xref: /freebsd/sys/netinet/ip_divert.c (revision 0f454b93f8502e85e8d2b26a383e40b5dc50cd27)
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_ipfw.h"
36 #include "opt_sctp.h"
37 #ifndef INET
38 #error "IPDIVERT requires INET."
39 #endif
40 #ifndef IPFIREWALL
41 #error "IPDIVERT requires IPFIREWALL"
42 #endif
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/module.h>
51 #include <sys/kernel.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/protosw.h>
55 #include <sys/rwlock.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sx.h>
60 #include <sys/sysctl.h>
61 #include <sys/systm.h>
62 
63 #include <vm/uma.h>
64 
65 #include <net/if.h>
66 #include <net/netisr.h>
67 #include <net/route.h>
68 #include <net/vnet.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip.h>
75 #include <netinet/ip_divert.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_fw.h>
78 #include <netinet/ipfw/ip_fw_private.h>
79 #ifdef SCTP
80 #include <netinet/sctp_crc32.h>
81 #endif
82 
83 #include <security/mac/mac_framework.h>
84 
85 /*
86  * Divert sockets
87  */
88 
89 /*
90  * Allocate enough space to hold a full IP packet
91  */
92 #define	DIVSNDQ		(65536 + 100)
93 #define	DIVRCVQ		(65536 + 100)
94 
95 /*
96  * Divert sockets work in conjunction with ipfw, see the divert(4)
97  * manpage for features.
98  * Internally, packets selected by ipfw in ip_input() or ip_output(),
99  * and never diverted before, are passed to the input queue of the
100  * divert socket with a given 'divert_port' number (as specified in
101  * the matching ipfw rule), and they are tagged with a 16 bit cookie
102  * (representing the rule number of the matching ipfw rule), which
103  * is passed to process reading from the socket.
104  *
105  * Packets written to the divert socket are again tagged with a cookie
106  * (usually the same as above) and a destination address.
107  * If the destination address is INADDR_ANY then the packet is
108  * treated as outgoing and sent to ip_output(), otherwise it is
109  * treated as incoming and sent to ip_input().
110  * In both cases, the packet is tagged with the cookie.
111  *
112  * On reinjection, processing in ip_input() and ip_output()
113  * will be exactly the same as for the original packet, except that
114  * ipfw processing will start at the rule number after the one
115  * written in the cookie (so, tagging a packet with a cookie of 0
116  * will cause it to be effectively considered as a standard packet).
117  */
118 
119 /* Internal variables. */
120 static VNET_DEFINE(struct inpcbhead, divcb);
121 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
122 
123 #define	V_divcb				VNET(divcb)
124 #define	V_divcbinfo			VNET(divcbinfo)
125 
126 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
127 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
128 
129 static eventhandler_tag ip_divert_event_tag;
130 
131 /*
132  * Initialize divert connection block queue.
133  */
134 static void
135 div_zone_change(void *tag)
136 {
137 
138 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
139 }
140 
141 static int
142 div_inpcb_init(void *mem, int size, int flags)
143 {
144 	struct inpcb *inp = mem;
145 
146 	INP_LOCK_INIT(inp, "inp", "divinp");
147 	return (0);
148 }
149 
150 static void
151 div_inpcb_fini(void *mem, int size)
152 {
153 	struct inpcb *inp = mem;
154 
155 	INP_LOCK_DESTROY(inp);
156 }
157 
158 static void
159 div_init(void)
160 {
161 
162 	INP_INFO_LOCK_INIT(&V_divcbinfo, "div");
163 	LIST_INIT(&V_divcb);
164 	V_divcbinfo.ipi_listhead = &V_divcb;
165 #ifdef VIMAGE
166 	V_divcbinfo.ipi_vnet = curvnet;
167 #endif
168 	/*
169 	 * XXX We don't use the hash list for divert IP, but it's easier
170 	 * to allocate a one entry hash list than it is to check all
171 	 * over the place for hashbase == NULL.
