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: releng/6.2/sys/netinet/ip_divert.c 158641 2006-05-16 07:27:49Z 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. */
110 static struct inpcbhead divcb;
111 static struct inpcbinfo divcbinfo;
112
113 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
114 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
115
116 /*
117 * Initialize divert connection block queue.
118 */
119 static void
120 div_zone_change(void *tag)
121 {
122
123 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
124 }
125
126 void
127 div_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 */
150 void
151 div_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 */
163 static void
164 divert_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 */
281 static int
282 div_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
395 cantsend:
396 m_freem(m);
397 return error;
398 }
399
400 static int
401 div_attach(struct socket *so, int proto, struct thread *td)
402 {
403 struct inpcb *inp;
404 int error;
405
406 INP_INFO_WLOCK(&divcbinfo);
407 inp = sotoinpcb(so);
408 if (inp != 0) {
409 INP_INFO_WUNLOCK(&divcbinfo);
410 return EINVAL;
411 }
412 if (td && (error = suser(td)) != 0) {
413 INP_INFO_WUNLOCK(&divcbinfo);
414 return error;
415 }
416 error = soreserve(so, div_sendspace, div_recvspace);
417 if (error) {
418 INP_INFO_WUNLOCK(&divcbinfo);
419 return error;
420 }
421 error = in_pcballoc(so, &divcbinfo, "divinp");
422 if (error) {
423 INP_INFO_WUNLOCK(&divcbinfo);
424 return error;
425 }
426 inp = (struct inpcb *)so->so_pcb;
427 INP_LOCK(inp);
428 INP_INFO_WUNLOCK(&divcbinfo);
429 inp->inp_ip_p = proto;
430 inp->inp_vflag |= INP_IPV4;
431 inp->inp_flags |= INP_HDRINCL;
432 INP_UNLOCK(inp);
433 return 0;
434 }
435
436 static int
437 div_detach(struct socket *so)
438 {
439 struct inpcb *inp;
440
441 INP_INFO_WLOCK(&divcbinfo);
442 inp = sotoinpcb(so);
443 if (inp == 0) {
444 INP_INFO_WUNLOCK(&divcbinfo);
445 return EINVAL;
446 }
447 INP_LOCK(inp);
448 in_pcbdetach(inp);
449 INP_INFO_WUNLOCK(&divcbinfo);
450 return 0;
451 }
452
453 static int
454 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
455 {
456 struct inpcb *inp;
457 int error;
458
459 INP_INFO_WLOCK(&divcbinfo);
460 inp = sotoinpcb(so);
461 if (inp == 0) {
462 INP_INFO_WUNLOCK(&divcbinfo);
463 return EINVAL;
464 }
465 /* in_pcbbind assumes that nam is a sockaddr_in
466 * and in_pcbbind requires a valid address. Since divert
467 * sockets don't we need to make sure the address is
468 * filled in properly.
469 * XXX -- divert should not be abusing in_pcbind
470 * and should probably have its own family.
471 */
472 if (nam->sa_family != AF_INET)
473 error = EAFNOSUPPORT;
474 else {
475 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
476 INP_LOCK(inp);
477 error = in_pcbbind(inp, nam, td->td_ucred);
478 INP_UNLOCK(inp);
479 }
480 INP_INFO_WUNLOCK(&divcbinfo);
481 return error;
482 }
483
484 static int
485 div_shutdown(struct socket *so)
486 {
487 struct inpcb *inp;
488
489 INP_INFO_RLOCK(&divcbinfo);
490 inp = sotoinpcb(so);
491 if (inp == 0) {
492 INP_INFO_RUNLOCK(&divcbinfo);
493 return EINVAL;
494 }
495 INP_LOCK(inp);
496 INP_INFO_RUNLOCK(&divcbinfo);
497 socantsendmore(so);
498 INP_UNLOCK(inp);
499 return 0;
500 }
501
502 static int
503 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
504 struct mbuf *control, struct thread *td)
505 {
506 /* Packet must have a header (but that's about it) */
507 if (m->m_len < sizeof (struct ip) &&
508 (m = m_pullup(m, sizeof (struct ip))) == 0) {
509 ipstat.ips_toosmall++;
510 m_freem(m);
511 return EINVAL;
512 }
513
514 /* Send packet */
515 return div_output(so, m, (struct sockaddr_in *)nam, control);
516 }
517
518 void
519 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
520 {
521 struct in_addr faddr;
522
523 faddr = ((struct sockaddr_in *)sa)->sin_addr;
524 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
525 return;
526 if (PRC_IS_REDIRECT(cmd))
527 return;
528 }
529
530 static int
531 div_pcblist(SYSCTL_HANDLER_ARGS)
532 {
533 int error, i, n;
534 struct inpcb *inp, **inp_list;
535 inp_gen_t gencnt;
536 struct xinpgen xig;
537
538 /*
539 * The process of preparing the TCB list is too time-consuming and
540 * resource-intensive to repeat twice on every request.
541 */
542 if (req->oldptr == 0) {
543 n = divcbinfo.ipi_count;
544 req->oldidx = 2 * (sizeof xig)
545 + (n + n/8) * sizeof(struct xinpcb);
546 return 0;
547 }
548
549 if (req->newptr != 0)
550 return EPERM;
551
552 /*
553 * OK, now we're committed to doing something.
