xref: /freebsd/sys/netinet/ip_divert.c (revision 84dfba8d183d31e3412639ecb4b8ad4433cf7e80)
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 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_sctp.h"
36 #ifndef INET
37 #error "IPDIVERT requires INET"
38 #endif
39 
40 #include <sys/param.h>
41 #include <sys/eventhandler.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/if_var.h>
58 #include <net/netisr.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip_var.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/ip6_var.h>
69 #endif
70 #ifdef SCTP
71 #include <netinet/sctp_crc32.h>
72 #endif
73 
74 #include <security/mac/mac_framework.h>
75 
76 /*
77  * Divert sockets
78  */
79 
80 /*
81  * Allocate enough space to hold a full IP packet
82  */
83 #define	DIVSNDQ		(65536 + 100)
84 #define	DIVRCVQ		(65536 + 100)
85 
86 /*
87  * Divert sockets work in conjunction with ipfw or other packet filters,
88  * see the divert(4) manpage for features.
89  * Packets are selected by the packet filter and tagged with an
90  * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
91  * the packet filter) and information on the matching filter rule for
92  * subsequent reinjection. The divert_port is used to put the packet
93  * on the corresponding divert socket, while the rule number is passed
94  * up (at least partially) as the sin_port in the struct sockaddr.
95  *
96  * Packets written to the divert socket carry in sin_addr a
97  * destination address, and in sin_port the number of the filter rule
98  * after which to continue processing.
99  * If the destination address is INADDR_ANY, the packet is treated as
100  * as outgoing and sent to ip_output(); otherwise it is treated as
101  * incoming and sent to ip_input().
102  * Further, sin_zero carries some information on the interface,
103  * which can be used in the reinject -- see comments in the code.
104  *
105  * On reinjection, processing in ip_input() and ip_output()
106  * will be exactly the same as for the original packet, except that
107  * packet filter processing will start at the rule number after the one
108  * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
109  * will apply the entire ruleset to the packet).
110  */
111 
112 /* Internal variables. */
113 static VNET_DEFINE(struct inpcbhead, divcb);
114 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
115 
116 #define	V_divcb				VNET(divcb)
117 #define	V_divcbinfo			VNET(divcbinfo)
118 
119 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
120 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
121 
122 static eventhandler_tag ip_divert_event_tag;
123 
124 /*
125  * Initialize divert connection block queue.
126  */
127 static void
128 div_zone_change(void *tag)
129 {
130 
131 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
132 }
133 
134 static int
135 div_inpcb_init(void *mem, int size, int flags)
136 {
137 	struct inpcb *inp = mem;
138 
139 	INP_LOCK_INIT(inp, "inp", "divinp");
140 	return (0);
141 }
142 
143 static void
144 div_inpcb_fini(void *mem, int size)
145 {
146 	struct inpcb *inp = mem;
147 
148 	INP_LOCK_DESTROY(inp);
149 }
150 
151 static void
152 div_init(void)
153 {
154 
155 	/*
156 	 * XXX We don't use the hash list for divert IP, but it's easier to
157 	 * allocate one-entry hash lists than it is to check all over the
158 	 * place for hashbase == NULL.
159 	 */
160 	in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
161 	    div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE,
162 	    IPI_HASHFIELDS_NONE);
163 }
164 
165 static void
166 div_destroy(void)
167 {
168 
169 	in_pcbinfo_destroy(&V_divcbinfo);
170 }
171 
172 /*
173  * IPPROTO_DIVERT is not in the real IP protocol number space; this
174  * function should never be called.  Just in case, drop any packets.
175  */
176 static void
177 div_input(struct mbuf *m, int off)
178 {
179 
180 	KMOD_IPSTAT_INC(ips_noproto);
181 	m_freem(m);
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, int 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 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
201 	if (mtag == NULL) {
202 		m_freem(m);
203 		return;
204 	}
205 	/* Assure header */
206 	if (m->m_len < sizeof(struct ip) &&
207 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
208 		return;
209 	ip = mtod(m, struct ip *);
210 
211 	/* Delayed checksums are currently not compatible with divert. */
212 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
213 		in_delayed_cksum(m);
214 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
215 	}
216 #ifdef SCTP
217 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
218 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
219 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
220 	}
221 #endif
222 	bzero(&divsrc, sizeof(divsrc));
223 	divsrc.sin_len = sizeof(divsrc);
224 	divsrc.sin_family = AF_INET;
225 	/* record matching rule, in host format */
226 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
227 	/*
228 	 * Record receive interface address, if any.
