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
div_zone_change(void * tag)128 div_zone_change(void *tag)
129 {
130
131 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
132 }
133
134 static int
div_inpcb_init(void * mem,int size,int flags)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
div_inpcb_fini(void * mem,int size)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
div_init(void)152 div_init(void)
153 {
154
155 /*
156 * XXX We don't use the hash list for divert IP, but it's easier to
157 * allocate one-entry hash lists than it is to check all over the
158 * place for hashbase == NULL.
159 */
160 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
161 div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE,
162 IPI_HASHFIELDS_NONE);
163 }
164
165 static void
div_destroy(void)166 div_destroy(void)
167 {
168
169 in_pcbinfo_destroy(&V_divcbinfo);
170 }
171
172 /*
173 * IPPROTO_DIVERT is not in the real IP protocol number space; this
174 * function should never be called. Just in case, drop any packets.
175 */
176 static int
div_input(struct mbuf ** mp,int * offp,int proto)177 div_input(struct mbuf **mp, int *offp, int proto)
178 {
179 struct mbuf *m = *mp;
180
181 KMOD_IPSTAT_INC(ips_noproto);
182 m_freem(m);
183 return (IPPROTO_DONE);
184 }
185
186 /*
187 * Divert a packet by passing it up to the divert socket at port 'port'.
188 *
189 * Setup generic address and protocol structures for div_input routine,
190 * then pass them along with mbuf chain.
191 */
192 static void
divert_packet(struct mbuf * m,int incoming)193 divert_packet(struct mbuf *m, int incoming)
194 {
195 struct ip *ip;
196 struct inpcb *inp;
197 struct socket *sa;
198 u_int16_t nport;
199 struct sockaddr_in divsrc;
200 struct m_tag *mtag;
201
202 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
203 if (mtag == NULL) {
204 m_freem(m);
205 return;
206 }
207 /* Assure header */
208 if (m->m_len < sizeof(struct ip) &&
209 (m = m_pullup(m, sizeof(struct ip))) == 0)
210 return;
211 ip = mtod(m, struct ip *);
212
213 /* Delayed checksums are currently not compatible with divert. */
214 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
215 in_delayed_cksum(m);
216 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
217 }
218 #ifdef SCTP
219 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
220 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
221 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
222 }
223 #endif
224 bzero(&divsrc, sizeof(divsrc));
225 divsrc.sin_len = sizeof(divsrc);
226 divsrc.sin_family = AF_INET;
227 /* record matching rule, in host format */
228 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
229 /*
230 * Record receive interface address, if any.
231 * But only for incoming packets.
232 */
233 if (incoming) {
234 struct ifaddr *ifa;
235 struct ifnet *ifp;
236
237 /* Sanity check */
238 M_ASSERTPKTHDR(m);
239
240 /* Find IP address for receive interface */
241 ifp = m->m_pkthdr.rcvif;
242 if_addr_rlock(ifp);
243 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
244 if (ifa->ifa_addr->sa_family != AF_INET)
245 continue;
246 divsrc.sin_addr =
247 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
248 break;
249 }
250 if_addr_runlock(ifp);
251 }
252 /*
253 * Record the incoming interface name whenever we have one.
254 */
255 if (m->m_pkthdr.rcvif) {
256 /*
257 * Hide the actual interface name in there in the
258 * sin_zero array. XXX This needs to be moved to a
259 * different sockaddr type for divert, e.g.
260 * sockaddr_div with multiple fields like
261 * sockaddr_dl. Presently we have only 7 bytes
262 * but that will do for now as most interfaces
263 * are 4 or less + 2 or less bytes for unit.
264 * There is probably a faster way of doing this,
265 * possibly taking it from the sockaddr_dl on the iface.
266 * This solves the problem of a P2P link and a LAN interface
267 * having the same address, which can result in the wrong
268 * interface being assigned to the packet when fed back
269 * into the divert socket. Theoretically if the daemon saves
270 * and re-uses the sockaddr_in as suggested in the man pages,
271 * this iface name will come along for the ride.
272 * (see div_output for the other half of this.)
