xref: /freebsd/sys/netinet/ip_divert.c (revision bfcc09ddd422c95a1a2e4e794b63ee54c4902398)
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 		in_delayed_cksum(m);
216 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
217 	}
218 #if defined(SCTP) || defined(SCTP_SUPPORT)
219 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
220 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
221 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
222 	}
223 #endif
224 #ifdef INET6
225 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
226 		in6_delayed_cksum(m, m->m_pkthdr.len -
227 		    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
228 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
229 	}
230 #if defined(SCTP) || defined(SCTP_SUPPORT)
231 	if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
232 		sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
233 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
234 	}
235 #endif
236 #endif /* INET6 */
237 	bzero(&divsrc, sizeof(divsrc));
238 	divsrc.sin_len = sizeof(divsrc);
239 	divsrc.sin_family = AF_INET;
240 	/* record matching rule, in host format */
241 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
242 	/*
243 	 * Record receive interface address, if any.
244 	 * But only for incoming packets.
245 	 */
246 	if (incoming) {
247 		struct ifaddr *ifa;
248 		struct ifnet *ifp;
249 
250 		/* Sanity check */
251 		M_ASSERTPKTHDR(m);
252 
253 		/* Find IP address for receive interface */
254 		ifp = m->m_pkthdr.rcvif;
255 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
256 			if (ifa->ifa_addr->sa_family != AF_INET)
257 				continue;
258 			divsrc.sin_addr =
259 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
260 			break;
261 		}
262 	}
263 	/*
264 	 * Record the incoming interface name whenever we have one.
265 	 */
266 	if (m->m_pkthdr.rcvif) {
267 		/*
268 		 * Hide the actual interface name in there in the
269 		 * sin_zero array. XXX This needs to be moved to a
270 		 * different sockaddr type for divert, e.g.
271 		 * sockaddr_div with multiple fields like
272 		 * sockaddr_dl. Presently we have only 7 bytes
273 		 * but that will do for now as most interfaces
274 		 * are 4 or less + 2 or less bytes for unit.
275 		 * There is probably a faster way of doing this,
276 		 * possibly taking it from the sockaddr_dl on the iface.
277 		 * This solves the problem of a P2P link and a LAN interface
278 		 * having the same address, which can result in the wrong
279 		 * interface being assigned to the packet when fed back
280 		 * into the divert socket. Theoretically if the daemon saves
281 		 * and re-uses the sockaddr_in as suggested in the man pages,
282 		 * this iface name will come along for the ride.
283 		 * (see div_output for the other half of this.)
284 		 */
285 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
286 		    sizeof(divsrc.sin_zero));
287 	}
288 
289 	/* Put packet on socket queue, if any */
290 	sa = NULL;
291 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
292 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
293 		/* XXX why does only one socket match? */
294 		if (inp->inp_lport == nport) {
295 			INP_RLOCK(inp);
296 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
297 				INP_RUNLOCK(inp);
298 				continue;
299 			}
300 			sa = inp->inp_socket;
301 			SOCKBUF_LOCK(&sa->so_rcv);
302 			if (sbappendaddr_locked(&sa->so_rcv,
303 			    (struct sockaddr *)&divsrc, m,
304 			    (struct mbuf *)0) == 0) {
305 				soroverflow_locked(sa);
306 				sa = NULL;	/* force mbuf reclaim below */
307 			} else
308 				sorwakeup_locked(sa);
309 			INP_RUNLOCK(inp);
310 			break;
311 		}
312 	}
313 	if (sa == NULL) {
314 		m_freem(m);
315 		KMOD_IPSTAT_INC(ips_noproto);
316 		KMOD_IPSTAT_DEC(ips_delivered);
317         }
318 }
319 
320 /*
321  * Deliver packet back into the IP processing machinery.
322  *
323  * If no address specified, or address is 0.0.0.0, send to ip_output();
324  * otherwise, send to ip_input() and mark as having been received on
325  * the interface with that address.
