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/10.0/sys/netinet/in_pcb.c 253282 2013-07-12 19:08:33Z 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 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
166 "IP Ports");
167
168 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
169 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowfirstauto), 0,
170 &sysctl_net_ipport_check, "I", "");
171 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
172 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowlastauto), 0,
173 &sysctl_net_ipport_check, "I", "");
174 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, first,
175 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_firstauto), 0,
176 &sysctl_net_ipport_check, "I", "");
177 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, last,
178 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lastauto), 0,
179 &sysctl_net_ipport_check, "I", "");
180 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
181 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hifirstauto), 0,
182 &sysctl_net_ipport_check, "I", "");
183 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
184 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hilastauto), 0,
185 &sysctl_net_ipport_check, "I", "");
186 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
187 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedhigh), 0, "");
188 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
189 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
190 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_RW,
191 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
192 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_RW,
193 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
194 "allocations before switching to a sequental one");
195 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_RW,
196 &VNET_NAME(ipport_randomtime), 0,
197 "Minimum time to keep sequental port "
198 "allocation before switching to a random one");
199 #endif /* INET */
200
201 /*
202 * in_pcb.c: manage the Protocol Control Blocks.
203 *
204 * NOTE: It is assumed that most of these functions will be called with
205 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
206 * functions often modify hash chains or addresses in pcbs.
207 */
208
209 /*
210 * Initialize an inpcbinfo -- we should be able to reduce the number of
211 * arguments in time.
212 */
213 void
214 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
215 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
216 char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
217 uint32_t inpcbzone_flags, u_int hashfields)
218 {
219
220 INP_INFO_LOCK_INIT(pcbinfo, name);
221 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
222 #ifdef VIMAGE
223 pcbinfo->ipi_vnet = curvnet;
224 #endif
225 pcbinfo->ipi_listhead = listhead;
226 LIST_INIT(pcbinfo->ipi_listhead);
227 pcbinfo->ipi_count = 0;
228 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
229 &pcbinfo->ipi_hashmask);
230 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
231 &pcbinfo->ipi_porthashmask);
232 #ifdef PCBGROUP
233 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
234 #endif
235 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
236 NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
237 inpcbzone_flags);
238 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
239 uma_zone_set_warning(pcbinfo->ipi_zone,
240 "kern.ipc.maxsockets limit reached");
241 }
242
243 /*
244 * Destroy an inpcbinfo.
245 */
246 void
247 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
248 {
249
250 KASSERT(pcbinfo->ipi_count == 0,
251 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
252
253 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
254 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
255 pcbinfo->ipi_porthashmask);
256 #ifdef PCBGROUP
257 in_pcbgroup_destroy(pcbinfo);
258 #endif
259 uma_zdestroy(pcbinfo->ipi_zone);
260 INP_HASH_LOCK_DESTROY(pcbinfo);
261 INP_INFO_LOCK_DESTROY(pcbinfo);
262 }
263
264 /*
265 * Allocate a PCB and associate it with the socket.
266 * On success return with the PCB locked.
267 */
268 int
269 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
270 {
271 struct inpcb *inp;
272 int error;
273
274 INP_INFO_WLOCK_ASSERT(pcbinfo);
275 error = 0;
276 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
277 if (inp == NULL)
278 return (ENOBUFS);
279 bzero(inp, inp_zero_size);
280 inp->inp_pcbinfo = pcbinfo;
281 inp->inp_socket = so;
282 inp->inp_cred = crhold(so->so_cred);
283 inp->inp_inc.inc_fibnum = so->so_fibnum;
284 #ifdef MAC
285 error = mac_inpcb_init(inp, M_NOWAIT);
286 if (error != 0)
287 goto out;
288 mac_inpcb_create(so, inp);
289 #endif
290 #ifdef IPSEC
291 error = ipsec_init_policy(so, &inp->inp_sp);
292 if (error != 0) {
293 #ifdef MAC
294 mac_inpcb_destroy(inp);
295 #endif
296 goto out;
297 }
298 #endif /*IPSEC*/
299 #ifdef INET6
300 if (INP_SOCKAF(so) == AF_INET6) {
301 inp->inp_vflag |= INP_IPV6PROTO;
302 if (V_ip6_v6only)
303 inp->inp_flags |= IN6P_IPV6_V6ONLY;
304 }
305 #endif
306 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
307 pcbinfo->ipi_count++;
308 so->so_pcb = (caddr_t)inp;
309 #ifdef INET6
310 if (V_ip6_auto_flowlabel)
311 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
312 #endif
313 INP_WLOCK(inp);
314 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
315 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
316 #if defined(IPSEC) || defined(MAC)
317 out:
318 if (error != 0) {
319 crfree(inp->inp_cred);
320 uma_zfree(pcbinfo->ipi_zone, inp);
321 }
322 #endif
323 return (error);
324 }
325
326 #ifdef INET
327 int
328 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
329 {
330 int anonport, error;
331
332 INP_WLOCK_ASSERT(inp);
333 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
334
335 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
336 return (EINVAL);
337 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
338 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
339 &inp->inp_lport, cred);
340 if (error)
341 return (error);
342 if (in_pcbinshash(inp) != 0) {
343 inp->inp_laddr.s_addr = INADDR_ANY;
344 inp->inp_lport = 0;
345 return (EAGAIN);
346 }
347 if (anonport)
348 inp->inp_flags |= INP_ANONPORT;
349 return (0);
350 }
351 #endif
352
353 #if defined(INET) || defined(INET6)
354 int
355 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
356 struct ucred *cred, int lookupflags)
357 {
358 struct inpcbinfo *pcbinfo;
359 struct inpcb *tmpinp;
360 unsigned short *lastport;
361 int count, dorandom, error;
362 u_short aux, first, last, lport;
363 #ifdef INET
364 struct in_addr laddr;
365 #endif
366
367 pcbinfo = inp->inp_pcbinfo;
368
369 /*
370 * Because no actual state changes occur here, a global write lock on
371 * the pcbinfo isn't required.
372 */
373 INP_LOCK_ASSERT(inp);
374 INP_HASH_LOCK_ASSERT(pcbinfo);
375
376 if (inp->inp_flags & INP_HIGHPORT) {
377 first = V_ipport_hifirstauto; /* sysctl */
378 last = V_ipport_hilastauto;
379 lastport = &pcbinfo->ipi_lasthi;
380 } else if (inp->inp_flags & INP_LOWPORT) {
381 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
382 if (error)
383 return (error);
384 first = V_ipport_lowfirstauto; /* 1023 */
385 last = V_ipport_lowlastauto; /* 600 */
386 lastport = &pcbinfo->ipi_lastlow;
387 } else {
388 first = V_ipport_firstauto; /* sysctl */
389 last = V_ipport_lastauto;
390 lastport = &pcbinfo->ipi_lastport;
391 }
392 /*
393 * For UDP, use random port allocation as long as the user
394 * allows it. For TCP (and as of yet unknown) connections,
395 * use random port allocation only if the user allows it AND
396 * ipport_tick() allows it.
397 */
398 if (V_ipport_randomized &&
399 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo))
400 dorandom = 1;
401 else
402 dorandom = 0;
403 /*
404 * It makes no sense to do random port allocation if
405 * we have the only port available.
406 */
407 if (first == last)
408 dorandom = 0;
409 /* Make sure to not include UDP packets in the count. */
410 if (pcbinfo != &V_udbinfo)
411 V_ipport_tcpallocs++;
412 /*
413 * Instead of having two loops further down counting up or down
414 * make sure that first is always <= last and go with only one
415 * code path implementing all logic.
416 */
417 if (first > last) {
418 aux = first;
419 first = last;
420 last = aux;
421 }
422
423 #ifdef INET
424 /* Make the compiler happy. */
425 laddr.s_addr = 0;
426 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
427 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
428 __func__, inp));
429 laddr = *laddrp;
430 }
431 #endif
432 tmpinp = NULL; /* Make compiler happy. */
433 lport = *lportp;
434
435 if (dorandom)
436 *lastport = first + (arc4random() % (last - first));
437
438 count = last - first;
439
440 do {
441 if (count-- < 0) /* completely used? */
442 return (EADDRNOTAVAIL);
443 ++*lastport;
444 if (*lastport < first || *lastport > last)
445 *lastport = first;
446 lport = htons(*lastport);
447
448 #ifdef INET6
449 if ((inp->inp_vflag & INP_IPV6) != 0)
450 tmpinp = in6_pcblookup_local(pcbinfo,
451 &inp->in6p_laddr, lport, lookupflags, cred);
452 #endif
453 #if defined(INET) && defined(INET6)
454 else
455 #endif
456 #ifdef INET
457 tmpinp = in_pcblookup_local(pcbinfo, laddr,
458 lport, lookupflags, cred);
459 #endif
460 } while (tmpinp != NULL);
461
462 #ifdef INET
463 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
464 laddrp->s_addr = laddr.s_addr;
465 #endif
466 *lportp = lport;
467
468 return (0);
469 }
470
471 /*
472 * Return cached socket options.
