xref: /freebsd/sys/netinet/ip_divert.c (revision ac099daf6742ead81ea7ea86351a8ef4e783041b)
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 #if defined(SCTP) || defined(SCTP_SUPPORT)
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 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
126     struct sockaddr_in *sin);
127 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
128 
129 /*
130  * Initialize divert connection block queue.
131  */
132 static void
133 div_zone_change(void *tag)
134 {
135 
136 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
137 }
138 
139 static int
140 div_inpcb_init(void *mem, int size, int flags)
141 {
142 	struct inpcb *inp = mem;
143 
144 	INP_LOCK_INIT(inp, "inp", "divinp");
145 	return (0);
146 }
147 
148 static void
149 div_init(void)
150 {
151 
152 	/*
153 	 * XXX We don't use the hash list for divert IP, but it's easier to
154 	 * allocate one-entry hash lists than it is to check all over the
155 	 * place for hashbase == NULL.
156 	 */
157 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
158 	    div_inpcb_init, IPI_HASHFIELDS_NONE);
159 }
160 
161 static void
162 div_destroy(void *unused __unused)
163 {
164 
165 	in_pcbinfo_destroy(&V_divcbinfo);
166 }
167 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
168     div_destroy, NULL);
169 
170 /*
171  * IPPROTO_DIVERT is not in the real IP protocol number space; this
172  * function should never be called.  Just in case, drop any packets.
173  */
174 static int
175 div_input(struct mbuf **mp, int *offp, int proto)
176 {
177 	struct mbuf *m = *mp;
178 
179 	KMOD_IPSTAT_INC(ips_noproto);
180 	m_freem(m);
181 	return (IPPROTO_DONE);
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, bool 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 	NET_EPOCH_ASSERT();
201 
202 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
203 	if (mtag == NULL) {
204 		m_freem(m);
205 		return;
206 	}
207 	/* Assure header */
208 	if (m->m_len < sizeof(struct ip) &&
209 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
210 		return;
211 	ip = mtod(m, struct ip *);
212 
213 	/* Delayed checksums are currently not compatible with divert. */
214 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
215 		m = mb_unmapped_to_ext(m);
216 		if (m == NULL)
217 			return;
218 		in_delayed_cksum(m);
219 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
220 	}
221 #if defined(SCTP) || defined(SCTP_SUPPORT)
222 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
223 		m = mb_unmapped_to_ext(m);
224 		if (m == NULL)
225 			return;
226 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
227 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
228 	}
229 #endif
230 	bzero(&divsrc, sizeof(divsrc));
231 	divsrc.sin_len = sizeof(divsrc);
232 	divsrc.sin_family = AF_INET;
233 	/* record matching rule, in host format */
234 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
235 	/*
236 	 * Record receive interface address, if any.
237 	 * But only for incoming packets.
238 	 */
239 	if (incoming) {
240 		struct ifaddr *ifa;
241 		struct ifnet *ifp;
242 
243 		/* Sanity check */
244 		M_ASSERTPKTHDR(m);
245 
246 		/* Find IP address for receive interface */
247 		ifp = m->m_pkthdr.rcvif;
248 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
249 			if (ifa->ifa_addr->sa_family != AF_INET)
250 				continue;
251 			divsrc.sin_addr =
252 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
253 			break;
254 		}
255 	}
256 	/*
257 	 * Record the incoming interface name whenever we have one.
258 	 */
259 	if (m->m_pkthdr.rcvif) {
260 		/*
261 		 * Hide the actual interface name in there in the
262 		 * sin_zero array. XXX This needs to be moved to a
263 		 * different sockaddr type for divert, e.g.
264 		 * sockaddr_div with multiple fields like
265 		 * sockaddr_dl. Presently we have only 7 bytes
266 		 * but that will do for now as most interfaces
267 		 * are 4 or less + 2 or less bytes for unit.
268 		 * There is probably a faster way of doing this,
269 		 * possibly taking it from the sockaddr_dl on the iface.
270 		 * This solves the problem of a P2P link and a LAN interface
271 		 * having the same address, which can result in the wrong
272 		 * interface being assigned to the packet when fed back
273 		 * into the divert socket. Theoretically if the daemon saves
274 		 * and re-uses the sockaddr_in as suggested in the man pages,
275 		 * this iface name will come along for the ride.
276 		 * (see div_output for the other half of this.)
