xref: /freebsd/sys/netinet/ip_divert.c (revision f024dd3bf0c99ff38239ef8e92c12a9fc6bce14c)
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 #ifdef SCTP
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 /*
126  * Initialize divert connection block queue.
127  */
128 static void
129 div_zone_change(void *tag)
130 {
131 
132 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
133 }
134 
135 static int
136 div_inpcb_init(void *mem, int size, int flags)
137 {
138 	struct inpcb *inp = mem;
139 
140 	INP_LOCK_INIT(inp, "inp", "divinp");
141 	return (0);
142 }
143 
144 static void
145 div_init(void)
146 {
147 
148 	/*
149 	 * XXX We don't use the hash list for divert IP, but it's easier to
150 	 * allocate one-entry hash lists than it is to check all over the
151 	 * place for hashbase == NULL.
152 	 */
153 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
154 	    div_inpcb_init, IPI_HASHFIELDS_NONE);
155 }
156 
157 static void
158 div_destroy(void *unused __unused)
159 {
160 
161 	in_pcbinfo_destroy(&V_divcbinfo);
162 }
163 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
164     div_destroy, NULL);
165 
166 /*
167  * IPPROTO_DIVERT is not in the real IP protocol number space; this
168  * function should never be called.  Just in case, drop any packets.
169  */
170 static int
171 div_input(struct mbuf **mp, int *offp, int proto)
172 {
173 	struct mbuf *m = *mp;
174 
175 	KMOD_IPSTAT_INC(ips_noproto);
176 	m_freem(m);
177 	return (IPPROTO_DONE);
178 }
179 
180 /*
181  * Divert a packet by passing it up to the divert socket at port 'port'.
182  *
183  * Setup generic address and protocol structures for div_input routine,
184  * then pass them along with mbuf chain.
185  */
186 static void
187 divert_packet(struct mbuf *m, int incoming)
188 {
189 	struct ip *ip;
190 	struct inpcb *inp;
191 	struct socket *sa;
192 	u_int16_t nport;
193 	struct sockaddr_in divsrc;
194 	struct m_tag *mtag;
195 	struct epoch_tracker et;
196 
197 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
198 	if (mtag == NULL) {
199 		m_freem(m);
200 		return;
201 	}
202 	/* Assure header */
203 	if (m->m_len < sizeof(struct ip) &&
204 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
205 		return;
206 	ip = mtod(m, struct ip *);
207 
208 	/* Delayed checksums are currently not compatible with divert. */
209 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
210 		in_delayed_cksum(m);
211 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
212 	}
213 #ifdef SCTP
214 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
215 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
216 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
217 	}
218 #endif
219 	bzero(&divsrc, sizeof(divsrc));
220 	divsrc.sin_len = sizeof(divsrc);
221 	divsrc.sin_family = AF_INET;
222 	/* record matching rule, in host format */
223 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
224 	/*
225 	 * Record receive interface address, if any.
226 	 * But only for incoming packets.
227 	 */
228 	if (incoming) {
229 		struct ifaddr *ifa;
230 		struct ifnet *ifp;
231 
232 		/* Sanity check */
233 		M_ASSERTPKTHDR(m);
234 
235 		/* Find IP address for receive interface */
236 		ifp = m->m_pkthdr.rcvif;
237 		if_addr_rlock(ifp);
238 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
239 			if (ifa->ifa_addr->sa_family != AF_INET)
240 				continue;
241 			divsrc.sin_addr =
242 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
243 			break;
244 		}
245 		if_addr_runlock(ifp);
246 	}
247 	/*
248 	 * Record the incoming interface name whenever we have one.
249 	 */
250 	if (m->m_pkthdr.rcvif) {
251 		/*
252 		 * Hide the actual interface name in there in the
253 		 * sin_zero array. XXX This needs to be moved to a
254 		 * different sockaddr type for divert, e.g.
255 		 * sockaddr_div with multiple fields like
256 		 * sockaddr_dl. Presently we have only 7 bytes
257 		 * but that will do for now as most interfaces
258 		 * are 4 or less + 2 or less bytes for unit.
259 		 * There is probably a faster way of doing this,
260 		 * possibly taking it from the sockaddr_dl on the iface.
261 		 * This solves the problem of a P2P link and a LAN interface
262 		 * having the same address, which can result in the wrong
263 		 * interface being assigned to the packet when fed back
264 		 * into the divert socket. Theoretically if the daemon saves
265 		 * and re-uses the sockaddr_in as suggested in the man pages,
266 		 * this iface name will come along for the ride.
