FreeBSD/Linux Kernel Cross Reference
sys/netinet/in_pcb.c
1 /*-
2 * Copyright (c) 1982, 1986, 1991, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2007-2009 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Juniper Networks, Inc.
6 * All rights reserved.
7 *
8 * Portions of this software were developed by Robert N. M. Watson under
9 * contract to Juniper Networks, Inc.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
36 */
37
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD: releng/9.0/sys/netinet/in_pcb.c 227428 2011-11-10 20:28:30Z trociny $");
40
41 #include "opt_ddb.h"
42 #include "opt_ipsec.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_pcbgroup.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/callout.h>
52 #include <sys/domain.h>
53 #include <sys/protosw.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/refcount.h>
59 #include <sys/jail.h>
60 #include <sys/kernel.h>
61 #include <sys/sysctl.h>
62
63 #ifdef DDB
64 #include <ddb/ddb.h>
65 #endif
66
67 #include <vm/uma.h>
68
69 #include <net/if.h>
70 #include <net/if_types.h>
71 #include <net/route.h>
72 #include <net/vnet.h>
73
74 #if defined(INET) || defined(INET6)
75 #include <netinet/in.h>
76 #include <netinet/in_pcb.h>
77 #include <netinet/ip_var.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet/udp.h>
80 #include <netinet/udp_var.h>
81 #endif
82 #ifdef INET
83 #include <netinet/in_var.h>
84 #endif
85 #ifdef INET6
86 #include <netinet/ip6.h>
87 #include <netinet6/in6_pcb.h>
88 #include <netinet6/in6_var.h>
89 #include <netinet6/ip6_var.h>
90 #endif /* INET6 */
91
92
93 #ifdef IPSEC
94 #include <netipsec/ipsec.h>
95 #include <netipsec/key.h>
96 #endif /* IPSEC */
97
98 #include <security/mac/mac_framework.h>
99
100 static struct callout ipport_tick_callout;
101
102 /*
103 * These configure the range of local port addresses assigned to
104 * "unspecified" outgoing connections/packets/whatever.
105 */
106 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
107 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
108 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
109 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
110 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
111 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
112
113 /*
114 * Reserved ports accessible only to root. There are significant
115 * security considerations that must be accounted for when changing these,
116 * but the security benefits can be great. Please be careful.
117 */
118 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
119 VNET_DEFINE(int, ipport_reservedlow);
120
121 /* Variables dealing with random ephemeral port allocation. */
122 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
123 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
124 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
125 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
126 VNET_DEFINE(int, ipport_tcpallocs);
127 static VNET_DEFINE(int, ipport_tcplastcount);
128
129 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
130
131 static void in_pcbremlists(struct inpcb *inp);
132 #ifdef INET
133 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
134 struct in_addr faddr, u_int fport_arg,
135 struct in_addr laddr, u_int lport_arg,
136 int lookupflags, struct ifnet *ifp);
137
138 #define RANGECHK(var, min, max) \
139 if ((var) < (min)) { (var) = (min); } \
140 else if ((var) > (max)) { (var) = (max); }
141
142 static int
143 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
144 {
145 int error;
146
147 #ifdef VIMAGE
148 error = vnet_sysctl_handle_int(oidp, arg1, arg2, req);
149 #else
150 error = sysctl_handle_int(oidp, arg1, arg2, req);
151 #endif
152 if (error == 0) {
153 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
154 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
155 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
156 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
157 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
158 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
159 }
160 return (error);
161 }
162
163 #undef RANGECHK
164
165 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
166
167 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
168 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowfirstauto), 0,
169 &sysctl_net_ipport_check, "I", "");
170 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
171 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowlastauto), 0,
172 &sysctl_net_ipport_check, "I", "");
173 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, first,
174 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_firstauto), 0,
175 &sysctl_net_ipport_check, "I", "");
176 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, last,
177 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lastauto), 0,
178 &sysctl_net_ipport_check, "I", "");
179 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
180 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hifirstauto), 0,
181 &sysctl_net_ipport_check, "I", "");
182 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
183 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hilastauto), 0,
184 &sysctl_net_ipport_check, "I", "");
185 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
186 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedhigh), 0, "");
187 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
188 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
189 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_RW,
190 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
191 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_RW,
192 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
193 "allocations before switching to a sequental one");
194 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_RW,
195 &VNET_NAME(ipport_randomtime), 0,
196 "Minimum time to keep sequental port "
197 "allocation before switching to a random one");
198 #endif
199
200 /*
201 * in_pcb.c: manage the Protocol Control Blocks.
202 *
203 * NOTE: It is assumed that most of these functions will be called with
204 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
205 * functions often modify hash chains or addresses in pcbs.
206 */
207
208 /*
209 * Initialize an inpcbinfo -- we should be able to reduce the number of
210 * arguments in time.
211 */
212 void
213 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
214 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
215 char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
216 uint32_t inpcbzone_flags, u_int hashfields)
217 {
218
219 INP_INFO_LOCK_INIT(pcbinfo, name);
220 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
221 #ifdef VIMAGE
222 pcbinfo->ipi_vnet = curvnet;
223 #endif
224 pcbinfo->ipi_listhead = listhead;
225 LIST_INIT(pcbinfo->ipi_listhead);
226 pcbinfo->ipi_count = 0;
227 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
228 &pcbinfo->ipi_hashmask);
229 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
230 &pcbinfo->ipi_porthashmask);
231 #ifdef PCBGROUP
232 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
233 #endif
234 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
235 NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
236 inpcbzone_flags);
237 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
238 }
239
240 /*
241 * Destroy an inpcbinfo.
242 */
243 void
244 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
245 {
246
247 KASSERT(pcbinfo->ipi_count == 0,
248 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
249
250 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
251 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
252 pcbinfo->ipi_porthashmask);
253 #ifdef PCBGROUP
254 in_pcbgroup_destroy(pcbinfo);
255 #endif
256 uma_zdestroy(pcbinfo->ipi_zone);
257 INP_HASH_LOCK_DESTROY(pcbinfo);
258 INP_INFO_LOCK_DESTROY(pcbinfo);
259 }
260
261 /*
262 * Allocate a PCB and associate it with the socket.
263 * On success return with the PCB locked.
264 */
265 int
266 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
267 {
268 struct inpcb *inp;
269 int error;
270
271 INP_INFO_WLOCK_ASSERT(pcbinfo);
272 error = 0;
273 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
274 if (inp == NULL)
275 return (ENOBUFS);
276 bzero(inp, inp_zero_size);
277 inp->inp_pcbinfo = pcbinfo;
278 inp->inp_socket = so;
279 inp->inp_cred = crhold(so->so_cred);
280 inp->inp_inc.inc_fibnum = so->so_fibnum;
281 #ifdef MAC
282 error = mac_inpcb_init(inp, M_NOWAIT);
283 if (error != 0)
284 goto out;
285 mac_inpcb_create(so, inp);
286 #endif
287 #ifdef IPSEC
288 error = ipsec_init_policy(so, &inp->inp_sp);
289 if (error != 0) {
290 #ifdef MAC
291 mac_inpcb_destroy(inp);
292 #endif
293 goto out;
294 }
295 #endif /*IPSEC*/
296 #ifdef INET6
297 if (INP_SOCKAF(so) == AF_INET6) {
298 inp->inp_vflag |= INP_IPV6PROTO;
299 if (V_ip6_v6only)
300 inp->inp_flags |= IN6P_IPV6_V6ONLY;
301 }
302 #endif
303 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
304 pcbinfo->ipi_count++;
305 so->so_pcb = (caddr_t)inp;
306 #ifdef INET6
307 if (V_ip6_auto_flowlabel)
308 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
309 #endif
310 INP_WLOCK(inp);
311 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
312 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
313 #if defined(IPSEC) || defined(MAC)
314 out:
315 if (error != 0) {
316 crfree(inp->inp_cred);
317 uma_zfree(pcbinfo->ipi_zone, inp);
318 }
319 #endif
320 return (error);
321 }
322
323 #ifdef INET
324 int
325 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
326 {
327 int anonport, error;
328
329 INP_WLOCK_ASSERT(inp);
330 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
331
332 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
333 return (EINVAL);
334 anonport = inp->inp_lport == 0 && (nam == NULL ||
335 ((struct sockaddr_in *)nam)->sin_port == 0);
336 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
337 &inp->inp_lport, cred);
338 if (error)
339 return (error);
340 if (in_pcbinshash(inp) != 0) {
341 inp->inp_laddr.s_addr = INADDR_ANY;
342 inp->inp_lport = 0;
343 return (EAGAIN);
344 }
345 if (anonport)
346 inp->inp_flags |= INP_ANONPORT;
347 return (0);
348 }
349 #endif
350
351 #if defined(INET) || defined(INET6)
352 int
353 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
354 struct ucred *cred, int lookupflags)
355 {
356 struct inpcbinfo *pcbinfo;
357 struct inpcb *tmpinp;
358 unsigned short *lastport;
359 int count, dorandom, error;
360 u_short aux, first, last, lport;
361 #ifdef INET
362 struct in_addr laddr;
363 #endif
364
365 pcbinfo = inp->inp_pcbinfo;
366
367 /*
368 * Because no actual state changes occur here, a global write lock on
369 * the pcbinfo isn't required.
