xref: /freebsd/sys/netinet/ip_divert.c (revision 5bb3134a8c21cb87b30e135ef168483f0333dabb)
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 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
224 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
225 	}
226 #endif
227 #ifdef INET6
228 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
229 		m = mb_unmapped_to_ext(m);
230 		if (m == NULL)
231 			return;
232 		in6_delayed_cksum(m, m->m_pkthdr.len -
233 		    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
234 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
235 	}
236 #if defined(SCTP) || defined(SCTP_SUPPORT)
237 	if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
238 		sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
239 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
240 	}
241 #endif
242 #endif /* INET6 */
243 	bzero(&divsrc, sizeof(divsrc));
244 	divsrc.sin_len = sizeof(divsrc);
245 	divsrc.sin_family = AF_INET;
246 	/* record matching rule, in host format */
247 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
248 	/*
249 	 * Record receive interface address, if any.
250 	 * But only for incoming packets.
251 	 */
252 	if (incoming) {
253 		struct ifaddr *ifa;
254 		struct ifnet *ifp;
255 
256 		/* Sanity check */
257 		M_ASSERTPKTHDR(m);
258 
259 		/* Find IP address for receive interface */
260 		ifp = m->m_pkthdr.rcvif;
261 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
262 			if (ifa->ifa_addr->sa_family != AF_INET)
263 				continue;
264 			divsrc.sin_addr =
265 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
266 			break;
267 		}
268 	}
269 	/*
270 	 * Record the incoming interface name whenever we have one.
271 	 */
272 	if (m->m_pkthdr.rcvif) {
273 		/*
274 		 * Hide the actual interface name in there in the
275 		 * sin_zero array. XXX This needs to be moved to a
276 		 * different sockaddr type for divert, e.g.
277 		 * sockaddr_div with multiple fields like
278 		 * sockaddr_dl. Presently we have only 7 bytes
279 		 * but that will do for now as most interfaces
280 		 * are 4 or less + 2 or less bytes for unit.
281 		 * There is probably a faster way of doing this,
282 		 * possibly taking it from the sockaddr_dl on the iface.
283 		 * This solves the problem of a P2P link and a LAN interface
284 		 * having the same address, which can result in the wrong
285 		 * interface being assigned to the packet when fed back
286 		 * into the divert socket. Theoretically if the daemon saves
287 		 * and re-uses the sockaddr_in as suggested in the man pages,
288 		 * this iface name will come along for the ride.
289 		 * (see div_output for the other half of this.)
290 		 */
291 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
292 		    sizeof(divsrc.sin_zero));
293 	}
294 
295 	/* Put packet on socket queue, if any */
296 	sa = NULL;
297 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
298 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
299 		/* XXX why does only one socket match? */
300 		if (inp->inp_lport == nport) {
301 			INP_RLOCK(inp);
302 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
303 				INP_RUNLOCK(inp);
304 				continue;
305 			}
306 			sa = inp->inp_socket;
307 			SOCKBUF_LOCK(&sa->so_rcv);
308 			if (sbappendaddr_locked(&sa->so_rcv,
309 			    (struct sockaddr *)&divsrc, m,
310 			    (struct mbuf *)0) == 0) {
311 				soroverflow_locked(sa);
312 				sa = NULL;	/* force mbuf reclaim below */
313 			} else
314 				sorwakeup_locked(sa);
315 			INP_RUNLOCK(inp);
316 			break;
317 		}
318 	}
319 	if (sa == NULL) {
320 		m_freem(m);
321 		KMOD_IPSTAT_INC(ips_noproto);
322 		KMOD_IPSTAT_DEC(ips_delivered);
323         }
324 }
325 
326 /*
327  * Deliver packet back into the IP processing machinery.
328  *
329  * If no address specified, or address is 0.0.0.0, send to ip_output();
330  * otherwise, send to ip_input() and mark as having been received on
331  * the interface with that address.
