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. 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: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
30 */
31
32 #define _IP_VHL
33
34 #include "opt_inet.h"
35 #include "opt_ipdivert.h"
36
37 #ifndef INET
38 #error "IPDIVERT requires INET."
39 #endif
40
41 #include <sys/param.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/socket.h>
46 #include <sys/protosw.h>
47 #include <sys/socketvar.h>
48 #include <sys/socketvar2.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/proc.h>
52 #include <sys/caps.h>
53 #include <sys/in_cksum.h>
54 #include <sys/lock.h>
55 #include <sys/msgport.h>
56
57 #include <net/if.h>
58 #include <net/route.h>
59
60 #include <net/netmsg2.h>
61 #include <net/netisr2.h>
62
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_divert.h>
70
71 /*
72 * Divert sockets
73 */
74
75 /*
76 * Allocate enough space to hold a full IP packet
77 */
78 #define DIVSNDQ (65536 + 100)
79 #define DIVRCVQ (65536 + 100)
80
81 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
82
83 #define DIV_OUTPUT 0x10000
84 #define DIV_INPUT 0x20000
85
86 /*
87 * Divert sockets work in conjunction with ipfw, see the divert(4)
88 * manpage for features.
89 * Internally, packets selected by ipfw in ip_input() or ip_output(),
90 * and never diverted before, are passed to the input queue of the
91 * divert socket with a given 'divert_port' number (as specified in
92 * the matching ipfw rule), and they are tagged with a 16 bit cookie
93 * (representing the rule number of the matching ipfw rule), which
94 * is passed to process reading from the socket.
95 *
96 * Packets written to the divert socket are again tagged with a cookie
97 * (usually the same as above) and a destination address.
98 * If the destination address is INADDR_ANY then the packet is
99 * treated as outgoing and sent to ip_output(), otherwise it is
100 * treated as incoming and sent to ip_input().
101 * In both cases, the packet is tagged with the cookie.
102 *
103 * On reinjection, processing in ip_input() and ip_output()
104 * will be exactly the same as for the original packet, except that
105 * ipfw processing will start at the rule number after the one
106 * written in the cookie (so, tagging a packet with a cookie of 0
107 * will cause it to be effectively considered as a standard packet).
108 */
109
110 /* Internal variables */
111 static struct inpcbinfo divcbinfo;
112 static struct inpcbportinfo divcbportinfo;
113
114 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
115 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
116
117 static struct mbuf *ip_divert(struct mbuf *, int, int);
118
119 /*
120 * Initialize divert connection block queue.
121 */
122 void
div_init(void)123 div_init(void)
124 {
125 in_pcbinfo_init(&divcbinfo, 0, FALSE);
126 in_pcbportinfo_init(&divcbportinfo, 1, 0);
127 /*
128 * XXX We don't use the hash list for divert IP, but it's easier
129 * to allocate a one entry hash list than it is to check all
130 * over the place for hashbase == NULL.
131 */
132 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
133 in_pcbportinfo_set(&divcbinfo, &divcbportinfo, 1);
134 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
135 &divcbinfo.wildcardhashmask);
136 divcbinfo.ipi_size = sizeof(struct inpcb);
137 ip_divert_p = ip_divert;
138 }
139
140 /*
141 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
142 * with that protocol number to enter the system from the outside.
143 */
144 int
div_input(struct mbuf ** mp,int * offp,int proto)145 div_input(struct mbuf **mp, int *offp, int proto)
146 {
147 struct mbuf *m = *mp;
148
149 ipstat.ips_noproto++;
150 m_freem(m);
151 return(IPPROTO_DONE);
152 }
153
154 /*
155 * Divert a packet by passing it up to the divert socket at port 'port'.
156 *
157 * Setup generic address and protocol structures for div_input routine,
158 * then pass them along with mbuf chain.
159 */
160 static void
div_packet(struct mbuf * m,int incoming,int port)161 div_packet(struct mbuf *m, int incoming, int port)
162 {
163 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
164 struct inpcb *inp;
165 struct socket *sa;
166 struct m_tag *mtag;
167 struct divert_info *divinfo;
168 u_int16_t nport;
169
170 ASSERT_NETISR0;
171
172 /* Locate the divert info */
173 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
174 divinfo = m_tag_data(mtag);
175 divsrc.sin_port = divinfo->skipto;
176
177 /*
178 * Record receive interface address, if any.
179 * But only for incoming packets.
180 */
181 divsrc.sin_addr.s_addr = 0;
182 if (incoming) {
183 struct ifaddr_container *ifac;
184
185 /* Find IP address for receive interface */
186 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
187 ifa_link) {
188 struct ifaddr *ifa = ifac->ifa;
189
190 if (ifa->ifa_addr == NULL)
191 continue;
192 if (ifa->ifa_addr->sa_family != AF_INET)
193 continue;
194 divsrc.sin_addr =
195 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
196 break;
197 }
198 }
199 /*
200 * Record the incoming interface name whenever we have one.
