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("$FreeBSD$");
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 #include "opt_rss.h"
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/callout.h>
54 #include <sys/eventhandler.h>
55 #include <sys/domain.h>
56 #include <sys/protosw.h>
57 #include <sys/rmlock.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/refcount.h>
63 #include <sys/jail.h>
64 #include <sys/kernel.h>
65 #include <sys/sysctl.h>
66
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #endif
70
71 #include <vm/uma.h>
72
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/if_types.h>
76 #include <net/if_llatbl.h>
77 #include <net/route.h>
78 #include <net/rss_config.h>
79 #include <net/vnet.h>
80
81 #if defined(INET) || defined(INET6)
82 #include <netinet/in.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/ip_var.h>
85 #include <netinet/tcp_var.h>
86 #include <netinet/udp.h>
87 #include <netinet/udp_var.h>
88 #endif
89 #ifdef INET
90 #include <netinet/in_var.h>
91 #endif
92 #ifdef INET6
93 #include <netinet/ip6.h>
94 #include <netinet6/in6_pcb.h>
95 #include <netinet6/in6_var.h>
96 #include <netinet6/ip6_var.h>
97 #endif /* INET6 */
98
99 #include <netipsec/ipsec_support.h>
100
101 #include <security/mac/mac_framework.h>
102
103 static struct callout ipport_tick_callout;
104
105 /*
106 * These configure the range of local port addresses assigned to
107 * "unspecified" outgoing connections/packets/whatever.
108 */
109 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
110 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
111 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
112 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
113 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
114 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
115
116 /*
117 * Reserved ports accessible only to root. There are significant
118 * security considerations that must be accounted for when changing these,
119 * but the security benefits can be great. Please be careful.
120 */
121 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
122 VNET_DEFINE(int, ipport_reservedlow);
123
124 /* Variables dealing with random ephemeral port allocation. */
125 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
126 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
127 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
128 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
129 VNET_DEFINE(int, ipport_tcpallocs);
130 static VNET_DEFINE(int, ipport_tcplastcount);
131
132 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
133
134 static void in_pcbremlists(struct inpcb *inp);
135 #ifdef INET
136 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
137 struct in_addr faddr, u_int fport_arg,
138 struct in_addr laddr, u_int lport_arg,
139 int lookupflags, struct ifnet *ifp);
140
141 #define RANGECHK(var, min, max) \
142 if ((var) < (min)) { (var) = (min); } \
143 else if ((var) > (max)) { (var) = (max); }
144
145 static int
sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)146 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
147 {
148 int error;
149
150 error = sysctl_handle_int(oidp, arg1, arg2, req);
151 if (error == 0) {
152 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
153 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
154 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
155 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
156 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
157 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
158 }
159 return (error);
160 }
161
162 #undef RANGECHK
163
164 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
165 "IP Ports");
166
167 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
168 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
169 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
170 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
171 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
172 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
175 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
178 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
180 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
181 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
183 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
184 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
185 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
186 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
187 &VNET_NAME(ipport_reservedhigh), 0, "");
188 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
189 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
190 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
191 CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
193 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
194 CTLFLAG_VNET | CTLFLAG_RW,
195 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
196 "allocations before switching to a sequental one");
197 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
198 CTLFLAG_VNET | CTLFLAG_RW,
199 &VNET_NAME(ipport_randomtime), 0,
200 "Minimum time to keep sequental port "
201 "allocation before switching to a random one");
202 #endif /* INET */
203
204 /*
205 * in_pcb.c: manage the Protocol Control Blocks.
206 *
207 * NOTE: It is assumed that most of these functions will be called with
208 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
209 * functions often modify hash chains or addresses in pcbs.
210 */
211
212 /*
213 * Initialize an inpcbinfo -- we should be able to reduce the number of
214 * arguments in time.
215 */
216 void
in_pcbinfo_init(struct inpcbinfo * pcbinfo,const char * name,struct inpcbhead * listhead,int hash_nelements,int porthash_nelements,char * inpcbzone_name,uma_init inpcbzone_init,uma_fini inpcbzone_fini,uint32_t inpcbzone_flags,u_int hashfields)217 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
218 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
219 char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
220 uint32_t inpcbzone_flags, u_int hashfields)
221 {
222
223 INP_INFO_LOCK_INIT(pcbinfo, name);
224 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
225 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
226 #ifdef VIMAGE
227 pcbinfo->ipi_vnet = curvnet;
228 #endif
229 pcbinfo->ipi_listhead = listhead;
230 LIST_INIT(pcbinfo->ipi_listhead);
231 pcbinfo->ipi_count = 0;
232 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
233 &pcbinfo->ipi_hashmask);
234 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
235 &pcbinfo->ipi_porthashmask);
236 #ifdef PCBGROUP
237 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
238 #endif
239 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
240 NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
241 inpcbzone_flags);
242 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
243 uma_zone_set_warning(pcbinfo->ipi_zone,
244 "kern.ipc.maxsockets limit reached");
245 }
246
247 /*
248 * Destroy an inpcbinfo.
249 */
250 void
in_pcbinfo_destroy(struct inpcbinfo * pcbinfo)251 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
252 {
253
254 KASSERT(pcbinfo->ipi_count == 0,
255 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
256
257 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
258 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
259 pcbinfo->ipi_porthashmask);
260 #ifdef PCBGROUP
261 in_pcbgroup_destroy(pcbinfo);
262 #endif
263 uma_zdestroy(pcbinfo->ipi_zone);
264 INP_LIST_LOCK_DESTROY(pcbinfo);
265 INP_HASH_LOCK_DESTROY(pcbinfo);
266 INP_INFO_LOCK_DESTROY(pcbinfo);
267 }
268
269 /*
270 * Allocate a PCB and associate it with the socket.
271 * On success return with the PCB locked.
272 */
273 int
in_pcballoc(struct socket * so,struct inpcbinfo * pcbinfo)274 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
275 {
276 struct inpcb *inp;
277 int error;
278
279 #ifdef INVARIANTS
280 if (pcbinfo == &V_tcbinfo) {
281 INP_INFO_RLOCK_ASSERT(pcbinfo);
282 } else {
283 INP_INFO_WLOCK_ASSERT(pcbinfo);
284 }
285 #endif
286
287 error = 0;
288 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
289 if (inp == NULL)
290 return (ENOBUFS);
291 bzero(inp, inp_zero_size);
292 inp->inp_pcbinfo = pcbinfo;
293 inp->inp_socket = so;
294 inp->inp_cred = crhold(so->so_cred);
295 inp->inp_inc.inc_fibnum = so->so_fibnum;
296 #ifdef MAC
297 error = mac_inpcb_init(inp, M_NOWAIT);
298 if (error != 0)
299 goto out;
300 mac_inpcb_create(so, inp);
301 #endif
302 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
303 error = ipsec_init_pcbpolicy(inp);
304 if (error != 0) {
305 #ifdef MAC
306 mac_inpcb_destroy(inp);
307 #endif
308 goto out;
309 }
310 #endif /*IPSEC*/
311 #ifdef INET6
312 if (INP_SOCKAF(so) == AF_INET6) {
313 inp->inp_vflag |= INP_IPV6PROTO;
314 if (V_ip6_v6only)
315 inp->inp_flags |= IN6P_IPV6_V6ONLY;
316 }
317 #endif
318 INP_WLOCK(inp);
319 INP_LIST_WLOCK(pcbinfo);
320 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
321 pcbinfo->ipi_count++;
322 so->so_pcb = (caddr_t)inp;
323 #ifdef INET6
324 if (V_ip6_auto_flowlabel)
325 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
326 #endif
327 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
328 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
329
330 /*
331 * Routes in inpcb's can cache L2 as well; they are guaranteed
332 * to be cleaned up.
333 */
334 inp->inp_route.ro_flags = RT_LLE_CACHE;
335 INP_LIST_WUNLOCK(pcbinfo);
336 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
337 out:
338 if (error != 0) {
339 crfree(inp->inp_cred);
340 uma_zfree(pcbinfo->ipi_zone, inp);
341 }
342 #endif
343 return (error);
344 }
345
346 #ifdef INET
347 int
in_pcbbind(struct inpcb * inp,struct sockaddr * nam,struct ucred * cred)348 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
349 {
350 int anonport, error;
351
352 INP_WLOCK_ASSERT(inp);
353 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
354
355 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
356 return (EINVAL);
357 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
358 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
359 &inp->inp_lport, cred);
360 if (error)
361 return (error);
362 if (in_pcbinshash(inp) != 0) {
363 inp->inp_laddr.s_addr = INADDR_ANY;
364 inp->inp_lport = 0;
365 return (EAGAIN);
366 }
367 if (anonport)
368 inp->inp_flags |= INP_ANONPORT;
369 return (0);
370 }
371 #endif
372
373 /*
374 * Select a local port (number) to use.
375 */
376 #if defined(INET) || defined(INET6)
377 int
in_pcb_lport(struct inpcb * inp,struct in_addr * laddrp,u_short * lportp,struct ucred * cred,int lookupflags)378 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
379 struct ucred *cred, int lookupflags)
380 {
381 struct inpcbinfo *pcbinfo;
382 struct inpcb *tmpinp;
383 unsigned short *lastport;
384 int count, dorandom, error;
385 u_short aux, first, last, lport;
386 #ifdef INET
387 struct in_addr laddr;
388 #endif
389
390 pcbinfo = inp->inp_pcbinfo;
391
392 /*
393 * Because no actual state changes occur here, a global write lock on
394 * the pcbinfo isn't required.
395 */
396 INP_LOCK_ASSERT(inp);
397 INP_HASH_LOCK_ASSERT(pcbinfo);
398
399 if (inp->inp_flags & INP_HIGHPORT) {
400 first = V_ipport_hifirstauto; /* sysctl */
401 last = V_ipport_hilastauto;
402 lastport = &pcbinfo->ipi_lasthi;
403 } else if (inp->inp_flags & INP_LOWPORT) {
404 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
405 if (error)
406 return (error);
407 first = V_ipport_lowfirstauto; /* 1023 */
408 last = V_ipport_lowlastauto; /* 600 */
409 lastport = &pcbinfo->ipi_lastlow;
410 } else {
411 first = V_ipport_firstauto; /* sysctl */
412 last = V_ipport_lastauto;
413 lastport = &pcbinfo->ipi_lastport;
414 }
415 /*
416 * For UDP(-Lite), use random port allocation as long as the user
417 * allows it. For TCP (and as of yet unknown) connections,
418 * use random port allocation only if the user allows it AND
419 * ipport_tick() allows it.