267 		 * (see div_output for the other half of this.)
268 		 */
269 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
270 		    sizeof(divsrc.sin_zero));
271 	}
272 
273 	/* Put packet on socket queue, if any */
274 	sa = NULL;
275 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
276 	INP_INFO_RLOCK_ET(&V_divcbinfo, et);
277 	CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
278 		/* XXX why does only one socket match? */
279 		if (inp->inp_lport == nport) {
280 			INP_RLOCK(inp);
281 			sa = inp->inp_socket;
282 			SOCKBUF_LOCK(&sa->so_rcv);
283 			if (sbappendaddr_locked(&sa->so_rcv,
284 			    (struct sockaddr *)&divsrc, m,
285 			    (struct mbuf *)0) == 0) {
286 				SOCKBUF_UNLOCK(&sa->so_rcv);
287 				sa = NULL;	/* force mbuf reclaim below */
288 			} else
289 				sorwakeup_locked(sa);
290 			INP_RUNLOCK(inp);
291 			break;
292 		}
293 	}
294 	INP_INFO_RUNLOCK_ET(&V_divcbinfo, et);
295 	if (sa == NULL) {
296 		m_freem(m);
297 		KMOD_IPSTAT_INC(ips_noproto);
298 		KMOD_IPSTAT_DEC(ips_delivered);
299         }
300 }
301 
302 /*
303  * Deliver packet back into the IP processing machinery.
304  *
305  * If no address specified, or address is 0.0.0.0, send to ip_output();
306  * otherwise, send to ip_input() and mark as having been received on
307  * the interface with that address.
308  */
309 static int
310 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
311     struct mbuf *control)
312 {
313 	struct ip *const ip = mtod(m, struct ip *);
314 	struct m_tag *mtag;
315 	struct ipfw_rule_ref *dt;
316 	int error = 0;
317 
318 	/*
319 	 * An mbuf may hasn't come from userland, but we pretend
320 	 * that it has.
321 	 */
322 	m->m_pkthdr.rcvif = NULL;
323 	m->m_nextpkt = NULL;
324 	M_SETFIB(m, so->so_fibnum);
325 
326 	if (control)
327 		m_freem(control);		/* XXX */
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 			error = ENOBUFS;
336 			goto cantsend;
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 		/*
355 		 * Find receive interface with the given name, stuffed
356 		 * (if it exists) in the sin_zero[] field.
357 		 * The name is user supplied data so don't trust its size
358 		 * or that it is zero terminated.
359 		 */
360 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
361 			;
362 		if ( i > 0 && i < sizeof(sin->sin_zero))
363 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
364 	}
365 
366 	/* Reinject packet into the system as incoming or outgoing */
367 	if (!sin || sin->sin_addr.s_addr == 0) {
368 		struct mbuf *options = NULL;
369 		struct inpcb *inp;
370 
371 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
372 		inp = sotoinpcb(so);
373 		INP_RLOCK(inp);
374 		switch (ip->ip_v) {
375 		case IPVERSION:
376 			/*
377 			 * Don't allow both user specified and setsockopt
378 			 * options, and don't allow packet length sizes that
379 			 * will crash.
380 			 */
381 			if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
382 			    inp->inp_options != NULL) ||
383 			    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
384 				error = EINVAL;
385 				INP_RUNLOCK(inp);
386 				goto cantsend;
387 			}
388 			break;
389 #ifdef INET6
390 		case IPV6_VERSION >> 4:
391 		    {
392 			struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
393 
394 			/* Don't allow packet length sizes that will crash */
395 			if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
396 				error = EINVAL;
397 				INP_RUNLOCK(inp);
398 				goto cantsend;
399 			}
400 			break;
401 		    }
402 #endif
403 		default:
404 			error = EINVAL;
405 			INP_RUNLOCK(inp);
406 			goto cantsend;
407 		}
408 
409 		/* Send packet to output processing */
410 		KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
411 
412 #ifdef MAC
413 		mac_inpcb_create_mbuf(inp, m);
414 #endif
415 		/*
416 		 * Get ready to inject the packet into ip_output().
417 		 * Just in case socket options were specified on the
418 		 * divert socket, we duplicate them.  This is done
419 		 * to avoid having to hold the PCB locks over the call
420 		 * to ip_output(), as doing this results in a number of
421 		 * lock ordering complexities.