473 */
474 short
475 inp_so_options(const struct inpcb *inp)
476 {
477 short so_options;
478
479 so_options = 0;
480
481 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
482 so_options |= SO_REUSEPORT;
483 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
484 so_options |= SO_REUSEADDR;
485 return (so_options);
486 }
487 #endif /* INET || INET6 */
488
489 #ifdef INET
490 /*
491 * Set up a bind operation on a PCB, performing port allocation
492 * as required, but do not actually modify the PCB. Callers can
493 * either complete the bind by setting inp_laddr/inp_lport and
494 * calling in_pcbinshash(), or they can just use the resulting
495 * port and address to authorise the sending of a once-off packet.
496 *
497 * On error, the values of *laddrp and *lportp are not changed.
498 */
499 int
500 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
501 u_short *lportp, struct ucred *cred)
502 {
503 struct socket *so = inp->inp_socket;
504 struct sockaddr_in *sin;
505 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
506 struct in_addr laddr;
507 u_short lport = 0;
508 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
509 int error;
510
511 /*
512 * No state changes, so read locks are sufficient here.
513 */
514 INP_LOCK_ASSERT(inp);
515 INP_HASH_LOCK_ASSERT(pcbinfo);
516
517 if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
518 return (EADDRNOTAVAIL);
519 laddr.s_addr = *laddrp;
520 if (nam != NULL && laddr.s_addr != INADDR_ANY)
521 return (EINVAL);
522 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
523 lookupflags = INPLOOKUP_WILDCARD;
524 if (nam == NULL) {
525 if ((error = prison_local_ip4(cred, &laddr)) != 0)
526 return (error);
527 } else {
528 sin = (struct sockaddr_in *)nam;
529 if (nam->sa_len != sizeof (*sin))
530 return (EINVAL);
531 #ifdef notdef
532 /*
533 * We should check the family, but old programs
534 * incorrectly fail to initialize it.
535 */
536 if (sin->sin_family != AF_INET)
537 return (EAFNOSUPPORT);
538 #endif
539 error = prison_local_ip4(cred, &sin->sin_addr);
540 if (error)
541 return (error);
542 if (sin->sin_port != *lportp) {
543 /* Don't allow the port to change. */
544 if (*lportp != 0)
545 return (EINVAL);
546 lport = sin->sin_port;
547 }
548 /* NB: lport is left as 0 if the port isn't being changed. */
549 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
550 /*
551 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
552 * allow complete duplication of binding if
553 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
554 * and a multicast address is bound on both
555 * new and duplicated sockets.
556 */
557 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
558 reuseport = SO_REUSEADDR|SO_REUSEPORT;
559 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
560 sin->sin_port = 0; /* yech... */
561 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
562 /*
563 * Is the address a local IP address?
564 * If INP_BINDANY is set, then the socket may be bound
565 * to any endpoint address, local or not.
566 */
567 if ((inp->inp_flags & INP_BINDANY) == 0 &&
568 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
569 return (EADDRNOTAVAIL);
570 }
571 laddr = sin->sin_addr;
572 if (lport) {
573 struct inpcb *t;
574 struct tcptw *tw;
575
576 /* GROSS */
577 if (ntohs(lport) <= V_ipport_reservedhigh &&
578 ntohs(lport) >= V_ipport_reservedlow &&
579 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
580 0))
581 return (EACCES);
582 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
583 priv_check_cred(inp->inp_cred,
584 PRIV_NETINET_REUSEPORT, 0) != 0) {
585 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
586 lport, INPLOOKUP_WILDCARD, cred);
587 /*
588 * XXX
589 * This entire block sorely needs a rewrite.
590 */
591 if (t &&
592 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
593 (so->so_type != SOCK_STREAM ||
594 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
595 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
596 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
597 (t->inp_flags2 & INP_REUSEPORT) == 0) &&
598 (inp->inp_cred->cr_uid !=
599 t->inp_cred->cr_uid))
600 return (EADDRINUSE);
601 }
602 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
603 lport, lookupflags, cred);
604 if (t && (t->inp_flags & INP_TIMEWAIT)) {
605 /*
606 * XXXRW: If an incpb has had its timewait
607 * state recycled, we treat the address as
608 * being in use (for now). This is better
609 * than a panic, but not desirable.
610 */
611 tw = intotw(t);
612 if (tw == NULL ||
613 (reuseport & tw->tw_so_options) == 0)
614 return (EADDRINUSE);
615 } else if (t && (reuseport & inp_so_options(t)) == 0) {
616 #ifdef INET6
617 if (ntohl(sin->sin_addr.s_addr) !=
618 INADDR_ANY ||
619 ntohl(t->inp_laddr.s_addr) !=
620 INADDR_ANY ||
621 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
622 (t->inp_vflag & INP_IPV6PROTO) == 0)
623 #endif
624 return (EADDRINUSE);
625 }
626 }
627 }
628 if (*lportp != 0)
629 lport = *lportp;
630 if (lport == 0) {
631 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
632 if (error != 0)
633 return (error);
634
635 }
636 *laddrp = laddr.s_addr;
637 *lportp = lport;
638 return (0);
639 }
640
641 /*
642 * Connect from a socket to a specified address.
643 * Both address and port must be specified in argument sin.
644 * If don't have a local address for this socket yet,
645 * then pick one.
646 */
647 int
648 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
649 struct ucred *cred, struct mbuf *m)
650 {
651 u_short lport, fport;
652 in_addr_t laddr, faddr;
653 int anonport, error;
654
655 INP_WLOCK_ASSERT(inp);
656 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
657
658 lport = inp->inp_lport;
659 laddr = inp->inp_laddr.s_addr;
660 anonport = (lport == 0);
661 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
662 NULL, cred);
663 if (error)
664 return (error);
665
666 /* Do the initial binding of the local address if required. */
667 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
668 inp->inp_lport = lport;
669 inp->inp_laddr.s_addr = laddr;
670 if (in_pcbinshash(inp) != 0) {
671 inp->inp_laddr.s_addr = INADDR_ANY;
672 inp->inp_lport = 0;
673 return (EAGAIN);
674 }
675 }
676
677 /* Commit the remaining changes. */
678 inp->inp_lport = lport;
679 inp->inp_laddr.s_addr = laddr;
680 inp->inp_faddr.s_addr = faddr;
681 inp->inp_fport = fport;
682 in_pcbrehash_mbuf(inp, m);
683
684 if (anonport)
685 inp->inp_flags |= INP_ANONPORT;
686 return (0);
687 }
688
689 int
690 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
691 {
692
693 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
694 }
695
696 /*
697 * Do proper source address selection on an unbound socket in case
698 * of connect. Take jails into account as well.
699 */
700 static int
701 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
702 struct ucred *cred)
703 {
704 struct ifaddr *ifa;
705 struct sockaddr *sa;
706 struct sockaddr_in *sin;
707 struct route sro;
708 int error;
709
710 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
711
712 /*
713 * Bypass source address selection and use the primary jail IP
714 * if requested.
715 */
716 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
717 return (0);
718
719 error = 0;
720 bzero(&sro, sizeof(sro));
721
722 sin = (struct sockaddr_in *)&sro.ro_dst;
723 sin->sin_family = AF_INET;
724 sin->sin_len = sizeof(struct sockaddr_in);
725 sin->sin_addr.s_addr = faddr->s_addr;
726
727 /*
728 * If route is known our src addr is taken from the i/f,
729 * else punt.
730 *
731 * Find out route to destination.
732 */
733 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
734 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
735
736 /*
737 * If we found a route, use the address corresponding to
738 * the outgoing interface.
739 *
740 * Otherwise assume faddr is reachable on a directly connected
741 * network and try to find a corresponding interface to take
742 * the source address from.