277 		 */
278 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
279 		    sizeof(divsrc.sin_zero));
280 	}
281 
282 	/* Put packet on socket queue, if any */
283 	sa = NULL;
284 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
285 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
286 		/* XXX why does only one socket match? */
287 		if (inp->inp_lport == nport) {
288 			INP_RLOCK(inp);
289 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
290 				INP_RUNLOCK(inp);
291 				continue;
292 			}
293 			sa = inp->inp_socket;
294 			SOCKBUF_LOCK(&sa->so_rcv);
295 			if (sbappendaddr_locked(&sa->so_rcv,
296 			    (struct sockaddr *)&divsrc, m,
297 			    (struct mbuf *)0) == 0) {
298 				SOCKBUF_UNLOCK(&sa->so_rcv);
299 				sa = NULL;	/* force mbuf reclaim below */
300 			} else
301 				sorwakeup_locked(sa);
302 			INP_RUNLOCK(inp);
303 			break;
304 		}
305 	}
306 	if (sa == NULL) {
307 		m_freem(m);
308 		KMOD_IPSTAT_INC(ips_noproto);
309 		KMOD_IPSTAT_DEC(ips_delivered);
310         }
311 }
312 
313 /*
314  * Deliver packet back into the IP processing machinery.
315  *
316  * If no address specified, or address is 0.0.0.0, send to ip_output();
317  * otherwise, send to ip_input() and mark as having been received on
318  * the interface with that address.
319  */
320 static int
321 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
322     struct mbuf *control)
323 {
324 	struct epoch_tracker et;
325 	const struct ip *ip;
326 	struct m_tag *mtag;
327 	struct ipfw_rule_ref *dt;
328 	int error, family;
329 
330 	/*
331 	 * An mbuf may hasn't come from userland, but we pretend
332 	 * that it has.
333 	 */
334 	m->m_pkthdr.rcvif = NULL;
335 	m->m_nextpkt = NULL;
336 	M_SETFIB(m, so->so_fibnum);
337 
338 	if (control)
339 		m_freem(control);		/* XXX */
340 
341 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
342 	if (mtag == NULL) {
343 		/* this should be normal */
344 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
345 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
346 		if (mtag == NULL) {
347 			m_freem(m);
348 			return (ENOBUFS);
349 		}
350 		m_tag_prepend(m, mtag);
351 	}
352 	dt = (struct ipfw_rule_ref *)(mtag+1);
353 
354 	/* Loopback avoidance and state recovery */
355 	if (sin) {
356 		int i;
357 
358 		/* set the starting point. We provide a non-zero slot,
359 		 * but a non_matching chain_id to skip that info and use
360 		 * the rulenum/rule_id.
361 		 */
362 		dt->slot = 1; /* dummy, chain_id is invalid */
363 		dt->chain_id = 0;
364 		dt->rulenum = sin->sin_port+1; /* host format ? */
365 		dt->rule_id = 0;
366 		/* XXX: broken for IPv6 */
367 		/*
368 		 * Find receive interface with the given name, stuffed
369 		 * (if it exists) in the sin_zero[] field.
370 		 * The name is user supplied data so don't trust its size
371 		 * or that it is zero terminated.
372 		 */
373 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
374 			;
375 		if ( i > 0 && i < sizeof(sin->sin_zero))
376 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
377 	}
378 
379 	ip = mtod(m, struct ip *);
380 	switch (ip->ip_v) {
381 	case IPVERSION:
382 		family = AF_INET;
383 		break;
384 #ifdef INET6
385 	case IPV6_VERSION >> 4:
386 		family = AF_INET6;
387 		break;
388 #endif
389 	default:
390 		m_freem(m);
391 		return (EAFNOSUPPORT);
392 	}
393 
394 	/* Reinject packet into the system as incoming or outgoing */
395 	NET_EPOCH_ENTER(et);
396 	if (!sin || sin->sin_addr.s_addr == 0) {
397 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
398 		error = div_output_outbound(family, so, m);
399 	} else {
400 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
401 		error = div_output_inbound(family, so, m, sin);
402 	}
403 	NET_EPOCH_EXIT(et);
404 
405 	if (error != 0)
406 		m_freem(m);
407 
408 	return (error);
409 }
410 
411 /*
412  * Sends mbuf @m to the wire via ip[6]_output().
413  *
414  * Returns 0 on success, @m is consumed.
415  * On failure, returns error code. It is caller responsibility to free @m.
