xref: /freebsd/sys/netinet/ip_divert.c (revision c17d43407fe04133a94055b0dbc7ea8965654a9f)
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/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/proc.h>
50 #include <sys/socket.h>
51 #include <sys/protosw.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
54 #include <sys/systm.h>
55 
56 #include <vm/uma.h>
57 
58 #include <net/if.h>
59 #include <net/route.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 
68 /*
69  * Divert sockets
70  */
71 
72 /*
73  * Allocate enough space to hold a full IP packet
74  */
75 #define	DIVSNDQ		(65536 + 100)
76 #define	DIVRCVQ		(65536 + 100)
77 
78 /*
79  * A 16 bit cookie is passed to and from the user process.
80  * The user process can send it back to help the caller know
81  * something about where the packet originally came from.
82  *
83  * In the case of ipfw, then the cookie is the rule that sent
84  * us here. On reinjection is is the rule after which processing
85  * should continue. Leaving it the same will make processing start
86  * at the rule number after that which sent it here. Setting it to
87  * 0 will restart processing at the beginning.
88  *
89  * For divert_packet(), ip_divert_cookie is an input value only.
90  * For div_output(), ip_divert_cookie is an output value only.
91  */
92 u_int16_t ip_divert_cookie;
93 
94 /* Internal variables */
95 static struct inpcbhead divcb;
96 static struct inpcbinfo divcbinfo;
97 
98 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
99 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
100 
101 /* Optimization: have this preinitialized */
102 static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET };
103 
104 /* Internal functions */
105 static int div_output(struct socket *so,
106 		struct mbuf *m, struct sockaddr *addr, struct mbuf *control);
107 
108 /*
109  * Initialize divert connection block queue.
110  */
111 void
112 div_init(void)
113 {
114 	LIST_INIT(&divcb);
115 	divcbinfo.listhead = &divcb;
116 	/*
117 	 * XXX We don't use the hash list for divert IP, but it's easier
118 	 * to allocate a one entry hash list than it is to check all
119 	 * over the place for hashbase == NULL.
120 	 */
121 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
122 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
123 	divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
124 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
125 	uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
126 }
127 
128 /*
129  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
130  * with that protocol number to enter the system from the outside.
131  */
132 void
133 div_input(struct mbuf *m, int off)
134 {
135 	ipstat.ips_noproto++;
136 	m_freem(m);
137 }
138 
139 /*
140  * Divert a packet by passing it up to the divert socket at port 'port'.
141  *
142  * Setup generic address and protocol structures for div_input routine,
143  * then pass them along with mbuf chain.
144  */
145 void
146 divert_packet(struct mbuf *m, int incoming, int port)
147 {
148 	struct ip *ip;
149 	struct inpcb *inp;
150 	struct socket *sa;
151 	u_int16_t nport;
152 
153 	/* Sanity check */
154 	KASSERT(port != 0, ("%s: port=0", __func__));
155 
156 	/* Record and reset divert cookie */
157 	divsrc.sin_port = ip_divert_cookie;
158 	ip_divert_cookie = 0;
159 
160 	/* Assure header */
161 	if (m->m_len < sizeof(struct ip) &&
162 	    (m = m_pullup(m, sizeof(struct ip))) == 0) {
163 		return;
164 	}
165 	ip = mtod(m, struct ip *);
166 
167 	/*
168 	 * Record receive interface address, if any.
169 	 * But only for incoming packets.
170 	 */
171 	divsrc.sin_addr.s_addr = 0;
172 	if (incoming) {
173 		struct ifaddr *ifa;
174 
175 		/* Sanity check */
176 		KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __func__));
177 
178 		/* Find IP address for receive interface */
179 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
180 			if (ifa->ifa_addr == NULL)
181 				continue;
182 			if (ifa->ifa_addr->sa_family != AF_INET)
183 				continue;
184 			divsrc.sin_addr =
185 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
186 			break;
187 		}
188 	}
189 	/*
190 	 * Record the incoming interface name whenever we have one.
