xref: /freebsd/sys/netinet/ip_divert.c (revision 7aa383846770374466b1dcb2cefd71bde9acf463)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #if !defined(KLD_MODULE)
34 #include "opt_inet.h"
35 #include "opt_sctp.h"
36 #ifndef INET
37 #error "IPDIVERT requires INET."
38 #endif
39 #endif
40 
41 #include <sys/param.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/module.h>
47 #include <sys/kernel.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
54 #include <net/vnet.h>
55 
56 #include <net/if.h>
57 #include <net/netisr.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_var.h>
65 #ifdef SCTP
66 #include <netinet/sctp_crc32.h>
67 #endif
68 
69 #include <security/mac/mac_framework.h>
70 
71 /*
72  * Divert sockets
73  */
74 
75 /*
76  * Allocate enough space to hold a full IP packet
77  */
78 #define	DIVSNDQ		(65536 + 100)
79 #define	DIVRCVQ		(65536 + 100)
80 
81 /*
82  * Divert sockets work in conjunction with ipfw or other packet filters,
83  * see the divert(4) manpage for features.
84  * Packets are selected by the packet filter and tagged with an
85  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
86  * the packet filter) and information on the matching filter rule for
87  * subsequent reinjection. The divert_port is used to put the packet
88  * on the corresponding divert socket, while the rule number is passed
89  * up (at least partially) as the sin_port in the struct sockaddr.
90  *
91  * Packets written to the divert socket carry in sin_addr a
92  * destination address, and in sin_port the number of the filter rule
93  * after which to continue processing.
94  * If the destination address is INADDR_ANY, the packet is treated as
95  * as outgoing and sent to ip_output(); otherwise it is treated as
96  * incoming and sent to ip_input().
97  * Further, sin_zero carries some information on the interface,
98  * which can be used in the reinject -- see comments in the code.
99  *
100  * On reinjection, processing in ip_input() and ip_output()
101  * will be exactly the same as for the original packet, except that
102  * packet filter processing will start at the rule number after the one
103  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
104  * will apply the entire ruleset to the packet).
105  */
106 
107 /* Internal variables. */
108 static VNET_DEFINE(struct inpcbhead, divcb);
109 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
110 
111 #define	V_divcb				VNET(divcb)
112 #define	V_divcbinfo			VNET(divcbinfo)
113 
114 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
115 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
116 
117 static eventhandler_tag ip_divert_event_tag;
118 
119 /*
120  * Initialize divert connection block queue.
121  */
122 static void
123 div_zone_change(void *tag)
124 {
125 
126 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
127 }
128 
129 static int
130 div_inpcb_init(void *mem, int size, int flags)
131 {
132 	struct inpcb *inp = mem;
133 
134 	INP_LOCK_INIT(inp, "inp", "divinp");
135 	return (0);
136 }
137 
138 static void
139 div_inpcb_fini(void *mem, int size)
140 {
141 	struct inpcb *inp = mem;
142 
143 	INP_LOCK_DESTROY(inp);
144 }
145 
146 static void
147 div_init(void)
148 {
149 
150 	/*
151 	 * XXX We don't use the hash list for divert IP, but it's easier to
152 	 * allocate one-entry hash lists than it is to check all over the
153 	 * place for hashbase == NULL.
154 	 */
155 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
156 	    div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE);
157 }
158 
159 static void
160 div_destroy(void)
161 {
162 
163 	in_pcbinfo_destroy(&V_divcbinfo);
164 }
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 void
171 div_input(struct mbuf *m, int off)
172 {
173 
174 	KMOD_IPSTAT_INC(ips_noproto);
175 	m_freem(m);
176 }
177 
178 /*
179  * Divert a packet by passing it up to the divert socket at port 'port'.
180  *
181  * Setup generic address and protocol structures for div_input routine,
182  * then pass them along with mbuf chain.
