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 * $FreeBSD: head/sys/netinet/ip_divert.c 157927 2006-04-21 09:25:40Z ps $
30 */
31
32#if !defined(KLD_MODULE)
33#include "opt_inet.h"
34#include "opt_ipfw.h"
35#include "opt_mac.h"
36#ifndef INET
37#error "IPDIVERT requires INET."
38#endif
39#ifndef IPFIREWALL
40#error "IPDIVERT requires IPFIREWALL"
41#endif
42#endif
43
44#include <sys/param.h>
45#include <sys/kernel.h>
46#include <sys/lock.h>
47#include <sys/malloc.h>
48#include <sys/mac.h>
49#include <sys/mbuf.h>
50#include <sys/module.h>
51#include <sys/kernel.h>
52#include <sys/proc.h>
53#include <sys/protosw.h>
54#include <sys/signalvar.h>
55#include <sys/socket.h>
56#include <sys/socketvar.h>
57#include <sys/sx.h>
58#include <sys/sysctl.h>
59#include <sys/systm.h>
60
61#include <vm/uma.h>
62
63#include <net/if.h>
64#include <net/route.h>
65
66#include <netinet/in.h>
67#include <netinet/in_pcb.h>
68#include <netinet/in_systm.h>
69#include <netinet/in_var.h>
70#include <netinet/ip.h>
71#include <netinet/ip_divert.h>
72#include <netinet/ip_var.h>
73#include <netinet/ip_fw.h>
74
75/*
76 * Divert sockets
77 */
78
79/*
80 * Allocate enough space to hold a full IP packet
81 */
82#define DIVSNDQ (65536 + 100)
83#define DIVRCVQ (65536 + 100)
84
85/*
86 * Divert sockets work in conjunction with ipfw, see the divert(4)
87 * manpage for features.
88 * Internally, packets selected by ipfw in ip_input() or ip_output(),
89 * and never diverted before, are passed to the input queue of the
90 * divert socket with a given 'divert_port' number (as specified in
91 * the matching ipfw rule), and they are tagged with a 16 bit cookie
92 * (representing the rule number of the matching ipfw rule), which
93 * is passed to process reading from the socket.
94 *
95 * Packets written to the divert socket are again tagged with a cookie
96 * (usually the same as above) and a destination address.
97 * If the destination address is INADDR_ANY then the packet is
98 * treated as outgoing and sent to ip_output(), otherwise it is
99 * treated as incoming and sent to ip_input().
100 * In both cases, the packet is tagged with the cookie.
101 *
102 * On reinjection, processing in ip_input() and ip_output()
103 * will be exactly the same as for the original packet, except that
104 * ipfw processing will start at the rule number after the one
105 * written in the cookie (so, tagging a packet with a cookie of 0
106 * will cause it to be effectively considered as a standard packet).
107 */
108
109/* Internal variables. */
110static struct inpcbhead divcb;
111static struct inpcbinfo divcbinfo;
112
113static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
114static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
115
116/*
117 * Initialize divert connection block queue.
118 */
119static void
120div_zone_change(void *tag)
121{
122
123 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
124}
125
126void
127div_init(void)
128{
129 INP_INFO_LOCK_INIT(&divcbinfo, "div");
130 LIST_INIT(&divcb);
131 divcbinfo.listhead = &divcb;
132 /*
133 * XXX We don't use the hash list for divert IP, but it's easier
134 * to allocate a one entry hash list than it is to check all
135 * over the place for hashbase == NULL.
136 */
137 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
138 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
139 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
140 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
141 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
142 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
143 NULL, EVENTHANDLER_PRI_ANY);
144}
145
146/*
147 * IPPROTO_DIVERT is not in the real IP protocol number space; this
148 * function should never be called. Just in case, drop any packets.
149 */
150void
151div_input(struct mbuf *m, int off)
152{
153 ipstat.ips_noproto++;
154 m_freem(m);
155}
156
157/*
158 * Divert a packet by passing it up to the divert socket at port 'port'.
159 *
160 * Setup generic address and protocol structures for div_input routine,
161 * then pass them along with mbuf chain.
162 */
163static void
164divert_packet(struct mbuf *m, int incoming)
165{
166 struct ip *ip;
167 struct inpcb *inp;
168 struct socket *sa;
169 u_int16_t nport;
170 struct sockaddr_in divsrc;
171 struct m_tag *mtag;
172
173 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
174 if (mtag == NULL) {
175 printf("%s: no divert tag\n", __func__);
176 m_freem(m);
177 return;
178 }
179 /* Assure header */
180 if (m->m_len < sizeof(struct ip) &&
181 (m = m_pullup(m, sizeof(struct ip))) == 0)
182 return;
183 ip = mtod(m, struct ip *);
184
185 /* Delayed checksums are currently not compatible with divert. */
186 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
187 ip->ip_len = ntohs(ip->ip_len);
188 in_delayed_cksum(m);
189 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
190 ip->ip_len = htons(ip->ip_len);
191 }
192
193 /*
194 * Record receive interface address, if any.
195 * But only for incoming packets.
196 */
197 bzero(&divsrc, sizeof(divsrc));
198 divsrc.sin_len = sizeof(divsrc);
199 divsrc.sin_family = AF_INET;
200 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
201 if (incoming) {
202 struct ifaddr *ifa;
203
204 /* Sanity check */
205 M_ASSERTPKTHDR(m);
206
207 /* Find IP address for receive interface */
208 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
209 if (ifa->ifa_addr == NULL)
210 continue;
211 if (ifa->ifa_addr->sa_family != AF_INET)
212 continue;
213 divsrc.sin_addr =
214 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
215 break;
216 }
217 }
218 /*
219 * Record the incoming interface name whenever we have one.
