xref: /freebsd/sys/netinet/ip_divert.c (revision d4eeb02986980bf33dd56c41ceb9fc5f180c0d47)
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 inpcbinfo, divcbinfo);
115 #define	V_divcbinfo			VNET(divcbinfo)
116 
117 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
118 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
119 
120 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
121     struct sockaddr_in *sin);
122 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
123 
124 /*
125  * Initialize divert connection block queue.
126  */
127 INPCBSTORAGE_DEFINE(divcbstor, "divinp", "divcb", "div", "divhash");
128 
129 static void
130 div_init(void *arg __unused)
131 {
132 
133 	/*
134 	 * XXX We don't use the hash list for divert IP, but it's easier to
135 	 * allocate one-entry hash lists than it is to check all over the
136 	 * place for hashbase == NULL.
137 	 */
138 	in_pcbinfo_init(&V_divcbinfo, &divcbstor, 1, 1);
139 }
140 VNET_SYSINIT(div_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, div_init, NULL);
141 
142 static void
143 div_destroy(void *unused __unused)
144 {
145 
146 	in_pcbinfo_destroy(&V_divcbinfo);
147 }
148 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, div_destroy, NULL);
149 
150 static bool
151 div_port_match(const struct inpcb *inp, void *v)
152 {
153 	uint16_t nport = *(uint16_t *)v;
154 
155 	return (inp->inp_lport == nport);
156 }
157 
158 /*
159  * Divert a packet by passing it up to the divert socket at port 'port'.
160  */
161 static void
162 divert_packet(struct mbuf *m, bool incoming)
163 {
164 #if defined(SCTP) || defined(SCTP_SUPPORT)
165 	struct ip *ip;
166 #endif
167 	struct inpcb *inp;
168 	struct socket *sa;
169 	u_int16_t nport;
170 	struct sockaddr_in divsrc;
171 	struct inpcb_iterator inpi = INP_ITERATOR(&V_divcbinfo,
172 	    INPLOOKUP_RLOCKPCB, div_port_match, &nport);
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 
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 		ip = mtod(m, struct ip *);
195 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
196 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
197 	}
198 #endif
199 #ifdef INET6
200 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
201 		in6_delayed_cksum(m, m->m_pkthdr.len -
202 		    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
203 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
204 	}
205 #if defined(SCTP) || defined(SCTP_SUPPORT)
206 	if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
207 		sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
208 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
209 	}
210 #endif
211 #endif /* INET6 */
212 	bzero(&divsrc, sizeof(divsrc));
213 	divsrc.sin_len = sizeof(divsrc);
214 	divsrc.sin_family = AF_INET;
215 	/* record matching rule, in host format */
216 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
217 	/*
218 	 * Record receive interface address, if any.
219 	 * But only for incoming packets.
220 	 */
221 	if (incoming) {
222 		struct ifaddr *ifa;
223 		struct ifnet *ifp;
224 
225 		/* Sanity check */
226 		M_ASSERTPKTHDR(m);
227 
228 		/* Find IP address for receive interface */
229 		ifp = m->m_pkthdr.rcvif;
230 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
231 			if (ifa->ifa_addr->sa_family != AF_INET)
232 				continue;
233 			divsrc.sin_addr =
234 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
235 			break;
236 		}
237 	}
238 	/*
239 	 * Record the incoming interface name whenever we have one.
240 	 */
241 	if (m->m_pkthdr.rcvif) {
242 		/*
243 		 * Hide the actual interface name in there in the
244 		 * sin_zero array. XXX This needs to be moved to a
245 		 * different sockaddr type for divert, e.g.
246 		 * sockaddr_div with multiple fields like
247 		 * sockaddr_dl. Presently we have only 7 bytes
248 		 * but that will do for now as most interfaces
249 		 * are 4 or less + 2 or less bytes for unit.
250 		 * There is probably a faster way of doing this,
251 		 * possibly taking it from the sockaddr_dl on the iface.
252 		 * This solves the problem of a P2P link and a LAN interface
253 		 * having the same address, which can result in the wrong
254 		 * interface being assigned to the packet when fed back
255 		 * into the divert socket. Theoretically if the daemon saves
256 		 * and re-uses the sockaddr_in as suggested in the man pages,
257 		 * this iface name will come along for the ride.
258 		 * (see div_output for the other half of this.)
