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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
| 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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
|
34 */
35
36#include "opt_inet.h"
37#include "opt_ipfw.h"
38#include "opt_ipdivert.h"
39#include "opt_ipsec.h"
40#include "opt_mac.h"
41
42#ifndef INET
43#error "IPDIVERT requires INET."
44#endif
45
46#include <sys/param.h>
47#include <sys/kernel.h>
48#include <sys/lock.h>
49#include <sys/malloc.h>
50#include <sys/mac.h>
51#include <sys/mbuf.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
74/*
75 * Divert sockets
76 */
77
78/*
79 * Allocate enough space to hold a full IP packet
80 */
81#define DIVSNDQ (65536 + 100)
82#define DIVRCVQ (65536 + 100)
83
84/*
85 * Divert sockets work in conjunction with ipfw, see the divert(4)
86 * manpage for features.
87 * Internally, packets selected by ipfw in ip_input() or ip_output(),
88 * and never diverted before, are passed to the input queue of the
89 * divert socket with a given 'divert_port' number (as specified in
90 * the matching ipfw rule), and they are tagged with a 16 bit cookie
91 * (representing the rule number of the matching ipfw rule), which
92 * is passed to process reading from the socket.
93 *
94 * Packets written to the divert socket are again tagged with a cookie
95 * (usually the same as above) and a destination address.
96 * If the destination address is INADDR_ANY then the packet is
97 * treated as outgoing and sent to ip_output(), otherwise it is
98 * treated as incoming and sent to ip_input().
99 * In both cases, the packet is tagged with the cookie.
100 *
101 * On reinjection, processing in ip_input() and ip_output()
102 * will be exactly the same as for the original packet, except that
103 * ipfw processing will start at the rule number after the one
104 * written in the cookie (so, tagging a packet with a cookie of 0
105 * will cause it to be effectively considered as a standard packet).
106 */
107
108/* Internal variables */
109static struct inpcbhead divcb;
110static struct inpcbinfo divcbinfo;
111
112static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
113static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
114
115/*
116 * Initialize divert connection block queue.
117 */
118void
119div_init(void)
120{
121 INP_INFO_LOCK_INIT(&divcbinfo, "div");
122 LIST_INIT(&divcb);
123 divcbinfo.listhead = &divcb;
124 /*
125 * XXX We don't use the hash list for divert IP, but it's easier
126 * to allocate a one entry hash list than it is to check all
127 * over the place for hashbase == NULL.
128 */
129 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
130 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
131 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
132 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
133 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
134}
135
136/*
137 * IPPROTO_DIVERT is not in the real IP protocol number space; this
138 * function should never be called. Just in case, drop any packets.
139 */
140void
141div_input(struct mbuf *m, int off)
142{
143 ipstat.ips_noproto++;
144 m_freem(m);
145}
146
147/*
148 * Divert a packet by passing it up to the divert socket at port 'port'.
149 *
150 * Setup generic address and protocol structures for div_input routine,
151 * then pass them along with mbuf chain.
152 */
153void
154divert_packet(struct mbuf *m, int incoming)
155{
156 struct ip *ip;
157 struct inpcb *inp;
158 struct socket *sa;
159 u_int16_t nport;
160 struct sockaddr_in divsrc;
161 struct m_tag *mtag;
162
163 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
164 if (mtag == NULL) {
165 printf("%s: no divert tag\n", __func__);
166 m_freem(m);
167 return;
168 }
169 /* Assure header */
170 if (m->m_len < sizeof(struct ip) &&
171 (m = m_pullup(m, sizeof(struct ip))) == 0)
172 return;
173 ip = mtod(m, struct ip *);
174
175 /*
176 * Record receive interface address, if any.
177 * But only for incoming packets.
178 */
179 bzero(&divsrc, sizeof(divsrc));
180 divsrc.sin_len = sizeof(divsrc);
181 divsrc.sin_family = AF_INET;
182 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
183 if (incoming) {
184 struct ifaddr *ifa;
185
186 /* Sanity check */
187 M_ASSERTPKTHDR(m);
188
189 /* Find IP address for receive interface */
190 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
191 if (ifa->ifa_addr == NULL)
192 continue;
193 if (ifa->ifa_addr->sa_family != AF_INET)
194 continue;
195 divsrc.sin_addr =
196 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
197 break;
198 }
199 }
200 /*
201 * Record the incoming interface name whenever we have one.
202 */
203 if (m->m_pkthdr.rcvif) {
204 /*
205 * Hide the actual interface name in there in the
206 * sin_zero array. XXX This needs to be moved to a
207 * different sockaddr type for divert, e.g.
208 * sockaddr_div with multiple fields like
209 * sockaddr_dl. Presently we have only 7 bytes
210 * but that will do for now as most interfaces
211 * are 4 or less + 2 or less bytes for unit.
212 * There is probably a faster way of doing this,
213 * possibly taking it from the sockaddr_dl on the iface.