273 */
274 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
275 sizeof(divsrc.sin_zero));
276 }
277
278 /* Put packet on socket queue, if any */
279 sa = NULL;
280 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
281 INP_INFO_RLOCK(&V_divcbinfo);
282 LIST_FOREACH(inp, &V_divcb, inp_list) {
283 /* XXX why does only one socket match? */
284 if (inp->inp_lport == nport) {
285 INP_RLOCK(inp);
286 sa = inp->inp_socket;
287 SOCKBUF_LOCK(&sa->so_rcv);
288 if (sbappendaddr_locked(&sa->so_rcv,
289 (struct sockaddr *)&divsrc, m,
290 (struct mbuf *)0) == 0) {
291 SOCKBUF_UNLOCK(&sa->so_rcv);
292 sa = NULL; /* force mbuf reclaim below */
293 } else
294 sorwakeup_locked(sa);
295 INP_RUNLOCK(inp);
296 break;
297 }
298 }
299 INP_INFO_RUNLOCK(&V_divcbinfo);
300 if (sa == NULL) {
301 m_freem(m);
302 KMOD_IPSTAT_INC(ips_noproto);
303 KMOD_IPSTAT_DEC(ips_delivered);
304 }
305 }
306
307 /*
308 * Deliver packet back into the IP processing machinery.
309 *
310 * If no address specified, or address is 0.0.0.0, send to ip_output();
311 * otherwise, send to ip_input() and mark as having been received on
312 * the interface with that address.
313 */
314 static int
div_output(struct socket * so,struct mbuf * m,struct sockaddr_in * sin,struct mbuf * control)315 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
316 struct mbuf *control)
317 {
318 struct ip *const ip = mtod(m, struct ip *);
319 struct m_tag *mtag;
320 struct ipfw_rule_ref *dt;
321 int error = 0;
322
323 /*
324 * An mbuf may hasn't come from userland, but we pretend
325 * that it has.
326 */
327 m->m_pkthdr.rcvif = NULL;
328 m->m_nextpkt = NULL;
329 M_SETFIB(m, so->so_fibnum);
330
331 if (control)
332 m_freem(control); /* XXX */
333
334 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
335 if (mtag == NULL) {
336 /* this should be normal */
337 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
338 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
339 if (mtag == NULL) {
340 error = ENOBUFS;
341 goto cantsend;
342 }
343 m_tag_prepend(m, mtag);
344 }
345 dt = (struct ipfw_rule_ref *)(mtag+1);
346
347 /* Loopback avoidance and state recovery */
348 if (sin) {
349 int i;
350
351 /* set the starting point. We provide a non-zero slot,
352 * but a non_matching chain_id to skip that info and use
353 * the rulenum/rule_id.
354 */
355 dt->slot = 1; /* dummy, chain_id is invalid */
356 dt->chain_id = 0;
357 dt->rulenum = sin->sin_port+1; /* host format ? */
358 dt->rule_id = 0;
359 /*
360 * Find receive interface with the given name, stuffed
361 * (if it exists) in the sin_zero[] field.
362 * The name is user supplied data so don't trust its size
363 * or that it is zero terminated.
364 */
365 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
366 ;
367 if ( i > 0 && i < sizeof(sin->sin_zero))
368 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
369 }
370
371 /* Reinject packet into the system as incoming or outgoing */
372 if (!sin || sin->sin_addr.s_addr == 0) {
373 struct mbuf *options = NULL;
374 struct inpcb *inp;
375
376 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
377 inp = sotoinpcb(so);
378 INP_RLOCK(inp);
379 switch (ip->ip_v) {
380 case IPVERSION:
381 /*
382 * Don't allow both user specified and setsockopt
383 * options, and don't allow packet length sizes that
384 * will crash.
385 */
386 if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
387 inp->inp_options != NULL) ||
388 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
389 error = EINVAL;
390 INP_RUNLOCK(inp);
391 goto cantsend;
392 }
393 break;
394 #ifdef INET6
395 case IPV6_VERSION >> 4:
396 {
397 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
398
399 /* Don't allow packet length sizes that will crash */
400 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
401 error = EINVAL;
402 INP_RUNLOCK(inp);
403 goto cantsend;
404 }
405 break;
406 }
407 #endif
408 default:
409 error = EINVAL;
410 INP_RUNLOCK(inp);
411 goto cantsend;
412 }
413
414 /* Send packet to output processing */
415 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
416
417 #ifdef MAC
418 mac_inpcb_create_mbuf(inp, m);
419 #endif
420 /*
421 * Get ready to inject the packet into ip_output().
422 * Just in case socket options were specified on the
423 * divert socket, we duplicate them. This is done
424 * to avoid having to hold the PCB locks over the call
425 * to ip_output(), as doing this results in a number of
426 * lock ordering complexities.
