1 /*
2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
4 *
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34 /*
35 * Copyright (c) 1982, 1986, 1991, 1993, 1995
36 * The Regents of the University of California. All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. Neither the name of the University nor the names of its contributors
47 * may be used to endorse or promote products derived from this software
48 * without specific prior written permission.
49 *
50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60 * SUCH DAMAGE.
61 *
62 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
63 * $FreeBSD: src/sys/netinet/in_pcb.c,v 1.59.2.27 2004/01/02 04:06:42 ambrisko Exp $
64 */
65
66 #include "opt_inet6.h"
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/malloc.h>
71 #include <sys/mbuf.h>
72 #include <sys/domain.h>
73 #include <sys/protosw.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/proc.h>
77 #include <sys/caps.h>
78 #include <sys/jail.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81
82 #include <sys/socketvar2.h>
83 #include <sys/msgport2.h>
84
85 #include <machine/limits.h>
86
87 #include <net/if.h>
88 #include <net/if_types.h>
89 #include <net/route.h>
90 #include <net/netisr2.h>
91 #include <net/toeplitz2.h>
92
93 #include <netinet/in.h>
94 #include <netinet/in_pcb.h>
95 #include <netinet/in_var.h>
96 #include <netinet/ip_var.h>
97 #ifdef INET6
98 #include <netinet/ip6.h>
99 #include <netinet6/ip6_var.h>
100 #endif /* INET6 */
101
102 #define INP_LOCALGROUP_SIZMIN 8
103 #define INP_LOCALGROUP_SIZMAX 256
104
105 static struct inpcb *in_pcblookup_local(struct inpcbporthead *porthash,
106 struct in_addr laddr, u_int lport_arg, int wild_okay,
107 struct ucred *cred);
108
109 struct in_addr zeroin_addr;
110
111 /*
112 * These configure the range of local port addresses assigned to
113 * "unspecified" outgoing connections/packets/whatever.
114 */
115 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */
116 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */
117
118 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */
119 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */
120
121 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */
122 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */
123
124 #define RANGECHK(var, min, max) \
125 if ((var) < (min)) { (var) = (min); } \
126 else if ((var) > (max)) { (var) = (max); }
127
128 int udpencap_enable = 1; /* enabled by default */
129 int udpencap_port = 4500; /* triggers decapsulation */
130
131 /*
132 * Per-netisr inpcb markers.
133 * NOTE: they should only be used in netisrs.
134 */
135 static struct inpcb *in_pcbmarkers;
136 static struct inpcontainer *in_pcbcontainer_markers;
137
138 static int
sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)139 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
140 {
141 int error;
142
143 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
144 if (!error) {
145 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
146 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
147
148 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX);
149 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX);
150
151 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX);
152 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX);
153 }
154 return (error);
155 }
156
157 #undef RANGECHK
158
159 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
160
161 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW,
162 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", "");
163 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW,
164 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", "");
165 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW,
166 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", "");
167 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW,
168 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", "");
169 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW,
170 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", "");
171 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW,
172 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", "");
173
174 /* Initialized by ip_init() */
175 int ip_porthash_trycount;
176 SYSCTL_INT(_net_inet_ip, OID_AUTO, porthash_trycount, CTLFLAG_RW,
177 &ip_porthash_trycount, 0,
178 "Number of tries to find local port matching hash of 4-tuple");
179
180 /*
181 * in_pcb.c: manage the Protocol Control Blocks.
182 *
183 * NOTE: It is assumed that most of these functions will be called from
184 * a critical section. XXX - There are, unfortunately, a few exceptions
185 * to this rule that should be fixed.
186 *
187 * NOTE: The caller should initialize the cpu field to the cpu running the
188 * protocol stack associated with this inpcbinfo.
189 */
190
191 void
in_pcbinfo_init(struct inpcbinfo * pcbinfo,int cpu,boolean_t shared)192 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared)
193 {
194 KASSERT(cpu >= 0 && cpu < netisr_ncpus, ("invalid cpu%d", cpu));
195 pcbinfo->cpu = cpu;
196
197 LIST_INIT(&pcbinfo->pcblisthead);
198 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB,
199 M_WAITOK | M_ZERO);
200
201 if (shared) {
202 pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token),
203 M_PCB, M_WAITOK);
204 lwkt_token_init(pcbinfo->infotoken, "infotoken");
205 } else {
206 pcbinfo->infotoken = NULL;
207 }
208 }
209
210 void
in_pcbportinfo_set(struct inpcbinfo * pcbinfo,struct inpcbportinfo * portinfo,int portinfo_cnt)211 in_pcbportinfo_set(struct inpcbinfo *pcbinfo, struct inpcbportinfo *portinfo,
212 int portinfo_cnt)
213 {
214
215 KASSERT(portinfo_cnt > 0, ("invalid portinfo_cnt %d", portinfo_cnt));
216 pcbinfo->portinfo = portinfo;
217 pcbinfo->portinfo_cnt = portinfo_cnt;
218 }
219
220 struct baddynamicports baddynamicports;
221
222 /*
223 * Check if the specified port is invalid for dynamic allocation.
