xref: /freebsd/sys/netinet/ip_divert.c (revision 3332f1b444d4a73238e9f59cca27bfc95fe936bd)
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 #ifdef INET6
231 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
232 		m = mb_unmapped_to_ext(m);
233 		if (m == NULL)
234 			return;
235 		in6_delayed_cksum(m, m->m_pkthdr.len -
236 		    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
237 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
238 	}
239 #if defined(SCTP) || defined(SCTP_SUPPORT)
240 	if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
241 		m = mb_unmapped_to_ext(m);
242 		if (m == NULL)
243 			return;
244 		sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
245 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
246 	}
247 #endif
248 #endif /* INET6 */
249 	bzero(&divsrc, sizeof(divsrc));
250 	divsrc.sin_len = sizeof(divsrc);
251 	divsrc.sin_family = AF_INET;
252 	/* record matching rule, in host format */
253 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
254 	/*
255 	 * Record receive interface address, if any.
256 	 * But only for incoming packets.
257 	 */
258 	if (incoming) {
259 		struct ifaddr *ifa;
260 		struct ifnet *ifp;
261 
262 		/* Sanity check */
263 		M_ASSERTPKTHDR(m);
264 
265 		/* Find IP address for receive interface */
266 		ifp = m->m_pkthdr.rcvif;
267 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
268 			if (ifa->ifa_addr->sa_family != AF_INET)
269 				continue;
270 			divsrc.sin_addr =
271 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
272 			break;
273 		}
274 	}
275 	/*
276 	 * Record the incoming interface name whenever we have one.
277 	 */
278 	if (m->m_pkthdr.rcvif) {
279 		/*
280 		 * Hide the actual interface name in there in the
281 		 * sin_zero array. XXX This needs to be moved to a
282 		 * different sockaddr type for divert, e.g.
283 		 * sockaddr_div with multiple fields like
284 		 * sockaddr_dl. Presently we have only 7 bytes
285 		 * but that will do for now as most interfaces
286 		 * are 4 or less + 2 or less bytes for unit.
287 		 * There is probably a faster way of doing this,
288 		 * possibly taking it from the sockaddr_dl on the iface.
289 		 * This solves the problem of a P2P link and a LAN interface
290 		 * having the same address, which can result in the wrong
291 		 * interface being assigned to the packet when fed back
292 		 * into the divert socket. Theoretically if the daemon saves
293 		 * and re-uses the sockaddr_in as suggested in the man pages,
294 		 * this iface name will come along for the ride.
295 		 * (see div_output for the other half of this.)
296 		 */
297 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
298 		    sizeof(divsrc.sin_zero));
299 	}
300 
301 	/* Put packet on socket queue, if any */
302 	sa = NULL;
303 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
304 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
305 		/* XXX why does only one socket match? */
306 		if (inp->inp_lport == nport) {
307 			INP_RLOCK(inp);
308 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
309 				INP_RUNLOCK(inp);
310 				continue;
311 			}
312 			sa = inp->inp_socket;
313 			SOCKBUF_LOCK(&sa->so_rcv);
314 			if (sbappendaddr_locked(&sa->so_rcv,
315 			    (struct sockaddr *)&divsrc, m,
316 			    (struct mbuf *)0) == 0) {
317 				soroverflow_locked(sa);
318 				sa = NULL;	/* force mbuf reclaim below */
319 			} else
320 				sorwakeup_locked(sa);
321 			INP_RUNLOCK(inp);
322 			break;
323 		}
324 	}
325 	if (sa == NULL) {
326 		m_freem(m);
327 		KMOD_IPSTAT_INC(ips_noproto);
328 		KMOD_IPSTAT_DEC(ips_delivered);
329         }
330 }
331 
332 /*
333  * Deliver packet back into the IP processing machinery.
334  *
335  * If no address specified, or address is 0.0.0.0, send to ip_output();
336  * otherwise, send to ip_input() and mark as having been received on
337  * the interface with that address.
