ip_divert.c revision 169454
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 * $FreeBSD: head/sys/netinet/ip_divert.c 169454 2007-05-10 15:58:48Z rwatson $
30 */
31
32#if !defined(KLD_MODULE)
33#include "opt_inet.h"
34#include "opt_ipfw.h"
35#include "opt_mac.h"
36#ifndef INET
37#error "IPDIVERT requires INET."
38#endif
39#ifndef IPFIREWALL
40#error "IPDIVERT requires IPFIREWALL"
41#endif
42#endif
43
44#include <sys/param.h>
45#include <sys/kernel.h>
46#include <sys/lock.h>
47#include <sys/malloc.h>
48#include <sys/mbuf.h>
49#include <sys/module.h>
50#include <sys/kernel.h>
51#include <sys/priv.h>
52#include <sys/proc.h>
53#include <sys/protosw.h>
54#include <sys/signalvar.h>
55#include <sys/socket.h>
56#include <sys/socketvar.h>
57#include <sys/sx.h>
58#include <sys/sysctl.h>
59#include <sys/systm.h>
60
61#include <vm/uma.h>
62
63#include <net/if.h>
64#include <net/route.h>
65
66#include <netinet/in.h>
67#include <netinet/in_pcb.h>
68#include <netinet/in_systm.h>
69#include <netinet/in_var.h>
70#include <netinet/ip.h>
71#include <netinet/ip_divert.h>
72#include <netinet/ip_var.h>
73#include <netinet/ip_fw.h>
74
75#include <security/mac/mac_framework.h>
76
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, see the divert(4)
89 * manpage for features.
90 * Internally, packets selected by ipfw in ip_input() or ip_output(),
91 * and never diverted before, are passed to the input queue of the
92 * divert socket with a given 'divert_port' number (as specified in
93 * the matching ipfw rule), and they are tagged with a 16 bit cookie
94 * (representing the rule number of the matching ipfw rule), which
95 * is passed to process reading from the socket.
96 *
97 * Packets written to the divert socket are again tagged with a cookie
98 * (usually the same as above) and a destination address.
99 * If the destination address is INADDR_ANY then the packet is
100 * treated as outgoing and sent to ip_output(), otherwise it is
101 * treated as incoming and sent to ip_input().
102 * In both cases, the packet is tagged with the cookie.
103 *
104 * On reinjection, processing in ip_input() and ip_output()
105 * will be exactly the same as for the original packet, except that
106 * ipfw processing will start at the rule number after the one
107 * written in the cookie (so, tagging a packet with a cookie of 0
108 * will cause it to be effectively considered as a standard packet).
109 */
110
111/* Internal variables. */
112static struct inpcbhead divcb;
113static struct inpcbinfo divcbinfo;
114
115static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
116static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
117
118/*
119 * Initialize divert connection block queue.
120 */
121static void
122div_zone_change(void *tag)
123{
124
125	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
126}
127
128static int
129div_inpcb_init(void *mem, int size, int flags)
130{
131	struct inpcb *inp = mem;
132
133	INP_LOCK_INIT(inp, "inp", "divinp");
134	return (0);
135}
136
137static void
138div_inpcb_fini(void *mem, int size)
139{
140	struct inpcb *inp = mem;
141
142	INP_LOCK_DESTROY(inp);
143}
144
145void
146div_init(void)
147{
148
149	INP_INFO_LOCK_INIT(&divcbinfo, "div");
150	LIST_INIT(&divcb);
151	divcbinfo.ipi_listhead = &divcb;
152	/*
153	 * XXX We don't use the hash list for divert IP, but it's easier
154	 * to allocate a one entry hash list than it is to check all
155	 * over the place for hashbase == NULL.
156	 */
157	divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &divcbinfo.ipi_hashmask);
158	divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
159	    &divcbinfo.ipi_porthashmask);
160	divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
161	    NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
162	    UMA_ZONE_NOFREE);
163	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
164	EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
165		NULL, EVENTHANDLER_PRI_ANY);
166}
167
168/*
169 * IPPROTO_DIVERT is not in the real IP protocol number space; this
170 * function should never be called.  Just in case, drop any packets.
171 */
172void
173div_input(struct mbuf *m, int off)
174{
175	ipstat.ips_noproto++;
176	m_freem(m);
177}
178
179/*
180 * Divert a packet by passing it up to the divert socket at port 'port'.
181 *
182 * Setup generic address and protocol structures for div_input routine,
183 * then pass them along with mbuf chain.