224 */
225 int
in_baddynamic(u_int16_t port,u_int16_t proto)226 in_baddynamic(u_int16_t port, u_int16_t proto)
227 {
228 switch (proto) {
229 case IPPROTO_TCP:
230 return (DP_ISSET(baddynamicports.tcp, port));
231 case IPPROTO_UDP:
232 return (DP_ISSET(baddynamicports.udp, port));
233 default:
234 return (0);
235 }
236 }
237
238 void
in_pcbonlist(struct inpcb * inp)239 in_pcbonlist(struct inpcb *inp)
240 {
241 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
242
243 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
244 ("not in the correct netisr"));
245 KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist"));
246 inp->inp_flags |= INP_ONLIST;
247
248 GET_PCBINFO_TOKEN(pcbinfo);
249 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
250 pcbinfo->ipi_count++;
251 REL_PCBINFO_TOKEN(pcbinfo);
252 }
253
254 void
in_pcbofflist(struct inpcb * inp)255 in_pcbofflist(struct inpcb *inp)
256 {
257 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
258
259 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
260 ("not in the correct netisr"));
261 KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist"));
262 inp->inp_flags &= ~INP_ONLIST;
263
264 GET_PCBINFO_TOKEN(pcbinfo);
265 LIST_REMOVE(inp, inp_list);
266 KASSERT(pcbinfo->ipi_count > 0,
267 ("invalid inpcb count %d", pcbinfo->ipi_count));
268 pcbinfo->ipi_count--;
269 REL_PCBINFO_TOKEN(pcbinfo);
270 }
271
272 /*
273 * Allocate a PCB and associate it with the socket.
274 */
275 int
in_pcballoc(struct socket * so,struct inpcbinfo * pcbinfo)276 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
277 {
278 struct inpcb *inp;
279
280 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK);
281 if (inp == NULL)
282 return (ENOMEM);
283 inp->inp_lgrpindex = -1;
284 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
285 inp->inp_pcbinfo = pcbinfo;
286 inp->inp_socket = so;
287 #ifdef INET6
288 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
289 if (ip6_auto_flowlabel)
290 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
291 inp->inp_af = AF_INET6;
292 } else
293 #endif
294 inp->inp_af = AF_INET;
295 soreference(so);
296 so->so_pcb = inp;
297
298 in_pcbonlist(inp);
299 return (0);
300 }
301
302 /*
303 * Unlink a pcb with the intention of moving it to another cpu with a
304 * different pcbinfo. While unlinked nothing should attempt to dereference
305 * inp_pcbinfo, NULL it out so we assert if it does.
306 */
307 void
in_pcbunlink_flags(struct inpcb * inp,struct inpcbinfo * pcbinfo,int flags)308 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
309 {
310 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
311 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
312 ("already linked"));
313
314 in_pcbofflist(inp);
315 inp->inp_pcbinfo = NULL;
316 }
317
318 void
in_pcbunlink(struct inpcb * inp,struct inpcbinfo * pcbinfo)319 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
320 {
321 in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD);
322 }
323
324 /*
325 * Relink a pcb into a new pcbinfo.
326 */
327 void
in_pcblink_flags(struct inpcb * inp,struct inpcbinfo * pcbinfo,int flags)328 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
329 {
330 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
331 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
332 ("already linked"));
333
334 inp->inp_pcbinfo = pcbinfo;
335 in_pcbonlist(inp);
336 }
337
338 void
in_pcblink(struct inpcb * inp,struct inpcbinfo * pcbinfo)339 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
340 {
341 return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD);
342 }
343
344 static boolean_t
in_pcbporthash_update(struct inpcbportinfo * portinfo,struct inpcb * inp,u_short lport,struct ucred * cred,int wild)345 in_pcbporthash_update(struct inpcbportinfo *portinfo,
346 struct inpcb *inp, u_short lport, struct ucred *cred, int wild)
347 {
348 struct inpcbporthead *porthash;
349
350 /*
351 * This has to be atomic. If the porthash is shared across multiple
352 * protocol threads, e.g. tcp and udp, then the token must be held.