214 * This solves the problem of a P2P link and a LAN interface
215 * having the same address, which can result in the wrong
216 * interface being assigned to the packet when fed back
217 * into the divert socket. Theoretically if the daemon saves
218 * and re-uses the sockaddr_in as suggested in the man pages,
219 * this iface name will come along for the ride.
220 * (see div_output for the other half of this.)
221 */
222 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
223 sizeof(divsrc.sin_zero));
224 }
225
226 /*
227 * XXX sbappendaddr must be protected by Giant until
228 * we have locking at the socket layer. When entered
229 * from below we come in w/o Giant and must take it
230 * here. Unfortunately we cannot tell whether we're
231 * entering from above (already holding Giant),
232 * below (potentially without Giant), or otherwise
233 * (e.g. from tcp_syncache through a timeout) so we
234 * have to grab it regardless. This causes a LOR with
235 * the tcp lock, at least, and possibly others. For
236 * the moment we're ignoring this. Once sockets are
237 * locked this cruft can be removed.
238 */
239 mtx_lock(&Giant);
240 /* Put packet on socket queue, if any */
241 sa = NULL;
242 nport = htons((u_int16_t)divert_info(mtag));
243 INP_INFO_RLOCK(&divcbinfo);
244 LIST_FOREACH(inp, &divcb, inp_list) {
245 INP_LOCK(inp);
246 /* XXX why does only one socket match? */
247 if (inp->inp_lport == nport) {
248 sa = inp->inp_socket;
249 if (sbappendaddr(&sa->so_rcv,
250 (struct sockaddr *)&divsrc, m,
251 (struct mbuf *)0) == 0)
252 sa = NULL; /* force mbuf reclaim below */
253 else
254 sorwakeup(sa);
255 INP_UNLOCK(inp);
256 break;
257 }
258 INP_UNLOCK(inp);
259 }
260 INP_INFO_RUNLOCK(&divcbinfo);
261 mtx_unlock(&Giant);
262 if (sa == NULL) {
263 m_freem(m);
264 ipstat.ips_noproto++;
265 ipstat.ips_delivered--;
266 }
267}
268
269/*
270 * Deliver packet back into the IP processing machinery.
271 *
272 * If no address specified, or address is 0.0.0.0, send to ip_output();
273 * otherwise, send to ip_input() and mark as having been received on
274 * the interface with that address.
275 */
276static int
277div_output(struct socket *so, struct mbuf *m,
278 struct sockaddr_in *sin, struct mbuf *control)
279{
280 int error = 0;
281
282 KASSERT(m->m_pkthdr.rcvif == NULL, ("rcvif not null"));
283
284#ifdef MAC
285 mac_create_mbuf_from_socket(so, m);
286#endif
287
288 if (control)
289 m_freem(control); /* XXX */
290
291 /* Loopback avoidance and state recovery */
292 if (sin) {
293 struct m_tag *mtag;
294 struct divert_tag *dt;
295 int i;
296
297 mtag = m_tag_get(PACKET_TAG_DIVERT,
298 sizeof(struct divert_tag), M_NOWAIT);
299 if (mtag == NULL) {
300 error = ENOBUFS;
301 goto cantsend;
302 }
303 dt = (struct divert_tag *)(mtag+1);
304 dt->info = 0;
305 dt->cookie = sin->sin_port;
306 m_tag_prepend(m, mtag);
307
308 /*
309 * Find receive interface with the given name, stuffed
310 * (if it exists) in the sin_zero[] field.
311 * The name is user supplied data so don't trust its size
312 * or that it is zero terminated.
313 */
314 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
315 ;
316 if ( i > 0 && i < sizeof(sin->sin_zero))
317 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
318 }
319
320 /* Reinject packet into the system as incoming or outgoing */
321 if (!sin || sin->sin_addr.s_addr == 0) {
322 struct ip *const ip = mtod(m, struct ip *);
323 struct inpcb *inp;
324
325 INP_INFO_WLOCK(&divcbinfo);
326 inp = sotoinpcb(so);
327 INP_LOCK(inp);
328 /*
329 * Don't allow both user specified and setsockopt options,
330 * and don't allow packet length sizes that will crash
331 */
332 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
333 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
334 error = EINVAL;
335 m_freem(m);
336 } else {
337 /* Convert fields to host order for ip_output() */
338 ip->ip_len = ntohs(ip->ip_len);
339 ip->ip_off = ntohs(ip->ip_off);
340
341 /* Send packet to output processing */
342 ipstat.ips_rawout++; /* XXX */
343
344 error = ip_output(m,
345 inp->inp_options, NULL,
346 (so->so_options & SO_DONTROUTE) |
347 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
348 inp->inp_moptions, NULL);
349 }
350 INP_UNLOCK(inp);
351 INP_INFO_WUNLOCK(&divcbinfo);
352 } else {
353 if (m->m_pkthdr.rcvif == NULL) {
354 /*
355 * No luck with the name, check by IP address.
356 * Clear the port and the ifname to make sure
357 * there are no distractions for ifa_ifwithaddr.