326  */
327 static int
328 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
329     struct mbuf *control)
330 {
331 	struct epoch_tracker et;
332 	const struct ip *ip;
333 	struct m_tag *mtag;
334 	struct ipfw_rule_ref *dt;
335 	int error, family;
336 
337 	if (control) {
338 		m_freem(control);		/* XXX */
339 		control = NULL;
340 	}
341 
342 	if (sin != NULL) {
343 		if (sin->sin_family != AF_INET) {
344 			m_freem(m);
345 			return (EAFNOSUPPORT);
346 		}
347 		if (sin->sin_len != sizeof(*sin)) {
348 			m_freem(m);
349 			return (EINVAL);
350 		}
351 	}
352 
353 	/*
354 	 * An mbuf may hasn't come from userland, but we pretend
355 	 * that it has.
356 	 */
357 	m->m_pkthdr.rcvif = NULL;
358 	m->m_nextpkt = NULL;
359 	M_SETFIB(m, so->so_fibnum);
360 
361 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
362 	if (mtag == NULL) {
363 		/* this should be normal */
364 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
365 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
366 		if (mtag == NULL) {
367 			m_freem(m);
368 			return (ENOBUFS);
369 		}
370 		m_tag_prepend(m, mtag);
371 	}
372 	dt = (struct ipfw_rule_ref *)(mtag+1);
373 
374 	/* Loopback avoidance and state recovery */
375 	if (sin) {
376 		int i;
377 
378 		/* set the starting point. We provide a non-zero slot,
379 		 * but a non_matching chain_id to skip that info and use
380 		 * the rulenum/rule_id.
381 		 */
382 		dt->slot = 1; /* dummy, chain_id is invalid */
383 		dt->chain_id = 0;
384 		dt->rulenum = sin->sin_port+1; /* host format ? */
385 		dt->rule_id = 0;
386 		/* XXX: broken for IPv6 */
387 		/*
388 		 * Find receive interface with the given name, stuffed
389 		 * (if it exists) in the sin_zero[] field.
390 		 * The name is user supplied data so don't trust its size
391 		 * or that it is zero terminated.
392 		 */
393 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
394 			;
395 		if ( i > 0 && i < sizeof(sin->sin_zero))
396 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
397 	}
398 
399 	ip = mtod(m, struct ip *);
400 	switch (ip->ip_v) {
401 	case IPVERSION:
402 		family = AF_INET;
403 		break;
404 #ifdef INET6
405 	case IPV6_VERSION >> 4:
406 		family = AF_INET6;
407 		break;
408 #endif
409 	default:
410 		m_freem(m);
411 		return (EAFNOSUPPORT);
412 	}
413 
414 	/* Reinject packet into the system as incoming or outgoing */
415 	NET_EPOCH_ENTER(et);
416 	if (!sin || sin->sin_addr.s_addr == 0) {
417 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
418 		error = div_output_outbound(family, so, m);
419 	} else {
420 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
421 		error = div_output_inbound(family, so, m, sin);
422 	}
423 	NET_EPOCH_EXIT(et);
424 
425 	return (error);
426 }
427 
428 /*
429  * Sends mbuf @m to the wire via ip[6]_output().
430  *
431  * Returns 0 on success or an errno value on failure.  @m is always consumed.
432  */
433 static int
434 div_output_outbound(int family, struct socket *so, struct mbuf *m)
435 {
436 	struct ip *const ip = mtod(m, struct ip *);
437 	struct mbuf *options;
438 	struct inpcb *inp;
439 	int error;
440 
441 	inp = sotoinpcb(so);
442 	INP_RLOCK(inp);
443 	switch (family) {
444 	case AF_INET:
445 		/*
446 		 * Don't allow both user specified and setsockopt
447 		 * options, and don't allow packet length sizes that
448 		 * will crash.
449 		 */
450 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
451 		    inp->inp_options != NULL) ||
452 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
453 			INP_RUNLOCK(inp);
454 			m_freem(m);
455 			return (EINVAL);
456 		}
457 		break;
458 #ifdef INET6
459 	case AF_INET6:
460 	    {
461 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
462 
463 		/* Don't allow packet length sizes that will crash */
464 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
465 			INP_RUNLOCK(inp);
466 			m_freem(m);
467 			return (EINVAL);
468 		}
469 		break;
470 	    }
471 #endif
472 	}
473 
474 	/* Send packet to output processing */
475 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
476 
477 #ifdef MAC
478 	mac_inpcb_create_mbuf(inp, m);
479 #endif
480 	/*
481 	 * Get ready to inject the packet into ip_output().
