xref: /freebsd/sys/netinet/ip_divert.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
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 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_sctp.h"
36 
37 #include <sys/param.h>
38 #include <sys/eventhandler.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/module.h>
44 #include <sys/kernel.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <net/vnet.h>
53 
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_private.h>
57 #include <net/netisr.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/ip_divert.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/ip6_var.h>
69 #endif
70 #if defined(SCTP) || defined(SCTP_SUPPORT)
71 #include <netinet/sctp_crc32.h>
72 #endif
73 
74 #include <security/mac/mac_framework.h>
75 /*
76  * Divert sockets
77  */
78 
79 /*
80  * Allocate enough space to hold a full IP packet
81  */
82 #define	DIVSNDQ		(65536 + 100)
83 #define	DIVRCVQ		(65536 + 100)
84 
85 /*
86  * Usually a system has very few divert ports.  Previous implementation
87  * used a linked list.
88  */
89 #define	DIVHASHSIZE	(1 << 3)	/* 8 entries, one cache line. */
90 #define	DIVHASH(port)	(port % DIVHASHSIZE)
91 #define	DCBHASH(dcb)	((dcb)->dcb_port % DIVHASHSIZE)
92 
93 /*
94  * Divert sockets work in conjunction with ipfw or other packet filters,
95  * see the divert(4) manpage for features.
96  * Packets are selected by the packet filter and tagged with an
97  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
98  * the packet filter) and information on the matching filter rule for
99  * subsequent reinjection. The divert_port is used to put the packet
100  * on the corresponding divert socket, while the rule number is passed
101  * up (at least partially) as the sin_port in the struct sockaddr.
102  *
103  * Packets written to the divert socket carry in sin_addr a
104  * destination address, and in sin_port the number of the filter rule
105  * after which to continue processing.
106  * If the destination address is INADDR_ANY, the packet is treated as
107  * as outgoing and sent to ip_output(); otherwise it is treated as
108  * incoming and sent to ip_input().
109  * Further, sin_zero carries some information on the interface,
110  * which can be used in the reinject -- see comments in the code.
111  *
112  * On reinjection, processing in ip_input() and ip_output()
113  * will be exactly the same as for the original packet, except that
114  * packet filter processing will start at the rule number after the one
115  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
116  * will apply the entire ruleset to the packet).
117  */
118 static SYSCTL_NODE(_net_inet, OID_AUTO, divert, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
119     "divert(4)");
120 
121 VNET_PCPUSTAT_DEFINE_STATIC(struct divstat, divstat);
122 VNET_PCPUSTAT_SYSINIT(divstat);
123 #ifdef VIMAGE
124 VNET_PCPUSTAT_SYSUNINIT(divstat);
125 #endif
126 SYSCTL_VNET_PCPUSTAT(_net_inet_divert, OID_AUTO, stats, struct divstat,
127     divstat, "divert(4) socket statistics");
128 #define	DIVSTAT_INC(name)	\
129     VNET_PCPUSTAT_ADD(struct divstat, divstat, div_ ## name, 1)
130 
131 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
132 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
133 
134 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
135     struct sockaddr_in *sin);
136 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
137 
138 struct divcb {
139 	union {
140 		SLIST_ENTRY(divcb)	dcb_next;
141 		intptr_t		dcb_bound;
142 #define	DCB_UNBOUND	((intptr_t)-1)
143 	};
144 	struct socket		*dcb_socket;
145 	uint16_t		 dcb_port;
146 	uint64_t		 dcb_gencnt;
147 	struct epoch_context	 dcb_epochctx;
148 };
149 
150 SLIST_HEAD(divhashhead, divcb);
151 
152 VNET_DEFINE_STATIC(struct divhashhead, divhash[DIVHASHSIZE]) = {};
153 #define	V_divhash	VNET(divhash)
154 VNET_DEFINE_STATIC(uint64_t, dcb_count) = 0;
155 #define	V_dcb_count	VNET(dcb_count)
156 VNET_DEFINE_STATIC(uint64_t, dcb_gencnt) = 0;
157 #define	V_dcb_gencnt	VNET(dcb_gencnt)
158 
159 static struct mtx divert_mtx;
160 MTX_SYSINIT(divert, &divert_mtx, "divert(4) socket pcb lists", MTX_DEF);
161 #define	DIVERT_LOCK()	mtx_lock(&divert_mtx)
162 #define	DIVERT_UNLOCK()	mtx_unlock(&divert_mtx)
163 
164 /*
165  * Divert a packet by passing it up to the divert socket at port 'port'.