370 */
371 INP_LOCK_ASSERT(inp);
372 INP_HASH_LOCK_ASSERT(pcbinfo);
373
374 if (inp->inp_flags & INP_HIGHPORT) {
375 first = V_ipport_hifirstauto; /* sysctl */
376 last = V_ipport_hilastauto;
377 lastport = &pcbinfo->ipi_lasthi;
378 } else if (inp->inp_flags & INP_LOWPORT) {
379 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
380 if (error)
381 return (error);
382 first = V_ipport_lowfirstauto; /* 1023 */
383 last = V_ipport_lowlastauto; /* 600 */
384 lastport = &pcbinfo->ipi_lastlow;
385 } else {
386 first = V_ipport_firstauto; /* sysctl */
387 last = V_ipport_lastauto;
388 lastport = &pcbinfo->ipi_lastport;
389 }
390 /*
391 * For UDP, use random port allocation as long as the user
392 * allows it. For TCP (and as of yet unknown) connections,
393 * use random port allocation only if the user allows it AND
394 * ipport_tick() allows it.
395 */
396 if (V_ipport_randomized &&
397 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo))
398 dorandom = 1;
399 else
400 dorandom = 0;
401 /*
402 * It makes no sense to do random port allocation if
403 * we have the only port available.
404 */
405 if (first == last)
406 dorandom = 0;
407 /* Make sure to not include UDP packets in the count. */
408 if (pcbinfo != &V_udbinfo)
409 V_ipport_tcpallocs++;
410 /*
411 * Instead of having two loops further down counting up or down
412 * make sure that first is always <= last and go with only one
413 * code path implementing all logic.
414 */
415 if (first > last) {
416 aux = first;
417 first = last;
418 last = aux;
419 }
420
421 #ifdef INET
422 /* Make the compiler happy. */
423 laddr.s_addr = 0;
424 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
425 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
426 __func__, inp));
427 laddr = *laddrp;
428 }
429 #endif
430 tmpinp = NULL; /* Make compiler happy. */
431 lport = *lportp;
432
433 if (dorandom)
434 *lastport = first + (arc4random() % (last - first));
435
436 count = last - first;
437
438 do {
439 if (count-- < 0) /* completely used? */
440 return (EADDRNOTAVAIL);
441 ++*lastport;
442 if (*lastport < first || *lastport > last)
443 *lastport = first;
444 lport = htons(*lastport);
445
446 #ifdef INET6
447 if ((inp->inp_vflag & INP_IPV6) != 0)
448 tmpinp = in6_pcblookup_local(pcbinfo,
449 &inp->in6p_laddr, lport, lookupflags, cred);
450 #endif
451 #if defined(INET) && defined(INET6)
452 else
453 #endif
454 #ifdef INET
455 tmpinp = in_pcblookup_local(pcbinfo, laddr,
456 lport, lookupflags, cred);
457 #endif
458 } while (tmpinp != NULL);
459
460 #ifdef INET
461 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
462 laddrp->s_addr = laddr.s_addr;
463 #endif
464 *lportp = lport;
465
466 return (0);
467 }
468 #endif /* INET || INET6 */
469
470 #ifdef INET
471 /*
472 * Set up a bind operation on a PCB, performing port allocation
473 * as required, but do not actually modify the PCB. Callers can
474 * either complete the bind by setting inp_laddr/inp_lport and
475 * calling in_pcbinshash(), or they can just use the resulting
476 * port and address to authorise the sending of a once-off packet.
477 *
478 * On error, the values of *laddrp and *lportp are not changed.
479 */
480 int
481 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
482 u_short *lportp, struct ucred *cred)
483 {
484 struct socket *so = inp->inp_socket;
485 struct sockaddr_in *sin;
486 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
487 struct in_addr laddr;
488 u_short lport = 0;
489 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
490 int error;
491
492 /*
493 * No state changes, so read locks are sufficient here.
494 */
495 INP_LOCK_ASSERT(inp);
496 INP_HASH_LOCK_ASSERT(pcbinfo);
497
498 if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
499 return (EADDRNOTAVAIL);
500 laddr.s_addr = *laddrp;
501 if (nam != NULL && laddr.s_addr != INADDR_ANY)
502 return (EINVAL);
503 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
504 lookupflags = INPLOOKUP_WILDCARD;
505 if (nam == NULL) {
506 if ((error = prison_local_ip4(cred, &laddr)) != 0)
507 return (error);
508 } else {
509 sin = (struct sockaddr_in *)nam;
510 if (nam->sa_len != sizeof (*sin))
511 return (EINVAL);
512 #ifdef notdef
513 /*
514 * We should check the family, but old programs
515 * incorrectly fail to initialize it.
516 */
517 if (sin->sin_family != AF_INET)
518 return (EAFNOSUPPORT);
519 #endif
520 error = prison_local_ip4(cred, &sin->sin_addr);
521 if (error)
522 return (error);
523 if (sin->sin_port != *lportp) {
524 /* Don't allow the port to change. */
525 if (*lportp != 0)
526 return (EINVAL);
527 lport = sin->sin_port;
528 }
529 /* NB: lport is left as 0 if the port isn't being changed. */
530 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
531 /*
532 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
533 * allow complete duplication of binding if
534 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
535 * and a multicast address is bound on both
536 * new and duplicated sockets.
537 */
538 if (so->so_options & SO_REUSEADDR)
539 reuseport = SO_REUSEADDR|SO_REUSEPORT;
540 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
541 sin->sin_port = 0; /* yech... */
542 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
543 /*
544 * Is the address a local IP address?
545 * If INP_BINDANY is set, then the socket may be bound
546 * to any endpoint address, local or not.
547 */
548 if ((inp->inp_flags & INP_BINDANY) == 0 &&
549 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
550 return (EADDRNOTAVAIL);
551 }
552 laddr = sin->sin_addr;
553 if (lport) {
554 struct inpcb *t;
555 struct tcptw *tw;
556
557 /* GROSS */
558 if (ntohs(lport) <= V_ipport_reservedhigh &&
559 ntohs(lport) >= V_ipport_reservedlow &&
560 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
561 0))
562 return (EACCES);
563 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
564 priv_check_cred(inp->inp_cred,
565 PRIV_NETINET_REUSEPORT, 0) != 0) {
566 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
567 lport, INPLOOKUP_WILDCARD, cred);
568 /*
569 * XXX
570 * This entire block sorely needs a rewrite.
571 */
572 if (t &&
573 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
574 (so->so_type != SOCK_STREAM ||
575 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
576 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
577 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
578 (t->inp_flags2 & INP_REUSEPORT) == 0) &&
579 (inp->inp_cred->cr_uid !=
580 t->inp_cred->cr_uid))
581 return (EADDRINUSE);
582 }
583 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
584 lport, lookupflags, cred);
585 if (t && (t->inp_flags & INP_TIMEWAIT)) {
586 /*
587 * XXXRW: If an incpb has had its timewait
588 * state recycled, we treat the address as
589 * being in use (for now). This is better
590 * than a panic, but not desirable.
591 */
592 tw = intotw(t);
593 if (tw == NULL ||
594 (reuseport & tw->tw_so_options) == 0)
595 return (EADDRINUSE);
596 } else if (t && (reuseport == 0 ||
597 (t->inp_flags2 & INP_REUSEPORT) == 0)) {
598 #ifdef INET6
599 if (ntohl(sin->sin_addr.s_addr) !=
600 INADDR_ANY ||
601 ntohl(t->inp_laddr.s_addr) !=
602 INADDR_ANY ||
603 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
604 (t->inp_vflag & INP_IPV6PROTO) == 0)
605 #endif
606 return (EADDRINUSE);
607 }
608 }
609 }
610 if (*lportp != 0)
611 lport = *lportp;
612 if (lport == 0) {
613 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
614 if (error != 0)
615 return (error);
616
617 }
618 *laddrp = laddr.s_addr;
619 *lportp = lport;
620 return (0);
621 }
622
623 /*
624 * Connect from a socket to a specified address.
625 * Both address and port must be specified in argument sin.
626 * If don't have a local address for this socket yet,
627 * then pick one.