183  */
184 static void
185 divert_packet(struct mbuf *m, int incoming)
186 {
187 	struct ip *ip;
188 	struct inpcb *inp;
189 	struct socket *sa;
190 	u_int16_t nport;
191 	struct sockaddr_in divsrc;
192 	struct m_tag *mtag;
193 
194 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
195 	if (mtag == NULL) {
196 		m_freem(m);
197 		return;
198 	}
199 	/* Assure header */
200 	if (m->m_len < sizeof(struct ip) &&
201 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
202 		return;
203 	ip = mtod(m, struct ip *);
204 
205 	/* Delayed checksums are currently not compatible with divert. */
206 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
207 		ip->ip_len = ntohs(ip->ip_len);
208 		in_delayed_cksum(m);
209 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
210 		ip->ip_len = htons(ip->ip_len);
211 	}
212 #ifdef SCTP
213 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
214 		ip->ip_len = ntohs(ip->ip_len);
215 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
216 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
217 		ip->ip_len = htons(ip->ip_len);
218 	}
219 #endif
220 	bzero(&divsrc, sizeof(divsrc));
221 	divsrc.sin_len = sizeof(divsrc);
222 	divsrc.sin_family = AF_INET;
223 	/* record matching rule, in host format */
224 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
225 	/*
226 	 * Record receive interface address, if any.
227 	 * But only for incoming packets.
228 	 */
229 	if (incoming) {
230 		struct ifaddr *ifa;
231 		struct ifnet *ifp;
232 
233 		/* Sanity check */
234 		M_ASSERTPKTHDR(m);
235 
236 		/* Find IP address for receive interface */
237 		ifp = m->m_pkthdr.rcvif;
238 		if_addr_rlock(ifp);
239 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
240 			if (ifa->ifa_addr->sa_family != AF_INET)
241 				continue;
242 			divsrc.sin_addr =
243 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
244 			break;
245 		}
246 		if_addr_runlock(ifp);
247 	}
248 	/*
249 	 * Record the incoming interface name whenever we have one.
250 	 */
251 	if (m->m_pkthdr.rcvif) {
252 		/*
253 		 * Hide the actual interface name in there in the
254 		 * sin_zero array. XXX This needs to be moved to a
255 		 * different sockaddr type for divert, e.g.
256 		 * sockaddr_div with multiple fields like
257 		 * sockaddr_dl. Presently we have only 7 bytes
258 		 * but that will do for now as most interfaces
259 		 * are 4 or less + 2 or less bytes for unit.
260 		 * There is probably a faster way of doing this,
261 		 * possibly taking it from the sockaddr_dl on the iface.
262 		 * This solves the problem of a P2P link and a LAN interface
263 		 * having the same address, which can result in the wrong
264 		 * interface being assigned to the packet when fed back
265 		 * into the divert socket. Theoretically if the daemon saves
266 		 * and re-uses the sockaddr_in as suggested in the man pages,
267 		 * this iface name will come along for the ride.
268 		 * (see div_output for the other half of this.)
269 		 */
270 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
271 		    sizeof(divsrc.sin_zero));
272 	}
273 
274 	/* Put packet on socket queue, if any */
275 	sa = NULL;
276 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
277 	INP_INFO_RLOCK(&V_divcbinfo);
278 	LIST_FOREACH(inp, &V_divcb, inp_list) {
279 		/* XXX why does only one socket match? */
280 		if (inp->inp_lport == nport) {
281 			INP_RLOCK(inp);
282 			sa = inp->inp_socket;
283 			SOCKBUF_LOCK(&sa->so_rcv);
284 			if (sbappendaddr_locked(&sa->so_rcv,
285 			    (struct sockaddr *)&divsrc, m,
286 			    (struct mbuf *)0) == 0) {
287 				SOCKBUF_UNLOCK(&sa->so_rcv);
288 				sa = NULL;	/* force mbuf reclaim below */
289 			} else
290 				sorwakeup_locked(sa);
291 			INP_RUNLOCK(inp);
292 			break;
293 		}
294 	}
295 	INP_INFO_RUNLOCK(&V_divcbinfo);
296 	if (sa == NULL) {
297 		m_freem(m);
298 		KMOD_IPSTAT_INC(ips_noproto);
299 		KMOD_IPSTAT_DEC(ips_delivered);
300         }
301 }
302 
303 /*
304  * Deliver packet back into the IP processing machinery.