172 	 */
173 	V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask);
174 	V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
175 	    &V_divcbinfo.ipi_porthashmask);
176 	V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
177 	    NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
178 	    UMA_ZONE_NOFREE);
179 	uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
180 }
181 
182 static void
183 div_destroy(void)
184 {
185 
186 	INP_INFO_LOCK_DESTROY(&V_divcbinfo);
187 	uma_zdestroy(V_divcbinfo.ipi_zone);
188 	hashdestroy(V_divcbinfo.ipi_hashbase, M_PCB, V_divcbinfo.ipi_hashmask);
189 	hashdestroy(V_divcbinfo.ipi_porthashbase, M_PCB,
190 	    V_divcbinfo.ipi_porthashmask);
191 }
192 
193 /*
194  * IPPROTO_DIVERT is not in the real IP protocol number space; this
195  * function should never be called.  Just in case, drop any packets.
196  */
197 void
198 div_input(struct mbuf *m, int off)
199 {
200 
201 	KMOD_IPSTAT_INC(ips_noproto);
202 	m_freem(m);
203 }
204 
205 /*
206  * Divert a packet by passing it up to the divert socket at port 'port'.
207  *
208  * Setup generic address and protocol structures for div_input routine,
209  * then pass them along with mbuf chain.
210  */
211 static void
212 divert_packet(struct mbuf *m, int incoming)
213 {
214 	struct ip *ip;
215 	struct inpcb *inp;
216 	struct socket *sa;
217 	u_int16_t nport;
218 	struct sockaddr_in divsrc;
219 	struct m_tag *mtag;
220 
221 	mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
222 	if (mtag == NULL) {
223 		printf("%s: no divert tag\n", __func__);
224 		m_freem(m);
225 		return;
226 	}
227 	/* Assure header */
228 	if (m->m_len < sizeof(struct ip) &&
229 	    (m = m_pullup(m, sizeof(struct ip))) == 0)
230 		return;
231 	ip = mtod(m, struct ip *);
232 
233 	/* Delayed checksums are currently not compatible with divert. */
234 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
235 		ip->ip_len = ntohs(ip->ip_len);
236 		in_delayed_cksum(m);
237 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
238 		ip->ip_len = htons(ip->ip_len);
239 	}
240 #ifdef SCTP
241 	if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
242 		ip->ip_len = ntohs(ip->ip_len);
243 		sctp_delayed_cksum(m);
244 		m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
245 		ip->ip_len = htons(ip->ip_len);
246 	}
247 #endif
248 	/*
249 	 * Record receive interface address, if any.
250 	 * But only for incoming packets.
251 	 */
252 	bzero(&divsrc, sizeof(divsrc));
253 	divsrc.sin_len = sizeof(divsrc);
254 	divsrc.sin_family = AF_INET;
255 	divsrc.sin_port = divert_cookie(mtag);	/* record matching rule */
256 	if (incoming) {
257 		struct ifaddr *ifa;
258 		struct ifnet *ifp;
259 
260 		/* Sanity check */
261 		M_ASSERTPKTHDR(m);
262 
263 		/* Find IP address for receive interface */
264 		ifp = m->m_pkthdr.rcvif;
265 		if_addr_rlock(ifp);
266 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
267 			if (ifa->ifa_addr->sa_family != AF_INET)
268 				continue;
269 			divsrc.sin_addr =
270 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
271 			break;
272 		}
273 		if_addr_runlock(ifp);
274 	}
275 	/*
276 	 * Record the incoming interface name whenever we have one.
277 	 */
278 	if (m->m_pkthdr.rcvif) {
279 		/*
280 		 * Hide the actual interface name in there in the
281 		 * sin_zero array. XXX This needs to be moved to a
282 		 * different sockaddr type for divert, e.g.
283 		 * sockaddr_div with multiple fields like
284 		 * sockaddr_dl. Presently we have only 7 bytes
285 		 * but that will do for now as most interfaces
286 		 * are 4 or less + 2 or less bytes for unit.
287 		 * There is probably a faster way of doing this,
288 		 * possibly taking it from the sockaddr_dl on the iface.
289 		 * This solves the problem of a P2P link and a LAN interface
290 		 * having the same address, which can result in the wrong
291 		 * interface being assigned to the packet when fed back
292 		 * into the divert socket. Theoretically if the daemon saves
293 		 * and re-uses the sockaddr_in as suggested in the man pages,
294 		 * this iface name will come along for the ride.
295 		 * (see div_output for the other half of this.)