201 */
202 if (m->m_pkthdr.rcvif) {
203 /*
204 * Hide the actual interface name in there in the
205 * sin_zero array. XXX This needs to be moved to a
206 * different sockaddr type for divert, e.g.
207 * sockaddr_div with multiple fields like
208 * sockaddr_dl. Presently we have only 7 bytes
209 * but that will do for now as most interfaces
210 * are 4 or less + 2 or less bytes for unit.
211 * There is probably a faster way of doing this,
212 * possibly taking it from the sockaddr_dl on the iface.
213 * This solves the problem of a P2P link and a LAN interface
214 * having the same address, which can result in the wrong
215 * interface being assigned to the packet when fed back
216 * into the divert socket. Theoretically if the daemon saves
217 * and re-uses the sockaddr_in as suggested in the man pages,
218 * this iface name will come along for the ride.
219 * (see div_output for the other half of this.)
220 */
221 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero, "%s",
222 m->m_pkthdr.rcvif->if_xname);
223 }
224
225 /* Put packet on socket queue, if any */
226 sa = NULL;
227 nport = htons((u_int16_t)port);
228
229 /*
230 * Following loop to locate the inpcb is MPSAFE since the inpcb
231 * insertion/removal happens on the same CPU (CPU0).
232 */
233 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
234 if (inp->inp_flags & INP_PLACEMARKER)
235 continue;
236 if (inp->inp_lport == nport)
237 sa = inp->inp_socket;
238 }
239 if (sa) {
240 lwkt_gettoken(&sa->so_rcv.ssb_token);
241 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0) {
242 m_freem(m);
243 soroverflow(sa);
244 } else
245 sorwakeup(sa);
246 lwkt_reltoken(&sa->so_rcv.ssb_token);
247 } else {
248 m_freem(m);
249 ipstat.ips_noproto++;
250 ipstat.ips_delivered--;
251 }
252 }
253
254 static void
div_packet_handler(netmsg_t msg)255 div_packet_handler(netmsg_t msg)
256 {
257 struct mbuf *m;
258 int port, incoming = 0;
259
260 m = msg->packet.nm_packet;
261
262 port = msg->lmsg.u.ms_result32 & 0xffff;
263 if (msg->lmsg.u.ms_result32 & DIV_INPUT)
264 incoming = 1;
265 div_packet(m, incoming, port);
266 /* no reply, msg embedded in mbuf */
267 }
268
269 static void
divert_packet(struct mbuf * m,int incoming)270 divert_packet(struct mbuf *m, int incoming)
271 {
272 struct m_tag *mtag;
273 struct divert_info *divinfo;
274 int port;
275
276 M_ASSERTPKTHDR(m);
277
278 /* Assure header */
279 if (m->m_len < sizeof(struct ip) &&
280 (m = m_pullup(m, sizeof(struct ip))) == NULL)
281 return;
282
283 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
284 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
285 divinfo = m_tag_data(mtag);
286
287 port = divinfo->port;
288 KASSERT(port != 0, ("%s: port=0", __func__));
289
290 if (mycpuid != 0) {
291 struct netmsg_packet *nmp;
292
293 nmp = &m->m_hdr.mh_netmsg;
294 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
295 0, div_packet_handler);
296 nmp->nm_packet = m;
297
298 nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
299 if (incoming)
300 nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
301 else
302 nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
303
304 lwkt_sendmsg(netisr_cpuport(0), &nmp->base.lmsg);
305 } else {
306 div_packet(m, incoming, port);
307 }
308 }
309
310 /*
311 * Deliver packet back into the IP processing machinery.
312 *
313 * If no address specified, or address is 0.0.0.0, send to ip_output();
314 * otherwise, send to ip_input() and mark as having been received on
315 * the interface with that address.
316 */
317 static int
div_output(struct socket * so,struct mbuf * m,struct sockaddr_in * sin,struct mbuf * control)318 div_output(struct socket *so, struct mbuf *m,
319 struct sockaddr_in *sin, struct mbuf *control)
320 {
321 int error = 0;
322 struct m_tag *mtag;
323 struct divert_info *divinfo;
324
325 ASSERT_NETISR0;
326
327 if (control)
328 m_freem(control); /* XXX */
329
330 /*
331 * Prepare the tag for divert info. Note that a packet
332 * with a 0 tag in mh_data is effectively untagged,
333 * so we could optimize that case.