416  */
417 static int
418 div_output_outbound(int family, struct socket *so, struct mbuf *m)
419 {
420 	struct ip *const ip = mtod(m, struct ip *);
421 	struct mbuf *options;
422 	struct inpcb *inp;
423 	int error;
424 
425 	inp = sotoinpcb(so);
426 	INP_RLOCK(inp);
427 	switch (family) {
428 	case AF_INET:
429 		/*
430 		 * Don't allow both user specified and setsockopt
431 		 * options, and don't allow packet length sizes that
432 		 * will crash.
433 		 */
434 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
435 		    inp->inp_options != NULL) ||
436 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
437 			INP_RUNLOCK(inp);
438 			return (EINVAL);
439 		}
440 		break;
441 #ifdef INET6
442 	case AF_INET6:
443 	    {
444 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
445 
446 		/* Don't allow packet length sizes that will crash */
447 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
448 			INP_RUNLOCK(inp);
449 			return (EINVAL);
450 		}
451 		break;
452 	    }
453 #endif
454 	}
455 
456 	/* Send packet to output processing */
457 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
458 
459 #ifdef MAC
460 	mac_inpcb_create_mbuf(inp, m);
461 #endif
462 	/*
463 	 * Get ready to inject the packet into ip_output().
464 	 * Just in case socket options were specified on the
465 	 * divert socket, we duplicate them.  This is done
466 	 * to avoid having to hold the PCB locks over the call
467 	 * to ip_output(), as doing this results in a number of
468 	 * lock ordering complexities.
469 	 *
470 	 * Note that we set the multicast options argument for
471 	 * ip_output() to NULL since it should be invariant that
472 	 * they are not present.
473 	 */
474 	KASSERT(inp->inp_moptions == NULL,
475 	    ("multicast options set on a divert socket"));
476 	/*
477 	 * XXXCSJP: It is unclear to me whether or not it makes
478 	 * sense for divert sockets to have options.  However,
479 	 * for now we will duplicate them with the INP locks
480 	 * held so we can use them in ip_output() without
481 	 * requring a reference to the pcb.
482 	 */
483 	options = NULL;
484 	if (inp->inp_options != NULL) {
485 		options = m_dup(inp->inp_options, M_NOWAIT);
486 		if (options == NULL) {
487 			INP_RUNLOCK(inp);
488 			return (ENOBUFS);
489 		}
490 	}
491 	INP_RUNLOCK(inp);
492 
493 	error = 0;
494 	switch (family) {
495 	case AF_INET:
496 		error = ip_output(m, options, NULL,
497 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
498 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
499 		break;
500 #ifdef INET6
501 	case AF_INET6:
502 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
503 		break;
504 #endif
505 	}
506 	if (options != NULL)
507 		m_freem(options);
508 
509 	return (error);
510 }
511 
512 /*
513  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
514  *
515  * Returns 0 on success, @m is consumed.
516  * Returns error code on failure. It is caller responsibility to free @m.
517  */
518 static int
519 div_output_inbound(int family, struct socket *so, struct mbuf *m,
520     struct sockaddr_in *sin)
521 {
522 	const struct ip *ip;
523 	struct ifaddr *ifa;
524 
525 	if (m->m_pkthdr.rcvif == NULL) {
526 		/*
527 		 * No luck with the name, check by IP address.
528 		 * Clear the port and the ifname to make sure
529 		 * there are no distractions for ifa_ifwithaddr.
530 		 */
531 
532 		/* XXX: broken for IPv6 */
533 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
534 		sin->sin_port = 0;
535 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
536 		if (ifa == NULL)
537 			return (EADDRNOTAVAIL);
538 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
539 	}
540 #ifdef MAC
541 	mac_socket_create_mbuf(so, m);
542 #endif
543 	/* Send packet to input processing via netisr */
544 	switch (family) {
545 	case AF_INET:
546 		ip = mtod(m, struct ip *);
547 		/*
548 		 * Restore M_BCAST flag when destination address is
549 		 * broadcast. It is expected by ip_tryforward().