296 		 */
297 		strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
298 		    sizeof(divsrc.sin_zero));
299 	}
300 
301 	/* Put packet on socket queue, if any */
302 	sa = NULL;
303 	nport = htons((u_int16_t)divert_info(mtag));
304 	INP_INFO_RLOCK(&V_divcbinfo);
305 	LIST_FOREACH(inp, &V_divcb, inp_list) {
306 		/* XXX why does only one socket match? */
307 		if (inp->inp_lport == nport) {
308 			INP_RLOCK(inp);
309 			sa = inp->inp_socket;
310 			SOCKBUF_LOCK(&sa->so_rcv);
311 			if (sbappendaddr_locked(&sa->so_rcv,
312 			    (struct sockaddr *)&divsrc, m,
313 			    (struct mbuf *)0) == 0) {
314 				SOCKBUF_UNLOCK(&sa->so_rcv);
315 				sa = NULL;	/* force mbuf reclaim below */
316 			} else
317 				sorwakeup_locked(sa);
318 			INP_RUNLOCK(inp);
319 			break;
320 		}
321 	}
322 	INP_INFO_RUNLOCK(&V_divcbinfo);
323 	if (sa == NULL) {
324 		m_freem(m);
325 		KMOD_IPSTAT_INC(ips_noproto);
326 		KMOD_IPSTAT_DEC(ips_delivered);
327         }
328 }
329 
330 /*
331  * Deliver packet back into the IP processing machinery.
332  *
333  * If no address specified, or address is 0.0.0.0, send to ip_output();
334  * otherwise, send to ip_input() and mark as having been received on
335  * the interface with that address.
336  */
337 static int
338 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
339     struct mbuf *control)
340 {
341 	struct m_tag *mtag;
342 	struct divert_tag *dt;
343 	int error = 0;
344 	struct mbuf *options;
345 
346 	/*
347 	 * An mbuf may hasn't come from userland, but we pretend
348 	 * that it has.
349 	 */
350 	m->m_pkthdr.rcvif = NULL;
351 	m->m_nextpkt = NULL;
352 	M_SETFIB(m, so->so_fibnum);
353 
354 	if (control)
355 		m_freem(control);		/* XXX */
356 
357 	if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
358 		mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
359 		    M_NOWAIT | M_ZERO);
360 		if (mtag == NULL) {
361 			error = ENOBUFS;
362 			goto cantsend;
363 		}
364 		dt = (struct divert_tag *)(mtag+1);
365 		m_tag_prepend(m, mtag);
366 	} else
367 		dt = (struct divert_tag *)(mtag+1);
368 
369 	/* Loopback avoidance and state recovery */
370 	if (sin) {
371 		int i;
372 
373 		dt->cookie = sin->sin_port;
374 		/*
375 		 * Find receive interface with the given name, stuffed
376 		 * (if it exists) in the sin_zero[] field.
377 		 * The name is user supplied data so don't trust its size
378 		 * or that it is zero terminated.
379 		 */
380 		for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
381 			;
382 		if ( i > 0 && i < sizeof(sin->sin_zero))
383 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
384 	}
385 
386 	/* Reinject packet into the system as incoming or outgoing */
387 	if (!sin || sin->sin_addr.s_addr == 0) {
388 		struct ip *const ip = mtod(m, struct ip *);
389 		struct inpcb *inp;
390 
391 		dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
392 		INP_INFO_WLOCK(&V_divcbinfo);
393 		inp = sotoinpcb(so);
394 		INP_RLOCK(inp);
395 		/*
396 		 * Don't allow both user specified and setsockopt options,
397 		 * and don't allow packet length sizes that will crash
398 		 */
399 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
400 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
401 			error = EINVAL;
402 			INP_RUNLOCK(inp);
403 			INP_INFO_WUNLOCK(&V_divcbinfo);
404 			m_freem(m);
405 		} else {
406 			/* Convert fields to host order for ip_output() */
407 			ip->ip_len = ntohs(ip->ip_len);
408 			ip->ip_off = ntohs(ip->ip_off);
409 
410 			/* Send packet to output processing */
411 			KMOD_IPSTAT_INC(ips_rawout);		/* XXX */
412 
413 #ifdef MAC
414 			mac_inpcb_create_mbuf(inp, m);
415 #endif
416 			/*
417 			 * Get ready to inject the packet into ip_output().
418 			 * Just in case socket options were specified on the
419 			 * divert socket, we duplicate them.  This is done
420 			 * to avoid having to hold the PCB locks over the call
421 			 * to ip_output(), as doing this results in a number of
422 			 * lock ordering complexities.