743 */
744 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
745 struct in_ifaddr *ia;
746 struct ifnet *ifp;
747
748 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin));
749 if (ia == NULL)
750 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0));
751 if (ia == NULL) {
752 error = ENETUNREACH;
753 goto done;
754 }
755
756 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
757 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
758 ifa_free(&ia->ia_ifa);
759 goto done;
760 }
761
762 ifp = ia->ia_ifp;
763 ifa_free(&ia->ia_ifa);
764 ia = NULL;
765 IF_ADDR_RLOCK(ifp);
766 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
767
768 sa = ifa->ifa_addr;
769 if (sa->sa_family != AF_INET)
770 continue;
771 sin = (struct sockaddr_in *)sa;
772 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
773 ia = (struct in_ifaddr *)ifa;
774 break;
775 }
776 }
777 if (ia != NULL) {
778 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
779 IF_ADDR_RUNLOCK(ifp);
780 goto done;
781 }
782 IF_ADDR_RUNLOCK(ifp);
783
784 /* 3. As a last resort return the 'default' jail address. */
785 error = prison_get_ip4(cred, laddr);
786 goto done;
787 }
788
789 /*
790 * If the outgoing interface on the route found is not
791 * a loopback interface, use the address from that interface.
792 * In case of jails do those three steps:
793 * 1. check if the interface address belongs to the jail. If so use it.
794 * 2. check if we have any address on the outgoing interface
795 * belonging to this jail. If so use it.
796 * 3. as a last resort return the 'default' jail address.
797 */
798 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
799 struct in_ifaddr *ia;
800 struct ifnet *ifp;
801
802 /* If not jailed, use the default returned. */
803 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
804 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
805 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
806 goto done;
807 }
808
809 /* Jailed. */
810 /* 1. Check if the iface address belongs to the jail. */
811 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
812 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
813 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
814 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
815 goto done;
816 }
817
818 /*
819 * 2. Check if we have any address on the outgoing interface
820 * belonging to this jail.
821 */
822 ia = NULL;
823 ifp = sro.ro_rt->rt_ifp;
824 IF_ADDR_RLOCK(ifp);
825 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
826 sa = ifa->ifa_addr;
827 if (sa->sa_family != AF_INET)
828 continue;
829 sin = (struct sockaddr_in *)sa;
830 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
831 ia = (struct in_ifaddr *)ifa;
832 break;
833 }
834 }
835 if (ia != NULL) {
836 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
837 IF_ADDR_RUNLOCK(ifp);
838 goto done;
839 }
840 IF_ADDR_RUNLOCK(ifp);
841
842 /* 3. As a last resort return the 'default' jail address. */
843 error = prison_get_ip4(cred, laddr);
844 goto done;
845 }
846
847 /*
848 * The outgoing interface is marked with 'loopback net', so a route
849 * to ourselves is here.
850 * Try to find the interface of the destination address and then
851 * take the address from there. That interface is not necessarily
852 * a loopback interface.
853 * In case of jails, check that it is an address of the jail
854 * and if we cannot find, fall back to the 'default' jail address.
855 */
856 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
857 struct sockaddr_in sain;
858 struct in_ifaddr *ia;
859
860 bzero(&sain, sizeof(struct sockaddr_in));
861 sain.sin_family = AF_INET;
862 sain.sin_len = sizeof(struct sockaddr_in);
863 sain.sin_addr.s_addr = faddr->s_addr;
864
865 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain)));
866 if (ia == NULL)
867 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0));
868 if (ia == NULL)
869 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
870
871 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
872 if (ia == NULL) {
873 error = ENETUNREACH;
874 goto done;
875 }
876 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
877 ifa_free(&ia->ia_ifa);
878 goto done;
879 }
880
881 /* Jailed. */
882 if (ia != NULL) {
883 struct ifnet *ifp;
884
885 ifp = ia->ia_ifp;
886 ifa_free(&ia->ia_ifa);
887 ia = NULL;
888 IF_ADDR_RLOCK(ifp);
889 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
890
891 sa = ifa->ifa_addr;
892 if (sa->sa_family != AF_INET)
893 continue;
894 sin = (struct sockaddr_in *)sa;
895 if (prison_check_ip4(cred,
896 &sin->sin_addr) == 0) {
897 ia = (struct in_ifaddr *)ifa;
898 break;
899 }
900 }
901 if (ia != NULL) {
902 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
903 IF_ADDR_RUNLOCK(ifp);
904 goto done;
905 }
906 IF_ADDR_RUNLOCK(ifp);
907 }
908
909 /* 3. As a last resort return the 'default' jail address. */
910 error = prison_get_ip4(cred, laddr);
911 goto done;
912 }
913
914 done:
915 if (sro.ro_rt != NULL)
916 RTFREE(sro.ro_rt);
917 return (error);
918 }
919
920 /*
921 * Set up for a connect from a socket to the specified address.
922 * On entry, *laddrp and *lportp should contain the current local
923 * address and port for the PCB; these are updated to the values
924 * that should be placed in inp_laddr and inp_lport to complete
925 * the connect.
926 *
927 * On success, *faddrp and *fportp will be set to the remote address
928 * and port. These are not updated in the error case.
929 *
930 * If the operation fails because the connection already exists,
931 * *oinpp will be set to the PCB of that connection so that the
932 * caller can decide to override it. In all other cases, *oinpp
933 * is set to NULL.
934 */
935 int
936 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
937 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
938 struct inpcb **oinpp, struct ucred *cred)
939 {
940 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
941 struct in_ifaddr *ia;
942 struct inpcb *oinp;
943 struct in_addr laddr, faddr;
944 u_short lport, fport;
945 int error;
946
947 /*
948 * Because a global state change doesn't actually occur here, a read
949 * lock is sufficient.
950 */
951 INP_LOCK_ASSERT(inp);
952 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
953
954 if (oinpp != NULL)
955 *oinpp = NULL;
956 if (nam->sa_len != sizeof (*sin))
957 return (EINVAL);
958 if (sin->sin_family != AF_INET)
959 return (EAFNOSUPPORT);
960 if (sin->sin_port == 0)
961 return (EADDRNOTAVAIL);
962 laddr.s_addr = *laddrp;
963 lport = *lportp;
964 faddr = sin->sin_addr;
965 fport = sin->sin_port;
966
967 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
968 /*
969 * If the destination address is INADDR_ANY,
970 * use the primary local address.
971 * If the supplied address is INADDR_BROADCAST,
972 * and the primary interface supports broadcast,
973 * choose the broadcast address for that interface.
974 */
975 if (faddr.s_addr == INADDR_ANY) {
976 IN_IFADDR_RLOCK();
977 faddr =
978 IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
979 IN_IFADDR_RUNLOCK();
980 if (cred != NULL &&
981 (error = prison_get_ip4(cred, &faddr)) != 0)
982 return (error);
983 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
984 IN_IFADDR_RLOCK();
985 if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
986 IFF_BROADCAST)
987 faddr = satosin(&TAILQ_FIRST(
988 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
989 IN_IFADDR_RUNLOCK();
990 }
991 }
992 if (laddr.s_addr == INADDR_ANY) {
993 error = in_pcbladdr(inp, &faddr, &laddr, cred);
994 /*
995 * If the destination address is multicast and an outgoing
996 * interface has been set as a multicast option, prefer the
997 * address of that interface as our source address.
998 */
999 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1000 inp->inp_moptions != NULL) {
1001 struct ip_moptions *imo;
1002 struct ifnet *ifp;
1003
1004 imo = inp->inp_moptions;
1005 if (imo->imo_multicast_ifp != NULL) {
1006 ifp = imo->imo_multicast_ifp;
1007 IN_IFADDR_RLOCK();
1008 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1009 if ((ia->ia_ifp == ifp) &&
1010 (cred == NULL ||
1011 prison_check_ip4(cred,
1012 &ia->ia_addr.sin_addr) == 0))
1013 break;
1014 }
1015 if (ia == NULL)
1016 error = EADDRNOTAVAIL;
1017 else {
1018 laddr = ia->ia_addr.sin_addr;
1019 error = 0;
1020 }
1021 IN_IFADDR_RUNLOCK();
1022 }
1023 }
1024 if (error)
1025 return (error);
1026 }
1027 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1028 laddr, lport, 0, NULL);
1029 if (oinp != NULL) {
1030 if (oinpp != NULL)
1031 *oinpp = oinp;
1032 return (EADDRINUSE);
1033 }
1034 if (lport == 0) {
1035 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1036 cred);
1037 if (error)
1038 return (error);
1039 }
1040 *laddrp = laddr.s_addr;
1041 *lportp = lport;
1042 *faddrp = faddr.s_addr;
1043 *fportp = fport;
1044 return (0);
1045 }
1046
1047 void
1048 in_pcbdisconnect(struct inpcb *inp)
1049 {
1050
1051 INP_WLOCK_ASSERT(inp);
1052 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1053
1054 inp->inp_faddr.s_addr = INADDR_ANY;
1055 inp->inp_fport = 0;
1056 in_pcbrehash(inp);
1057 }
1058 #endif /* INET */
1059
1060 /*
1061 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1062 * For most protocols, this will be invoked immediately prior to calling
1063 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1064 * socket, in which case in_pcbfree() is deferred.