422 		 *
423 		 * Note that we set the multicast options argument for
424 		 * ip_output() to NULL since it should be invariant that
425 		 * they are not present.
426 		 */
427 		KASSERT(inp->inp_moptions == NULL,
428 		    ("multicast options set on a divert socket"));
429 		/*
430 		 * XXXCSJP: It is unclear to me whether or not it makes
431 		 * sense for divert sockets to have options.  However,
432 		 * for now we will duplicate them with the INP locks
433 		 * held so we can use them in ip_output() without
434 		 * requring a reference to the pcb.
435 		 */
436 		if (inp->inp_options != NULL) {
437 			options = m_dup(inp->inp_options, M_NOWAIT);
438 			if (options == NULL) {
439 				INP_RUNLOCK(inp);
440 				error = ENOBUFS;
441 				goto cantsend;
442 			}
443 		}
444 		INP_RUNLOCK(inp);
445 
446 		switch (ip->ip_v) {
447 		case IPVERSION:
448 			error = ip_output(m, options, NULL,
449 			    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
450 			    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
451 			break;
452 #ifdef INET6
453 		case IPV6_VERSION >> 4:
454 			error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
455 			break;
456 #endif
457 		}
458 		if (options != NULL)
459 			m_freem(options);
460 	} else {
461 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
462 		if (m->m_pkthdr.rcvif == NULL) {
463 			/*
464 			 * No luck with the name, check by IP address.
465 			 * Clear the port and the ifname to make sure
466 			 * there are no distractions for ifa_ifwithaddr.
467 			 */
468 			struct	ifaddr *ifa;
469 
470 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
471 			sin->sin_port = 0;
472 			NET_EPOCH_ENTER();
473 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
474 			if (ifa == NULL) {
475 				error = EADDRNOTAVAIL;
476 				NET_EPOCH_EXIT();
477 				goto cantsend;
478 			}
479 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
480 			NET_EPOCH_EXIT();
481 		}
482 #ifdef MAC
483 		mac_socket_create_mbuf(so, m);
484 #endif
485 		/* Send packet to input processing via netisr */
486 		switch (ip->ip_v) {
487 		case IPVERSION:
488 			/*
489 			 * Restore M_BCAST flag when destination address is
490 			 * broadcast. It is expected by ip_tryforward().
491 			 */
492 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
493 				m->m_flags |= M_MCAST;
494 			else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
495 				m->m_flags |= M_BCAST;
496 			netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
497 			break;
498 #ifdef INET6
499 		case IPV6_VERSION >> 4:
500 			netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
501 			break;
502 #endif
503 		default:
504 			error = EINVAL;
505 			goto cantsend;
506 		}
507 	}
508 
509 	return (error);
510 
511 cantsend:
512 	m_freem(m);
513 	return (error);
514 }
515 
516 static int
517 div_attach(struct socket *so, int proto, struct thread *td)
518 {
519 	struct inpcb *inp;
520 	int error;
521 
522 	inp  = sotoinpcb(so);
523 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
524 	if (td != NULL) {
525 		error = priv_check(td, PRIV_NETINET_DIVERT);
526 		if (error)
527 			return (error);
528 	}
529 	error = soreserve(so, div_sendspace, div_recvspace);
530 	if (error)
531 		return error;
532 	INP_INFO_WLOCK(&V_divcbinfo);
533 	error = in_pcballoc(so, &V_divcbinfo);
534 	if (error) {
535 		INP_INFO_WUNLOCK(&V_divcbinfo);
536 		return error;
537 	}
538 	inp = (struct inpcb *)so->so_pcb;
539 	INP_INFO_WUNLOCK(&V_divcbinfo);
540 	inp->inp_ip_p = proto;
541 	inp->inp_vflag |= INP_IPV4;
542 	inp->inp_flags |= INP_HDRINCL;
543 	INP_WUNLOCK(inp);
544 	return 0;
545 }
546 
547 static void
548 div_detach(struct socket *so)
549 {
550 	struct inpcb *inp;
551 
552 	inp = sotoinpcb(so);
553 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
554 	INP_INFO_WLOCK(&V_divcbinfo);
555 	INP_WLOCK(inp);
556 	in_pcbdetach(inp);
557 	in_pcbfree(inp);
558 	INP_INFO_WUNLOCK(&V_divcbinfo);
559 }
560 
561 static int
562 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
563 {
564 	struct inpcb *inp;
565 	int error;
566 
567 	inp = sotoinpcb(so);
568 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
569 	/* in_pcbbind assumes that nam is a sockaddr_in
570 	 * and in_pcbbind requires a valid address. Since divert
571 	 * sockets don't we need to make sure the address is
572 	 * filled in properly.