427 *
428 * Note that we set the multicast options argument for
429 * ip_output() to NULL since it should be invariant that
430 * they are not present.
431 */
432 KASSERT(inp->inp_moptions == NULL,
433 ("multicast options set on a divert socket"));
434 /*
435 * XXXCSJP: It is unclear to me whether or not it makes
436 * sense for divert sockets to have options. However,
437 * for now we will duplicate them with the INP locks
438 * held so we can use them in ip_output() without
439 * requring a reference to the pcb.
440 */
441 if (inp->inp_options != NULL) {
442 options = m_dup(inp->inp_options, M_NOWAIT);
443 if (options == NULL) {
444 INP_RUNLOCK(inp);
445 error = ENOBUFS;
446 goto cantsend;
447 }
448 }
449 INP_RUNLOCK(inp);
450
451 switch (ip->ip_v) {
452 case IPVERSION:
453 error = ip_output(m, options, NULL,
454 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
455 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
456 break;
457 #ifdef INET6
458 case IPV6_VERSION >> 4:
459 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
460 break;
461 #endif
462 }
463 if (options != NULL)
464 m_freem(options);
465 } else {
466 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
467 if (m->m_pkthdr.rcvif == NULL) {
468 /*
469 * No luck with the name, check by IP address.
470 * Clear the port and the ifname to make sure
471 * there are no distractions for ifa_ifwithaddr.
472 */
473 struct ifaddr *ifa;
474
475 bzero(sin->sin_zero, sizeof(sin->sin_zero));
476 sin->sin_port = 0;
477 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
478 if (ifa == NULL) {
479 error = EADDRNOTAVAIL;
480 goto cantsend;
481 }
482 m->m_pkthdr.rcvif = ifa->ifa_ifp;
483 ifa_free(ifa);
484 }
485 #ifdef MAC
486 mac_socket_create_mbuf(so, m);
487 #endif
488 /* Send packet to input processing via netisr */
489 switch (ip->ip_v) {
490 case IPVERSION:
491 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
492 break;
493 #ifdef INET6
494 case IPV6_VERSION >> 4:
495 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
496 break;
497 #endif
498 default:
499 error = EINVAL;
500 goto cantsend;
501 }
502 }
503
504 return (error);
505
506 cantsend:
507 m_freem(m);
508 return (error);
509 }
510
511 static int
div_attach(struct socket * so,int proto,struct thread * td)512 div_attach(struct socket *so, int proto, struct thread *td)
513 {
514 struct inpcb *inp;
515 int error;
516
517 inp = sotoinpcb(so);
518 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
519 if (td != NULL) {
520 error = priv_check(td, PRIV_NETINET_DIVERT);
521 if (error)
522 return (error);
523 }
524 error = soreserve(so, div_sendspace, div_recvspace);
525 if (error)
526 return error;
527 INP_INFO_WLOCK(&V_divcbinfo);
528 error = in_pcballoc(so, &V_divcbinfo);
529 if (error) {
530 INP_INFO_WUNLOCK(&V_divcbinfo);
531 return error;
532 }
533 inp = (struct inpcb *)so->so_pcb;
534 INP_INFO_WUNLOCK(&V_divcbinfo);
535 inp->inp_ip_p = proto;
536 inp->inp_vflag |= INP_IPV4;
537 inp->inp_flags |= INP_HDRINCL;
538 INP_WUNLOCK(inp);
539 return 0;
540 }
541
542 static void
div_detach(struct socket * so)543 div_detach(struct socket *so)
544 {
545 struct inpcb *inp;
546
547 inp = sotoinpcb(so);
548 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
549 INP_INFO_WLOCK(&V_divcbinfo);
550 INP_WLOCK(inp);
551 in_pcbdetach(inp);
552 in_pcbfree(inp);
553 INP_INFO_WUNLOCK(&V_divcbinfo);
554 }
555
556 static int
div_bind(struct socket * so,struct sockaddr * nam,struct thread * td)557 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
558 {
559 struct inpcb *inp;
560 int error;
561
562 inp = sotoinpcb(so);
563 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
564 /* in_pcbbind assumes that nam is a sockaddr_in
565 * and in_pcbbind requires a valid address. Since divert
566 * sockets don't we need to make sure the address is
567 * filled in properly.
568 * XXX -- divert should not be abusing in_pcbind
569 * and should probably have its own family.
