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_var.h>
72
73/*
74 * Divert sockets
75 */
76
77/*
78 * Allocate enough space to hold a full IP packet
79 */
80#define DIVSNDQ (65536 + 100)
81#define DIVRCVQ (65536 + 100)
82
83/*
84 * Divert sockets work in conjunction with ipfw, see the divert(4)
85 * manpage for features.
86 * Internally, packets selected by ipfw in ip_input() or ip_output(),
87 * and never diverted before, are passed to the input queue of the
88 * divert socket with a given 'divert_port' number (as specified in
89 * the matching ipfw rule), and they are tagged with a 16 bit cookie
90 * (representing the rule number of the matching ipfw rule), which
91 * is passed to process reading from the socket.
92 *
93 * Packets written to the divert socket are again tagged with a cookie
94 * (usually the same as above) and a destination address.
95 * If the destination address is INADDR_ANY then the packet is
96 * treated as outgoing and sent to ip_output(), otherwise it is
97 * treated as incoming and sent to ip_input().
98 * In both cases, the packet is tagged with the cookie.
99 *
100 * On reinjection, processing in ip_input() and ip_output()
101 * will be exactly the same as for the original packet, except that
102 * ipfw processing will start at the rule number after the one
103 * written in the cookie (so, tagging a packet with a cookie of 0
104 * will cause it to be effectively considered as a standard packet).
105 */
106
107/* Internal variables */
108static struct inpcbhead divcb;
109static struct inpcbinfo divcbinfo;
110
111static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
112static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
113
114/*
115 * Initialize divert connection block queue.
116 */
117void
118div_init(void)
119{
120 INP_INFO_LOCK_INIT(&divcbinfo, "div");
121 LIST_INIT(&divcb);
122 divcbinfo.listhead = &divcb;
123 /*
124 * XXX We don't use the hash list for divert IP, but it's easier
125 * to allocate a one entry hash list than it is to check all
126 * over the place for hashbase == NULL.
127 */
128 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
129 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
130 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
131 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
132 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
133}
134
135/*
136 * IPPROTO_DIVERT is not in the real IP protocol number space; this
137 * function should never be called. Just in case, drop any packets.
138 */
139void
140div_input(struct mbuf *m, int off)
141{
142 ipstat.ips_noproto++;
143 m_freem(m);
144}
145
146/*
147 * Divert a packet by passing it up to the divert socket at port 'port'.
148 *
149 * Setup generic address and protocol structures for div_input routine,
150 * then pass them along with mbuf chain.
151 */
152void
153divert_packet(struct mbuf *m, int incoming, int port, int rule)
154{
155 struct ip *ip;
156 struct inpcb *inp;
157 struct socket *sa;
158 u_int16_t nport;
159 struct sockaddr_in divsrc;
160
161 /* Sanity check */
162 KASSERT(port != 0, ("%s: port=0", __func__));
163
164 /* Assure header */
165 if (m->m_len < sizeof(struct ip) &&
166 (m = m_pullup(m, sizeof(struct ip))) == 0)
167 return;
168 ip = mtod(m, struct ip *);
169
170 /*
171 * Record receive interface address, if any.
172 * But only for incoming packets.
173 */
174 bzero(&divsrc, sizeof(divsrc));
175 divsrc.sin_len = sizeof(divsrc);
176 divsrc.sin_family = AF_INET;
177 divsrc.sin_port = rule; /* record matching rule */
178 if (incoming) {
179 struct ifaddr *ifa;
180
181 /* Sanity check */
182 M_ASSERTPKTHDR(m);
183
184 /* Find IP address for receive interface */
185 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
186 if (ifa->ifa_addr == NULL)
187 continue;
188 if (ifa->ifa_addr->sa_family != AF_INET)
189 continue;
190 divsrc.sin_addr =
191 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
192 break;
193 }
194 }
195 /*
196 * Record the incoming interface name whenever we have one.
197 */
198 if (m->m_pkthdr.rcvif) {
199 /*
200 * Hide the actual interface name in there in the
201 * sin_zero array. XXX This needs to be moved to a
202 * different sockaddr type for divert, e.g.
203 * sockaddr_div with multiple fields like
204 * sockaddr_dl. Presently we have only 7 bytes
205 * but that will do for now as most interfaces
206 * are 4 or less + 2 or less bytes for unit.
207 * There is probably a faster way of doing this,
208 * possibly taking it from the sockaddr_dl on the iface.
209 * This solves the problem of a P2P link and a LAN interface
210 * having the same address, which can result in the wrong
211 * interface being assigned to the packet when fed back
212 * into the divert socket. Theoretically if the daemon saves
213 * and re-uses the sockaddr_in as suggested in the man pages,
214 * this iface name will come along for the ride.
215 * (see div_output for the other half of this.)