1065 */
1066 void
1067 in_pcbdetach(struct inpcb *inp)
1068 {
1069
1070 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1071
1072 inp->inp_socket->so_pcb = NULL;
1073 inp->inp_socket = NULL;
1074 }
1075
1076 /*
1077 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1078 * stability of an inpcb pointer despite the inpcb lock being released. This
1079 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1080 * but where the inpcb lock may already held, or when acquiring a reference
1081 * via a pcbgroup.
1082 *
1083 * in_pcbref() should be used only to provide brief memory stability, and
1084 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1085 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1086 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1087 * lock and rele are the *only* safe operations that may be performed on the
1088 * inpcb.
1089 *
1090 * While the inpcb will not be freed, releasing the inpcb lock means that the
1091 * connection's state may change, so the caller should be careful to
1092 * revalidate any cached state on reacquiring the lock. Drop the reference
1093 * using in_pcbrele().
1094 */
1095 void
1096 in_pcbref(struct inpcb *inp)
1097 {
1098
1099 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1100
1101 refcount_acquire(&inp->inp_refcount);
1102 }
1103
1104 /*
1105 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1106 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1107 * return a flag indicating whether or not the inpcb remains valid. If it is
1108 * valid, we return with the inpcb lock held.
1109 *
1110 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1111 * reference on an inpcb. Historically more work was done here (actually, in
1112 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1113 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1114 * about memory stability (and continued use of the write lock).
1115 */
1116 int
1117 in_pcbrele_rlocked(struct inpcb *inp)
1118 {
1119 struct inpcbinfo *pcbinfo;
1120
1121 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1122
1123 INP_RLOCK_ASSERT(inp);
1124
1125 if (refcount_release(&inp->inp_refcount) == 0) {
1126 /*
1127 * If the inpcb has been freed, let the caller know, even if
1128 * this isn't the last reference.
1129 */
1130 if (inp->inp_flags2 & INP_FREED) {
1131 INP_RUNLOCK(inp);
1132 return (1);
1133 }
1134 return (0);
1135 }
1136
1137 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1138
1139 INP_RUNLOCK(inp);
1140 pcbinfo = inp->inp_pcbinfo;
1141 uma_zfree(pcbinfo->ipi_zone, inp);
1142 return (1);
1143 }
1144
1145 int
1146 in_pcbrele_wlocked(struct inpcb *inp)
1147 {
1148 struct inpcbinfo *pcbinfo;
1149
1150 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1151
1152 INP_WLOCK_ASSERT(inp);
1153
1154 if (refcount_release(&inp->inp_refcount) == 0)
1155 return (0);
1156
1157 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1158
1159 INP_WUNLOCK(inp);
1160 pcbinfo = inp->inp_pcbinfo;
1161 uma_zfree(pcbinfo->ipi_zone, inp);
1162 return (1);
1163 }
1164
1165 /*
1166 * Temporary wrapper.
1167 */
1168 int
1169 in_pcbrele(struct inpcb *inp)
1170 {
1171
1172 return (in_pcbrele_wlocked(inp));
1173 }
1174
1175 /*
1176 * Unconditionally schedule an inpcb to be freed by decrementing its
1177 * reference count, which should occur only after the inpcb has been detached
1178 * from its socket. If another thread holds a temporary reference (acquired
1179 * using in_pcbref()) then the free is deferred until that reference is
1180 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1181 * work, including removal from global lists, is done in this context, where
1182 * the pcbinfo lock is held.
1183 */
1184 void
1185 in_pcbfree(struct inpcb *inp)
1186 {
1187 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1188
1189 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1190
1191 INP_INFO_WLOCK_ASSERT(pcbinfo);
1192 INP_WLOCK_ASSERT(inp);
1193
1194 /* XXXRW: Do as much as possible here. */
1195 #ifdef IPSEC
1196 if (inp->inp_sp != NULL)
1197 ipsec_delete_pcbpolicy(inp);
1198 #endif
1199 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1200 in_pcbremlists(inp);
1201 #ifdef INET6
1202 if (inp->inp_vflag & INP_IPV6PROTO) {
1203 ip6_freepcbopts(inp->in6p_outputopts);
1204 if (inp->in6p_moptions != NULL)
1205 ip6_freemoptions(inp->in6p_moptions);
1206 }
1207 #endif
1208 if (inp->inp_options)
1209 (void)m_free(inp->inp_options);
1210 #ifdef INET
1211 if (inp->inp_moptions != NULL)
1212 inp_freemoptions(inp->inp_moptions);
1213 #endif
1214 inp->inp_vflag = 0;
1215 inp->inp_flags2 |= INP_FREED;
1216 crfree(inp->inp_cred);
1217 #ifdef MAC
1218 mac_inpcb_destroy(inp);
1219 #endif
1220 if (!in_pcbrele_wlocked(inp))
1221 INP_WUNLOCK(inp);
1222 }
1223
1224 /*
1225 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1226 * port reservation, and preventing it from being returned by inpcb lookups.
1227 *
1228 * It is used by TCP to mark an inpcb as unused and avoid future packet
1229 * delivery or event notification when a socket remains open but TCP has
1230 * closed. This might occur as a result of a shutdown()-initiated TCP close
1231 * or a RST on the wire, and allows the port binding to be reused while still
1232 * maintaining the invariant that so_pcb always points to a valid inpcb until
1233 * in_pcbdetach().
1234 *
1235 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1236 * in_pcbnotifyall() and in_pcbpurgeif0()?
1237 */
1238 void
1239 in_pcbdrop(struct inpcb *inp)
1240 {
1241
1242 INP_WLOCK_ASSERT(inp);
1243
1244 /*
1245 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1246 * the hash lock...?
1247 */
1248 inp->inp_flags |= INP_DROPPED;
1249 if (inp->inp_flags & INP_INHASHLIST) {
1250 struct inpcbport *phd = inp->inp_phd;
1251
1252 INP_HASH_WLOCK(inp->inp_pcbinfo);
1253 LIST_REMOVE(inp, inp_hash);
1254 LIST_REMOVE(inp, inp_portlist);
1255 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1256 LIST_REMOVE(phd, phd_hash);
1257 free(phd, M_PCB);
1258 }
1259 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1260 inp->inp_flags &= ~INP_INHASHLIST;
1261 #ifdef PCBGROUP
1262 in_pcbgroup_remove(inp);
1263 #endif
1264 }
1265 }
1266
1267 #ifdef INET
1268 /*
1269 * Common routines to return the socket addresses associated with inpcbs.
1270 */
1271 struct sockaddr *
1272 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1273 {
1274 struct sockaddr_in *sin;
1275
1276 sin = malloc(sizeof *sin, M_SONAME,
1277 M_WAITOK | M_ZERO);
1278 sin->sin_family = AF_INET;
1279 sin->sin_len = sizeof(*sin);
1280 sin->sin_addr = *addr_p;
1281 sin->sin_port = port;
1282
1283 return (struct sockaddr *)sin;
1284 }
1285
1286 int
1287 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1288 {
1289 struct inpcb *inp;
1290 struct in_addr addr;
1291 in_port_t port;
1292
1293 inp = sotoinpcb(so);
1294 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1295
1296 INP_RLOCK(inp);
1297 port = inp->inp_lport;
1298 addr = inp->inp_laddr;
1299 INP_RUNLOCK(inp);
1300
1301 *nam = in_sockaddr(port, &addr);
1302 return 0;
1303 }
1304
1305 int
1306 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1307 {
1308 struct inpcb *inp;
1309 struct in_addr addr;
1310 in_port_t port;
1311
1312 inp = sotoinpcb(so);
1313 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1314
1315 INP_RLOCK(inp);
1316 port = inp->inp_fport;
1317 addr = inp->inp_faddr;
1318 INP_RUNLOCK(inp);
1319
1320 *nam = in_sockaddr(port, &addr);
1321 return 0;
1322 }
1323
1324 void
1325 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1326 struct inpcb *(*notify)(struct inpcb *, int))
1327 {
1328 struct inpcb *inp, *inp_temp;
1329
1330 INP_INFO_WLOCK(pcbinfo);
1331 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1332 INP_WLOCK(inp);
1333 #ifdef INET6
1334 if ((inp->inp_vflag & INP_IPV4) == 0) {
1335 INP_WUNLOCK(inp);
1336 continue;
1337 }
1338 #endif
1339 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1340 inp->inp_socket == NULL) {
1341 INP_WUNLOCK(inp);
1342 continue;
1343 }
1344 if ((*notify)(inp, errno))
1345 INP_WUNLOCK(inp);
1346 }
1347 INP_INFO_WUNLOCK(pcbinfo);
1348 }
1349
1350 void
1351 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1352 {
1353 struct inpcb *inp;
1354 struct ip_moptions *imo;
1355 int i, gap;
1356
1357 INP_INFO_RLOCK(pcbinfo);
1358 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1359 INP_WLOCK(inp);
1360 imo = inp->inp_moptions;
1361 if ((inp->inp_vflag & INP_IPV4) &&
1362 imo != NULL) {
1363 /*
1364 * Unselect the outgoing interface if it is being
1365 * detached.