482 	 * Just in case socket options were specified on the
483 	 * divert socket, we duplicate them.  This is done
484 	 * to avoid having to hold the PCB locks over the call
485 	 * to ip_output(), as doing this results in a number of
486 	 * lock ordering complexities.
487 	 *
488 	 * Note that we set the multicast options argument for
489 	 * ip_output() to NULL since it should be invariant that
490 	 * they are not present.
491 	 */
492 	KASSERT(inp->inp_moptions == NULL,
493 	    ("multicast options set on a divert socket"));
494 	/*
495 	 * XXXCSJP: It is unclear to me whether or not it makes
496 	 * sense for divert sockets to have options.  However,
497 	 * for now we will duplicate them with the INP locks
498 	 * held so we can use them in ip_output() without
499 	 * requring a reference to the pcb.
500 	 */
501 	options = NULL;
502 	if (inp->inp_options != NULL) {
503 		options = m_dup(inp->inp_options, M_NOWAIT);
504 		if (options == NULL) {
505 			INP_RUNLOCK(inp);
506 			m_freem(m);
507 			return (ENOBUFS);
508 		}
509 	}
510 	INP_RUNLOCK(inp);
511 
512 	error = 0;
513 	switch (family) {
514 	case AF_INET:
515 		error = ip_output(m, options, NULL,
516 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
517 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
518 		break;
519 #ifdef INET6
520 	case AF_INET6:
521 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
522 		break;
523 #endif
524 	}
525 	if (options != NULL)
526 		m_freem(options);
527 
528 	return (error);
529 }
530 
531 /*
532  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
533  *
534  * Returns 0 on success or an errno value on failure.  @m is always consumed.
535  */
536 static int
537 div_output_inbound(int family, struct socket *so, struct mbuf *m,
538     struct sockaddr_in *sin)
539 {
540 	const struct ip *ip;
541 	struct ifaddr *ifa;
542 
543 	if (m->m_pkthdr.rcvif == NULL) {
544 		/*
545 		 * No luck with the name, check by IP address.
546 		 * Clear the port and the ifname to make sure
547 		 * there are no distractions for ifa_ifwithaddr.
548 		 */
549 
550 		/* XXX: broken for IPv6 */
551 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
552 		sin->sin_port = 0;
553 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
554 		if (ifa == NULL) {
555 			m_freem(m);
556 			return (EADDRNOTAVAIL);
557 		}
558 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
559 	}
560 #ifdef MAC
561 	mac_socket_create_mbuf(so, m);
562 #endif
563 	/* Send packet to input processing via netisr */
564 	switch (family) {
565 	case AF_INET:
566 		ip = mtod(m, struct ip *);
567 		/*
568 		 * Restore M_BCAST flag when destination address is
569 		 * broadcast. It is expected by ip_tryforward().