216 */
217 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
218 sizeof(divsrc.sin_zero));
219 }
220
221 /*
222 * XXX sbappendaddr must be protected by Giant until
223 * we have locking at the socket layer. When entered
224 * from below we come in w/o Giant and must take it
225 * here. Unfortunately we cannot tell whether we're
226 * entering from above (already holding Giant),
227 * below (potentially without Giant), or otherwise
228 * (e.g. from tcp_syncache through a timeout) so we
229 * have to grab it regardless. This causes a LOR with
230 * the tcp lock, at least, and possibly others. For
231 * the moment we're ignoring this. Once sockets are
232 * locked this cruft can be removed.
233 */
234 mtx_lock(&Giant);
235 /* Put packet on socket queue, if any */
236 sa = NULL;
237 nport = htons((u_int16_t)port);
238 INP_INFO_RLOCK(&divcbinfo);
239 LIST_FOREACH(inp, &divcb, inp_list) {
240 INP_LOCK(inp);
241 /* XXX why does only one socket match? */
242 if (inp->inp_lport == nport) {
243 sa = inp->inp_socket;
244 if (sbappendaddr(&sa->so_rcv,
245 (struct sockaddr *)&divsrc, m,
246 (struct mbuf *)0) == 0)
247 sa = NULL; /* force mbuf reclaim below */
248 else
249 sorwakeup(sa);
250 INP_UNLOCK(inp);
251 break;
252 }
253 INP_UNLOCK(inp);
254 }
255 INP_INFO_RUNLOCK(&divcbinfo);
256 mtx_unlock(&Giant);
257 if (sa == NULL) {
258 m_freem(m);
259 ipstat.ips_noproto++;
260 ipstat.ips_delivered--;
261 }
262}
263
264/*
265 * Deliver packet back into the IP processing machinery.
266 *
267 * If no address specified, or address is 0.0.0.0, send to ip_output();
268 * otherwise, send to ip_input() and mark as having been received on
269 * the interface with that address.
270 */
271static int
272div_output(struct socket *so, struct mbuf *m,
273 struct sockaddr_in *sin, struct mbuf *control)
274{
275 int error = 0;
276 struct m_hdr divert_tag;
277
278 /*
279 * Prepare the tag for divert info. Note that a packet
280 * with a 0 tag in mh_data is effectively untagged,
281 * so we could optimize that case.
282 */
283 divert_tag.mh_type = MT_TAG;
284 divert_tag.mh_flags = PACKET_TAG_DIVERT;
285 divert_tag.mh_next = m;
286 divert_tag.mh_data = 0; /* the matching rule # */
287 divert_tag.mh_nextpkt = NULL;
288 m->m_pkthdr.rcvif = NULL; /* XXX is it necessary ? */
289
290#ifdef MAC
291 mac_create_mbuf_from_socket(so, m);
292#endif
293
294 if (control)
295 m_freem(control); /* XXX */
296
297 /* Loopback avoidance and state recovery */
298 if (sin) {
299 int i;
300
301 divert_tag.mh_data = (caddr_t)(uintptr_t)sin->sin_port;
302 /*
303 * Find receive interface with the given name, stuffed
304 * (if it exists) in the sin_zero[] field.
305 * The name is user supplied data so don't trust its size
306 * or that it is zero terminated.
307 */
308 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
309 ;
310 if ( i > 0 && i < sizeof(sin->sin_zero))
311 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
312 }
313
314 /* Reinject packet into the system as incoming or outgoing */
315 if (!sin || sin->sin_addr.s_addr == 0) {
316 struct ip *const ip = mtod(m, struct ip *);
317 struct inpcb *inp;
318
319 INP_INFO_WLOCK(&divcbinfo);
320 inp = sotoinpcb(so);
321 INP_LOCK(inp);
322 /*
323 * Don't allow both user specified and setsockopt options,
324 * and don't allow packet length sizes that will crash
325 */
326 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
327 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
328 error = EINVAL;
329 m_freem(m);
330 } else {
331 /* Convert fields to host order for ip_output() */
332 ip->ip_len = ntohs(ip->ip_len);
333 ip->ip_off = ntohs(ip->ip_off);
334
335 /* Send packet to output processing */
336 ipstat.ips_rawout++; /* XXX */
337
338 error = ip_output((struct mbuf *)&divert_tag,
339 inp->inp_options, &inp->inp_route,
340 (so->so_options & SO_DONTROUTE) |
341 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
342 inp->inp_moptions, NULL);
343 }
344 INP_UNLOCK(inp);
345 INP_INFO_WUNLOCK(&divcbinfo);
346 } else {
347 if (m->m_pkthdr.rcvif == NULL) {
348 /*
349 * No luck with the name, check by IP address.
350 * Clear the port and the ifname to make sure
351 * there are no distractions for ifa_ifwithaddr.