1366 */
1367 if (imo->imo_multicast_ifp == ifp)
1368 imo->imo_multicast_ifp = NULL;
1369
1370 /*
1371 * Drop multicast group membership if we joined
1372 * through the interface being detached.
1373 */
1374 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1375 i++) {
1376 if (imo->imo_membership[i]->inm_ifp == ifp) {
1377 in_delmulti(imo->imo_membership[i]);
1378 gap++;
1379 } else if (gap != 0)
1380 imo->imo_membership[i - gap] =
1381 imo->imo_membership[i];
1382 }
1383 imo->imo_num_memberships -= gap;
1384 }
1385 INP_WUNLOCK(inp);
1386 }
1387 INP_INFO_RUNLOCK(pcbinfo);
1388 }
1389
1390 /*
1391 * Lookup a PCB based on the local address and port. Caller must hold the
1392 * hash lock. No inpcb locks or references are acquired.
1393 */
1394 #define INP_LOOKUP_MAPPED_PCB_COST 3
1395 struct inpcb *
1396 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1397 u_short lport, int lookupflags, struct ucred *cred)
1398 {
1399 struct inpcb *inp;
1400 #ifdef INET6
1401 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1402 #else
1403 int matchwild = 3;
1404 #endif
1405 int wildcard;
1406
1407 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1408 ("%s: invalid lookup flags %d", __func__, lookupflags));
1409
1410 INP_HASH_LOCK_ASSERT(pcbinfo);
1411
1412 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1413 struct inpcbhead *head;
1414 /*
1415 * Look for an unconnected (wildcard foreign addr) PCB that
1416 * matches the local address and port we're looking for.
1417 */
1418 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1419 0, pcbinfo->ipi_hashmask)];
1420 LIST_FOREACH(inp, head, inp_hash) {
1421 #ifdef INET6
1422 /* XXX inp locking */
1423 if ((inp->inp_vflag & INP_IPV4) == 0)
1424 continue;
1425 #endif
1426 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1427 inp->inp_laddr.s_addr == laddr.s_addr &&
1428 inp->inp_lport == lport) {
1429 /*
1430 * Found?
1431 */
1432 if (cred == NULL ||
1433 prison_equal_ip4(cred->cr_prison,
1434 inp->inp_cred->cr_prison))
1435 return (inp);
1436 }
1437 }
1438 /*
1439 * Not found.
1440 */
1441 return (NULL);
1442 } else {
1443 struct inpcbporthead *porthash;
1444 struct inpcbport *phd;
1445 struct inpcb *match = NULL;
1446 /*
1447 * Best fit PCB lookup.
1448 *
1449 * First see if this local port is in use by looking on the
1450 * port hash list.
1451 */
1452 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1453 pcbinfo->ipi_porthashmask)];
1454 LIST_FOREACH(phd, porthash, phd_hash) {
1455 if (phd->phd_port == lport)
1456 break;
1457 }
1458 if (phd != NULL) {
1459 /*
1460 * Port is in use by one or more PCBs. Look for best
1461 * fit.
1462 */
1463 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1464 wildcard = 0;
1465 if (cred != NULL &&
1466 !prison_equal_ip4(inp->inp_cred->cr_prison,
1467 cred->cr_prison))
1468 continue;
1469 #ifdef INET6
1470 /* XXX inp locking */
1471 if ((inp->inp_vflag & INP_IPV4) == 0)
1472 continue;
1473 /*
1474 * We never select the PCB that has
1475 * INP_IPV6 flag and is bound to :: if
1476 * we have another PCB which is bound
1477 * to 0.0.0.0. If a PCB has the
1478 * INP_IPV6 flag, then we set its cost
1479 * higher than IPv4 only PCBs.
1480 *
1481 * Note that the case only happens
1482 * when a socket is bound to ::, under
1483 * the condition that the use of the
1484 * mapped address is allowed.
1485 */
1486 if ((inp->inp_vflag & INP_IPV6) != 0)
1487 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1488 #endif
1489 if (inp->inp_faddr.s_addr != INADDR_ANY)
1490 wildcard++;
1491 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1492 if (laddr.s_addr == INADDR_ANY)
1493 wildcard++;
1494 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1495 continue;
1496 } else {
1497 if (laddr.s_addr != INADDR_ANY)
1498 wildcard++;
1499 }
1500 if (wildcard < matchwild) {
1501 match = inp;
1502 matchwild = wildcard;
1503 if (matchwild == 0)
1504 break;
1505 }
1506 }
1507 }
1508 return (match);
1509 }
1510 }
1511 #undef INP_LOOKUP_MAPPED_PCB_COST
1512
1513 #ifdef PCBGROUP
1514 /*
1515 * Lookup PCB in hash list, using pcbgroup tables.
1516 */
1517 static struct inpcb *
1518 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1519 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1520 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1521 {
1522 struct inpcbhead *head;
1523 struct inpcb *inp, *tmpinp;
1524 u_short fport = fport_arg, lport = lport_arg;
1525
1526 /*
1527 * First look for an exact match.
1528 */
1529 tmpinp = NULL;
1530 INP_GROUP_LOCK(pcbgroup);
1531 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1532 pcbgroup->ipg_hashmask)];
1533 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1534 #ifdef INET6
1535 /* XXX inp locking */
1536 if ((inp->inp_vflag & INP_IPV4) == 0)
1537 continue;
1538 #endif
1539 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1540 inp->inp_laddr.s_addr == laddr.s_addr &&
1541 inp->inp_fport == fport &&
1542 inp->inp_lport == lport) {
1543 /*
1544 * XXX We should be able to directly return
1545 * the inp here, without any checks.
1546 * Well unless both bound with SO_REUSEPORT?
1547 */
1548 if (prison_flag(inp->inp_cred, PR_IP4))
1549 goto found;
1550 if (tmpinp == NULL)
1551 tmpinp = inp;
1552 }
1553 }
1554 if (tmpinp != NULL) {
1555 inp = tmpinp;
1556 goto found;
1557 }
1558
1559 /*
1560 * Then look for a wildcard match, if requested.
1561 */
1562 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1563 struct inpcb *local_wild = NULL, *local_exact = NULL;
1564 #ifdef INET6
1565 struct inpcb *local_wild_mapped = NULL;
1566 #endif
1567 struct inpcb *jail_wild = NULL;
1568 struct inpcbhead *head;
1569 int injail;
1570
1571 /*
1572 * Order of socket selection - we always prefer jails.
1573 * 1. jailed, non-wild.
1574 * 2. jailed, wild.
1575 * 3. non-jailed, non-wild.
1576 * 4. non-jailed, wild.