166  */
167 static void
168 divert_packet(struct mbuf *m, bool incoming)
169 {
170 	struct divcb *dcb;
171 	u_int16_t nport;
172 	struct sockaddr_in divsrc;
173 	struct m_tag *mtag;
174 
175 	NET_EPOCH_ASSERT();
176 
177 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
178 	if (mtag == NULL) {
179 		m_freem(m);
180 		return;
181 	}
182 	/* Assure header */
183 	if (m->m_len < sizeof(struct ip) &&
184 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
185 		return;
186 #ifdef INET
187 	/* Delayed checksums are currently not compatible with divert. */
188 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
189 		in_delayed_cksum(m);
190 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
191 	}
192 #if defined(SCTP) || defined(SCTP_SUPPORT)
193 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
194 		struct ip *ip;
195 
196 		ip = mtod(m, struct ip *);
197 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
198 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
199 	}
200 #endif
201 #endif
202 #ifdef INET6
203 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
204 		in6_delayed_cksum(m, m->m_pkthdr.len -
205 		    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
206 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
207 	}
208 #if defined(SCTP) || defined(SCTP_SUPPORT)
209 	if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
210 		sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
211 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
212 	}
213 #endif
214 #endif /* INET6 */
215 	bzero(&divsrc, sizeof(divsrc));
216 	divsrc.sin_len = sizeof(divsrc);
217 	divsrc.sin_family = AF_INET;
218 	/* record matching rule, in host format */
219 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
220 	/*
221 	 * Record receive interface address, if any.
222 	 * But only for incoming packets.
223 	 */
224 	if (incoming) {
225 		struct ifaddr *ifa;
226 		struct ifnet *ifp;
227 
228 		/* Sanity check */
229 		M_ASSERTPKTHDR(m);
230 
231 		/* Find IP address for receive interface */
232 		ifp = m->m_pkthdr.rcvif;
233 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
234 			if (ifa->ifa_addr->sa_family != AF_INET)
235 				continue;
236 			divsrc.sin_addr =
237 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
238 			break;
239 		}
240 	}
241 	/*
242 	 * Record the incoming interface name whenever we have one.
243 	 */
244 	if (m->m_pkthdr.rcvif) {
245 		/*
246 		 * Hide the actual interface name in there in the
247 		 * sin_zero array. XXX This needs to be moved to a
248 		 * different sockaddr type for divert, e.g.
249 		 * sockaddr_div with multiple fields like
250 		 * sockaddr_dl. Presently we have only 7 bytes
251 		 * but that will do for now as most interfaces
252 		 * are 4 or less + 2 or less bytes for unit.
253 		 * There is probably a faster way of doing this,
254 		 * possibly taking it from the sockaddr_dl on the iface.
255 		 * This solves the problem of a P2P link and a LAN interface
256 		 * having the same address, which can result in the wrong
257 		 * interface being assigned to the packet when fed back
258 		 * into the divert socket. Theoretically if the daemon saves
259 		 * and re-uses the sockaddr_in as suggested in the man pages,
260 		 * this iface name will come along for the ride.
261 		 * (see div_output for the other half of this.)
262 		 */
263 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
264 		    sizeof(divsrc.sin_zero));
265 	}
266 
267 	/* Put packet on socket queue, if any */
268 	nport = htons((uint16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
269 	SLIST_FOREACH(dcb, &V_divhash[DIVHASH(nport)], dcb_next)
270 		if (dcb->dcb_port == nport)
271 			break;
272 
273 	if (dcb != NULL) {
274 		struct socket *sa = dcb->dcb_socket;
275 
276 		SOCKBUF_LOCK(&sa->so_rcv);
277 		if (sbappendaddr_locked(&sa->so_rcv,
278 		    (struct sockaddr *)&divsrc, m, NULL) == 0) {
279 			soroverflow_locked(sa);
280 			m_freem(m);
281 		} else {
282 			sorwakeup_locked(sa);
283 			DIVSTAT_INC(diverted);
284 		}
285 	} else {
286 		DIVSTAT_INC(noport);
287 		m_freem(m);
288 	}
289 }
290 
291 /*
292  * Deliver packet back into the IP processing machinery.
293  *
294  * If no address specified, or address is 0.0.0.0, send to ip_output();
295  * otherwise, send to ip_input() and mark as having been received on
296  * the interface with that address.