420 */
421 if (V_ipport_randomized &&
422 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
423 pcbinfo == &V_ulitecbinfo))
424 dorandom = 1;
425 else
426 dorandom = 0;
427 /*
428 * It makes no sense to do random port allocation if
429 * we have the only port available.
430 */
431 if (first == last)
432 dorandom = 0;
433 /* Make sure to not include UDP(-Lite) packets in the count. */
434 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
435 V_ipport_tcpallocs++;
436 /*
437 * Instead of having two loops further down counting up or down
438 * make sure that first is always <= last and go with only one
439 * code path implementing all logic.
440 */
441 if (first > last) {
442 aux = first;
443 first = last;
444 last = aux;
445 }
446
447 #ifdef INET
448 /* Make the compiler happy. */
449 laddr.s_addr = 0;
450 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
451 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
452 __func__, inp));
453 laddr = *laddrp;
454 }
455 #endif
456 tmpinp = NULL; /* Make compiler happy. */
457 lport = *lportp;
458
459 if (dorandom)
460 *lastport = first + (arc4random() % (last - first));
461
462 count = last - first;
463
464 do {
465 if (count-- < 0) /* completely used? */
466 return (EADDRNOTAVAIL);
467 ++*lastport;
468 if (*lastport < first || *lastport > last)
469 *lastport = first;
470 lport = htons(*lastport);
471
472 #ifdef INET6
473 if ((inp->inp_vflag & INP_IPV6) != 0)
474 tmpinp = in6_pcblookup_local(pcbinfo,
475 &inp->in6p_laddr, lport, lookupflags, cred);
476 #endif
477 #if defined(INET) && defined(INET6)
478 else
479 #endif
480 #ifdef INET
481 tmpinp = in_pcblookup_local(pcbinfo, laddr,
482 lport, lookupflags, cred);
483 #endif
484 } while (tmpinp != NULL);
485
486 #ifdef INET
487 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
488 laddrp->s_addr = laddr.s_addr;
489 #endif
490 *lportp = lport;
491
492 return (0);
493 }
494
495 /*
496 * Return cached socket options.
497 */
498 short
inp_so_options(const struct inpcb * inp)499 inp_so_options(const struct inpcb *inp)
500 {
501 short so_options;
502
503 so_options = 0;
504
505 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
506 so_options |= SO_REUSEPORT;
507 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
508 so_options |= SO_REUSEADDR;
509 return (so_options);
510 }
511 #endif /* INET || INET6 */
512
513 /*
514 * Check if a new BINDMULTI socket is allowed to be created.
515 *
516 * ni points to the new inp.
517 * oi points to the exisitng inp.
518 *
519 * This checks whether the existing inp also has BINDMULTI and
520 * whether the credentials match.
521 */
522 int
in_pcbbind_check_bindmulti(const struct inpcb * ni,const struct inpcb * oi)523 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
524 {
525 /* Check permissions match */
526 if ((ni->inp_flags2 & INP_BINDMULTI) &&
527 (ni->inp_cred->cr_uid !=
528 oi->inp_cred->cr_uid))
529 return (0);
530
531 /* Check the existing inp has BINDMULTI set */
532 if ((ni->inp_flags2 & INP_BINDMULTI) &&
533 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
534 return (0);
535
536 /*
537 * We're okay - either INP_BINDMULTI isn't set on ni, or
538 * it is and it matches the checks.
539 */
540 return (1);
541 }
542
543 #ifdef INET
544 /*
545 * Set up a bind operation on a PCB, performing port allocation
546 * as required, but do not actually modify the PCB. Callers can
547 * either complete the bind by setting inp_laddr/inp_lport and
548 * calling in_pcbinshash(), or they can just use the resulting
549 * port and address to authorise the sending of a once-off packet.
550 *
551 * On error, the values of *laddrp and *lportp are not changed.
552 */
553 int
in_pcbbind_setup(struct inpcb * inp,struct sockaddr * nam,in_addr_t * laddrp,u_short * lportp,struct ucred * cred)554 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
555 u_short *lportp, struct ucred *cred)
556 {
557 struct socket *so = inp->inp_socket;
558 struct sockaddr_in *sin;
559 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
560 struct in_addr laddr;
561 u_short lport = 0;
562 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
563 int error;
564
565 /*
566 * No state changes, so read locks are sufficient here.
567 */
568 INP_LOCK_ASSERT(inp);
569 INP_HASH_LOCK_ASSERT(pcbinfo);
570
571 if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
572 return (EADDRNOTAVAIL);
573 laddr.s_addr = *laddrp;
574 if (nam != NULL && laddr.s_addr != INADDR_ANY)
575 return (EINVAL);
576 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
577 lookupflags = INPLOOKUP_WILDCARD;
578 if (nam == NULL) {
579 if ((error = prison_local_ip4(cred, &laddr)) != 0)
580 return (error);
581 } else {
582 sin = (struct sockaddr_in *)nam;
583 if (nam->sa_len != sizeof (*sin))
584 return (EINVAL);
585 #ifdef notdef
586 /*
587 * We should check the family, but old programs
588 * incorrectly fail to initialize it.
589 */
590 if (sin->sin_family != AF_INET)
591 return (EAFNOSUPPORT);
592 #endif
593 error = prison_local_ip4(cred, &sin->sin_addr);
594 if (error)
595 return (error);
596 if (sin->sin_port != *lportp) {
597 /* Don't allow the port to change. */
598 if (*lportp != 0)
599 return (EINVAL);
600 lport = sin->sin_port;
601 }
602 /* NB: lport is left as 0 if the port isn't being changed. */
603 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
604 /*
605 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
606 * allow complete duplication of binding if
607 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
608 * and a multicast address is bound on both
609 * new and duplicated sockets.
610 */
611 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
612 reuseport = SO_REUSEADDR|SO_REUSEPORT;
613 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
614 sin->sin_port = 0; /* yech... */
615 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
616 /*
617 * Is the address a local IP address?
618 * If INP_BINDANY is set, then the socket may be bound
619 * to any endpoint address, local or not.
620 */
621 if ((inp->inp_flags & INP_BINDANY) == 0 &&
622 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
623 return (EADDRNOTAVAIL);
624 }
625 laddr = sin->sin_addr;
626 if (lport) {
627 struct inpcb *t;
628 struct tcptw *tw;
629
630 /* GROSS */
631 if (ntohs(lport) <= V_ipport_reservedhigh &&
632 ntohs(lport) >= V_ipport_reservedlow &&
633 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
634 0))
635 return (EACCES);
636 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
637 priv_check_cred(inp->inp_cred,
638 PRIV_NETINET_REUSEPORT, 0) != 0) {
639 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
640 lport, INPLOOKUP_WILDCARD, cred);
641 /*
642 * XXX
643 * This entire block sorely needs a rewrite.
644 */
645 if (t &&
646 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
647 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
648 (so->so_type != SOCK_STREAM ||
649 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
650 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
651 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
652 (t->inp_flags2 & INP_REUSEPORT) == 0) &&
653 (inp->inp_cred->cr_uid !=
654 t->inp_cred->cr_uid))
655 return (EADDRINUSE);
656
657 /*
658 * If the socket is a BINDMULTI socket, then
659 * the credentials need to match and the
660 * original socket also has to have been bound
661 * with BINDMULTI.
662 */
663 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
664 return (EADDRINUSE);
665 }
666 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
667 lport, lookupflags, cred);
668 if (t && (t->inp_flags & INP_TIMEWAIT)) {
669 /*
670 * XXXRW: If an incpb has had its timewait
671 * state recycled, we treat the address as
672 * being in use (for now). This is better
673 * than a panic, but not desirable.
674 */
675 tw = intotw(t);
676 if (tw == NULL ||
677 (reuseport & tw->tw_so_options) == 0)
678 return (EADDRINUSE);
679 } else if (t &&
680 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
681 (reuseport & inp_so_options(t)) == 0) {
682 #ifdef INET6
683 if (ntohl(sin->sin_addr.s_addr) !=
684 INADDR_ANY ||
685 ntohl(t->inp_laddr.s_addr) !=
686 INADDR_ANY ||
687 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
688 (t->inp_vflag & INP_IPV6PROTO) == 0)
689 #endif
690 return (EADDRINUSE);
691 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
692 return (EADDRINUSE);
693 }
694 }
695 }
696 if (*lportp != 0)
697 lport = *lportp;
698 if (lport == 0) {
699 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
700 if (error != 0)
701 return (error);
702
703 }
704 *laddrp = laddr.s_addr;
705 *lportp = lport;
706 return (0);
707 }
708
709 /*
710 * Connect from a socket to a specified address.
711 * Both address and port must be specified in argument sin.
712 * If don't have a local address for this socket yet,
713 * then pick one.
714 */
715 int
in_pcbconnect_mbuf(struct inpcb * inp,struct sockaddr * nam,struct ucred * cred,struct mbuf * m)716 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
717 struct ucred *cred, struct mbuf *m)
718 {
719 u_short lport, fport;
720 in_addr_t laddr, faddr;
721 int anonport, error;
722
723 INP_WLOCK_ASSERT(inp);
724 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
725
726 lport = inp->inp_lport;
727 laddr = inp->inp_laddr.s_addr;
728 anonport = (lport == 0);
729 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
730 NULL, cred);
731 if (error)
732 return (error);
733
734 /* Do the initial binding of the local address if required. */
735 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
736 inp->inp_lport = lport;
737 inp->inp_laddr.s_addr = laddr;
738 if (in_pcbinshash(inp) != 0) {
739 inp->inp_laddr.s_addr = INADDR_ANY;
740 inp->inp_lport = 0;
741 return (EAGAIN);
742 }
743 }
744
745 /* Commit the remaining changes. */
746 inp->inp_lport = lport;
747 inp->inp_laddr.s_addr = laddr;
748 inp->inp_faddr.s_addr = faddr;
749 inp->inp_fport = fport;
750 in_pcbrehash_mbuf(inp, m);
751
752 if (anonport)
753 inp->inp_flags |= INP_ANONPORT;
754 return (0);
755 }
756
757 int
in_pcbconnect(struct inpcb * inp,struct sockaddr * nam,struct ucred * cred)758 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
759 {
760
761 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
762 }
763
764 /*
765 * Do proper source address selection on an unbound socket in case
766 * of connect. Take jails into account as well.