570 		 */
571 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
572 			m->m_flags |= M_MCAST;
573 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
574 			m->m_flags |= M_BCAST;
575 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
576 		break;
577 #ifdef INET6
578 	case AF_INET6:
579 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
580 		break;
581 #endif
582 	default:
583 		m_freem(m);
584 		return (EINVAL);
585 	}
586 
587 	return (0);
588 }
589 
590 static int
591 div_attach(struct socket *so, int proto, struct thread *td)
592 {
593 	struct inpcb *inp;
594 	int error;
595 
596 	inp  = sotoinpcb(so);
597 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
598 	if (td != NULL) {
599 		error = priv_check(td, PRIV_NETINET_DIVERT);
600 		if (error)
601 			return (error);
602 	}
603 	error = soreserve(so, div_sendspace, div_recvspace);
604 	if (error)
605 		return error;
606 	INP_INFO_WLOCK(&V_divcbinfo);
607 	error = in_pcballoc(so, &V_divcbinfo);
608 	if (error) {
609 		INP_INFO_WUNLOCK(&V_divcbinfo);
610 		return error;
611 	}
612 	inp = (struct inpcb *)so->so_pcb;
613 	INP_INFO_WUNLOCK(&V_divcbinfo);
614 	inp->inp_ip_p = proto;
615 	inp->inp_vflag |= INP_IPV4;
616 	inp->inp_flags |= INP_HDRINCL;
617 	INP_WUNLOCK(inp);
618 	return 0;
619 }
620 
621 static void
622 div_detach(struct socket *so)
623 {
624 	struct inpcb *inp;
625 
626 	inp = sotoinpcb(so);
627 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
628 	INP_INFO_WLOCK(&V_divcbinfo);
629 	INP_WLOCK(inp);
630 	in_pcbdetach(inp);
631 	in_pcbfree(inp);
632 	INP_INFO_WUNLOCK(&V_divcbinfo);
633 }
634 
635 static int
636 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
637 {
638 	struct inpcb *inp;
639 	int error;
640 
641 	inp = sotoinpcb(so);
642 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
643 	/* in_pcbbind assumes that nam is a sockaddr_in
644 	 * and in_pcbbind requires a valid address. Since divert
645 	 * sockets don't we need to make sure the address is
646 	 * filled in properly.
647 	 * XXX -- divert should not be abusing in_pcbind
648 	 * and should probably have its own family.
649 	 */
650 	if (nam->sa_family != AF_INET)
651 		return EAFNOSUPPORT;
652 	if (nam->sa_len != sizeof(struct sockaddr_in))
653 		return EINVAL;
654 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
655 	INP_INFO_WLOCK(&V_divcbinfo);
656 	INP_WLOCK(inp);
657 	INP_HASH_WLOCK(&V_divcbinfo);
658 	error = in_pcbbind(inp, nam, td->td_ucred);
659 	INP_HASH_WUNLOCK(&V_divcbinfo);
660 	INP_WUNLOCK(inp);
661 	INP_INFO_WUNLOCK(&V_divcbinfo);
662 	return error;
663 }
664 
665 static int
666 div_shutdown(struct socket *so)
667 {
668 	struct inpcb *inp;
669 
670 	inp = sotoinpcb(so);
671 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
672 	INP_WLOCK(inp);
673 	socantsendmore(so);
674 	INP_WUNLOCK(inp);
675 	return 0;
676 }
677 
678 static int
679 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
680     struct mbuf *control, struct thread *td)
681 {
682 
683 	/* Packet must have a header (but that's about it) */
684 	if (m->m_len < sizeof (struct ip) &&
685 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
686 		KMOD_IPSTAT_INC(ips_toosmall);
687 		if (control != NULL)
688 			m_freem(control);
689 		m_freem(m);
690 		return EINVAL;
691 	}
692 
693 	/* Send packet */
694 	return div_output(so, m, (struct sockaddr_in *)nam, control);
695 }
696 
697 static int
698 div_pcblist(SYSCTL_HANDLER_ARGS)
699 {
700 	struct xinpgen xig;
701 	struct epoch_tracker et;
702 	struct inpcb *inp;
703 	int error;
704 
705 	if (req->newptr != 0)
706 		return EPERM;
707 
708 	if (req->oldptr == 0) {
709 		int n;
710 
711 		n = V_divcbinfo.ipi_count;
712 		n += imax(n / 8, 10);
713 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
714 		return 0;
715 	}
716 
717 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
718 		return (error);
719 
720 	bzero(&xig, sizeof(xig));
721 	xig.xig_len = sizeof xig;
722 	xig.xig_count = V_divcbinfo.ipi_count;
723 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
724 	xig.xig_sogen = so_gencnt;
725 	error = SYSCTL_OUT(req, &xig, sizeof xig);
726 	if (error)
727 		return error;
728 
729 	NET_EPOCH_ENTER(et);
730 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
731 	    inp != NULL;
732 	    inp = CK_LIST_NEXT(inp, inp_list)) {
733 		INP_RLOCK(inp);
734 		if (inp->inp_gencnt <= xig.xig_gen) {
735 			struct xinpcb xi;
736 
737 			in_pcbtoxinpcb(inp, &xi);
738 			INP_RUNLOCK(inp);
739 			error = SYSCTL_OUT(req, &xi, sizeof xi);
740 		} else
741 			INP_RUNLOCK(inp);
742 	}
743 	NET_EPOCH_EXIT(et);
744 
745 	if (!error) {
746 		/*
747 		 * Give the user an updated idea of our state.