191 	 */
192 	bzero(&divsrc.sin_zero, sizeof(divsrc.sin_zero));
193 	if (m->m_pkthdr.rcvif) {
194 		/*
195 		 * Hide the actual interface name in there in the
196 		 * sin_zero array. XXX This needs to be moved to a
197 		 * different sockaddr type for divert, e.g.
198 		 * sockaddr_div with multiple fields like
199 		 * sockaddr_dl. Presently we have only 7 bytes
200 		 * but that will do for now as most interfaces
201 		 * are 4 or less + 2 or less bytes for unit.
202 		 * There is probably a faster way of doing this,
203 		 * possibly taking it from the sockaddr_dl on the iface.
204 		 * This solves the problem of a P2P link and a LAN interface
205 		 * having the same address, which can result in the wrong
206 		 * interface being assigned to the packet when fed back
207 		 * into the divert socket. Theoretically if the daemon saves
208 		 * and re-uses the sockaddr_in as suggested in the man pages,
209 		 * this iface name will come along for the ride.
210 		 * (see div_output for the other half of this.)
211 		 */
212 		snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero),
213 			"%s%d", m->m_pkthdr.rcvif->if_name,
214 			m->m_pkthdr.rcvif->if_unit);
215 	}
216 
217 	/* Put packet on socket queue, if any */
218 	sa = NULL;
219 	nport = htons((u_int16_t)port);
220 	LIST_FOREACH(inp, &divcb, inp_list) {
221 		if (inp->inp_lport == nport)
222 			sa = inp->inp_socket;
223 	}
224 	if (sa) {
225 		if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc,
226 				m, (struct mbuf *)0) == 0)
227 			m_freem(m);
228 		else
229 			sorwakeup(sa);
230 	} else {
231 		m_freem(m);
232 		ipstat.ips_noproto++;
233 		ipstat.ips_delivered--;
234         }
235 }
236 
237 /*
238  * Deliver packet back into the IP processing machinery.
239  *
240  * If no address specified, or address is 0.0.0.0, send to ip_output();
241  * otherwise, send to ip_input() and mark as having been received on
242  * the interface with that address.
243  */
244 static int
245 div_output(so, m, addr, control)
246 	struct socket *so;
247 	register struct mbuf *m;
248 	struct sockaddr *addr;
249 	struct mbuf *control;
250 {
251 	register struct inpcb *const inp = sotoinpcb(so);
252 	register struct ip *const ip = mtod(m, struct ip *);
253 	struct sockaddr_in *sin = (struct sockaddr_in *)addr;
254 	int error = 0;
255 
256 	if (control)
257 		m_freem(control);		/* XXX */
258 
259 	/* Loopback avoidance and state recovery */
260 	if (sin) {
261 		int	len = 0;
262 		char	*c = sin->sin_zero;
263 
264 		ip_divert_cookie = sin->sin_port;
265 
266 		/*
267 		 * Find receive interface with the given name or IP address.
268 		 * The name is user supplied data so don't trust it's size or
269 		 * that it is zero terminated. The name has priority.
270 		 * We are presently assuming that the sockaddr_in
271 		 * has not been replaced by a sockaddr_div, so we limit it
272 		 * to 16 bytes in total. the name is stuffed (if it exists)
273 		 * in the sin_zero[] field.