305  *
306  * If no address specified, or address is 0.0.0.0, send to ip_output();
307  * otherwise, send to ip_input() and mark as having been received on
308  * the interface with that address.
309  */
310 static int
311 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
312     struct mbuf *control)
313 {
314 	struct m_tag *mtag;
315 	struct ipfw_rule_ref *dt;
316 	int error = 0;
317 	struct mbuf *options;
318 
319 	/*
320 	 * An mbuf may hasn't come from userland, but we pretend
321 	 * that it has.
322 	 */
323 	m->m_pkthdr.rcvif = NULL;
324 	m->m_nextpkt = NULL;
325 	M_SETFIB(m, so->so_fibnum);
326 
327 	if (control)
328 		m_freem(control);		/* XXX */
329 
330 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
331 	if (mtag == NULL) {
332 		/* this should be normal */
333 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
334 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
335 		if (mtag == NULL) {
336 			error = ENOBUFS;
337 			goto cantsend;
338 		}
339 		m_tag_prepend(m, mtag);
340 	}
341 	dt = (struct ipfw_rule_ref *)(mtag+1);
342 
343 	/* Loopback avoidance and state recovery */
344 	if (sin) {
345 		int i;
346 
347 		/* set the starting point. We provide a non-zero slot,
348 		 * but a non_matching chain_id to skip that info and use
349 		 * the rulenum/rule_id.
350 		 */
351 		dt->slot = 1; /* dummy, chain_id is invalid */
352 		dt->chain_id = 0;
353 		dt->rulenum = sin->sin_port+1; /* host format ? */
354 		dt->rule_id = 0;
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 	/* Reinject packet into the system as incoming or outgoing */
368 	if (!sin || sin->sin_addr.s_addr == 0) {
369 		struct ip *const ip = mtod(m, struct ip *);
370 		struct inpcb *inp;
371 
372 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
373 		inp = sotoinpcb(so);
374 		INP_RLOCK(inp);
375 		/*
376 		 * Don't allow both user specified and setsockopt options,
377 		 * and don't allow packet length sizes that will crash
378 		 */
379 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
380 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
381 			error = EINVAL;
382 			INP_RUNLOCK(inp);
383 			m_freem(m);
384 		} else {
385 			/* Convert fields to host order for ip_output() */
386 			ip->ip_len = ntohs(ip->ip_len);
387 			ip->ip_off = ntohs(ip->ip_off);
388 
389 			/* Send packet to output processing */
390 			KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
391 
392 #ifdef MAC
393 			mac_inpcb_create_mbuf(inp, m);
394 #endif
395 			/*
396 			 * Get ready to inject the packet into ip_output().
397 			 * Just in case socket options were specified on the
398 			 * divert socket, we duplicate them.  This is done
399 			 * to avoid having to hold the PCB locks over the call
400 			 * to ip_output(), as doing this results in a number of
401 			 * lock ordering complexities.
402 			 *
403 			 * Note that we set the multicast options argument for
404 			 * ip_output() to NULL since it should be invariant that
405 			 * they are not present.
406 			 */
407 			KASSERT(inp->inp_moptions == NULL,
408 			    ("multicast options set on a divert socket"));
409 			options = NULL;
410 			/*
411 			 * XXXCSJP: It is unclear to me whether or not it makes
412 			 * sense for divert sockets to have options.  However,
413 			 * for now we will duplicate them with the INP locks
414 			 * held so we can use them in ip_output() without
415 			 * requring a reference to the pcb.
416 			 */
417 			if (inp->inp_options != NULL) {
418 				options = m_dup(inp->inp_options, M_DONTWAIT);
419 				if (options == NULL)
420 					error = ENOBUFS;
421 			}
422 			INP_RUNLOCK(inp);
423 			if (error == ENOBUFS) {
424 				m_freem(m);
425 				return (error);
426 			}
427 			error = ip_output(m, options, NULL,
428 			    ((so->so_options & SO_DONTROUTE) ?