423 			 *
424 			 * Note that we set the multicast options argument for
425 			 * ip_output() to NULL since it should be invariant that
426 			 * they are not present.
427 			 */
428 			KASSERT(inp->inp_moptions == NULL,
429 			    ("multicast options set on a divert socket"));
430 			options = NULL;
431 			/*
432 			 * XXXCSJP: It is unclear to me whether or not it makes
433 			 * sense for divert sockets to have options.  However,
434 			 * for now we will duplicate them with the INP locks
435 			 * held so we can use them in ip_output() without
436 			 * requring a reference to the pcb.
437 			 */
438 			if (inp->inp_options != NULL) {
439 				options = m_dup(inp->inp_options, M_DONTWAIT);
440 				if (options == NULL)
441 					error = ENOBUFS;
442 			}
443 			INP_RUNLOCK(inp);
444 			INP_INFO_WUNLOCK(&V_divcbinfo);
445 			if (error == ENOBUFS) {
446 				m_freem(m);
447 				return (error);
448 			}
449 			error = ip_output(m, options, NULL,
450 			    ((so->so_options & SO_DONTROUTE) ?
451 			    IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
452 			    IP_RAWOUTPUT, NULL, NULL);
453 			if (options != NULL)
454 				m_freem(options);
455 		}
456 	} else {
457 		dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
458 		if (m->m_pkthdr.rcvif == NULL) {
459 			/*
460 			 * No luck with the name, check by IP address.
461 			 * Clear the port and the ifname to make sure
462 			 * there are no distractions for ifa_ifwithaddr.
463 			 */
464 			struct	ifaddr *ifa;
465 
466 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
467 			sin->sin_port = 0;
468 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
469 			if (ifa == NULL) {
470 				error = EADDRNOTAVAIL;
471 				goto cantsend;
472 			}
473 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
474 			ifa_free(ifa);
475 		}
476 #ifdef MAC
477 		mac_socket_create_mbuf(so, m);
478 #endif
479 		/* Send packet to input processing via netisr */
480 		netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
481 	}
482 
483 	return error;
484 
485 cantsend:
486 	m_freem(m);
487 	return error;
488 }
489 
490 static int
491 div_attach(struct socket *so, int proto, struct thread *td)
492 {
493 	struct inpcb *inp;
494 	int error;
495 
496 	inp  = sotoinpcb(so);
497 	KASSERT(inp == NULL, ("div_attach: inp != NULL"));
498 	if (td != NULL) {
499 		error = priv_check(td, PRIV_NETINET_DIVERT);
500 		if (error)
501 			return (error);
502 	}
503 	error = soreserve(so, div_sendspace, div_recvspace);
504 	if (error)
505 		return error;
506 	INP_INFO_WLOCK(&V_divcbinfo);
507 	error = in_pcballoc(so, &V_divcbinfo);
508 	if (error) {
509 		INP_INFO_WUNLOCK(&V_divcbinfo);
510 		return error;
511 	}
512 	inp = (struct inpcb *)so->so_pcb;
513 	INP_INFO_WUNLOCK(&V_divcbinfo);
514 	inp->inp_ip_p = proto;
515 	inp->inp_vflag |= INP_IPV4;
516 	inp->inp_flags |= INP_HDRINCL;
517 	INP_WUNLOCK(inp);
518 	return 0;
519 }
520 
521 static void
522 div_detach(struct socket *so)
523 {
524 	struct inpcb *inp;
525 
526 	inp = sotoinpcb(so);
527 	KASSERT(inp != NULL, ("div_detach: inp == NULL"));
528 	INP_INFO_WLOCK(&V_divcbinfo);
529 	INP_WLOCK(inp);
530 	in_pcbdetach(inp);
531 	in_pcbfree(inp);
532 	INP_INFO_WUNLOCK(&V_divcbinfo);
533 }
534 
535 static int
536 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
537 {
538 	struct inpcb *inp;
539 	int error;
540 
541 	inp = sotoinpcb(so);
542 	KASSERT(inp != NULL, ("div_bind: inp == NULL"));
543 	/* in_pcbbind assumes that nam is a sockaddr_in
544 	 * and in_pcbbind requires a valid address. Since divert
545 	 * sockets don't we need to make sure the address is
546 	 * filled in properly.