628 */
629 int
630 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
631 struct ucred *cred, struct mbuf *m)
632 {
633 u_short lport, fport;
634 in_addr_t laddr, faddr;
635 int anonport, error;
636
637 INP_WLOCK_ASSERT(inp);
638 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
639
640 lport = inp->inp_lport;
641 laddr = inp->inp_laddr.s_addr;
642 anonport = (lport == 0);
643 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
644 NULL, cred);
645 if (error)
646 return (error);
647
648 /* Do the initial binding of the local address if required. */
649 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
650 inp->inp_lport = lport;
651 inp->inp_laddr.s_addr = laddr;
652 if (in_pcbinshash(inp) != 0) {
653 inp->inp_laddr.s_addr = INADDR_ANY;
654 inp->inp_lport = 0;
655 return (EAGAIN);
656 }
657 }
658
659 /* Commit the remaining changes. */
660 inp->inp_lport = lport;
661 inp->inp_laddr.s_addr = laddr;
662 inp->inp_faddr.s_addr = faddr;
663 inp->inp_fport = fport;
664 in_pcbrehash_mbuf(inp, m);
665
666 if (anonport)
667 inp->inp_flags |= INP_ANONPORT;
668 return (0);
669 }
670
671 int
672 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
673 {
674
675 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
676 }
677
678 /*
679 * Do proper source address selection on an unbound socket in case
680 * of connect. Take jails into account as well.
681 */
682 static int
683 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
684 struct ucred *cred)
685 {
686 struct ifaddr *ifa;
687 struct sockaddr *sa;
688 struct sockaddr_in *sin;
689 struct route sro;
690 int error;
691
692 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
693
694 /*
695 * Bypass source address selection and use the primary jail IP
696 * if requested.
697 */
698 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
699 return (0);
700
701 error = 0;
702 bzero(&sro, sizeof(sro));
703
704 sin = (struct sockaddr_in *)&sro.ro_dst;
705 sin->sin_family = AF_INET;
706 sin->sin_len = sizeof(struct sockaddr_in);
707 sin->sin_addr.s_addr = faddr->s_addr;
708
709 /*
710 * If route is known our src addr is taken from the i/f,
711 * else punt.
712 *
713 * Find out route to destination.
714 */
715 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
716 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
717
718 /*
719 * If we found a route, use the address corresponding to
720 * the outgoing interface.
721 *
722 * Otherwise assume faddr is reachable on a directly connected
723 * network and try to find a corresponding interface to take
724 * the source address from.
725 */
726 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
727 struct in_ifaddr *ia;
728 struct ifnet *ifp;
729
730 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin));
731 if (ia == NULL)
732 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0));
733 if (ia == NULL) {
734 error = ENETUNREACH;
735 goto done;
736 }
737
738 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
739 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
740 ifa_free(&ia->ia_ifa);
741 goto done;
742 }
743
744 ifp = ia->ia_ifp;
745 ifa_free(&ia->ia_ifa);
746 ia = NULL;
747 IF_ADDR_LOCK(ifp);
748 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
749
750 sa = ifa->ifa_addr;
751 if (sa->sa_family != AF_INET)
752 continue;
753 sin = (struct sockaddr_in *)sa;
754 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
755 ia = (struct in_ifaddr *)ifa;
756 break;
757 }
758 }
759 if (ia != NULL) {
760 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
761 IF_ADDR_UNLOCK(ifp);
762 goto done;
763 }
764 IF_ADDR_UNLOCK(ifp);
765
766 /* 3. As a last resort return the 'default' jail address. */
767 error = prison_get_ip4(cred, laddr);
768 goto done;
769 }
770
771 /*
772 * If the outgoing interface on the route found is not
773 * a loopback interface, use the address from that interface.
774 * In case of jails do those three steps:
775 * 1. check if the interface address belongs to the jail. If so use it.
776 * 2. check if we have any address on the outgoing interface
777 * belonging to this jail. If so use it.
778 * 3. as a last resort return the 'default' jail address.
779 */
780 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
781 struct in_ifaddr *ia;
782 struct ifnet *ifp;
783
784 /* If not jailed, use the default returned. */
785 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
786 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
787 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
788 goto done;
789 }
790
791 /* Jailed. */
792 /* 1. Check if the iface address belongs to the jail. */
793 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
794 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
795 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
796 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
797 goto done;
798 }
799
800 /*
801 * 2. Check if we have any address on the outgoing interface
802 * belonging to this jail.
803 */
804 ia = NULL;
805 ifp = sro.ro_rt->rt_ifp;
806 IF_ADDR_LOCK(ifp);
807 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
808 sa = ifa->ifa_addr;
809 if (sa->sa_family != AF_INET)
810 continue;
811 sin = (struct sockaddr_in *)sa;
812 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
813 ia = (struct in_ifaddr *)ifa;
814 break;
815 }
816 }
817 if (ia != NULL) {
818 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
819 IF_ADDR_UNLOCK(ifp);
820 goto done;
821 }
822 IF_ADDR_UNLOCK(ifp);
823
824 /* 3. As a last resort return the 'default' jail address. */
825 error = prison_get_ip4(cred, laddr);
826 goto done;
827 }
828
829 /*
830 * The outgoing interface is marked with 'loopback net', so a route
831 * to ourselves is here.
832 * Try to find the interface of the destination address and then
833 * take the address from there. That interface is not necessarily
834 * a loopback interface.
835 * In case of jails, check that it is an address of the jail
836 * and if we cannot find, fall back to the 'default' jail address.
837 */
838 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
839 struct sockaddr_in sain;
840 struct in_ifaddr *ia;
841
842 bzero(&sain, sizeof(struct sockaddr_in));
843 sain.sin_family = AF_INET;
844 sain.sin_len = sizeof(struct sockaddr_in);
845 sain.sin_addr.s_addr = faddr->s_addr;
846
847 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain)));
848 if (ia == NULL)
849 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0));
850 if (ia == NULL)
851 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
852
853 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
854 if (ia == NULL) {
855 error = ENETUNREACH;
856 goto done;
857 }
858 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
859 ifa_free(&ia->ia_ifa);
860 goto done;
861 }
862
863 /* Jailed. */
864 if (ia != NULL) {
865 struct ifnet *ifp;
866
867 ifp = ia->ia_ifp;
868 ifa_free(&ia->ia_ifa);
869 ia = NULL;
870 IF_ADDR_LOCK(ifp);
871 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
872
873 sa = ifa->ifa_addr;
874 if (sa->sa_family != AF_INET)
875 continue;
876 sin = (struct sockaddr_in *)sa;
877 if (prison_check_ip4(cred,
878 &sin->sin_addr) == 0) {
879 ia = (struct in_ifaddr *)ifa;
880 break;
881 }
882 }
883 if (ia != NULL) {
884 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
885 IF_ADDR_UNLOCK(ifp);
886 goto done;
887 }
888 IF_ADDR_UNLOCK(ifp);
889 }
890
891 /* 3. As a last resort return the 'default' jail address. */
892 error = prison_get_ip4(cred, laddr);
893 goto done;
894 }
895
896 done:
897 if (sro.ro_rt != NULL)
898 RTFREE(sro.ro_rt);
899 return (error);
900 }
901
902 /*
903 * Set up for a connect from a socket to the specified address.
904 * On entry, *laddrp and *lportp should contain the current local
905 * address and port for the PCB; these are updated to the values
906 * that should be placed in inp_laddr and inp_lport to complete
907 * the connect.
908 *
909 * On success, *faddrp and *fportp will be set to the remote address
910 * and port. These are not updated in the error case.
911 *
912 * If the operation fails because the connection already exists,
913 * *oinpp will be set to the PCB of that connection so that the
914 * caller can decide to override it. In all other cases, *oinpp
915 * is set to NULL.
916 */
917 int
918 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
919 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
920 struct inpcb **oinpp, struct ucred *cred)
921 {
922 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
923 struct in_ifaddr *ia;
924 struct inpcb *oinp;
925 struct in_addr laddr, faddr;
926 u_short lport, fport;
927 int error;
928
929 /*
930 * Because a global state change doesn't actually occur here, a read
931 * lock is sufficient.
932 */
933 INP_LOCK_ASSERT(inp);
934 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
935
936 if (oinpp != NULL)
937 *oinpp = NULL;
938 if (nam->sa_len != sizeof (*sin))
939 return (EINVAL);
940 if (sin->sin_family != AF_INET)
941 return (EAFNOSUPPORT);
942 if (sin->sin_port == 0)
943 return (EADDRNOTAVAIL);
944 laddr.s_addr = *laddrp;
945 lport = *lportp;
946 faddr = sin->sin_addr;
947 fport = sin->sin_port;
948
949 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
950 /*
951 * If the destination address is INADDR_ANY,
952 * use the primary local address.
953 * If the supplied address is INADDR_BROADCAST,
954 * and the primary interface supports broadcast,
955 * choose the broadcast address for that interface.