767 */
768 int
in_pcbladdr(struct inpcb * inp,struct in_addr * faddr,struct in_addr * laddr,struct ucred * cred)769 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
770 struct ucred *cred)
771 {
772 struct ifaddr *ifa;
773 struct sockaddr *sa;
774 struct sockaddr_in *sin;
775 struct route sro;
776 int error;
777
778 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
779
780 /*
781 * Bypass source address selection and use the primary jail IP
782 * if requested.
783 */
784 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
785 return (0);
786
787 error = 0;
788 bzero(&sro, sizeof(sro));
789
790 sin = (struct sockaddr_in *)&sro.ro_dst;
791 sin->sin_family = AF_INET;
792 sin->sin_len = sizeof(struct sockaddr_in);
793 sin->sin_addr.s_addr = faddr->s_addr;
794
795 /*
796 * If route is known our src addr is taken from the i/f,
797 * else punt.
798 *
799 * Find out route to destination.
800 */
801 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
802 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
803
804 /*
805 * If we found a route, use the address corresponding to
806 * the outgoing interface.
807 *
808 * Otherwise assume faddr is reachable on a directly connected
809 * network and try to find a corresponding interface to take
810 * the source address from.
811 */
812 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
813 struct in_ifaddr *ia;
814 struct ifnet *ifp;
815
816 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
817 inp->inp_socket->so_fibnum));
818 if (ia == NULL)
819 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
820 inp->inp_socket->so_fibnum));
821 if (ia == NULL) {
822 error = ENETUNREACH;
823 goto done;
824 }
825
826 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
827 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
828 ifa_free(&ia->ia_ifa);
829 goto done;
830 }
831
832 ifp = ia->ia_ifp;
833 ifa_free(&ia->ia_ifa);
834 ia = NULL;
835 IF_ADDR_RLOCK(ifp);
836 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
837
838 sa = ifa->ifa_addr;
839 if (sa->sa_family != AF_INET)
840 continue;
841 sin = (struct sockaddr_in *)sa;
842 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
843 ia = (struct in_ifaddr *)ifa;
844 break;
845 }
846 }
847 if (ia != NULL) {
848 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
849 IF_ADDR_RUNLOCK(ifp);
850 goto done;
851 }
852 IF_ADDR_RUNLOCK(ifp);
853
854 /* 3. As a last resort return the 'default' jail address. */
855 error = prison_get_ip4(cred, laddr);
856 goto done;
857 }
858
859 /*
860 * If the outgoing interface on the route found is not
861 * a loopback interface, use the address from that interface.
862 * In case of jails do those three steps:
863 * 1. check if the interface address belongs to the jail. If so use it.
864 * 2. check if we have any address on the outgoing interface
865 * belonging to this jail. If so use it.
866 * 3. as a last resort return the 'default' jail address.
867 */
868 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
869 struct in_ifaddr *ia;
870 struct ifnet *ifp;
871
872 /* If not jailed, use the default returned. */
873 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
874 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
875 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
876 goto done;
877 }
878
879 /* Jailed. */
880 /* 1. Check if the iface address belongs to the jail. */
881 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
882 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
883 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
884 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
885 goto done;
886 }
887
888 /*
889 * 2. Check if we have any address on the outgoing interface
890 * belonging to this jail.
891 */
892 ia = NULL;
893 ifp = sro.ro_rt->rt_ifp;
894 IF_ADDR_RLOCK(ifp);
895 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
896 sa = ifa->ifa_addr;
897 if (sa->sa_family != AF_INET)
898 continue;
899 sin = (struct sockaddr_in *)sa;
900 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
901 ia = (struct in_ifaddr *)ifa;
902 break;
903 }
904 }
905 if (ia != NULL) {
906 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
907 IF_ADDR_RUNLOCK(ifp);
908 goto done;
909 }
910 IF_ADDR_RUNLOCK(ifp);
911
912 /* 3. As a last resort return the 'default' jail address. */
913 error = prison_get_ip4(cred, laddr);
914 goto done;
915 }
916
917 /*
918 * The outgoing interface is marked with 'loopback net', so a route
919 * to ourselves is here.
920 * Try to find the interface of the destination address and then
921 * take the address from there. That interface is not necessarily
922 * a loopback interface.
923 * In case of jails, check that it is an address of the jail
924 * and if we cannot find, fall back to the 'default' jail address.
925 */
926 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
927 struct sockaddr_in sain;
928 struct in_ifaddr *ia;
929
930 bzero(&sain, sizeof(struct sockaddr_in));
931 sain.sin_family = AF_INET;
932 sain.sin_len = sizeof(struct sockaddr_in);
933 sain.sin_addr.s_addr = faddr->s_addr;
934
935 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
936 inp->inp_socket->so_fibnum));
937 if (ia == NULL)
938 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
939 inp->inp_socket->so_fibnum));
940 if (ia == NULL)
941 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
942
943 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
944 if (ia == NULL) {
945 error = ENETUNREACH;
946 goto done;
947 }
948 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
949 ifa_free(&ia->ia_ifa);
950 goto done;
951 }
952
953 /* Jailed. */
954 if (ia != NULL) {
955 struct ifnet *ifp;
956
957 ifp = ia->ia_ifp;
958 ifa_free(&ia->ia_ifa);
959 ia = NULL;
960 IF_ADDR_RLOCK(ifp);
961 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
962
963 sa = ifa->ifa_addr;
964 if (sa->sa_family != AF_INET)
965 continue;
966 sin = (struct sockaddr_in *)sa;
967 if (prison_check_ip4(cred,
968 &sin->sin_addr) == 0) {
969 ia = (struct in_ifaddr *)ifa;
970 break;
971 }
972 }
973 if (ia != NULL) {
974 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
975 IF_ADDR_RUNLOCK(ifp);
976 goto done;
977 }
978 IF_ADDR_RUNLOCK(ifp);
979 }
980
981 /* 3. As a last resort return the 'default' jail address. */
982 error = prison_get_ip4(cred, laddr);
983 goto done;
984 }
985
986 done:
987 if (sro.ro_rt != NULL)
988 RTFREE(sro.ro_rt);
989 return (error);
990 }
991
992 /*
993 * Set up for a connect from a socket to the specified address.
994 * On entry, *laddrp and *lportp should contain the current local
995 * address and port for the PCB; these are updated to the values
996 * that should be placed in inp_laddr and inp_lport to complete
997 * the connect.
998 *
999 * On success, *faddrp and *fportp will be set to the remote address
1000 * and port. These are not updated in the error case.
1001 *
1002 * If the operation fails because the connection already exists,
1003 * *oinpp will be set to the PCB of that connection so that the
1004 * caller can decide to override it. In all other cases, *oinpp
1005 * is set to NULL.
1006 */
1007 int
in_pcbconnect_setup(struct inpcb * inp,struct sockaddr * nam,in_addr_t * laddrp,u_short * lportp,in_addr_t * faddrp,u_short * fportp,struct inpcb ** oinpp,struct ucred * cred)1008 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1009 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1010 struct inpcb **oinpp, struct ucred *cred)
1011 {
1012 struct rm_priotracker in_ifa_tracker;
1013 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1014 struct in_ifaddr *ia;
1015 struct inpcb *oinp;
1016 struct in_addr laddr, faddr;
1017 u_short lport, fport;
1018 int error;
1019
1020 /*
1021 * Because a global state change doesn't actually occur here, a read
1022 * lock is sufficient.
1023 */
1024 INP_LOCK_ASSERT(inp);
1025 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1026
1027 if (oinpp != NULL)
1028 *oinpp = NULL;
1029 if (nam->sa_len != sizeof (*sin))
1030 return (EINVAL);
1031 if (sin->sin_family != AF_INET)
1032 return (EAFNOSUPPORT);
1033 if (sin->sin_port == 0)
1034 return (EADDRNOTAVAIL);
1035 laddr.s_addr = *laddrp;
1036 lport = *lportp;
1037 faddr = sin->sin_addr;
1038 fport = sin->sin_port;
1039
1040 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1041 /*
1042 * If the destination address is INADDR_ANY,
1043 * use the primary local address.
1044 * If the supplied address is INADDR_BROADCAST,
1045 * and the primary interface supports broadcast,
1046 * choose the broadcast address for that interface.
1047 */
1048 if (faddr.s_addr == INADDR_ANY) {
1049 IN_IFADDR_RLOCK(&in_ifa_tracker);
1050 faddr =
1051 IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1052 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1053 if (cred != NULL &&
1054 (error = prison_get_ip4(cred, &faddr)) != 0)
1055 return (error);
1056 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1057 IN_IFADDR_RLOCK(&in_ifa_tracker);
1058 if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1059 IFF_BROADCAST)
1060 faddr = satosin(&TAILQ_FIRST(
1061 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1062 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1063 }
1064 }
1065 if (laddr.s_addr == INADDR_ANY) {
1066 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1067 /*
1068 * If the destination address is multicast and an outgoing
1069 * interface has been set as a multicast option, prefer the
1070 * address of that interface as our source address.
