xref: /freebsd/sys/netinet/ip_divert.c (revision f157ca4696f5922275d5d451736005b9332eb136)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_sctp.h"
38 #ifndef INET
39 #error "IPDIVERT requires INET"
40 #endif
41 
42 #include <sys/param.h>
43 #include <sys/eventhandler.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/module.h>
49 #include <sys/kernel.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
56 #include <net/vnet.h>
57 
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/netisr.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_var.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #endif
72 #ifdef SCTP
73 #include <netinet/sctp_crc32.h>
74 #endif
75 
76 #include <security/mac/mac_framework.h>
77 /*
78  * Divert sockets
79  */
80 
81 /*
82  * Allocate enough space to hold a full IP packet
83  */
84 #define	DIVSNDQ		(65536 + 100)
85 #define	DIVRCVQ		(65536 + 100)
86 
87 /*
88  * Divert sockets work in conjunction with ipfw or other packet filters,
89  * see the divert(4) manpage for features.
90  * Packets are selected by the packet filter and tagged with an
91  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
92  * the packet filter) and information on the matching filter rule for
93  * subsequent reinjection. The divert_port is used to put the packet
94  * on the corresponding divert socket, while the rule number is passed
95  * up (at least partially) as the sin_port in the struct sockaddr.
96  *
97  * Packets written to the divert socket carry in sin_addr a
98  * destination address, and in sin_port the number of the filter rule
99  * after which to continue processing.
100  * If the destination address is INADDR_ANY, the packet is treated as
101  * as outgoing and sent to ip_output(); otherwise it is treated as
102  * incoming and sent to ip_input().
103  * Further, sin_zero carries some information on the interface,
104  * which can be used in the reinject -- see comments in the code.
105  *
106  * On reinjection, processing in ip_input() and ip_output()
107  * will be exactly the same as for the original packet, except that
108  * packet filter processing will start at the rule number after the one
109  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
110  * will apply the entire ruleset to the packet).
111  */
112 
113 /* Internal variables. */
114 VNET_DEFINE_STATIC(struct inpcbhead, divcb);
115 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo);
116 
117 #define	V_divcb				VNET(divcb)
118 #define	V_divcbinfo			VNET(divcbinfo)
119 
120 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
121 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
122 
123 static eventhandler_tag ip_divert_event_tag;
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_init(void)
146 {
147 
148 	/*
149 	 * XXX We don't use the hash list for divert IP, but it's easier to
150 	 * allocate one-entry hash lists than it is to check all over the
151 	 * place for hashbase == NULL.
152 	 */
153 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
154 	    div_inpcb_init, IPI_HASHFIELDS_NONE);
155 }
156 
157 static void
158 div_destroy(void *unused __unused)
159 {
160 
161 	in_pcbinfo_destroy(&V_divcbinfo);
162 }
163 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
164     div_destroy, NULL);
165 
166 /*
167  * IPPROTO_DIVERT is not in the real IP protocol number space; this
168  * function should never be called.  Just in case, drop any packets.
169  */
170 static int
171 div_input(struct mbuf **mp, int *offp, int proto)
172 {
173 	struct mbuf *m = *mp;
174 
175 	KMOD_IPSTAT_INC(ips_noproto);
176 	m_freem(m);
177 	return (IPPROTO_DONE);
178 }
179 
180 /*
181  * Divert a packet by passing it up to the divert socket at port 'port'.
182  *
183  * Setup generic address and protocol structures for div_input routine,
184  * then pass them along with mbuf chain.
185  */
186 static void
187 divert_packet(struct mbuf *m, bool incoming)
188 {
189 	struct ip *ip;
190 	struct inpcb *inp;
191 	struct socket *sa;
192 	u_int16_t nport;
193 	struct sockaddr_in divsrc;
194 	struct m_tag *mtag;
195 
196 	NET_EPOCH_ASSERT();
197 
198 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
199 	if (mtag == NULL) {
200 		m_freem(m);
201 		return;
202 	}
203 	/* Assure header */
204 	if (m->m_len < sizeof(struct ip) &&
205 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
206 		return;
207 	ip = mtod(m, struct ip *);
208 
209 	/* Delayed checksums are currently not compatible with divert. */
210 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
211 		in_delayed_cksum(m);
212 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
213 	}
214 #ifdef SCTP
215 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
216 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
217 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
218 	}
219 #endif
220 	bzero(&divsrc, sizeof(divsrc));
221 	divsrc.sin_len = sizeof(divsrc);
222 	divsrc.sin_family = AF_INET;
223 	/* record matching rule, in host format */
224 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
225 	/*
226 	 * Record receive interface address, if any.