352 */
353 struct ifaddr *ifa;
354
355 bzero(sin->sin_zero, sizeof(sin->sin_zero));
356 sin->sin_port = 0;
357 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
358 if (ifa == NULL) {
359 error = EADDRNOTAVAIL;
360 goto cantsend;
361 }
362 m->m_pkthdr.rcvif = ifa->ifa_ifp;
363 }
364 /* Send packet to input processing */
365 ip_input((struct mbuf *)&divert_tag);
366 }
367
368 return error;
369
370cantsend:
371 m_freem(m);
372 return error;
373}
374
375static int
376div_attach(struct socket *so, int proto, struct thread *td)
377{
378 struct inpcb *inp;
379 int error;
380
381 INP_INFO_WLOCK(&divcbinfo);
382 inp = sotoinpcb(so);
383 if (inp != 0) {
384 INP_INFO_WUNLOCK(&divcbinfo);
385 return EINVAL;
386 }
387 if (td && (error = suser(td)) != 0) {
388 INP_INFO_WUNLOCK(&divcbinfo);
389 return error;
390 }
391 error = soreserve(so, div_sendspace, div_recvspace);
392 if (error) {
393 INP_INFO_WUNLOCK(&divcbinfo);
394 return error;
395 }
396 error = in_pcballoc(so, &divcbinfo, td);
397 if (error) {
398 INP_INFO_WUNLOCK(&divcbinfo);
399 return error;
400 }
401 inp = (struct inpcb *)so->so_pcb;
402 INP_LOCK(inp);
403 INP_INFO_WUNLOCK(&divcbinfo);
404 inp->inp_ip_p = proto;
405 inp->inp_vflag |= INP_IPV4;
406 inp->inp_flags |= INP_HDRINCL;
407 /* The socket is always "connected" because
408 we always know "where" to send the packet */
409 INP_UNLOCK(inp);
410 so->so_state |= SS_ISCONNECTED;
411 return 0;
412}
413
414static int
415div_detach(struct socket *so)
416{
417 struct inpcb *inp;
418
419 INP_INFO_WLOCK(&divcbinfo);
420 inp = sotoinpcb(so);
421 if (inp == 0) {
422 INP_INFO_WUNLOCK(&divcbinfo);
423 return EINVAL;
424 }
425 INP_LOCK(inp);
426 in_pcbdetach(inp);
427 INP_INFO_WUNLOCK(&divcbinfo);
428 return 0;
429}
430
431static int
432div_abort(struct socket *so)
433{
434 struct inpcb *inp;
435
436 INP_INFO_WLOCK(&divcbinfo);
437 inp = sotoinpcb(so);
438 if (inp == 0) {
439 INP_INFO_WUNLOCK(&divcbinfo);
440 return EINVAL; /* ??? possible? panic instead? */
441 }
442 INP_LOCK(inp);
443 soisdisconnected(so);
444 in_pcbdetach(inp);
445 INP_INFO_WUNLOCK(&divcbinfo);
446 return 0;
447}
448
449static int
450div_disconnect(struct socket *so)
451{
452 if ((so->so_state & SS_ISCONNECTED) == 0)
453 return ENOTCONN;
454 return div_abort(so);
455}
456
457static int
458div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
459{
460 struct inpcb *inp;
461 int error;
462
463 INP_INFO_WLOCK(&divcbinfo);
464 inp = sotoinpcb(so);
465 if (inp == 0) {
466 INP_INFO_WUNLOCK(&divcbinfo);
467 return EINVAL;
468 }
469 /* in_pcbbind assumes that nam is a sockaddr_in
470 * and in_pcbbind requires a valid address. Since divert
471 * sockets don't we need to make sure the address is
472 * filled in properly.
473 * XXX -- divert should not be abusing in_pcbind
474 * and should probably have its own family.