1071 */
1072 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1073 inp->inp_moptions != NULL) {
1074 struct ip_moptions *imo;
1075 struct ifnet *ifp;
1076
1077 imo = inp->inp_moptions;
1078 if (imo->imo_multicast_ifp != NULL) {
1079 ifp = imo->imo_multicast_ifp;
1080 IN_IFADDR_RLOCK(&in_ifa_tracker);
1081 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1082 if ((ia->ia_ifp == ifp) &&
1083 (cred == NULL ||
1084 prison_check_ip4(cred,
1085 &ia->ia_addr.sin_addr) == 0))
1086 break;
1087 }
1088 if (ia == NULL)
1089 error = EADDRNOTAVAIL;
1090 else {
1091 laddr = ia->ia_addr.sin_addr;
1092 error = 0;
1093 }
1094 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1095 }
1096 }
1097 if (error)
1098 return (error);
1099 }
1100 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1101 laddr, lport, 0, NULL);
1102 if (oinp != NULL) {
1103 if (oinpp != NULL)
1104 *oinpp = oinp;
1105 return (EADDRINUSE);
1106 }
1107 if (lport == 0) {
1108 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1109 cred);
1110 if (error)
1111 return (error);
1112 }
1113 *laddrp = laddr.s_addr;
1114 *lportp = lport;
1115 *faddrp = faddr.s_addr;
1116 *fportp = fport;
1117 return (0);
1118 }
1119
1120 void
in_pcbdisconnect(struct inpcb * inp)1121 in_pcbdisconnect(struct inpcb *inp)
1122 {
1123
1124 INP_WLOCK_ASSERT(inp);
1125 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1126
1127 inp->inp_faddr.s_addr = INADDR_ANY;
1128 inp->inp_fport = 0;
1129 in_pcbrehash(inp);
1130 }
1131 #endif /* INET */
1132
1133 /*
1134 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1135 * For most protocols, this will be invoked immediately prior to calling
1136 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1137 * socket, in which case in_pcbfree() is deferred.
1138 */
1139 void
in_pcbdetach(struct inpcb * inp)1140 in_pcbdetach(struct inpcb *inp)
1141 {
1142
1143 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1144
1145 inp->inp_socket->so_pcb = NULL;
1146 inp->inp_socket = NULL;
1147 }
1148
1149 /*
1150 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1151 * stability of an inpcb pointer despite the inpcb lock being released. This
1152 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1153 * but where the inpcb lock may already held, or when acquiring a reference
1154 * via a pcbgroup.
1155 *
1156 * in_pcbref() should be used only to provide brief memory stability, and
1157 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1158 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1159 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1160 * lock and rele are the *only* safe operations that may be performed on the
1161 * inpcb.
1162 *
1163 * While the inpcb will not be freed, releasing the inpcb lock means that the
1164 * connection's state may change, so the caller should be careful to
1165 * revalidate any cached state on reacquiring the lock. Drop the reference
1166 * using in_pcbrele().
1167 */
1168 void
in_pcbref(struct inpcb * inp)1169 in_pcbref(struct inpcb *inp)
1170 {
1171
1172 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1173
1174 refcount_acquire(&inp->inp_refcount);
1175 }
1176
1177 /*
1178 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1179 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1180 * return a flag indicating whether or not the inpcb remains valid. If it is
1181 * valid, we return with the inpcb lock held.
1182 *
1183 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1184 * reference on an inpcb. Historically more work was done here (actually, in
1185 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1186 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1187 * about memory stability (and continued use of the write lock).
1188 */
1189 int
in_pcbrele_rlocked(struct inpcb * inp)1190 in_pcbrele_rlocked(struct inpcb *inp)
1191 {
1192 struct inpcbinfo *pcbinfo;
1193
1194 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1195
1196 INP_RLOCK_ASSERT(inp);
1197
1198 if (refcount_release(&inp->inp_refcount) == 0) {
1199 /*
1200 * If the inpcb has been freed, let the caller know, even if
1201 * this isn't the last reference.
1202 */
1203 if (inp->inp_flags2 & INP_FREED) {
1204 INP_RUNLOCK(inp);
1205 return (1);
1206 }
1207 return (0);
1208 }
1209
1210 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1211
1212 INP_RUNLOCK(inp);
1213 pcbinfo = inp->inp_pcbinfo;
1214 uma_zfree(pcbinfo->ipi_zone, inp);
1215 return (1);
1216 }
1217
1218 int
in_pcbrele_wlocked(struct inpcb * inp)1219 in_pcbrele_wlocked(struct inpcb *inp)
1220 {
1221 struct inpcbinfo *pcbinfo;
1222
1223 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1224
1225 INP_WLOCK_ASSERT(inp);
1226
1227 if (refcount_release(&inp->inp_refcount) == 0) {
1228 /*
1229 * If the inpcb has been freed, let the caller know, even if
1230 * this isn't the last reference.
1231 */
1232 if (inp->inp_flags2 & INP_FREED) {
1233 INP_WUNLOCK(inp);
1234 return (1);
1235 }
1236 return (0);
1237 }
1238
1239 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1240
1241 INP_WUNLOCK(inp);
1242 pcbinfo = inp->inp_pcbinfo;
1243 uma_zfree(pcbinfo->ipi_zone, inp);
1244 return (1);
1245 }
1246
1247 /*
1248 * Temporary wrapper.
1249 */
1250 int
in_pcbrele(struct inpcb * inp)1251 in_pcbrele(struct inpcb *inp)
1252 {
1253
1254 return (in_pcbrele_wlocked(inp));
1255 }
1256
1257 /*
1258 * Unconditionally schedule an inpcb to be freed by decrementing its
1259 * reference count, which should occur only after the inpcb has been detached
1260 * from its socket. If another thread holds a temporary reference (acquired
1261 * using in_pcbref()) then the free is deferred until that reference is
1262 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1263 * work, including removal from global lists, is done in this context, where
1264 * the pcbinfo lock is held.
1265 */
1266 void
in_pcbfree(struct inpcb * inp)1267 in_pcbfree(struct inpcb *inp)
1268 {
1269 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1270
1271 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1272
1273 #ifdef INVARIANTS
1274 if (pcbinfo == &V_tcbinfo) {
1275 INP_INFO_LOCK_ASSERT(pcbinfo);
1276 } else {
1277 INP_INFO_WLOCK_ASSERT(pcbinfo);
1278 }
1279 #endif
1280 INP_WLOCK_ASSERT(inp);
1281
1282 /* XXXRW: Do as much as possible here. */
1283 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1284 if (inp->inp_sp != NULL)
1285 ipsec_delete_pcbpolicy(inp);
1286 #endif
1287 INP_LIST_WLOCK(pcbinfo);
1288 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1289 in_pcbremlists(inp);
1290 INP_LIST_WUNLOCK(pcbinfo);
1291 #ifdef INET6
1292 if (inp->inp_vflag & INP_IPV6PROTO) {
1293 ip6_freepcbopts(inp->in6p_outputopts);
1294 if (inp->in6p_moptions != NULL)
1295 ip6_freemoptions(inp->in6p_moptions);
1296 }
1297 #endif
1298 if (inp->inp_options)
1299 (void)m_free(inp->inp_options);
1300 #ifdef INET
1301 if (inp->inp_moptions != NULL)
1302 inp_freemoptions(inp->inp_moptions);
1303 #endif
1304 if (inp->inp_route.ro_rt) {
1305 RTFREE(inp->inp_route.ro_rt);
1306 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1307 }
1308 if (inp->inp_route.ro_lle)
1309 LLE_FREE(inp->inp_route.ro_lle); /* zeros ro_lle */
1310
1311 inp->inp_vflag = 0;
1312 inp->inp_flags2 |= INP_FREED;
1313 crfree(inp->inp_cred);
1314 #ifdef MAC
1315 mac_inpcb_destroy(inp);
1316 #endif
1317 if (!in_pcbrele_wlocked(inp))
1318 INP_WUNLOCK(inp);
1319 }
1320
1321 /*
1322 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1323 * port reservation, and preventing it from being returned by inpcb lookups.
1324 *
1325 * It is used by TCP to mark an inpcb as unused and avoid future packet
1326 * delivery or event notification when a socket remains open but TCP has
1327 * closed. This might occur as a result of a shutdown()-initiated TCP close
1328 * or a RST on the wire, and allows the port binding to be reused while still
1329 * maintaining the invariant that so_pcb always points to a valid inpcb until
1330 * in_pcbdetach().
1331 *
1332 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1333 * in_pcbnotifyall() and in_pcbpurgeif0()?
1334 */
1335 void
in_pcbdrop(struct inpcb * inp)1336 in_pcbdrop(struct inpcb *inp)
1337 {
1338
1339 INP_WLOCK_ASSERT(inp);
1340
1341 /*
1342 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1343 * the hash lock...?
1344 */
1345 inp->inp_flags |= INP_DROPPED;
1346 if (inp->inp_flags & INP_INHASHLIST) {
1347 struct inpcbport *phd = inp->inp_phd;
1348
1349 INP_HASH_WLOCK(inp->inp_pcbinfo);
1350 LIST_REMOVE(inp, inp_hash);
1351 LIST_REMOVE(inp, inp_portlist);
1352 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1353 LIST_REMOVE(phd, phd_hash);
1354 free(phd, M_PCB);
1355 }
1356 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1357 inp->inp_flags &= ~INP_INHASHLIST;
1358 #ifdef PCBGROUP
1359 in_pcbgroup_remove(inp);
1360 #endif
1361 }
1362 }
1363
1364 #ifdef INET
1365 /*
1366 * Common routines to return the socket addresses associated with inpcbs.
1367 */
1368 struct sockaddr *
in_sockaddr(in_port_t port,struct in_addr * addr_p)1369 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1370 {
1371 struct sockaddr_in *sin;
1372
1373 sin = malloc(sizeof *sin, M_SONAME,
1374 M_WAITOK | M_ZERO);
1375 sin->sin_family = AF_INET;
1376 sin->sin_len = sizeof(*sin);
1377 sin->sin_addr = *addr_p;
1378 sin->sin_port = port;
1379
1380 return (struct sockaddr *)sin;
1381 }
1382
1383 int
in_getsockaddr(struct socket * so,struct sockaddr ** nam)1384 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1385 {
1386 struct inpcb *inp;
1387 struct in_addr addr;
1388 in_port_t port;
1389
1390 inp = sotoinpcb(so);
1391 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1392
1393 INP_RLOCK(inp);
1394 port = inp->inp_lport;
1395 addr = inp->inp_laddr;
1396 INP_RUNLOCK(inp);
1397
1398 *nam = in_sockaddr(port, &addr);
1399 return 0;
1400 }
1401
1402 int
in_getpeeraddr(struct socket * so,struct sockaddr ** nam)1403 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1404 {
1405 struct inpcb *inp;
1406 struct in_addr addr;
1407 in_port_t port;
1408
1409 inp = sotoinpcb(so);
1410 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1411
1412 INP_RLOCK(inp);
1413 port = inp->inp_fport;
1414 addr = inp->inp_faddr;
1415 INP_RUNLOCK(inp);
1416
1417 *nam = in_sockaddr(port, &addr);
1418 return 0;
1419 }
1420
1421 void
in_pcbnotifyall(struct inpcbinfo * pcbinfo,struct in_addr faddr,int errno,struct inpcb * (* notify)(struct inpcb *,int))1422 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1423 struct inpcb *(*notify)(struct inpcb *, int))
1424 {
1425 struct inpcb *inp, *inp_temp;
1426
1427 INP_INFO_WLOCK(pcbinfo);
1428 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1429 INP_WLOCK(inp);
1430 #ifdef INET6
1431 if ((inp->inp_vflag & INP_IPV4) == 0) {
1432 INP_WUNLOCK(inp);
1433 continue;
1434 }
1435 #endif
1436 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1437 inp->inp_socket == NULL) {
1438 INP_WUNLOCK(inp);
1439 continue;
1440 }
1441 if ((*notify)(inp, errno))
1442 INP_WUNLOCK(inp);
1443 }
1444 INP_INFO_WUNLOCK(pcbinfo);
1445 }
1446
1447 void
in_pcbpurgeif0(struct inpcbinfo * pcbinfo,struct ifnet * ifp)1448 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1449 {
1450 struct inpcb *inp;
1451 struct ip_moptions *imo;
1452 int i, gap;
1453
1454 INP_INFO_WLOCK(pcbinfo);
1455 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1456 INP_WLOCK(inp);
1457 imo = inp->inp_moptions;
1458 if ((inp->inp_vflag & INP_IPV4) &&
1459 imo != NULL) {
1460 /*
1461 * Unselect the outgoing interface if it is being
1462 * detached.