358 */
359 struct ifaddr *ifa;
360
361 bzero(sin->sin_zero, sizeof(sin->sin_zero));
362 sin->sin_port = 0;
363 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
364 if (ifa == NULL) {
365 error = EADDRNOTAVAIL;
366 goto cantsend;
367 }
368 m->m_pkthdr.rcvif = ifa->ifa_ifp;
369 }
370 /* Send packet to input processing */
371 ip_input(m);
372 }
373
374 return error;
375
376cantsend:
377 m_freem(m);
378 return error;
379}
380
381/*
382 * Return a copy of the specified packet, but without
383 * the divert tag. This is used when packets are ``tee'd''
384 * and we want the cloned copy to not have divert processing.
385 */
386struct mbuf *
387divert_clone(struct mbuf *m)
388{
389 struct mbuf *clone;
390 struct m_tag *mtag;
391
392 clone = m_dup(m, M_DONTWAIT);
393 if (clone != NULL) {
394 /* strip divert tag from copy */
395 mtag = m_tag_find(clone, PACKET_TAG_DIVERT, NULL);
396 if (mtag != NULL)
397 m_tag_delete(clone, mtag);
398 }
399 return clone;
400}
401
402static int
403div_attach(struct socket *so, int proto, struct thread *td)
404{
405 struct inpcb *inp;
406 int error;
407
408 INP_INFO_WLOCK(&divcbinfo);
409 inp = sotoinpcb(so);
410 if (inp != 0) {
411 INP_INFO_WUNLOCK(&divcbinfo);
412 return EINVAL;
413 }
414 if (td && (error = suser(td)) != 0) {
415 INP_INFO_WUNLOCK(&divcbinfo);
416 return error;
417 }
418 error = soreserve(so, div_sendspace, div_recvspace);
419 if (error) {
420 INP_INFO_WUNLOCK(&divcbinfo);
421 return error;
422 }
423 error = in_pcballoc(so, &divcbinfo, td, "divinp");
424 if (error) {
425 INP_INFO_WUNLOCK(&divcbinfo);
426 return error;
427 }
428 inp = (struct inpcb *)so->so_pcb;
429 INP_LOCK(inp);
430 INP_INFO_WUNLOCK(&divcbinfo);
431 inp->inp_ip_p = proto;
432 inp->inp_vflag |= INP_IPV4;
433 inp->inp_flags |= INP_HDRINCL;
434 /* The socket is always "connected" because
435 we always know "where" to send the packet */
436 INP_UNLOCK(inp);
437 so->so_state |= SS_ISCONNECTED;
438 return 0;
439}
440
441static int
442div_detach(struct socket *so)
443{
444 struct inpcb *inp;
445
446 INP_INFO_WLOCK(&divcbinfo);
447 inp = sotoinpcb(so);
448 if (inp == 0) {
449 INP_INFO_WUNLOCK(&divcbinfo);
450 return EINVAL;
451 }
452 INP_LOCK(inp);
453 in_pcbdetach(inp);
454 INP_INFO_WUNLOCK(&divcbinfo);
455 return 0;
456}
457
458static int
459div_abort(struct socket *so)
460{
461 struct inpcb *inp;
462
463 INP_INFO_WLOCK(&divcbinfo);
464 inp = sotoinpcb(so);
465 if (inp == 0) {
466 INP_INFO_WUNLOCK(&divcbinfo);
467 return EINVAL; /* ??? possible? panic instead? */
468 }
469 INP_LOCK(inp);
470 soisdisconnected(so);
471 in_pcbdetach(inp);
472 INP_INFO_WUNLOCK(&divcbinfo);
473 return 0;
474}
475
476static int
477div_disconnect(struct socket *so)
478{
479 if ((so->so_state & SS_ISCONNECTED) == 0)
480 return ENOTCONN;
481 return div_abort(so);
482}
483
484static int
485div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
486{
487 struct inpcb *inp;
488 int error;
489
490 INP_INFO_WLOCK(&divcbinfo);
491 inp = sotoinpcb(so);
492 if (inp == 0) {
493 INP_INFO_WUNLOCK(&divcbinfo);
494 return EINVAL;
495 }
496 /* in_pcbbind assumes that nam is a sockaddr_in
497 * and in_pcbbind requires a valid address. Since divert
498 * sockets don't we need to make sure the address is
499 * filled in properly.
500 * XXX -- divert should not be abusing in_pcbind
501 * and should probably have its own family.