475 */
476 if (nam->sa_family != AF_INET)
477 error = EAFNOSUPPORT;
478 else {
479 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
480 INP_LOCK(inp);
481 error = in_pcbbind(inp, nam, td);
482 INP_UNLOCK(inp);
483 }
484 INP_INFO_WUNLOCK(&divcbinfo);
485 return error;
486}
487
488static int
489div_shutdown(struct socket *so)
490{
491 struct inpcb *inp;
492
493 INP_INFO_RLOCK(&divcbinfo);
494 inp = sotoinpcb(so);
495 if (inp == 0) {
496 INP_INFO_RUNLOCK(&divcbinfo);
497 return EINVAL;
498 }
499 INP_LOCK(inp);
500 INP_INFO_RUNLOCK(&divcbinfo);
501 socantsendmore(so);
502 INP_UNLOCK(inp);
503 return 0;
504}
505
506static int
507div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
508 struct mbuf *control, struct thread *td)
509{
510 /* Packet must have a header (but that's about it) */
511 if (m->m_len < sizeof (struct ip) &&
512 (m = m_pullup(m, sizeof (struct ip))) == 0) {
513 ipstat.ips_toosmall++;
514 m_freem(m);
515 return EINVAL;
516 }
517
518 /* Send packet */
519 return div_output(so, m, (struct sockaddr_in *)nam, control);
520}
521
522void
523div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
524{
525 struct in_addr faddr;
526
527 faddr = ((struct sockaddr_in *)sa)->sin_addr;
528 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
529 return;
530 if (PRC_IS_REDIRECT(cmd)) {
531 /* flush held routes */
532 in_pcbnotifyall(&divcbinfo, faddr,
533 inetctlerrmap[cmd], in_rtchange);
534 }
535}
536
537static int
538div_pcblist(SYSCTL_HANDLER_ARGS)
539{
540 int error, i, n;
541 struct inpcb *inp, **inp_list;
542 inp_gen_t gencnt;
543 struct xinpgen xig;
544
545 /*
546 * The process of preparing the TCB list is too time-consuming and
547 * resource-intensive to repeat twice on every request.
548 */
549 if (req->oldptr == 0) {
550 n = divcbinfo.ipi_count;
551 req->oldidx = 2 * (sizeof xig)
552 + (n + n/8) * sizeof(struct xinpcb);
553 return 0;
554 }
555
556 if (req->newptr != 0)
557 return EPERM;
558
559 /*
560 * OK, now we're committed to doing something.
561 */
562 INP_INFO_RLOCK(&divcbinfo);
563 gencnt = divcbinfo.ipi_gencnt;
564 n = divcbinfo.ipi_count;
565 INP_INFO_RUNLOCK(&divcbinfo);
566
567 sysctl_wire_old_buffer(req, 2 * sizeof(xig) + n*sizeof(struct xinpcb));
568
569 xig.xig_len = sizeof xig;
570 xig.xig_count = n;
571 xig.xig_gen = gencnt;
572 xig.xig_sogen = so_gencnt;
573 error = SYSCTL_OUT(req, &xig, sizeof xig);
574 if (error)
575 return error;
576
577 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
578 if (inp_list == 0)
579 return ENOMEM;
580
581 INP_INFO_RLOCK(&divcbinfo);
582 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
583 inp = LIST_NEXT(inp, inp_list)) {
584 INP_LOCK(inp);
585 if (inp->inp_gencnt <= gencnt &&
586 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
587 inp_list[i++] = inp;
588 INP_UNLOCK(inp);
589 }
590 INP_INFO_RUNLOCK(&divcbinfo);
591 n = i;
592
593 error = 0;
594 for (i = 0; i < n; i++) {
595 inp = inp_list[i];
596 if (inp->inp_gencnt <= gencnt) {
597 struct xinpcb xi;
598 xi.xi_len = sizeof xi;
599 /* XXX should avoid extra copy */
600 bcopy(inp, &xi.xi_inp, sizeof *inp);
601 if (inp->inp_socket)
602 sotoxsocket(inp->inp_socket, &xi.xi_socket);
603 error = SYSCTL_OUT(req, &xi, sizeof xi);
604 }
605 }
606 if (!error) {
607 /*
608 * Give the user an updated idea of our state.
609 * If the generation differs from what we told
610 * her before, she knows that something happened
611 * while we were processing this request, and it
612 * might be necessary to retry.
613 */
614 INP_INFO_RLOCK(&divcbinfo);
615 xig.xig_gen = divcbinfo.ipi_gencnt;
616 xig.xig_sogen = so_gencnt;
617 xig.xig_count = divcbinfo.ipi_count;
618 INP_INFO_RUNLOCK(&divcbinfo);
619 error = SYSCTL_OUT(req, &xig, sizeof xig);
620 }
621 free(inp_list, M_TEMP);
622 return error;
623}
624
625/*
626 * This is the wrapper function for in_setsockaddr. We just pass down
627 * the pcbinfo for in_setpeeraddr to lock.
628 */
629static int
630div_sockaddr(struct socket *so, struct sockaddr **nam)
631{
632 return (in_setsockaddr(so, nam, &divcbinfo));
633}
634
635/*
636 * This is the wrapper function for in_setpeeraddr. We just pass down
637 * the pcbinfo for in_setpeeraddr to lock.
638 */
639static int
640div_peeraddr(struct socket *so, struct sockaddr **nam)
641{
642 return (in_setpeeraddr(so, nam, &divcbinfo));
643}
644
645
646SYSCTL_DECL(_net_inet_divert);
647SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
648 div_pcblist, "S,xinpcb", "List of active divert sockets");
649
650struct pr_usrreqs div_usrreqs = {
651 div_abort, pru_accept_notsupp, div_attach, div_bind,
652 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
653 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
654 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
| 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_var.h>
72
73/*
74 * Divert sockets
75 */
76
77/*
78 * Allocate enough space to hold a full IP packet
79 */
80#define DIVSNDQ (65536 + 100)
81#define DIVRCVQ (65536 + 100)
82
83/*
84 * Divert sockets work in conjunction with ipfw, see the divert(4)
85 * manpage for features.