184 */
185static void
186divert_packet(struct mbuf *m, int incoming)
187{
188	struct ip *ip;
189	struct inpcb *inp;
190	struct socket *sa;
191	u_int16_t nport;
192	struct sockaddr_in divsrc;
193	struct m_tag *mtag;
194
195	mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
196	if (mtag == NULL) {
197		printf("%s: no divert tag\n", __func__);
198		m_freem(m);
199		return;
200	}
201	/* Assure header */
202	if (m->m_len < sizeof(struct ip) &&
203	    (m = m_pullup(m, sizeof(struct ip))) == 0)
204		return;
205	ip = mtod(m, struct ip *);
206
207	/* Delayed checksums are currently not compatible with divert. */
208	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
209		ip->ip_len = ntohs(ip->ip_len);
210		in_delayed_cksum(m);
211		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
212		ip->ip_len = htons(ip->ip_len);
213	}
214
215	/*
216	 * Record receive interface address, if any.
217	 * But only for incoming packets.
218	 */
219	bzero(&divsrc, sizeof(divsrc));
220	divsrc.sin_len = sizeof(divsrc);
221	divsrc.sin_family = AF_INET;
222	divsrc.sin_port = divert_cookie(mtag);	/* record matching rule */
223	if (incoming) {
224		struct ifaddr *ifa;
225
226		/* Sanity check */
227		M_ASSERTPKTHDR(m);
228
229		/* Find IP address for receive interface */
230		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
231			if (ifa->ifa_addr->sa_family != AF_INET)
232				continue;
233			divsrc.sin_addr =
234			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
235			break;
236		}
237	}
238	/*
239	 * Record the incoming interface name whenever we have one.
240	 */
241	if (m->m_pkthdr.rcvif) {
242		/*
243		 * Hide the actual interface name in there in the
244		 * sin_zero array. XXX This needs to be moved to a
245		 * different sockaddr type for divert, e.g.
246		 * sockaddr_div with multiple fields like
247		 * sockaddr_dl. Presently we have only 7 bytes
248		 * but that will do for now as most interfaces
249		 * are 4 or less + 2 or less bytes for unit.
250		 * There is probably a faster way of doing this,
251		 * possibly taking it from the sockaddr_dl on the iface.
252		 * This solves the problem of a P2P link and a LAN interface
253		 * having the same address, which can result in the wrong
254		 * interface being assigned to the packet when fed back
255		 * into the divert socket. Theoretically if the daemon saves
256		 * and re-uses the sockaddr_in as suggested in the man pages,
257		 * this iface name will come along for the ride.
258		 * (see div_output for the other half of this.)
259		 */
260		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
261		    sizeof(divsrc.sin_zero));
262	}
263
264	/* Put packet on socket queue, if any */
265	sa = NULL;
266	nport = htons((u_int16_t)divert_info(mtag));
267	INP_INFO_RLOCK(&divcbinfo);
268	LIST_FOREACH(inp, &divcb, inp_list) {
269		INP_LOCK(inp);
270		/* XXX why does only one socket match? */
271		if (inp->inp_lport == nport) {
272			sa = inp->inp_socket;
273			SOCKBUF_LOCK(&sa->so_rcv);
274			if (sbappendaddr_locked(&sa->so_rcv,
275			    (struct sockaddr *)&divsrc, m,
276			    (struct mbuf *)0) == 0) {
277				SOCKBUF_UNLOCK(&sa->so_rcv);
278				sa = NULL;	/* force mbuf reclaim below */
279			} else
280				sorwakeup_locked(sa);
281			INP_UNLOCK(inp);
282			break;
283		}
284		INP_UNLOCK(inp);
285	}
286	INP_INFO_RUNLOCK(&divcbinfo);
287	if (sa == NULL) {
288		m_freem(m);
289		ipstat.ips_noproto++;
290		ipstat.ips_delivered--;
291        }
292}
293
294/*
295 * Deliver packet back into the IP processing machinery.
296 *
297 * If no address specified, or address is 0.0.0.0, send to ip_output();
298 * otherwise, send to ip_input() and mark as having been received on
299 * the interface with that address.
300 */
301static int
302div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
303    struct mbuf *control)
304{
305	struct m_tag *mtag;
306	struct divert_tag *dt;
307	int error = 0;
308
309	/*
310	 * An mbuf may hasn't come from userland, but we pretend
311	 * that it has.