429 			    IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
430 			    IP_RAWOUTPUT, NULL, NULL);
431 			if (options != NULL)
432 				m_freem(options);
433 		}
434 	} else {
435 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
436 		if (m->m_pkthdr.rcvif == NULL) {
437 			/*
438 			 * No luck with the name, check by IP address.
439 			 * Clear the port and the ifname to make sure
440 			 * there are no distractions for ifa_ifwithaddr.
441 			 */
442 			struct	ifaddr *ifa;
443 
444 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
445 			sin->sin_port = 0;
446 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
447 			if (ifa == NULL) {
448 				error = EADDRNOTAVAIL;
449 				goto cantsend;
450 			}
451 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
452 			ifa_free(ifa);
453 		}
454 #ifdef MAC
455 		mac_socket_create_mbuf(so, m);
456 #endif
457 		/* Send packet to input processing via netisr */
458 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
459 	}
460 
461 	return error;
462 
463 cantsend:
464 	m_freem(m);
465 	return error;
466 }
467 
468 static int
469 div_attach(struct socket *so, int proto, struct thread *td)
470 {
471 	struct inpcb *inp;
472 	int error;
473 
474 	inp  = sotoinpcb(so);
475 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
476 	if (td != NULL) {
477 		error = priv_check(td, PRIV_NETINET_DIVERT);
478 		if (error)
479 			return (error);
480 	}
481 	error = soreserve(so, div_sendspace, div_recvspace);
482 	if (error)
483 		return error;
484 	INP_INFO_WLOCK(&V_divcbinfo);
485 	error = in_pcballoc(so, &V_divcbinfo);
486 	if (error) {
487 		INP_INFO_WUNLOCK(&V_divcbinfo);
488 		return error;
489 	}
490 	inp = (struct inpcb *)so->so_pcb;
491 	INP_INFO_WUNLOCK(&V_divcbinfo);
492 	inp->inp_ip_p = proto;
493 	inp->inp_vflag |= INP_IPV4;
494 	inp->inp_flags |= INP_HDRINCL;
495 	INP_WUNLOCK(inp);
496 	return 0;
497 }
498 
499 static void
500 div_detach(struct socket *so)
501 {
502 	struct inpcb *inp;
503 
504 	inp = sotoinpcb(so);
505 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
506 	INP_INFO_WLOCK(&V_divcbinfo);
507 	INP_WLOCK(inp);
508 	in_pcbdetach(inp);
509 	in_pcbfree(inp);
510 	INP_INFO_WUNLOCK(&V_divcbinfo);
511 }
512 
513 static int
514 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
515 {
516 	struct inpcb *inp;
517 	int error;
518 
519 	inp = sotoinpcb(so);
520 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
521 	/* in_pcbbind assumes that nam is a sockaddr_in
522 	 * and in_pcbbind requires a valid address. Since divert
523 	 * sockets don't we need to make sure the address is
524 	 * filled in properly.
525 	 * XXX -- divert should not be abusing in_pcbind
526 	 * and should probably have its own family.
