xref: /freebsd/sys/netinet/ip_divert.c (revision 3ff369fed2a08f32dda232c10470b949bef9489f)
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  *
33  * $FreeBSD$
34  */
35 
36 #include "opt_inet.h"
37 #include "opt_ipfw.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipsec.h"
40 
41 #ifndef INET
42 #error "IPDIVERT requires INET."
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/proc.h>
51 #include <sys/protosw.h>
52 #include <sys/signalvar.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sx.h>
56 #include <sys/sysctl.h>
57 #include <sys/systm.h>
58 
59 #include <vm/uma.h>
60 
61 #include <net/if.h>
62 #include <net/route.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip_var.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 /*
82  * A 16 bit cookie is passed to and from the user process.
83  * The user process can send it back to help the caller know
84  * something about where the packet originally came from.
85  *
86  * In the case of ipfw, then the cookie is the rule that sent
87  * us here. On reinjection is is the rule after which processing
88  * should continue. Leaving it the same will make processing start
89  * at the rule number after that which sent it here. Setting it to
90  * 0 will restart processing at the beginning.
91  *
92  * For divert_packet(), ip_divert_cookie is an input value only.
93  * For div_output(), ip_divert_cookie is an output value only.
94  */
95 u_int16_t ip_divert_cookie;
96 
97 /* Internal variables */
98 static struct inpcbhead divcb;
99 static struct inpcbinfo divcbinfo;
100 
101 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
102 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
103 
104 /* Optimization: have this preinitialized */
105 static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET };
106 
107 /* Internal functions */
108 static int div_output(struct socket *so,
109 		struct mbuf *m, struct sockaddr *addr, struct mbuf *control);
110 
111 /*
112  * Initialize divert connection block queue.
113  */
114 void
115 div_init(void)
116 {
117 	INP_INFO_LOCK_INIT(&divcbinfo, "div");
118 	LIST_INIT(&divcb);
119 	divcbinfo.listhead = &divcb;
120 	/*
121 	 * XXX We don't use the hash list for divert IP, but it's easier
122 	 * to allocate a one entry hash list than it is to check all
123 	 * over the place for hashbase == NULL.
124 	 */
125 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
126 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
127 	divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
128 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
129 	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
130 }
131 
132 /*
133  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
134  * with that protocol number to enter the system from the outside.
135  */
136 void
137 div_input(struct mbuf *m, int off)
138 {
139 	ipstat.ips_noproto++;
140 	m_freem(m);
141 }
142 
143 /*
144  * Divert a packet by passing it up to the divert socket at port 'port'.
145  *
146  * Setup generic address and protocol structures for div_input routine,
147  * then pass them along with mbuf chain.
148  */
149 void
150 divert_packet(struct mbuf *m, int incoming, int port)
151 {
152 	struct ip *ip;
153 	struct inpcb *inp;
154 	struct socket *sa;
155 	u_int16_t nport;
156 
157 	/* Sanity check */
158 	KASSERT(port != 0, ("%s: port=0", __func__));
159 
160 	/* Record and reset divert cookie */
161 	divsrc.sin_port = ip_divert_cookie;
162 	ip_divert_cookie = 0;
163 
164 	/* Assure header */
165 	if (m->m_len < sizeof(struct ip) &&
166 	    (m = m_pullup(m, sizeof(struct ip))) == 0) {
167 		return;
168 	}
169 	ip = mtod(m, struct ip *);
170 
171 	/*
172 	 * Record receive interface address, if any.
173 	 * But only for incoming packets.
174 	 */
175 	divsrc.sin_addr.s_addr = 0;
176 	if (incoming) {
177 		struct ifaddr *ifa;
178 
179 		/* Sanity check */
180 		KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __func__));
181 
182 		/* Find IP address for receive interface */
183 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
184 			if (ifa->ifa_addr == NULL)
185 				continue;
186 			if (ifa->ifa_addr->sa_family != AF_INET)
187 				continue;
188 			divsrc.sin_addr =
189 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
190 			break;
191 		}
192 	}
193 	/*
194 	 * Record the incoming interface name whenever we have one.