332  */
333 static int
334 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
335     struct mbuf *control)
336 {
337 	struct epoch_tracker et;
338 	const struct ip *ip;
339 	struct m_tag *mtag;
340 	struct ipfw_rule_ref *dt;
341 	int error, family;
342 
343 	if (control) {
344 		m_freem(control);		/* XXX */
345 		control = NULL;
346 	}
347 
348 	if (sin != NULL) {
349 		if (sin->sin_family != AF_INET) {
350 			m_freem(m);
351 			return (EAFNOSUPPORT);
352 		}
353 		if (sin->sin_len != sizeof(*sin)) {
354 			m_freem(m);
355 			return (EINVAL);
356 		}
357 	}
358 
359 	/*
360 	 * An mbuf may hasn't come from userland, but we pretend
361 	 * that it has.
362 	 */
363 	m->m_pkthdr.rcvif = NULL;
364 	m->m_nextpkt = NULL;
365 	M_SETFIB(m, so->so_fibnum);
366 
367 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
368 	if (mtag == NULL) {
369 		/* this should be normal */
370 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
371 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
372 		if (mtag == NULL) {
373 			m_freem(m);
374 			return (ENOBUFS);
375 		}
376 		m_tag_prepend(m, mtag);
377 	}
378 	dt = (struct ipfw_rule_ref *)(mtag+1);
379 
380 	/* Loopback avoidance and state recovery */
381 	if (sin) {
382 		int i;
383 
384 		/* set the starting point. We provide a non-zero slot,
385 		 * but a non_matching chain_id to skip that info and use
386 		 * the rulenum/rule_id.
387 		 */
388 		dt->slot = 1; /* dummy, chain_id is invalid */
389 		dt->chain_id = 0;
390 		dt->rulenum = sin->sin_port+1; /* host format ? */
391 		dt->rule_id = 0;
392 		/* XXX: broken for IPv6 */
393 		/*
394 		 * Find receive interface with the given name, stuffed
395 		 * (if it exists) in the sin_zero[] field.
396 		 * The name is user supplied data so don't trust its size
397 		 * or that it is zero terminated.
398 		 */
399 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
400 			;
401 		if ( i > 0 && i < sizeof(sin->sin_zero))
402 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
403 	}
404 
405 	ip = mtod(m, struct ip *);
406 	switch (ip->ip_v) {
407 	case IPVERSION:
408 		family = AF_INET;
409 		break;
410 #ifdef INET6
411 	case IPV6_VERSION >> 4:
412 		family = AF_INET6;
413 		break;
414 #endif
415 	default:
416 		m_freem(m);
417 		return (EAFNOSUPPORT);
418 	}
419 
420 	/* Reinject packet into the system as incoming or outgoing */
421 	NET_EPOCH_ENTER(et);
422 	if (!sin || sin->sin_addr.s_addr == 0) {
423 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
424 		error = div_output_outbound(family, so, m);
425 	} else {
426 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
427 		error = div_output_inbound(family, so, m, sin);
428 	}
429 	NET_EPOCH_EXIT(et);
430 
431 	return (error);
432 }
433 
434 /*
435  * Sends mbuf @m to the wire via ip[6]_output().
436  *
437  * Returns 0 on success or an errno value on failure.  @m is always consumed.
438  */
439 static int
440 div_output_outbound(int family, struct socket *so, struct mbuf *m)
441 {
442 	struct ip *const ip = mtod(m, struct ip *);
443 	struct mbuf *options;
444 	struct inpcb *inp;
445 	int error;
446 
447 	inp = sotoinpcb(so);
448 	INP_RLOCK(inp);
449 	switch (family) {
450 	case AF_INET:
451 		/*
452 		 * Don't allow both user specified and setsockopt
453 		 * options, and don't allow packet length sizes that
454 		 * will crash.
455 		 */
456 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
457 		    inp->inp_options != NULL) ||
458 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
459 			INP_RUNLOCK(inp);
460 			m_freem(m);
461 			return (EINVAL);
462 		}
463 		break;
464 #ifdef INET6
465 	case AF_INET6:
466 	    {
467 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
468 
469 		/* Don't allow packet length sizes that will crash */
470 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
471 			INP_RUNLOCK(inp);
472 			m_freem(m);
473 			return (EINVAL);
474 		}
475 		break;
476 	    }
477 #endif
478 	}
479 
480 	/* Send packet to output processing */
481 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
482 
483 #ifdef MAC
484 	mac_inpcb_create_mbuf(inp, m);
485 #endif
486 	/*
487 	 * Get ready to inject the packet into ip_output().