573 	 * XXX -- divert should not be abusing in_pcbind
574 	 * and should probably have its own family.
575 	 */
576 	if (nam->sa_family != AF_INET)
577 		return EAFNOSUPPORT;
578 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
579 	INP_INFO_WLOCK(&V_divcbinfo);
580 	INP_WLOCK(inp);
581 	INP_HASH_WLOCK(&V_divcbinfo);
582 	error = in_pcbbind(inp, nam, td->td_ucred);
583 	INP_HASH_WUNLOCK(&V_divcbinfo);
584 	INP_WUNLOCK(inp);
585 	INP_INFO_WUNLOCK(&V_divcbinfo);
586 	return error;
587 }
588 
589 static int
590 div_shutdown(struct socket *so)
591 {
592 	struct inpcb *inp;
593 
594 	inp = sotoinpcb(so);
595 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
596 	INP_WLOCK(inp);
597 	socantsendmore(so);
598 	INP_WUNLOCK(inp);
599 	return 0;
600 }
601 
602 static int
603 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
604     struct mbuf *control, struct thread *td)
605 {
606 
607 	/* Packet must have a header (but that's about it) */
608 	if (m->m_len < sizeof (struct ip) &&
609 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
610 		KMOD_IPSTAT_INC(ips_toosmall);
611 		m_freem(m);
612 		return EINVAL;
613 	}
614 
615 	/* Send packet */
616 	return div_output(so, m, (struct sockaddr_in *)nam, control);
617 }
618 
619 static void
620 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
621 {
622         struct in_addr faddr;
623 
624 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
625 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
626         	return;
627 	if (PRC_IS_REDIRECT(cmd))
628 		return;
629 }
630 
631 static int
632 div_pcblist(SYSCTL_HANDLER_ARGS)
633 {
634 	int error, i, n;
635 	struct inpcb *inp, **inp_list;
636 	inp_gen_t gencnt;
637 	struct xinpgen xig;
638 	struct epoch_tracker et;
639 
640 	/*
641 	 * The process of preparing the TCB list is too time-consuming and
642 	 * resource-intensive to repeat twice on every request.
643 	 */
644 	if (req->oldptr == 0) {
645 		n = V_divcbinfo.ipi_count;
646 		n += imax(n / 8, 10);
647 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
648 		return 0;
649 	}
650 
651 	if (req->newptr != 0)
652 		return EPERM;
653 
654 	/*
655 	 * OK, now we're committed to doing something.
656 	 */
657 	INP_INFO_WLOCK(&V_divcbinfo);
658 	gencnt = V_divcbinfo.ipi_gencnt;
659 	n = V_divcbinfo.ipi_count;
660 	INP_INFO_WUNLOCK(&V_divcbinfo);
661 
662 	error = sysctl_wire_old_buffer(req,
663 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
664 	if (error != 0)
665 		return (error);
666 
667 	xig.xig_len = sizeof xig;
668 	xig.xig_count = n;
669 	xig.xig_gen = gencnt;
670 	xig.xig_sogen = so_gencnt;
671 	error = SYSCTL_OUT(req, &xig, sizeof xig);
672 	if (error)
673 		return error;
674 
675 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
676 	if (inp_list == NULL)
677 		return ENOMEM;
678 
679 	INP_INFO_RLOCK_ET(&V_divcbinfo, et);
680 	for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
681 	     inp = CK_LIST_NEXT(inp, inp_list)) {
682 		INP_WLOCK(inp);
683 		if (inp->inp_gencnt <= gencnt &&
684 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
685 			in_pcbref(inp);
686 			inp_list[i++] = inp;
687 		}
688 		INP_WUNLOCK(inp);
689 	}
690 	INP_INFO_RUNLOCK_ET(&V_divcbinfo, et);
691 	n = i;
692 
693 	error = 0;
694 	for (i = 0; i < n; i++) {
695 		inp = inp_list[i];
696 		INP_RLOCK(inp);
697 		if (inp->inp_gencnt <= gencnt) {
698 			struct xinpcb xi;
699 
700 			in_pcbtoxinpcb(inp, &xi);
701 			INP_RUNLOCK(inp);
702 			error = SYSCTL_OUT(req, &xi, sizeof xi);
703 		} else
704 			INP_RUNLOCK(inp);
705 	}
706 	INP_INFO_WLOCK(&V_divcbinfo);
707 	for (i = 0; i < n; i++) {
708 		inp = inp_list[i];
709 		INP_RLOCK(inp);
710 		if (!in_pcbrele_rlocked(inp))
711 			INP_RUNLOCK(inp);
712 	}
713 	INP_INFO_WUNLOCK(&V_divcbinfo);
714 
715 	if (!error) {
716 		struct epoch_tracker et;
717 		/*
718 		 * Give the user an updated idea of our state.