227 	 * But only for incoming packets.
228 	 */
229 	if (incoming) {
230 		struct ifaddr *ifa;
231 		struct ifnet *ifp;
232 
233 		/* Sanity check */
234 		M_ASSERTPKTHDR(m);
235 
236 		/* Find IP address for receive interface */
237 		ifp = m->m_pkthdr.rcvif;
238 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
239 			if (ifa->ifa_addr->sa_family != AF_INET)
240 				continue;
241 			divsrc.sin_addr =
242 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
243 			break;
244 		}
245 	}
246 	/*
247 	 * Record the incoming interface name whenever we have one.
248 	 */
249 	if (m->m_pkthdr.rcvif) {
250 		/*
251 		 * Hide the actual interface name in there in the
252 		 * sin_zero array. XXX This needs to be moved to a
253 		 * different sockaddr type for divert, e.g.
254 		 * sockaddr_div with multiple fields like
255 		 * sockaddr_dl. Presently we have only 7 bytes
256 		 * but that will do for now as most interfaces
257 		 * are 4 or less + 2 or less bytes for unit.
258 		 * There is probably a faster way of doing this,
259 		 * possibly taking it from the sockaddr_dl on the iface.
260 		 * This solves the problem of a P2P link and a LAN interface
261 		 * having the same address, which can result in the wrong
262 		 * interface being assigned to the packet when fed back
263 		 * into the divert socket. Theoretically if the daemon saves
264 		 * and re-uses the sockaddr_in as suggested in the man pages,
265 		 * this iface name will come along for the ride.
266 		 * (see div_output for the other half of this.)
267 		 */
268 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
269 		    sizeof(divsrc.sin_zero));
270 	}
271 
272 	/* Put packet on socket queue, if any */
273 	sa = NULL;
274 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
275 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
276 		/* XXX why does only one socket match? */
277 		if (inp->inp_lport == nport) {
278 			INP_RLOCK(inp);
279 			sa = inp->inp_socket;
280 			SOCKBUF_LOCK(&sa->so_rcv);
281 			if (sbappendaddr_locked(&sa->so_rcv,
282 			    (struct sockaddr *)&divsrc, m,
283 			    (struct mbuf *)0) == 0) {
284 				SOCKBUF_UNLOCK(&sa->so_rcv);
285 				sa = NULL;	/* force mbuf reclaim below */
286 			} else
287 				sorwakeup_locked(sa);
288 			INP_RUNLOCK(inp);
289 			break;
290 		}
291 	}
292 	if (sa == NULL) {
293 		m_freem(m);
294 		KMOD_IPSTAT_INC(ips_noproto);
295 		KMOD_IPSTAT_DEC(ips_delivered);
296         }
297 }
298 
299 /*
300  * Deliver packet back into the IP processing machinery.
301  *
302  * If no address specified, or address is 0.0.0.0, send to ip_output();
303  * otherwise, send to ip_input() and mark as having been received on
304  * the interface with that address.
305  */
306 static int
307 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
308     struct mbuf *control)
309 {
310 	struct ip *const ip = mtod(m, struct ip *);
311 	struct m_tag *mtag;
312 	struct ipfw_rule_ref *dt;
313 	int error = 0;
314 
315 	/*
316 	 * An mbuf may hasn't come from userland, but we pretend
317 	 * that it has.
318 	 */
319 	m->m_pkthdr.rcvif = NULL;
320 	m->m_nextpkt = NULL;
321 	M_SETFIB(m, so->so_fibnum);
322 
323 	if (control)
324 		m_freem(control);		/* XXX */
325 
326 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
327 	if (mtag == NULL) {
328 		/* this should be normal */
329 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
330 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
331 		if (mtag == NULL) {
332 			error = ENOBUFS;
333 			goto cantsend;
334 		}
335 		m_tag_prepend(m, mtag);
336 	}
337 	dt = (struct ipfw_rule_ref *)(mtag+1);
338 
339 	/* Loopback avoidance and state recovery */
340 	if (sin) {
341 		int i;
342 
343 		/* set the starting point. We provide a non-zero slot,
344 		 * but a non_matching chain_id to skip that info and use
345 		 * the rulenum/rule_id.