1577 */
1578 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1579 0, pcbinfo->ipi_wildmask)];
1580 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1581 #ifdef INET6
1582 /* XXX inp locking */
1583 if ((inp->inp_vflag & INP_IPV4) == 0)
1584 continue;
1585 #endif
1586 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1587 inp->inp_lport != lport)
1588 continue;
1589
1590 /* XXX inp locking */
1591 if (ifp && ifp->if_type == IFT_FAITH &&
1592 (inp->inp_flags & INP_FAITH) == 0)
1593 continue;
1594
1595 injail = prison_flag(inp->inp_cred, PR_IP4);
1596 if (injail) {
1597 if (prison_check_ip4(inp->inp_cred,
1598 &laddr) != 0)
1599 continue;
1600 } else {
1601 if (local_exact != NULL)
1602 continue;
1603 }
1604
1605 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1606 if (injail)
1607 goto found;
1608 else
1609 local_exact = inp;
1610 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1611 #ifdef INET6
1612 /* XXX inp locking, NULL check */
1613 if (inp->inp_vflag & INP_IPV6PROTO)
1614 local_wild_mapped = inp;
1615 else
1616 #endif
1617 if (injail)
1618 jail_wild = inp;
1619 else
1620 local_wild = inp;
1621 }
1622 } /* LIST_FOREACH */
1623 inp = jail_wild;
1624 if (inp == NULL)
1625 inp = local_exact;
1626 if (inp == NULL)
1627 inp = local_wild;
1628 #ifdef INET6
1629 if (inp == NULL)
1630 inp = local_wild_mapped;
1631 #endif
1632 if (inp != NULL)
1633 goto found;
1634 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1635 INP_GROUP_UNLOCK(pcbgroup);
1636 return (NULL);
1637
1638 found:
1639 in_pcbref(inp);
1640 INP_GROUP_UNLOCK(pcbgroup);
1641 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1642 INP_WLOCK(inp);
1643 if (in_pcbrele_wlocked(inp))
1644 return (NULL);
1645 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1646 INP_RLOCK(inp);
1647 if (in_pcbrele_rlocked(inp))
1648 return (NULL);
1649 } else
1650 panic("%s: locking bug", __func__);
1651 return (inp);
1652 }
1653 #endif /* PCBGROUP */
1654
1655 /*
1656 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
1657 * that the caller has locked the hash list, and will not perform any further
1658 * locking or reference operations on either the hash list or the connection.
1659 */
1660 static struct inpcb *
1661 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1662 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1663 struct ifnet *ifp)
1664 {
1665 struct inpcbhead *head;
1666 struct inpcb *inp, *tmpinp;
1667 u_short fport = fport_arg, lport = lport_arg;
1668
1669 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1670 ("%s: invalid lookup flags %d", __func__, lookupflags));
1671
1672 INP_HASH_LOCK_ASSERT(pcbinfo);
1673
1674 /*
1675 * First look for an exact match.
1676 */
1677 tmpinp = NULL;
1678 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1679 pcbinfo->ipi_hashmask)];
1680 LIST_FOREACH(inp, head, inp_hash) {
1681 #ifdef INET6
1682 /* XXX inp locking */
1683 if ((inp->inp_vflag & INP_IPV4) == 0)
1684 continue;
1685 #endif
1686 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1687 inp->inp_laddr.s_addr == laddr.s_addr &&
1688 inp->inp_fport == fport &&
1689 inp->inp_lport == lport) {
1690 /*
1691 * XXX We should be able to directly return
1692 * the inp here, without any checks.
1693 * Well unless both bound with SO_REUSEPORT?
1694 */
1695 if (prison_flag(inp->inp_cred, PR_IP4))
1696 return (inp);
1697 if (tmpinp == NULL)
1698 tmpinp = inp;
1699 }
1700 }
1701 if (tmpinp != NULL)
1702 return (tmpinp);
1703
1704 /*
1705 * Then look for a wildcard match, if requested.
1706 */
1707 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1708 struct inpcb *local_wild = NULL, *local_exact = NULL;
1709 #ifdef INET6
1710 struct inpcb *local_wild_mapped = NULL;
1711 #endif
1712 struct inpcb *jail_wild = NULL;
1713 int injail;
1714
1715 /*
1716 * Order of socket selection - we always prefer jails.
1717 * 1. jailed, non-wild.
1718 * 2. jailed, wild.
1719 * 3. non-jailed, non-wild.
1720 * 4. non-jailed, wild.
1721 */
1722
1723 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1724 0, pcbinfo->ipi_hashmask)];
1725 LIST_FOREACH(inp, head, inp_hash) {
1726 #ifdef INET6
1727 /* XXX inp locking */
1728 if ((inp->inp_vflag & INP_IPV4) == 0)
1729 continue;
1730 #endif
1731 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1732 inp->inp_lport != lport)
1733 continue;
1734
1735 /* XXX inp locking */
1736 if (ifp && ifp->if_type == IFT_FAITH &&
1737 (inp->inp_flags & INP_FAITH) == 0)
1738 continue;
1739
1740 injail = prison_flag(inp->inp_cred, PR_IP4);
1741 if (injail) {
1742 if (prison_check_ip4(inp->inp_cred,
1743 &laddr) != 0)
1744 continue;
1745 } else {
1746 if (local_exact != NULL)
1747 continue;
1748 }
1749
1750 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1751 if (injail)
1752 return (inp);
1753 else
1754 local_exact = inp;
1755 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1756 #ifdef INET6
1757 /* XXX inp locking, NULL check */
1758 if (inp->inp_vflag & INP_IPV6PROTO)
1759 local_wild_mapped = inp;
1760 else
1761 #endif
1762 if (injail)
1763 jail_wild = inp;
1764 else
1765 local_wild = inp;
1766 }
1767 } /* LIST_FOREACH */
1768 if (jail_wild != NULL)
1769 return (jail_wild);
1770 if (local_exact != NULL)
1771 return (local_exact);
1772 if (local_wild != NULL)
1773 return (local_wild);
1774 #ifdef INET6
1775 if (local_wild_mapped != NULL)
1776 return (local_wild_mapped);
1777 #endif
1778 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1779
1780 return (NULL);
1781 }
1782
1783 /*
1784 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
1785 * hash list lock, and will return the inpcb locked (i.e., requires
1786 * INPLOOKUP_LOCKPCB).
1787 */
1788 static struct inpcb *
1789 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1790 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1791 struct ifnet *ifp)
1792 {
1793 struct inpcb *inp;
1794
1795 INP_HASH_RLOCK(pcbinfo);
1796 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1797 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1798 if (inp != NULL) {
1799 in_pcbref(inp);
1800 INP_HASH_RUNLOCK(pcbinfo);
1801 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1802 INP_WLOCK(inp);
1803 if (in_pcbrele_wlocked(inp))
1804 return (NULL);
1805 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1806 INP_RLOCK(inp);
1807 if (in_pcbrele_rlocked(inp))
1808 return (NULL);
1809 } else
1810 panic("%s: locking bug", __func__);
1811 } else
1812 INP_HASH_RUNLOCK(pcbinfo);
1813 return (inp);
1814 }
1815
1816 /*
1817 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1818 * from which a pre-calculated hash value may be extracted.
1819 *
1820 * Possibly more of this logic should be in in_pcbgroup.c.
1821 */
1822 struct inpcb *
1823 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1824 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1825 {
1826 #if defined(PCBGROUP)
1827 struct inpcbgroup *pcbgroup;
1828 #endif
1829
1830 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1831 ("%s: invalid lookup flags %d", __func__, lookupflags));
1832 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1833 ("%s: LOCKPCB not set", __func__));
1834
1835 #if defined(PCBGROUP)
1836 if (in_pcbgroup_enabled(pcbinfo)) {
1837 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1838 fport);
1839 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1840 laddr, lport, lookupflags, ifp));
1841 }
1842 #endif
1843 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1844 lookupflags, ifp));
1845 }
1846
1847 struct inpcb *
1848 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1849 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1850 struct ifnet *ifp, struct mbuf *m)
1851 {
1852 #ifdef PCBGROUP
1853 struct inpcbgroup *pcbgroup;
1854 #endif
1855
1856 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1857 ("%s: invalid lookup flags %d", __func__, lookupflags));
1858 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1859 ("%s: LOCKPCB not set", __func__));
1860
1861 #ifdef PCBGROUP
1862 if (in_pcbgroup_enabled(pcbinfo)) {
1863 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
1864 m->m_pkthdr.flowid);
1865 if (pcbgroup != NULL)
1866 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
1867 fport, laddr, lport, lookupflags, ifp));
1868 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1869 fport);
1870 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1871 laddr, lport, lookupflags, ifp));
1872 }
1873 #endif
1874 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1875 lookupflags, ifp));
1876 }
1877 #endif /* INET */
1878
1879 /*
1880 * Insert PCB onto various hash lists.