229 	 * But only for incoming packets.
230 	 */
231 	if (incoming) {
232 		struct ifaddr *ifa;
233 		struct ifnet *ifp;
234 
235 		/* Sanity check */
236 		M_ASSERTPKTHDR(m);
237 
238 		/* Find IP address for receive interface */
239 		ifp = m->m_pkthdr.rcvif;
240 		if_addr_rlock(ifp);
241 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
242 			if (ifa->ifa_addr->sa_family != AF_INET)
243 				continue;
244 			divsrc.sin_addr =
245 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
246 			break;
247 		}
248 		if_addr_runlock(ifp);
249 	}
250 	/*
251 	 * Record the incoming interface name whenever we have one.
252 	 */
253 	if (m->m_pkthdr.rcvif) {
254 		/*
255 		 * Hide the actual interface name in there in the
256 		 * sin_zero array. XXX This needs to be moved to a
257 		 * different sockaddr type for divert, e.g.
258 		 * sockaddr_div with multiple fields like
259 		 * sockaddr_dl. Presently we have only 7 bytes
260 		 * but that will do for now as most interfaces
261 		 * are 4 or less + 2 or less bytes for unit.
262 		 * There is probably a faster way of doing this,
263 		 * possibly taking it from the sockaddr_dl on the iface.
264 		 * This solves the problem of a P2P link and a LAN interface
265 		 * having the same address, which can result in the wrong
266 		 * interface being assigned to the packet when fed back
267 		 * into the divert socket. Theoretically if the daemon saves
268 		 * and re-uses the sockaddr_in as suggested in the man pages,
269 		 * this iface name will come along for the ride.
270 		 * (see div_output for the other half of this.)
271 		 */
272 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
273 		    sizeof(divsrc.sin_zero));
274 	}
275 
276 	/* Put packet on socket queue, if any */
277 	sa = NULL;
278 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
279 	INP_INFO_RLOCK(&V_divcbinfo);
280 	LIST_FOREACH(inp, &V_divcb, inp_list) {
281 		/* XXX why does only one socket match? */
282 		if (inp->inp_lport == nport) {
283 			INP_RLOCK(inp);
284 			sa = inp->inp_socket;
285 			SOCKBUF_LOCK(&sa->so_rcv);
286 			if (sbappendaddr_locked(&sa->so_rcv,
287 			    (struct sockaddr *)&divsrc, m,
288 			    (struct mbuf *)0) == 0) {
289 				SOCKBUF_UNLOCK(&sa->so_rcv);
290 				sa = NULL;	/* force mbuf reclaim below */
291 			} else
292 				sorwakeup_locked(sa);
293 			INP_RUNLOCK(inp);
294 			break;
295 		}
296 	}
297 	INP_INFO_RUNLOCK(&V_divcbinfo);
298 	if (sa == NULL) {
299 		m_freem(m);
300 		KMOD_IPSTAT_INC(ips_noproto);
301 		KMOD_IPSTAT_DEC(ips_delivered);
302         }
303 }
304 
305 /*
306  * Deliver packet back into the IP processing machinery.
307  *
308  * If no address specified, or address is 0.0.0.0, send to ip_output();
309  * otherwise, send to ip_input() and mark as having been received on
310  * the interface with that address.