297  */
298 static int
299 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
300     struct mbuf *control, struct thread *td)
301 {
302 	struct epoch_tracker et;
303 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
304 	const struct ip *ip;
305 	struct m_tag *mtag;
306 	struct ipfw_rule_ref *dt;
307 	int error, family;
308 
309 	if (control)
310 		m_freem(control);
311 
312 	/* Packet must have a header (but that's about it) */
313 	if (m->m_len < sizeof (struct ip) &&
314 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
315 		m_freem(m);
316 		return (EINVAL);
317 	}
318 
319 	if (sin != NULL) {
320 		if (sin->sin_family != AF_INET) {
321 			m_freem(m);
322 			return (EAFNOSUPPORT);
323 		}
324 		if (sin->sin_len != sizeof(*sin)) {
325 			m_freem(m);
326 			return (EINVAL);
327 		}
328 	}
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 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
339 	if (mtag == NULL) {
340 		/* this should be normal */
341 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
342 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
343 		if (mtag == NULL) {
344 			m_freem(m);
345 			return (ENOBUFS);
346 		}
347 		m_tag_prepend(m, mtag);
348 	}
349 	dt = (struct ipfw_rule_ref *)(mtag+1);
350 
351 	/* Loopback avoidance and state recovery */
352 	if (sin) {
353 		int i;
354 
355 		/* set the starting point. We provide a non-zero slot,
356 		 * but a non_matching chain_id to skip that info and use
357 		 * the rulenum/rule_id.
358 		 */
359 		dt->slot = 1; /* dummy, chain_id is invalid */
360 		dt->chain_id = 0;
361 		dt->rulenum = sin->sin_port+1; /* host format ? */
362 		dt->rule_id = 0;
363 		/* XXX: broken for IPv6 */
364 		/*
365 		 * Find receive interface with the given name, stuffed
366 		 * (if it exists) in the sin_zero[] field.
367 		 * The name is user supplied data so don't trust its size
368 		 * or that it is zero terminated.
369 		 */
370 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
371 			;
372 		if ( i > 0 && i < sizeof(sin->sin_zero))
373 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
374 	}
375 
376 	ip = mtod(m, struct ip *);
377 	switch (ip->ip_v) {
378 #ifdef INET
379 	case IPVERSION:
380 		family = AF_INET;
381 		break;
382 #endif
383 #ifdef INET6
384 	case IPV6_VERSION >> 4:
385 		family = AF_INET6;
386 		break;
387 #endif
388 	default:
389 		m_freem(m);
390 		return (EAFNOSUPPORT);
391 	}
392 
393 	/* Reinject packet into the system as incoming or outgoing */
394 	NET_EPOCH_ENTER(et);
395 	if (!sin || sin->sin_addr.s_addr == 0) {
396 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
397 		error = div_output_outbound(family, so, m);
398 	} else {
399 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
400 		error = div_output_inbound(family, so, m, sin);
401 	}
402 	NET_EPOCH_EXIT(et);
403 
404 	return (error);
405 }
406 
407 /*
408  * Sends mbuf @m to the wire via ip[6]_output().
409  *
410  * Returns 0 on success or an errno value on failure.  @m is always consumed.
411  */
412 static int
413 div_output_outbound(int family, struct socket *so, struct mbuf *m)
414 {
415 	int error;
416 
417 	switch (family) {
418 #ifdef INET
419 	case AF_INET:
420 	    {
421 		struct ip *const ip = mtod(m, struct ip *);
422 
423 		/* Don't allow packet length sizes that will crash. */
424 		if (((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
425 			m_freem(m);
426 			return (EINVAL);
427 		}
428 		break;
429 	    }
430 #endif
431 #ifdef INET6
432 	case AF_INET6:
433 	    {
434 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
435 
436 		/* Don't allow packet length sizes that will crash */
437 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
438 			m_freem(m);
439 			return (EINVAL);
440 		}
441 		break;
442 	    }
443 #endif
444 	}
445 
446 #ifdef MAC
447 	mac_socket_create_mbuf(so, m);
448 #endif
449 
450 	error = 0;
451 	switch (family) {
452 #ifdef INET
453 	case AF_INET:
454 		error = ip_output(m, NULL, NULL,
455 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
456 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
457 		break;
458 #endif
459 #ifdef INET6
460 	case AF_INET6:
461 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
462 		break;
463 #endif
464 	}
465 	if (error == 0)
466 		DIVSTAT_INC(outbound);
467 
468 	return (error);
469 }
470 
471 /*
472  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
473  *
474  * Returns 0 on success or an errno value on failure.  @m is always consumed.
475  */
476 static int
477 div_output_inbound(int family, struct socket *so, struct mbuf *m,
478     struct sockaddr_in *sin)
479 {
480 	struct ifaddr *ifa;
481 
482 	if (m->m_pkthdr.rcvif == NULL) {
483 		/*
484 		 * No luck with the name, check by IP address.