956 */
957 if (faddr.s_addr == INADDR_ANY) {
958 IN_IFADDR_RLOCK();
959 faddr =
960 IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
961 IN_IFADDR_RUNLOCK();
962 if (cred != NULL &&
963 (error = prison_get_ip4(cred, &faddr)) != 0)
964 return (error);
965 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
966 IN_IFADDR_RLOCK();
967 if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
968 IFF_BROADCAST)
969 faddr = satosin(&TAILQ_FIRST(
970 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
971 IN_IFADDR_RUNLOCK();
972 }
973 }
974 if (laddr.s_addr == INADDR_ANY) {
975 error = in_pcbladdr(inp, &faddr, &laddr, cred);
976 /*
977 * If the destination address is multicast and an outgoing
978 * interface has been set as a multicast option, prefer the
979 * address of that interface as our source address.
980 */
981 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
982 inp->inp_moptions != NULL) {
983 struct ip_moptions *imo;
984 struct ifnet *ifp;
985
986 imo = inp->inp_moptions;
987 if (imo->imo_multicast_ifp != NULL) {
988 ifp = imo->imo_multicast_ifp;
989 IN_IFADDR_RLOCK();
990 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
991 if ((ia->ia_ifp == ifp) &&
992 (cred == NULL ||
993 prison_check_ip4(cred,
994 &ia->ia_addr.sin_addr) == 0))
995 break;
996 }
997 if (ia == NULL)
998 error = EADDRNOTAVAIL;
999 else {
1000 laddr = ia->ia_addr.sin_addr;
1001 error = 0;
1002 }
1003 IN_IFADDR_RUNLOCK();
1004 }
1005 }
1006 if (error)
1007 return (error);
1008 }
1009 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1010 laddr, lport, 0, NULL);
1011 if (oinp != NULL) {
1012 if (oinpp != NULL)
1013 *oinpp = oinp;
1014 return (EADDRINUSE);
1015 }
1016 if (lport == 0) {
1017 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1018 cred);
1019 if (error)
1020 return (error);
1021 }
1022 *laddrp = laddr.s_addr;
1023 *lportp = lport;
1024 *faddrp = faddr.s_addr;
1025 *fportp = fport;
1026 return (0);
1027 }
1028
1029 void
1030 in_pcbdisconnect(struct inpcb *inp)
1031 {
1032
1033 INP_WLOCK_ASSERT(inp);
1034 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1035
1036 inp->inp_faddr.s_addr = INADDR_ANY;
1037 inp->inp_fport = 0;
1038 in_pcbrehash(inp);
1039 }
1040 #endif
1041
1042 /*
1043 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1044 * For most protocols, this will be invoked immediately prior to calling
1045 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1046 * socket, in which case in_pcbfree() is deferred.
1047 */
1048 void
1049 in_pcbdetach(struct inpcb *inp)
1050 {
1051
1052 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1053
1054 inp->inp_socket->so_pcb = NULL;
1055 inp->inp_socket = NULL;
1056 }
1057
1058 /*
1059 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1060 * stability of an inpcb pointer despite the inpcb lock being released. This
1061 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1062 * but where the inpcb lock may already held, or when acquiring a reference
1063 * via a pcbgroup.
1064 *
1065 * in_pcbref() should be used only to provide brief memory stability, and
1066 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1067 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1068 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1069 * lock and rele are the *only* safe operations that may be performed on the
1070 * inpcb.
1071 *
1072 * While the inpcb will not be freed, releasing the inpcb lock means that the
1073 * connection's state may change, so the caller should be careful to
1074 * revalidate any cached state on reacquiring the lock. Drop the reference
1075 * using in_pcbrele().
1076 */
1077 void
1078 in_pcbref(struct inpcb *inp)
1079 {
1080
1081 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1082
1083 refcount_acquire(&inp->inp_refcount);
1084 }
1085
1086 /*
1087 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1088 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1089 * return a flag indicating whether or not the inpcb remains valid. If it is
1090 * valid, we return with the inpcb lock held.
1091 *
1092 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1093 * reference on an inpcb. Historically more work was done here (actually, in
1094 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1095 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1096 * about memory stability (and continued use of the write lock).
1097 */
1098 int
1099 in_pcbrele_rlocked(struct inpcb *inp)
1100 {
1101 struct inpcbinfo *pcbinfo;
1102
1103 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1104
1105 INP_RLOCK_ASSERT(inp);
1106
1107 if (refcount_release(&inp->inp_refcount) == 0)
1108 return (0);
1109
1110 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1111
1112 INP_RUNLOCK(inp);
1113 pcbinfo = inp->inp_pcbinfo;
1114 uma_zfree(pcbinfo->ipi_zone, inp);
1115 return (1);
1116 }
1117
1118 int
1119 in_pcbrele_wlocked(struct inpcb *inp)
1120 {
1121 struct inpcbinfo *pcbinfo;
1122
1123 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1124
1125 INP_WLOCK_ASSERT(inp);
1126
1127 if (refcount_release(&inp->inp_refcount) == 0)
1128 return (0);
1129
1130 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1131
1132 INP_WUNLOCK(inp);
1133 pcbinfo = inp->inp_pcbinfo;
1134 uma_zfree(pcbinfo->ipi_zone, inp);
1135 return (1);
1136 }
1137
1138 /*
1139 * Temporary wrapper.
1140 */
1141 int
1142 in_pcbrele(struct inpcb *inp)
1143 {
1144
1145 return (in_pcbrele_wlocked(inp));
1146 }
1147
1148 /*
1149 * Unconditionally schedule an inpcb to be freed by decrementing its
1150 * reference count, which should occur only after the inpcb has been detached
1151 * from its socket. If another thread holds a temporary reference (acquired
1152 * using in_pcbref()) then the free is deferred until that reference is
1153 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1154 * work, including removal from global lists, is done in this context, where
1155 * the pcbinfo lock is held.
1156 */
1157 void
1158 in_pcbfree(struct inpcb *inp)
1159 {
1160 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1161
1162 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1163
1164 INP_INFO_WLOCK_ASSERT(pcbinfo);
1165 INP_WLOCK_ASSERT(inp);
1166
1167 /* XXXRW: Do as much as possible here. */
1168 #ifdef IPSEC
1169 if (inp->inp_sp != NULL)
1170 ipsec_delete_pcbpolicy(inp);
1171 #endif /* IPSEC */
1172 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1173 in_pcbremlists(inp);
1174 #ifdef INET6
1175 if (inp->inp_vflag & INP_IPV6PROTO) {
1176 ip6_freepcbopts(inp->in6p_outputopts);
1177 if (inp->in6p_moptions != NULL)
1178 ip6_freemoptions(inp->in6p_moptions);
1179 }
1180 #endif
1181 if (inp->inp_options)
1182 (void)m_free(inp->inp_options);
1183 #ifdef INET
1184 if (inp->inp_moptions != NULL)
1185 inp_freemoptions(inp->inp_moptions);
1186 #endif
1187 inp->inp_vflag = 0;
1188 crfree(inp->inp_cred);
1189 #ifdef MAC
1190 mac_inpcb_destroy(inp);
1191 #endif
1192 if (!in_pcbrele_wlocked(inp))
1193 INP_WUNLOCK(inp);
1194 }
1195
1196 /*
1197 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1198 * port reservation, and preventing it from being returned by inpcb lookups.
1199 *
1200 * It is used by TCP to mark an inpcb as unused and avoid future packet
1201 * delivery or event notification when a socket remains open but TCP has
1202 * closed. This might occur as a result of a shutdown()-initiated TCP close
1203 * or a RST on the wire, and allows the port binding to be reused while still
1204 * maintaining the invariant that so_pcb always points to a valid inpcb until
1205 * in_pcbdetach().
1206 *
1207 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1208 * in_pcbnotifyall() and in_pcbpurgeif0()?
1209 */
1210 void
1211 in_pcbdrop(struct inpcb *inp)
1212 {
1213
1214 INP_WLOCK_ASSERT(inp);
1215
1216 /*
1217 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1218 * the hash lock...?
1219 */
1220 inp->inp_flags |= INP_DROPPED;
1221 if (inp->inp_flags & INP_INHASHLIST) {
1222 struct inpcbport *phd = inp->inp_phd;
1223
1224 INP_HASH_WLOCK(inp->inp_pcbinfo);
1225 LIST_REMOVE(inp, inp_hash);
1226 LIST_REMOVE(inp, inp_portlist);
1227 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1228 LIST_REMOVE(phd, phd_hash);
1229 free(phd, M_PCB);
1230 }
1231 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1232 inp->inp_flags &= ~INP_INHASHLIST;
1233 #ifdef PCBGROUP
1234 in_pcbgroup_remove(inp);
1235 #endif
1236 }
1237 }
1238
1239 #ifdef INET
1240 /*
1241 * Common routines to return the socket addresses associated with inpcbs.