311  */
312 static int
313 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
314     struct mbuf *control)
315 {
316 	struct ip *const ip = mtod(m, struct ip *);
317 	struct m_tag *mtag;
318 	struct ipfw_rule_ref *dt;
319 	int error = 0;
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 	if (control)
330 		m_freem(control);		/* XXX */
331 
332 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
333 	if (mtag == NULL) {
334 		/* this should be normal */
335 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
336 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
337 		if (mtag == NULL) {
338 			error = ENOBUFS;
339 			goto cantsend;
340 		}
341 		m_tag_prepend(m, mtag);
342 	}
343 	dt = (struct ipfw_rule_ref *)(mtag+1);
344 
345 	/* Loopback avoidance and state recovery */
346 	if (sin) {
347 		int i;
348 
349 		/* set the starting point. We provide a non-zero slot,
350 		 * but a non_matching chain_id to skip that info and use
351 		 * the rulenum/rule_id.
352 		 */
353 		dt->slot = 1; /* dummy, chain_id is invalid */
354 		dt->chain_id = 0;
355 		dt->rulenum = sin->sin_port+1; /* host format ? */
356 		dt->rule_id = 0;
357 		/*
358 		 * Find receive interface with the given name, stuffed
359 		 * (if it exists) in the sin_zero[] field.
360 		 * The name is user supplied data so don't trust its size
361 		 * or that it is zero terminated.
362 		 */
363 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
364 			;
365 		if ( i > 0 && i < sizeof(sin->sin_zero))
366 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
367 	}
368 
369 	/* Reinject packet into the system as incoming or outgoing */
370 	if (!sin || sin->sin_addr.s_addr == 0) {
371 		struct mbuf *options = NULL;
372 		struct inpcb *inp;
373 
374 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
375 		inp = sotoinpcb(so);
376 		INP_RLOCK(inp);
377 		switch (ip->ip_v) {
378 		case IPVERSION:
379 			/*
380 			 * Don't allow both user specified and setsockopt
381 			 * options, and don't allow packet length sizes that
382 			 * will crash.
383 			 */
384 			if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
385 			    inp->inp_options != NULL) ||
386 			    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
387 				error = EINVAL;
388 				INP_RUNLOCK(inp);
389 				goto cantsend;
390 			}
391 			break;
392 #ifdef INET6
393 		case IPV6_VERSION >> 4:
394 		    {
395 			struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
396 
397 			/* Don't allow packet length sizes that will crash */
398 			if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
399 				error = EINVAL;
400 				INP_RUNLOCK(inp);
401 				goto cantsend;
402 			}
403 			break;
404 		    }
405 #endif
406 		default:
407 			error = EINVAL;
408 			INP_RUNLOCK(inp);
409 			goto cantsend;
410 		}
411 
412 		/* Send packet to output processing */
413 		KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
414 
415 #ifdef MAC
416 		mac_inpcb_create_mbuf(inp, m);
417 #endif
418 		/*
419 		 * Get ready to inject the packet into ip_output().
420 		 * Just in case socket options were specified on the
421 		 * divert socket, we duplicate them.  This is done
422 		 * to avoid having to hold the PCB locks over the call
423 		 * to ip_output(), as doing this results in a number of
424 		 * lock ordering complexities.
425 		 *
426 		 * Note that we set the multicast options argument for
427 		 * ip_output() to NULL since it should be invariant that
428 		 * they are not present.
429 		 */
430 		KASSERT(inp->inp_moptions == NULL,
431 		    ("multicast options set on a divert socket"));
432 		/*
433 		 * XXXCSJP: It is unclear to me whether or not it makes
434 		 * sense for divert sockets to have options.  However,
435 		 * for now we will duplicate them with the INP locks
436 		 * held so we can use them in ip_output() without
437 		 * requring a reference to the pcb.
438 		 */
439 		if (inp->inp_options != NULL) {
440 			options = m_dup(inp->inp_options, M_NOWAIT);
441 			if (options == NULL) {
442 				INP_RUNLOCK(inp);
443 				error = ENOBUFS;
444 				goto cantsend;
445 			}
446 		}
447 		INP_RUNLOCK(inp);
448 
449 		switch (ip->ip_v) {
450 		case IPVERSION:
451 			error = ip_output(m, options, NULL,
452 			    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
453 			    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
454 			break;
455 #ifdef INET6
456 		case IPV6_VERSION >> 4:
457 			error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
458 			break;
459 #endif
460 		}
461 		if (options != NULL)
462 			m_freem(options);
463 	} else {
464 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
465 		if (m->m_pkthdr.rcvif == NULL) {
466 			/*
467 			 * No luck with the name, check by IP address.