748 		 * If the generation differs from what we told
749 		 * her before, she knows that something happened
750 		 * while we were processing this request, and it
751 		 * might be necessary to retry.
752 		 */
753 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
754 		xig.xig_sogen = so_gencnt;
755 		xig.xig_count = V_divcbinfo.ipi_count;
756 		error = SYSCTL_OUT(req, &xig, sizeof xig);
757 	}
758 
759 	return (error);
760 }
761 
762 #ifdef SYSCTL_NODE
763 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
764     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
765     "IPDIVERT");
766 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
767    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
768     NULL, 0, div_pcblist, "S,xinpcb",
769     "List of active divert sockets");
770 #endif
771 
772 struct pr_usrreqs div_usrreqs = {
773 	.pru_attach =		div_attach,
774 	.pru_bind =		div_bind,
775 	.pru_control =		in_control,
776 	.pru_detach =		div_detach,
777 	.pru_peeraddr =		in_getpeeraddr,
778 	.pru_send =		div_send,
779 	.pru_shutdown =		div_shutdown,
780 	.pru_sockaddr =		in_getsockaddr,
781 	.pru_sosetlabel =	in_pcbsosetlabel
782 };
783 
784 struct protosw div_protosw = {
785 	.pr_type =		SOCK_RAW,
786 	.pr_protocol =		IPPROTO_DIVERT,
787 	.pr_flags =		PR_ATOMIC|PR_ADDR,
788 	.pr_input =		div_input,
789 	.pr_init =		div_init,
790 	.pr_usrreqs =		&div_usrreqs
791 };
792 
793 static int
794 div_modevent(module_t mod, int type, void *unused)
795 {
796 	int err = 0;
797 
798 	switch (type) {
799 	case MOD_LOAD:
800 		/*
801 		 * Protocol will be initialized by pf_proto_register().
802 		 * We don't have to register ip_protox because we are not
803 		 * a true IP protocol that goes over the wire.
804 		 */
805 		err = pf_proto_register(PF_INET, &div_protosw);
806 		if (err != 0)
807 			return (err);
808 		ip_divert_ptr = divert_packet;
809 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
810 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
811 		break;
812 	case MOD_QUIESCE:
813 		/*
814 		 * IPDIVERT may normally not be unloaded because of the
815 		 * potential race conditions.  Tell kldunload we can't be
816 		 * unloaded unless the unload is forced.
817 		 */
818 		err = EPERM;
819 		break;
820 	case MOD_UNLOAD:
821 		/*
822 		 * Forced unload.
823 		 *
824 		 * Module ipdivert can only be unloaded if no sockets are
825 		 * connected.  Maybe this can be changed later to forcefully
826 		 * disconnect any open sockets.
827 		 *
828 		 * XXXRW: Note that there is a slight race here, as a new
829 		 * socket open request could be spinning on the lock and then
830 		 * we destroy the lock.
831 		 */
832 		INP_INFO_WLOCK(&V_divcbinfo);
833 		if (V_divcbinfo.ipi_count != 0) {
834 			err = EBUSY;
835 			INP_INFO_WUNLOCK(&V_divcbinfo);
836 			break;
837 		}
838 		ip_divert_ptr = NULL;
839 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
840 		INP_INFO_WUNLOCK(&V_divcbinfo);
841 #ifndef VIMAGE
842 		div_destroy(NULL);
843 #endif
844 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
845 		break;
846 	default:
847 		err = EOPNOTSUPP;
848 		break;
849 	}
850 	return err;
851 }
852 
853 static moduledata_t ipdivertmod = {
854         "ipdivert",
855         div_modevent,
856         0
857 };
858 
859 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
860 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
861 MODULE_VERSION(ipdivert, 1);
862