259 		 */
260 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
261 		    sizeof(divsrc.sin_zero));
262 	}
263 
264 	/* Put packet on socket queue, if any */
265 	sa = NULL;
266 	/* nport is inp_next's context. */
267 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
268 	while ((inp = inp_next(&inpi)) != NULL) {
269 		sa = inp->inp_socket;
270 		SOCKBUF_LOCK(&sa->so_rcv);
271 		if (sbappendaddr_locked(&sa->so_rcv,
272 		    (struct sockaddr *)&divsrc, m, NULL) == 0) {
273 			soroverflow_locked(sa);
274 			sa = NULL;	/* force mbuf reclaim below */
275 		} else
276 			sorwakeup_locked(sa);
277 		/* XXX why does only one socket match? */
278 		INP_RUNLOCK(inp);
279 		break;
280 	}
281 	if (sa == NULL) {
282 		m_freem(m);
283 		KMOD_IPSTAT_INC(ips_noproto);
284 		KMOD_IPSTAT_DEC(ips_delivered);
285         }
286 }
287 
288 /*
289  * Deliver packet back into the IP processing machinery.
290  *
291  * If no address specified, or address is 0.0.0.0, send to ip_output();
292  * otherwise, send to ip_input() and mark as having been received on
293  * the interface with that address.
294  */
295 static int
296 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
297     struct mbuf *control)
298 {
299 	struct epoch_tracker et;
300 	const struct ip *ip;
301 	struct m_tag *mtag;
302 	struct ipfw_rule_ref *dt;
303 	int error, family;
304 
305 	if (control) {
306 		m_freem(control);		/* XXX */
307 		control = NULL;
308 	}
309 
310 	if (sin != NULL) {
311 		if (sin->sin_family != AF_INET) {
312 			m_freem(m);
313 			return (EAFNOSUPPORT);
314 		}
315 		if (sin->sin_len != sizeof(*sin)) {
316 			m_freem(m);
317 			return (EINVAL);
318 		}
319 	}
320 
321 	/*
322 	 * An mbuf may hasn't come from userland, but we pretend
323 	 * that it has.
324 	 */
325 	m->m_pkthdr.rcvif = NULL;
326 	m->m_nextpkt = NULL;
327 	M_SETFIB(m, so->so_fibnum);
328 
329 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
330 	if (mtag == NULL) {
331 		/* this should be normal */
332 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
333 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
334 		if (mtag == NULL) {
335 			m_freem(m);
336 			return (ENOBUFS);
337 		}
338 		m_tag_prepend(m, mtag);
339 	}
340 	dt = (struct ipfw_rule_ref *)(mtag+1);
341 
342 	/* Loopback avoidance and state recovery */
343 	if (sin) {
344 		int i;
345 
346 		/* set the starting point. We provide a non-zero slot,
347 		 * but a non_matching chain_id to skip that info and use
348 		 * the rulenum/rule_id.
349 		 */
350 		dt->slot = 1; /* dummy, chain_id is invalid */
351 		dt->chain_id = 0;
352 		dt->rulenum = sin->sin_port+1; /* host format ? */
353 		dt->rule_id = 0;
354 		/* XXX: broken for IPv6 */
355 		/*
356 		 * Find receive interface with the given name, stuffed
357 		 * (if it exists) in the sin_zero[] field.
358 		 * The name is user supplied data so don't trust its size
359 		 * or that it is zero terminated.
360 		 */
361 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
362 			;
363 		if ( i > 0 && i < sizeof(sin->sin_zero))
364 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
365 	}
366 
367 	ip = mtod(m, struct ip *);
368 	switch (ip->ip_v) {
369 	case IPVERSION:
370 		family = AF_INET;
371 		break;
372 #ifdef INET6
373 	case IPV6_VERSION >> 4:
374 		family = AF_INET6;
375 		break;
376 #endif
377 	default:
378 		m_freem(m);
379 		return (EAFNOSUPPORT);
380 	}
381 
382 	/* Reinject packet into the system as incoming or outgoing */
383 	NET_EPOCH_ENTER(et);
384 	if (!sin || sin->sin_addr.s_addr == 0) {
385 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
386 		error = div_output_outbound(family, so, m);
387 	} else {
388 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
389 		error = div_output_inbound(family, so, m, sin);
390 	}
391 	NET_EPOCH_EXIT(et);
392 
393 	return (error);
394 }
395 
396 /*
397  * Sends mbuf @m to the wire via ip[6]_output().