468 			 * Clear the port and the ifname to make sure
469 			 * there are no distractions for ifa_ifwithaddr.
470 			 */
471 			struct	ifaddr *ifa;
472 
473 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
474 			sin->sin_port = 0;
475 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
476 			if (ifa == NULL) {
477 				error = EADDRNOTAVAIL;
478 				goto cantsend;
479 			}
480 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
481 			ifa_free(ifa);
482 		}
483 #ifdef MAC
484 		mac_socket_create_mbuf(so, m);
485 #endif
486 		/* Send packet to input processing via netisr */
487 		switch (ip->ip_v) {
488 		case IPVERSION:
489 			netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
490 			break;
491 #ifdef INET6
492 		case IPV6_VERSION >> 4:
493 			netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
494 			break;
495 #endif
496 		default:
497 			error = EINVAL;
498 			goto cantsend;
499 		}
500 	}
501 
502 	return (error);
503 
504 cantsend:
505 	m_freem(m);
506 	return (error);
507 }
508 
509 static int
510 div_attach(struct socket *so, int proto, struct thread *td)
511 {
512 	struct inpcb *inp;
513 	int error;
514 
515 	inp  = sotoinpcb(so);
516 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
517 	if (td != NULL) {
518 		error = priv_check(td, PRIV_NETINET_DIVERT);
519 		if (error)
520 			return (error);
521 	}
522 	error = soreserve(so, div_sendspace, div_recvspace);
523 	if (error)
524 		return error;
525 	INP_INFO_WLOCK(&V_divcbinfo);
526 	error = in_pcballoc(so, &V_divcbinfo);
527 	if (error) {
528 		INP_INFO_WUNLOCK(&V_divcbinfo);
529 		return error;
530 	}
531 	inp = (struct inpcb *)so->so_pcb;
532 	INP_INFO_WUNLOCK(&V_divcbinfo);
533 	inp->inp_ip_p = proto;
534 	inp->inp_vflag |= INP_IPV4;
535 	inp->inp_flags |= INP_HDRINCL;
536 	INP_WUNLOCK(inp);
537 	return 0;
538 }
539 
540 static void
541 div_detach(struct socket *so)
542 {
543 	struct inpcb *inp;
544 
545 	inp = sotoinpcb(so);
546 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
547 	INP_INFO_WLOCK(&V_divcbinfo);
548 	INP_WLOCK(inp);
549 	in_pcbdetach(inp);
550 	in_pcbfree(inp);
551 	INP_INFO_WUNLOCK(&V_divcbinfo);
552 }
553 
554 static int
555 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
556 {
557 	struct inpcb *inp;
558 	int error;
559 
560 	inp = sotoinpcb(so);
561 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
562 	/* in_pcbbind assumes that nam is a sockaddr_in
563 	 * and in_pcbbind requires a valid address. Since divert
564 	 * sockets don't we need to make sure the address is
565 	 * filled in properly.
566 	 * XXX -- divert should not be abusing in_pcbind
567 	 * and should probably have its own family.