488 	 * Just in case socket options were specified on the
489 	 * divert socket, we duplicate them.  This is done
490 	 * to avoid having to hold the PCB locks over the call
491 	 * to ip_output(), as doing this results in a number of
492 	 * lock ordering complexities.
493 	 *
494 	 * Note that we set the multicast options argument for
495 	 * ip_output() to NULL since it should be invariant that
496 	 * they are not present.
497 	 */
498 	KASSERT(inp->inp_moptions == NULL,
499 	    ("multicast options set on a divert socket"));
500 	/*
501 	 * XXXCSJP: It is unclear to me whether or not it makes
502 	 * sense for divert sockets to have options.  However,
503 	 * for now we will duplicate them with the INP locks
504 	 * held so we can use them in ip_output() without
505 	 * requring a reference to the pcb.
506 	 */
507 	options = NULL;
508 	if (inp->inp_options != NULL) {
509 		options = m_dup(inp->inp_options, M_NOWAIT);
510 		if (options == NULL) {
511 			INP_RUNLOCK(inp);
512 			m_freem(m);
513 			return (ENOBUFS);
514 		}
515 	}
516 	INP_RUNLOCK(inp);
517 
518 	error = 0;
519 	switch (family) {
520 	case AF_INET:
521 		error = ip_output(m, options, NULL,
522 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
523 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
524 		break;
525 #ifdef INET6
526 	case AF_INET6:
527 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
528 		break;
529 #endif
530 	}
531 	if (options != NULL)
532 		m_freem(options);
533 
534 	return (error);
535 }
536 
537 /*
538  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
539  *
540  * Returns 0 on success or an errno value on failure.  @m is always consumed.
541  */
542 static int
543 div_output_inbound(int family, struct socket *so, struct mbuf *m,
544     struct sockaddr_in *sin)
545 {
546 	const struct ip *ip;
547 	struct ifaddr *ifa;
548 
549 	if (m->m_pkthdr.rcvif == NULL) {
550 		/*
551 		 * No luck with the name, check by IP address.
552 		 * Clear the port and the ifname to make sure
553 		 * there are no distractions for ifa_ifwithaddr.
554 		 */
555 
556 		/* XXX: broken for IPv6 */
557 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
558 		sin->sin_port = 0;
559 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
560 		if (ifa == NULL) {
561 			m_freem(m);
562 			return (EADDRNOTAVAIL);
563 		}
564 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
565 	}
566 #ifdef MAC
567 	mac_socket_create_mbuf(so, m);
568 #endif
569 	/* Send packet to input processing via netisr */
570 	switch (family) {
571 	case AF_INET:
572 		ip = mtod(m, struct ip *);
573 		/*
574 		 * Restore M_BCAST flag when destination address is
575 		 * broadcast. It is expected by ip_tryforward().