550 		 */
551 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
552 			m->m_flags |= M_MCAST;
553 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
554 			m->m_flags |= M_BCAST;
555 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
556 		break;
557 #ifdef INET6
558 	case AF_INET6:
559 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
560 		break;
561 #endif
562 	default:
563 		return (EINVAL);
564 	}
565 
566 	return (0);
567 }
568 
569 static int
570 div_attach(struct socket *so, int proto, struct thread *td)
571 {
572 	struct inpcb *inp;
573 	int error;
574 
575 	inp  = sotoinpcb(so);
576 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
577 	if (td != NULL) {
578 		error = priv_check(td, PRIV_NETINET_DIVERT);
579 		if (error)
580 			return (error);
581 	}
582 	error = soreserve(so, div_sendspace, div_recvspace);
583 	if (error)
584 		return error;
585 	INP_INFO_WLOCK(&V_divcbinfo);
586 	error = in_pcballoc(so, &V_divcbinfo);
587 	if (error) {
588 		INP_INFO_WUNLOCK(&V_divcbinfo);
589 		return error;
590 	}
591 	inp = (struct inpcb *)so->so_pcb;
592 	INP_INFO_WUNLOCK(&V_divcbinfo);
593 	inp->inp_ip_p = proto;
594 	inp->inp_vflag |= INP_IPV4;
595 	inp->inp_flags |= INP_HDRINCL;
596 	INP_WUNLOCK(inp);
597 	return 0;
598 }
599 
600 static void
601 div_detach(struct socket *so)
602 {
603 	struct inpcb *inp;
604 
605 	inp = sotoinpcb(so);
606 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
607 	INP_INFO_WLOCK(&V_divcbinfo);
608 	INP_WLOCK(inp);
609 	in_pcbdetach(inp);
610 	in_pcbfree(inp);
611 	INP_INFO_WUNLOCK(&V_divcbinfo);
612 }
613 
614 static int
615 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
616 {
617 	struct inpcb *inp;
618 	int error;
619 
620 	inp = sotoinpcb(so);
621 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
622 	/* in_pcbbind assumes that nam is a sockaddr_in
623 	 * and in_pcbbind requires a valid address. Since divert
624 	 * sockets don't we need to make sure the address is
625 	 * filled in properly.
626 	 * XXX -- divert should not be abusing in_pcbind
627 	 * and should probably have its own family.
628 	 */
629 	if (nam->sa_family != AF_INET)
630 		return EAFNOSUPPORT;
631 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
632 	INP_INFO_WLOCK(&V_divcbinfo);
633 	INP_WLOCK(inp);
634 	INP_HASH_WLOCK(&V_divcbinfo);
635 	error = in_pcbbind(inp, nam, td->td_ucred);
636 	INP_HASH_WUNLOCK(&V_divcbinfo);
637 	INP_WUNLOCK(inp);
638 	INP_INFO_WUNLOCK(&V_divcbinfo);
639 	return error;
640 }
641 
642 static int
643 div_shutdown(struct socket *so)
644 {
645 	struct inpcb *inp;
646 
647 	inp = sotoinpcb(so);
648 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
649 	INP_WLOCK(inp);
650 	socantsendmore(so);
651 	INP_WUNLOCK(inp);
652 	return 0;
653 }
654 
655 static int
656 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
657     struct mbuf *control, struct thread *td)
658 {
659 
660 	/* Packet must have a header (but that's about it) */
661 	if (m->m_len < sizeof (struct ip) &&
662 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
663 		KMOD_IPSTAT_INC(ips_toosmall);
664 		m_freem(m);
665 		return EINVAL;
666 	}
667 
668 	/* Send packet */
669 	return div_output(so, m, (struct sockaddr_in *)nam, control);
670 }
671 
672 static void
673 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
674 {
675         struct in_addr faddr;
676 
677 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
678 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
679         	return;
680 	if (PRC_IS_REDIRECT(cmd))
681 		return;
682 }
683 
684 static int
685 div_pcblist(SYSCTL_HANDLER_ARGS)
686 {
687 	struct xinpgen xig;
688 	struct epoch_tracker et;
689 	struct inpcb *inp;
690 	int error;
691 
692 	if (req->newptr != 0)
693 		return EPERM;
694 
695 	if (req->oldptr == 0) {
696 		int n;
697 
698 		n = V_divcbinfo.ipi_count;
699 		n += imax(n / 8, 10);
700 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
701 		return 0;
702 	}
703 
704 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
705 		return (error);
706 
707 	bzero(&xig, sizeof(xig));
708 	xig.xig_len = sizeof xig;
709 	xig.xig_count = V_divcbinfo.ipi_count;
710 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
711 	xig.xig_sogen = so_gencnt;
712 	error = SYSCTL_OUT(req, &xig, sizeof xig);
713 	if (error)
714 		return error;
715 
716 	NET_EPOCH_ENTER(et);
717 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
718 	    inp != NULL;
719 	    inp = CK_LIST_NEXT(inp, inp_list)) {
720 		INP_RLOCK(inp);
721 		if (inp->inp_gencnt <= xig.xig_gen) {
722 			struct xinpcb xi;
723 
724 			in_pcbtoxinpcb(inp, &xi);
725 			INP_RUNLOCK(inp);
726 			error = SYSCTL_OUT(req, &xi, sizeof xi);
727 		} else
728 			INP_RUNLOCK(inp);
729 	}
730 	NET_EPOCH_EXIT(et);
731 
732 	if (!error) {
733 		/*
734 		 * Give the user an updated idea of our state.