195 	 */
196 	bzero(&divsrc.sin_zero, sizeof(divsrc.sin_zero));
197 	if (m->m_pkthdr.rcvif) {
198 		/*
199 		 * Hide the actual interface name in there in the
200 		 * sin_zero array. XXX This needs to be moved to a
201 		 * different sockaddr type for divert, e.g.
202 		 * sockaddr_div with multiple fields like
203 		 * sockaddr_dl. Presently we have only 7 bytes
204 		 * but that will do for now as most interfaces
205 		 * are 4 or less + 2 or less bytes for unit.
206 		 * There is probably a faster way of doing this,
207 		 * possibly taking it from the sockaddr_dl on the iface.
208 		 * This solves the problem of a P2P link and a LAN interface
209 		 * having the same address, which can result in the wrong
210 		 * interface being assigned to the packet when fed back
211 		 * into the divert socket. Theoretically if the daemon saves
212 		 * and re-uses the sockaddr_in as suggested in the man pages,
213 		 * this iface name will come along for the ride.
214 		 * (see div_output for the other half of this.)
215 		 */
216 		snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero),
217 			"%s%d", m->m_pkthdr.rcvif->if_name,
218 			m->m_pkthdr.rcvif->if_unit);
219 	}
220 
221 	/* Put packet on socket queue, if any */
222 	sa = NULL;
223 	nport = htons((u_int16_t)port);
224 	LIST_FOREACH(inp, &divcb, inp_list) {
225 		if (inp->inp_lport == nport)
226 			sa = inp->inp_socket;
227 	}
228 	if (sa) {
229 		if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc,
230 				m, (struct mbuf *)0) == 0)
231 			m_freem(m);
232 		else
233 			sorwakeup(sa);
234 	} else {
235 		m_freem(m);
236 		ipstat.ips_noproto++;
237 		ipstat.ips_delivered--;
238         }
239 }
240 
241 /*
242  * Deliver packet back into the IP processing machinery.
243  *
244  * If no address specified, or address is 0.0.0.0, send to ip_output();
245  * otherwise, send to ip_input() and mark as having been received on
246  * the interface with that address.
247  */
248 static int
249 div_output(so, m, addr, control)
250 	struct socket *so;
251 	register struct mbuf *m;
252 	struct sockaddr *addr;
253 	struct mbuf *control;
254 {
255 	register struct inpcb *const inp = sotoinpcb(so);
256 	register struct ip *const ip = mtod(m, struct ip *);
257 	struct sockaddr_in *sin = (struct sockaddr_in *)addr;
258 	int error = 0;
259 
260 	if (control)
261 		m_freem(control);		/* XXX */
262 
263 	/* Loopback avoidance and state recovery */
264 	if (sin) {
265 		int	len = 0;
266 		char	*c = sin->sin_zero;
267 
268 		ip_divert_cookie = sin->sin_port;
269 
270 		/*
271 		 * Find receive interface with the given name or IP address.
272 		 * The name is user supplied data so don't trust it's size or
273 		 * that it is zero terminated. The name has priority.
274 		 * We are presently assuming that the sockaddr_in
275 		 * has not been replaced by a sockaddr_div, so we limit it
276 		 * to 16 bytes in total. the name is stuffed (if it exists)
277 		 * in the sin_zero[] field.
278 		 */
279 		while (*c++ && (len++ < sizeof(sin->sin_zero)));
280 		if ((len > 0) && (len < sizeof(sin->sin_zero)))
281 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
282 	} else {
283 		ip_divert_cookie = 0;
284 	}
285 
286 	/* Reinject packet into the system as incoming or outgoing */
287 	if (!sin || sin->sin_addr.s_addr == 0) {
288 		/*
289 		 * Don't allow both user specified and setsockopt options,
290 		 * and don't allow packet length sizes that will crash
291 		 */
292 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
293 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
294 			error = EINVAL;
295 			goto cantsend;
296 		}
297 
298 		/* Convert fields to host order for ip_output() */
299 		ip->ip_len = ntohs(ip->ip_len);
300 		ip->ip_off = ntohs(ip->ip_off);
301 
302 		/* Send packet to output processing */
303 		ipstat.ips_rawout++;			/* XXX */
304 		error = ip_output(m, inp->inp_options, &inp->inp_route,
305 			(so->so_options & SO_DONTROUTE) |
306 			IP_ALLOWBROADCAST | IP_RAWOUTPUT,
307 			inp->inp_moptions);
308 	} else {
309 		struct	ifaddr *ifa;
310 
311 		/* If no luck with the name above. check by IP address.  */
312 		if (m->m_pkthdr.rcvif == NULL) {
313 			/*
314 			 * Make sure there are no distractions
315 			 * for ifa_ifwithaddr. Clear the port and the ifname.