502 */
503 if (nam->sa_family != AF_INET)
504 error = EAFNOSUPPORT;
505 else {
506 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
507 INP_LOCK(inp);
508 error = in_pcbbind(inp, nam, td);
509 INP_UNLOCK(inp);
510 }
511 INP_INFO_WUNLOCK(&divcbinfo);
512 return error;
513}
514
515static int
516div_shutdown(struct socket *so)
517{
518 struct inpcb *inp;
519
520 INP_INFO_RLOCK(&divcbinfo);
521 inp = sotoinpcb(so);
522 if (inp == 0) {
523 INP_INFO_RUNLOCK(&divcbinfo);
524 return EINVAL;
525 }
526 INP_LOCK(inp);
527 INP_INFO_RUNLOCK(&divcbinfo);
528 socantsendmore(so);
529 INP_UNLOCK(inp);
530 return 0;
531}
532
533static int
534div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
535 struct mbuf *control, struct thread *td)
536{
537 /* Packet must have a header (but that's about it) */
538 if (m->m_len < sizeof (struct ip) &&
539 (m = m_pullup(m, sizeof (struct ip))) == 0) {
540 ipstat.ips_toosmall++;
541 m_freem(m);
542 return EINVAL;
543 }
544
545 /* Send packet */
546 return div_output(so, m, (struct sockaddr_in *)nam, control);
547}
548
549void
550div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
551{
552 struct in_addr faddr;
553
554 faddr = ((struct sockaddr_in *)sa)->sin_addr;
555 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
556 return;
557 if (PRC_IS_REDIRECT(cmd))
558 return;
559}
560
561static int
562div_pcblist(SYSCTL_HANDLER_ARGS)
563{
564 int error, i, n;
565 struct inpcb *inp, **inp_list;
566 inp_gen_t gencnt;
567 struct xinpgen xig;
568
569 /*
570 * The process of preparing the TCB list is too time-consuming and
571 * resource-intensive to repeat twice on every request.
572 */
573 if (req->oldptr == 0) {
574 n = divcbinfo.ipi_count;
575 req->oldidx = 2 * (sizeof xig)
576 + (n + n/8) * sizeof(struct xinpcb);
577 return 0;
578 }
579
580 if (req->newptr != 0)
581 return EPERM;
582
583 /*
584 * OK, now we're committed to doing something.
585 */
586 INP_INFO_RLOCK(&divcbinfo);
587 gencnt = divcbinfo.ipi_gencnt;
588 n = divcbinfo.ipi_count;
589 INP_INFO_RUNLOCK(&divcbinfo);
590
| 34 */
35
36#include "opt_inet.h"
37#include "opt_ipfw.h"
38#include "opt_ipdivert.h"
39#include "opt_ipsec.h"
40#include "opt_mac.h"
41
42#ifndef INET
43#error "IPDIVERT requires INET."
44#endif
45
46#include <sys/param.h>
47#include <sys/kernel.h>
48#include <sys/lock.h>
49#include <sys/malloc.h>
50#include <sys/mac.h>
51#include <sys/mbuf.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
74/*
75 * Divert sockets
76 */
77
78/*
79 * Allocate enough space to hold a full IP packet
80 */
81#define DIVSNDQ (65536 + 100)
82#define DIVRCVQ (65536 + 100)
83
84/*
85 * Divert sockets work in conjunction with ipfw, see the divert(4)
86 * manpage for features.
87 * Internally, packets selected by ipfw in ip_input() or ip_output(),
88 * and never diverted before, are passed to the input queue of the
89 * divert socket with a given 'divert_port' number (as specified in
90 * the matching ipfw rule), and they are tagged with a 16 bit cookie
91 * (representing the rule number of the matching ipfw rule), which
92 * is passed to process reading from the socket.
93 *
94 * Packets written to the divert socket are again tagged with a cookie
95 * (usually the same as above) and a destination address.
96 * If the destination address is INADDR_ANY then the packet is
97 * treated as outgoing and sent to ip_output(), otherwise it is
98 * treated as incoming and sent to ip_input().
99 * In both cases, the packet is tagged with the cookie.
100 *
101 * On reinjection, processing in ip_input() and ip_output()
102 * will be exactly the same as for the original packet, except that
103 * ipfw processing will start at the rule number after the one
104 * written in the cookie (so, tagging a packet with a cookie of 0
105 * will cause it to be effectively considered as a standard packet).
106 */
107
108/* Internal variables */
109static struct inpcbhead divcb;
110static struct inpcbinfo divcbinfo;
111
112static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
113static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
114
115/*
116 * Initialize divert connection block queue.
117 */
118void
119div_init(void)
120{
121 INP_INFO_LOCK_INIT(&divcbinfo, "div");
122 LIST_INIT(&divcb);
123 divcbinfo.listhead = &divcb;
124 /*
125 * XXX We don't use the hash list for divert IP, but it's easier
126 * to allocate a one entry hash list than it is to check all
127 * over the place for hashbase == NULL.
128 */
129 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
130 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
131 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
132 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
133 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
134}
135
136/*
137 * IPPROTO_DIVERT is not in the real IP protocol number space; this
138 * function should never be called. Just in case, drop any packets.
139 */
140void
141div_input(struct mbuf *m, int off)
142{
143 ipstat.ips_noproto++;
144 m_freem(m);
145}
146
147/*
148 * Divert a packet by passing it up to the divert socket at port 'port'.
149 *
150 * Setup generic address and protocol structures for div_input routine,
151 * then pass them along with mbuf chain.
