xref: /freebsd/sys/netinet/ip_divert.c (revision a321cc5dc908a14d42e57e2468923937f18c21fc)
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, 0, IPI_HASHFIELDS_NONE);
162 }
163 
164 static void
165 div_destroy(void)
166 {
167 
168 	in_pcbinfo_destroy(&V_divcbinfo);
169 }
170 
171 /*
172  * IPPROTO_DIVERT is not in the real IP protocol number space; this
173  * function should never be called.  Just in case, drop any packets.
174  */
175 static int
176 div_input(struct mbuf **mp, int *offp, int proto)
177 {
178 	struct mbuf *m = *mp;
179 
180 	KMOD_IPSTAT_INC(ips_noproto);
181 	m_freem(m);
182 	return (IPPROTO_DONE);
183 }
184 
185 /*
186  * Divert a packet by passing it up to the divert socket at port 'port'.
187  *
188  * Setup generic address and protocol structures for div_input routine,
189  * then pass them along with mbuf chain.
190  */
191 static void
192 divert_packet(struct mbuf *m, int incoming)
193 {
194 	struct ip *ip;
195 	struct inpcb *inp;
196 	struct socket *sa;
197 	u_int16_t nport;
198 	struct sockaddr_in divsrc;
199 	struct m_tag *mtag;
200 
201 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
202 	if (mtag == NULL) {
203 		m_freem(m);
204 		return;
205 	}
206 	/* Assure header */
207 	if (m->m_len < sizeof(struct ip) &&
208 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
209 		return;
210 	ip = mtod(m, struct ip *);
211 
212 	/* Delayed checksums are currently not compatible with divert. */
213 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
214 		in_delayed_cksum(m);
215 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
216 	}
217 #ifdef SCTP
218 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
219 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
220 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
221 	}
222 #endif
223 	bzero(&divsrc, sizeof(divsrc));
224 	divsrc.sin_len = sizeof(divsrc);
225 	divsrc.sin_family = AF_INET;
226 	/* record matching rule, in host format */
227 	divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
228 	/*
229 	 * Record receive interface address, if any.
230 	 * But only for incoming packets.
231 	 */
232 	if (incoming) {
233 		struct ifaddr *ifa;
234 		struct ifnet *ifp;
235 
236 		/* Sanity check */
237 		M_ASSERTPKTHDR(m);
238 
239 		/* Find IP address for receive interface */
240 		ifp = m->m_pkthdr.rcvif;
241 		if_addr_rlock(ifp);
242 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
243 			if (ifa->ifa_addr->sa_family != AF_INET)
244 				continue;
245 			divsrc.sin_addr =
246 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
247 			break;
248 		}
249 		if_addr_runlock(ifp);
250 	}
251 	/*
252 	 * Record the incoming interface name whenever we have one.
253 	 */
254 	if (m->m_pkthdr.rcvif) {
255 		/*
256 		 * Hide the actual interface name in there in the
257 		 * sin_zero array. XXX This needs to be moved to a
258 		 * different sockaddr type for divert, e.g.
259 		 * sockaddr_div with multiple fields like
260 		 * sockaddr_dl. Presently we have only 7 bytes
261 		 * but that will do for now as most interfaces
262 		 * are 4 or less + 2 or less bytes for unit.
263 		 * There is probably a faster way of doing this,
264 		 * possibly taking it from the sockaddr_dl on the iface.
265 		 * This solves the problem of a P2P link and a LAN interface
266 		 * having the same address, which can result in the wrong
267 		 * interface being assigned to the packet when fed back
268 		 * into the divert socket. Theoretically if the daemon saves
269 		 * and re-uses the sockaddr_in as suggested in the man pages,
270 		 * this iface name will come along for the ride.
271 		 * (see div_output for the other half of this.)