312	 */
313	m->m_pkthdr.rcvif = NULL;
314	m->m_nextpkt = NULL;
315
316	if (control)
317		m_freem(control);		/* XXX */
318
319	if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
320		mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
321		    M_NOWAIT | M_ZERO);
322		if (mtag == NULL) {
323			error = ENOBUFS;
324			goto cantsend;
325		}
326		dt = (struct divert_tag *)(mtag+1);
327		m_tag_prepend(m, mtag);
328	} else
329		dt = (struct divert_tag *)(mtag+1);
330
331	/* Loopback avoidance and state recovery */
332	if (sin) {
333		int i;
334
335		dt->cookie = sin->sin_port;
336		/*
337		 * Find receive interface with the given name, stuffed
338		 * (if it exists) in the sin_zero[] field.
339		 * The name is user supplied data so don't trust its size
340		 * or that it is zero terminated.
341		 */
342		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
343			;
344		if ( i > 0 && i < sizeof(sin->sin_zero))
345			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
346	}
347
348	/* Reinject packet into the system as incoming or outgoing */
349	if (!sin || sin->sin_addr.s_addr == 0) {
350		struct ip *const ip = mtod(m, struct ip *);
351		struct inpcb *inp;
352
353		dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
354		INP_INFO_WLOCK(&divcbinfo);
355		inp = sotoinpcb(so);
356		INP_LOCK(inp);
357		/*
358		 * Don't allow both user specified and setsockopt options,
359		 * and don't allow packet length sizes that will crash
360		 */
361		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
362		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
363			error = EINVAL;
364			m_freem(m);
365		} else {
366			/* Convert fields to host order for ip_output() */
367			ip->ip_len = ntohs(ip->ip_len);
368			ip->ip_off = ntohs(ip->ip_off);
369
370			/* Send packet to output processing */
371			ipstat.ips_rawout++;			/* XXX */
372
373#ifdef MAC
374			mac_create_mbuf_from_inpcb(inp, m);
375#endif
376			error = ip_output(m,
377				    inp->inp_options, NULL,
378				    ((so->so_options & SO_DONTROUTE) ?
379				    IP_ROUTETOIF : 0) |
380				    IP_ALLOWBROADCAST | IP_RAWOUTPUT,
381				    inp->inp_moptions, NULL);
382		}
383		INP_UNLOCK(inp);
384		INP_INFO_WUNLOCK(&divcbinfo);
385	} else {
386		dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
387		if (m->m_pkthdr.rcvif == NULL) {
388			/*
389			 * No luck with the name, check by IP address.
390			 * Clear the port and the ifname to make sure
391			 * there are no distractions for ifa_ifwithaddr.
392			 */
393			struct	ifaddr *ifa;
394
395			bzero(sin->sin_zero, sizeof(sin->sin_zero));
396			sin->sin_port = 0;
397			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
398			if (ifa == NULL) {
399				error = EADDRNOTAVAIL;
400				goto cantsend;
401			}
402			m->m_pkthdr.rcvif = ifa->ifa_ifp;
403		}
404#ifdef MAC
405		SOCK_LOCK(so);
406		mac_create_mbuf_from_socket(so, m);
407		SOCK_UNLOCK(so);
408#endif
409		/* Send packet to input processing */
410		ip_input(m);
411	}
412
413	return error;
414
415cantsend:
416	m_freem(m);
417	return error;
418}
419
420static int
421div_attach(struct socket *so, int proto, struct thread *td)
422{
423	struct inpcb *inp;
424	int error;
425
426	inp  = sotoinpcb(so);
427	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
428	if (td != NULL) {
429		error = priv_check(td, PRIV_NETINET_DIVERT);
430		if (error)
431			return (error);
432	}
433	error = soreserve(so, div_sendspace, div_recvspace);
434	if (error)
435		return error;
436	INP_INFO_WLOCK(&divcbinfo);
437	error = in_pcballoc(so, &divcbinfo);
438	if (error) {
439		INP_INFO_WUNLOCK(&divcbinfo);
440		return error;
441	}
442	inp = (struct inpcb *)so->so_pcb;
443	INP_INFO_WUNLOCK(&divcbinfo);
444	inp->inp_ip_p = proto;
445	inp->inp_vflag |= INP_IPV4;
446	inp->inp_flags |= INP_HDRINCL;
447	INP_UNLOCK(inp);
448	return 0;
449}
450
451static void
452div_detach(struct socket *so)
453{
454	struct inpcb *inp;
455
456	inp = sotoinpcb(so);
457	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
458	INP_INFO_WLOCK(&divcbinfo);
459	INP_LOCK(inp);
460	in_pcbdetach(inp);
461	in_pcbfree(inp);
462	INP_INFO_WUNLOCK(&divcbinfo);
463}
464
465static int
466div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
467{
468	struct inpcb *inp;
469	int error;
470
471	inp = sotoinpcb(so);
472	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
473	/* in_pcbbind assumes that nam is a sockaddr_in
474	 * and in_pcbbind requires a valid address. Since divert
475	 * sockets don't we need to make sure the address is
476	 * filled in properly.