485 		 * Clear the port and the ifname to make sure
486 		 * there are no distractions for ifa_ifwithaddr.
487 		 */
488 
489 		/* XXX: broken for IPv6 */
490 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
491 		sin->sin_port = 0;
492 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
493 		if (ifa == NULL) {
494 			m_freem(m);
495 			return (EADDRNOTAVAIL);
496 		}
497 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
498 	}
499 #ifdef MAC
500 	mac_socket_create_mbuf(so, m);
501 #endif
502 	/* Send packet to input processing via netisr */
503 	switch (family) {
504 #ifdef INET
505 	case AF_INET:
506 	    {
507 		const struct ip *ip;
508 
509 		ip = mtod(m, struct ip *);
510 		/*
511 		 * Restore M_BCAST flag when destination address is
512 		 * broadcast. It is expected by ip_tryforward().
513 		 */
514 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
515 			m->m_flags |= M_MCAST;
516 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
517 			m->m_flags |= M_BCAST;
518 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
519 		DIVSTAT_INC(inbound);
520 		break;
521 	    }
522 #endif
523 #ifdef INET6
524 	case AF_INET6:
525 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
526 		DIVSTAT_INC(inbound);
527 		break;
528 #endif
529 	default:
530 		m_freem(m);
531 		return (EINVAL);
532 	}
533 
534 	return (0);
535 }
536 
537 static int
538 div_attach(struct socket *so, int proto, struct thread *td)
539 {
540 	struct divcb *dcb;
541 	int error;
542 
543 	if (td != NULL) {
544 		error = priv_check(td, PRIV_NETINET_DIVERT);
545 		if (error)
546 			return (error);
547 	}
548 	error = soreserve(so, div_sendspace, div_recvspace);
549 	if (error)
550 		return error;
551 	dcb = malloc(sizeof(*dcb), M_PCB, M_WAITOK);
552 	dcb->dcb_bound = DCB_UNBOUND;
553 	dcb->dcb_socket = so;
554 	DIVERT_LOCK();
555 	V_dcb_count++;
556 	dcb->dcb_gencnt = ++V_dcb_gencnt;
557 	DIVERT_UNLOCK();
558 	so->so_pcb = dcb;
559 
560 	return (0);
561 }
562 
563 static void
564 div_free(epoch_context_t ctx)
565 {
566 	struct divcb *dcb = __containerof(ctx, struct divcb, dcb_epochctx);
567 
568 	free(dcb, M_PCB);
569 }
570 
571 static void
572 div_detach(struct socket *so)
573 {
574 	struct divcb *dcb = so->so_pcb;
575 
576 	so->so_pcb = NULL;
577 	DIVERT_LOCK();
578 	if (dcb->dcb_bound != DCB_UNBOUND)
579 		SLIST_REMOVE(&V_divhash[DCBHASH(dcb)], dcb, divcb, dcb_next);
580 	V_dcb_count--;
581 	V_dcb_gencnt++;
582 	DIVERT_UNLOCK();
583 	NET_EPOCH_CALL(div_free, &dcb->dcb_epochctx);
584 }
585 
586 static int
587 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
588 {
589 	struct divcb *dcb;
590 	uint16_t port;
591 
592 	if (nam->sa_family != AF_INET)
593 		return EAFNOSUPPORT;
594 	if (nam->sa_len != sizeof(struct sockaddr_in))
595 		return EINVAL;
596 	port = ((struct sockaddr_in *)nam)->sin_port;
597 	DIVERT_LOCK();
598 	SLIST_FOREACH(dcb, &V_divhash[DIVHASH(port)], dcb_next)
599 		if (dcb->dcb_port == port) {
600 			DIVERT_UNLOCK();
601 			return (EADDRINUSE);
602 		}
603 	dcb = so->so_pcb;
604 	if (dcb->dcb_bound != DCB_UNBOUND)
605 		SLIST_REMOVE(&V_divhash[DCBHASH(dcb)], dcb, divcb, dcb_next);
606 	dcb->dcb_port = port;
607 	SLIST_INSERT_HEAD(&V_divhash[DIVHASH(port)], dcb, dcb_next);
608 	DIVERT_UNLOCK();
609 
610 	return (0);
611 }
612 
613 static int
614 div_shutdown(struct socket *so)
615 {
616 
617 	socantsendmore(so);
618 	return 0;
619 }
620 
621 static int
622 div_pcblist(SYSCTL_HANDLER_ARGS)
623 {
624 	struct xinpgen xig;
625 	struct divcb *dcb;
626 	int error;
627 
628 	if (req->newptr != 0)
629 		return EPERM;
630 
631 	if (req->oldptr == 0) {
632 		u_int n;
633 
634 		n = V_dcb_count;