398  *
399  * Returns 0 on success or an errno value on failure.  @m is always consumed.
400  */
401 static int
402 div_output_outbound(int family, struct socket *so, struct mbuf *m)
403 {
404 	struct ip *const ip = mtod(m, struct ip *);
405 	struct mbuf *options;
406 	struct inpcb *inp;
407 	int error;
408 
409 	inp = sotoinpcb(so);
410 	INP_RLOCK(inp);
411 	switch (family) {
412 	case AF_INET:
413 		/*
414 		 * Don't allow both user specified and setsockopt
415 		 * options, and don't allow packet length sizes that
416 		 * will crash.
417 		 */
418 		if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
419 		    inp->inp_options != NULL) ||
420 		    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
421 			INP_RUNLOCK(inp);
422 			m_freem(m);
423 			return (EINVAL);
424 		}
425 		break;
426 #ifdef INET6
427 	case AF_INET6:
428 	    {
429 		struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
430 
431 		/* Don't allow packet length sizes that will crash */
432 		if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
433 			INP_RUNLOCK(inp);
434 			m_freem(m);
435 			return (EINVAL);
436 		}
437 		break;
438 	    }
439 #endif
440 	}
441 
442 	/* Send packet to output processing */
443 	KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
444 
445 #ifdef MAC
446 	mac_inpcb_create_mbuf(inp, m);
447 #endif
448 	/*
449 	 * Get ready to inject the packet into ip_output().
450 	 * Just in case socket options were specified on the
451 	 * divert socket, we duplicate them.  This is done
452 	 * to avoid having to hold the PCB locks over the call
453 	 * to ip_output(), as doing this results in a number of
454 	 * lock ordering complexities.
455 	 *
456 	 * Note that we set the multicast options argument for
457 	 * ip_output() to NULL since it should be invariant that
458 	 * they are not present.
459 	 */
460 	KASSERT(inp->inp_moptions == NULL,
461 	    ("multicast options set on a divert socket"));
462 	/*
463 	 * XXXCSJP: It is unclear to me whether or not it makes
464 	 * sense for divert sockets to have options.  However,
465 	 * for now we will duplicate them with the INP locks
466 	 * held so we can use them in ip_output() without
467 	 * requring a reference to the pcb.
468 	 */
469 	options = NULL;
470 	if (inp->inp_options != NULL) {
471 		options = m_dup(inp->inp_options, M_NOWAIT);
472 		if (options == NULL) {
473 			INP_RUNLOCK(inp);
474 			m_freem(m);
475 			return (ENOBUFS);
476 		}
477 	}
478 	INP_RUNLOCK(inp);
479 
480 	error = 0;
481 	switch (family) {
482 	case AF_INET:
483 		error = ip_output(m, options, NULL,
484 		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
485 		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
486 		break;
487 #ifdef INET6
488 	case AF_INET6:
489 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
490 		break;
491 #endif
492 	}
493 	if (options != NULL)
494 		m_freem(options);
495 
496 	return (error);
497 }
498 
499 /*
500  * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
501  *
502  * Returns 0 on success or an errno value on failure.  @m is always consumed.
503  */
504 static int
505 div_output_inbound(int family, struct socket *so, struct mbuf *m,
506     struct sockaddr_in *sin)
507 {
508 	const struct ip *ip;
509 	struct ifaddr *ifa;
510 
511 	if (m->m_pkthdr.rcvif == NULL) {
512 		/*
513 		 * No luck with the name, check by IP address.
514 		 * Clear the port and the ifname to make sure
515 		 * there are no distractions for ifa_ifwithaddr.
516 		 */
517 
518 		/* XXX: broken for IPv6 */
519 		bzero(sin->sin_zero, sizeof(sin->sin_zero));
520 		sin->sin_port = 0;
521 		ifa = ifa_ifwithaddr((struct sockaddr *) sin);
522 		if (ifa == NULL) {
523 			m_freem(m);
524 			return (EADDRNOTAVAIL);
525 		}
526 		m->m_pkthdr.rcvif = ifa->ifa_ifp;
527 	}
528 #ifdef MAC
529 	mac_socket_create_mbuf(so, m);
530 #endif
531 	/* Send packet to input processing via netisr */
532 	switch (family) {
533 	case AF_INET:
534 		ip = mtod(m, struct ip *);
535 		/*
536 		 * Restore M_BCAST flag when destination address is
537 		 * broadcast. It is expected by ip_tryforward().