735 		 * If the generation differs from what we told
736 		 * her before, she knows that something happened
737 		 * while we were processing this request, and it
738 		 * might be necessary to retry.
739 		 */
740 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
741 		xig.xig_sogen = so_gencnt;
742 		xig.xig_count = V_divcbinfo.ipi_count;
743 		error = SYSCTL_OUT(req, &xig, sizeof xig);
744 	}
745 
746 	return (error);
747 }
748 
749 #ifdef SYSCTL_NODE
750 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
751     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
752     "IPDIVERT");
753 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
754    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
755     NULL, 0, div_pcblist, "S,xinpcb",
756     "List of active divert sockets");
757 #endif
758 
759 struct pr_usrreqs div_usrreqs = {
760 	.pru_attach =		div_attach,
761 	.pru_bind =		div_bind,
762 	.pru_control =		in_control,
763 	.pru_detach =		div_detach,
764 	.pru_peeraddr =		in_getpeeraddr,
765 	.pru_send =		div_send,
766 	.pru_shutdown =		div_shutdown,
767 	.pru_sockaddr =		in_getsockaddr,
768 	.pru_sosetlabel =	in_pcbsosetlabel
769 };
770 
771 struct protosw div_protosw = {
772 	.pr_type =		SOCK_RAW,
773 	.pr_protocol =		IPPROTO_DIVERT,
774 	.pr_flags =		PR_ATOMIC|PR_ADDR,
775 	.pr_input =		div_input,
776 	.pr_ctlinput =		div_ctlinput,
777 	.pr_ctloutput =		ip_ctloutput,
778 	.pr_init =		div_init,
779 	.pr_usrreqs =		&div_usrreqs
780 };
781 
782 static int
783 div_modevent(module_t mod, int type, void *unused)
784 {
785 	int err = 0;
786 
787 	switch (type) {
788 	case MOD_LOAD:
789 		/*
790 		 * Protocol will be initialized by pf_proto_register().
791 		 * We don't have to register ip_protox because we are not
792 		 * a true IP protocol that goes over the wire.
793 		 */
794 		err = pf_proto_register(PF_INET, &div_protosw);
795 		if (err != 0)
796 			return (err);
797 		ip_divert_ptr = divert_packet;
798 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
799 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
800 		break;
801 	case MOD_QUIESCE:
802 		/*
803 		 * IPDIVERT may normally not be unloaded because of the
804 		 * potential race conditions.  Tell kldunload we can't be
805 		 * unloaded unless the unload is forced.
806 		 */
807 		err = EPERM;
808 		break;
809 	case MOD_UNLOAD:
810 		/*
811 		 * Forced unload.
812 		 *
813 		 * Module ipdivert can only be unloaded if no sockets are
814 		 * connected.  Maybe this can be changed later to forcefully
815 		 * disconnect any open sockets.
816 		 *
817 		 * XXXRW: Note that there is a slight race here, as a new
818 		 * socket open request could be spinning on the lock and then
819 		 * we destroy the lock.
820 		 */
821 		INP_INFO_WLOCK(&V_divcbinfo);
822 		if (V_divcbinfo.ipi_count != 0) {
823 			err = EBUSY;
824 			INP_INFO_WUNLOCK(&V_divcbinfo);
825 			break;
826 		}
827 		ip_divert_ptr = NULL;
828 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
829 		INP_INFO_WUNLOCK(&V_divcbinfo);
830 #ifndef VIMAGE
831 		div_destroy(NULL);
832 #endif
833 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
834 		break;
835 	default:
836 		err = EOPNOTSUPP;
837 		break;
838 	}
839 	return err;
840 }
841 
842 static moduledata_t ipdivertmod = {
843         "ipdivert",
844         div_modevent,
845         0
846 };
847 
848 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
849 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
850 MODULE_VERSION(ipdivert, 1);
851