527 	 */
528 	if (nam->sa_family != AF_INET)
529 		return EAFNOSUPPORT;
530 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
531 	INP_INFO_WLOCK(&V_divcbinfo);
532 	INP_WLOCK(inp);
533 	error = in_pcbbind(inp, nam, td->td_ucred);
534 	INP_WUNLOCK(inp);
535 	INP_INFO_WUNLOCK(&V_divcbinfo);
536 	return error;
537 }
538 
539 static int
540 div_shutdown(struct socket *so)
541 {
542 	struct inpcb *inp;
543 
544 	inp = sotoinpcb(so);
545 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
546 	INP_WLOCK(inp);
547 	socantsendmore(so);
548 	INP_WUNLOCK(inp);
549 	return 0;
550 }
551 
552 static int
553 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
554     struct mbuf *control, struct thread *td)
555 {
556 
557 	/* Packet must have a header (but that's about it) */
558 	if (m->m_len < sizeof (struct ip) &&
559 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
560 		KMOD_IPSTAT_INC(ips_toosmall);
561 		m_freem(m);
562 		return EINVAL;
563 	}
564 
565 	/* Send packet */
566 	return div_output(so, m, (struct sockaddr_in *)nam, control);
567 }
568 
569 static void
570 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
571 {
572         struct in_addr faddr;
573 
574 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
575 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
576         	return;
577 	if (PRC_IS_REDIRECT(cmd))
578 		return;
579 }
580 
581 static int
582 div_pcblist(SYSCTL_HANDLER_ARGS)
583 {
584 	int error, i, n;
585 	struct inpcb *inp, **inp_list;
586 	inp_gen_t gencnt;
587 	struct xinpgen xig;
588 
589 	/*
590 	 * The process of preparing the TCB list is too time-consuming and
591 	 * resource-intensive to repeat twice on every request.
592 	 */
593 	if (req->oldptr == 0) {
594 		n = V_divcbinfo.ipi_count;
595 		req->oldidx = 2 * (sizeof xig)
596 			+ (n + n/8) * sizeof(struct xinpcb);
597 		return 0;
598 	}
599 
600 	if (req->newptr != 0)
601 		return EPERM;
602 
603 	/*
604 	 * OK, now we're committed to doing something.
605 	 */
606 	INP_INFO_RLOCK(&V_divcbinfo);
607 	gencnt = V_divcbinfo.ipi_gencnt;
608 	n = V_divcbinfo.ipi_count;
609 	INP_INFO_RUNLOCK(&V_divcbinfo);
610 
611 	error = sysctl_wire_old_buffer(req,
612 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
613 	if (error != 0)
614 		return (error);
615 
616 	xig.xig_len = sizeof xig;
617 	xig.xig_count = n;
618 	xig.xig_gen = gencnt;
619 	xig.xig_sogen = so_gencnt;
620 	error = SYSCTL_OUT(req, &xig, sizeof xig);
621 	if (error)
622 		return error;
623 
624 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
625 	if (inp_list == 0)
626 		return ENOMEM;
627 
628 	INP_INFO_RLOCK(&V_divcbinfo);
629 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
630 	     inp = LIST_NEXT(inp, inp_list)) {
631 		INP_WLOCK(inp);
632 		if (inp->inp_gencnt <= gencnt &&
633 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
634 			in_pcbref(inp);
635 			inp_list[i++] = inp;
636 		}
637 		INP_WUNLOCK(inp);
638 	}
639 	INP_INFO_RUNLOCK(&V_divcbinfo);
640 	n = i;
641 
642 	error = 0;
643 	for (i = 0; i < n; i++) {
644 		inp = inp_list[i];
645 		INP_RLOCK(inp);
646 		if (inp->inp_gencnt <= gencnt) {
647 			struct xinpcb xi;
648 			bzero(&xi, sizeof(xi));
649 			xi.xi_len = sizeof xi;
650 			/* XXX should avoid extra copy */
651 			bcopy(inp, &xi.xi_inp, sizeof *inp);
652 			if (inp->inp_socket)
653 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
654 			INP_RUNLOCK(inp);
655 			error = SYSCTL_OUT(req, &xi, sizeof xi);
656 		} else
657 			INP_RUNLOCK(inp);
658 	}
659 	INP_INFO_WLOCK(&V_divcbinfo);
660 	for (i = 0; i < n; i++) {
661 		inp = inp_list[i];
662 		INP_WLOCK(inp);
663 		if (!in_pcbrele(inp))
664 			INP_WUNLOCK(inp);
665 	}
666 	INP_INFO_WUNLOCK(&V_divcbinfo);
667 
668 	if (!error) {
669 		/*
670 		 * Give the user an updated idea of our state.
671 		 * If the generation differs from what we told
672 		 * her before, she knows that something happened
673 		 * while we were processing this request, and it
674 		 * might be necessary to retry.