576 		 */
577 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
578 			m->m_flags |= M_MCAST;
579 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
580 			m->m_flags |= M_BCAST;
581 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
582 		break;
583 #ifdef INET6
584 	case AF_INET6:
585 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
586 		break;
587 #endif
588 	default:
589 		m_freem(m);
590 		return (EINVAL);
591 	}
592 
593 	return (0);
594 }
595 
596 static int
597 div_attach(struct socket *so, int proto, struct thread *td)
598 {
599 	struct inpcb *inp;
600 	int error;
601 
602 	inp  = sotoinpcb(so);
603 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
604 	if (td != NULL) {
605 		error = priv_check(td, PRIV_NETINET_DIVERT);
606 		if (error)
607 			return (error);
608 	}
609 	error = soreserve(so, div_sendspace, div_recvspace);
610 	if (error)
611 		return error;
612 	INP_INFO_WLOCK(&V_divcbinfo);
613 	error = in_pcballoc(so, &V_divcbinfo);
614 	if (error) {
615 		INP_INFO_WUNLOCK(&V_divcbinfo);
616 		return error;
617 	}
618 	inp = (struct inpcb *)so->so_pcb;
619 	INP_INFO_WUNLOCK(&V_divcbinfo);
620 	inp->inp_ip_p = proto;
621 	inp->inp_vflag |= INP_IPV4;
622 	inp->inp_flags |= INP_HDRINCL;
623 	INP_WUNLOCK(inp);
624 	return 0;
625 }
626 
627 static void
628 div_detach(struct socket *so)
629 {
630 	struct inpcb *inp;
631 
632 	inp = sotoinpcb(so);
633 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
634 	INP_INFO_WLOCK(&V_divcbinfo);
635 	INP_WLOCK(inp);
636 	in_pcbdetach(inp);
637 	in_pcbfree(inp);
638 	INP_INFO_WUNLOCK(&V_divcbinfo);
639 }
640 
641 static int
642 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
643 {
644 	struct inpcb *inp;
645 	int error;
646 
647 	inp = sotoinpcb(so);
648 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
649 	/* in_pcbbind assumes that nam is a sockaddr_in
650 	 * and in_pcbbind requires a valid address. Since divert
651 	 * sockets don't we need to make sure the address is
652 	 * filled in properly.
653 	 * XXX -- divert should not be abusing in_pcbind
654 	 * and should probably have its own family.
655 	 */
656 	if (nam->sa_family != AF_INET)
657 		return EAFNOSUPPORT;
658 	if (nam->sa_len != sizeof(struct sockaddr_in))
659 		return EINVAL;
660 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
661 	INP_INFO_WLOCK(&V_divcbinfo);
662 	INP_WLOCK(inp);
663 	INP_HASH_WLOCK(&V_divcbinfo);
664 	error = in_pcbbind(inp, nam, td->td_ucred);
665 	INP_HASH_WUNLOCK(&V_divcbinfo);
666 	INP_WUNLOCK(inp);
667 	INP_INFO_WUNLOCK(&V_divcbinfo);
668 	return error;
669 }
670 
671 static int
672 div_shutdown(struct socket *so)
673 {
674 	struct inpcb *inp;
675 
676 	inp = sotoinpcb(so);
677 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
678 	INP_WLOCK(inp);
679 	socantsendmore(so);
680 	INP_WUNLOCK(inp);
681 	return 0;
682 }
683 
684 static int
685 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
686     struct mbuf *control, struct thread *td)
687 {
688 
689 	/* Packet must have a header (but that's about it) */
690 	if (m->m_len < sizeof (struct ip) &&
691 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
692 		KMOD_IPSTAT_INC(ips_toosmall);
693 		if (control != NULL)
694 			m_freem(control);
695 		m_freem(m);
696 		return EINVAL;
697 	}
698 
699 	/* Send packet */
700 	return div_output(so, m, (struct sockaddr_in *)nam, control);
701 }
702 
703 static int
704 div_pcblist(SYSCTL_HANDLER_ARGS)
705 {
706 	struct xinpgen xig;
707 	struct epoch_tracker et;
708 	struct inpcb *inp;
709 	int error;
710 
711 	if (req->newptr != 0)
712 		return EPERM;
713 
714 	if (req->oldptr == 0) {
715 		int n;
716 
717 		n = V_divcbinfo.ipi_count;
718 		n += imax(n / 8, 10);
719 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
720 		return 0;
721 	}
722 
723 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
724 		return (error);
725 
726 	bzero(&xig, sizeof(xig));
727 	xig.xig_len = sizeof xig;
728 	xig.xig_count = V_divcbinfo.ipi_count;
729 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
730 	xig.xig_sogen = so_gencnt;
731 	error = SYSCTL_OUT(req, &xig, sizeof xig);
732 	if (error)
733 		return error;
734 
735 	NET_EPOCH_ENTER(et);
736 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
737 	    inp != NULL;
738 	    inp = CK_LIST_NEXT(inp, inp_list)) {
739 		INP_RLOCK(inp);
740 		if (inp->inp_gencnt <= xig.xig_gen) {
741 			struct xinpcb xi;
742 
743 			in_pcbtoxinpcb(inp, &xi);
744 			INP_RUNLOCK(inp);
745 			error = SYSCTL_OUT(req, &xi, sizeof xi);
746 		} else
747 			INP_RUNLOCK(inp);
748 	}
749 	NET_EPOCH_EXIT(et);
750 
751 	if (!error) {
752 		/*
753 		 * Give the user an updated idea of our state.