338  */
339 static int
340 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
341     struct mbuf *control)
342 {
343 	struct epoch_tracker et;
344 	const struct ip *ip;
345 	struct m_tag *mtag;
346 	struct ipfw_rule_ref *dt;
347 	int error, family;
348 
349 	if (control) {
350 		m_freem(control);		/* XXX */
351 		control = NULL;
352 	}
353 
354 	if (sin != NULL) {
355 		if (sin->sin_family != AF_INET) {
356 			m_freem(m);
357 			return (EAFNOSUPPORT);
358 		}
359 		if (sin->sin_len != sizeof(*sin)) {
360 			m_freem(m);
361 			return (EINVAL);
362 		}
363 	}
364 
365 	/*
366 	 * An mbuf may hasn't come from userland, but we pretend
367 	 * that it has.
368 	 */
369 	m->m_pkthdr.rcvif = NULL;
370 	m->m_nextpkt = NULL;
371 	M_SETFIB(m, so->so_fibnum);
372 
373 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
374 	if (mtag == NULL) {
375 		/* this should be normal */
376 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
377 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
378 		if (mtag == NULL) {
379 			m_freem(m);
380 			return (ENOBUFS);
381 		}
382 		m_tag_prepend(m, mtag);
383 	}
384 	dt = (struct ipfw_rule_ref *)(mtag+1);
385 
386 	/* Loopback avoidance and state recovery */
387 	if (sin) {
388 		int i;
389 
390 		/* set the starting point. We provide a non-zero slot,
391 		 * but a non_matching chain_id to skip that info and use
392 		 * the rulenum/rule_id.
393 		 */
394 		dt->slot = 1; /* dummy, chain_id is invalid */
395 		dt->chain_id = 0;
396 		dt->rulenum = sin->sin_port+1; /* host format ? */
397 		dt->rule_id = 0;
398 		/* XXX: broken for IPv6 */
399 		/*
400 		 * Find receive interface with the given name, stuffed
401 		 * (if it exists) in the sin_zero[] field.
402 		 * The name is user supplied data so don't trust its size
403 		 * or that it is zero terminated.
404 		 */
405 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
406 			;
407 		if ( i > 0 && i < sizeof(sin->sin_zero))
408 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
409 	}
410 
411 	ip = mtod(m, struct ip *);
412 	switch (ip->ip_v) {
413 	case IPVERSION:
414 		family = AF_INET;
415 		break;
416 #ifdef INET6
417 	case IPV6_VERSION >> 4:
418 		family = AF_INET6;
419 		break;
420 #endif
421 	default:
422 		m_freem(m);
423 		return (EAFNOSUPPORT);
424 	}
425 
426 	/* Reinject packet into the system as incoming or outgoing */
427 	NET_EPOCH_ENTER(et);
428 	if (!sin || sin->sin_addr.s_addr == 0) {
429 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
430 		error = div_output_outbound(family, so, m);
431 	} else {
432 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
433 		error = div_output_inbound(family, so, m, sin);
434 	}
435 	NET_EPOCH_EXIT(et);
436 
437 	return (error);
438 }
439 
440 /*
441  * Sends mbuf @m to the wire via ip[6]_output().
442  *
443  * Returns 0 on success or an errno value on failure.  @m is always consumed.
444  */
445 static int
446 div_output_outbound(int family, struct socket *so, struct mbuf *m)
447 {
448 	struct ip *const ip = mtod(m, struct ip *);
449 	struct mbuf *options;
450 	struct inpcb *inp;
451 	int error;
452 
453 	inp = sotoinpcb(so);
454 	INP_RLOCK(inp);
455 	switch (family) {
456 	case AF_INET:
457 		/*
458 		 * Don't allow both user specified and setsockopt
459 		 * options, and don't allow packet length sizes that
460 		 * will crash.
461 		 */
462 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
463 		    inp->inp_options != NULL) ||
464 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
465 			INP_RUNLOCK(inp);
466 			m_freem(m);
467 			return (EINVAL);
468 		}
469 		break;
470 #ifdef INET6
471 	case AF_INET6:
472 	    {
473 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
474 
475 		/* Don't allow packet length sizes that will crash */
476 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
477 			INP_RUNLOCK(inp);
478 			m_freem(m);
479 			return (EINVAL);
480 		}
481 		break;
482 	    }
483 #endif
484 	}
485 
486 	/* Send packet to output processing */
487 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
488 
489 #ifdef MAC
490 	mac_inpcb_create_mbuf(inp, m);
491 #endif
492 	/*
493 	 * Get ready to inject the packet into ip_output().