1242 */
1243 struct sockaddr *
1244 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1245 {
1246 struct sockaddr_in *sin;
1247
1248 sin = malloc(sizeof *sin, M_SONAME,
1249 M_WAITOK | M_ZERO);
1250 sin->sin_family = AF_INET;
1251 sin->sin_len = sizeof(*sin);
1252 sin->sin_addr = *addr_p;
1253 sin->sin_port = port;
1254
1255 return (struct sockaddr *)sin;
1256 }
1257
1258 int
1259 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1260 {
1261 struct inpcb *inp;
1262 struct in_addr addr;
1263 in_port_t port;
1264
1265 inp = sotoinpcb(so);
1266 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1267
1268 INP_RLOCK(inp);
1269 port = inp->inp_lport;
1270 addr = inp->inp_laddr;
1271 INP_RUNLOCK(inp);
1272
1273 *nam = in_sockaddr(port, &addr);
1274 return 0;
1275 }
1276
1277 int
1278 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1279 {
1280 struct inpcb *inp;
1281 struct in_addr addr;
1282 in_port_t port;
1283
1284 inp = sotoinpcb(so);
1285 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1286
1287 INP_RLOCK(inp);
1288 port = inp->inp_fport;
1289 addr = inp->inp_faddr;
1290 INP_RUNLOCK(inp);
1291
1292 *nam = in_sockaddr(port, &addr);
1293 return 0;
1294 }
1295
1296 void
1297 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1298 struct inpcb *(*notify)(struct inpcb *, int))
1299 {
1300 struct inpcb *inp, *inp_temp;
1301
1302 INP_INFO_WLOCK(pcbinfo);
1303 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1304 INP_WLOCK(inp);
1305 #ifdef INET6
1306 if ((inp->inp_vflag & INP_IPV4) == 0) {
1307 INP_WUNLOCK(inp);
1308 continue;
1309 }
1310 #endif
1311 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1312 inp->inp_socket == NULL) {
1313 INP_WUNLOCK(inp);
1314 continue;
1315 }
1316 if ((*notify)(inp, errno))
1317 INP_WUNLOCK(inp);
1318 }
1319 INP_INFO_WUNLOCK(pcbinfo);
1320 }
1321
1322 void
1323 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1324 {
1325 struct inpcb *inp;
1326 struct ip_moptions *imo;
1327 int i, gap;
1328
1329 INP_INFO_RLOCK(pcbinfo);
1330 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1331 INP_WLOCK(inp);
1332 imo = inp->inp_moptions;
1333 if ((inp->inp_vflag & INP_IPV4) &&
1334 imo != NULL) {
1335 /*
1336 * Unselect the outgoing interface if it is being
1337 * detached.
1338 */
1339 if (imo->imo_multicast_ifp == ifp)
1340 imo->imo_multicast_ifp = NULL;
1341
1342 /*
1343 * Drop multicast group membership if we joined
1344 * through the interface being detached.
1345 */
1346 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1347 i++) {
1348 if (imo->imo_membership[i]->inm_ifp == ifp) {
1349 in_delmulti(imo->imo_membership[i]);
1350 gap++;
1351 } else if (gap != 0)
1352 imo->imo_membership[i - gap] =
1353 imo->imo_membership[i];
1354 }
1355 imo->imo_num_memberships -= gap;
1356 }
1357 INP_WUNLOCK(inp);
1358 }
1359 INP_INFO_RUNLOCK(pcbinfo);
1360 }
1361
1362 /*
1363 * Lookup a PCB based on the local address and port. Caller must hold the
1364 * hash lock. No inpcb locks or references are acquired.
1365 */
1366 #define INP_LOOKUP_MAPPED_PCB_COST 3
1367 struct inpcb *
1368 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1369 u_short lport, int lookupflags, struct ucred *cred)
1370 {
1371 struct inpcb *inp;
1372 #ifdef INET6
1373 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1374 #else
1375 int matchwild = 3;
1376 #endif
1377 int wildcard;
1378
1379 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1380 ("%s: invalid lookup flags %d", __func__, lookupflags));
1381
1382 INP_HASH_LOCK_ASSERT(pcbinfo);
1383
1384 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1385 struct inpcbhead *head;
1386 /*
1387 * Look for an unconnected (wildcard foreign addr) PCB that
1388 * matches the local address and port we're looking for.
1389 */
1390 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1391 0, pcbinfo->ipi_hashmask)];
1392 LIST_FOREACH(inp, head, inp_hash) {
1393 #ifdef INET6
1394 /* XXX inp locking */
1395 if ((inp->inp_vflag & INP_IPV4) == 0)
1396 continue;
1397 #endif
1398 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1399 inp->inp_laddr.s_addr == laddr.s_addr &&
1400 inp->inp_lport == lport) {
1401 /*
1402 * Found?
1403 */
1404 if (cred == NULL ||
1405 prison_equal_ip4(cred->cr_prison,
1406 inp->inp_cred->cr_prison))
1407 return (inp);
1408 }
1409 }
1410 /*
1411 * Not found.
1412 */
1413 return (NULL);
1414 } else {
1415 struct inpcbporthead *porthash;
1416 struct inpcbport *phd;
1417 struct inpcb *match = NULL;
1418 /*
1419 * Best fit PCB lookup.
1420 *
1421 * First see if this local port is in use by looking on the
1422 * port hash list.
1423 */
1424 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1425 pcbinfo->ipi_porthashmask)];
1426 LIST_FOREACH(phd, porthash, phd_hash) {
1427 if (phd->phd_port == lport)
1428 break;
1429 }
1430 if (phd != NULL) {
1431 /*
1432 * Port is in use by one or more PCBs. Look for best
1433 * fit.
1434 */
1435 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1436 wildcard = 0;
1437 if (cred != NULL &&
1438 !prison_equal_ip4(inp->inp_cred->cr_prison,
1439 cred->cr_prison))
1440 continue;
1441 #ifdef INET6
1442 /* XXX inp locking */
1443 if ((inp->inp_vflag & INP_IPV4) == 0)
1444 continue;
1445 /*
1446 * We never select the PCB that has
1447 * INP_IPV6 flag and is bound to :: if
1448 * we have another PCB which is bound
1449 * to 0.0.0.0. If a PCB has the
1450 * INP_IPV6 flag, then we set its cost
1451 * higher than IPv4 only PCBs.
1452 *
1453 * Note that the case only happens
1454 * when a socket is bound to ::, under
1455 * the condition that the use of the
1456 * mapped address is allowed.
1457 */
1458 if ((inp->inp_vflag & INP_IPV6) != 0)
1459 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1460 #endif
1461 if (inp->inp_faddr.s_addr != INADDR_ANY)
1462 wildcard++;
1463 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1464 if (laddr.s_addr == INADDR_ANY)
1465 wildcard++;
1466 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1467 continue;
1468 } else {
1469 if (laddr.s_addr != INADDR_ANY)
1470 wildcard++;
1471 }
1472 if (wildcard < matchwild) {
1473 match = inp;
1474 matchwild = wildcard;
1475 if (matchwild == 0)
1476 break;
1477 }
1478 }
1479 }
1480 return (match);
1481 }
1482 }
1483 #undef INP_LOOKUP_MAPPED_PCB_COST
1484
1485 #ifdef PCBGROUP
1486 /*
1487 * Lookup PCB in hash list, using pcbgroup tables.
1488 */
1489 static struct inpcb *
1490 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1491 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1492 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1493 {
1494 struct inpcbhead *head;
1495 struct inpcb *inp, *tmpinp;
1496 u_short fport = fport_arg, lport = lport_arg;
1497
1498 /*
1499 * First look for an exact match.
1500 */
1501 tmpinp = NULL;
1502 INP_GROUP_LOCK(pcbgroup);
1503 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1504 pcbgroup->ipg_hashmask)];
1505 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1506 #ifdef INET6
1507 /* XXX inp locking */
1508 if ((inp->inp_vflag & INP_IPV4) == 0)
1509 continue;
1510 #endif
1511 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1512 inp->inp_laddr.s_addr == laddr.s_addr &&
1513 inp->inp_fport == fport &&
1514 inp->inp_lport == lport) {
1515 /*
1516 * XXX We should be able to directly return
1517 * the inp here, without any checks.
1518 * Well unless both bound with SO_REUSEPORT?
1519 */
1520 if (prison_flag(inp->inp_cred, PR_IP4))
1521 goto found;
1522 if (tmpinp == NULL)
1523 tmpinp = inp;
1524 }
1525 }
1526 if (tmpinp != NULL) {
1527 inp = tmpinp;
1528 goto found;
1529 }
1530
1531 /*
1532 * Then look for a wildcard match, if requested.
1533 */
1534 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1535 struct inpcb *local_wild = NULL, *local_exact = NULL;
1536 #ifdef INET6
1537 struct inpcb *local_wild_mapped = NULL;
1538 #endif
1539 struct inpcb *jail_wild = NULL;
1540 struct inpcbhead *head;
1541 int injail;
1542
1543 /*
1544 * Order of socket selection - we always prefer jails.