1881 */
1882 static int
1883 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
1884 {
1885 struct inpcbhead *pcbhash;
1886 struct inpcbporthead *pcbporthash;
1887 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1888 struct inpcbport *phd;
1889 u_int32_t hashkey_faddr;
1890
1891 INP_WLOCK_ASSERT(inp);
1892 INP_HASH_WLOCK_ASSERT(pcbinfo);
1893
1894 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
1895 ("in_pcbinshash: INP_INHASHLIST"));
1896
1897 #ifdef INET6
1898 if (inp->inp_vflag & INP_IPV6)
1899 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */;
1900 else
1901 #endif
1902 hashkey_faddr = inp->inp_faddr.s_addr;
1903
1904 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
1905 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
1906
1907 pcbporthash = &pcbinfo->ipi_porthashbase[
1908 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
1909
1910 /*
1911 * Go through port list and look for a head for this lport.
1912 */
1913 LIST_FOREACH(phd, pcbporthash, phd_hash) {
1914 if (phd->phd_port == inp->inp_lport)
1915 break;
1916 }
1917 /*
1918 * If none exists, malloc one and tack it on.
1919 */
1920 if (phd == NULL) {
1921 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
1922 if (phd == NULL) {
1923 return (ENOBUFS); /* XXX */
1924 }
1925 phd->phd_port = inp->inp_lport;
1926 LIST_INIT(&phd->phd_pcblist);
1927 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1928 }
1929 inp->inp_phd = phd;
1930 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1931 LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
1932 inp->inp_flags |= INP_INHASHLIST;
1933 #ifdef PCBGROUP
1934 if (do_pcbgroup_update)
1935 in_pcbgroup_update(inp);
1936 #endif
1937 return (0);
1938 }
1939
1940 /*
1941 * For now, there are two public interfaces to insert an inpcb into the hash
1942 * lists -- one that does update pcbgroups, and one that doesn't. The latter
1943 * is used only in the TCP syncache, where in_pcbinshash is called before the
1944 * full 4-tuple is set for the inpcb, and we don't want to install in the
1945 * pcbgroup until later.
1946 *
1947 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
1948 * connection groups, and partially initialised inpcbs should not be exposed
1949 * to either reservation hash tables or pcbgroups.
1950 */
1951 int
1952 in_pcbinshash(struct inpcb *inp)
1953 {
1954
1955 return (in_pcbinshash_internal(inp, 1));
1956 }
1957
1958 int
1959 in_pcbinshash_nopcbgroup(struct inpcb *inp)
1960 {
1961
1962 return (in_pcbinshash_internal(inp, 0));
1963 }
1964
1965 /*
1966 * Move PCB to the proper hash bucket when { faddr, fport } have been
1967 * changed. NOTE: This does not handle the case of the lport changing (the
1968 * hashed port list would have to be updated as well), so the lport must
1969 * not change after in_pcbinshash() has been called.
1970 */
1971 void
1972 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
1973 {
1974 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1975 struct inpcbhead *head;
1976 u_int32_t hashkey_faddr;
1977
1978 INP_WLOCK_ASSERT(inp);
1979 INP_HASH_WLOCK_ASSERT(pcbinfo);
1980
1981 KASSERT(inp->inp_flags & INP_INHASHLIST,
1982 ("in_pcbrehash: !INP_INHASHLIST"));
1983
1984 #ifdef INET6
1985 if (inp->inp_vflag & INP_IPV6)
1986 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */;
1987 else
1988 #endif
1989 hashkey_faddr = inp->inp_faddr.s_addr;
1990
1991 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
1992 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
1993
1994 LIST_REMOVE(inp, inp_hash);
1995 LIST_INSERT_HEAD(head, inp, inp_hash);
1996
1997 #ifdef PCBGROUP
1998 if (m != NULL)
1999 in_pcbgroup_update_mbuf(inp, m);
2000 else
2001 in_pcbgroup_update(inp);
2002 #endif
2003 }
2004
2005 void
2006 in_pcbrehash(struct inpcb *inp)
2007 {
2008
2009 in_pcbrehash_mbuf(inp, NULL);
2010 }
2011
2012 /*
2013 * Remove PCB from various lists.
2014 */
2015 static void
2016 in_pcbremlists(struct inpcb *inp)
2017 {
2018 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2019
2020 INP_INFO_WLOCK_ASSERT(pcbinfo);
2021 INP_WLOCK_ASSERT(inp);
2022
2023 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2024 if (inp->inp_flags & INP_INHASHLIST) {
2025 struct inpcbport *phd = inp->inp_phd;
2026
2027 INP_HASH_WLOCK(pcbinfo);
2028 LIST_REMOVE(inp, inp_hash);
2029 LIST_REMOVE(inp, inp_portlist);
2030 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2031 LIST_REMOVE(phd, phd_hash);
2032 free(phd, M_PCB);
2033 }
2034 INP_HASH_WUNLOCK(pcbinfo);
2035 inp->inp_flags &= ~INP_INHASHLIST;
2036 }
2037 LIST_REMOVE(inp, inp_list);
2038 pcbinfo->ipi_count--;
2039 #ifdef PCBGROUP
2040 in_pcbgroup_remove(inp);
2041 #endif
2042 }
2043
2044 /*
2045 * A set label operation has occurred at the socket layer, propagate the
2046 * label change into the in_pcb for the socket.
2047 */
2048 void
2049 in_pcbsosetlabel(struct socket *so)
2050 {
2051 #ifdef MAC
2052 struct inpcb *inp;
2053
2054 inp = sotoinpcb(so);
2055 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2056
2057 INP_WLOCK(inp);
2058 SOCK_LOCK(so);
2059 mac_inpcb_sosetlabel(so, inp);
2060 SOCK_UNLOCK(so);
2061 INP_WUNLOCK(inp);
2062 #endif
2063 }
2064
2065 /*
2066 * ipport_tick runs once per second, determining if random port allocation
2067 * should be continued. If more than ipport_randomcps ports have been
2068 * allocated in the last second, then we return to sequential port
2069 * allocation. We return to random allocation only once we drop below
2070 * ipport_randomcps for at least ipport_randomtime seconds.
2071 */
2072 static void
2073 ipport_tick(void *xtp)
2074 {
2075 VNET_ITERATOR_DECL(vnet_iter);
2076
2077 VNET_LIST_RLOCK_NOSLEEP();
2078 VNET_FOREACH(vnet_iter) {
2079 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2080 if (V_ipport_tcpallocs <=
2081 V_ipport_tcplastcount + V_ipport_randomcps) {
2082 if (V_ipport_stoprandom > 0)
2083 V_ipport_stoprandom--;
2084 } else
2085 V_ipport_stoprandom = V_ipport_randomtime;
2086 V_ipport_tcplastcount = V_ipport_tcpallocs;
2087 CURVNET_RESTORE();
2088 }
2089 VNET_LIST_RUNLOCK_NOSLEEP();
2090 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2091 }
2092
2093 static void
2094 ip_fini(void *xtp)
2095 {
2096
2097 callout_stop(&ipport_tick_callout);
2098 }
2099
2100 /*
2101 * The ipport_callout should start running at about the time we attach the
2102 * inet or inet6 domains.