494 	 * Just in case socket options were specified on the
495 	 * divert socket, we duplicate them.  This is done
496 	 * to avoid having to hold the PCB locks over the call
497 	 * to ip_output(), as doing this results in a number of
498 	 * lock ordering complexities.
499 	 *
500 	 * Note that we set the multicast options argument for
501 	 * ip_output() to NULL since it should be invariant that
502 	 * they are not present.
503 	 */
504 	KASSERT(inp->inp_moptions == NULL,
505 	    ("multicast options set on a divert socket"));
506 	/*
507 	 * XXXCSJP: It is unclear to me whether or not it makes
508 	 * sense for divert sockets to have options.  However,
509 	 * for now we will duplicate them with the INP locks
510 	 * held so we can use them in ip_output() without
511 	 * requring a reference to the pcb.
512 	 */
513 	options = NULL;
514 	if (inp->inp_options != NULL) {
515 		options = m_dup(inp->inp_options, M_NOWAIT);
516 		if (options == NULL) {
517 			INP_RUNLOCK(inp);
518 			m_freem(m);
519 			return (ENOBUFS);
520 		}
521 	}
522 	INP_RUNLOCK(inp);
523 
524 	error = 0;
525 	switch (family) {
526 	case AF_INET:
527 		error = ip_output(m, options, NULL,
528 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
529 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
530 		break;
531 #ifdef INET6
532 	case AF_INET6:
533 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
534 		break;
535 #endif
536 	}
537 	if (options != NULL)
538 		m_freem(options);
539 
540 	return (error);
541 }
542 
543 /*
544  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
545  *
546  * Returns 0 on success or an errno value on failure.  @m is always consumed.
547  */
548 static int
549 div_output_inbound(int family, struct socket *so, struct mbuf *m,
550     struct sockaddr_in *sin)
551 {
552 	const struct ip *ip;
553 	struct ifaddr *ifa;
554 
555 	if (m->m_pkthdr.rcvif == NULL) {
556 		/*
557 		 * No luck with the name, check by IP address.
558 		 * Clear the port and the ifname to make sure
559 		 * there are no distractions for ifa_ifwithaddr.
560 		 */
561 
562 		/* XXX: broken for IPv6 */
563 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
564 		sin->sin_port = 0;
565 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
566 		if (ifa == NULL) {
567 			m_freem(m);
568 			return (EADDRNOTAVAIL);
569 		}
570 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
571 	}
572 #ifdef MAC
573 	mac_socket_create_mbuf(so, m);
574 #endif
575 	/* Send packet to input processing via netisr */
576 	switch (family) {
577 	case AF_INET:
578 		ip = mtod(m, struct ip *);
579 		/*
580 		 * Restore M_BCAST flag when destination address is
581 		 * broadcast. It is expected by ip_tryforward().