570 */
571 if (nam->sa_family != AF_INET)
572 return EAFNOSUPPORT;
573 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
574 INP_INFO_WLOCK(&V_divcbinfo);
575 INP_WLOCK(inp);
576 INP_HASH_WLOCK(&V_divcbinfo);
577 error = in_pcbbind(inp, nam, td->td_ucred);
578 INP_HASH_WUNLOCK(&V_divcbinfo);
579 INP_WUNLOCK(inp);
580 INP_INFO_WUNLOCK(&V_divcbinfo);
581 return error;
582 }
583
584 static int
div_shutdown(struct socket * so)585 div_shutdown(struct socket *so)
586 {
587 struct inpcb *inp;
588
589 inp = sotoinpcb(so);
590 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
591 INP_WLOCK(inp);
592 socantsendmore(so);
593 INP_WUNLOCK(inp);
594 return 0;
595 }
596
597 static int
div_send(struct socket * so,int flags,struct mbuf * m,struct sockaddr * nam,struct mbuf * control,struct thread * td)598 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
599 struct mbuf *control, struct thread *td)
600 {
601
602 /* Packet must have a header (but that's about it) */
603 if (m->m_len < sizeof (struct ip) &&
604 (m = m_pullup(m, sizeof (struct ip))) == 0) {
605 KMOD_IPSTAT_INC(ips_toosmall);
606 m_freem(m);
607 return EINVAL;
608 }
609
610 /* Send packet */
611 return div_output(so, m, (struct sockaddr_in *)nam, control);
612 }
613
614 static void
div_ctlinput(int cmd,struct sockaddr * sa,void * vip)615 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
616 {
617 struct in_addr faddr;
618
619 faddr = ((struct sockaddr_in *)sa)->sin_addr;
620 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
621 return;
622 if (PRC_IS_REDIRECT(cmd))
623 return;
624 }
625
626 static int
div_pcblist(SYSCTL_HANDLER_ARGS)627 div_pcblist(SYSCTL_HANDLER_ARGS)
628 {
629 int error, i, n;
630 struct inpcb *inp, **inp_list;
631 inp_gen_t gencnt;
632 struct xinpgen xig;
633
634 /*
635 * The process of preparing the TCB list is too time-consuming and
636 * resource-intensive to repeat twice on every request.
637 */
638 if (req->oldptr == 0) {
639 n = V_divcbinfo.ipi_count;
640 n += imax(n / 8, 10);
641 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
642 return 0;
643 }
644
645 if (req->newptr != 0)
646 return EPERM;
647
648 /*
649 * OK, now we're committed to doing something.
650 */
651 INP_INFO_RLOCK(&V_divcbinfo);
652 gencnt = V_divcbinfo.ipi_gencnt;
653 n = V_divcbinfo.ipi_count;
654 INP_INFO_RUNLOCK(&V_divcbinfo);
655
656 error = sysctl_wire_old_buffer(req,
657 2 * sizeof(xig) + n*sizeof(struct xinpcb));
658 if (error != 0)
659 return (error);
660
661 xig.xig_len = sizeof xig;
662 xig.xig_count = n;
663 xig.xig_gen = gencnt;
664 xig.xig_sogen = so_gencnt;
665 error = SYSCTL_OUT(req, &xig, sizeof xig);
666 if (error)
667 return error;
668
669 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
670 if (inp_list == 0)
671 return ENOMEM;
672
673 INP_INFO_RLOCK(&V_divcbinfo);
674 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
675 inp = LIST_NEXT(inp, inp_list)) {
676 INP_WLOCK(inp);
677 if (inp->inp_gencnt <= gencnt &&
678 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
679 in_pcbref(inp);
680 inp_list[i++] = inp;
681 }
682 INP_WUNLOCK(inp);
683 }
684 INP_INFO_RUNLOCK(&V_divcbinfo);
685 n = i;
686
687 error = 0;
688 for (i = 0; i < n; i++) {
689 inp = inp_list[i];
690 INP_RLOCK(inp);
691 if (inp->inp_gencnt <= gencnt) {
692 struct xinpcb xi;
693 bzero(&xi, sizeof(xi));
694 xi.xi_len = sizeof xi;
695 /* XXX should avoid extra copy */
696 bcopy(inp, &xi.xi_inp, sizeof *inp);
697 if (inp->inp_socket)
698 sotoxsocket(inp->inp_socket, &xi.xi_socket);
699 INP_RUNLOCK(inp);
700 error = SYSCTL_OUT(req, &xi, sizeof xi);
701 } else
702 INP_RUNLOCK(inp);
703 }
704 INP_INFO_WLOCK(&V_divcbinfo);
705 for (i = 0; i < n; i++) {
706 inp = inp_list[i];
707 INP_RLOCK(inp);
708 if (!in_pcbrele_rlocked(inp))
709 INP_RUNLOCK(inp);
710 }
711 INP_INFO_WUNLOCK(&V_divcbinfo);
712
713 if (!error) {
714 /*
715 * Give the user an updated idea of our state.
