserve(so, div_sendspace, div_recvspace);
526 	if (error)
527 		return error;
528 	INP_INFO_WLOCK(&V_divcbinfo);
529 	error = in_pcballoc(so, &V_divcbinfo);
530 	if (error) {
531 		INP_INFO_WUNLOCK(&V_divcbinfo);
532 		return error;
533 	}
534 	inp = (struct inpcb *)so->so_pcb;
535 	INP_INFO_WUNLOCK(&V_divcbinfo);
536 	inp->inp_ip_p = proto;
537 	inp->inp_vflag |= INP_IPV4;
538 	inp->inp_flags |= INP_HDRINCL;
539 	INP_WUNLOCK(inp);
540 	return 0;
541 }
542 
543 static void
544 div_detach(struct socket *so)
545 {
546 	struct inpcb *inp;
547 
548 	inp = sotoinpcb(so);
549 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
550 	INP_INFO_WLOCK(&V_divcbinfo);
551 	INP_WLOCK(inp);
552 	in_pcbdetach(inp);
553 	in_pcbfree(inp);
554 	INP_INFO_WUNLOCK(&V_divcbinfo);
555 }
556 
557 static int
558 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
559 {
560 	struct inpcb *inp;
561 	int error;
562 
563 	inp = sotoinpcb(so);
564 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
565 	/* in_pcbbind assumes that nam is a sockaddr_in
566 	 * and in_pcbbind requires a valid address. Since divert
567 	 * sockets don't we need to make sure the address is
568 	 * filled in properly.
569 	 * XXX -- divert should not be abusing in_pcbind
570 	 * and should probably have its own family.
571 	 */
572 	if (nam->sa_family != AF_INET)
573 		return EAFNOSUPPORT;
574 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
575 	INP_INFO_WLOCK(&V_divcbinfo);
576 	INP_WLOCK(inp);
577 	INP_HASH_WLOCK(&V_divcbinfo);
578 	error = in_pcbbind(inp, nam, td->td_ucred);
579 	INP_HASH_WUNLOCK(&V_divcbinfo);
580 	INP_WUNLOCK(inp);
581 	INP_INFO_WUNLOCK(&V_divcbinfo);
582 	return error;
583 }
584 
585 static int
586 div_shutdown(struct socket *so)
587 {
588 	struct inpcb *inp;
589 
590 	inp = sotoinpcb(so);
591 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
592 	INP_WLOCK(inp);
593 	socantsendmore(so);
594 	INP_WUNLOCK(inp);
595 	return 0;
596 }
597 
598 static int
599 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
600     struct mbuf *control, struct thread *td)
601 {
602 
603 	/* Packet must have a header (but that's about it) */
604 	if (m->m_len < sizeof (struct ip) &&
605 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
606 		KMOD_IPSTAT_INC(ips_toosmall);
607 		m_freem(m);
608 		return EINVAL;
609 	}
610 
611 	/* Send packet */
612 	return div_output(so, m, (struct sockaddr_in *)nam, control);
613 }
614 
615 static void
616 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
617 {
618         struct in_addr faddr;
619 
620 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
621 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
622         	return;
623 	if (PRC_IS_REDIRECT(cmd))
624 		return;
625 }
626 
627 static int
628 div_pcblist(SYSCTL_HANDLER_ARGS)
629 {
630 	int error, i, n;
631 	struct inpcb *inp, **inp_list;
632 	inp_gen_t gencnt;
633 	struct xinpgen xig;
634 
635 	/*
636 	 * The process of preparing the TCB list is too time-consuming and
637 	 * resource-intensive to repeat twice on every request.
638 	 */
639 	if (req->oldptr == 0) {
640 		n = V_divcbinfo.ipi_count;
641 		n += imax(n / 8, 10);
642 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
643 		return 0;
644 	}
645 
646 	if (req->newptr != 0)
647 		return EPERM;
648 
649 	/*
650 	 * OK, now we're committed to doing something.