272 		 */
273 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
274 		    sizeof(divsrc.sin_zero));
275 	}
276 
277 	/* Put packet on socket queue, if any */
278 	sa = NULL;
279 	nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
280 	INP_INFO_RLOCK(&V_divcbinfo);
281 	LIST_FOREACH(inp, &V_divcb, inp_list) {
282 		/* XXX why does only one socket match? */
283 		if (inp->inp_lport == nport) {
284 			INP_RLOCK(inp);
285 			sa = inp->inp_socket;
286 			SOCKBUF_LOCK(&sa->so_rcv);
287 			if (sbappendaddr_locked(&sa->so_rcv,
288 			    (struct sockaddr *)&divsrc, m,
289 			    (struct mbuf *)0) == 0) {
290 				SOCKBUF_UNLOCK(&sa->so_rcv);
291 				sa = NULL;	/* force mbuf reclaim below */
292 			} else
293 				sorwakeup_locked(sa);
294 			INP_RUNLOCK(inp);
295 			break;
296 		}
297 	}
298 	INP_INFO_RUNLOCK(&V_divcbinfo);
299 	if (sa == NULL) {
300 		m_freem(m);
301 		KMOD_IPSTAT_INC(ips_noproto);
302 		KMOD_IPSTAT_DEC(ips_delivered);
303         }
304 }
305 
306 /*
307  * Deliver packet back into the IP processing machinery.
308  *
309  * If no address specified, or address is 0.0.0.0, send to ip_output();
310  * otherwise, send to ip_input() and mark as having been received on
311  * the interface with that address.
312  */
313 static int
314 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
315     struct mbuf *control)
316 {
317 	struct ip *const ip = mtod(m, struct ip *);
318 	struct m_tag *mtag;
319 	struct ipfw_rule_ref *dt;
320 	int error = 0;
321 
322 	/*
323 	 * An mbuf may hasn't come from userland, but we pretend
324 	 * that it has.
325 	 */
326 	m->m_pkthdr.rcvif = NULL;
327 	m->m_nextpkt = NULL;
328 	M_SETFIB(m, so->so_fibnum);
329 
330 	if (control)
331 		m_freem(control);		/* XXX */
332 
333 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
334 	if (mtag == NULL) {
335 		/* this should be normal */
336 		mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
337 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
338 		if (mtag == NULL) {
339 			error = ENOBUFS;
340 			goto cantsend;
341 		}
342 		m_tag_prepend(m, mtag);
343 	}
344 	dt = (struct ipfw_rule_ref *)(mtag+1);
345 
346 	/* Loopback avoidance and state recovery */
347 	if (sin) {
348 		int i;
349 
350 		/* set the starting point. We provide a non-zero slot,
351 		 * but a non_matching chain_id to skip that info and use
352 		 * the rulenum/rule_id.
353 		 */
354 		dt->slot = 1; /* dummy, chain_id is invalid */
355 		dt->chain_id = 0;
356 		dt->rulenum = sin->sin_port+1; /* host format ? */
357 		dt->rule_id = 0;
358 		/*
359 		 * Find receive interface with the given name, stuffed
360 		 * (if it exists) in the sin_zero[] field.
361 		 * The name is user supplied data so don't trust its size
362 		 * or that it is zero terminated.
363 		 */
364 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
365 			;
366 		if ( i > 0 && i < sizeof(sin->sin_zero))
367 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
368 	}
369 
370 	/* Reinject packet into the system as incoming or outgoing */
371 	if (!sin || sin->sin_addr.s_addr == 0) {
372 		struct mbuf *options = NULL;
373 		struct inpcb *inp;
374 
375 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
376 		inp = sotoinpcb(so);
377 		INP_RLOCK(inp);
378 		switch (ip->ip_v) {
379 		case IPVERSION:
380 			/*
381 			 * Don't allow both user specified and setsockopt
382 			 * options, and don't allow packet length sizes that
383 			 * will crash.
384 			 */
385 			if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
386 			    inp->inp_options != NULL) ||
387 			    ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
388 				error = EINVAL;
389 				INP_RUNLOCK(inp);
390 				goto cantsend;
391 			}
392 			break;
393 #ifdef INET6
394 		case IPV6_VERSION >> 4:
395 		    {
396 			struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
397 
398 			/* Don't allow packet length sizes that will crash */
399 			if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
400 				error = EINVAL;
401 				INP_RUNLOCK(inp);
402 				goto cantsend;
403 			}
404 			break;
405 		    }
406 #endif
407 		default:
408 			error = EINVAL;
409 			INP_RUNLOCK(inp);
410 			goto cantsend;
411 		}
412 
413 		/* Send packet to output processing */
414 		KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
415 
416 #ifdef MAC
417 		mac_inpcb_create_mbuf(inp, m);
418 #endif
419 		/*
420 		 * Get ready to inject the packet into ip_output().
421 		 * Just in case socket options were specified on the
422 		 * divert socket, we duplicate them.  This is done
423 		 * to avoid having to hold the PCB locks over the call
424 		 * to ip_output(), as doing this results in a number of
425 		 * lock ordering complexities.