754 		 * If the generation differs from what we told
755 		 * her before, she knows that something happened
756 		 * while we were processing this request, and it
757 		 * might be necessary to retry.
758 		 */
759 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
760 		xig.xig_sogen = so_gencnt;
761 		xig.xig_count = V_divcbinfo.ipi_count;
762 		error = SYSCTL_OUT(req, &xig, sizeof xig);
763 	}
764 
765 	return (error);
766 }
767 
768 #ifdef SYSCTL_NODE
769 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
770     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
771     "IPDIVERT");
772 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
773    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
774     NULL, 0, div_pcblist, "S,xinpcb",
775     "List of active divert sockets");
776 #endif
777 
778 struct pr_usrreqs div_usrreqs = {
779 	.pru_attach =		div_attach,
780 	.pru_bind =		div_bind,
781 	.pru_control =		in_control,
782 	.pru_detach =		div_detach,
783 	.pru_peeraddr =		in_getpeeraddr,
784 	.pru_send =		div_send,
785 	.pru_shutdown =		div_shutdown,
786 	.pru_sockaddr =		in_getsockaddr,
787 	.pru_sosetlabel =	in_pcbsosetlabel
788 };
789 
790 struct protosw div_protosw = {
791 	.pr_type =		SOCK_RAW,
792 	.pr_protocol =		IPPROTO_DIVERT,
793 	.pr_flags =		PR_ATOMIC|PR_ADDR,
794 	.pr_input =		div_input,
795 	.pr_init =		div_init,
796 	.pr_usrreqs =		&div_usrreqs
797 };
798 
799 static int
800 div_modevent(module_t mod, int type, void *unused)
801 {
802 	int err = 0;
803 
804 	switch (type) {
805 	case MOD_LOAD:
806 		/*
807 		 * Protocol will be initialized by pf_proto_register().
808 		 * We don't have to register ip_protox because we are not
809 		 * a true IP protocol that goes over the wire.
810 		 */
811 		err = pf_proto_register(PF_INET, &div_protosw);
812 		if (err != 0)
813 			return (err);
814 		ip_divert_ptr = divert_packet;
815 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
816 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
817 		break;
818 	case MOD_QUIESCE:
819 		/*
820 		 * IPDIVERT may normally not be unloaded because of the
821 		 * potential race conditions.  Tell kldunload we can't be
822 		 * unloaded unless the unload is forced.
823 		 */
824 		err = EPERM;
825 		break;
826 	case MOD_UNLOAD:
827 		/*
828 		 * Forced unload.
829 		 *
830 		 * Module ipdivert can only be unloaded if no sockets are
831 		 * connected.  Maybe this can be changed later to forcefully
832 		 * disconnect any open sockets.
833 		 *
834 		 * XXXRW: Note that there is a slight race here, as a new
835 		 * socket open request could be spinning on the lock and then
836 		 * we destroy the lock.
837 		 */
838 		INP_INFO_WLOCK(&V_divcbinfo);
839 		if (V_divcbinfo.ipi_count != 0) {
840 			err = EBUSY;
841 			INP_INFO_WUNLOCK(&V_divcbinfo);
842 			break;
843 		}
844 		ip_divert_ptr = NULL;
845 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
846 		INP_INFO_WUNLOCK(&V_divcbinfo);
847 #ifndef VIMAGE
848 		div_destroy(NULL);
849 #endif
850 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
851 		break;
852 	default:
853 		err = EOPNOTSUPP;
854 		break;
855 	}
856 	return err;
857 }
858 
859 static moduledata_t ipdivertmod = {
860         "ipdivert",
861         div_modevent,
862         0
863 };
864 
865 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
866 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
867 MODULE_VERSION(ipdivert, 1);
868