547 	 * XXX -- divert should not be abusing in_pcbind
548 	 * and should probably have its own family.
549 	 */
550 	if (nam->sa_family != AF_INET)
551 		return EAFNOSUPPORT;
552 	((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
553 	INP_INFO_WLOCK(&V_divcbinfo);
554 	INP_WLOCK(inp);
555 	error = in_pcbbind(inp, nam, td->td_ucred);
556 	INP_WUNLOCK(inp);
557 	INP_INFO_WUNLOCK(&V_divcbinfo);
558 	return error;
559 }
560 
561 static int
562 div_shutdown(struct socket *so)
563 {
564 	struct inpcb *inp;
565 
566 	inp = sotoinpcb(so);
567 	KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
568 	INP_WLOCK(inp);
569 	socantsendmore(so);
570 	INP_WUNLOCK(inp);
571 	return 0;
572 }
573 
574 static int
575 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
576     struct mbuf *control, struct thread *td)
577 {
578 
579 	/* Packet must have a header (but that's about it) */
580 	if (m->m_len < sizeof (struct ip) &&
581 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
582 		KMOD_IPSTAT_INC(ips_toosmall);
583 		m_freem(m);
584 		return EINVAL;
585 	}
586 
587 	/* Send packet */
588 	return div_output(so, m, (struct sockaddr_in *)nam, control);
589 }
590 
591 void
592 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
593 {
594         struct in_addr faddr;
595 
596 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
597 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
598         	return;
599 	if (PRC_IS_REDIRECT(cmd))
600 		return;
601 }
602 
603 static int
604 div_pcblist(SYSCTL_HANDLER_ARGS)
605 {
606 	int error, i, n;
607 	struct inpcb *inp, **inp_list;
608 	inp_gen_t gencnt;
609 	struct xinpgen xig;
610 
611 	/*
612 	 * The process of preparing the TCB list is too time-consuming and
613 	 * resource-intensive to repeat twice on every request.
614 	 */
615 	if (req->oldptr == 0) {
616 		n = V_divcbinfo.ipi_count;
617 		req->oldidx = 2 * (sizeof xig)
618 			+ (n + n/8) * sizeof(struct xinpcb);
619 		return 0;
620 	}
621 
622 	if (req->newptr != 0)
623 		return EPERM;
624 
625 	/*
626 	 * OK, now we're committed to doing something.
627 	 */
628 	INP_INFO_RLOCK(&V_divcbinfo);
629 	gencnt = V_divcbinfo.ipi_gencnt;
630 	n = V_divcbinfo.ipi_count;
631 	INP_INFO_RUNLOCK(&V_divcbinfo);
632 
633 	error = sysctl_wire_old_buffer(req,
634 	    2 * sizeof(xig) + n*sizeof(struct xinpcb));
635 	if (error != 0)
636 		return (error);
637 
638 	xig.xig_len = sizeof xig;
639 	xig.xig_count = n;
640 	xig.xig_gen = gencnt;
641 	xig.xig_sogen = so_gencnt;
642 	error = SYSCTL_OUT(req, &xig, sizeof xig);
643 	if (error)
644 		return error;
645 
646 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
647 	if (inp_list == 0)
648 		return ENOMEM;
649 
650 	INP_INFO_RLOCK(&V_divcbinfo);
651 	for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
652 	     inp = LIST_NEXT(inp, inp_list)) {
653 		INP_RLOCK(inp);
654 		if (inp->inp_gencnt <= gencnt &&
655 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
656 			inp_list[i++] = inp;
657 		INP_RUNLOCK(inp);
658 	}
659 	INP_INFO_RUNLOCK(&V_divcbinfo);
660 	n = i;
661 
662 	error = 0;
663 	for (i = 0; i < n; i++) {
664 		inp = inp_list[i];
665 		INP_RLOCK(inp);
666 		if (inp->inp_gencnt <= gencnt) {
667 			struct xinpcb xi;
668 			bzero(&xi, sizeof(xi));
669 			xi.xi_len = sizeof xi;
670 			/* XXX should avoid extra copy */
671 			bcopy(inp, &xi.xi_inp, sizeof *inp);
672 			if (inp->inp_socket)
673 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
674 			INP_RUNLOCK(inp);
675 			error = SYSCTL_OUT(req, &xi, sizeof xi);
676 		} else
677 			INP_RUNLOCK(inp);
678 	}
679 	if (!error) {
680 		/*
681 		 * Give the user an updated idea of our state.