316 			 * Maybe zap all 8 bytes at once using a 64bit write?
317 			 */
318 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
319 			/* *((u_int64_t *)sin->sin_zero) = 0; */ /* XXX ?? */
320 			sin->sin_port = 0;
321 			if (!(ifa = ifa_ifwithaddr((struct sockaddr *) sin))) {
322 				error = EADDRNOTAVAIL;
323 				goto cantsend;
324 			}
325 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
326 		}
327 
328 		/* Send packet to input processing */
329 		ip_input(m);
330 	}
331 
332 	/* paranoid: Reset for next time (and other packets) */
333 	/* almost definitly already done in the ipfw filter but.. */
334 	ip_divert_cookie = 0;
335 	return error;
336 
337 cantsend:
338 	m_freem(m);
339 	ip_divert_cookie = 0;
340 	return error;
341 }
342 
343 static int
344 div_attach(struct socket *so, int proto, struct thread *td)
345 {
346 	struct inpcb *inp;
347 	int error, s;
348 
349 	inp  = sotoinpcb(so);
350 	if (inp)
351 		panic("div_attach");
352 	if (td && (error = suser(td)) != 0)
353 		return error;
354 
355 	error = soreserve(so, div_sendspace, div_recvspace);
356 	if (error)
357 		return error;
358 	s = splnet();
359 	error = in_pcballoc(so, &divcbinfo, td);
360 	splx(s);
361 	if (error)
362 		return error;
363 	inp = (struct inpcb *)so->so_pcb;
364 	inp->inp_ip_p = proto;
365 	inp->inp_vflag |= INP_IPV4;
366 	inp->inp_flags |= INP_HDRINCL;
367 	/* The socket is always "connected" because
368 	   we always know "where" to send the packet */
369 	so->so_state |= SS_ISCONNECTED;
370 	return 0;
371 }
372 
373 static int
374 div_detach(struct socket *so)
375 {
376 	struct inpcb *inp;
377 
378 	inp = sotoinpcb(so);
379 	if (inp == 0)
380 		panic("div_detach");
381 	in_pcbdetach(inp);
382 	return 0;
383 }
384 
385 static int
386 div_abort(struct socket *so)
387 {
388 	soisdisconnected(so);
389 	return div_detach(so);
390 }
391 
392 static int
393 div_disconnect(struct socket *so)
394 {
395 	if ((so->so_state & SS_ISCONNECTED) == 0)
396 		return ENOTCONN;
397 	return div_abort(so);
398 }
399 
400 static int
401 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
402 {
403 	struct inpcb *inp;
404 	int s;
405 	int error;
406 
407 	s = splnet();
408 	inp = sotoinpcb(so);
409 	/* in_pcbbind assumes that the socket is a sockaddr_in
410 	 * and in_pcbbind requires a valid address. Since divert
411 	 * sockets don't we need to make sure the address is
412 	 * filled in properly.
413 	 * XXX -- divert should not be abusing in_pcbind
414 	 * and should probably have its own family.