152 */
153void
154divert_packet(struct mbuf *m, int incoming)
155{
156 struct ip *ip;
157 struct inpcb *inp;
158 struct socket *sa;
159 u_int16_t nport;
160 struct sockaddr_in divsrc;
161 struct m_tag *mtag;
162
163 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
164 if (mtag == NULL) {
165 printf("%s: no divert tag\n", __func__);
166 m_freem(m);
167 return;
168 }
169 /* Assure header */
170 if (m->m_len < sizeof(struct ip) &&
171 (m = m_pullup(m, sizeof(struct ip))) == 0)
172 return;
173 ip = mtod(m, struct ip *);
174
175 /*
176 * Record receive interface address, if any.
177 * But only for incoming packets.
178 */
179 bzero(&divsrc, sizeof(divsrc));
180 divsrc.sin_len = sizeof(divsrc);
181 divsrc.sin_family = AF_INET;
182 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
183 if (incoming) {
184 struct ifaddr *ifa;
185
186 /* Sanity check */
187 M_ASSERTPKTHDR(m);
188
189 /* Find IP address for receive interface */
190 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
191 if (ifa->ifa_addr == NULL)
192 continue;
193 if (ifa->ifa_addr->sa_family != AF_INET)
194 continue;
195 divsrc.sin_addr =
196 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
197 break;
198 }
199 }
200 /*
201 * Record the incoming interface name whenever we have one.
202 */
203 if (m->m_pkthdr.rcvif) {
204 /*
205 * Hide the actual interface name in there in the
206 * sin_zero array. XXX This needs to be moved to a
207 * different sockaddr type for divert, e.g.
208 * sockaddr_div with multiple fields like
209 * sockaddr_dl. Presently we have only 7 bytes
210 * but that will do for now as most interfaces
211 * are 4 or less + 2 or less bytes for unit.
212 * There is probably a faster way of doing this,
213 * possibly taking it from the sockaddr_dl on the iface.
214 * This solves the problem of a P2P link and a LAN interface
215 * having the same address, which can result in the wrong
216 * interface being assigned to the packet when fed back
217 * into the divert socket. Theoretically if the daemon saves
218 * and re-uses the sockaddr_in as suggested in the man pages,
219 * this iface name will come along for the ride.
220 * (see div_output for the other half of this.)
221 */
222 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
223 sizeof(divsrc.sin_zero));
224 }
225
226 /*
227 * XXX sbappendaddr must be protected by Giant until
228 * we have locking at the socket layer. When entered
229 * from below we come in w/o Giant and must take it
230 * here. Unfortunately we cannot tell whether we're
231 * entering from above (already holding Giant),
232 * below (potentially without Giant), or otherwise
233 * (e.g. from tcp_syncache through a timeout) so we
234 * have to grab it regardless. This causes a LOR with
235 * the tcp lock, at least, and possibly others. For
236 * the moment we're ignoring this. Once sockets are
237 * locked this cruft can be removed.
238 */
239 mtx_lock(&Giant);
240 /* Put packet on socket queue, if any */
241 sa = NULL;
242 nport = htons((u_int16_t)divert_info(mtag));
243 INP_INFO_RLOCK(&divcbinfo);
244 LIST_FOREACH(inp, &divcb, inp_list) {
245 INP_LOCK(inp);
246 /* XXX why does only one socket match? */
247 if (inp->inp_lport == nport) {
248 sa = inp->inp_socket;
249 if (sbappendaddr(&sa->so_rcv,
250 (struct sockaddr *)&divsrc, m,
251 (struct mbuf *)0) == 0)
252 sa = NULL; /* force mbuf reclaim below */
253 else
254 sorwakeup(sa);
255 INP_UNLOCK(inp);
256 break;
257 }
258 INP_UNLOCK(inp);
259 }
260 INP_INFO_RUNLOCK(&divcbinfo);
261 mtx_unlock(&Giant);
262 if (sa == NULL) {
263 m_freem(m);
264 ipstat.ips_noproto++;
265 ipstat.ips_delivered--;
266 }
267}
268
269/*
270 * Deliver packet back into the IP processing machinery.
271 *
272 * If no address specified, or address is 0.0.0.0, send to ip_output();
273 * otherwise, send to ip_input() and mark as having been received on
274 * the interface with that address.
275 */
276static int
277div_output(struct socket *so, struct mbuf *m,
278 struct sockaddr_in *sin, struct mbuf *control)
279{
280 int error = 0;
281
282 KASSERT(m->m_pkthdr.rcvif == NULL, ("rcvif not null"));
283
284#ifdef MAC
285 mac_create_mbuf_from_socket(so, m);
286#endif
287
288 if (control)
289 m_freem(control); /* XXX */
290
291 /* Loopback avoidance and state recovery */
292 if (sin) {
293 struct m_tag *mtag;
294 struct divert_tag *dt;
295 int i;
296
297 mtag = m_tag_get(PACKET_TAG_DIVERT,
298 sizeof(struct divert_tag), M_NOWAIT);
299 if (mtag == NULL) {
300 error = ENOBUFS;
301 goto cantsend;
302 }
303 dt = (struct divert_tag *)(mtag+1);
304 dt->info = 0;
305 dt->cookie = sin->sin_port;
306 m_tag_prepend(m, mtag);
307
308 /*
309 * Find receive interface with the given name, stuffed
310 * (if it exists) in the sin_zero[] field.