220 */
221 if (m->m_pkthdr.rcvif) {
222 /*
223 * Hide the actual interface name in there in the
224 * sin_zero array. XXX This needs to be moved to a
225 * different sockaddr type for divert, e.g.
226 * sockaddr_div with multiple fields like
227 * sockaddr_dl. Presently we have only 7 bytes
228 * but that will do for now as most interfaces
229 * are 4 or less + 2 or less bytes for unit.
230 * There is probably a faster way of doing this,
231 * possibly taking it from the sockaddr_dl on the iface.
232 * This solves the problem of a P2P link and a LAN interface
233 * having the same address, which can result in the wrong
234 * interface being assigned to the packet when fed back
235 * into the divert socket. Theoretically if the daemon saves
236 * and re-uses the sockaddr_in as suggested in the man pages,
237 * this iface name will come along for the ride.
238 * (see div_output for the other half of this.)
239 */
240 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
241 sizeof(divsrc.sin_zero));
242 }
243
244 /* Put packet on socket queue, if any */
245 sa = NULL;
246 nport = htons((u_int16_t)divert_info(mtag));
247 INP_INFO_RLOCK(&divcbinfo);
248 LIST_FOREACH(inp, &divcb, inp_list) {
249 INP_LOCK(inp);
250 /* XXX why does only one socket match? */
251 if (inp->inp_lport == nport) {
252 sa = inp->inp_socket;
253 SOCKBUF_LOCK(&sa->so_rcv);
254 if (sbappendaddr_locked(&sa->so_rcv,
255 (struct sockaddr *)&divsrc, m,
256 (struct mbuf *)0) == 0) {
257 SOCKBUF_UNLOCK(&sa->so_rcv);
258 sa = NULL; /* force mbuf reclaim below */
259 } else
260 sorwakeup_locked(sa);
261 INP_UNLOCK(inp);
262 break;
263 }
264 INP_UNLOCK(inp);
265 }
266 INP_INFO_RUNLOCK(&divcbinfo);
267 if (sa == NULL) {
268 m_freem(m);
269 ipstat.ips_noproto++;
270 ipstat.ips_delivered--;
271 }
272}
273
274/*
275 * Deliver packet back into the IP processing machinery.
276 *
277 * If no address specified, or address is 0.0.0.0, send to ip_output();
278 * otherwise, send to ip_input() and mark as having been received on
279 * the interface with that address.
280 */
281static int
282div_output(struct socket *so, struct mbuf *m,
283 struct sockaddr_in *sin, struct mbuf *control)
284{
285 struct m_tag *mtag;
286 struct divert_tag *dt;
287 int error = 0;
288
289 /*
290 * An mbuf may hasn't come from userland, but we pretend
291 * that it has.
292 */
293 m->m_pkthdr.rcvif = NULL;
294 m->m_nextpkt = NULL;
295
296 if (control)
297 m_freem(control); /* XXX */
298
299 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
300 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
301 M_NOWAIT | M_ZERO);
302 if (mtag == NULL) {
303 error = ENOBUFS;
304 goto cantsend;
305 }
306 dt = (struct divert_tag *)(mtag+1);
307 m_tag_prepend(m, mtag);
308 } else
309 dt = (struct divert_tag *)(mtag+1);
310
311 /* Loopback avoidance and state recovery */
312 if (sin) {
313 int i;
314
315 dt->cookie = sin->sin_port;
316 /*
317 * Find receive interface with the given name, stuffed
318 * (if it exists) in the sin_zero[] field.
319 * The name is user supplied data so don't trust its size
320 * or that it is zero terminated.
321 */
322 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
323 ;
324 if ( i > 0 && i < sizeof(sin->sin_zero))
325 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
326 }
327
328 /* Reinject packet into the system as incoming or outgoing */
329 if (!sin || sin->sin_addr.s_addr == 0) {
330 struct ip *const ip = mtod(m, struct ip *);
331 struct inpcb *inp;
332
333 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
334 INP_INFO_WLOCK(&divcbinfo);
335 inp = sotoinpcb(so);
336 INP_LOCK(inp);
337 /*
338 * Don't allow both user specified and setsockopt options,
339 * and don't allow packet length sizes that will crash
340 */
341 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
342 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
343 error = EINVAL;
344 m_freem(m);
345 } else {
346 /* Convert fields to host order for ip_output() */
347 ip->ip_len = ntohs(ip->ip_len);
348 ip->ip_off = ntohs(ip->ip_off);
349
350 /* Send packet to output processing */
351 ipstat.ips_rawout++; /* XXX */
352
353#ifdef MAC
354 mac_create_mbuf_from_inpcb(inp, m);
355#endif
356 error = ip_output(m,
357 inp->inp_options, NULL,
358 ((so->so_options & SO_DONTROUTE) ?
359 IP_ROUTETOIF : 0) |
360 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
361 inp->inp_moptions, NULL);
362 }
363 INP_UNLOCK(inp);
364 INP_INFO_WUNLOCK(&divcbinfo);
365 } else {
366 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
367 if (m->m_pkthdr.rcvif == NULL) {
368 /*
369 * No luck with the name, check by IP address.