477	 * XXX -- divert should not be abusing in_pcbind
478	 * and should probably have its own family.
479	 */
480	if (nam->sa_family != AF_INET)
481		return EAFNOSUPPORT;
482	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
483	INP_INFO_WLOCK(&divcbinfo);
484	INP_LOCK(inp);
485	error = in_pcbbind(inp, nam, td->td_ucred);
486	INP_UNLOCK(inp);
487	INP_INFO_WUNLOCK(&divcbinfo);
488	return error;
489}
490
491static int
492div_shutdown(struct socket *so)
493{
494	struct inpcb *inp;
495
496	inp = sotoinpcb(so);
497	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
498	INP_LOCK(inp);
499	socantsendmore(so);
500	INP_UNLOCK(inp);
501	return 0;
502}
503
504static int
505div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
506    struct mbuf *control, struct thread *td)
507{
508	/* Packet must have a header (but that's about it) */
509	if (m->m_len < sizeof (struct ip) &&
510	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
511		ipstat.ips_toosmall++;
512		m_freem(m);
513		return EINVAL;
514	}
515
516	/* Send packet */
517	return div_output(so, m, (struct sockaddr_in *)nam, control);
518}
519
520void
521div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
522{
523        struct in_addr faddr;
524
525	faddr = ((struct sockaddr_in *)sa)->sin_addr;
526	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
527        	return;
528	if (PRC_IS_REDIRECT(cmd))
529		return;
530}
531
532static int
533div_pcblist(SYSCTL_HANDLER_ARGS)
534{
535	int error, i, n;
536	struct inpcb *inp, **inp_list;
537	inp_gen_t gencnt;
538	struct xinpgen xig;
539
540	/*
541	 * The process of preparing the TCB list is too time-consuming and
542	 * resource-intensive to repeat twice on every request.
543	 */
544	if (req->oldptr == 0) {
545		n = divcbinfo.ipi_count;
546		req->oldidx = 2 * (sizeof xig)
547			+ (n + n/8) * sizeof(struct xinpcb);
548		return 0;
549	}
550
551	if (req->newptr != 0)
552		return EPERM;
553
554	/*
555	 * OK, now we're committed to doing something.
556	 */
557	INP_INFO_RLOCK(&divcbinfo);
558	gencnt = divcbinfo.ipi_gencnt;
559	n = divcbinfo.ipi_count;
560	INP_INFO_RUNLOCK(&divcbinfo);
561
562	error = sysctl_wire_old_buffer(req,
563	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
564	if (error != 0)
565		return (error);
566
567	xig.xig_len = sizeof xig;
568	xig.xig_count = n;
569	xig.xig_gen = gencnt;
570	xig.xig_sogen = so_gencnt;
571	error = SYSCTL_OUT(req, &xig, sizeof xig);
572	if (error)
573		return error;
574
575	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
576	if (inp_list == 0)
577		return ENOMEM;
578
579	INP_INFO_RLOCK(&divcbinfo);
580	for (inp = LIST_FIRST(divcbinfo.ipi_listhead), i = 0; inp && i < n;
581	     inp = LIST_NEXT(inp, inp_list)) {
582		INP_LOCK(inp);
583		if (inp->inp_gencnt <= gencnt &&
584		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
585			inp_list[i++] = inp;
586		INP_UNLOCK(inp);
587	}
588	INP_INFO_RUNLOCK(&divcbinfo);
589	n = i;
590
591	error = 0;
592	for (i = 0; i < n; i++) {
593		inp = inp_list[i];
594		INP_LOCK(inp);
595		if (inp->inp_gencnt <= gencnt) {
596			struct xinpcb xi;
597			bzero(&xi, sizeof(xi));
598			xi.xi_len = sizeof xi;
599			/* XXX should avoid extra copy */
600			bcopy(inp, &xi.xi_inp, sizeof *inp);
601			if (inp->inp_socket)
602				sotoxsocket(inp->inp_socket, &xi.xi_socket);
603			INP_UNLOCK(inp);
604			error = SYSCTL_OUT(req, &xi, sizeof xi);
605		} else
606			INP_UNLOCK(inp);
607	}
608	if (!error) {
609		/*
610		 * Give the user an updated idea of our state.