311 * The name is user supplied data so don't trust its size
312 * or that it is zero terminated.
313 */
314 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
315 ;
316 if ( i > 0 && i < sizeof(sin->sin_zero))
317 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
318 }
319
320 /* Reinject packet into the system as incoming or outgoing */
321 if (!sin || sin->sin_addr.s_addr == 0) {
322 struct ip *const ip = mtod(m, struct ip *);
323 struct inpcb *inp;
324
325 INP_INFO_WLOCK(&divcbinfo);
326 inp = sotoinpcb(so);
327 INP_LOCK(inp);
328 /*
329 * Don't allow both user specified and setsockopt options,
330 * and don't allow packet length sizes that will crash
331 */
332 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
333 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
334 error = EINVAL;
335 m_freem(m);
336 } else {
337 /* Convert fields to host order for ip_output() */
338 ip->ip_len = ntohs(ip->ip_len);
339 ip->ip_off = ntohs(ip->ip_off);
340
341 /* Send packet to output processing */
342 ipstat.ips_rawout++; /* XXX */
343
344 error = ip_output(m,
345 inp->inp_options, NULL,
346 (so->so_options & SO_DONTROUTE) |
347 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
348 inp->inp_moptions, NULL);
349 }
350 INP_UNLOCK(inp);
351 INP_INFO_WUNLOCK(&divcbinfo);
352 } else {
353 if (m->m_pkthdr.rcvif == NULL) {
354 /*
355 * No luck with the name, check by IP address.
356 * Clear the port and the ifname to make sure
357 * there are no distractions for ifa_ifwithaddr.
358 */
359 struct ifaddr *ifa;
360
361 bzero(sin->sin_zero, sizeof(sin->sin_zero));
362 sin->sin_port = 0;
363 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
364 if (ifa == NULL) {
365 error = EADDRNOTAVAIL;
366 goto cantsend;
367 }
368 m->m_pkthdr.rcvif = ifa->ifa_ifp;
369 }
370 /* Send packet to input processing */
371 ip_input(m);
372 }
373
374 return error;
375
376cantsend:
377 m_freem(m);
378 return error;
379}
380
381/*
382 * Return a copy of the specified packet, but without
383 * the divert tag. This is used when packets are ``tee'd''
384 * and we want the cloned copy to not have divert processing.
385 */
386struct mbuf *
387divert_clone(struct mbuf *m)
388{
389 struct mbuf *clone;
390 struct m_tag *mtag;
391
392 clone = m_dup(m, M_DONTWAIT);
393 if (clone != NULL) {
394 /* strip divert tag from copy */
395 mtag = m_tag_find(clone, PACKET_TAG_DIVERT, NULL);
396 if (mtag != NULL)
397 m_tag_delete(clone, mtag);
398 }
399 return clone;
400}
401
402static int
403div_attach(struct socket *so, int proto, struct thread *td)
404{
405 struct inpcb *inp;
406 int error;
407
408 INP_INFO_WLOCK(&divcbinfo);
409 inp = sotoinpcb(so);
410 if (inp != 0) {
411 INP_INFO_WUNLOCK(&divcbinfo);
412 return EINVAL;
413 }
414 if (td && (error = suser(td)) != 0) {
415 INP_INFO_WUNLOCK(&divcbinfo);
416 return error;
417 }
418 error = soreserve(so, div_sendspace, div_recvspace);
419 if (error) {
420 INP_INFO_WUNLOCK(&divcbinfo);
421 return error;
422 }
423 error = in_pcballoc(so, &divcbinfo, td, "divinp");
424 if (error) {
425 INP_INFO_WUNLOCK(&divcbinfo);
426 return error;
427 }
428 inp = (struct inpcb *)so->so_pcb;
429 INP_LOCK(inp);
430 INP_INFO_WUNLOCK(&divcbinfo);
431 inp->inp_ip_p = proto;
432 inp->inp_vflag |= INP_IPV4;
433 inp->inp_flags |= INP_HDRINCL;
434 /* The socket is always "connected" because
435 we always know "where" to send the packet */
436 INP_UNLOCK(inp);
437 so->so_state |= SS_ISCONNECTED;
438 return 0;
439}
440
441static int
442div_detach(struct socket *so)
443{
444 struct inpcb *inp;
445
446 INP_INFO_WLOCK(&divcbinfo);
447 inp = sotoinpcb(so);
448 if (inp == 0) {
449 INP_INFO_WUNLOCK(&divcbinfo);
450 return EINVAL;
451 }
452 INP_LOCK(inp);
453 in_pcbdetach(inp);
454 INP_INFO_WUNLOCK(&divcbinfo);
455 return 0;
456}
457
458static int
459div_abort(struct socket *so)
460{
461 struct inpcb *inp;
462
463 INP_INFO_WLOCK(&divcbinfo);
464 inp = sotoinpcb(so);
465 if (inp == 0) {
466 INP_INFO_WUNLOCK(&divcbinfo);
467 return EINVAL; /* ??? possible? panic instead? */
468 }
469 INP_LOCK(inp);
470 soisdisconnected(so);
471 in_pcbdetach(inp);
472 INP_INFO_WUNLOCK(&divcbinfo);
473 return 0;
474}
475
476static int
477div_disconnect(struct socket *so)
478{
479 if ((so->so_state & SS_ISCONNECTED) == 0)
480 return ENOTCONN;
481 return div_abort(so);
482}
483
484static int
485div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
486{
487 struct inpcb *inp;
488 int error;
489
490 INP_INFO_WLOCK(&divcbinfo);
491 inp = sotoinpcb(so);
492 if (inp == 0) {
493 INP_INFO_WUNLOCK(&divcbinfo);
494 return EINVAL;
495 }
496 /* in_pcbbind assumes that nam is a sockaddr_in
497 * and in_pcbbind requires a valid address. Since divert
498 * sockets don't we need to make sure the address is
499 * filled in properly.