370 * Clear the port and the ifname to make sure
371 * there are no distractions for ifa_ifwithaddr.
372 */
373 struct ifaddr *ifa;
374
375 bzero(sin->sin_zero, sizeof(sin->sin_zero));
376 sin->sin_port = 0;
377 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
378 if (ifa == NULL) {
379 error = EADDRNOTAVAIL;
380 goto cantsend;
381 }
382 m->m_pkthdr.rcvif = ifa->ifa_ifp;
383 }
384#ifdef MAC
385 SOCK_LOCK(so);
386 mac_create_mbuf_from_socket(so, m);
387 SOCK_UNLOCK(so);
388#endif
389 /* Send packet to input processing */
390 ip_input(m);
391 }
392
393 return error;
394
395cantsend:
396 m_freem(m);
397 return error;
398}
399
400static int
401div_attach(struct socket *so, int proto, struct thread *td)
402{
403 struct inpcb *inp;
404 int error;
405
406 inp = sotoinpcb(so);
407 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
408 if (td && (error = suser(td)) != 0)
409 return error;
410 error = soreserve(so, div_sendspace, div_recvspace);
411 if (error)
412 return error;
413 INP_INFO_WLOCK(&divcbinfo);
414 error = in_pcballoc(so, &divcbinfo, "divinp");
415 if (error) {
416 INP_INFO_WUNLOCK(&divcbinfo);
417 return error;
418 }
419 inp = (struct inpcb *)so->so_pcb;
420 INP_LOCK(inp);
421 INP_INFO_WUNLOCK(&divcbinfo);
422 inp->inp_ip_p = proto;
423 inp->inp_vflag |= INP_IPV4;
424 inp->inp_flags |= INP_HDRINCL;
425 INP_UNLOCK(inp);
426 return 0;
427}
428
429static void
430div_detach(struct socket *so)
431{
432 struct inpcb *inp;
433
434 inp = sotoinpcb(so);
435 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
436 INP_INFO_WLOCK(&divcbinfo);
437 INP_LOCK(inp);
438 in_pcbdetach(inp);
439 in_pcbfree(inp);
440 INP_INFO_WUNLOCK(&divcbinfo);
441}
442
443static int
444div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
445{
446 struct inpcb *inp;
447 int error;
448
449 inp = sotoinpcb(so);
450 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
451 /* in_pcbbind assumes that nam is a sockaddr_in
452 * and in_pcbbind requires a valid address. Since divert
453 * sockets don't we need to make sure the address is
454 * filled in properly.
455 * XXX -- divert should not be abusing in_pcbind
456 * and should probably have its own family.
457 */
458 if (nam->sa_family != AF_INET)
459 return EAFNOSUPPORT;
460 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
461 INP_INFO_WLOCK(&divcbinfo);
462 INP_LOCK(inp);
463 error = in_pcbbind(inp, nam, td->td_ucred);
464 INP_UNLOCK(inp);
465 INP_INFO_WUNLOCK(&divcbinfo);
466 return error;
467}
468
469static int
470div_shutdown(struct socket *so)
471{
472 struct inpcb *inp;
473
474 inp = sotoinpcb(so);
475 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
476 INP_LOCK(inp);
477 socantsendmore(so);
478 INP_UNLOCK(inp);
479 return 0;
480}
481
482static int
483div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
484 struct mbuf *control, struct thread *td)
485{
486 /* Packet must have a header (but that's about it) */
487 if (m->m_len < sizeof (struct ip) &&
488 (m = m_pullup(m, sizeof (struct ip))) == 0) {
489 ipstat.ips_toosmall++;
490 m_freem(m);
491 return EINVAL;
492 }
493
494 /* Send packet */
495 return div_output(so, m, (struct sockaddr_in *)nam, control);
496}
497
498void
499div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
500{
501 struct in_addr faddr;
502
503 faddr = ((struct sockaddr_in *)sa)->sin_addr;
504 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
505 return;
506 if (PRC_IS_REDIRECT(cmd))
507 return;
508}
509
510static int
511div_pcblist(SYSCTL_HANDLER_ARGS)
512{
513 int error, i, n;
514 struct inpcb *inp, **inp_list;
515 inp_gen_t gencnt;
516 struct xinpgen xig;
517
518 /*
519 * The process of preparing the TCB list is too time-consuming and
520 * resource-intensive to repeat twice on every request.
521 */
522 if (req->oldptr == 0) {
523 n = divcbinfo.ipi_count;
524 req->oldidx = 2 * (sizeof xig)
525 + (n + n/8) * sizeof(struct xinpcb);
526 return 0;
527 }
528
529 if (req->newptr != 0)
530 return EPERM;
531
532 /*
533 * OK, now we're committed to doing something.