346 		 */
347 		dt->slot = 1; /* dummy, chain_id is invalid */
348 		dt->chain_id = 0;
349 		dt->rulenum = sin->sin_port+1; /* host format ? */
350 		dt->rule_id = 0;
351 		/*
352 		 * Find receive interface with the given name, stuffed
353 		 * (if it exists) in the sin_zero[] field.
354 		 * The name is user supplied data so don't trust its size
355 		 * or that it is zero terminated.
356 		 */
357 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
358 			;
359 		if ( i > 0 && i < sizeof(sin->sin_zero))
360 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
361 	}
362 
363 	/* Reinject packet into the system as incoming or outgoing */
364 	if (!sin || sin->sin_addr.s_addr == 0) {
365 		struct mbuf *options = NULL;
366 		struct inpcb *inp;
367 
368 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
369 		inp = sotoinpcb(so);
370 		INP_RLOCK(inp);
371 		switch (ip->ip_v) {
372 		case IPVERSION:
373 			/*
374 			 * Don't allow both user specified and setsockopt
375 			 * options, and don't allow packet length sizes that
376 			 * will crash.
377 			 */
378 			if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
379 			    inp->inp_options != NULL) ||
380 			    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
381 				error = EINVAL;
382 				INP_RUNLOCK(inp);
383 				goto cantsend;
384 			}
385 			break;
386 #ifdef INET6
387 		case IPV6_VERSION >> 4:
388 		    {
389 			struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
390 
391 			/* Don't allow packet length sizes that will crash */
392 			if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
393 				error = EINVAL;
394 				INP_RUNLOCK(inp);
395 				goto cantsend;
396 			}
397 			break;
398 		    }
399 #endif
400 		default:
401 			error = EINVAL;
402 			INP_RUNLOCK(inp);
403 			goto cantsend;
404 		}
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 		/*
427 		 * XXXCSJP: It is unclear to me whether or not it makes
428 		 * sense for divert sockets to have options.  However,
429 		 * for now we will duplicate them with the INP locks
430 		 * held so we can use them in ip_output() without
431 		 * requring a reference to the pcb.
432 		 */
433 		if (inp->inp_options != NULL) {
434 			options = m_dup(inp->inp_options, M_NOWAIT);
435 			if (options == NULL) {
436 				INP_RUNLOCK(inp);
437 				error = ENOBUFS;
438 				goto cantsend;
439 			}
440 		}
441 		INP_RUNLOCK(inp);
442 
443 		switch (ip->ip_v) {
444 		case IPVERSION:
445 			error = ip_output(m, options, NULL,
446 			    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
447 			    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
448 			break;
449 #ifdef INET6
450 		case IPV6_VERSION >> 4:
451 			error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
452 			break;
453 #endif
454 		}
455 		if (options != NULL)
456 			m_freem(options);
457 	} else {
458 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
459 		if (m->m_pkthdr.rcvif == NULL) {
460 			/*
461 			 * No luck with the name, check by IP address.
462 			 * Clear the port and the ifname to make sure
463 			 * there are no distractions for ifa_ifwithaddr.
464 			 */
465 			struct epoch_tracker et;
466 			struct	ifaddr *ifa;
467 
468 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
469 			sin->sin_port = 0;
470 			NET_EPOCH_ENTER(et);
471 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
472 			if (ifa == NULL) {
473 				error = EADDRNOTAVAIL;
474 				NET_EPOCH_EXIT(et);
475 				goto cantsend;
476 			}
477 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
478 			NET_EPOCH_EXIT(et);
479 		}
480 #ifdef MAC
481 		mac_socket_create_mbuf(so, m);
482 #endif
483 		/* Send packet to input processing via netisr */
484 		switch (ip->ip_v) {
485 		case IPVERSION:
486 			/*
487 			 * Restore M_BCAST flag when destination address is
488 			 * broadcast. It is expected by ip_tryforward().