86 * Internally, packets selected by ipfw in ip_input() or ip_output(),
87 * and never diverted before, are passed to the input queue of the
88 * divert socket with a given 'divert_port' number (as specified in
89 * the matching ipfw rule), and they are tagged with a 16 bit cookie
90 * (representing the rule number of the matching ipfw rule), which
91 * is passed to process reading from the socket.
92 *
93 * Packets written to the divert socket are again tagged with a cookie
94 * (usually the same as above) and a destination address.
95 * If the destination address is INADDR_ANY then the packet is
96 * treated as outgoing and sent to ip_output(), otherwise it is
97 * treated as incoming and sent to ip_input().
98 * In both cases, the packet is tagged with the cookie.
99 *
100 * On reinjection, processing in ip_input() and ip_output()
101 * will be exactly the same as for the original packet, except that
102 * ipfw processing will start at the rule number after the one
103 * written in the cookie (so, tagging a packet with a cookie of 0
104 * will cause it to be effectively considered as a standard packet).
105 */
106
107/* Internal variables */
108static struct inpcbhead divcb;
109static struct inpcbinfo divcbinfo;
110
111static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
112static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
113
114/*
115 * Initialize divert connection block queue.
116 */
117void
118div_init(void)
119{
120 INP_INFO_LOCK_INIT(&divcbinfo, "div");
121 LIST_INIT(&divcb);
122 divcbinfo.listhead = &divcb;
123 /*
124 * XXX We don't use the hash list for divert IP, but it's easier
125 * to allocate a one entry hash list than it is to check all
126 * over the place for hashbase == NULL.
127 */
128 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
129 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
130 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
131 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
132 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
133}
134
135/*
136 * IPPROTO_DIVERT is not in the real IP protocol number space; this
137 * function should never be called. Just in case, drop any packets.
138 */
139void
140div_input(struct mbuf *m, int off)
141{
142 ipstat.ips_noproto++;
143 m_freem(m);
144}
145
146/*
147 * Divert a packet by passing it up to the divert socket at port 'port'.
148 *
149 * Setup generic address and protocol structures for div_input routine,
150 * then pass them along with mbuf chain.
151 */
152void
153divert_packet(struct mbuf *m, int incoming, int port, int rule)
154{
155 struct ip *ip;
156 struct inpcb *inp;
157 struct socket *sa;
158 u_int16_t nport;
159 struct sockaddr_in divsrc;
160
161 /* Sanity check */
162 KASSERT(port != 0, ("%s: port=0", __func__));
163
164 /* Assure header */
165 if (m->m_len < sizeof(struct ip) &&
166 (m = m_pullup(m, sizeof(struct ip))) == 0)
167 return;
168 ip = mtod(m, struct ip *);
169
170 /*
171 * Record receive interface address, if any.
172 * But only for incoming packets.
173 */
174 bzero(&divsrc, sizeof(divsrc));
175 divsrc.sin_len = sizeof(divsrc);
176 divsrc.sin_family = AF_INET;
177 divsrc.sin_port = rule; /* record matching rule */
178 if (incoming) {
179 struct ifaddr *ifa;
180
181 /* Sanity check */
182 M_ASSERTPKTHDR(m);
183
184 /* Find IP address for receive interface */
185 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
186 if (ifa->ifa_addr == NULL)
187 continue;
188 if (ifa->ifa_addr->sa_family != AF_INET)
189 continue;
190 divsrc.sin_addr =
191 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
192 break;
193 }
194 }
195 /*
196 * Record the incoming interface name whenever we have one.
197 */
198 if (m->m_pkthdr.rcvif) {
199 /*
200 * Hide the actual interface name in there in the
201 * sin_zero array. XXX This needs to be moved to a
202 * different sockaddr type for divert, e.g.
203 * sockaddr_div with multiple fields like
204 * sockaddr_dl. Presently we have only 7 bytes
205 * but that will do for now as most interfaces
206 * are 4 or less + 2 or less bytes for unit.
207 * There is probably a faster way of doing this,
208 * possibly taking it from the sockaddr_dl on the iface.
209 * This solves the problem of a P2P link and a LAN interface
210 * having the same address, which can result in the wrong
211 * interface being assigned to the packet when fed back
212 * into the divert socket. Theoretically if the daemon saves
213 * and re-uses the sockaddr_in as suggested in the man pages,
214 * this iface name will come along for the ride.
215 * (see div_output for the other half of this.)