500 * XXX -- divert should not be abusing in_pcbind
501 * and should probably have its own family.
502 */
503 if (nam->sa_family != AF_INET)
504 error = EAFNOSUPPORT;
505 else {
506 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
507 INP_LOCK(inp);
508 error = in_pcbbind(inp, nam, td);
509 INP_UNLOCK(inp);
510 }
511 INP_INFO_WUNLOCK(&divcbinfo);
512 return error;
513}
514
515static int
516div_shutdown(struct socket *so)
517{
518 struct inpcb *inp;
519
520 INP_INFO_RLOCK(&divcbinfo);
521 inp = sotoinpcb(so);
522 if (inp == 0) {
523 INP_INFO_RUNLOCK(&divcbinfo);
524 return EINVAL;
525 }
526 INP_LOCK(inp);
527 INP_INFO_RUNLOCK(&divcbinfo);
528 socantsendmore(so);
529 INP_UNLOCK(inp);
530 return 0;
531}
532
533static int
534div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
535 struct mbuf *control, struct thread *td)
536{
537 /* Packet must have a header (but that's about it) */
538 if (m->m_len < sizeof (struct ip) &&
539 (m = m_pullup(m, sizeof (struct ip))) == 0) {
540 ipstat.ips_toosmall++;
541 m_freem(m);
542 return EINVAL;
543 }
544
545 /* Send packet */
546 return div_output(so, m, (struct sockaddr_in *)nam, control);
547}
548
549void
550div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
551{
552 struct in_addr faddr;
553
554 faddr = ((struct sockaddr_in *)sa)->sin_addr;
555 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
556 return;
557 if (PRC_IS_REDIRECT(cmd))
558 return;
559}
560
561static int
562div_pcblist(SYSCTL_HANDLER_ARGS)
563{
564 int error, i, n;
565 struct inpcb *inp, **inp_list;
566 inp_gen_t gencnt;
567 struct xinpgen xig;
568
569 /*
570 * The process of preparing the TCB list is too time-consuming and
571 * resource-intensive to repeat twice on every request.
572 */
573 if (req->oldptr == 0) {
574 n = divcbinfo.ipi_count;
575 req->oldidx = 2 * (sizeof xig)
576 + (n + n/8) * sizeof(struct xinpcb);
577 return 0;
578 }
579
580 if (req->newptr != 0)
581 return EPERM;
582
583 /*
584 * OK, now we're committed to doing something.
585 */
586 INP_INFO_RLOCK(&divcbinfo);
587 gencnt = divcbinfo.ipi_gencnt;
588 n = divcbinfo.ipi_count;
589 INP_INFO_RUNLOCK(&divcbinfo);
590
|
592
593 xig.xig_len = sizeof xig;
594 xig.xig_count = n;
595 xig.xig_gen = gencnt;
596 xig.xig_sogen = so_gencnt;
597 error = SYSCTL_OUT(req, &xig, sizeof xig);
598 if (error)
599 return error;
600
601 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
602 if (inp_list == 0)
603 return ENOMEM;
604
605 INP_INFO_RLOCK(&divcbinfo);
606 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
607 inp = LIST_NEXT(inp, inp_list)) {
608 INP_LOCK(inp);
609 if (inp->inp_gencnt <= gencnt &&
610 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
611 inp_list[i++] = inp;
612 INP_UNLOCK(inp);
613 }
614 INP_INFO_RUNLOCK(&divcbinfo);
615 n = i;
616
617 error = 0;
618 for (i = 0; i < n; i++) {
619 inp = inp_list[i];
620 if (inp->inp_gencnt <= gencnt) {
621 struct xinpcb xi;
622 xi.xi_len = sizeof xi;
623 /* XXX should avoid extra copy */
624 bcopy(inp, &xi.xi_inp, sizeof *inp);
625 if (inp->inp_socket)
626 sotoxsocket(inp->inp_socket, &xi.xi_socket);
627 error = SYSCTL_OUT(req, &xi, sizeof xi);
628 }
629 }
630 if (!error) {
631 /*
632 * Give the user an updated idea of our state.