1545 * 1. jailed, non-wild.
1546 * 2. jailed, wild.
1547 * 3. non-jailed, non-wild.
1548 * 4. non-jailed, wild.
1549 */
1550 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1551 0, pcbinfo->ipi_wildmask)];
1552 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1553 #ifdef INET6
1554 /* XXX inp locking */
1555 if ((inp->inp_vflag & INP_IPV4) == 0)
1556 continue;
1557 #endif
1558 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1559 inp->inp_lport != lport)
1560 continue;
1561
1562 /* XXX inp locking */
1563 if (ifp && ifp->if_type == IFT_FAITH &&
1564 (inp->inp_flags & INP_FAITH) == 0)
1565 continue;
1566
1567 injail = prison_flag(inp->inp_cred, PR_IP4);
1568 if (injail) {
1569 if (prison_check_ip4(inp->inp_cred,
1570 &laddr) != 0)
1571 continue;
1572 } else {
1573 if (local_exact != NULL)
1574 continue;
1575 }
1576
1577 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1578 if (injail)
1579 goto found;
1580 else
1581 local_exact = inp;
1582 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1583 #ifdef INET6
1584 /* XXX inp locking, NULL check */
1585 if (inp->inp_vflag & INP_IPV6PROTO)
1586 local_wild_mapped = inp;
1587 else
1588 #endif /* INET6 */
1589 if (injail)
1590 jail_wild = inp;
1591 else
1592 local_wild = inp;
1593 }
1594 } /* LIST_FOREACH */
1595 inp = jail_wild;
1596 if (inp == NULL)
1597 inp = local_exact;
1598 if (inp == NULL)
1599 inp = local_wild;
1600 #ifdef INET6
1601 if (inp == NULL)
1602 inp = local_wild_mapped;
1603 #endif /* defined(INET6) */
1604 if (inp != NULL)
1605 goto found;
1606 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1607 INP_GROUP_UNLOCK(pcbgroup);
1608 return (NULL);
1609
1610 found:
1611 in_pcbref(inp);
1612 INP_GROUP_UNLOCK(pcbgroup);
1613 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1614 INP_WLOCK(inp);
1615 if (in_pcbrele_wlocked(inp))
1616 return (NULL);
1617 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1618 INP_RLOCK(inp);
1619 if (in_pcbrele_rlocked(inp))
1620 return (NULL);
1621 } else
1622 panic("%s: locking bug", __func__);
1623 return (inp);
1624 }
1625 #endif /* PCBGROUP */
1626
1627 /*
1628 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
1629 * that the caller has locked the hash list, and will not perform any further
1630 * locking or reference operations on either the hash list or the connection.
1631 */
1632 static struct inpcb *
1633 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1634 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1635 struct ifnet *ifp)
1636 {
1637 struct inpcbhead *head;
1638 struct inpcb *inp, *tmpinp;
1639 u_short fport = fport_arg, lport = lport_arg;
1640
1641 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1642 ("%s: invalid lookup flags %d", __func__, lookupflags));
1643
1644 INP_HASH_LOCK_ASSERT(pcbinfo);
1645
1646 /*
1647 * First look for an exact match.
1648 */
1649 tmpinp = NULL;
1650 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1651 pcbinfo->ipi_hashmask)];
1652 LIST_FOREACH(inp, head, inp_hash) {
1653 #ifdef INET6
1654 /* XXX inp locking */
1655 if ((inp->inp_vflag & INP_IPV4) == 0)
1656 continue;
1657 #endif
1658 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1659 inp->inp_laddr.s_addr == laddr.s_addr &&
1660 inp->inp_fport == fport &&
1661 inp->inp_lport == lport) {
1662 /*
1663 * XXX We should be able to directly return
1664 * the inp here, without any checks.
1665 * Well unless both bound with SO_REUSEPORT?
1666 */
1667 if (prison_flag(inp->inp_cred, PR_IP4))
1668 return (inp);
1669 if (tmpinp == NULL)
1670 tmpinp = inp;
1671 }
1672 }
1673 if (tmpinp != NULL)
1674 return (tmpinp);
1675
1676 /*
1677 * Then look for a wildcard match, if requested.
1678 */
1679 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1680 struct inpcb *local_wild = NULL, *local_exact = NULL;
1681 #ifdef INET6
1682 struct inpcb *local_wild_mapped = NULL;
1683 #endif
1684 struct inpcb *jail_wild = NULL;
1685 int injail;
1686
1687 /*
1688 * Order of socket selection - we always prefer jails.
1689 * 1. jailed, non-wild.
1690 * 2. jailed, wild.
1691 * 3. non-jailed, non-wild.
1692 * 4. non-jailed, wild.
1693 */
1694
1695 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1696 0, pcbinfo->ipi_hashmask)];
1697 LIST_FOREACH(inp, head, inp_hash) {
1698 #ifdef INET6
1699 /* XXX inp locking */
1700 if ((inp->inp_vflag & INP_IPV4) == 0)
1701 continue;
1702 #endif
1703 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1704 inp->inp_lport != lport)
1705 continue;
1706
1707 /* XXX inp locking */
1708 if (ifp && ifp->if_type == IFT_FAITH &&
1709 (inp->inp_flags & INP_FAITH) == 0)
1710 continue;
1711
1712 injail = prison_flag(inp->inp_cred, PR_IP4);
1713 if (injail) {
1714 if (prison_check_ip4(inp->inp_cred,
1715 &laddr) != 0)
1716 continue;
1717 } else {
1718 if (local_exact != NULL)
1719 continue;
1720 }
1721
1722 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1723 if (injail)
1724 return (inp);
1725 else
1726 local_exact = inp;
1727 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1728 #ifdef INET6
1729 /* XXX inp locking, NULL check */
1730 if (inp->inp_vflag & INP_IPV6PROTO)
1731 local_wild_mapped = inp;
1732 else
1733 #endif /* INET6 */
1734 if (injail)
1735 jail_wild = inp;
1736 else
1737 local_wild = inp;
1738 }
1739 } /* LIST_FOREACH */
1740 if (jail_wild != NULL)
1741 return (jail_wild);
1742 if (local_exact != NULL)
1743 return (local_exact);
1744 if (local_wild != NULL)
1745 return (local_wild);
1746 #ifdef INET6
1747 if (local_wild_mapped != NULL)
1748 return (local_wild_mapped);
1749 #endif /* defined(INET6) */
1750 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1751
1752 return (NULL);
1753 }
1754
1755 /*
1756 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
1757 * hash list lock, and will return the inpcb locked (i.e., requires
1758 * INPLOOKUP_LOCKPCB).
1759 */
1760 static struct inpcb *
1761 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1762 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1763 struct ifnet *ifp)
1764 {
1765 struct inpcb *inp;
1766
1767 INP_HASH_RLOCK(pcbinfo);
1768 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1769 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1770 if (inp != NULL) {
1771 in_pcbref(inp);
1772 INP_HASH_RUNLOCK(pcbinfo);
1773 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1774 INP_WLOCK(inp);
1775 if (in_pcbrele_wlocked(inp))
1776 return (NULL);
1777 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1778 INP_RLOCK(inp);
1779 if (in_pcbrele_rlocked(inp))
1780 return (NULL);
1781 } else
1782 panic("%s: locking bug", __func__);
1783 } else
1784 INP_HASH_RUNLOCK(pcbinfo);
1785 return (inp);
1786 }
1787
1788 /*
1789 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1790 * from which a pre-calculated hash value may be extracted.
1791 *
1792 * Possibly more of this logic should be in in_pcbgroup.c.
1793 */
1794 struct inpcb *
1795 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1796 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1797 {
1798 #if defined(PCBGROUP)
1799 struct inpcbgroup *pcbgroup;
1800 #endif
1801
1802 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1803 ("%s: invalid lookup flags %d", __func__, lookupflags));
1804 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1805 ("%s: LOCKPCB not set", __func__));
1806
1807 #if defined(PCBGROUP)
1808 if (in_pcbgroup_enabled(pcbinfo)) {
1809 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1810 fport);
1811 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1812 laddr, lport, lookupflags, ifp));
1813 }
1814 #endif
1815 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1816 lookupflags, ifp));
1817 }
1818
1819 struct inpcb *
1820 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1821 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1822 struct ifnet *ifp, struct mbuf *m)
1823 {
1824 #ifdef PCBGROUP
1825 struct inpcbgroup *pcbgroup;
1826 #endif
1827
1828 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1829 ("%s: invalid lookup flags %d", __func__, lookupflags));
1830 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1831 ("%s: LOCKPCB not set", __func__));
1832
1833 #ifdef PCBGROUP
1834 if (in_pcbgroup_enabled(pcbinfo)) {
1835 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
1836 m->m_pkthdr.flowid);
1837 if (pcbgroup != NULL)
1838 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
1839 fport, laddr, lport, lookupflags, ifp));
1840 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1841 fport);
1842 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1843 laddr, lport, lookupflags, ifp));
1844 }
1845 #endif
1846 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1847 lookupflags, ifp));
1848 }
1849 #endif /* INET */
1850
1851 /*
1852 * Insert PCB onto various hash lists.