534 */
535 INP_INFO_RLOCK(&divcbinfo);
536 gencnt = divcbinfo.ipi_gencnt;
537 n = divcbinfo.ipi_count;
538 INP_INFO_RUNLOCK(&divcbinfo);
539
540 error = sysctl_wire_old_buffer(req,
541 2 * sizeof(xig) + n*sizeof(struct xinpcb));
542 if (error != 0)
543 return (error);
544
545 xig.xig_len = sizeof xig;
546 xig.xig_count = n;
547 xig.xig_gen = gencnt;
548 xig.xig_sogen = so_gencnt;
549 error = SYSCTL_OUT(req, &xig, sizeof xig);
550 if (error)
551 return error;
552
553 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
554 if (inp_list == 0)
555 return ENOMEM;
556
557 INP_INFO_RLOCK(&divcbinfo);
558 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
559 inp = LIST_NEXT(inp, inp_list)) {
560 INP_LOCK(inp);
561 if (inp->inp_gencnt <= gencnt &&
562 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
563 inp_list[i++] = inp;
564 INP_UNLOCK(inp);
565 }
566 INP_INFO_RUNLOCK(&divcbinfo);
567 n = i;
568
569 error = 0;
570 for (i = 0; i < n; i++) {
571 inp = inp_list[i];
572 if (inp->inp_gencnt <= gencnt) {
573 struct xinpcb xi;
574 bzero(&xi, sizeof(xi));
575 xi.xi_len = sizeof xi;
576 /* XXX should avoid extra copy */
577 bcopy(inp, &xi.xi_inp, sizeof *inp);
578 if (inp->inp_socket)
579 sotoxsocket(inp->inp_socket, &xi.xi_socket);
580 error = SYSCTL_OUT(req, &xi, sizeof xi);
581 }
582 }
583 if (!error) {
584 /*
585 * Give the user an updated idea of our state.
586 * If the generation differs from what we told
587 * her before, she knows that something happened
588 * while we were processing this request, and it
589 * might be necessary to retry.
590 */
591 INP_INFO_RLOCK(&divcbinfo);
592 xig.xig_gen = divcbinfo.ipi_gencnt;
593 xig.xig_sogen = so_gencnt;
594 xig.xig_count = divcbinfo.ipi_count;
595 INP_INFO_RUNLOCK(&divcbinfo);
596 error = SYSCTL_OUT(req, &xig, sizeof xig);
597 }
598 free(inp_list, M_TEMP);
599 return error;
600}
601
602/*
603 * This is the wrapper function for in_setsockaddr. We just pass down
604 * the pcbinfo for in_setpeeraddr to lock.
605 */
606static int
607div_sockaddr(struct socket *so, struct sockaddr **nam)
608{
609 return (in_setsockaddr(so, nam, &divcbinfo));
610}
611
612/*
613 * This is the wrapper function for in_setpeeraddr. We just pass down
614 * the pcbinfo for in_setpeeraddr to lock.
615 */
616static int
617div_peeraddr(struct socket *so, struct sockaddr **nam)
618{
619 return (in_setpeeraddr(so, nam, &divcbinfo));
620}
621
622#ifdef SYSCTL_NODE
623SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
624SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
625 div_pcblist, "S,xinpcb", "List of active divert sockets");
626#endif
627
628struct pr_usrreqs div_usrreqs = {
629 .pru_attach = div_attach,
630 .pru_bind = div_bind,
631 .pru_control = in_control,
632 .pru_detach = div_detach,
633 .pru_peeraddr = div_peeraddr,
634 .pru_send = div_send,
635 .pru_shutdown = div_shutdown,
636 .pru_sockaddr = div_sockaddr,
637 .pru_sosetlabel = in_pcbsosetlabel
638};
639
640struct protosw div_protosw = {
641 .pr_type = SOCK_RAW,
642 .pr_protocol = IPPROTO_DIVERT,
643 .pr_flags = PR_ATOMIC|PR_ADDR,
644 .pr_input = div_input,
645 .pr_ctlinput = div_ctlinput,
646 .pr_ctloutput = ip_ctloutput,
647 .pr_init = div_init,
648 .pr_usrreqs = &div_usrreqs
649};
650
651static int
652div_modevent(module_t mod, int type, void *unused)
653{
654 int err = 0;
655 int n;
656
657 switch (type) {
658 case MOD_LOAD:
659 /*
660 * Protocol will be initialized by pf_proto_register().
661 * We don't have to register ip_protox because we are not
662 * a true IP protocol that goes over the wire.
663 */
664 err = pf_proto_register(PF_INET, &div_protosw);
665 ip_divert_ptr = divert_packet;
666 break;
667 case MOD_QUIESCE:
668 /*
669 * IPDIVERT may normally not be unloaded because of the
670 * potential race conditions. Tell kldunload we can't be
671 * unloaded unless the unload is forced.
672 */
673 err = EPERM;
674 break;
675 case MOD_UNLOAD:
676 /*
677 * Forced unload.
678 *
679 * Module ipdivert can only be unloaded if no sockets are
680 * connected. Maybe this can be changed later to forcefully
681 * disconnect any open sockets.
682 *
683 * XXXRW: Note that there is a slight race here, as a new
684 * socket open request could be spinning on the lock and then
685 * we destroy the lock.
686 */
687 INP_INFO_WLOCK(&divcbinfo);
688 n = divcbinfo.ipi_count;
689 if (n != 0) {
690 err = EBUSY;
691 INP_INFO_WUNLOCK(&divcbinfo);
692 break;
693 }
694 ip_divert_ptr = NULL;
695 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
696 INP_INFO_WUNLOCK(&divcbinfo);
697 INP_INFO_LOCK_DESTROY(&divcbinfo);
698 uma_zdestroy(divcbinfo.ipi_zone);
699 break;
700 default:
701 err = EOPNOTSUPP;
702 break;
703 }
704 return err;
705}
706
707static moduledata_t ipdivertmod = {
708 "ipdivert",
709 div_modevent,
710 0
711};
712
713DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
714MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
715MODULE_VERSION(ipdivert, 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 * $FreeBSD: head/sys/netinet/ip_divert.c 157927 2006-04-21 09:25:40Z ps $
30 */
31
32#if !defined(KLD_MODULE)
33#include "opt_inet.h"
34#include "opt_ipfw.h"
35#include "opt_mac.h"
36#ifndef INET
37#error "IPDIVERT requires INET."