415 	 */
416 	if (nam->sa_family != AF_INET) {
417 		error = EAFNOSUPPORT;
418 	} else {
419 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
420 		error = in_pcbbind(inp, nam, td);
421 	}
422 	splx(s);
423 	return error;
424 }
425 
426 static int
427 div_shutdown(struct socket *so)
428 {
429 	socantsendmore(so);
430 	return 0;
431 }
432 
433 static int
434 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
435 	 struct mbuf *control, struct thread *td)
436 {
437 	/* Packet must have a header (but that's about it) */
438 	if (m->m_len < sizeof (struct ip) &&
439 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
440 		ipstat.ips_toosmall++;
441 		m_freem(m);
442 		return EINVAL;
443 	}
444 
445 	/* Send packet */
446 	return div_output(so, m, nam, control);
447 }
448 
449 static int
450 div_pcblist(SYSCTL_HANDLER_ARGS)
451 {
452 	int error, i, n, s;
453 	struct inpcb *inp, **inp_list;
454 	inp_gen_t gencnt;
455 	struct xinpgen xig;
456 
457 	/*
458 	 * The process of preparing the TCB list is too time-consuming and
459 	 * resource-intensive to repeat twice on every request.
460 	 */
461 	if (req->oldptr == 0) {
462 		n = divcbinfo.ipi_count;
463 		req->oldidx = 2 * (sizeof xig)
464 			+ (n + n/8) * sizeof(struct xinpcb);
465 		return 0;
466 	}
467 
468 	if (req->newptr != 0)
469 		return EPERM;
470 
471 	/*
472 	 * OK, now we're committed to doing something.
473 	 */
474 	s = splnet();
475 	gencnt = divcbinfo.ipi_gencnt;
476 	n = divcbinfo.ipi_count;
477 	splx(s);
478 
479 	xig.xig_len = sizeof xig;
480 	xig.xig_count = n;
481 	xig.xig_gen = gencnt;
482 	xig.xig_sogen = so_gencnt;
483 	error = SYSCTL_OUT(req, &xig, sizeof xig);
484 	if (error)
485 		return error;
486 
487 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
488 	if (inp_list == 0)
489 		return ENOMEM;
490 
491 	s = splnet();
492 	for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
493 	     inp = LIST_NEXT(inp, inp_list)) {
494 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->td, inp))
495 			inp_list[i++] = inp;
496 	}
497 	splx(s);
498 	n = i;
499 
500 	error = 0;
501 	for (i = 0; i < n; i++) {
502 		inp = inp_list[i];
503 		if (inp->inp_gencnt <= gencnt) {
504 			struct xinpcb xi;
505 			xi.xi_len = sizeof xi;
506 			/* XXX should avoid extra copy */
507 			bcopy(inp, &xi.xi_inp, sizeof *inp);
508 			if (inp->inp_socket)
509 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
510 			error = SYSCTL_OUT(req, &xi, sizeof xi);
511 		}
512 	}
513 	if (!error) {
514 		/*
515 		 * Give the user an updated idea of our state.
516 		 * If the generation differs from what we told
517 		 * her before, she knows that something happened
518 		 * while we were processing this request, and it
519 		 * might be necessary to retry.
520 		 */
521 		s = splnet();
522 		xig.xig_gen = divcbinfo.ipi_gencnt;
523 		xig.xig_sogen = so_gencnt;
524 		xig.xig_count = divcbinfo.ipi_count;
525 		splx(s);
526 		error = SYSCTL_OUT(req, &xig, sizeof xig);
527 	}
528 	free(inp_list, M_TEMP);
529 	return error;
530 }
531 
532 /*
533  * This is the wrapper function for in_setsockaddr.  We just pass down
534  * the pcbinfo for in_setpeeraddr to lock.
535  */
536 static int
537 div_sockaddr(struct socket *so, struct sockaddr **nam)
538 {
539 	return (in_setsockaddr(so, nam, &divcbinfo));
540 }
541 
542 /*
543  * This is the wrapper function for in_setpeeraddr. We just pass down
544  * the pcbinfo for in_setpeeraddr to lock.
545  */
546 static int
547 div_peeraddr(struct socket *so, struct sockaddr **nam)
548 {
549 	return (in_setpeeraddr(so, nam, &divcbinfo));
550 }
551 
552 
553 SYSCTL_DECL(_net_inet_divert);
554 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
555 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
556 
557 struct pr_usrreqs div_usrreqs = {
558 	div_abort, pru_accept_notsupp, div_attach, div_bind,
559 	pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
560 	div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
561 	pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
562 	div_sockaddr, sosend, soreceive, sopoll
563 };
564