1853 */
1854 static int
1855 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
1856 {
1857 struct inpcbhead *pcbhash;
1858 struct inpcbporthead *pcbporthash;
1859 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1860 struct inpcbport *phd;
1861 u_int32_t hashkey_faddr;
1862
1863 INP_WLOCK_ASSERT(inp);
1864 INP_HASH_WLOCK_ASSERT(pcbinfo);
1865
1866 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
1867 ("in_pcbinshash: INP_INHASHLIST"));
1868
1869 #ifdef INET6
1870 if (inp->inp_vflag & INP_IPV6)
1871 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */;
1872 else
1873 #endif /* INET6 */
1874 hashkey_faddr = inp->inp_faddr.s_addr;
1875
1876 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
1877 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
1878
1879 pcbporthash = &pcbinfo->ipi_porthashbase[
1880 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
1881
1882 /*
1883 * Go through port list and look for a head for this lport.
1884 */
1885 LIST_FOREACH(phd, pcbporthash, phd_hash) {
1886 if (phd->phd_port == inp->inp_lport)
1887 break;
1888 }
1889 /*
1890 * If none exists, malloc one and tack it on.
1891 */
1892 if (phd == NULL) {
1893 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
1894 if (phd == NULL) {
1895 return (ENOBUFS); /* XXX */
1896 }
1897 phd->phd_port = inp->inp_lport;
1898 LIST_INIT(&phd->phd_pcblist);
1899 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1900 }
1901 inp->inp_phd = phd;
1902 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1903 LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
1904 inp->inp_flags |= INP_INHASHLIST;
1905 #ifdef PCBGROUP
1906 if (do_pcbgroup_update)
1907 in_pcbgroup_update(inp);
1908 #endif
1909 return (0);
1910 }
1911
1912 /*
1913 * For now, there are two public interfaces to insert an inpcb into the hash
1914 * lists -- one that does update pcbgroups, and one that doesn't. The latter
1915 * is used only in the TCP syncache, where in_pcbinshash is called before the
1916 * full 4-tuple is set for the inpcb, and we don't want to install in the
1917 * pcbgroup until later.
1918 *
1919 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
1920 * connection groups, and partially initialised inpcbs should not be exposed
1921 * to either reservation hash tables or pcbgroups.
1922 */
1923 int
1924 in_pcbinshash(struct inpcb *inp)
1925 {
1926
1927 return (in_pcbinshash_internal(inp, 1));
1928 }
1929
1930 int
1931 in_pcbinshash_nopcbgroup(struct inpcb *inp)
1932 {
1933
1934 return (in_pcbinshash_internal(inp, 0));
1935 }
1936
1937 /*
1938 * Move PCB to the proper hash bucket when { faddr, fport } have been
1939 * changed. NOTE: This does not handle the case of the lport changing (the
1940 * hashed port list would have to be updated as well), so the lport must
1941 * not change after in_pcbinshash() has been called.
1942 */
1943 void
1944 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
1945 {
1946 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1947 struct inpcbhead *head;
1948 u_int32_t hashkey_faddr;
1949
1950 INP_WLOCK_ASSERT(inp);
1951 INP_HASH_WLOCK_ASSERT(pcbinfo);
1952
1953 KASSERT(inp->inp_flags & INP_INHASHLIST,
1954 ("in_pcbrehash: !INP_INHASHLIST"));
1955
1956 #ifdef INET6
1957 if (inp->inp_vflag & INP_IPV6)
1958 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */;
1959 else
1960 #endif /* INET6 */
1961 hashkey_faddr = inp->inp_faddr.s_addr;
1962
1963 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
1964 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
1965
1966 LIST_REMOVE(inp, inp_hash);
1967 LIST_INSERT_HEAD(head, inp, inp_hash);
1968
1969 #ifdef PCBGROUP
1970 if (m != NULL)
1971 in_pcbgroup_update_mbuf(inp, m);
1972 else
1973 in_pcbgroup_update(inp);
1974 #endif
1975 }
1976
1977 void
1978 in_pcbrehash(struct inpcb *inp)
1979 {
1980
1981 in_pcbrehash_mbuf(inp, NULL);
1982 }
1983
1984 /*
1985 * Remove PCB from various lists.
1986 */
1987 static void
1988 in_pcbremlists(struct inpcb *inp)
1989 {
1990 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1991
1992 INP_INFO_WLOCK_ASSERT(pcbinfo);
1993 INP_WLOCK_ASSERT(inp);
1994
1995 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1996 if (inp->inp_flags & INP_INHASHLIST) {
1997 struct inpcbport *phd = inp->inp_phd;
1998
1999 INP_HASH_WLOCK(pcbinfo);
2000 LIST_REMOVE(inp, inp_hash);
2001 LIST_REMOVE(inp, inp_portlist);
2002 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2003 LIST_REMOVE(phd, phd_hash);
2004 free(phd, M_PCB);
2005 }
2006 INP_HASH_WUNLOCK(pcbinfo);
2007 inp->inp_flags &= ~INP_INHASHLIST;
2008 }
2009 LIST_REMOVE(inp, inp_list);
2010 pcbinfo->ipi_count--;
2011 #ifdef PCBGROUP
2012 in_pcbgroup_remove(inp);
2013 #endif
2014 }
2015
2016 /*
2017 * A set label operation has occurred at the socket layer, propagate the
2018 * label change into the in_pcb for the socket.
2019 */
2020 void
2021 in_pcbsosetlabel(struct socket *so)
2022 {
2023 #ifdef MAC
2024 struct inpcb *inp;
2025
2026 inp = sotoinpcb(so);
2027 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2028
2029 INP_WLOCK(inp);
2030 SOCK_LOCK(so);
2031 mac_inpcb_sosetlabel(so, inp);
2032 SOCK_UNLOCK(so);
2033 INP_WUNLOCK(inp);
2034 #endif
2035 }
2036
2037 /*
2038 * ipport_tick runs once per second, determining if random port allocation
2039 * should be continued. If more than ipport_randomcps ports have been
2040 * allocated in the last second, then we return to sequential port
2041 * allocation. We return to random allocation only once we drop below
2042 * ipport_randomcps for at least ipport_randomtime seconds.
2043 */
2044 static void
2045 ipport_tick(void *xtp)
2046 {
2047 VNET_ITERATOR_DECL(vnet_iter);
2048
2049 VNET_LIST_RLOCK_NOSLEEP();
2050 VNET_FOREACH(vnet_iter) {
2051 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2052 if (V_ipport_tcpallocs <=
2053 V_ipport_tcplastcount + V_ipport_randomcps) {
2054 if (V_ipport_stoprandom > 0)
2055 V_ipport_stoprandom--;
2056 } else
2057 V_ipport_stoprandom = V_ipport_randomtime;
2058 V_ipport_tcplastcount = V_ipport_tcpallocs;
2059 CURVNET_RESTORE();
2060 }
2061 VNET_LIST_RUNLOCK_NOSLEEP();
2062 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2063 }
2064
2065 static void
2066 ip_fini(void *xtp)
2067 {
2068
2069 callout_stop(&ipport_tick_callout);
2070 }
2071
2072 /*
2073 * The ipport_callout should start running at about the time we attach the
2074 * inet or inet6 domains.