716 * If the generation differs from what we told
717 * her before, she knows that something happened
718 * while we were processing this request, and it
719 * might be necessary to retry.
720 */
721 INP_INFO_RLOCK(&V_divcbinfo);
722 xig.xig_gen = V_divcbinfo.ipi_gencnt;
723 xig.xig_sogen = so_gencnt;
724 xig.xig_count = V_divcbinfo.ipi_count;
725 INP_INFO_RUNLOCK(&V_divcbinfo);
726 error = SYSCTL_OUT(req, &xig, sizeof xig);
727 }
728 free(inp_list, M_TEMP);
729 return error;
730 }
731
732 #ifdef SYSCTL_NODE
733 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
734 "IPDIVERT");
735 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
736 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
737 #endif
738
739 struct pr_usrreqs div_usrreqs = {
740 .pru_attach = div_attach,
741 .pru_bind = div_bind,
742 .pru_control = in_control,
743 .pru_detach = div_detach,
744 .pru_peeraddr = in_getpeeraddr,
745 .pru_send = div_send,
746 .pru_shutdown = div_shutdown,
747 .pru_sockaddr = in_getsockaddr,
748 .pru_sosetlabel = in_pcbsosetlabel
749 };
750
751 struct protosw div_protosw = {
752 .pr_type = SOCK_RAW,
753 .pr_protocol = IPPROTO_DIVERT,
754 .pr_flags = PR_ATOMIC|PR_ADDR,
755 .pr_input = div_input,
756 .pr_ctlinput = div_ctlinput,
757 .pr_ctloutput = ip_ctloutput,
758 .pr_init = div_init,
759 #ifdef VIMAGE
760 .pr_destroy = div_destroy,
761 #endif
762 .pr_usrreqs = &div_usrreqs
763 };
764
765 static int
div_modevent(module_t mod,int type,void * unused)766 div_modevent(module_t mod, int type, void *unused)
767 {
768 int err = 0;
769 #ifndef VIMAGE
770 int n;
771 #endif
772
773 switch (type) {
774 case MOD_LOAD:
775 /*
776 * Protocol will be initialized by pf_proto_register().
777 * We don't have to register ip_protox because we are not
778 * a true IP protocol that goes over the wire.
779 */
780 err = pf_proto_register(PF_INET, &div_protosw);
781 if (err != 0)
782 return (err);
783 ip_divert_ptr = divert_packet;
784 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
785 div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
786 break;
787 case MOD_QUIESCE:
788 /*
789 * IPDIVERT may normally not be unloaded because of the
790 * potential race conditions. Tell kldunload we can't be
791 * unloaded unless the unload is forced.
792 */
793 err = EPERM;
794 break;
795 case MOD_UNLOAD:
796 #ifdef VIMAGE
797 err = EPERM;
798 break;
799 #else
800 /*
801 * Forced unload.
802 *
803 * Module ipdivert can only be unloaded if no sockets are
804 * connected. Maybe this can be changed later to forcefully
805 * disconnect any open sockets.
806 *
807 * XXXRW: Note that there is a slight race here, as a new
808 * socket open request could be spinning on the lock and then
809 * we destroy the lock.
810 */
811 INP_INFO_WLOCK(&V_divcbinfo);
812 n = V_divcbinfo.ipi_count;
813 if (n != 0) {
814 err = EBUSY;
815 INP_INFO_WUNLOCK(&V_divcbinfo);
816 break;
817 }
818 ip_divert_ptr = NULL;
819 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
820 INP_INFO_WUNLOCK(&V_divcbinfo);
821 div_destroy();
822 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
823 break;
824 #endif /* !VIMAGE */
825 default:
826 err = EOPNOTSUPP;
827 break;
828 }
829 return err;
830 }
831
832 static moduledata_t ipdivertmod = {
833 "ipdivert",
834 div_modevent,
835 0
836 };
837
838 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
839 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
840 MODULE_VERSION(ipdivert, 1);
841