651 	 */
652 	INP_INFO_RLOCK(&V_divcbinfo);
653 	gencnt = V_divcbinfo.ipi_gencnt;
654 	n = V_divcbinfo.ipi_count;
655 	INP_INFO_RUNLOCK(&V_divcbinfo);
656 
657 	error = sysctl_wire_old_buffer(req,
658 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
659 	if (error != 0)
660 		return (error);
661 
662 	xig.xig_len = sizeof xig;
663 	xig.xig_count = n;
664 	xig.xig_gen = gencnt;
665 	xig.xig_sogen = so_gencnt;
666 	error = SYSCTL_OUT(req, &xig, sizeof xig);
667 	if (error)
668 		return error;
669 
670 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
671 	if (inp_list == NULL)
672 		return ENOMEM;
673 
674 	INP_INFO_RLOCK(&V_divcbinfo);
675 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
676 	     inp = LIST_NEXT(inp, inp_list)) {
677 		INP_WLOCK(inp);
678 		if (inp->inp_gencnt <= gencnt &&
679 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
680 			in_pcbref(inp);
681 			inp_list[i++] = inp;
682 		}
683 		INP_WUNLOCK(inp);
684 	}
685 	INP_INFO_RUNLOCK(&V_divcbinfo);
686 	n = i;
687 
688 	error = 0;
689 	for (i = 0; i < n; i++) {
690 		inp = inp_list[i];
691 		INP_RLOCK(inp);
692 		if (inp->inp_gencnt <= gencnt) {
693 			struct xinpcb xi;
694 			bzero(&xi, sizeof(xi));
695 			xi.xi_len = sizeof xi;
696 			/* XXX should avoid extra copy */
697 			bcopy(inp, &xi.xi_inp, sizeof *inp);
698 			if (inp->inp_socket)
699 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
700 			INP_RUNLOCK(inp);
701 			error = SYSCTL_OUT(req, &xi, sizeof xi);
702 		} else
703 			INP_RUNLOCK(inp);
704 	}
705 	INP_INFO_WLOCK(&V_divcbinfo);
706 	for (i = 0; i < n; i++) {
707 		inp = inp_list[i];
708 		INP_RLOCK(inp);
709 		if (!in_pcbrele_rlocked(inp))
710 			INP_RUNLOCK(inp);
711 	}
712 	INP_INFO_WUNLOCK(&V_divcbinfo);
713 
714 	if (!error) {
715 		/*
716 		 * Give the user an updated idea of our state.
717 		 * If the generation differs from what we told
718 		 * her before, she knows that something happened
719 		 * while we were processing this request, and it
720 		 * might be necessary to retry.
721 		 */
722 		INP_INFO_RLOCK(&V_divcbinfo);
723 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
724 		xig.xig_sogen = so_gencnt;
725 		xig.xig_count = V_divcbinfo.ipi_count;
726 		INP_INFO_RUNLOCK(&V_divcbinfo);
727 		error = SYSCTL_OUT(req, &xig, sizeof xig);
728 	}
729 	free(inp_list, M_TEMP);
730 	return error;
731 }
732 
733 #ifdef SYSCTL_NODE
734 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
735     "IPDIVERT");
736 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
737     NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
738 #endif
739 
740 struct pr_usrreqs div_usrreqs = {
741 	.pru_attach =		div_attach,
742 	.pru_bind =		div_bind,
743 	.pru_control =		in_control,
744 	.pru_detach =		div_detach,
745 	.pru_peeraddr =		in_getpeeraddr,
746 	.pru_send =		div_send,
747 	.pru_shutdown =		div_shutdown,
748 	.pru_sockaddr =		in_getsockaddr,
749 	.pru_sosetlabel =	in_pcbsosetlabel
750 };
751 
752 struct protosw div_protosw = {
753 	.pr_type =		SOCK_RAW,
754 	.pr_protocol =		IPPROTO_DIVERT,
755 	.pr_flags =		PR_ATOMIC|PR_ADDR,
756 	.pr_input =		div_input,
757 	.pr_ctlinput =		div_ctlinput,
758 	.pr_ctloutput =		ip_ctloutput,
759 	.pr_init =		div_init,
760 	.pr_usrreqs =		&div_usrreqs
761 };
762 
763 static int
764 div_modevent(module_t mod, int type, void *unused)
765 {
766 	int err = 0;
767 
768 	switch (type) {
769 	case MOD_LOAD:
770 		/*
771 		 * Protocol will be initialized by pf_proto_register().
772 		 * We don't have to register ip_protox because we are not
773 		 * a true IP protocol that goes over the wire.
774 		 */
775 		err = pf_proto_register(PF_INET, &div_protosw);
776 		if (err != 0)
777 			return (err);
778 		ip_divert_ptr = divert_packet;
779 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
780 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
781 		break;
782 	case MOD_QUIESCE:
783 		/*
784 		 * IPDIVERT may normally not be unloaded because of the
785 		 * potential race conditions.  Tell kldunload we can't be
786 		 * unloaded unless the unload is forced.
787 		 */
788 		err = EPERM;
789 		break;
790 	case MOD_UNLOAD:
791 		/*
792 		 * Forced unload.
793 		 *
794 		 * Module ipdivert can only be unloaded if no sockets are
795 		 * connected.  Maybe this can be changed later to forcefully
796 		 * disconnect any open sockets.
797 		 *
798 		 * XXXRW: Note that there is a slight race here, as a new
799 		 * socket open request could be spinning on the lock and then
800 		 * we destroy the lock.
801 		 */
802 		INP_INFO_WLOCK(&V_divcbinfo);
803 		if (V_divcbinfo.ipi_count != 0) {
804 			err = EBUSY;
805 			INP_INFO_WUNLOCK(&V_divcbinfo);
806 			break;
807 		}
808 		ip_divert_ptr = NULL;
809 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
810 		INP_INFO_WUNLOCK(&V_divcbinfo);
811 #ifndef VIMAGE
812 		div_destroy(NULL);
813 #endif
814 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
815 		break;
816 	default:
817 		err = EOPNOTSUPP;
818 		break;
819 	}
820 	return err;
821 }
822 
823 static moduledata_t ipdivertmod = {
824         "ipdivert",
825         div_modevent,
826         0
827 };
828 
829 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
830 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
831 MODULE_VERSION(ipdivert, 1);
832