426 		 *
427 		 * Note that we set the multicast options argument for
428 		 * ip_output() to NULL since it should be invariant that
429 		 * they are not present.
430 		 */
431 		KASSERT(inp->inp_moptions == NULL,
432 		    ("multicast options set on a divert socket"));
433 		/*
434 		 * XXXCSJP: It is unclear to me whether or not it makes
435 		 * sense for divert sockets to have options.  However,
436 		 * for now we will duplicate them with the INP locks
437 		 * held so we can use them in ip_output() without
438 		 * requring a reference to the pcb.
439 		 */
440 		if (inp->inp_options != NULL) {
441 			options = m_dup(inp->inp_options, M_NOWAIT);
442 			if (options == NULL) {
443 				INP_RUNLOCK(inp);
444 				error = ENOBUFS;
445 				goto cantsend;
446 			}
447 		}
448 		INP_RUNLOCK(inp);
449 
450 		switch (ip->ip_v) {
451 		case IPVERSION:
452 			error = ip_output(m, options, NULL,
453 			    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
454 			    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
455 			break;
456 #ifdef INET6
457 		case IPV6_VERSION >> 4:
458 			error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
459 			break;
460 #endif
461 		}
462 		if (options != NULL)
463 			m_freem(options);
464 	} else {
465 		dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
466 		if (m->m_pkthdr.rcvif == NULL) {
467 			/*
468 			 * No luck with the name, check by IP address.
469 			 * Clear the port and the ifname to make sure
470 			 * there are no distractions for ifa_ifwithaddr.
471 			 */
472 			struct	ifaddr *ifa;
473 
474 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
475 			sin->sin_port = 0;
476 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
477 			if (ifa == NULL) {
478 				error = EADDRNOTAVAIL;
479 				goto cantsend;
480 			}
481 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
482 			ifa_free(ifa);
483 		}
484 #ifdef MAC
485 		mac_socket_create_mbuf(so, m);
486 #endif
487 		/* Send packet to input processing via netisr */
488 		switch (ip->ip_v) {
489 		case IPVERSION:
490 			netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
491 			break;
492 #ifdef INET6
493 		case IPV6_VERSION >> 4:
494 			netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
495 			break;
496 #endif
497 		default:
498 			error = EINVAL;
499 			goto cantsend;
500 		}
501 	}
502 
503 	return (error);
504 
505 cantsend:
506 	m_freem(m);
507 	return (error);
508 }
509 
510 static int
511 div_attach(struct socket *so, int proto, struct thread *td)
512 {
513 	struct inpcb *inp;
514 	int error;
515 
516 	inp  = sotoinpcb(so);
517 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
518 	if (td != NULL) {
519 		error = priv_check(td, PRIV_NETINET_DIVERT);
520 		if (error)
521 			return (error);
522 	}
523 	error = soreserve(so, div_sendspace, div_recvspace);
524 	if (error)
525 		return error;
526 	INP_INFO_WLOCK(&V_divcbinfo);
527 	error = in_pcballoc(so, &V_divcbinfo);
528 	if (error) {
529 		INP_INFO_WUNLOCK(&V_divcbinfo);
530 		return error;
531 	}
532 	inp = (struct inpcb *)so->so_pcb;
533 	INP_INFO_WUNLOCK(&V_divcbinfo);
534 	inp->inp_ip_p = proto;
535 	inp->inp_vflag |= INP_IPV4;
536 	inp->inp_flags |= INP_HDRINCL;
537 	INP_WUNLOCK(inp);
538 	return 0;
539 }
540 
541 static void
542 div_detach(struct socket *so)
543 {
544 	struct inpcb *inp;
545 
546 	inp = sotoinpcb(so);
547 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
548 	INP_INFO_WLOCK(&V_divcbinfo);
549 	INP_WLOCK(inp);
550 	in_pcbdetach(inp);
551 	in_pcbfree(inp);
552 	INP_INFO_WUNLOCK(&V_divcbinfo);
553 }
554 
555 static int
556 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
557 {
558 	struct inpcb *inp;
559 	int error;
560 
561 	inp = sotoinpcb(so);
562 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
563 	/* in_pcbbind assumes that nam is a sockaddr_in
564 	 * and in_pcbbind requires a valid address. Since divert
565 	 * sockets don't we need to make sure the address is
566 	 * filled in properly.
567 	 * XXX -- divert should not be abusing in_pcbind
568 	 * and should probably have its own family.