2075 */
2076 static void
2077 ipport_tick_init(const void *unused __unused)
2078 {
2079
2080 /* Start ipport_tick. */
2081 callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
2082 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2083 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2084 SHUTDOWN_PRI_DEFAULT);
2085 }
2086 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2087 ipport_tick_init, NULL);
2088
2089 void
2090 inp_wlock(struct inpcb *inp)
2091 {
2092
2093 INP_WLOCK(inp);
2094 }
2095
2096 void
2097 inp_wunlock(struct inpcb *inp)
2098 {
2099
2100 INP_WUNLOCK(inp);
2101 }
2102
2103 void
2104 inp_rlock(struct inpcb *inp)
2105 {
2106
2107 INP_RLOCK(inp);
2108 }
2109
2110 void
2111 inp_runlock(struct inpcb *inp)
2112 {
2113
2114 INP_RUNLOCK(inp);
2115 }
2116
2117 #ifdef INVARIANTS
2118 void
2119 inp_lock_assert(struct inpcb *inp)
2120 {
2121
2122 INP_WLOCK_ASSERT(inp);
2123 }
2124
2125 void
2126 inp_unlock_assert(struct inpcb *inp)
2127 {
2128
2129 INP_UNLOCK_ASSERT(inp);
2130 }
2131 #endif
2132
2133 void
2134 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2135 {
2136 struct inpcb *inp;
2137
2138 INP_INFO_RLOCK(&V_tcbinfo);
2139 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2140 INP_WLOCK(inp);
2141 func(inp, arg);
2142 INP_WUNLOCK(inp);
2143 }
2144 INP_INFO_RUNLOCK(&V_tcbinfo);
2145 }
2146
2147 struct socket *
2148 inp_inpcbtosocket(struct inpcb *inp)
2149 {
2150
2151 INP_WLOCK_ASSERT(inp);
2152 return (inp->inp_socket);
2153 }
2154
2155 struct tcpcb *
2156 inp_inpcbtotcpcb(struct inpcb *inp)
2157 {
2158
2159 INP_WLOCK_ASSERT(inp);
2160 return ((struct tcpcb *)inp->inp_ppcb);
2161 }
2162
2163 int
2164 inp_ip_tos_get(const struct inpcb *inp)
2165 {
2166
2167 return (inp->inp_ip_tos);
2168 }
2169
2170 void
2171 inp_ip_tos_set(struct inpcb *inp, int val)
2172 {
2173
2174 inp->inp_ip_tos = val;
2175 }
2176
2177 void
2178 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2179 uint32_t *faddr, uint16_t *fp)
2180 {
2181
2182 INP_LOCK_ASSERT(inp);
2183 *laddr = inp->inp_laddr.s_addr;
2184 *faddr = inp->inp_faddr.s_addr;
2185 *lp = inp->inp_lport;
2186 *fp = inp->inp_fport;
2187 }
2188
2189 struct inpcb *
2190 so_sotoinpcb(struct socket *so)
2191 {
2192
2193 return (sotoinpcb(so));
2194 }
2195
2196 struct tcpcb *
2197 so_sototcpcb(struct socket *so)
2198 {
2199
2200 return (sototcpcb(so));
2201 }
2202
2203 #ifdef DDB
2204 static void
2205 db_print_indent(int indent)
2206 {
2207 int i;
2208
2209 for (i = 0; i < indent; i++)
2210 db_printf(" ");
2211 }
2212
2213 static void
2214 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2215 {
2216 char faddr_str[48], laddr_str[48];
2217
2218 db_print_indent(indent);
2219 db_printf("%s at %p\n", name, inc);
2220
2221 indent += 2;
2222
2223 #ifdef INET6
2224 if (inc->inc_flags & INC_ISIPV6) {
2225 /* IPv6. */
2226 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2227 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2228 } else {
2229 #endif
2230 /* IPv4. */
2231 inet_ntoa_r(inc->inc_laddr, laddr_str);
2232 inet_ntoa_r(inc->inc_faddr, faddr_str);
2233 #ifdef INET6
2234 }
2235 #endif
2236 db_print_indent(indent);
2237 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2238 ntohs(inc->inc_lport));
2239 db_print_indent(indent);
2240 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2241 ntohs(inc->inc_fport));
2242 }
2243
2244 static void
2245 db_print_inpflags(int inp_flags)
2246 {
2247 int comma;
2248
2249 comma = 0;
2250 if (inp_flags & INP_RECVOPTS) {
2251 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2252 comma = 1;
2253 }
2254 if (inp_flags & INP_RECVRETOPTS) {
2255 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2256 comma = 1;
2257 }
2258 if (inp_flags & INP_RECVDSTADDR) {
2259 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2260 comma = 1;
2261 }
2262 if (inp_flags & INP_HDRINCL) {
2263 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2264 comma = 1;
2265 }
2266 if (inp_flags & INP_HIGHPORT) {
2267 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2268 comma = 1;
2269 }
2270 if (inp_flags & INP_LOWPORT) {
2271 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2272 comma = 1;
2273 }
2274 if (inp_flags & INP_ANONPORT) {
2275 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2276 comma = 1;
2277 }
2278 if (inp_flags & INP_RECVIF) {
2279 db_printf("%sINP_RECVIF", comma ? ", " : "");
2280 comma = 1;
2281 }
2282 if (inp_flags & INP_MTUDISC) {
2283 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2284 comma = 1;
2285 }
2286 if (inp_flags & INP_FAITH) {
2287 db_printf("%sINP_FAITH", comma ? ", " : "");
2288 comma = 1;
2289 }
2290 if (inp_flags & INP_RECVTTL) {
2291 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2292 comma = 1;
2293 }
2294 if (inp_flags & INP_DONTFRAG) {
2295 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2296 comma = 1;
2297 }
2298 if (inp_flags & IN6P_IPV6_V6ONLY) {
2299 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2300 comma = 1;
2301 }
2302 if (inp_flags & IN6P_PKTINFO) {
2303 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2304 comma = 1;
2305 }
2306 if (inp_flags & IN6P_HOPLIMIT) {
2307 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2308 comma = 1;
2309 }
2310 if (inp_flags & IN6P_HOPOPTS) {
2311 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2312 comma = 1;
2313 }
2314 if (inp_flags & IN6P_DSTOPTS) {
2315 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2316 comma = 1;
2317 }
2318 if (inp_flags & IN6P_RTHDR) {
2319 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2320 comma = 1;
2321 }
2322 if (inp_flags & IN6P_RTHDRDSTOPTS) {
2323 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2324 comma = 1;
2325 }
2326 if (inp_flags & IN6P_TCLASS) {
2327 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2328 comma = 1;
2329 }
2330 if (inp_flags & IN6P_AUTOFLOWLABEL) {
2331 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2332 comma = 1;
2333 }
2334 if (inp_flags & INP_TIMEWAIT) {
2335 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2336 comma = 1;
2337 }
2338 if (inp_flags & INP_ONESBCAST) {
2339 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2340 comma = 1;
2341 }
2342 if (inp_flags & INP_DROPPED) {
2343 db_printf("%sINP_DROPPED", comma ? ", " : "");
2344 comma = 1;
2345 }
2346 if (inp_flags & INP_SOCKREF) {
2347 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2348 comma = 1;
2349 }
2350 if (inp_flags & IN6P_RFC2292) {
2351 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2352 comma = 1;
2353 }
2354 if (inp_flags & IN6P_MTU) {
2355 db_printf("IN6P_MTU%s", comma ? ", " : "");
2356 comma = 1;
2357 }
2358 }
2359
2360 static void
2361 db_print_inpvflag(u_char inp_vflag)
2362 {
2363 int comma;
2364
2365 comma = 0;
2366 if (inp_vflag & INP_IPV4) {
2367 db_printf("%sINP_IPV4", comma ? ", " : "");
2368 comma = 1;
2369 }
2370 if (inp_vflag & INP_IPV6) {
2371 db_printf("%sINP_IPV6", comma ? ", " : "");
2372 comma = 1;
2373 }
2374 if (inp_vflag & INP_IPV6PROTO) {
2375 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2376 comma = 1;
2377 }
2378 }
2379
2380 static void
2381 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2382 {
2383
2384 db_print_indent(indent);
2385 db_printf("%s at %p\n", name, inp);
2386
2387 indent += 2;
2388
2389 db_print_indent(indent);
2390 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2391
2392 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2393
2394 db_print_indent(indent);
2395 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
2396 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2397
2398 db_print_indent(indent);
2399 db_printf("inp_label: %p inp_flags: 0x%x (",
2400 inp->inp_label, inp->inp_flags);
2401 db_print_inpflags(inp->inp_flags);
2402 db_printf(")\n");
2403
2404 db_print_indent(indent);
2405 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
2406 inp->inp_vflag);
2407 db_print_inpvflag(inp->inp_vflag);
2408 db_printf(")\n");
2409
2410 db_print_indent(indent);
2411 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
2412 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2413
2414 db_print_indent(indent);
2415 #ifdef INET6
2416 if (inp->inp_vflag & INP_IPV6) {
2417 db_printf("in6p_options: %p in6p_outputopts: %p "
2418 "in6p_moptions: %p\n", inp->in6p_options,
2419 inp->in6p_outputopts, inp->in6p_moptions);
2420 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
2421 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2422 inp->in6p_hops);
2423 } else
2424 #endif
2425 {
2426 db_printf("inp_ip_tos: %d inp_ip_options: %p "
2427 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2428 inp->inp_options, inp->inp_moptions);
2429 }
2430
2431 db_print_indent(indent);
2432 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
2433 (uintmax_t)inp->inp_gencnt);
2434 }
2435
2436 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2437 {
2438 struct inpcb *inp;
2439
2440 if (!have_addr) {
2441 db_printf("usage: show inpcb <addr>\n");
2442 return;
2443 }
2444 inp = (struct inpcb *)addr;
2445
2446 db_print_inpcb(inp, "inpcb", 0);
2447 }
2448 #endif
Cache object: 2a9feb60ea908783b6690f6a2959b349
|