38#endif
39#ifndef IPFIREWALL
40#error "IPDIVERT requires IPFIREWALL"
41#endif
42#endif
43
44#include <sys/param.h>
45#include <sys/kernel.h>
46#include <sys/lock.h>
47#include <sys/malloc.h>
48#include <sys/mac.h>
49#include <sys/mbuf.h>
50#include <sys/module.h>
51#include <sys/kernel.h>
52#include <sys/proc.h>
53#include <sys/protosw.h>
54#include <sys/signalvar.h>
55#include <sys/socket.h>
56#include <sys/socketvar.h>
57#include <sys/sx.h>
58#include <sys/sysctl.h>
59#include <sys/systm.h>
60
61#include <vm/uma.h>
62
63#include <net/if.h>
64#include <net/route.h>
65
66#include <netinet/in.h>
67#include <netinet/in_pcb.h>
68#include <netinet/in_systm.h>
69#include <netinet/in_var.h>
70#include <netinet/ip.h>
71#include <netinet/ip_divert.h>
72#include <netinet/ip_var.h>
73#include <netinet/ip_fw.h>
74
75/*
76 * Divert sockets
77 */
78
79/*
80 * Allocate enough space to hold a full IP packet
81 */
82#define DIVSNDQ (65536 + 100)
83#define DIVRCVQ (65536 + 100)
84
85/*
86 * Divert sockets work in conjunction with ipfw, see the divert(4)
87 * manpage for features.
88 * Internally, packets selected by ipfw in ip_input() or ip_output(),
89 * and never diverted before, are passed to the input queue of the
90 * divert socket with a given 'divert_port' number (as specified in
91 * the matching ipfw rule), and they are tagged with a 16 bit cookie
92 * (representing the rule number of the matching ipfw rule), which
93 * is passed to process reading from the socket.
94 *
95 * Packets written to the divert socket are again tagged with a cookie
96 * (usually the same as above) and a destination address.
97 * If the destination address is INADDR_ANY then the packet is
98 * treated as outgoing and sent to ip_output(), otherwise it is
99 * treated as incoming and sent to ip_input().
100 * In both cases, the packet is tagged with the cookie.
101 *
102 * On reinjection, processing in ip_input() and ip_output()
103 * will be exactly the same as for the original packet, except that
104 * ipfw processing will start at the rule number after the one
105 * written in the cookie (so, tagging a packet with a cookie of 0
106 * will cause it to be effectively considered as a standard packet).
107 */
108
109/* Internal variables. */
110static struct inpcbhead divcb;
111static struct inpcbinfo divcbinfo;
112
113static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
114static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
115
116/*
117 * Initialize divert connection block queue.
118 */
119static void
120div_zone_change(void *tag)
121{
122
123 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
124}
125
126void
127div_init(void)
128{
129 INP_INFO_LOCK_INIT(&divcbinfo, "div");
130 LIST_INIT(&divcb);
131 divcbinfo.listhead = &divcb;
132 /*
133 * XXX We don't use the hash list for divert IP, but it's easier
134 * to allocate a one entry hash list than it is to check all
135 * over the place for hashbase == NULL.
136 */
137 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
138 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
139 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
140 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
141 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
142 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
143 NULL, EVENTHANDLER_PRI_ANY);
144}
145
146/*
147 * IPPROTO_DIVERT is not in the real IP protocol number space; this
148 * function should never be called. Just in case, drop any packets.
149 */
150void
151div_input(struct mbuf *m, int off)
152{
153 ipstat.ips_noproto++;
154 m_freem(m);
155}
156
157/*
158 * Divert a packet by passing it up to the divert socket at port 'port'.
159 *
160 * Setup generic address and protocol structures for div_input routine,
161 * then pass them along with mbuf chain.
162 */
163static void
164divert_packet(struct mbuf *m, int incoming)
165{
166 struct ip *ip;
167 struct inpcb *inp;
168 struct socket *sa;
169 u_int16_t nport;
170 struct sockaddr_in divsrc;
171 struct m_tag *mtag;
172
173 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
174 if (mtag == NULL) {
175 printf("%s: no divert tag\n", __func__);
176 m_freem(m);
177 return;
178 }
179 /* Assure header */
180 if (m->m_len < sizeof(struct ip) &&
181 (m = m_pullup(m, sizeof(struct ip))) == 0)
182 return;
183 ip = mtod(m, struct ip *);
184
185 /* Delayed checksums are currently not compatible with divert. */
186 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
187 ip->ip_len = ntohs(ip->ip_len);
188 in_delayed_cksum(m);
189 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
190 ip->ip_len = htons(ip->ip_len);
191 }
192
193 /*
194 * Record receive interface address, if any.
195 * But only for incoming packets.
196 */
197 bzero(&divsrc, sizeof(divsrc));
198 divsrc.sin_len = sizeof(divsrc);
199 divsrc.sin_family = AF_INET;
200 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
201 if (incoming) {
202 struct ifaddr *ifa;
203
204 /* Sanity check */
205 M_ASSERTPKTHDR(m);
206
207 /* Find IP address for receive interface */
208 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
209 if (ifa->ifa_addr == NULL)
210 continue;
211 if (ifa->ifa_addr->sa_family != AF_INET)
212 continue;
213 divsrc.sin_addr =
214 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
215 break;
216 }
217 }
218 /*
219 * Record the incoming interface name whenever we have one.