216 */
217 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
218 sizeof(divsrc.sin_zero));
219 }
220
221 /*
222 * XXX sbappendaddr must be protected by Giant until
223 * we have locking at the socket layer. When entered
224 * from below we come in w/o Giant and must take it
225 * here. Unfortunately we cannot tell whether we're
226 * entering from above (already holding Giant),
227 * below (potentially without Giant), or otherwise
228 * (e.g. from tcp_syncache through a timeout) so we
229 * have to grab it regardless. This causes a LOR with
230 * the tcp lock, at least, and possibly others. For
231 * the moment we're ignoring this. Once sockets are
232 * locked this cruft can be removed.
233 */
234 mtx_lock(&Giant);
235 /* Put packet on socket queue, if any */
236 sa = NULL;
237 nport = htons((u_int16_t)port);
238 INP_INFO_RLOCK(&divcbinfo);
239 LIST_FOREACH(inp, &divcb, inp_list) {
240 INP_LOCK(inp);
241 /* XXX why does only one socket match? */
242 if (inp->inp_lport == nport) {
243 sa = inp->inp_socket;
244 if (sbappendaddr(&sa->so_rcv,
245 (struct sockaddr *)&divsrc, m,
246 (struct mbuf *)0) == 0)
247 sa = NULL; /* force mbuf reclaim below */
248 else
249 sorwakeup(sa);
250 INP_UNLOCK(inp);
251 break;
252 }
253 INP_UNLOCK(inp);
254 }
255 INP_INFO_RUNLOCK(&divcbinfo);
256 mtx_unlock(&Giant);
257 if (sa == NULL) {
258 m_freem(m);
259 ipstat.ips_noproto++;
260 ipstat.ips_delivered--;
261 }
262}
263
264/*
265 * Deliver packet back into the IP processing machinery.
266 *
267 * If no address specified, or address is 0.0.0.0, send to ip_output();
268 * otherwise, send to ip_input() and mark as having been received on
269 * the interface with that address.
270 */
271static int
272div_output(struct socket *so, struct mbuf *m,
273 struct sockaddr_in *sin, struct mbuf *control)
274{
275 int error = 0;
276 struct m_hdr divert_tag;
277
278 /*
279 * Prepare the tag for divert info. Note that a packet
280 * with a 0 tag in mh_data is effectively untagged,
281 * so we could optimize that case.
282 */
283 divert_tag.mh_type = MT_TAG;
284 divert_tag.mh_flags = PACKET_TAG_DIVERT;
285 divert_tag.mh_next = m;
286 divert_tag.mh_data = 0; /* the matching rule # */
287 divert_tag.mh_nextpkt = NULL;
288 m->m_pkthdr.rcvif = NULL; /* XXX is it necessary ? */
289
290#ifdef MAC
291 mac_create_mbuf_from_socket(so, m);
292#endif
293
294 if (control)
295 m_freem(control); /* XXX */
296
297 /* Loopback avoidance and state recovery */
298 if (sin) {
299 int i;
300
301 divert_tag.mh_data = (caddr_t)(uintptr_t)sin->sin_port;
302 /*
303 * Find receive interface with the given name, stuffed
304 * (if it exists) in the sin_zero[] field.
305 * The name is user supplied data so don't trust its size
306 * or that it is zero terminated.
307 */
308 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
309 ;
310 if ( i > 0 && i < sizeof(sin->sin_zero))
311 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
312 }
313
314 /* Reinject packet into the system as incoming or outgoing */
315 if (!sin || sin->sin_addr.s_addr == 0) {
316 struct ip *const ip = mtod(m, struct ip *);
317 struct inpcb *inp;
318
319 INP_INFO_WLOCK(&divcbinfo);
320 inp = sotoinpcb(so);
321 INP_LOCK(inp);
322 /*
323 * Don't allow both user specified and setsockopt options,
324 * and don't allow packet length sizes that will crash
325 */
326 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
327 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
328 error = EINVAL;
329 m_freem(m);
330 } else {
331 /* Convert fields to host order for ip_output() */
332 ip->ip_len = ntohs(ip->ip_len);
333 ip->ip_off = ntohs(ip->ip_off);
334
335 /* Send packet to output processing */
336 ipstat.ips_rawout++; /* XXX */
337
338 error = ip_output((struct mbuf *)&divert_tag,
339 inp->inp_options, &inp->inp_route,
340 (so->so_options & SO_DONTROUTE) |
341 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
342 inp->inp_moptions, NULL);
343 }
344 INP_UNLOCK(inp);
345 INP_INFO_WUNLOCK(&divcbinfo);
346 } else {
347 if (m->m_pkthdr.rcvif == NULL) {
348 /*
349 * No luck with the name, check by IP address.
350 * Clear the port and the ifname to make sure
351 * there are no distractions for ifa_ifwithaddr.