1463 */
1464 if (imo->imo_multicast_ifp == ifp)
1465 imo->imo_multicast_ifp = NULL;
1466
1467 /*
1468 * Drop multicast group membership if we joined
1469 * through the interface being detached.
1470 */
1471 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1472 i++) {
1473 if (imo->imo_membership[i]->inm_ifp == ifp) {
1474 in_delmulti(imo->imo_membership[i]);
1475 gap++;
1476 } else if (gap != 0)
1477 imo->imo_membership[i - gap] =
1478 imo->imo_membership[i];
1479 }
1480 imo->imo_num_memberships -= gap;
1481 }
1482 INP_WUNLOCK(inp);
1483 }
1484 INP_INFO_WUNLOCK(pcbinfo);
1485 }
1486
1487 /*
1488 * Lookup a PCB based on the local address and port. Caller must hold the
1489 * hash lock. No inpcb locks or references are acquired.
1490 */
1491 #define INP_LOOKUP_MAPPED_PCB_COST 3
1492 struct inpcb *
in_pcblookup_local(struct inpcbinfo * pcbinfo,struct in_addr laddr,u_short lport,int lookupflags,struct ucred * cred)1493 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1494 u_short lport, int lookupflags, struct ucred *cred)
1495 {
1496 struct inpcb *inp;
1497 #ifdef INET6
1498 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1499 #else
1500 int matchwild = 3;
1501 #endif
1502 int wildcard;
1503
1504 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1505 ("%s: invalid lookup flags %d", __func__, lookupflags));
1506
1507 INP_HASH_LOCK_ASSERT(pcbinfo);
1508
1509 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1510 struct inpcbhead *head;
1511 /*
1512 * Look for an unconnected (wildcard foreign addr) PCB that
1513 * matches the local address and port we're looking for.
1514 */
1515 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1516 0, pcbinfo->ipi_hashmask)];
1517 LIST_FOREACH(inp, head, inp_hash) {
1518 #ifdef INET6
1519 /* XXX inp locking */
1520 if ((inp->inp_vflag & INP_IPV4) == 0)
1521 continue;
1522 #endif
1523 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1524 inp->inp_laddr.s_addr == laddr.s_addr &&
1525 inp->inp_lport == lport) {
1526 /*
1527 * Found?
1528 */
1529 if (cred == NULL ||
1530 prison_equal_ip4(cred->cr_prison,
1531 inp->inp_cred->cr_prison))
1532 return (inp);
1533 }
1534 }
1535 /*
1536 * Not found.
1537 */
1538 return (NULL);
1539 } else {
1540 struct inpcbporthead *porthash;
1541 struct inpcbport *phd;
1542 struct inpcb *match = NULL;
1543 /*
1544 * Best fit PCB lookup.
1545 *
1546 * First see if this local port is in use by looking on the
1547 * port hash list.
1548 */
1549 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1550 pcbinfo->ipi_porthashmask)];
1551 LIST_FOREACH(phd, porthash, phd_hash) {
1552 if (phd->phd_port == lport)
1553 break;
1554 }
1555 if (phd != NULL) {
1556 /*
1557 * Port is in use by one or more PCBs. Look for best
1558 * fit.
1559 */
1560 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1561 wildcard = 0;
1562 if (cred != NULL &&
1563 !prison_equal_ip4(inp->inp_cred->cr_prison,
1564 cred->cr_prison))
1565 continue;
1566 #ifdef INET6
1567 /* XXX inp locking */
1568 if ((inp->inp_vflag & INP_IPV4) == 0)
1569 continue;
1570 /*
1571 * We never select the PCB that has
1572 * INP_IPV6 flag and is bound to :: if
1573 * we have another PCB which is bound
1574 * to 0.0.0.0. If a PCB has the
1575 * INP_IPV6 flag, then we set its cost
1576 * higher than IPv4 only PCBs.
1577 *
1578 * Note that the case only happens
1579 * when a socket is bound to ::, under
1580 * the condition that the use of the
1581 * mapped address is allowed.
1582 */
1583 if ((inp->inp_vflag & INP_IPV6) != 0)
1584 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1585 #endif
1586 if (inp->inp_faddr.s_addr != INADDR_ANY)
1587 wildcard++;
1588 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1589 if (laddr.s_addr == INADDR_ANY)
1590 wildcard++;
1591 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1592 continue;
1593 } else {
1594 if (laddr.s_addr != INADDR_ANY)
1595 wildcard++;
1596 }
1597 if (wildcard < matchwild) {
1598 match = inp;
1599 matchwild = wildcard;
1600 if (matchwild == 0)
1601 break;
1602 }
1603 }
1604 }
1605 return (match);
1606 }
1607 }
1608 #undef INP_LOOKUP_MAPPED_PCB_COST
1609
1610 #ifdef PCBGROUP
1611 /*
1612 * Lookup PCB in hash list, using pcbgroup tables.
1613 */
1614 static struct inpcb *
in_pcblookup_group(struct inpcbinfo * pcbinfo,struct inpcbgroup * pcbgroup,struct in_addr faddr,u_int fport_arg,struct in_addr laddr,u_int lport_arg,int lookupflags,struct ifnet * ifp)1615 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1616 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1617 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1618 {
1619 struct inpcbhead *head;
1620 struct inpcb *inp, *tmpinp;
1621 u_short fport = fport_arg, lport = lport_arg;
1622 bool locked;
1623
1624 /*
1625 * First look for an exact match.
1626 */
1627 tmpinp = NULL;
1628 INP_GROUP_LOCK(pcbgroup);
1629 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1630 pcbgroup->ipg_hashmask)];
1631 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1632 #ifdef INET6
1633 /* XXX inp locking */
1634 if ((inp->inp_vflag & INP_IPV4) == 0)
1635 continue;
1636 #endif
1637 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1638 inp->inp_laddr.s_addr == laddr.s_addr &&
1639 inp->inp_fport == fport &&
1640 inp->inp_lport == lport) {
1641 /*
1642 * XXX We should be able to directly return
1643 * the inp here, without any checks.
1644 * Well unless both bound with SO_REUSEPORT?
1645 */
1646 if (prison_flag(inp->inp_cred, PR_IP4))
1647 goto found;
1648 if (tmpinp == NULL)
1649 tmpinp = inp;
1650 }
1651 }
1652 if (tmpinp != NULL) {
1653 inp = tmpinp;
1654 goto found;
1655 }
1656
1657 #ifdef RSS
1658 /*
1659 * For incoming connections, we may wish to do a wildcard
1660 * match for an RSS-local socket.
1661 */
1662 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1663 struct inpcb *local_wild = NULL, *local_exact = NULL;
1664 #ifdef INET6
1665 struct inpcb *local_wild_mapped = NULL;
1666 #endif
1667 struct inpcb *jail_wild = NULL;
1668 struct inpcbhead *head;
1669 int injail;
1670
1671 /*
1672 * Order of socket selection - we always prefer jails.
1673 * 1. jailed, non-wild.
1674 * 2. jailed, wild.
1675 * 3. non-jailed, non-wild.
1676 * 4. non-jailed, wild.
1677 */
1678
1679 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1680 lport, 0, pcbgroup->ipg_hashmask)];
1681 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1682 #ifdef INET6
1683 /* XXX inp locking */
1684 if ((inp->inp_vflag & INP_IPV4) == 0)
1685 continue;
1686 #endif
1687 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1688 inp->inp_lport != lport)
1689 continue;
1690
1691 injail = prison_flag(inp->inp_cred, PR_IP4);
1692 if (injail) {
1693 if (prison_check_ip4(inp->inp_cred,
1694 &laddr) != 0)
1695 continue;
1696 } else {
1697 if (local_exact != NULL)
1698 continue;
1699 }
1700
1701 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1702 if (injail)
1703 goto found;
1704 else
1705 local_exact = inp;
1706 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1707 #ifdef INET6
1708 /* XXX inp locking, NULL check */
1709 if (inp->inp_vflag & INP_IPV6PROTO)
1710 local_wild_mapped = inp;
1711 else
1712 #endif
1713 if (injail)
1714 jail_wild = inp;
1715 else
1716 local_wild = inp;
1717 }
1718 } /* LIST_FOREACH */
1719
1720 inp = jail_wild;
1721 if (inp == NULL)
1722 inp = local_exact;
1723 if (inp == NULL)
1724 inp = local_wild;
1725 #ifdef INET6
1726 if (inp == NULL)
1727 inp = local_wild_mapped;
1728 #endif
1729 if (inp != NULL)
1730 goto found;
1731 }
1732 #endif
1733
1734 /*
1735 * Then look for a wildcard match, if requested.