489 			 */
490 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
491 				m->m_flags |= M_MCAST;
492 			else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
493 				m->m_flags |= M_BCAST;
494 			netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
495 			break;
496 #ifdef INET6
497 		case IPV6_VERSION >> 4:
498 			netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
499 			break;
500 #endif
501 		default:
502 			error = EINVAL;
503 			goto cantsend;
504 		}
505 	}
506 
507 	return (error);
508 
509 cantsend:
510 	m_freem(m);
511 	return (error);
512 }
513 
514 static int
515 div_attach(struct socket *so, int proto, struct thread *td)
516 {
517 	struct inpcb *inp;
518 	int error;
519 
520 	inp  = sotoinpcb(so);
521 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
522 	if (td != NULL) {
523 		error = priv_check(td, PRIV_NETINET_DIVERT);
524 		if (error)
525 			return (error);
526 	}
527 	error = soreserve(so, div_sendspace, div_recvspace);
528 	if (error)
529 		return error;
530 	INP_INFO_WLOCK(&V_divcbinfo);
531 	error = in_pcballoc(so, &V_divcbinfo);
532 	if (error) {
533 		INP_INFO_WUNLOCK(&V_divcbinfo);
534 		return error;
535 	}
536 	inp = (struct inpcb *)so->so_pcb;
537 	INP_INFO_WUNLOCK(&V_divcbinfo);
538 	inp->inp_ip_p = proto;
539 	inp->inp_vflag |= INP_IPV4;
540 	inp->inp_flags |= INP_HDRINCL;
541 	INP_WUNLOCK(inp);
542 	return 0;
543 }
544 
545 static void
546 div_detach(struct socket *so)
547 {
548 	struct inpcb *inp;
549 
550 	inp = sotoinpcb(so);
551 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
552 	INP_INFO_WLOCK(&V_divcbinfo);
553 	INP_WLOCK(inp);
554 	in_pcbdetach(inp);
555 	in_pcbfree(inp);
556 	INP_INFO_WUNLOCK(&V_divcbinfo);
557 }
558 
559 static int
560 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
561 {
562 	struct inpcb *inp;
563 	int error;
564 
565 	inp = sotoinpcb(so);
566 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
567 	/* in_pcbbind assumes that nam is a sockaddr_in
568 	 * and in_pcbbind requires a valid address. Since divert
569 	 * sockets don't we need to make sure the address is
570 	 * filled in properly.
571 	 * XXX -- divert should not be abusing in_pcbind
572 	 * and should probably have its own family.
573 	 */
574 	if (nam->sa_family != AF_INET)
575 		return EAFNOSUPPORT;
576 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
577 	INP_INFO_WLOCK(&V_divcbinfo);
578 	INP_WLOCK(inp);
579 	INP_HASH_WLOCK(&V_divcbinfo);
580 	error = in_pcbbind(inp, nam, td->td_ucred);
581 	INP_HASH_WUNLOCK(&V_divcbinfo);
582 	INP_WUNLOCK(inp);
583 	INP_INFO_WUNLOCK(&V_divcbinfo);
584 	return error;
585 }
586 
587 static int
588 div_shutdown(struct socket *so)
589 {
590 	struct inpcb *inp;
591 
592 	inp = sotoinpcb(so);
593 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
594 	INP_WLOCK(inp);
595 	socantsendmore(so);
596 	INP_WUNLOCK(inp);
597 	return 0;
598 }
599 
600 static int
601 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
602     struct mbuf *control, struct thread *td)
603 {
604 
605 	/* Packet must have a header (but that's about it) */
606 	if (m->m_len < sizeof (struct ip) &&
607 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
608 		KMOD_IPSTAT_INC(ips_toosmall);
609 		m_freem(m);
610 		return EINVAL;
611 	}
612 
613 	/* Send packet */
614 	return div_output(so, m, (struct sockaddr_in *)nam, control);
615 }
616 
617 static void
618 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
619 {
620         struct in_addr faddr;
621 
622 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
623 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
624         	return;
625 	if (PRC_IS_REDIRECT(cmd))
626 		return;
627 }
628 
629 static int
630 div_pcblist(SYSCTL_HANDLER_ARGS)
631 {
632 	struct xinpgen xig;
633 	struct epoch_tracker et;
634 	struct inpcb *inp;
635 	int error;
636 
637 	if (req->newptr != 0)
638 		return EPERM;
639 
640 	if (req->oldptr == 0) {
641 		int n;
642 
643 		n = V_divcbinfo.ipi_count;
644 		n += imax(n / 8, 10);
645 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
646 		return 0;
647 	}
648 
649 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
650 		return (error);
651 
652 	bzero(&xig, sizeof(xig));
653 	xig.xig_len = sizeof xig;
654 	xig.xig_count = V_divcbinfo.ipi_count;
655 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
656 	xig.xig_sogen = so_gencnt;
657 	error = SYSCTL_OUT(req, &xig, sizeof xig);
658 	if (error)
659 		return error;
660 
661 	NET_EPOCH_ENTER(et);
662 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
663 	    inp != NULL;
664 	    inp = CK_LIST_NEXT(inp, inp_list)) {
665 		INP_RLOCK(inp);
666 		if (inp->inp_gencnt <= xig.xig_gen) {
667 			struct xinpcb xi;
668 
669 			in_pcbtoxinpcb(inp, &xi);
670 			INP_RUNLOCK(inp);
671 			error = SYSCTL_OUT(req, &xi, sizeof xi);
672 		} else
673 			INP_RUNLOCK(inp);
674 	}
675 	NET_EPOCH_EXIT(et);
676 
677 	if (!error) {
678 		/*
679 		 * Give the user an updated idea of our state.