352 */
353 struct ifaddr *ifa;
354
355 bzero(sin->sin_zero, sizeof(sin->sin_zero));
356 sin->sin_port = 0;
357 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
358 if (ifa == NULL) {
359 error = EADDRNOTAVAIL;
360 goto cantsend;
361 }
362 m->m_pkthdr.rcvif = ifa->ifa_ifp;
363 }
364 /* Send packet to input processing */
365 ip_input((struct mbuf *)&divert_tag);
366 }
367
368 return error;
369
370cantsend:
371 m_freem(m);
372 return error;
373}
374
375static int
376div_attach(struct socket *so, int proto, struct thread *td)
377{
378 struct inpcb *inp;
379 int error;
380
381 INP_INFO_WLOCK(&divcbinfo);
382 inp = sotoinpcb(so);
383 if (inp != 0) {
384 INP_INFO_WUNLOCK(&divcbinfo);
385 return EINVAL;
386 }
387 if (td && (error = suser(td)) != 0) {
388 INP_INFO_WUNLOCK(&divcbinfo);
389 return error;
390 }
391 error = soreserve(so, div_sendspace, div_recvspace);
392 if (error) {
393 INP_INFO_WUNLOCK(&divcbinfo);
394 return error;
395 }
396 error = in_pcballoc(so, &divcbinfo, td);
397 if (error) {
398 INP_INFO_WUNLOCK(&divcbinfo);
399 return error;
400 }
401 inp = (struct inpcb *)so->so_pcb;
402 INP_LOCK(inp);
403 INP_INFO_WUNLOCK(&divcbinfo);
404 inp->inp_ip_p = proto;
405 inp->inp_vflag |= INP_IPV4;
406 inp->inp_flags |= INP_HDRINCL;
407 /* The socket is always "connected" because
408 we always know "where" to send the packet */
409 INP_UNLOCK(inp);
410 so->so_state |= SS_ISCONNECTED;
411 return 0;
412}
413
414static int
415div_detach(struct socket *so)
416{
417 struct inpcb *inp;
418
419 INP_INFO_WLOCK(&divcbinfo);
420 inp = sotoinpcb(so);
421 if (inp == 0) {
422 INP_INFO_WUNLOCK(&divcbinfo);
423 return EINVAL;
424 }
425 INP_LOCK(inp);
426 in_pcbdetach(inp);
427 INP_INFO_WUNLOCK(&divcbinfo);
428 return 0;
429}
430
431static int
432div_abort(struct socket *so)
433{
434 struct inpcb *inp;
435
436 INP_INFO_WLOCK(&divcbinfo);
437 inp = sotoinpcb(so);
438 if (inp == 0) {
439 INP_INFO_WUNLOCK(&divcbinfo);
440 return EINVAL; /* ??? possible? panic instead? */
441 }
442 INP_LOCK(inp);
443 soisdisconnected(so);
444 in_pcbdetach(inp);
445 INP_INFO_WUNLOCK(&divcbinfo);
446 return 0;
447}
448
449static int
450div_disconnect(struct socket *so)
451{
452 if ((so->so_state & SS_ISCONNECTED) == 0)
453 return ENOTCONN;
454 return div_abort(so);
455}
456
457static int
458div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
459{
460 struct inpcb *inp;
461 int error;
462
463 INP_INFO_WLOCK(&divcbinfo);
464 inp = sotoinpcb(so);
465 if (inp == 0) {
466 INP_INFO_WUNLOCK(&divcbinfo);
467 return EINVAL;
468 }
469 /* in_pcbbind assumes that nam is a sockaddr_in
470 * and in_pcbbind requires a valid address. Since divert
471 * sockets don't we need to make sure the address is
472 * filled in properly.
473 * XXX -- divert should not be abusing in_pcbind
474 * and should probably have its own family.