1736 */
1737 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1738 struct inpcb *local_wild = NULL, *local_exact = NULL;
1739 #ifdef INET6
1740 struct inpcb *local_wild_mapped = NULL;
1741 #endif
1742 struct inpcb *jail_wild = NULL;
1743 struct inpcbhead *head;
1744 int injail;
1745
1746 /*
1747 * Order of socket selection - we always prefer jails.
1748 * 1. jailed, non-wild.
1749 * 2. jailed, wild.
1750 * 3. non-jailed, non-wild.
1751 * 4. non-jailed, wild.
1752 */
1753 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1754 0, pcbinfo->ipi_wildmask)];
1755 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1756 #ifdef INET6
1757 /* XXX inp locking */
1758 if ((inp->inp_vflag & INP_IPV4) == 0)
1759 continue;
1760 #endif
1761 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1762 inp->inp_lport != lport)
1763 continue;
1764
1765 injail = prison_flag(inp->inp_cred, PR_IP4);
1766 if (injail) {
1767 if (prison_check_ip4(inp->inp_cred,
1768 &laddr) != 0)
1769 continue;
1770 } else {
1771 if (local_exact != NULL)
1772 continue;
1773 }
1774
1775 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1776 if (injail)
1777 goto found;
1778 else
1779 local_exact = inp;
1780 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1781 #ifdef INET6
1782 /* XXX inp locking, NULL check */
1783 if (inp->inp_vflag & INP_IPV6PROTO)
1784 local_wild_mapped = inp;
1785 else
1786 #endif
1787 if (injail)
1788 jail_wild = inp;
1789 else
1790 local_wild = inp;
1791 }
1792 } /* LIST_FOREACH */
1793 inp = jail_wild;
1794 if (inp == NULL)
1795 inp = local_exact;
1796 if (inp == NULL)
1797 inp = local_wild;
1798 #ifdef INET6
1799 if (inp == NULL)
1800 inp = local_wild_mapped;
1801 #endif
1802 if (inp != NULL)
1803 goto found;
1804 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1805 INP_GROUP_UNLOCK(pcbgroup);
1806 return (NULL);
1807
1808 found:
1809 if (lookupflags & INPLOOKUP_WLOCKPCB)
1810 locked = INP_TRY_WLOCK(inp);
1811 else if (lookupflags & INPLOOKUP_RLOCKPCB)
1812 locked = INP_TRY_RLOCK(inp);
1813 else
1814 panic("%s: locking bug", __func__);
1815 if (!locked)
1816 in_pcbref(inp);
1817 INP_GROUP_UNLOCK(pcbgroup);
1818 if (!locked) {
1819 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1820 INP_WLOCK(inp);
1821 if (in_pcbrele_wlocked(inp))
1822 return (NULL);
1823 } else {
1824 INP_RLOCK(inp);
1825 if (in_pcbrele_rlocked(inp))
1826 return (NULL);
1827 }
1828 }
1829 #ifdef INVARIANTS
1830 if (lookupflags & INPLOOKUP_WLOCKPCB)
1831 INP_WLOCK_ASSERT(inp);
1832 else
1833 INP_RLOCK_ASSERT(inp);
1834 #endif
1835 return (inp);
1836 }
1837 #endif /* PCBGROUP */
1838
1839 /*
1840 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
1841 * that the caller has locked the hash list, and will not perform any further
1842 * locking or reference operations on either the hash list or the connection.
1843 */
1844 static struct inpcb *
in_pcblookup_hash_locked(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport_arg,struct in_addr laddr,u_int lport_arg,int lookupflags,struct ifnet * ifp)1845 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1846 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1847 struct ifnet *ifp)
1848 {
1849 struct inpcbhead *head;
1850 struct inpcb *inp, *tmpinp;
1851 u_short fport = fport_arg, lport = lport_arg;
1852
1853 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1854 ("%s: invalid lookup flags %d", __func__, lookupflags));
1855
1856 INP_HASH_LOCK_ASSERT(pcbinfo);
1857
1858 /*
1859 * First look for an exact match.
1860 */
1861 tmpinp = NULL;
1862 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1863 pcbinfo->ipi_hashmask)];
1864 LIST_FOREACH(inp, head, inp_hash) {
1865 #ifdef INET6
1866 /* XXX inp locking */
1867 if ((inp->inp_vflag & INP_IPV4) == 0)
1868 continue;
1869 #endif
1870 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1871 inp->inp_laddr.s_addr == laddr.s_addr &&
1872 inp->inp_fport == fport &&
1873 inp->inp_lport == lport) {
1874 /*
1875 * XXX We should be able to directly return
1876 * the inp here, without any checks.
1877 * Well unless both bound with SO_REUSEPORT?
1878 */
1879 if (prison_flag(inp->inp_cred, PR_IP4))
1880 return (inp);
1881 if (tmpinp == NULL)
1882 tmpinp = inp;
1883 }
1884 }
1885 if (tmpinp != NULL)
1886 return (tmpinp);
1887
1888 /*
1889 * Then look for a wildcard match, if requested.
1890 */
1891 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1892 struct inpcb *local_wild = NULL, *local_exact = NULL;
1893 #ifdef INET6
1894 struct inpcb *local_wild_mapped = NULL;
1895 #endif
1896 struct inpcb *jail_wild = NULL;
1897 int injail;
1898
1899 /*
1900 * Order of socket selection - we always prefer jails.
1901 * 1. jailed, non-wild.
1902 * 2. jailed, wild.
1903 * 3. non-jailed, non-wild.
1904 * 4. non-jailed, wild.
1905 */
1906
1907 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1908 0, pcbinfo->ipi_hashmask)];
1909 LIST_FOREACH(inp, head, inp_hash) {
1910 #ifdef INET6
1911 /* XXX inp locking */
1912 if ((inp->inp_vflag & INP_IPV4) == 0)
1913 continue;
1914 #endif
1915 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1916 inp->inp_lport != lport)
1917 continue;
1918
1919 injail = prison_flag(inp->inp_cred, PR_IP4);
1920 if (injail) {
1921 if (prison_check_ip4(inp->inp_cred,
1922 &laddr) != 0)
1923 continue;
1924 } else {
1925 if (local_exact != NULL)
1926 continue;
1927 }
1928
1929 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1930 if (injail)
1931 return (inp);
1932 else
1933 local_exact = inp;
1934 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1935 #ifdef INET6
1936 /* XXX inp locking, NULL check */
1937 if (inp->inp_vflag & INP_IPV6PROTO)
1938 local_wild_mapped = inp;
1939 else
1940 #endif
1941 if (injail)
1942 jail_wild = inp;
1943 else
1944 local_wild = inp;
1945 }
1946 } /* LIST_FOREACH */
1947 if (jail_wild != NULL)
1948 return (jail_wild);
1949 if (local_exact != NULL)
1950 return (local_exact);
1951 if (local_wild != NULL)
1952 return (local_wild);
1953 #ifdef INET6
1954 if (local_wild_mapped != NULL)
1955 return (local_wild_mapped);
1956 #endif
1957 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1958
1959 return (NULL);
1960 }
1961
1962 /*
1963 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
1964 * hash list lock, and will return the inpcb locked (i.e., requires
1965 * INPLOOKUP_LOCKPCB).
1966 */
1967 static struct inpcb *
in_pcblookup_hash(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport,struct in_addr laddr,u_int lport,int lookupflags,struct ifnet * ifp)1968 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1969 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1970 struct ifnet *ifp)
1971 {
1972 struct inpcb *inp;
1973 bool locked;
1974
1975 INP_HASH_RLOCK(pcbinfo);
1976 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1977 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1978 if (inp != NULL) {
1979 if (lookupflags & INPLOOKUP_WLOCKPCB)
1980 locked = INP_TRY_WLOCK(inp);
1981 else if (lookupflags & INPLOOKUP_RLOCKPCB)
1982 locked = INP_TRY_RLOCK(inp);
1983 else
1984 panic("%s: locking bug", __func__);
1985 if (!locked)
1986 in_pcbref(inp);
1987 INP_HASH_RUNLOCK(pcbinfo);
1988 if (!locked) {
1989 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1990 INP_WLOCK(inp);
1991 if (in_pcbrele_wlocked(inp))
1992 return (NULL);
1993 } else {
1994 INP_RLOCK(inp);
1995 if (in_pcbrele_rlocked(inp))
1996 return (NULL);
1997 }
1998 }
1999 #ifdef INVARIANTS
2000 if (lookupflags & INPLOOKUP_WLOCKPCB)
2001 INP_WLOCK_ASSERT(inp);
2002 else
2003 INP_RLOCK_ASSERT(inp);
2004 #endif
2005 } else
2006 INP_HASH_RUNLOCK(pcbinfo);
2007 return (inp);
2008 }
2009
2010 /*
2011 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2012 * from which a pre-calculated hash value may be extracted.
2013 *
2014 * Possibly more of this logic should be in in_pcbgroup.c.
2015 */
2016 struct inpcb *
in_pcblookup(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport,struct in_addr laddr,u_int lport,int lookupflags,struct ifnet * ifp)2017 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2018 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2019 {
2020 #if defined(PCBGROUP) && !defined(RSS)
2021 struct inpcbgroup *pcbgroup;
2022 #endif
2023
2024 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2025 ("%s: invalid lookup flags %d", __func__, lookupflags));
2026 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2027 ("%s: LOCKPCB not set", __func__));
2028
2029 /*
2030 * When not using RSS, use connection groups in preference to the
2031 * reservation table when looking up 4-tuples. When using RSS, just
2032 * use the reservation table, due to the cost of the Toeplitz hash
2033 * in software.
2034 *
2035 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2036 * we could be doing RSS with a non-Toeplitz hash that is affordable
2037 * in software.
2038 */
2039 #if defined(PCBGROUP) && !defined(RSS)
2040 if (in_pcbgroup_enabled(pcbinfo)) {
2041 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2042 fport);
2043 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2044 laddr, lport, lookupflags, ifp));
2045 }
2046 #endif
2047 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2048 lookupflags, ifp));
2049 }
2050
2051 struct inpcb *
in_pcblookup_mbuf(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport,struct in_addr laddr,u_int lport,int lookupflags,struct ifnet * ifp,struct mbuf * m)2052 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2053 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2054 struct ifnet *ifp, struct mbuf *m)
2055 {
2056 #ifdef PCBGROUP
2057 struct inpcbgroup *pcbgroup;
2058 #endif
2059
2060 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2061 ("%s: invalid lookup flags %d", __func__, lookupflags));
2062 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2063 ("%s: LOCKPCB not set", __func__));
2064
2065 #ifdef PCBGROUP
2066 /*
2067 * If we can use a hardware-generated hash to look up the connection
2068 * group, use that connection group to find the inpcb. Otherwise
2069 * fall back on a software hash -- or the reservation table if we're
2070 * using RSS.