682 		 * If the generation differs from what we told
683 		 * her before, she knows that something happened
684 		 * while we were processing this request, and it
685 		 * might be necessary to retry.
686 		 */
687 		INP_INFO_RLOCK(&V_divcbinfo);
688 		xig.xig_gen = V_divcbinfo.ipi_gencnt;
689 		xig.xig_sogen = so_gencnt;
690 		xig.xig_count = V_divcbinfo.ipi_count;
691 		INP_INFO_RUNLOCK(&V_divcbinfo);
692 		error = SYSCTL_OUT(req, &xig, sizeof xig);
693 	}
694 	free(inp_list, M_TEMP);
695 	return error;
696 }
697 
698 #ifdef SYSCTL_NODE
699 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
700 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
701 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
702 #endif
703 
704 struct pr_usrreqs div_usrreqs = {
705 	.pru_attach =		div_attach,
706 	.pru_bind =		div_bind,
707 	.pru_control =		in_control,
708 	.pru_detach =		div_detach,
709 	.pru_peeraddr =		in_getpeeraddr,
710 	.pru_send =		div_send,
711 	.pru_shutdown =		div_shutdown,
712 	.pru_sockaddr =		in_getsockaddr,
713 	.pru_sosetlabel =	in_pcbsosetlabel
714 };
715 
716 struct protosw div_protosw = {
717 	.pr_type =		SOCK_RAW,
718 	.pr_protocol =		IPPROTO_DIVERT,
719 	.pr_flags =		PR_ATOMIC|PR_ADDR,
720 	.pr_input =		div_input,
721 	.pr_ctlinput =		div_ctlinput,
722 	.pr_ctloutput =		ip_ctloutput,
723 	.pr_init =		div_init,
724 #ifdef VIMAGE
725 	.pr_destroy =		div_destroy,
726 #endif
727 	.pr_usrreqs =		&div_usrreqs
728 };
729 
730 static int
731 div_modevent(module_t mod, int type, void *unused)
732 {
733 	int err = 0;
734 #ifndef VIMAGE
735 	int n;
736 #endif
737 
738 	switch (type) {
739 	case MOD_LOAD:
740 		/*
741 		 * Protocol will be initialized by pf_proto_register().
742 		 * We don't have to register ip_protox because we are not
743 		 * a true IP protocol that goes over the wire.
744 		 */
745 		err = pf_proto_register(PF_INET, &div_protosw);
746 		if (err != 0)
747 			return (err);
748 		ip_divert_ptr = divert_packet;
749 		ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
750 		    div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
751 		break;
752 	case MOD_QUIESCE:
753 		/*
754 		 * IPDIVERT may normally not be unloaded because of the
755 		 * potential race conditions.  Tell kldunload we can't be
756 		 * unloaded unless the unload is forced.
757 		 */
758 		err = EPERM;
759 		break;
760 	case MOD_UNLOAD:
761 #ifdef VIMAGE
762 		err = EPERM;
763 		break;
764 #else
765 		/*
766 		 * Forced unload.
767 		 *
768 		 * Module ipdivert can only be unloaded if no sockets are
769 		 * connected.  Maybe this can be changed later to forcefully
770 		 * disconnect any open sockets.
771 		 *
772 		 * XXXRW: Note that there is a slight race here, as a new
773 		 * socket open request could be spinning on the lock and then
774 		 * we destroy the lock.
775 		 */
776 		INP_INFO_WLOCK(&V_divcbinfo);
777 		n = V_divcbinfo.ipi_count;
778 		if (n != 0) {
779 			err = EBUSY;
780 			INP_INFO_WUNLOCK(&V_divcbinfo);
781 			break;
782 		}
783 		ip_divert_ptr = NULL;
784 		err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
785 		INP_INFO_WUNLOCK(&V_divcbinfo);
786 		div_destroy();
787 		EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
788 		break;
789 #endif /* !VIMAGE */
790 	default:
791 		err = EOPNOTSUPP;
792 		break;
793 	}
794 	return err;
795 }
796 
797 static moduledata_t ipdivertmod = {
798         "ipdivert",
799         div_modevent,
800         0
801 };
802 
803 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
804 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
805 MODULE_VERSION(ipdivert, 1);
806