554 */
555 INP_INFO_RLOCK(&divcbinfo);
556 gencnt = divcbinfo.ipi_gencnt;
557 n = divcbinfo.ipi_count;
558 INP_INFO_RUNLOCK(&divcbinfo);
559
560 error = sysctl_wire_old_buffer(req,
561 2 * sizeof(xig) + n*sizeof(struct xinpcb));
562 if (error != 0)
563 return (error);
564
565 xig.xig_len = sizeof xig;
566 xig.xig_count = n;
567 xig.xig_gen = gencnt;
568 xig.xig_sogen = so_gencnt;
569 error = SYSCTL_OUT(req, &xig, sizeof xig);
570 if (error)
571 return error;
572
573 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
574 if (inp_list == 0)
575 return ENOMEM;
576
577 INP_INFO_RLOCK(&divcbinfo);
578 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
579 inp = LIST_NEXT(inp, inp_list)) {
580 INP_LOCK(inp);
581 if (inp->inp_gencnt <= gencnt &&
582 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
583 inp_list[i++] = inp;
584 INP_UNLOCK(inp);
585 }
586 INP_INFO_RUNLOCK(&divcbinfo);
587 n = i;
588
589 error = 0;
590 for (i = 0; i < n; i++) {
591 inp = inp_list[i];
592 if (inp->inp_gencnt <= gencnt) {
593 struct xinpcb xi;
594 bzero(&xi, sizeof(xi));
595 xi.xi_len = sizeof xi;
596 /* XXX should avoid extra copy */
597 bcopy(inp, &xi.xi_inp, sizeof *inp);
598 if (inp->inp_socket)
599 sotoxsocket(inp->inp_socket, &xi.xi_socket);
600 error = SYSCTL_OUT(req, &xi, sizeof xi);
601 }
602 }
603 if (!error) {
604 /*
605 * Give the user an updated idea of our state.
606 * If the generation differs from what we told
607 * her before, she knows that something happened
608 * while we were processing this request, and it
609 * might be necessary to retry.
610 */
611 INP_INFO_RLOCK(&divcbinfo);
612 xig.xig_gen = divcbinfo.ipi_gencnt;
613 xig.xig_sogen = so_gencnt;
614 xig.xig_count = divcbinfo.ipi_count;
615 INP_INFO_RUNLOCK(&divcbinfo);
616 error = SYSCTL_OUT(req, &xig, sizeof xig);
617 }
618 free(inp_list, M_TEMP);
619 return error;
620 }
621
622 /*
623 * This is the wrapper function for in_setsockaddr. We just pass down
624 * the pcbinfo for in_setpeeraddr to lock.
625 */
626 static int
627 div_sockaddr(struct socket *so, struct sockaddr **nam)
628 {
629 return (in_setsockaddr(so, nam, &divcbinfo));
630 }
631
632 /*
633 * This is the wrapper function for in_setpeeraddr. We just pass down
634 * the pcbinfo for in_setpeeraddr to lock.
635 */
636 static int
637 div_peeraddr(struct socket *so, struct sockaddr **nam)
638 {
639 return (in_setpeeraddr(so, nam, &divcbinfo));
640 }
641
642 #ifdef SYSCTL_NODE
643 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
644 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
645 div_pcblist, "S,xinpcb", "List of active divert sockets");
646 #endif
647
648 struct pr_usrreqs div_usrreqs = {
649 .pru_attach = div_attach,
650 .pru_bind = div_bind,
651 .pru_control = in_control,
652 .pru_detach = div_detach,
653 .pru_peeraddr = div_peeraddr,
654 .pru_send = div_send,
655 .pru_shutdown = div_shutdown,
656 .pru_sockaddr = div_sockaddr,
657 .pru_sosetlabel = in_pcbsosetlabel
658 };
659
660 struct protosw div_protosw = {
661 .pr_type = SOCK_RAW,
662 .pr_protocol = IPPROTO_DIVERT,
663 .pr_flags = PR_ATOMIC|PR_ADDR,
664 .pr_input = div_input,
665 .pr_ctlinput = div_ctlinput,
666 .pr_ctloutput = ip_ctloutput,
667 .pr_init = div_init,
668 .pr_usrreqs = &div_usrreqs
669 };
670
671 static int
672 div_modevent(module_t mod, int type, void *unused)
673 {
674 int err = 0;
675 int n;
676
677 switch (type) {
678 case MOD_LOAD:
679 /*
680 * Protocol will be initialized by pf_proto_register().
681 * We don't have to register ip_protox because we are not
682 * a true IP protocol that goes over the wire.
683 */
684 err = pf_proto_register(PF_INET, &div_protosw);
685 ip_divert_ptr = divert_packet;
686 break;
687 case MOD_QUIESCE:
688 /*
689 * IPDIVERT may normally not be unloaded because of the
690 * potential race conditions. Tell kldunload we can't be
691 * unloaded unless the unload is forced.
692 */
693 err = EPERM;
694 break;
695 case MOD_UNLOAD:
696 /*
697 * Forced unload.
698 *
699 * Module ipdivert can only be unloaded if no sockets are
700 * connected. Maybe this can be changed later to forcefully
701 * disconnect any open sockets.
702 *
703 * XXXRW: Note that there is a slight race here, as a new
704 * socket open request could be spinning on the lock and then
705 * we destroy the lock.
706 */
707 INP_INFO_WLOCK(&divcbinfo);
708 n = divcbinfo.ipi_count;
709 if (n != 0) {
710 err = EBUSY;
711 INP_INFO_WUNLOCK(&divcbinfo);
712 break;
713 }
714 ip_divert_ptr = NULL;
715 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
716 INP_INFO_WUNLOCK(&divcbinfo);
717 INP_INFO_LOCK_DESTROY(&divcbinfo);
718 uma_zdestroy(divcbinfo.ipi_zone);
719 break;
720 default:
721 err = EOPNOTSUPP;
722 break;
723 }
724 return err;
725 }
726
727 static moduledata_t ipdivertmod = {
728 "ipdivert",
729 div_modevent,
730 0
731 };
732
733 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
734 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
735 MODULE_VERSION(ipdivert, 1);
Cache object: 6af130759f1976705edb456f17a50bd6
|