274 		 */
275 		while (*c++ && (len++ < sizeof(sin->sin_zero)));
276 		if ((len > 0) && (len < sizeof(sin->sin_zero)))
277 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
278 	} else {
279 		ip_divert_cookie = 0;
280 	}
281 
282 	/* Reinject packet into the system as incoming or outgoing */
283 	if (!sin || sin->sin_addr.s_addr == 0) {
284 		/*
285 		 * Don't allow both user specified and setsockopt options,
286 		 * and don't allow packet length sizes that will crash
287 		 */
288 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
289 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
290 			error = EINVAL;
291 			goto cantsend;
292 		}
293 
294 		/* Convert fields to host order for ip_output() */
295 		ip->ip_len = ntohs(ip->ip_len);
296 		ip->ip_off = ntohs(ip->ip_off);
297 
298 		/* Send packet to output processing */
299 		ipstat.ips_rawout++;			/* XXX */
300 		error = ip_output(m, inp->inp_options, &inp->inp_route,
301 			(so->so_options & SO_DONTROUTE) |
302 			IP_ALLOWBROADCAST | IP_RAWOUTPUT,
303 			inp->inp_moptions);
304 	} else {
305 		struct	ifaddr *ifa;
306 
307 		/* If no luck with the name above. check by IP address.  */
308 		if (m->m_pkthdr.rcvif == NULL) {
309 			/*
310 			 * Make sure there are no distractions
311 			 * for ifa_ifwithaddr. Clear the port and the ifname.
312 			 * Maybe zap all 8 bytes at once using a 64bit write?
313 			 */
314 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
315 			/* *((u_int64_t *)sin->sin_zero) = 0; */ /* XXX ?? */
316 			sin->sin_port = 0;
317 			if (!(ifa = ifa_ifwithaddr((struct sockaddr *) sin))) {
318 				error = EADDRNOTAVAIL;
319 				goto cantsend;
320 			}
321 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
322 		}
323 
324 		/* Send packet to input processing */
325 		ip_input(m);
326 	}
327 
328 	/* paranoid: Reset for next time (and other packets) */
329 	/* almost definitly already done in the ipfw filter but.. */
330 	ip_divert_cookie = 0;
331 	return error;
332 
333 cantsend:
334 	m_freem(m);
335 	ip_divert_cookie = 0;
336 	return error;
337 }
338 
339 static int
340 div_attach(struct socket *so, int proto, struct thread *td)
341 {
342 	struct inpcb *inp;
343 	int error, s;
344 
345 	inp  = sotoinpcb(so);
346 	if (inp)
347 		panic("div_attach");
348 	if (td && (error = suser_td(td)) != 0)
349 		return error;
350 
351 	error = soreserve(so, div_sendspace, div_recvspace);
352 	if (error)
353 		return error;
354 	s = splnet();
355 	error = in_pcballoc(so, &divcbinfo, td);
356 	splx(s);
357 	if (error)
358 		return error;
359 	inp = (struct inpcb *)so->so_pcb;
360 	inp->inp_ip_p = proto;
361 	inp->inp_vflag |= INP_IPV4;
362 	inp->inp_flags |= INP_HDRINCL;
363 	/* The socket is always "connected" because
364 	   we always know "where" to send the packet */
365 	so->so_state |= SS_ISCONNECTED;
366 	return 0;
367 }
368 
369 static int
370 div_detach(struct socket *so)
371 {
372 	struct inpcb *inp;
373 
374 	inp = sotoinpcb(so);
375 	if (inp == 0)
376 		panic("div_detach");
377 	in_pcbdetach(inp);
378 	return 0;
379 }
380 
381 static int
382 div_abort(struct socket *so)
383 {
384 	soisdisconnected(so);
385 	return div_detach(so);
386 }
387 
388 static int
389 div_disconnect(struct socket *so)
390 {
391 	if ((so->so_state & SS_ISCONNECTED) == 0)
392 		return ENOTCONN;
393 	return div_abort(so);
394 }
395 
396 static int
397 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
398 {
399 	struct inpcb *inp;
400 	int s;
401 	int error;
402 
403 	s = splnet();
404 	inp = sotoinpcb(so);
405 	/* in_pcbbind assumes that the socket is a sockaddr_in
406 	 * and in_pcbbind requires a valid address. Since divert
407 	 * sockets don't we need to make sure the address is
408 	 * filled in properly.
409 	 * XXX -- divert should not be abusing in_pcbind
410 	 * and should probably have its own family.