2103 */
2104 static void
2105 ipport_tick_init(const void *unused __unused)
2106 {
2107
2108 /* Start ipport_tick. */
2109 callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
2110 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2111 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2112 SHUTDOWN_PRI_DEFAULT);
2113 }
2114 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2115 ipport_tick_init, NULL);
2116
2117 void
2118 inp_wlock(struct inpcb *inp)
2119 {
2120
2121 INP_WLOCK(inp);
2122 }
2123
2124 void
2125 inp_wunlock(struct inpcb *inp)
2126 {
2127
2128 INP_WUNLOCK(inp);
2129 }
2130
2131 void
2132 inp_rlock(struct inpcb *inp)
2133 {
2134
2135 INP_RLOCK(inp);
2136 }
2137
2138 void
2139 inp_runlock(struct inpcb *inp)
2140 {
2141
2142 INP_RUNLOCK(inp);
2143 }
2144
2145 #ifdef INVARIANTS
2146 void
2147 inp_lock_assert(struct inpcb *inp)
2148 {
2149
2150 INP_WLOCK_ASSERT(inp);
2151 }
2152
2153 void
2154 inp_unlock_assert(struct inpcb *inp)
2155 {
2156
2157 INP_UNLOCK_ASSERT(inp);
2158 }
2159 #endif
2160
2161 void
2162 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2163 {
2164 struct inpcb *inp;
2165
2166 INP_INFO_RLOCK(&V_tcbinfo);
2167 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2168 INP_WLOCK(inp);
2169 func(inp, arg);
2170 INP_WUNLOCK(inp);
2171 }
2172 INP_INFO_RUNLOCK(&V_tcbinfo);
2173 }
2174
2175 struct socket *
2176 inp_inpcbtosocket(struct inpcb *inp)
2177 {
2178
2179 INP_WLOCK_ASSERT(inp);
2180 return (inp->inp_socket);
2181 }
2182
2183 struct tcpcb *
2184 inp_inpcbtotcpcb(struct inpcb *inp)
2185 {
2186
2187 INP_WLOCK_ASSERT(inp);
2188 return ((struct tcpcb *)inp->inp_ppcb);
2189 }
2190
2191 int
2192 inp_ip_tos_get(const struct inpcb *inp)
2193 {
2194
2195 return (inp->inp_ip_tos);
2196 }
2197
2198 void
2199 inp_ip_tos_set(struct inpcb *inp, int val)
2200 {
2201
2202 inp->inp_ip_tos = val;
2203 }
2204
2205 void
2206 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2207 uint32_t *faddr, uint16_t *fp)
2208 {
2209
2210 INP_LOCK_ASSERT(inp);
2211 *laddr = inp->inp_laddr.s_addr;
2212 *faddr = inp->inp_faddr.s_addr;
2213 *lp = inp->inp_lport;
2214 *fp = inp->inp_fport;
2215 }
2216
2217 struct inpcb *
2218 so_sotoinpcb(struct socket *so)
2219 {
2220
2221 return (sotoinpcb(so));
2222 }
2223
2224 struct tcpcb *
2225 so_sototcpcb(struct socket *so)
2226 {
2227
2228 return (sototcpcb(so));
2229 }
2230
2231 #ifdef DDB
2232 static void
2233 db_print_indent(int indent)
2234 {
2235 int i;
2236
2237 for (i = 0; i < indent; i++)
2238 db_printf(" ");
2239 }
2240
2241 static void
2242 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2243 {
2244 char faddr_str[48], laddr_str[48];
2245
2246 db_print_indent(indent);
2247 db_printf("%s at %p\n", name, inc);
2248
2249 indent += 2;
2250
2251 #ifdef INET6
2252 if (inc->inc_flags & INC_ISIPV6) {
2253 /* IPv6. */
2254 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2255 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2256 } else
2257 #endif
2258 {
2259 /* IPv4. */
2260 inet_ntoa_r(inc->inc_laddr, laddr_str);
2261 inet_ntoa_r(inc->inc_faddr, faddr_str);
2262 }
2263 db_print_indent(indent);
2264 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2265 ntohs(inc->inc_lport));
2266 db_print_indent(indent);
2267 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2268 ntohs(inc->inc_fport));
2269 }
2270
2271 static void
2272 db_print_inpflags(int inp_flags)
2273 {
2274 int comma;
2275
2276 comma = 0;
2277 if (inp_flags & INP_RECVOPTS) {
2278 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2279 comma = 1;
2280 }
2281 if (inp_flags & INP_RECVRETOPTS) {
2282 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2283 comma = 1;
2284 }
2285 if (inp_flags & INP_RECVDSTADDR) {
2286 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2287 comma = 1;
2288 }
2289 if (inp_flags & INP_HDRINCL) {
2290 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2291 comma = 1;
2292 }
2293 if (inp_flags & INP_HIGHPORT) {
2294 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2295 comma = 1;
2296 }
2297 if (inp_flags & INP_LOWPORT) {
2298 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2299 comma = 1;
2300 }
2301 if (inp_flags & INP_ANONPORT) {
2302 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2303 comma = 1;
2304 }
2305 if (inp_flags & INP_RECVIF) {
2306 db_printf("%sINP_RECVIF", comma ? ", " : "");
2307 comma = 1;
2308 }
2309 if (inp_flags & INP_MTUDISC) {
2310 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2311 comma = 1;
2312 }
2313 if (inp_flags & INP_FAITH) {
2314 db_printf("%sINP_FAITH", comma ? ", " : "");
2315 comma = 1;
2316 }
2317 if (inp_flags & INP_RECVTTL) {
2318 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2319 comma = 1;
2320 }
2321 if (inp_flags & INP_DONTFRAG) {
2322 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2323 comma = 1;
2324 }
2325 if (inp_flags & INP_RECVTOS) {
2326 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2327 comma = 1;
2328 }
2329 if (inp_flags & IN6P_IPV6_V6ONLY) {
2330 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2331 comma = 1;
2332 }
2333 if (inp_flags & IN6P_PKTINFO) {
2334 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2335 comma = 1;
2336 }
2337 if (inp_flags & IN6P_HOPLIMIT) {
2338 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2339 comma = 1;
2340 }
2341 if (inp_flags & IN6P_HOPOPTS) {
2342 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2343 comma = 1;
2344 }
2345 if (inp_flags & IN6P_DSTOPTS) {
2346 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2347 comma = 1;
2348 }
2349 if (inp_flags & IN6P_RTHDR) {
2350 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2351 comma = 1;
2352 }
2353 if (inp_flags & IN6P_RTHDRDSTOPTS) {
2354 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2355 comma = 1;
2356 }
2357 if (inp_flags & IN6P_TCLASS) {
2358 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2359 comma = 1;
2360 }
2361 if (inp_flags & IN6P_AUTOFLOWLABEL) {
2362 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2363 comma = 1;
2364 }
2365 if (inp_flags & INP_TIMEWAIT) {
2366 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2367 comma = 1;
2368 }
2369 if (inp_flags & INP_ONESBCAST) {
2370 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2371 comma = 1;
2372 }
2373 if (inp_flags & INP_DROPPED) {
2374 db_printf("%sINP_DROPPED", comma ? ", " : "");
2375 comma = 1;
2376 }
2377 if (inp_flags & INP_SOCKREF) {
2378 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2379 comma = 1;
2380 }
2381 if (inp_flags & IN6P_RFC2292) {
2382 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2383 comma = 1;
2384 }
2385 if (inp_flags & IN6P_MTU) {
2386 db_printf("IN6P_MTU%s", comma ? ", " : "");
2387 comma = 1;
2388 }
2389 }
2390
2391 static void
2392 db_print_inpvflag(u_char inp_vflag)
2393 {
2394 int comma;
2395
2396 comma = 0;
2397 if (inp_vflag & INP_IPV4) {
2398 db_printf("%sINP_IPV4", comma ? ", " : "");
2399 comma = 1;
2400 }
2401 if (inp_vflag & INP_IPV6) {
2402 db_printf("%sINP_IPV6", comma ? ", " : "");
2403 comma = 1;
2404 }
2405 if (inp_vflag & INP_IPV6PROTO) {
2406 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2407 comma = 1;
2408 }
2409 }
2410
2411 static void
2412 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2413 {
2414
2415 db_print_indent(indent);
2416 db_printf("%s at %p\n", name, inp);
2417
2418 indent += 2;
2419
2420 db_print_indent(indent);
2421 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2422
2423 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2424
2425 db_print_indent(indent);
2426 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
2427 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2428
2429 db_print_indent(indent);
2430 db_printf("inp_label: %p inp_flags: 0x%x (",
2431 inp->inp_label, inp->inp_flags);
2432 db_print_inpflags(inp->inp_flags);
2433 db_printf(")\n");
2434
2435 db_print_indent(indent);
2436 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
2437 inp->inp_vflag);
2438 db_print_inpvflag(inp->inp_vflag);
2439 db_printf(")\n");
2440
2441 db_print_indent(indent);
2442 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
2443 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2444
2445 db_print_indent(indent);
2446 #ifdef INET6
2447 if (inp->inp_vflag & INP_IPV6) {
2448 db_printf("in6p_options: %p in6p_outputopts: %p "
2449 "in6p_moptions: %p\n", inp->in6p_options,
2450 inp->in6p_outputopts, inp->in6p_moptions);
2451 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
2452 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2453 inp->in6p_hops);
2454 } else
2455 #endif
2456 {
2457 db_printf("inp_ip_tos: %d inp_ip_options: %p "
2458 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2459 inp->inp_options, inp->inp_moptions);
2460 }
2461
2462 db_print_indent(indent);
2463 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
2464 (uintmax_t)inp->inp_gencnt);
2465 }
2466
2467 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2468 {
2469 struct inpcb *inp;
2470
2471 if (!have_addr) {
2472 db_printf("usage: show inpcb <addr>\n");
2473 return;
2474 }
2475 inp = (struct inpcb *)addr;
2476
2477 db_print_inpcb(inp, "inpcb", 0);
2478 }
2479 #endif /* DDB */
Cache object: 923a37edbdc2d8530cfdaa0c31eaef71
|