538 		 */
539 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
540 			m->m_flags |= M_MCAST;
541 		else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
542 			m->m_flags |= M_BCAST;
543 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
544 		break;
545 #ifdef INET6
546 	case AF_INET6:
547 		netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
548 		break;
549 #endif
550 	default:
551 		m_freem(m);
552 		return (EINVAL);
553 	}
554 
555 	return (0);
556 }
557 
558 static int
559 div_attach(struct socket *so, int proto, struct thread *td)
560 {
561 	struct inpcb *inp;
562 	int error;
563 
564 	inp  = sotoinpcb(so);
565 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
566 	if (td != NULL) {
567 		error = priv_check(td, PRIV_NETINET_DIVERT);
568 		if (error)
569 			return (error);
570 	}
571 	error = soreserve(so, div_sendspace, div_recvspace);
572 	if (error)
573 		return error;
574 	error = in_pcballoc(so, &V_divcbinfo);
575 	if (error)
576 		return error;
577 	inp = (struct inpcb *)so->so_pcb;
578 	inp->inp_ip_p = proto;
579 	inp->inp_flags |= INP_HDRINCL;
580 	INP_WUNLOCK(inp);
581 	return 0;
582 }
583 
584 static void
585 div_detach(struct socket *so)
586 {
587 	struct inpcb *inp;
588 
589 	inp = sotoinpcb(so);
590 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
591 	INP_WLOCK(inp);
592 	in_pcbdetach(inp);
593 	in_pcbfree(inp);
594 }
595 
596 static int
597 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
598 {
599 	struct inpcb *inp;
600 	int error;
601 
602 	inp = sotoinpcb(so);
603 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
604 	/* in_pcbbind assumes that nam is a sockaddr_in
605 	 * and in_pcbbind requires a valid address. Since divert
606 	 * sockets don't we need to make sure the address is
607 	 * filled in properly.
608 	 * XXX -- divert should not be abusing in_pcbind
609 	 * and should probably have its own family.
610 	 */
611 	if (nam->sa_family != AF_INET)
612 		return EAFNOSUPPORT;
613 	if (nam->sa_len != sizeof(struct sockaddr_in))
614 		return EINVAL;
615 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
616 	INP_WLOCK(inp);
617 	INP_HASH_WLOCK(&V_divcbinfo);
618 	error = in_pcbbind(inp, nam, td->td_ucred);
619 	INP_HASH_WUNLOCK(&V_divcbinfo);
620 	INP_WUNLOCK(inp);
621 	return error;
622 }
623 
624 static int
625 div_shutdown(struct socket *so)
626 {
627 	struct inpcb *inp;
628 
629 	inp = sotoinpcb(so);
630 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
631 	INP_WLOCK(inp);
632 	socantsendmore(so);
633 	INP_WUNLOCK(inp);
634 	return 0;
635 }
636 
637 static int
638 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
639     struct mbuf *control, struct thread *td)
640 {
641 
642 	/* Packet must have a header (but that's about it) */
643 	if (m->m_len < sizeof (struct ip) &&
644 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
645 		KMOD_IPSTAT_INC(ips_toosmall);
646 		if (control != NULL)
647 			m_freem(control);
648 		m_freem(m);
649 		return EINVAL;
650 	}
651 
652 	/* Send packet */
653 	return div_output(so, m, (struct sockaddr_in *)nam, control);
654 }
655 
656 static int
657 div_pcblist(SYSCTL_HANDLER_ARGS)
658 {
659 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_divcbinfo,
660 	    INPLOOKUP_RLOCKPCB);
661 	struct xinpgen xig;
662 	struct inpcb *inp;
663 	int error;
664 
665 	if (req->newptr != 0)
666 		return EPERM;
667 
668 	if (req->oldptr == 0) {
669 		int n;
670 
671 		n = V_divcbinfo.ipi_count;
672 		n += imax(n / 8, 10);
673 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
674 		return 0;
675 	}
676 
677 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
678 		return (error);
679 
680 	bzero(&xig, sizeof(xig));
681 	xig.xig_len = sizeof xig;
682 	xig.xig_count = V_divcbinfo.ipi_count;
683 	xig.xig_gen = V_divcbinfo.ipi_gencnt;
684 	xig.xig_sogen = so_gencnt;
685 	error = SYSCTL_OUT(req, &xig, sizeof xig);
686 	if (error)
687 		return error;
688 
689 	while ((inp = inp_next(&inpi)) != NULL) {
690 		if (inp->inp_gencnt <= xig.xig_gen) {
691 			struct xinpcb xi;
692 
693 			in_pcbtoxinpcb(inp, &xi);
694 			error = SYSCTL_OUT(req, &xi, sizeof xi);
695 			if (error) {
696 				INP_RUNLOCK(inp);
697 				break;
698 			}
699 		}
700 	}
701 
702 	if (!error) {
703 		/*
704 		 * Give the user an updated idea of our state.