2071 *
2072 * XXXRW: As above, that policy belongs in the pcbgroup code.
2073 */
2074 if (in_pcbgroup_enabled(pcbinfo) &&
2075 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2076 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2077 m->m_pkthdr.flowid);
2078 if (pcbgroup != NULL)
2079 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2080 fport, laddr, lport, lookupflags, ifp));
2081 #ifndef RSS
2082 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2083 fport);
2084 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2085 laddr, lport, lookupflags, ifp));
2086 #endif
2087 }
2088 #endif
2089 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2090 lookupflags, ifp));
2091 }
2092 #endif /* INET */
2093
2094 /*
2095 * Insert PCB onto various hash lists.
2096 */
2097 static int
in_pcbinshash_internal(struct inpcb * inp,int do_pcbgroup_update)2098 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2099 {
2100 struct inpcbhead *pcbhash;
2101 struct inpcbporthead *pcbporthash;
2102 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2103 struct inpcbport *phd;
2104 u_int32_t hashkey_faddr;
2105
2106 INP_WLOCK_ASSERT(inp);
2107 INP_HASH_WLOCK_ASSERT(pcbinfo);
2108
2109 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2110 ("in_pcbinshash: INP_INHASHLIST"));
2111
2112 #ifdef INET6
2113 if (inp->inp_vflag & INP_IPV6)
2114 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2115 else
2116 #endif
2117 hashkey_faddr = inp->inp_faddr.s_addr;
2118
2119 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2120 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2121
2122 pcbporthash = &pcbinfo->ipi_porthashbase[
2123 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2124
2125 /*
2126 * Go through port list and look for a head for this lport.
2127 */
2128 LIST_FOREACH(phd, pcbporthash, phd_hash) {
2129 if (phd->phd_port == inp->inp_lport)
2130 break;
2131 }
2132 /*
2133 * If none exists, malloc one and tack it on.
2134 */
2135 if (phd == NULL) {
2136 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2137 if (phd == NULL) {
2138 return (ENOBUFS); /* XXX */
2139 }
2140 phd->phd_port = inp->inp_lport;
2141 LIST_INIT(&phd->phd_pcblist);
2142 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2143 }
2144 inp->inp_phd = phd;
2145 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2146 LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2147 inp->inp_flags |= INP_INHASHLIST;
2148 #ifdef PCBGROUP
2149 if (do_pcbgroup_update)
2150 in_pcbgroup_update(inp);
2151 #endif
2152 return (0);
2153 }
2154
2155 /*
2156 * For now, there are two public interfaces to insert an inpcb into the hash
2157 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2158 * is used only in the TCP syncache, where in_pcbinshash is called before the
2159 * full 4-tuple is set for the inpcb, and we don't want to install in the
2160 * pcbgroup until later.
2161 *
2162 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2163 * connection groups, and partially initialised inpcbs should not be exposed
2164 * to either reservation hash tables or pcbgroups.
2165 */
2166 int
in_pcbinshash(struct inpcb * inp)2167 in_pcbinshash(struct inpcb *inp)
2168 {
2169
2170 return (in_pcbinshash_internal(inp, 1));
2171 }
2172
2173 int
in_pcbinshash_nopcbgroup(struct inpcb * inp)2174 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2175 {
2176
2177 return (in_pcbinshash_internal(inp, 0));
2178 }
2179
2180 /*
2181 * Move PCB to the proper hash bucket when { faddr, fport } have been
2182 * changed. NOTE: This does not handle the case of the lport changing (the
2183 * hashed port list would have to be updated as well), so the lport must
2184 * not change after in_pcbinshash() has been called.
2185 */
2186 void
in_pcbrehash_mbuf(struct inpcb * inp,struct mbuf * m)2187 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2188 {
2189 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2190 struct inpcbhead *head;
2191 u_int32_t hashkey_faddr;
2192
2193 INP_WLOCK_ASSERT(inp);
2194 INP_HASH_WLOCK_ASSERT(pcbinfo);
2195
2196 KASSERT(inp->inp_flags & INP_INHASHLIST,
2197 ("in_pcbrehash: !INP_INHASHLIST"));
2198
2199 #ifdef INET6
2200 if (inp->inp_vflag & INP_IPV6)
2201 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2202 else
2203 #endif
2204 hashkey_faddr = inp->inp_faddr.s_addr;
2205
2206 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2207 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2208
2209 LIST_REMOVE(inp, inp_hash);
2210 LIST_INSERT_HEAD(head, inp, inp_hash);
2211
2212 #ifdef PCBGROUP
2213 if (m != NULL)
2214 in_pcbgroup_update_mbuf(inp, m);
2215 else
2216 in_pcbgroup_update(inp);
2217 #endif
2218 }
2219
2220 void
in_pcbrehash(struct inpcb * inp)2221 in_pcbrehash(struct inpcb *inp)
2222 {
2223
2224 in_pcbrehash_mbuf(inp, NULL);
2225 }
2226
2227 /*
2228 * Remove PCB from various lists.
2229 */
2230 static void
in_pcbremlists(struct inpcb * inp)2231 in_pcbremlists(struct inpcb *inp)
2232 {
2233 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2234
2235 #ifdef INVARIANTS
2236 if (pcbinfo == &V_tcbinfo) {
2237 INP_INFO_RLOCK_ASSERT(pcbinfo);
2238 } else {
2239 INP_INFO_WLOCK_ASSERT(pcbinfo);
2240 }
2241 #endif
2242
2243 INP_WLOCK_ASSERT(inp);
2244 INP_LIST_WLOCK_ASSERT(pcbinfo);
2245
2246 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2247 if (inp->inp_flags & INP_INHASHLIST) {
2248 struct inpcbport *phd = inp->inp_phd;
2249
2250 INP_HASH_WLOCK(pcbinfo);
2251 LIST_REMOVE(inp, inp_hash);
2252 LIST_REMOVE(inp, inp_portlist);
2253 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2254 LIST_REMOVE(phd, phd_hash);
2255 free(phd, M_PCB);
2256 }
2257 INP_HASH_WUNLOCK(pcbinfo);
2258 inp->inp_flags &= ~INP_INHASHLIST;
2259 }
2260 LIST_REMOVE(inp, inp_list);
2261 pcbinfo->ipi_count--;
2262 #ifdef PCBGROUP
2263 in_pcbgroup_remove(inp);
2264 #endif
2265 }
2266
2267 /*
2268 * Check for alternatives when higher level complains
2269 * about service problems. For now, invalidate cached
2270 * routing information. If the route was created dynamically
2271 * (by a redirect), time to try a default gateway again.
2272 */
2273 void
in_losing(struct inpcb * inp)2274 in_losing(struct inpcb *inp)
2275 {
2276
2277 if (inp->inp_route.ro_rt) {
2278 RTFREE(inp->inp_route.ro_rt);
2279 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2280 }
2281 if (inp->inp_route.ro_lle)
2282 LLE_FREE(inp->inp_route.ro_lle); /* zeros ro_lle */
2283 return;
2284 }
2285
2286 /*
2287 * A set label operation has occurred at the socket layer, propagate the
2288 * label change into the in_pcb for the socket.
2289 */
2290 void
in_pcbsosetlabel(struct socket * so)2291 in_pcbsosetlabel(struct socket *so)
2292 {
2293 #ifdef MAC
2294 struct inpcb *inp;
2295
2296 inp = sotoinpcb(so);
2297 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2298
2299 INP_WLOCK(inp);
2300 SOCK_LOCK(so);
2301 mac_inpcb_sosetlabel(so, inp);
2302 SOCK_UNLOCK(so);
2303 INP_WUNLOCK(inp);
2304 #endif
2305 }
2306
2307 /*
2308 * ipport_tick runs once per second, determining if random port allocation
2309 * should be continued. If more than ipport_randomcps ports have been
2310 * allocated in the last second, then we return to sequential port
2311 * allocation. We return to random allocation only once we drop below
2312 * ipport_randomcps for at least ipport_randomtime seconds.
2313 */
2314 static void
ipport_tick(void * xtp)2315 ipport_tick(void *xtp)
2316 {
2317 VNET_ITERATOR_DECL(vnet_iter);
2318
2319 VNET_LIST_RLOCK_NOSLEEP();
2320 VNET_FOREACH(vnet_iter) {
2321 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2322 if (V_ipport_tcpallocs <=
2323 V_ipport_tcplastcount + V_ipport_randomcps) {
2324 if (V_ipport_stoprandom > 0)
2325 V_ipport_stoprandom--;
2326 } else
2327 V_ipport_stoprandom = V_ipport_randomtime;
2328 V_ipport_tcplastcount = V_ipport_tcpallocs;
2329 CURVNET_RESTORE();
2330 }
2331 VNET_LIST_RUNLOCK_NOSLEEP();
2332 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2333 }
2334
2335 static void
ip_fini(void * xtp)2336 ip_fini(void *xtp)
2337 {
2338
2339 callout_stop(&ipport_tick_callout);
2340 }
2341
2342 /*
2343 * The ipport_callout should start running at about the time we attach the
2344 * inet or inet6 domains.