220 */
221 if (m->m_pkthdr.rcvif) {
222 /*
223 * Hide the actual interface name in there in the
224 * sin_zero array. XXX This needs to be moved to a
225 * different sockaddr type for divert, e.g.
226 * sockaddr_div with multiple fields like
227 * sockaddr_dl. Presently we have only 7 bytes
228 * but that will do for now as most interfaces
229 * are 4 or less + 2 or less bytes for unit.
230 * There is probably a faster way of doing this,
231 * possibly taking it from the sockaddr_dl on the iface.
232 * This solves the problem of a P2P link and a LAN interface
233 * having the same address, which can result in the wrong
234 * interface being assigned to the packet when fed back
235 * into the divert socket. Theoretically if the daemon saves
236 * and re-uses the sockaddr_in as suggested in the man pages,
237 * this iface name will come along for the ride.
238 * (see div_output for the other half of this.)
239 */
240 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
241 sizeof(divsrc.sin_zero));
242 }
243
244 /* Put packet on socket queue, if any */
245 sa = NULL;
246 nport = htons((u_int16_t)divert_info(mtag));
247 INP_INFO_RLOCK(&divcbinfo);
248 LIST_FOREACH(inp, &divcb, inp_list) {
249 INP_LOCK(inp);
250 /* XXX why does only one socket match? */
251 if (inp->inp_lport == nport) {
252 sa = inp->inp_socket;
253 SOCKBUF_LOCK(&sa->so_rcv);
254 if (sbappendaddr_locked(&sa->so_rcv,
255 (struct sockaddr *)&divsrc, m,
256 (struct mbuf *)0) == 0) {
257 SOCKBUF_UNLOCK(&sa->so_rcv);
258 sa = NULL; /* force mbuf reclaim below */
259 } else
260 sorwakeup_locked(sa);
261 INP_UNLOCK(inp);
262 break;
263 }
264 INP_UNLOCK(inp);
265 }
266 INP_INFO_RUNLOCK(&divcbinfo);
267 if (sa == NULL) {
268 m_freem(m);
269 ipstat.ips_noproto++;
270 ipstat.ips_delivered--;
271 }
272}
273
274/*
275 * Deliver packet back into the IP processing machinery.
276 *
277 * If no address specified, or address is 0.0.0.0, send to ip_output();
278 * otherwise, send to ip_input() and mark as having been received on
279 * the interface with that address.
280 */
281static int
282div_output(struct socket *so, struct mbuf *m,
283 struct sockaddr_in *sin, struct mbuf *control)
284{
285 struct m_tag *mtag;
286 struct divert_tag *dt;
287 int error = 0;
288
289 /*
290 * An mbuf may hasn't come from userland, but we pretend
291 * that it has.
292 */
293 m->m_pkthdr.rcvif = NULL;
294 m->m_nextpkt = NULL;
295
296 if (control)
297 m_freem(control); /* XXX */
298
299 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
300 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
301 M_NOWAIT | M_ZERO);
302 if (mtag == NULL) {
303 error = ENOBUFS;
304 goto cantsend;
305 }
306 dt = (struct divert_tag *)(mtag+1);
307 m_tag_prepend(m, mtag);
308 } else
309 dt = (struct divert_tag *)(mtag+1);
310
311 /* Loopback avoidance and state recovery */
312 if (sin) {
313 int i;
314
315 dt->cookie = sin->sin_port;
316 /*
317 * Find receive interface with the given name, stuffed
318 * (if it exists) in the sin_zero[] field.
319 * The name is user supplied data so don't trust its size
320 * or that it is zero terminated.
321 */
322 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
323 ;
324 if ( i > 0 && i < sizeof(sin->sin_zero))
325 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
326 }
327
328 /* Reinject packet into the system as incoming or outgoing */
329 if (!sin || sin->sin_addr.s_addr == 0) {
330 struct ip *const ip = mtod(m, struct ip *);
331 struct inpcb *inp;
332
333 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
334 INP_INFO_WLOCK(&divcbinfo);
335 inp = sotoinpcb(so);
336 INP_LOCK(inp);
337 /*
338 * Don't allow both user specified and setsockopt options,
339 * and don't allow packet length sizes that will crash
340 */
341 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
342 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
343 error = EINVAL;
344 m_freem(m);
345 } else {
346 /* Convert fields to host order for ip_output() */
347 ip->ip_len = ntohs(ip->ip_len);
348 ip->ip_off = ntohs(ip->ip_off);
349
350 /* Send packet to output processing */
351 ipstat.ips_rawout++; /* XXX */
352
353#ifdef MAC
354 mac_create_mbuf_from_inpcb(inp, m);
355#endif
356 error = ip_output(m,
357 inp->inp_options, NULL,
358 ((so->so_options & SO_DONTROUTE) ?
359 IP_ROUTETOIF : 0) |
360 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
361 inp->inp_moptions, NULL);
362 }
363 INP_UNLOCK(inp);
364 INP_INFO_WUNLOCK(&divcbinfo);
365 } else {
366 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
367 if (m->m_pkthdr.rcvif == NULL) {
368 /*
369 * No luck with the name, check by IP address.