611		 * If the generation differs from what we told
612		 * her before, she knows that something happened
613		 * while we were processing this request, and it
614		 * might be necessary to retry.
615		 */
616		INP_INFO_RLOCK(&divcbinfo);
617		xig.xig_gen = divcbinfo.ipi_gencnt;
618		xig.xig_sogen = so_gencnt;
619		xig.xig_count = divcbinfo.ipi_count;
620		INP_INFO_RUNLOCK(&divcbinfo);
621		error = SYSCTL_OUT(req, &xig, sizeof xig);
622	}
623	free(inp_list, M_TEMP);
624	return error;
625}
626
627/*
628 * This is the wrapper function for in_setsockaddr.  We just pass down
629 * the pcbinfo for in_setpeeraddr to lock.
630 */
631static int
632div_sockaddr(struct socket *so, struct sockaddr **nam)
633{
634
635	return (in_setsockaddr(so, nam));
636}
637
638/*
639 * This is the wrapper function for in_setpeeraddr. We just pass down
640 * the pcbinfo for in_setpeeraddr to lock.
641 */
642static int
643div_peeraddr(struct socket *so, struct sockaddr **nam)
644{
645
646	return (in_setpeeraddr(so, nam));
647}
648
649#ifdef SYSCTL_NODE
650SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
651SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
652	    div_pcblist, "S,xinpcb", "List of active divert sockets");
653#endif
654
655struct pr_usrreqs div_usrreqs = {
656	.pru_attach =		div_attach,
657	.pru_bind =		div_bind,
658	.pru_control =		in_control,
659	.pru_detach =		div_detach,
660	.pru_peeraddr =		div_peeraddr,
661	.pru_send =		div_send,
662	.pru_shutdown =		div_shutdown,
663	.pru_sockaddr =		div_sockaddr,
664	.pru_sosetlabel =	in_pcbsosetlabel
665};
666
667struct protosw div_protosw = {
668	.pr_type =		SOCK_RAW,
669	.pr_protocol =		IPPROTO_DIVERT,
670	.pr_flags =		PR_ATOMIC|PR_ADDR,
671	.pr_input =		div_input,
672	.pr_ctlinput =		div_ctlinput,
673	.pr_ctloutput =		ip_ctloutput,
674	.pr_init =		div_init,
675	.pr_usrreqs =		&div_usrreqs
676};
677
678static int
679div_modevent(module_t mod, int type, void *unused)
680{
681	int err = 0;
682	int n;
683
684	switch (type) {
685	case MOD_LOAD:
686		/*
687		 * Protocol will be initialized by pf_proto_register().
688		 * We don't have to register ip_protox because we are not
689		 * a true IP protocol that goes over the wire.
690		 */
691		err = pf_proto_register(PF_INET, &div_protosw);
692		ip_divert_ptr = divert_packet;
693		break;
694	case MOD_QUIESCE:
695		/*
696		 * IPDIVERT may normally not be unloaded because of the
697		 * potential race conditions.  Tell kldunload we can't be
698		 * unloaded unless the unload is forced.
699		 */
700		err = EPERM;
701		break;
702	case MOD_UNLOAD:
703		/*
704		 * Forced unload.
705		 *
706		 * Module ipdivert can only be unloaded if no sockets are
707		 * connected.  Maybe this can be changed later to forcefully
708		 * disconnect any open sockets.
709		 *
710		 * XXXRW: Note that there is a slight race here, as a new
711		 * socket open request could be spinning on the lock and then
712		 * we destroy the lock.
713		 */
714		INP_INFO_WLOCK(&divcbinfo);
715		n = divcbinfo.ipi_count;
716		if (n != 0) {
717			err = EBUSY;
718			INP_INFO_WUNLOCK(&divcbinfo);
719			break;
720		}
721		ip_divert_ptr = NULL;
722		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
723		INP_INFO_WUNLOCK(&divcbinfo);
724		INP_INFO_LOCK_DESTROY(&divcbinfo);
725		uma_zdestroy(divcbinfo.ipi_zone);
726		break;
727	default:
728		err = EOPNOTSUPP;
729		break;
730	}
731	return err;
732}
733
734static moduledata_t ipdivertmod = {
735        "ipdivert",
736        div_modevent,
737        0
738};
739
740DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
741MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
742MODULE_VERSION(ipdivert, 1);
743