568 	 */
569 	if (nam->sa_family != AF_INET)
570 		return EAFNOSUPPORT;
571 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
572 	INP_INFO_WLOCK(&V_divcbinfo);
573 	INP_WLOCK(inp);
574 	INP_HASH_WLOCK(&V_divcbinfo);
575 	error = in_pcbbind(inp, nam, td->td_ucred);
576 	INP_HASH_WUNLOCK(&V_divcbinfo);
577 	INP_WUNLOCK(inp);
578 	INP_INFO_WUNLOCK(&V_divcbinfo);
579 	return error;
580 }
581 
582 static int
583 div_shutdown(struct socket *so)
584 {
585 	struct inpcb *inp;
586 
587 	inp = sotoinpcb(so);
588 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
589 	INP_WLOCK(inp);
590 	socantsendmore(so);
591 	INP_WUNLOCK(inp);
592 	return 0;
593 }
594 
595 static int
596 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
597     struct mbuf *control, struct thread *td)
598 {
599 
600 	/* Packet must have a header (but that's about it) */
601 	if (m->m_len < sizeof (struct ip) &&
602 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
603 		KMOD_IPSTAT_INC(ips_toosmall);
604 		m_freem(m);
605 		return EINVAL;
606 	}
607 
608 	/* Send packet */
609 	return div_output(so, m, (struct sockaddr_in *)nam, control);
610 }
611 
612 static void
613 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
614 {
615         struct in_addr faddr;
616 
617 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
618 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
619         	return;
620 	if (PRC_IS_REDIRECT(cmd))
621 		return;
622 }
623 
624 static int
625 div_pcblist(SYSCTL_HANDLER_ARGS)
626 {
627 	int error, i, n;
628 	struct inpcb *inp, **inp_list;
629 	inp_gen_t gencnt;
630 	struct xinpgen xig;
631 
632 	/*
633 	 * The process of preparing the TCB list is too time-consuming and
634 	 * resource-intensive to repeat twice on every request.
635 	 */
636 	if (req->oldptr == 0) {
637 		n = V_divcbinfo.ipi_count;
638 		n += imax(n / 8, 10);
639 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
640 		return 0;
641 	}
642 
643 	if (req->newptr != 0)
644 		return EPERM;
645 
646 	/*
647 	 * OK, now we're committed to doing something.
648 	 */
649 	INP_INFO_RLOCK(&V_divcbinfo);
650 	gencnt = V_divcbinfo.ipi_gencnt;
651 	n = V_divcbinfo.ipi_count;
652 	INP_INFO_RUNLOCK(&V_divcbinfo);
653 
654 	error = sysctl_wire_old_buffer(req,
655 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
656 	if (error != 0)
657 		return (error);
658 
659 	xig.xig_len = sizeof xig;
660 	xig.xig_count = n;
661 	xig.xig_gen = gencnt;
662 	xig.xig_sogen = so_gencnt;
663 	error = SYSCTL_OUT(req, &xig, sizeof xig);
664 	if (error)
665 		return error;
666 
667 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
668 	if (inp_list == 0)
669 		return ENOMEM;
670 
671 	INP_INFO_RLOCK(&V_divcbinfo);
672 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
673 	     inp = LIST_NEXT(inp, inp_list)) {
674 		INP_WLOCK(inp);
675 		if (inp->inp_gencnt <= gencnt &&
676 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
677 			in_pcbref(inp);
678 			inp_list[i++] = inp;
679 		}
680 		INP_WUNLOCK(inp);
681 	}
682 	INP_INFO_RUNLOCK(&V_divcbinfo);
683 	n = i;
684 
685 	error = 0;
686 	for (i = 0; i < n; i++) {
687 		inp = inp_list[i];
688 		INP_RLOCK(inp);
689 		if (inp->inp_gencnt <= gencnt) {
690 			struct xinpcb xi;
691 			bzero(&xi, sizeof(xi));
692 			xi.xi_len = sizeof xi;
693 			/* XXX should avoid extra copy */
694 			bcopy(inp, &xi.xi_inp, sizeof *inp);
695 			if (inp->inp_socket)
696 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
697 			INP_RUNLOCK(inp);
698 			error = SYSCTL_OUT(req, &xi, sizeof xi);
699 		} else
700 			INP_RUNLOCK(inp);
701 	}
702 	INP_INFO_WLOCK(&V_divcbinfo);
703 	for (i = 0; i < n; i++) {
704 		inp = inp_list[i];
705 		INP_RLOCK(inp);
706 		if (!in_pcbrele_rlocked(inp))
707 			INP_RUNLOCK(inp);
708 	}
709 	INP_INFO_WUNLOCK(&V_divcbinfo);
710 
711 	if (!error) {
712 		/*
713 		 * Give the user an updated idea of our state.
714 		 * If the generation differs from what we told
715 		 * her before, she knows that something happened
716 		 * while we were processing this request, and it
717 		 * might be necessary to retry.