633 * If the generation differs from what we told
634 * her before, she knows that something happened
635 * while we were processing this request, and it
636 * might be necessary to retry.
637 */
638 INP_INFO_RLOCK(&divcbinfo);
639 xig.xig_gen = divcbinfo.ipi_gencnt;
640 xig.xig_sogen = so_gencnt;
641 xig.xig_count = divcbinfo.ipi_count;
642 INP_INFO_RUNLOCK(&divcbinfo);
643 error = SYSCTL_OUT(req, &xig, sizeof xig);
644 }
645 free(inp_list, M_TEMP);
646 return error;
647}
648
649/*
650 * This is the wrapper function for in_setsockaddr. We just pass down
651 * the pcbinfo for in_setpeeraddr to lock.
652 */
653static int
654div_sockaddr(struct socket *so, struct sockaddr **nam)
655{
656 return (in_setsockaddr(so, nam, &divcbinfo));
657}
658
659/*
660 * This is the wrapper function for in_setpeeraddr. We just pass down
661 * the pcbinfo for in_setpeeraddr to lock.
662 */
663static int
664div_peeraddr(struct socket *so, struct sockaddr **nam)
665{
666 return (in_setpeeraddr(so, nam, &divcbinfo));
667}
668
669
670SYSCTL_DECL(_net_inet_divert);
671SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
672 div_pcblist, "S,xinpcb", "List of active divert sockets");
673
674struct pr_usrreqs div_usrreqs = {
675 div_abort, pru_accept_notsupp, div_attach, div_bind,
676 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
677 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
678 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
679 div_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel
680};
| 595
596 xig.xig_len = sizeof xig;
597 xig.xig_count = n;
598 xig.xig_gen = gencnt;
599 xig.xig_sogen = so_gencnt;
600 error = SYSCTL_OUT(req, &xig, sizeof xig);
601 if (error)
602 return error;
603
604 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
605 if (inp_list == 0)
606 return ENOMEM;
607
608 INP_INFO_RLOCK(&divcbinfo);
609 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
610 inp = LIST_NEXT(inp, inp_list)) {
611 INP_LOCK(inp);
612 if (inp->inp_gencnt <= gencnt &&
613 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
614 inp_list[i++] = inp;
615 INP_UNLOCK(inp);
616 }
617 INP_INFO_RUNLOCK(&divcbinfo);
618 n = i;
619
620 error = 0;
621 for (i = 0; i < n; i++) {
622 inp = inp_list[i];
623 if (inp->inp_gencnt <= gencnt) {
624 struct xinpcb xi;
625 xi.xi_len = sizeof xi;
626 /* XXX should avoid extra copy */
627 bcopy(inp, &xi.xi_inp, sizeof *inp);
628 if (inp->inp_socket)
629 sotoxsocket(inp->inp_socket, &xi.xi_socket);
630 error = SYSCTL_OUT(req, &xi, sizeof xi);
631 }
632 }
633 if (!error) {
634 /*
635 * Give the user an updated idea of our state.
636 * If the generation differs from what we told
637 * her before, she knows that something happened
638 * while we were processing this request, and it
639 * might be necessary to retry.
640 */
641 INP_INFO_RLOCK(&divcbinfo);
642 xig.xig_gen = divcbinfo.ipi_gencnt;
643 xig.xig_sogen = so_gencnt;
644 xig.xig_count = divcbinfo.ipi_count;
645 INP_INFO_RUNLOCK(&divcbinfo);
646 error = SYSCTL_OUT(req, &xig, sizeof xig);
647 }
648 free(inp_list, M_TEMP);
649 return error;
650}
651
652/*
653 * This is the wrapper function for in_setsockaddr. We just pass down
654 * the pcbinfo for in_setpeeraddr to lock.
655 */
656static int
657div_sockaddr(struct socket *so, struct sockaddr **nam)
658{
659 return (in_setsockaddr(so, nam, &divcbinfo));
660}
661
662/*
663 * This is the wrapper function for in_setpeeraddr. We just pass down
664 * the pcbinfo for in_setpeeraddr to lock.
665 */
666static int
667div_peeraddr(struct socket *so, struct sockaddr **nam)
668{
669 return (in_setpeeraddr(so, nam, &divcbinfo));
670}
671
672
673SYSCTL_DECL(_net_inet_divert);
674SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
675 div_pcblist, "S,xinpcb", "List of active divert sockets");
676
677struct pr_usrreqs div_usrreqs = {
678 div_abort, pru_accept_notsupp, div_attach, div_bind,
679 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
680 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
681 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
682 div_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel
683};
|