569 	 */
570 	if (nam->sa_family != AF_INET)
571 		return EAFNOSUPPORT;
572 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
573 	INP_INFO_WLOCK(&V_divcbinfo);
574 	INP_WLOCK(inp);
575 	INP_HASH_WLOCK(&V_divcbinfo);
576 	error = in_pcbbind(inp, nam, td->td_ucred);
577 	INP_HASH_WUNLOCK(&V_divcbinfo);
578 	INP_WUNLOCK(inp);
579 	INP_INFO_WUNLOCK(&V_divcbinfo);
580 	return error;
581 }
582 
583 static int
584 div_shutdown(struct socket *so)
585 {
586 	struct inpcb *inp;
587 
588 	inp = sotoinpcb(so);
589 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
590 	INP_WLOCK(inp);
591 	socantsendmore(so);
592 	INP_WUNLOCK(inp);
593 	return 0;
594 }
595 
596 static int
597 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
598     struct mbuf *control, struct thread *td)
599 {
600 
601 	/* Packet must have a header (but that's about it) */
602 	if (m->m_len < sizeof (struct ip) &&
603 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
604 		KMOD_IPSTAT_INC(ips_toosmall);
605 		m_freem(m);
606 		return EINVAL;
607 	}
608 
609 	/* Send packet */
610 	return div_output(so, m, (struct sockaddr_in *)nam, control);
611 }
612 
613 static void
614 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
615 {
616         struct in_addr faddr;
617 
618 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
619 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
620         	return;
621 	if (PRC_IS_REDIRECT(cmd))
622 		return;
623 }
624 
625 static int
626 div_pcblist(SYSCTL_HANDLER_ARGS)
627 {
628 	int error, i, n;
629 	struct inpcb *inp, **inp_list;
630 	inp_gen_t gencnt;
631 	struct xinpgen xig;
632 
633 	/*
634 	 * The process of preparing the TCB list is too time-consuming and
635 	 * resource-intensive to repeat twice on every request.
636 	 */
637 	if (req->oldptr == 0) {
638 		n = V_divcbinfo.ipi_count;
639 		n += imax(n / 8, 10);
640 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
641 		return 0;
642 	}
643 
644 	if (req->newptr != 0)
645 		return EPERM;
646 
647 	/*
648 	 * OK, now we're committed to doing something.
649 	 */
650 	INP_INFO_RLOCK(&V_divcbinfo);
651 	gencnt = V_divcbinfo.ipi_gencnt;
652 	n = V_divcbinfo.ipi_count;
653 	INP_INFO_RUNLOCK(&V_divcbinfo);
654 
655 	error = sysctl_wire_old_buffer(req,
656 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
657 	if (error != 0)
658 		return (error);
659 
660 	xig.xig_len = sizeof xig;
661 	xig.xig_count = n;
662 	xig.xig_gen = gencnt;
663 	xig.xig_sogen = so_gencnt;
664 	error = SYSCTL_OUT(req, &xig, sizeof xig);
665 	if (error)
666 		return error;
667 
668 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
669 	if (inp_list == NULL)
670 		return ENOMEM;
671 
672 	INP_INFO_RLOCK(&V_divcbinfo);
673 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
674 	     inp = LIST_NEXT(inp, inp_list)) {
675 		INP_WLOCK(inp);
676 		if (inp->inp_gencnt <= gencnt &&
677 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
678 			in_pcbref(inp);
679 			inp_list[i++] = inp;
680 		}
681 		INP_WUNLOCK(inp);
682 	}
683 	INP_INFO_RUNLOCK(&V_divcbinfo);
684 	n = i;
685 
686 	error = 0;
687 	for (i = 0; i < n; i++) {
688 		inp = inp_list[i];
689 		INP_RLOCK(inp);
690 		if (inp->inp_gencnt <= gencnt) {
691 			struct xinpcb xi;
692 			bzero(&xi, sizeof(xi));
693 			xi.xi_len = sizeof xi;
694 			/* XXX should avoid extra copy */
695 			bcopy(inp, &xi.xi_inp, sizeof *inp);
696 			if (inp->inp_socket)
697 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
698 			INP_RUNLOCK(inp);
699 			error = SYSCTL_OUT(req, &xi, sizeof xi);
700 		} else
701 			INP_RUNLOCK(inp);
702 	}
703 	INP_INFO_WLOCK(&V_divcbinfo);
704 	for (i = 0; i < n; i++) {
705 		inp = inp_list[i];
706 		INP_RLOCK(inp);
707 		if (!in_pcbrele_rlocked(inp))
708 			INP_RUNLOCK(inp);
709 	}
710 	INP_INFO_WUNLOCK(&V_divcbinfo);
711 
712 	if (!error) {
713 		/*
714 		 * Give the user an updated idea of our state.