2345 */
2346 static void
ipport_tick_init(const void * unused __unused)2347 ipport_tick_init(const void *unused __unused)
2348 {
2349
2350 /* Start ipport_tick. */
2351 callout_init(&ipport_tick_callout, 1);
2352 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2353 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2354 SHUTDOWN_PRI_DEFAULT);
2355 }
2356 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2357 ipport_tick_init, NULL);
2358
2359 void
inp_wlock(struct inpcb * inp)2360 inp_wlock(struct inpcb *inp)
2361 {
2362
2363 INP_WLOCK(inp);
2364 }
2365
2366 void
inp_wunlock(struct inpcb * inp)2367 inp_wunlock(struct inpcb *inp)
2368 {
2369
2370 INP_WUNLOCK(inp);
2371 }
2372
2373 void
inp_rlock(struct inpcb * inp)2374 inp_rlock(struct inpcb *inp)
2375 {
2376
2377 INP_RLOCK(inp);
2378 }
2379
2380 void
inp_runlock(struct inpcb * inp)2381 inp_runlock(struct inpcb *inp)
2382 {
2383
2384 INP_RUNLOCK(inp);
2385 }
2386
2387 #ifdef INVARIANTS
2388 void
inp_lock_assert(struct inpcb * inp)2389 inp_lock_assert(struct inpcb *inp)
2390 {
2391
2392 INP_WLOCK_ASSERT(inp);
2393 }
2394
2395 void
inp_unlock_assert(struct inpcb * inp)2396 inp_unlock_assert(struct inpcb *inp)
2397 {
2398
2399 INP_UNLOCK_ASSERT(inp);
2400 }
2401 #endif
2402
2403 void
inp_apply_all(void (* func)(struct inpcb *,void *),void * arg)2404 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2405 {
2406 struct inpcb *inp;
2407
2408 INP_INFO_WLOCK(&V_tcbinfo);
2409 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2410 INP_WLOCK(inp);
2411 func(inp, arg);
2412 INP_WUNLOCK(inp);
2413 }
2414 INP_INFO_WUNLOCK(&V_tcbinfo);
2415 }
2416
2417 struct socket *
inp_inpcbtosocket(struct inpcb * inp)2418 inp_inpcbtosocket(struct inpcb *inp)
2419 {
2420
2421 INP_WLOCK_ASSERT(inp);
2422 return (inp->inp_socket);
2423 }
2424
2425 struct tcpcb *
inp_inpcbtotcpcb(struct inpcb * inp)2426 inp_inpcbtotcpcb(struct inpcb *inp)
2427 {
2428
2429 INP_WLOCK_ASSERT(inp);
2430 return ((struct tcpcb *)inp->inp_ppcb);
2431 }
2432
2433 int
inp_ip_tos_get(const struct inpcb * inp)2434 inp_ip_tos_get(const struct inpcb *inp)
2435 {
2436
2437 return (inp->inp_ip_tos);
2438 }
2439
2440 void
inp_ip_tos_set(struct inpcb * inp,int val)2441 inp_ip_tos_set(struct inpcb *inp, int val)
2442 {
2443
2444 inp->inp_ip_tos = val;
2445 }
2446
2447 void
inp_4tuple_get(struct inpcb * inp,uint32_t * laddr,uint16_t * lp,uint32_t * faddr,uint16_t * fp)2448 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2449 uint32_t *faddr, uint16_t *fp)
2450 {
2451
2452 INP_LOCK_ASSERT(inp);
2453 *laddr = inp->inp_laddr.s_addr;
2454 *faddr = inp->inp_faddr.s_addr;
2455 *lp = inp->inp_lport;
2456 *fp = inp->inp_fport;
2457 }
2458
2459 struct inpcb *
so_sotoinpcb(struct socket * so)2460 so_sotoinpcb(struct socket *so)
2461 {
2462
2463 return (sotoinpcb(so));
2464 }
2465
2466 struct tcpcb *
so_sototcpcb(struct socket * so)2467 so_sototcpcb(struct socket *so)
2468 {
2469
2470 return (sototcpcb(so));
2471 }
2472
2473 #ifdef DDB
2474 static void
db_print_indent(int indent)2475 db_print_indent(int indent)
2476 {
2477 int i;
2478
2479 for (i = 0; i < indent; i++)
2480 db_printf(" ");
2481 }
2482
2483 static void
db_print_inconninfo(struct in_conninfo * inc,const char * name,int indent)2484 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2485 {
2486 char faddr_str[48], laddr_str[48];
2487
2488 db_print_indent(indent);
2489 db_printf("%s at %p\n", name, inc);
2490
2491 indent += 2;
2492
2493 #ifdef INET6
2494 if (inc->inc_flags & INC_ISIPV6) {
2495 /* IPv6. */
2496 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2497 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2498 } else
2499 #endif
2500 {
2501 /* IPv4. */
2502 inet_ntoa_r(inc->inc_laddr, laddr_str);
2503 inet_ntoa_r(inc->inc_faddr, faddr_str);
2504 }
2505 db_print_indent(indent);
2506 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2507 ntohs(inc->inc_lport));
2508 db_print_indent(indent);
2509 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2510 ntohs(inc->inc_fport));
2511 }
2512
2513 static void
db_print_inpflags(int inp_flags)2514 db_print_inpflags(int inp_flags)
2515 {
2516 int comma;
2517
2518 comma = 0;
2519 if (inp_flags & INP_RECVOPTS) {
2520 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2521 comma = 1;
2522 }
2523 if (inp_flags & INP_RECVRETOPTS) {
2524 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2525 comma = 1;
2526 }
2527 if (inp_flags & INP_RECVDSTADDR) {
2528 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2529 comma = 1;
2530 }
2531 if (inp_flags & INP_HDRINCL) {
2532 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2533 comma = 1;
2534 }
2535 if (inp_flags & INP_HIGHPORT) {
2536 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2537 comma = 1;
2538 }
2539 if (inp_flags & INP_LOWPORT) {
2540 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2541 comma = 1;
2542 }
2543 if (inp_flags & INP_ANONPORT) {
2544 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2545 comma = 1;
2546 }
2547 if (inp_flags & INP_RECVIF) {
2548 db_printf("%sINP_RECVIF", comma ? ", " : "");
2549 comma = 1;
2550 }
2551 if (inp_flags & INP_MTUDISC) {
2552 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2553 comma = 1;
2554 }
2555 if (inp_flags & INP_RECVTTL) {
2556 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2557 comma = 1;
2558 }
2559 if (inp_flags & INP_DONTFRAG) {
2560 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2561 comma = 1;
2562 }
2563 if (inp_flags & INP_RECVTOS) {
2564 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2565 comma = 1;
2566 }
2567 if (inp_flags & IN6P_IPV6_V6ONLY) {
2568 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2569 comma = 1;
2570 }
2571 if (inp_flags & IN6P_PKTINFO) {
2572 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2573 comma = 1;
2574 }
2575 if (inp_flags & IN6P_HOPLIMIT) {
2576 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2577 comma = 1;
2578 }
2579 if (inp_flags & IN6P_HOPOPTS) {
2580 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2581 comma = 1;
2582 }
2583 if (inp_flags & IN6P_DSTOPTS) {
2584 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2585 comma = 1;
2586 }
2587 if (inp_flags & IN6P_RTHDR) {
2588 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2589 comma = 1;
2590 }
2591 if (inp_flags & IN6P_RTHDRDSTOPTS) {
2592 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2593 comma = 1;
2594 }
2595 if (inp_flags & IN6P_TCLASS) {
2596 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2597 comma = 1;
2598 }
2599 if (inp_flags & IN6P_AUTOFLOWLABEL) {
2600 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2601 comma = 1;
2602 }
2603 if (inp_flags & INP_TIMEWAIT) {
2604 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2605 comma = 1;
2606 }
2607 if (inp_flags & INP_ONESBCAST) {
2608 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2609 comma = 1;
2610 }
2611 if (inp_flags & INP_DROPPED) {
2612 db_printf("%sINP_DROPPED", comma ? ", " : "");
2613 comma = 1;
2614 }
2615 if (inp_flags & INP_SOCKREF) {
2616 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2617 comma = 1;
2618 }
2619 if (inp_flags & IN6P_RFC2292) {
2620 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2621 comma = 1;
2622 }
2623 if (inp_flags & IN6P_MTU) {
2624 db_printf("IN6P_MTU%s", comma ? ", " : "");
2625 comma = 1;
2626 }
2627 }
2628
2629 static void
db_print_inpvflag(u_char inp_vflag)2630 db_print_inpvflag(u_char inp_vflag)
2631 {
2632 int comma;
2633
2634 comma = 0;
2635 if (inp_vflag & INP_IPV4) {
2636 db_printf("%sINP_IPV4", comma ? ", " : "");
2637 comma = 1;
2638 }
2639 if (inp_vflag & INP_IPV6) {
2640 db_printf("%sINP_IPV6", comma ? ", " : "");
2641 comma = 1;
2642 }
2643 if (inp_vflag & INP_IPV6PROTO) {
2644 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2645 comma = 1;
2646 }
2647 }
2648
2649 static void
db_print_inpcb(struct inpcb * inp,const char * name,int indent)2650 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2651 {
2652
2653 db_print_indent(indent);
2654 db_printf("%s at %p\n", name, inp);
2655
2656 indent += 2;
2657
2658 db_print_indent(indent);
2659 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2660
2661 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2662
2663 db_print_indent(indent);
2664 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
2665 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2666
2667 db_print_indent(indent);
2668 db_printf("inp_label: %p inp_flags: 0x%x (",
2669 inp->inp_label, inp->inp_flags);
2670 db_print_inpflags(inp->inp_flags);
2671 db_printf(")\n");
2672
2673 db_print_indent(indent);
2674 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
2675 inp->inp_vflag);
2676 db_print_inpvflag(inp->inp_vflag);
2677 db_printf(")\n");
2678
2679 db_print_indent(indent);
2680 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
2681 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2682
2683 db_print_indent(indent);
2684 #ifdef INET6
2685 if (inp->inp_vflag & INP_IPV6) {
2686 db_printf("in6p_options: %p in6p_outputopts: %p "
2687 "in6p_moptions: %p\n", inp->in6p_options,
2688 inp->in6p_outputopts, inp->in6p_moptions);
2689 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
2690 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2691 inp->in6p_hops);
2692 } else
2693 #endif
2694 {
2695 db_printf("inp_ip_tos: %d inp_ip_options: %p "
2696 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2697 inp->inp_options, inp->inp_moptions);
2698 }
2699
2700 db_print_indent(indent);
2701 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
2702 (uintmax_t)inp->inp_gencnt);
2703 }
2704
DB_SHOW_COMMAND(inpcb,db_show_inpcb)2705 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2706 {
2707 struct inpcb *inp;
2708
2709 if (!have_addr) {
2710 db_printf("usage: show inpcb <addr>\n");
2711 return;
2712 }
2713 inp = (struct inpcb *)addr;
2714
2715 db_print_inpcb(inp, "inpcb", 0);
2716 }
2717 #endif /* DDB */
2718