475 */
476 if (nam->sa_family != AF_INET)
477 error = EAFNOSUPPORT;
478 else {
479 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
480 INP_LOCK(inp);
481 error = in_pcbbind(inp, nam, td);
482 INP_UNLOCK(inp);
483 }
484 INP_INFO_WUNLOCK(&divcbinfo);
485 return error;
486}
487
488static int
489div_shutdown(struct socket *so)
490{
491 struct inpcb *inp;
492
493 INP_INFO_RLOCK(&divcbinfo);
494 inp = sotoinpcb(so);
495 if (inp == 0) {
496 INP_INFO_RUNLOCK(&divcbinfo);
497 return EINVAL;
498 }
499 INP_LOCK(inp);
500 INP_INFO_RUNLOCK(&divcbinfo);
501 socantsendmore(so);
502 INP_UNLOCK(inp);
503 return 0;
504}
505
506static int
507div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
508 struct mbuf *control, struct thread *td)
509{
510 /* Packet must have a header (but that's about it) */
511 if (m->m_len < sizeof (struct ip) &&
512 (m = m_pullup(m, sizeof (struct ip))) == 0) {
513 ipstat.ips_toosmall++;
514 m_freem(m);
515 return EINVAL;
516 }
517
518 /* Send packet */
519 return div_output(so, m, (struct sockaddr_in *)nam, control);
520}
521
522void
523div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
524{
525 struct in_addr faddr;
526
527 faddr = ((struct sockaddr_in *)sa)->sin_addr;
528 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
529 return;
530 if (PRC_IS_REDIRECT(cmd)) {
531 /* flush held routes */
532 in_pcbnotifyall(&divcbinfo, faddr,
533 inetctlerrmap[cmd], in_rtchange);
534 }
535}
536
537static int
538div_pcblist(SYSCTL_HANDLER_ARGS)
539{
540 int error, i, n;
541 struct inpcb *inp, **inp_list;
542 inp_gen_t gencnt;
543 struct xinpgen xig;
544
545 /*
546 * The process of preparing the TCB list is too time-consuming and
547 * resource-intensive to repeat twice on every request.
548 */
549 if (req->oldptr == 0) {
550 n = divcbinfo.ipi_count;
551 req->oldidx = 2 * (sizeof xig)
552 + (n + n/8) * sizeof(struct xinpcb);
553 return 0;
554 }
555
556 if (req->newptr != 0)
557 return EPERM;
558
559 /*
560 * OK, now we're committed to doing something.
561 */
562 INP_INFO_RLOCK(&divcbinfo);
563 gencnt = divcbinfo.ipi_gencnt;
564 n = divcbinfo.ipi_count;
565 INP_INFO_RUNLOCK(&divcbinfo);
566
567 sysctl_wire_old_buffer(req, 2 * sizeof(xig) + n*sizeof(struct xinpcb));
568
569 xig.xig_len = sizeof xig;
570 xig.xig_count = n;
571 xig.xig_gen = gencnt;
572 xig.xig_sogen = so_gencnt;
573 error = SYSCTL_OUT(req, &xig, sizeof xig);
574 if (error)
575 return error;
576
577 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
578 if (inp_list == 0)
579 return ENOMEM;
580
581 INP_INFO_RLOCK(&divcbinfo);
582 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
583 inp = LIST_NEXT(inp, inp_list)) {
584 INP_LOCK(inp);
585 if (inp->inp_gencnt <= gencnt &&
586 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
587 inp_list[i++] = inp;
588 INP_UNLOCK(inp);
589 }
590 INP_INFO_RUNLOCK(&divcbinfo);
591 n = i;
592
593 error = 0;
594 for (i = 0; i < n; i++) {
595 inp = inp_list[i];
596 if (inp->inp_gencnt <= gencnt) {
597 struct xinpcb xi;
598 xi.xi_len = sizeof xi;
599 /* XXX should avoid extra copy */
600 bcopy(inp, &xi.xi_inp, sizeof *inp);
601 if (inp->inp_socket)
602 sotoxsocket(inp->inp_socket, &xi.xi_socket);
603 error = SYSCTL_OUT(req, &xi, sizeof xi);
604 }
605 }
606 if (!error) {
607 /*
608 * Give the user an updated idea of our state.
609 * If the generation differs from what we told
610 * her before, she knows that something happened
611 * while we were processing this request, and it
612 * might be necessary to retry.
613 */
614 INP_INFO_RLOCK(&divcbinfo);
615 xig.xig_gen = divcbinfo.ipi_gencnt;
616 xig.xig_sogen = so_gencnt;
617 xig.xig_count = divcbinfo.ipi_count;
618 INP_INFO_RUNLOCK(&divcbinfo);
619 error = SYSCTL_OUT(req, &xig, sizeof xig);
620 }
621 free(inp_list, M_TEMP);
622 return error;
623}
624
625/*
626 * This is the wrapper function for in_setsockaddr. We just pass down
627 * the pcbinfo for in_setpeeraddr to lock.
628 */
629static int
630div_sockaddr(struct socket *so, struct sockaddr **nam)
631{
632 return (in_setsockaddr(so, nam, &divcbinfo));
633}
634
635/*
636 * This is the wrapper function for in_setpeeraddr. We just pass down
637 * the pcbinfo for in_setpeeraddr to lock.
638 */
639static int
640div_peeraddr(struct socket *so, struct sockaddr **nam)
641{
642 return (in_setpeeraddr(so, nam, &divcbinfo));
643}
644
645
646SYSCTL_DECL(_net_inet_divert);
647SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
648 div_pcblist, "S,xinpcb", "List of active divert sockets");
649
650struct pr_usrreqs div_usrreqs = {
651 div_abort, pru_accept_notsupp, div_attach, div_bind,
652 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
653 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
654 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
|