411 	 */
412 	if (nam->sa_family != AF_INET) {
413 		error = EAFNOSUPPORT;
414 	} else {
415 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
416 		error = in_pcbbind(inp, nam, td);
417 	}
418 	splx(s);
419 	return error;
420 }
421 
422 static int
423 div_shutdown(struct socket *so)
424 {
425 	socantsendmore(so);
426 	return 0;
427 }
428 
429 static int
430 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
431 	 struct mbuf *control, struct thread *td)
432 {
433 	/* Packet must have a header (but that's about it) */
434 	if (m->m_len < sizeof (struct ip) &&
435 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
436 		ipstat.ips_toosmall++;
437 		m_freem(m);
438 		return EINVAL;
439 	}
440 
441 	/* Send packet */
442 	return div_output(so, m, nam, control);
443 }
444 
445 static int
446 div_pcblist(SYSCTL_HANDLER_ARGS)
447 {
448 	int error, i, n, s;
449 	struct inpcb *inp, **inp_list;
450 	inp_gen_t gencnt;
451 	struct xinpgen xig;
452 
453 	/*
454 	 * The process of preparing the TCB list is too time-consuming and
455 	 * resource-intensive to repeat twice on every request.
456 	 */
457 	if (req->oldptr == 0) {
458 		n = divcbinfo.ipi_count;
459 		req->oldidx = 2 * (sizeof xig)
460 			+ (n + n/8) * sizeof(struct xinpcb);
461 		return 0;
462 	}
463 
464 	if (req->newptr != 0)
465 		return EPERM;
466 
467 	/*
468 	 * OK, now we're committed to doing something.
469 	 */
470 	s = splnet();
471 	gencnt = divcbinfo.ipi_gencnt;
472 	n = divcbinfo.ipi_count;
473 	splx(s);
474 
475 	xig.xig_len = sizeof xig;
476 	xig.xig_count = n;
477 	xig.xig_gen = gencnt;
478 	xig.xig_sogen = so_gencnt;
479 	error = SYSCTL_OUT(req, &xig, sizeof xig);
480 	if (error)
481 		return error;
482 
483 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
484 	if (inp_list == 0)
485 		return ENOMEM;
486 
487 	s = splnet();
488 	for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
489 	     inp = LIST_NEXT(inp, inp_list)) {
490 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(
491 		    req->td->td_proc, inp))
492 			inp_list[i++] = inp;
493 	}
494 	splx(s);
495 	n = i;
496 
497 	error = 0;
498 	for (i = 0; i < n; i++) {
499 		inp = inp_list[i];
500 		if (inp->inp_gencnt <= gencnt) {
501 			struct xinpcb xi;
502 			xi.xi_len = sizeof xi;
503 			/* XXX should avoid extra copy */
504 			bcopy(inp, &xi.xi_inp, sizeof *inp);
505 			if (inp->inp_socket)
506 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
507 			error = SYSCTL_OUT(req, &xi, sizeof xi);
508 		}
509 	}
510 	if (!error) {
511 		/*
512 		 * Give the user an updated idea of our state.
513 		 * If the generation differs from what we told
514 		 * her before, she knows that something happened
515 		 * while we were processing this request, and it
516 		 * might be necessary to retry.
517 		 */
518 		s = splnet();
519 		xig.xig_gen = divcbinfo.ipi_gencnt;
520 		xig.xig_sogen = so_gencnt;
521 		xig.xig_count = divcbinfo.ipi_count;
522 		splx(s);
523 		error = SYSCTL_OUT(req, &xig, sizeof xig);
524 	}
525 	free(inp_list, M_TEMP);
526 	return error;
527 }
528 
529 SYSCTL_DECL(_net_inet_divert);
530 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
531 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
532 
533 struct pr_usrreqs div_usrreqs = {
534 	div_abort, pru_accept_notsupp, div_attach, div_bind,
535 	pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
536 	div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp,
537 	pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
538 	in_setsockaddr, sosend, soreceive, sopoll
539 };
540