680 		 * If the generation differs from what we told
681 		 * her before, she knows that something happened
682 		 * while we were processing this request, and it
683 		 * might be necessary to retry.
684 		 */
685 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
686 		xig.xig_sogen = so_gencnt;
687 		xig.xig_count = V_divcbinfo.ipi_count;
688 		error = SYSCTL_OUT(req, &xig, sizeof xig);
689 	}
690 
691 	return (error);
692 }
693 
694 #ifdef SYSCTL_NODE
695 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
696     "IPDIVERT");
697 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
698     NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
699 #endif
700 
701 struct pr_usrreqs div_usrreqs = {
702 	.pru_attach =		div_attach,
703 	.pru_bind =		div_bind,
704 	.pru_control =		in_control,
705 	.pru_detach =		div_detach,
706 	.pru_peeraddr =		in_getpeeraddr,
707 	.pru_send =		div_send,
708 	.pru_shutdown =		div_shutdown,
709 	.pru_sockaddr =		in_getsockaddr,
710 	.pru_sosetlabel =	in_pcbsosetlabel
711 };
712 
713 struct protosw div_protosw = {
714 	.pr_type =		SOCK_RAW,
715 	.pr_protocol =		IPPROTO_DIVERT,
716 	.pr_flags =		PR_ATOMIC|PR_ADDR,
717 	.pr_input =		div_input,
718 	.pr_ctlinput =		div_ctlinput,
719 	.pr_ctloutput =		ip_ctloutput,
720 	.pr_init =		div_init,
721 	.pr_usrreqs =		&div_usrreqs
722 };
723 
724 static int
725 div_modevent(module_t mod, int type, void *unused)
726 {
727 	int err = 0;
728 
729 	switch (type) {
730 	case MOD_LOAD:
731 		/*
732 		 * Protocol will be initialized by pf_proto_register().
733 		 * We don't have to register ip_protox because we are not
734 		 * a true IP protocol that goes over the wire.
735 		 */
736 		err = pf_proto_register(PF_INET, &div_protosw);
737 		if (err != 0)
738 			return (err);
739 		ip_divert_ptr = divert_packet;
740 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
741 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
742 		break;
743 	case MOD_QUIESCE:
744 		/*
745 		 * IPDIVERT may normally not be unloaded because of the
746 		 * potential race conditions.  Tell kldunload we can't be
747 		 * unloaded unless the unload is forced.
748 		 */
749 		err = EPERM;
750 		break;
751 	case MOD_UNLOAD:
752 		/*
753 		 * Forced unload.
754 		 *
755 		 * Module ipdivert can only be unloaded if no sockets are
756 		 * connected.  Maybe this can be changed later to forcefully
757 		 * disconnect any open sockets.
758 		 *
759 		 * XXXRW: Note that there is a slight race here, as a new
760 		 * socket open request could be spinning on the lock and then
761 		 * we destroy the lock.
762 		 */
763 		INP_INFO_WLOCK(&V_divcbinfo);
764 		if (V_divcbinfo.ipi_count != 0) {
765 			err = EBUSY;
766 			INP_INFO_WUNLOCK(&V_divcbinfo);
767 			break;
768 		}
769 		ip_divert_ptr = NULL;
770 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
771 		INP_INFO_WUNLOCK(&V_divcbinfo);
772 #ifndef VIMAGE
773 		div_destroy(NULL);
774 #endif
775 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
776 		break;
777 	default:
778 		err = EOPNOTSUPP;
779 		break;
780 	}
781 	return err;
782 }
783 
784 static moduledata_t ipdivertmod = {
785         "ipdivert",
786         div_modevent,
787         0
788 };
789 
790 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
791 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
792 MODULE_VERSION(ipdivert, 1);
793