334 */
335 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), M_NOWAIT);
336 if (mtag == NULL) {
337 error = ENOBUFS;
338 goto cantsend;
339 }
340 m_tag_prepend(m, mtag);
341
342 /* Loopback avoidance and state recovery */
343 divinfo = m_tag_data(mtag);
344 if (sin)
345 divinfo->skipto = sin->sin_port;
346 else
347 divinfo->skipto = 0;
348
349 /* Reinject packet into the system as incoming or outgoing */
350 if (DIV_IS_OUTPUT(sin)) {
351 struct ip *const ip = mtod(m, struct ip *);
352
353 /* Don't allow packet length sizes that will crash */
354 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
355 error = EINVAL;
356 goto cantsend;
357 }
358
359 /* Send packet to output processing */
360 ipstat.ips_rawout++; /* XXX */
361 error = ip_output(m, NULL, NULL,
362 (so->so_options & SO_DONTROUTE) |
363 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
364 NULL, NULL);
365 } else {
366 ip_input(m);
367 }
368 return error;
369
370 cantsend:
371 m_freem(m);
372 return error;
373 }
374
375 static void
div_attach(netmsg_t msg)376 div_attach(netmsg_t msg)
377 {
378 struct socket *so = msg->attach.base.nm_so;
379 int proto = msg->attach.nm_proto;
380 struct pru_attach_info *ai = msg->attach.nm_ai;
381 struct inpcb *inp;
382 int error;
383
384 ASSERT_NETISR0;
385
386 inp = so->so_pcb;
387 if (inp)
388 panic("div_attach");
389 error = caps_priv_check(ai->p_ucred, SYSCAP_RESTRICTEDROOT |
390 __SYSCAP_NULLCRED);
391 if (error)
392 goto out;
393
394 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
395 if (error)
396 goto out;
397 error = in_pcballoc(so, &divcbinfo);
398 if (error)
399 goto out;
400 inp = (struct inpcb *)so->so_pcb;
401 inp->inp_ip_p = proto;
402 inp->inp_flags |= INP_HDRINCL;
403 /*
404 * The socket is always "connected" because
405 * we always know "where" to send the packet.
406 */
407 sosetstate(so, SS_ISCONNECTED);
408 error = 0;
409 out:
410 lwkt_replymsg(&msg->attach.base.lmsg, error);
411 }
412
413 static void
div_detach(netmsg_t msg)414 div_detach(netmsg_t msg)
415 {
416 struct socket *so = msg->detach.base.nm_so;
417 struct inpcb *inp;
418
419 ASSERT_NETISR0;
420
421 inp = so->so_pcb;
422 if (inp == NULL)
423 panic("div_detach");
424 in_pcbdetach(inp);
425 lwkt_replymsg(&msg->detach.base.lmsg, 0);
426 }
427
428 static void
div_abort(netmsg_t msg)429 div_abort(netmsg_t msg)
430 {
431 /*
432 * Divert socket does not support listen(2),
433 * so this should never be called.
434 */
435 panic("div_abort is called");
436 }
437
438 static void
div_disconnect(netmsg_t msg)439 div_disconnect(netmsg_t msg)
440 {
441 struct socket *so = msg->disconnect.base.nm_so;
442 int error;
443
444 ASSERT_NETISR0;
445
446 if (so->so_state & SS_ISCONNECTED) {
447 soisdisconnected(so);
448 error = 0;
449 } else {
450 error = ENOTCONN;
451 }
452 lwkt_replymsg(&msg->disconnect.base.lmsg, error);
453 }
454
455 static void
div_bind(netmsg_t msg)456 div_bind(netmsg_t msg)
457 {
458 struct socket *so = msg->bind.base.nm_so;
459 struct sockaddr *nam = msg->bind.nm_nam;
460 int error;
461
462 ASSERT_NETISR0;
463
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 error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
477 }
478 lwkt_replymsg(&msg->bind.base.lmsg, error);
479 }
480
481 static void
div_shutdown(netmsg_t msg)482 div_shutdown(netmsg_t msg)
483 {
484 struct socket *so = msg->shutdown.base.nm_so;
485
486 ASSERT_NETISR0;
487
488 socantsendmore(so);
489
490 lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
491 }
492
493 static void
div_send(netmsg_t msg)494 div_send(netmsg_t msg)
495 {
496 struct socket *so = msg->send.base.nm_so;
497 struct mbuf *m = msg->send.nm_m;
498 struct sockaddr *nam = msg->send.nm_addr;
499 struct mbuf *control = msg->send.nm_control;
500 int error;
501
502 /* Length check already done in ip_hashfn() */
503 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
504
505 /* Send packet */
506 error = div_output(so, m, (struct sockaddr_in *)nam, control);
507 lwkt_replymsg(&msg->send.base.lmsg, error);
508 }
509
510 SYSCTL_DECL(_net_inet_divert);
511 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 1,