705 		 * If the generation differs from what we told
706 		 * her before, she knows that something happened
707 		 * while we were processing this request, and it
708 		 * might be necessary to retry.
709 		 */
710 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
711 		xig.xig_sogen = so_gencnt;
712 		xig.xig_count = V_divcbinfo.ipi_count;
713 		error = SYSCTL_OUT(req, &xig, sizeof xig);
714 	}
715 
716 	return (error);
717 }
718 
719 #ifdef SYSCTL_NODE
720 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
721     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
722     "IPDIVERT");
723 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
724    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
725     NULL, 0, div_pcblist, "S,xinpcb",
726     "List of active divert sockets");
727 #endif
728 
729 static struct protosw div_protosw = {
730 	.pr_type =		SOCK_RAW,
731 	.pr_protocol =		IPPROTO_DIVERT,
732 	.pr_flags =		PR_ATOMIC|PR_ADDR,
733 	.pr_attach =		div_attach,
734 	.pr_bind =		div_bind,
735 	.pr_control =		in_control,
736 	.pr_detach =		div_detach,
737 	.pr_peeraddr =		in_getpeeraddr,
738 	.pr_send =		div_send,
739 	.pr_shutdown =		div_shutdown,
740 	.pr_sockaddr =		in_getsockaddr,
741 	.pr_sosetlabel =	in_pcbsosetlabel
742 };
743 
744 static int
745 div_modevent(module_t mod, int type, void *unused)
746 {
747 	int err = 0;
748 
749 	switch (type) {
750 	case MOD_LOAD:
751 		/*
752 		 * Protocol will be initialized by pf_proto_register().
753 		 */
754 		err = protosw_register(&inetdomain, &div_protosw);
755 		if (err != 0)
756 			return (err);
757 		ip_divert_ptr = divert_packet;
758 		break;
759 	case MOD_QUIESCE:
760 		/*
761 		 * IPDIVERT may normally not be unloaded because of the
762 		 * potential race conditions.  Tell kldunload we can't be
763 		 * unloaded unless the unload is forced.
764 		 */
765 		err = EPERM;
766 		break;
767 	case MOD_UNLOAD:
768 		/*
769 		 * Forced unload.
770 		 *
771 		 * Module ipdivert can only be unloaded if no sockets are
772 		 * connected.  Maybe this can be changed later to forcefully
773 		 * disconnect any open sockets.
774 		 *
775 		 * XXXRW: Note that there is a slight race here, as a new
776 		 * socket open request could be spinning on the lock and then
777 		 * we destroy the lock.
778 		 */
779 		INP_INFO_WLOCK(&V_divcbinfo);
780 		if (V_divcbinfo.ipi_count != 0) {
781 			err = EBUSY;
782 			INP_INFO_WUNLOCK(&V_divcbinfo);
783 			break;
784 		}
785 		ip_divert_ptr = NULL;
786 		err = protosw_unregister(&div_protosw);
787 		INP_INFO_WUNLOCK(&V_divcbinfo);
788 #ifndef VIMAGE
789 		div_destroy(NULL);
790 #endif
791 		break;
792 	default:
793 		err = EOPNOTSUPP;
794 		break;
795 	}
796 	return err;
797 }
798 
799 static moduledata_t ipdivertmod = {
800         "ipdivert",
801         div_modevent,
802         0
803 };
804 
805 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
806 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
807 MODULE_VERSION(ipdivert, 1);
808