675 		 */
676 		INP_INFO_RLOCK(&V_divcbinfo);
677 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
678 		xig.xig_sogen = so_gencnt;
679 		xig.xig_count = V_divcbinfo.ipi_count;
680 		INP_INFO_RUNLOCK(&V_divcbinfo);
681 		error = SYSCTL_OUT(req, &xig, sizeof xig);
682 	}
683 	free(inp_list, M_TEMP);
684 	return error;
685 }
686 
687 #ifdef SYSCTL_NODE
688 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
689 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
690 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
691 #endif
692 
693 struct pr_usrreqs div_usrreqs = {
694 	.pru_attach =		div_attach,
695 	.pru_bind =		div_bind,
696 	.pru_control =		in_control,
697 	.pru_detach =		div_detach,
698 	.pru_peeraddr =		in_getpeeraddr,
699 	.pru_send =		div_send,
700 	.pru_shutdown =		div_shutdown,
701 	.pru_sockaddr =		in_getsockaddr,
702 	.pru_sosetlabel =	in_pcbsosetlabel
703 };
704 
705 struct protosw div_protosw = {
706 	.pr_type =		SOCK_RAW,
707 	.pr_protocol =		IPPROTO_DIVERT,
708 	.pr_flags =		PR_ATOMIC|PR_ADDR,
709 	.pr_input =		div_input,
710 	.pr_ctlinput =		div_ctlinput,
711 	.pr_ctloutput =		ip_ctloutput,
712 	.pr_init =		div_init,
713 #ifdef VIMAGE
714 	.pr_destroy =		div_destroy,
715 #endif
716 	.pr_usrreqs =		&div_usrreqs
717 };
718 
719 static int
720 div_modevent(module_t mod, int type, void *unused)
721 {
722 	int err = 0;
723 #ifndef VIMAGE
724 	int n;
725 #endif
726 
727 	switch (type) {
728 	case MOD_LOAD:
729 		/*
730 		 * Protocol will be initialized by pf_proto_register().
731 		 * We don't have to register ip_protox because we are not
732 		 * a true IP protocol that goes over the wire.
733 		 */
734 		err = pf_proto_register(PF_INET, &div_protosw);
735 		if (err != 0)
736 			return (err);
737 		ip_divert_ptr = divert_packet;
738 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
739 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
740 		break;
741 	case MOD_QUIESCE:
742 		/*
743 		 * IPDIVERT may normally not be unloaded because of the
744 		 * potential race conditions.  Tell kldunload we can't be
745 		 * unloaded unless the unload is forced.
746 		 */
747 		err = EPERM;
748 		break;
749 	case MOD_UNLOAD:
750 #ifdef VIMAGE
751 		err = EPERM;
752 		break;
753 #else
754 		/*
755 		 * Forced unload.
756 		 *
757 		 * Module ipdivert can only be unloaded if no sockets are
758 		 * connected.  Maybe this can be changed later to forcefully
759 		 * disconnect any open sockets.
760 		 *
761 		 * XXXRW: Note that there is a slight race here, as a new
762 		 * socket open request could be spinning on the lock and then
763 		 * we destroy the lock.
764 		 */
765 		INP_INFO_WLOCK(&V_divcbinfo);
766 		n = V_divcbinfo.ipi_count;
767 		if (n != 0) {
768 			err = EBUSY;
769 			INP_INFO_WUNLOCK(&V_divcbinfo);
770 			break;
771 		}
772 		ip_divert_ptr = NULL;
773 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
774 		INP_INFO_WUNLOCK(&V_divcbinfo);
775 		div_destroy();
776 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
777 		break;
778 #endif /* !VIMAGE */
779 	default:
780 		err = EOPNOTSUPP;
781 		break;
782 	}
783 	return err;
784 }
785 
786 static moduledata_t ipdivertmod = {
787         "ipdivert",
788         div_modevent,
789         0
790 };
791 
792 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
793 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
794 MODULE_VERSION(ipdivert, 1);
795