715 		 * If the generation differs from what we told
716 		 * her before, she knows that something happened
717 		 * while we were processing this request, and it
718 		 * might be necessary to retry.
719 		 */
720 		INP_INFO_RLOCK(&V_divcbinfo);
721 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
722 		xig.xig_sogen = so_gencnt;
723 		xig.xig_count = V_divcbinfo.ipi_count;
724 		INP_INFO_RUNLOCK(&V_divcbinfo);
725 		error = SYSCTL_OUT(req, &xig, sizeof xig);
726 	}
727 	free(inp_list, M_TEMP);
728 	return error;
729 }
730 
731 #ifdef SYSCTL_NODE
732 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
733     "IPDIVERT");
734 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
735     NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
736 #endif
737 
738 struct pr_usrreqs div_usrreqs = {
739 	.pru_attach =		div_attach,
740 	.pru_bind =		div_bind,
741 	.pru_control =		in_control,
742 	.pru_detach =		div_detach,
743 	.pru_peeraddr =		in_getpeeraddr,
744 	.pru_send =		div_send,
745 	.pru_shutdown =		div_shutdown,
746 	.pru_sockaddr =		in_getsockaddr,
747 	.pru_sosetlabel =	in_pcbsosetlabel
748 };
749 
750 struct protosw div_protosw = {
751 	.pr_type =		SOCK_RAW,
752 	.pr_protocol =		IPPROTO_DIVERT,
753 	.pr_flags =		PR_ATOMIC|PR_ADDR,
754 	.pr_input =		div_input,
755 	.pr_ctlinput =		div_ctlinput,
756 	.pr_ctloutput =		ip_ctloutput,
757 	.pr_init =		div_init,
758 #ifdef VIMAGE
759 	.pr_destroy =		div_destroy,
760 #endif
761 	.pr_usrreqs =		&div_usrreqs
762 };
763 
764 static int
765 div_modevent(module_t mod, int type, void *unused)
766 {
767 	int err = 0;
768 #ifndef VIMAGE
769 	int n;
770 #endif
771 
772 	switch (type) {
773 	case MOD_LOAD:
774 		/*
775 		 * Protocol will be initialized by pf_proto_register().
776 		 * We don't have to register ip_protox because we are not
777 		 * a true IP protocol that goes over the wire.
778 		 */
779 		err = pf_proto_register(PF_INET, &div_protosw);
780 		if (err != 0)
781 			return (err);
782 		ip_divert_ptr = divert_packet;
783 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
784 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
785 		break;
786 	case MOD_QUIESCE:
787 		/*
788 		 * IPDIVERT may normally not be unloaded because of the
789 		 * potential race conditions.  Tell kldunload we can't be
790 		 * unloaded unless the unload is forced.
791 		 */
792 		err = EPERM;
793 		break;
794 	case MOD_UNLOAD:
795 #ifdef VIMAGE
796 		err = EPERM;
797 		break;
798 #else
799 		/*
800 		 * Forced unload.
801 		 *
802 		 * Module ipdivert can only be unloaded if no sockets are
803 		 * connected.  Maybe this can be changed later to forcefully
804 		 * disconnect any open sockets.
805 		 *
806 		 * XXXRW: Note that there is a slight race here, as a new
807 		 * socket open request could be spinning on the lock and then
808 		 * we destroy the lock.
809 		 */
810 		INP_INFO_WLOCK(&V_divcbinfo);
811 		n = V_divcbinfo.ipi_count;
812 		if (n != 0) {
813 			err = EBUSY;
814 			INP_INFO_WUNLOCK(&V_divcbinfo);
815 			break;
816 		}
817 		ip_divert_ptr = NULL;
818 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
819 		INP_INFO_WUNLOCK(&V_divcbinfo);
820 		div_destroy();
821 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
822 		break;
823 #endif /* !VIMAGE */
824 	default:
825 		err = EOPNOTSUPP;
826 		break;
827 	}
828 	return err;
829 }
830 
831 static moduledata_t ipdivertmod = {
832         "ipdivert",
833         div_modevent,
834         0
835 };
836 
837 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
838 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
839 MODULE_VERSION(ipdivert, 1);
840