635 		n += imax(n / 8, 10);
636 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
637 		return 0;
638 	}
639 
640 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
641 		return (error);
642 
643 	bzero(&xig, sizeof(xig));
644 	xig.xig_len = sizeof xig;
645 	xig.xig_count = V_dcb_count;
646 	xig.xig_gen = V_dcb_gencnt;
647 	xig.xig_sogen = so_gencnt;
648 	error = SYSCTL_OUT(req, &xig, sizeof xig);
649 	if (error)
650 		return error;
651 
652 	DIVERT_LOCK();
653 	for (int i = 0; i < DIVHASHSIZE; i++)
654 		SLIST_FOREACH(dcb, &V_divhash[i], dcb_next) {
655 			if (dcb->dcb_gencnt <= xig.xig_gen) {
656 				struct xinpcb xi;
657 
658 				bzero(&xi, sizeof(xi));
659 				xi.xi_len = sizeof(struct xinpcb);
660 				sotoxsocket(dcb->dcb_socket, &xi.xi_socket);
661 				xi.inp_gencnt = dcb->dcb_gencnt;
662 				xi.inp_vflag = INP_IPV4; /* XXX: netstat(1) */
663 				xi.inp_inc.inc_ie.ie_lport = dcb->dcb_port;
664 				error = SYSCTL_OUT(req, &xi, sizeof xi);
665 				if (error)
666 					goto errout;
667 			}
668 		}
669 
670 	/*
671 	 * Give the user an updated idea of our state.
672 	 * If the generation differs from what we told
673 	 * her before, she knows that something happened
674 	 * while we were processing this request, and it
675 	 * might be necessary to retry.
676 	 */
677 	xig.xig_gen = V_dcb_gencnt;
678 	xig.xig_sogen = so_gencnt;
679 	xig.xig_count = V_dcb_count;
680 	error = SYSCTL_OUT(req, &xig, sizeof xig);
681 
682 errout:
683 	DIVERT_UNLOCK();
684 
685 	return (error);
686 }
687 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
688     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, div_pcblist,
689     "S,xinpcb", "List of active divert sockets");
690 
691 static struct protosw div_protosw = {
692 	.pr_type =		SOCK_RAW,
693 	.pr_flags =		PR_ATOMIC|PR_ADDR,
694 	.pr_attach =		div_attach,
695 	.pr_bind =		div_bind,
696 	.pr_detach =		div_detach,
697 	.pr_send =		div_send,
698 	.pr_shutdown =		div_shutdown,
699 };
700 
701 static struct domain divertdomain = {
702 	.dom_family =	PF_DIVERT,
703 	.dom_name =	"divert",
704 	.dom_nprotosw =	1,
705 	.dom_protosw =	{ &div_protosw },
706 };
707 
708 static int
709 div_modevent(module_t mod, int type, void *unused)
710 {
711 	int err = 0;
712 
713 	switch (type) {
714 	case MOD_LOAD:
715 		domain_add(&divertdomain);
716 		ip_divert_ptr = divert_packet;
717 		break;
718 	case MOD_QUIESCE:
719 		/*
720 		 * IPDIVERT may normally not be unloaded because of the
721 		 * potential race conditions.  Tell kldunload we can't be
722 		 * unloaded unless the unload is forced.
723 		 */
724 		err = EPERM;
725 		break;
726 	case MOD_UNLOAD:
727 		/*
728 		 * Forced unload.
729 		 *
730 		 * Module ipdivert can only be unloaded if no sockets are
731 		 * connected.  Maybe this can be changed later to forcefully
732 		 * disconnect any open sockets.
733 		 *
734 		 * XXXRW: Note that there is a slight race here, as a new
735 		 * socket open request could be spinning on the lock and then
736 		 * we destroy the lock.
737 		 *
738 		 * XXXGL: One more reason this code is incorrect is that it
739 		 * checks only the current vnet.
740 		 */
741 		DIVERT_LOCK();
742 		if (V_dcb_count != 0) {
743 			DIVERT_UNLOCK();
744 			err = EBUSY;
745 			break;
746 		}
747 		DIVERT_UNLOCK();
748 		ip_divert_ptr = NULL;
749 		domain_remove(&divertdomain);
750 		break;
751 	default:
752 		err = EOPNOTSUPP;
753 		break;
754 	}
755 	return err;
756 }
757 
758 static moduledata_t ipdivertmod = {
759         "ipdivert",
760         div_modevent,
761         0
762 };
763 
764 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
765 MODULE_VERSION(ipdivert, 1);
766