512 in_pcblist_range, "S,xinpcb", "List of active divert sockets");
513
514 struct pr_usrreqs div_usrreqs = {
515 .pru_abort = div_abort,
516 .pru_accept = pr_generic_notsupp,
517 .pru_attach = div_attach,
518 .pru_bind = div_bind,
519 .pru_connect = pr_generic_notsupp,
520 .pru_connect2 = pr_generic_notsupp,
521 .pru_control = in_control_dispatch,
522 .pru_detach = div_detach,
523 .pru_disconnect = div_disconnect,
524 .pru_listen = pr_generic_notsupp,
525 .pru_peeraddr = in_setpeeraddr_dispatch,
526 .pru_rcvd = pr_generic_notsupp,
527 .pru_rcvoob = pr_generic_notsupp,
528 .pru_send = div_send,
529 .pru_sense = pru_sense_null,
530 .pru_shutdown = div_shutdown,
531 .pru_sockaddr = in_setsockaddr_dispatch,
532 .pru_sosend = sosend,
533 .pru_soreceive = soreceive
534 };
535
536 static struct mbuf *
ip_divert_out(struct mbuf * m,int tee)537 ip_divert_out(struct mbuf *m, int tee)
538 {
539 struct mbuf *clone = NULL;
540 //struct ip *ip = mtod(m, struct ip *);
541
542 /* Clone packet if we're doing a 'tee' */
543 if (tee)
544 clone = m_dup(m, M_NOWAIT);
545
546 /*
547 * XXX
548 * delayed checksums are not currently compatible
549 * with divert sockets.
550 */
551 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
552 in_delayed_cksum(m);
553 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
554 }
555
556 /* Deliver packet to divert input routine */
557 divert_packet(m, 0);
558
559 /* If 'tee', continue with original packet */
560 return clone;
561 }
562
563 static struct mbuf *
ip_divert_in(struct mbuf * m,int tee)564 ip_divert_in(struct mbuf *m, int tee)
565 {
566 struct mbuf *clone = NULL;
567 struct ip *ip = mtod(m, struct ip *);
568 struct m_tag *mtag;
569
570 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
571 const struct divert_info *divinfo;
572 u_short frag_off;
573 int hlen;
574
575 /*
576 * Only trust divert info in the fragment
577 * at offset 0.
578 */
579 frag_off = ntohs(ip->ip_off) << 3;
580 if (frag_off != 0) {
581 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
582 m_tag_delete(m, mtag);
583 }
584
585 /*
586 * Attempt reassembly; if it succeeds, proceed.
587 * ip_reass() will return a different mbuf.
588 */
589 m = ip_reass(m);
590 if (m == NULL)
591 return NULL;
592 ip = mtod(m, struct ip *);
593
594 /* Caller need to redispatch the packet, if it is for us */
595 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
596
597 /*
598 * Get the header length of the reassembled
599 * packet
600 */
601 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
602
603 /*
604 * Restore original checksum before diverting
605 * packet
606 */
607 ip->ip_len = htons(ntohs(ip->ip_len) + hlen);
608 ip->ip_sum = 0;
609 if (hlen == sizeof(struct ip))
610 ip->ip_sum = in_cksum_hdr(ip);
611 else
612 ip->ip_sum = in_cksum(m, hlen);
613
614 /*
615 * Only use the saved divert info
616 */
617 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
618 if (mtag == NULL) {
619 /* Wrongly configured ipfw */
620 kprintf("ip_input no divert info\n");
621 m_freem(m);
622 return NULL;
623 }
624 divinfo = m_tag_data(mtag);
625 tee = divinfo->tee;
626 }
627
628 /*
629 * Divert or tee packet to the divert protocol if
630 * required.
631 */
632
633 /* Clone packet if we're doing a 'tee' */
634 if (tee)
635 clone = m_dup(m, M_NOWAIT);
636
637 /* Deliver packet to divert input routine */
638 divert_packet(m, 1);
639
640 /* Catch invalid reference */
641 m = NULL;
642 ip = NULL;
643
644 ipstat.ips_delivered++;
645
646 /* If 'tee', continue with original packet */
647 if (clone != NULL) {
648 /*
649 * Complete processing of the packet.
650 * XXX Better safe than sorry, remove the DIVERT tag.
651 */
652 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
653 KKASSERT(mtag != NULL);
654 m_tag_delete(clone, mtag);
655 }
656 return clone;
657 }
658
659 static struct mbuf *
ip_divert(struct mbuf * m,int tee,int incoming)660 ip_divert(struct mbuf *m, int tee, int incoming)
661 {
662 struct mbuf *ret;
663
664 if (incoming)
665 ret = ip_divert_in(m, tee);
666 else
667 ret = ip_divert_out(m, tee);
668 return ret;
669 }
670