719 		 * If the generation differs from what we told
720 		 * her before, she knows that something happened
721 		 * while we were processing this request, and it
722 		 * might be necessary to retry.
723 		 */
724 		INP_INFO_RLOCK_ET(&V_divcbinfo, et);
725 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
726 		xig.xig_sogen = so_gencnt;
727 		xig.xig_count = V_divcbinfo.ipi_count;
728 		INP_INFO_RUNLOCK_ET(&V_divcbinfo, et);
729 		error = SYSCTL_OUT(req, &xig, sizeof xig);
730 	}
731 	free(inp_list, M_TEMP);
732 	return error;
733 }
734 
735 #ifdef SYSCTL_NODE
736 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
737     "IPDIVERT");
738 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
739     NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
740 #endif
741 
742 struct pr_usrreqs div_usrreqs = {
743 	.pru_attach =		div_attach,
744 	.pru_bind =		div_bind,
745 	.pru_control =		in_control,
746 	.pru_detach =		div_detach,
747 	.pru_peeraddr =		in_getpeeraddr,
748 	.pru_send =		div_send,
749 	.pru_shutdown =		div_shutdown,
750 	.pru_sockaddr =		in_getsockaddr,
751 	.pru_sosetlabel =	in_pcbsosetlabel
752 };
753 
754 struct protosw div_protosw = {
755 	.pr_type =		SOCK_RAW,
756 	.pr_protocol =		IPPROTO_DIVERT,
757 	.pr_flags =		PR_ATOMIC|PR_ADDR,
758 	.pr_input =		div_input,
759 	.pr_ctlinput =		div_ctlinput,
760 	.pr_ctloutput =		ip_ctloutput,
761 	.pr_init =		div_init,
762 	.pr_usrreqs =		&div_usrreqs
763 };
764 
765 static int
766 div_modevent(module_t mod, int type, void *unused)
767 {
768 	int err = 0;
769 
770 	switch (type) {
771 	case MOD_LOAD:
772 		/*
773 		 * Protocol will be initialized by pf_proto_register().
774 		 * We don't have to register ip_protox because we are not
775 		 * a true IP protocol that goes over the wire.
776 		 */
777 		err = pf_proto_register(PF_INET, &div_protosw);
778 		if (err != 0)
779 			return (err);
780 		ip_divert_ptr = divert_packet;
781 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
782 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
783 		break;
784 	case MOD_QUIESCE:
785 		/*
786 		 * IPDIVERT may normally not be unloaded because of the
787 		 * potential race conditions.  Tell kldunload we can't be
788 		 * unloaded unless the unload is forced.
789 		 */
790 		err = EPERM;
791 		break;
792 	case MOD_UNLOAD:
793 		/*
794 		 * Forced unload.
795 		 *
796 		 * Module ipdivert can only be unloaded if no sockets are
797 		 * connected.  Maybe this can be changed later to forcefully
798 		 * disconnect any open sockets.
799 		 *
800 		 * XXXRW: Note that there is a slight race here, as a new
801 		 * socket open request could be spinning on the lock and then
802 		 * we destroy the lock.
803 		 */
804 		INP_INFO_WLOCK(&V_divcbinfo);
805 		if (V_divcbinfo.ipi_count != 0) {
806 			err = EBUSY;
807 			INP_INFO_WUNLOCK(&V_divcbinfo);
808 			break;
809 		}
810 		ip_divert_ptr = NULL;
811 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
812 		INP_INFO_WUNLOCK(&V_divcbinfo);
813 #ifndef VIMAGE
814 		div_destroy(NULL);
815 #endif
816 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
817 		break;
818 	default:
819 		err = EOPNOTSUPP;
820 		break;
821 	}
822 	return err;
823 }
824 
825 static moduledata_t ipdivertmod = {
826         "ipdivert",
827         div_modevent,
828         0
829 };
830 
831 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
832 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
833 MODULE_VERSION(ipdivert, 1);
834