353 */
354 porthash = in_pcbporthash_head(portinfo, lport);
355 GET_PORTHASH_TOKEN(porthash);
356
357 if (in_pcblookup_local(porthash, inp->inp_laddr, lport, wild, cred)) {
358 REL_PORTHASH_TOKEN(porthash);
359 return FALSE;
360 }
361 inp->inp_lport = lport;
362 in_pcbinsporthash(porthash, inp);
363
364 REL_PORTHASH_TOKEN(porthash);
365 return TRUE;
366 }
367
368 static int
in_pcbsetlport(struct inpcb * inp,int wild,struct ucred * cred)369 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
370 {
371 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
372 struct inpcbportinfo *portinfo;
373 u_short first, last, lport, step, first0, last0;
374 int count, error;
375 int portinfo_first, portinfo_idx;
376 uint32_t cut;
377
378 /*
379 * We force matches against wildcard ports in order to
380 * avoid auto-assigning lport to such ports, which would
381 * cause problems for same-machine connect()s.
382 */
383 wild = INPLOOKUP_WILDCARD;
384
385 inp->inp_flags |= INP_ANONPORT;
386
387 step = pcbinfo->portinfo_cnt;
388 portinfo_first = mycpuid % pcbinfo->portinfo_cnt;
389 portinfo_idx = portinfo_first;
390
391 if (inp->inp_flags & INP_HIGHPORT) {
392 first0 = ipport_hifirstauto; /* sysctl */
393 last0 = ipport_hilastauto;
394 } else if (inp->inp_flags & INP_LOWPORT) {
395 if (cred &&
396 (error = caps_priv_check(cred, SYSCAP_NONET_RESPORT)))
397 {
398 inp->inp_laddr.s_addr = INADDR_ANY;
399 return error;
400 }
401 first0 = ipport_lowfirstauto; /* 1023 */
402 last0 = ipport_lowlastauto; /* 600 */
403 } else {
404 first0 = ipport_firstauto; /* sysctl */
405 last0 = ipport_lastauto;
406 }
407 if (first0 > last0) {
408 lport = last0;
409 last0 = first0;
410 first0 = lport;
411 }
412 KKASSERT(last0 >= first0);
413
414 cut = karc4random();
415 loop:
416 portinfo = &pcbinfo->portinfo[portinfo_idx];
417 first = first0;
418 last = last0;
419
420 /*
421 * Simple check to ensure all ports are not used up causing
422 * a deadlock here.
423 */
424 in_pcbportrange(&last, &first, portinfo->offset, step);
425 lport = last - first;
426 count = lport / step;
427
428 lport = rounddown(cut % lport, step) + first;
429 KKASSERT(lport % step == portinfo->offset);
430
431 for (;;) {
432 if (count-- < 0) { /* completely used? */
433 error = EADDRNOTAVAIL;
434 break;
435 }
436
437 if (__predict_false(lport < first || lport > last)) {
438 lport = first;
439 KKASSERT(lport % step == portinfo->offset);
440 }
441
442 if (in_pcbporthash_update(portinfo, inp, htons(lport),
443 cred, wild)) {
444 error = 0;
445 break;
446 }
447
448 lport += step;
449 KKASSERT(lport % step == portinfo->offset);
450 }
451
452 if (error) {
453 /* Try next portinfo */
454 portinfo_idx++;
455 portinfo_idx %= pcbinfo->portinfo_cnt;
456 if (portinfo_idx != portinfo_first)
457 goto loop;
458 inp->inp_laddr.s_addr = INADDR_ANY;
459 }
460 return error;
461 }
462
463 static __inline struct inpcbporthead *
OBTAIN_LPORTHASH_TOKEN(struct inpcbinfo * pcbinfo,u_short lport)464 OBTAIN_LPORTHASH_TOKEN(struct inpcbinfo *pcbinfo, u_short lport)
465 {
466 struct inpcbportinfo *portinfo;
467 struct inpcbporthead *porthash;
468 u_short lport_ho = ntohs(lport);
469
470 /*
471 * Locate the proper portinfo based on lport
472 */
473 portinfo = &pcbinfo->portinfo[lport_ho % pcbinfo->portinfo_cnt];
474 KKASSERT((lport_ho % pcbinfo->portinfo_cnt) == portinfo->offset);
475
476 porthash = in_pcbporthash_head(portinfo, lport);
477 GET_PORTHASH_TOKEN(porthash);
478
479 return porthash;
480 }
481
482 static int
in_pcbbind_laddr(struct sockaddr_in * sin,struct in_addr * laddr,struct thread * td)483 in_pcbbind_laddr(struct sockaddr_in *sin, struct in_addr *laddr,
484 struct thread *td)
485 {
486 struct sockaddr_in jsin;
487
488 if (!prison_replace_wildcards(td, (struct sockaddr *)sin))
489 return (EINVAL);
490
491 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