370 * Clear the port and the ifname to make sure
371 * there are no distractions for ifa_ifwithaddr.
372 */
373 struct ifaddr *ifa;
374
375 bzero(sin->sin_zero, sizeof(sin->sin_zero));
376 sin->sin_port = 0;
377 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
378 if (ifa == NULL) {
379 error = EADDRNOTAVAIL;
380 goto cantsend;
381 }
382 m->m_pkthdr.rcvif = ifa->ifa_ifp;
383 }
384#ifdef MAC
385 SOCK_LOCK(so);
386 mac_create_mbuf_from_socket(so, m);
387 SOCK_UNLOCK(so);
388#endif
389 /* Send packet to input processing */
390 ip_input(m);
391 }
392
393 return error;
394
395cantsend:
396 m_freem(m);
397 return error;
398}
399
400static int
401div_attach(struct socket *so, int proto, struct thread *td)
402{
403 struct inpcb *inp;
404 int error;
405
406 inp = sotoinpcb(so);
407 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
408 if (td && (error = suser(td)) != 0)
409 return error;
410 error = soreserve(so, div_sendspace, div_recvspace);
411 if (error)
412 return error;
413 INP_INFO_WLOCK(&divcbinfo);
414 error = in_pcballoc(so, &divcbinfo, "divinp");
415 if (error) {
416 INP_INFO_WUNLOCK(&divcbinfo);
417 return error;
418 }
419 inp = (struct inpcb *)so->so_pcb;
420 INP_LOCK(inp);
421 INP_INFO_WUNLOCK(&divcbinfo);
422 inp->inp_ip_p = proto;
423 inp->inp_vflag |= INP_IPV4;
424 inp->inp_flags |= INP_HDRINCL;
425 INP_UNLOCK(inp);
426 return 0;
427}
428
429static void
430div_detach(struct socket *so)
431{
432 struct inpcb *inp;
433
434 inp = sotoinpcb(so);
435 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
436 INP_INFO_WLOCK(&divcbinfo);
437 INP_LOCK(inp);
438 in_pcbdetach(inp);
439 in_pcbfree(inp);
440 INP_INFO_WUNLOCK(&divcbinfo);
441}
442
443static int
444div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
445{
446 struct inpcb *inp;
447 int error;
448
449 inp = sotoinpcb(so);
450 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
451 /* in_pcbbind assumes that nam is a sockaddr_in
452 * and in_pcbbind requires a valid address. Since divert
453 * sockets don't we need to make sure the address is
454 * filled in properly.
455 * XXX -- divert should not be abusing in_pcbind
456 * and should probably have its own family.
457 */
458 if (nam->sa_family != AF_INET)
459 return EAFNOSUPPORT;
460 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
461 INP_INFO_WLOCK(&divcbinfo);
462 INP_LOCK(inp);
463 error = in_pcbbind(inp, nam, td->td_ucred);
464 INP_UNLOCK(inp);
465 INP_INFO_WUNLOCK(&divcbinfo);
466 return error;
467}
468
469static int
470div_shutdown(struct socket *so)
471{
472 struct inpcb *inp;
473
474 inp = sotoinpcb(so);
475 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
476 INP_LOCK(inp);
477 socantsendmore(so);
478 INP_UNLOCK(inp);
479 return 0;
480}
481
482static int
483div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
484 struct mbuf *control, struct thread *td)
485{
486 /* Packet must have a header (but that's about it) */
487 if (m->m_len < sizeof (struct ip) &&
488 (m = m_pullup(m, sizeof (struct ip))) == 0) {
489 ipstat.ips_toosmall++;
490 m_freem(m);
491 return EINVAL;
492 }
493
494 /* Send packet */
495 return div_output(so, m, (struct sockaddr_in *)nam, control);
496}
497
498void
499div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
500{
501 struct in_addr faddr;
502
503 faddr = ((struct sockaddr_in *)sa)->sin_addr;
504 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
505 return;
506 if (PRC_IS_REDIRECT(cmd))
507 return;
508}
509
510static int
511div_pcblist(SYSCTL_HANDLER_ARGS)
512{
513 int error, i, n;
514 struct inpcb *inp, **inp_list;
515 inp_gen_t gencnt;
516 struct xinpgen xig;
517
518 /*
519 * The process of preparing the TCB list is too time-consuming and
520 * resource-intensive to repeat twice on every request.
521 */
522 if (req->oldptr == 0) {
523 n = divcbinfo.ipi_count;
524 req->oldidx = 2 * (sizeof xig)
525 + (n + n/8) * sizeof(struct xinpcb);
526 return 0;
527 }
528
529 if (req->newptr != 0)
530 return EPERM;
531
532 /*
533 * OK, now we're committed to doing something.