718 		 */
719 		INP_INFO_RLOCK(&V_divcbinfo);
720 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
721 		xig.xig_sogen = so_gencnt;
722 		xig.xig_count = V_divcbinfo.ipi_count;
723 		INP_INFO_RUNLOCK(&V_divcbinfo);
724 		error = SYSCTL_OUT(req, &xig, sizeof xig);
725 	}
726 	free(inp_list, M_TEMP);
727 	return error;
728 }
729 
730 #ifdef SYSCTL_NODE
731 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
732     "IPDIVERT");
733 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
734     NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
735 #endif
736 
737 struct pr_usrreqs div_usrreqs = {
738 	.pru_attach =		div_attach,
739 	.pru_bind =		div_bind,
740 	.pru_control =		in_control,
741 	.pru_detach =		div_detach,
742 	.pru_peeraddr =		in_getpeeraddr,
743 	.pru_send =		div_send,
744 	.pru_shutdown =		div_shutdown,
745 	.pru_sockaddr =		in_getsockaddr,
746 	.pru_sosetlabel =	in_pcbsosetlabel
747 };
748 
749 struct protosw div_protosw = {
750 	.pr_type =		SOCK_RAW,
751 	.pr_protocol =		IPPROTO_DIVERT,
752 	.pr_flags =		PR_ATOMIC|PR_ADDR,
753 	.pr_input =		div_input,
754 	.pr_ctlinput =		div_ctlinput,
755 	.pr_ctloutput =		ip_ctloutput,
756 	.pr_init =		div_init,
757 #ifdef VIMAGE
758 	.pr_destroy =		div_destroy,
759 #endif
760 	.pr_usrreqs =		&div_usrreqs
761 };
762 
763 static int
764 div_modevent(module_t mod, int type, void *unused)
765 {
766 	int err = 0;
767 #ifndef VIMAGE
768 	int n;
769 #endif
770 
771 	switch (type) {
772 	case MOD_LOAD:
773 		/*
774 		 * Protocol will be initialized by pf_proto_register().
775 		 * We don't have to register ip_protox because we are not
776 		 * a true IP protocol that goes over the wire.
777 		 */
778 		err = pf_proto_register(PF_INET, &div_protosw);
779 		if (err != 0)
780 			return (err);
781 		ip_divert_ptr = divert_packet;
782 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
783 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
784 		break;
785 	case MOD_QUIESCE:
786 		/*
787 		 * IPDIVERT may normally not be unloaded because of the
788 		 * potential race conditions.  Tell kldunload we can't be
789 		 * unloaded unless the unload is forced.
790 		 */
791 		err = EPERM;
792 		break;
793 	case MOD_UNLOAD:
794 #ifdef VIMAGE
795 		err = EPERM;
796 		break;
797 #else
798 		/*
799 		 * Forced unload.
800 		 *
801 		 * Module ipdivert can only be unloaded if no sockets are
802 		 * connected.  Maybe this can be changed later to forcefully
803 		 * disconnect any open sockets.
804 		 *
805 		 * XXXRW: Note that there is a slight race here, as a new
806 		 * socket open request could be spinning on the lock and then
807 		 * we destroy the lock.
808 		 */
809 		INP_INFO_WLOCK(&V_divcbinfo);
810 		n = V_divcbinfo.ipi_count;
811 		if (n != 0) {
812 			err = EBUSY;
813 			INP_INFO_WUNLOCK(&V_divcbinfo);
814 			break;
815 		}
816 		ip_divert_ptr = NULL;
817 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
818 		INP_INFO_WUNLOCK(&V_divcbinfo);
819 		div_destroy();
820 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
821 		break;
822 #endif /* !VIMAGE */
823 	default:
824 		err = EOPNOTSUPP;
825 		break;
826 	}
827 	return err;
828 }
829 
830 static moduledata_t ipdivertmod = {
831         "ipdivert",
832         div_modevent,
833         0
834 };
835 
836 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
837 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
838 MODULE_VERSION(ipdivert, 1);
839