582 		 */
583 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
584 			m->m_flags |= M_MCAST;
585 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
586 			m->m_flags |= M_BCAST;
587 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
588 		break;
589 #ifdef INET6
590 	case AF_INET6:
591 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
592 		break;
593 #endif
594 	default:
595 		m_freem(m);
596 		return (EINVAL);
597 	}
598 
599 	return (0);
600 }
601 
602 static int
603 div_attach(struct socket *so, int proto, struct thread *td)
604 {
605 	struct inpcb *inp;
606 	int error;
607 
608 	inp  = sotoinpcb(so);
609 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
610 	if (td != NULL) {
611 		error = priv_check(td, PRIV_NETINET_DIVERT);
612 		if (error)
613 			return (error);
614 	}
615 	error = soreserve(so, div_sendspace, div_recvspace);
616 	if (error)
617 		return error;
618 	INP_INFO_WLOCK(&V_divcbinfo);
619 	error = in_pcballoc(so, &V_divcbinfo);
620 	if (error) {
621 		INP_INFO_WUNLOCK(&V_divcbinfo);
622 		return error;
623 	}
624 	inp = (struct inpcb *)so->so_pcb;
625 	INP_INFO_WUNLOCK(&V_divcbinfo);
626 	inp->inp_ip_p = proto;
627 	inp->inp_vflag |= INP_IPV4;
628 	inp->inp_flags |= INP_HDRINCL;
629 	INP_WUNLOCK(inp);
630 	return 0;
631 }
632 
633 static void
634 div_detach(struct socket *so)
635 {
636 	struct inpcb *inp;
637 
638 	inp = sotoinpcb(so);
639 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
640 	INP_INFO_WLOCK(&V_divcbinfo);
641 	INP_WLOCK(inp);
642 	in_pcbdetach(inp);
643 	in_pcbfree(inp);
644 	INP_INFO_WUNLOCK(&V_divcbinfo);
645 }
646 
647 static int
648 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
649 {
650 	struct inpcb *inp;
651 	int error;
652 
653 	inp = sotoinpcb(so);
654 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
655 	/* in_pcbbind assumes that nam is a sockaddr_in
656 	 * and in_pcbbind requires a valid address. Since divert
657 	 * sockets don't we need to make sure the address is
658 	 * filled in properly.
659 	 * XXX -- divert should not be abusing in_pcbind
660 	 * and should probably have its own family.
661 	 */
662 	if (nam->sa_family != AF_INET)
663 		return EAFNOSUPPORT;
664 	if (nam->sa_len != sizeof(struct sockaddr_in))
665 		return EINVAL;
666 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
667 	INP_INFO_WLOCK(&V_divcbinfo);
668 	INP_WLOCK(inp);
669 	INP_HASH_WLOCK(&V_divcbinfo);
670 	error = in_pcbbind(inp, nam, td->td_ucred);
671 	INP_HASH_WUNLOCK(&V_divcbinfo);
672 	INP_WUNLOCK(inp);
673 	INP_INFO_WUNLOCK(&V_divcbinfo);
674 	return error;
675 }
676 
677 static int
678 div_shutdown(struct socket *so)
679 {
680 	struct inpcb *inp;
681 
682 	inp = sotoinpcb(so);
683 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
684 	INP_WLOCK(inp);
685 	socantsendmore(so);
686 	INP_WUNLOCK(inp);
687 	return 0;
688 }
689 
690 static int
691 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
692     struct mbuf *control, struct thread *td)
693 {
694 
695 	/* Packet must have a header (but that's about it) */
696 	if (m->m_len < sizeof (struct ip) &&
697 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
698 		KMOD_IPSTAT_INC(ips_toosmall);
699 		if (control != NULL)
700 			m_freem(control);
701 		m_freem(m);
702 		return EINVAL;
703 	}
704 
705 	/* Send packet */
706 	return div_output(so, m, (struct sockaddr_in *)nam, control);
707 }
708 
709 static int
710 div_pcblist(SYSCTL_HANDLER_ARGS)
711 {
712 	struct xinpgen xig;
713 	struct epoch_tracker et;
714 	struct inpcb *inp;
715 	int error;
716 
717 	if (req->newptr != 0)
718 		return EPERM;
719 
720 	if (req->oldptr == 0) {
721 		int n;
722 
723 		n = V_divcbinfo.ipi_count;
724 		n += imax(n / 8, 10);
725 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
726 		return 0;
727 	}
728 
729 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
730 		return (error);
731 
732 	bzero(&xig, sizeof(xig));
733 	xig.xig_len = sizeof xig;
734 	xig.xig_count = V_divcbinfo.ipi_count;
735 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
736 	xig.xig_sogen = so_gencnt;
737 	error = SYSCTL_OUT(req, &xig, sizeof xig);
738 	if (error)
739 		return error;
740 
741 	NET_EPOCH_ENTER(et);
742 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
743 	    inp != NULL;
744 	    inp = CK_LIST_NEXT(inp, inp_list)) {
745 		INP_RLOCK(inp);
746 		if (inp->inp_gencnt <= xig.xig_gen) {
747 			struct xinpcb xi;
748 
749 			in_pcbtoxinpcb(inp, &xi);
750 			INP_RUNLOCK(inp);
751 			error = SYSCTL_OUT(req, &xi, sizeof xi);
752 		} else
753 			INP_RUNLOCK(inp);
754 	}
755 	NET_EPOCH_EXIT(et);
756 
757 	if (!error) {
758 		/*
759 		 * Give the user an updated idea of our state.