534 */
535 INP_INFO_RLOCK(&divcbinfo);
536 gencnt = divcbinfo.ipi_gencnt;
537 n = divcbinfo.ipi_count;
538 INP_INFO_RUNLOCK(&divcbinfo);
539
540 error = sysctl_wire_old_buffer(req,
541 2 * sizeof(xig) + n*sizeof(struct xinpcb));
542 if (error != 0)
543 return (error);
544
545 xig.xig_len = sizeof xig;
546 xig.xig_count = n;
547 xig.xig_gen = gencnt;
548 xig.xig_sogen = so_gencnt;
549 error = SYSCTL_OUT(req, &xig, sizeof xig);
550 if (error)
551 return error;
552
553 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
554 if (inp_list == 0)
555 return ENOMEM;
556
557 INP_INFO_RLOCK(&divcbinfo);
558 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
559 inp = LIST_NEXT(inp, inp_list)) {
560 INP_LOCK(inp);
561 if (inp->inp_gencnt <= gencnt &&
562 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
563 inp_list[i++] = inp;
564 INP_UNLOCK(inp);
565 }
566 INP_INFO_RUNLOCK(&divcbinfo);
567 n = i;
568
569 error = 0;
570 for (i = 0; i < n; i++) {
571 inp = inp_list[i];
572 if (inp->inp_gencnt <= gencnt) {
573 struct xinpcb xi;
574 bzero(&xi, sizeof(xi));
575 xi.xi_len = sizeof xi;
576 /* XXX should avoid extra copy */
577 bcopy(inp, &xi.xi_inp, sizeof *inp);
578 if (inp->inp_socket)
579 sotoxsocket(inp->inp_socket, &xi.xi_socket);
580 error = SYSCTL_OUT(req, &xi, sizeof xi);
581 }
582 }
583 if (!error) {
584 /*
585 * Give the user an updated idea of our state.
586 * If the generation differs from what we told
587 * her before, she knows that something happened
588 * while we were processing this request, and it
589 * might be necessary to retry.
590 */
591 INP_INFO_RLOCK(&divcbinfo);
592 xig.xig_gen = divcbinfo.ipi_gencnt;
593 xig.xig_sogen = so_gencnt;
594 xig.xig_count = divcbinfo.ipi_count;
595 INP_INFO_RUNLOCK(&divcbinfo);
596 error = SYSCTL_OUT(req, &xig, sizeof xig);
597 }
598 free(inp_list, M_TEMP);
599 return error;
600}
601
602/*
603 * This is the wrapper function for in_setsockaddr. We just pass down
604 * the pcbinfo for in_setpeeraddr to lock.
605 */
606static int
607div_sockaddr(struct socket *so, struct sockaddr **nam)
608{
609 return (in_setsockaddr(so, nam, &divcbinfo));
610}
611
612/*
613 * This is the wrapper function for in_setpeeraddr. We just pass down
614 * the pcbinfo for in_setpeeraddr to lock.
615 */
616static int
617div_peeraddr(struct socket *so, struct sockaddr **nam)
618{
619 return (in_setpeeraddr(so, nam, &divcbinfo));
620}
621
622#ifdef SYSCTL_NODE
623SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
624SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
625 div_pcblist, "S,xinpcb", "List of active divert sockets");
626#endif
627
628struct pr_usrreqs div_usrreqs = {
629 .pru_attach = div_attach,
630 .pru_bind = div_bind,
631 .pru_control = in_control,
632 .pru_detach = div_detach,
633 .pru_peeraddr = div_peeraddr,
634 .pru_send = div_send,
635 .pru_shutdown = div_shutdown,
636 .pru_sockaddr = div_sockaddr,
637 .pru_sosetlabel = in_pcbsosetlabel
638};
639
640struct protosw div_protosw = {
641 .pr_type = SOCK_RAW,
642 .pr_protocol = IPPROTO_DIVERT,
643 .pr_flags = PR_ATOMIC|PR_ADDR,
644 .pr_input = div_input,
645 .pr_ctlinput = div_ctlinput,
646 .pr_ctloutput = ip_ctloutput,
647 .pr_init = div_init,
648 .pr_usrreqs = &div_usrreqs
649};
650
651static int
652div_modevent(module_t mod, int type, void *unused)
653{
654 int err = 0;
655 int n;
656
657 switch (type) {
658 case MOD_LOAD:
659 /*
660 * Protocol will be initialized by pf_proto_register().
661 * We don't have to register ip_protox because we are not
662 * a true IP protocol that goes over the wire.
663 */
664 err = pf_proto_register(PF_INET, &div_protosw);
665 ip_divert_ptr = divert_packet;
666 break;
667 case MOD_QUIESCE:
668 /*
669 * IPDIVERT may normally not be unloaded because of the
670 * potential race conditions. Tell kldunload we can't be
671 * unloaded unless the unload is forced.
672 */
673 err = EPERM;
674 break;
675 case MOD_UNLOAD:
676 /*
677 * Forced unload.
678 *
679 * Module ipdivert can only be unloaded if no sockets are
680 * connected. Maybe this can be changed later to forcefully
681 * disconnect any open sockets.
682 *
683 * XXXRW: Note that there is a slight race here, as a new
684 * socket open request could be spinning on the lock and then
685 * we destroy the lock.
686 */
687 INP_INFO_WLOCK(&divcbinfo);
688 n = divcbinfo.ipi_count;
689 if (n != 0) {
690 err = EBUSY;
691 INP_INFO_WUNLOCK(&divcbinfo);
692 break;
693 }
694 ip_divert_ptr = NULL;
695 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
696 INP_INFO_WUNLOCK(&divcbinfo);
697 INP_INFO_LOCK_DESTROY(&divcbinfo);
698 uma_zdestroy(divcbinfo.ipi_zone);
699 break;
700 default:
701 err = EOPNOTSUPP;
702 break;
703 }
704 return err;
705}
706
707static moduledata_t ipdivertmod = {
708 "ipdivert",
709 div_modevent,
710 0
711};
712
713DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
714MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
715MODULE_VERSION(ipdivert, 1);