760 		 * If the generation differs from what we told
761 		 * her before, she knows that something happened
762 		 * while we were processing this request, and it
763 		 * might be necessary to retry.
764 		 */
765 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
766 		xig.xig_sogen = so_gencnt;
767 		xig.xig_count = V_divcbinfo.ipi_count;
768 		error = SYSCTL_OUT(req, &xig, sizeof xig);
769 	}
770 
771 	return (error);
772 }
773 
774 #ifdef SYSCTL_NODE
775 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
776     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
777     "IPDIVERT");
778 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
779    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
780     NULL, 0, div_pcblist, "S,xinpcb",
781     "List of active divert sockets");
782 #endif
783 
784 struct pr_usrreqs div_usrreqs = {
785 	.pru_attach =		div_attach,
786 	.pru_bind =		div_bind,
787 	.pru_control =		in_control,
788 	.pru_detach =		div_detach,
789 	.pru_peeraddr =		in_getpeeraddr,
790 	.pru_send =		div_send,
791 	.pru_shutdown =		div_shutdown,
792 	.pru_sockaddr =		in_getsockaddr,
793 	.pru_sosetlabel =	in_pcbsosetlabel
794 };
795 
796 struct protosw div_protosw = {
797 	.pr_type =		SOCK_RAW,
798 	.pr_protocol =		IPPROTO_DIVERT,
799 	.pr_flags =		PR_ATOMIC|PR_ADDR,
800 	.pr_input =		div_input,
801 	.pr_init =		div_init,
802 	.pr_usrreqs =		&div_usrreqs
803 };
804 
805 static int
806 div_modevent(module_t mod, int type, void *unused)
807 {
808 	int err = 0;
809 
810 	switch (type) {
811 	case MOD_LOAD:
812 		/*
813 		 * Protocol will be initialized by pf_proto_register().
814 		 * We don't have to register ip_protox because we are not
815 		 * a true IP protocol that goes over the wire.
816 		 */
817 		err = pf_proto_register(PF_INET, &div_protosw);
818 		if (err != 0)
819 			return (err);
820 		ip_divert_ptr = divert_packet;
821 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
822 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
823 		break;
824 	case MOD_QUIESCE:
825 		/*
826 		 * IPDIVERT may normally not be unloaded because of the
827 		 * potential race conditions.  Tell kldunload we can't be
828 		 * unloaded unless the unload is forced.
829 		 */
830 		err = EPERM;
831 		break;
832 	case MOD_UNLOAD:
833 		/*
834 		 * Forced unload.
835 		 *
836 		 * Module ipdivert can only be unloaded if no sockets are
837 		 * connected.  Maybe this can be changed later to forcefully
838 		 * disconnect any open sockets.
839 		 *
840 		 * XXXRW: Note that there is a slight race here, as a new
841 		 * socket open request could be spinning on the lock and then
842 		 * we destroy the lock.
843 		 */
844 		INP_INFO_WLOCK(&V_divcbinfo);
845 		if (V_divcbinfo.ipi_count != 0) {
846 			err = EBUSY;
847 			INP_INFO_WUNLOCK(&V_divcbinfo);
848 			break;
849 		}
850 		ip_divert_ptr = NULL;
851 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
852 		INP_INFO_WUNLOCK(&V_divcbinfo);
853 #ifndef VIMAGE
854 		div_destroy(NULL);
855 #endif
856 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
857 		break;
858 	default:
859 		err = EOPNOTSUPP;
860 		break;
861 	}
862 	return err;
863 }
864 
865 static moduledata_t ipdivertmod = {
866         "ipdivert",
867         div_modevent,
868         0
869 };
870 
871 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
872 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
873 MODULE_VERSION(ipdivert, 1);
874