492 sin->sin_addr.s_addr != INADDR_ANY) {
493 sin->sin_port = 0; /* yech... */
494 bzero(&sin->sin_zero, sizeof sin->sin_zero);
495 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
496 return (EADDRNOTAVAIL);
497 }
498
499 *laddr = sin->sin_addr;
500
501 jsin.sin_family = AF_INET;
502 jsin.sin_addr.s_addr = laddr->s_addr;
503 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
504 laddr->s_addr = INADDR_ANY;
505 return (EINVAL);
506 }
507 laddr->s_addr = jsin.sin_addr.s_addr;
508
509 return (0);
510 }
511
512 static int
in_pcbbind_laddrport_check(const struct socket * so,struct sockaddr_in * sin,struct inpcbporthead * porthash,int wild,struct ucred * cred,struct thread * td)513 in_pcbbind_laddrport_check(const struct socket *so, struct sockaddr_in *sin,
514 struct inpcbporthead *porthash, int wild, struct ucred *cred,
515 struct thread *td)
516 {
517 int reuseport = (so->so_options & SO_REUSEPORT);
518 struct inpcb *t;
519
520 ASSERT_PORTHASH_TOKEN_HELD(porthash);
521
522 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
523 /*
524 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
525 * allow complete duplication of binding if
526 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
527 * and a multicast address is bound on both
528 * new and duplicated sockets.
529 */
530 if (so->so_options & SO_REUSEADDR)
531 reuseport = SO_REUSEADDR | SO_REUSEPORT;
532 }
533
534 if (so->so_cred->cr_uid != 0 &&
535 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
536 t = in_pcblookup_local(porthash, sin->sin_addr, sin->sin_port,
537 INPLOOKUP_WILDCARD, cred);
538 if (t && t->inp_socket != so &&
539 (so->so_cred->cr_uid != t->inp_socket->so_cred->cr_uid))
540 return (EADDRINUSE);
541 }
542 if (cred && !prison_replace_wildcards(td, (struct sockaddr *)sin))
543 return (EADDRNOTAVAIL);
544
545 /*
546 * When binding to a local port if the best match is against
547 * an accepted socket we generally want to allow the binding.
548 * This means that there is no longer any specific socket
549 * bound or bound for listening.
550 */
551 t = in_pcblookup_local(porthash, sin->sin_addr, sin->sin_port,
552 wild, cred);
553 if (t && t->inp_socket != so &&
554 (reuseport & t->inp_socket->so_options) == 0 &&
555 (t->inp_socket->so_state & SS_ACCEPTMECH) == 0)
556 return (EADDRINUSE);
557
558 return (0);
559 }
560
561 int
in_pcbsrcaddr_check(const struct inpcb * inp,struct sockaddr_in * sin,struct in_addr * laddr,struct thread * td)562 in_pcbsrcaddr_check(const struct inpcb *inp, struct sockaddr_in *sin,
563 struct in_addr *laddr, struct thread *td)
564 {
565 const struct socket *so = inp->inp_socket;
566 struct inpcbporthead *porthash;
567 struct ucred *cred = NULL;
568 int wild = 0;
569 int error;
570
571 /* inp must be bound beforehand. */
572 KKASSERT(inp->inp_lport != 0);
573 KKASSERT(sin->sin_len == sizeof(*sin));
574
575 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
576 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
577 if (td->td_proc)
578 cred = td->td_proc->p_ucred;
579
580 /* Always use inp_lport */
581 sin->sin_port = inp->inp_lport;
582
583 error = in_pcbbind_laddr(sin, laddr, td);
584 if (error)
585 return (error);
586
587 if (IN_MULTICAST(ntohl(laddr->s_addr))) {
588 /* Unlike bind, multicast src address is not allowed. */
589 return (EINVAL);
590 }
591
592 if (inp->inp_laddr.s_addr == laddr->s_addr) {
593 /*
594 * src address is same as what we bound to.
595 *
596 * inp_laddr == INADDR_ANY && srcaddr == INADDR_ANY
597 * is allowed, which does not really matter.
598 */
599 return (0);
600 } else if (inp->inp_laddr.s_addr != INADDR_ANY &&
601 !IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) {
602 /* Already bound to a specific address */
603 return (EINVAL);
604 }
605
606 /*
607 * This has to be atomic. If the porthash is shared across
608 * multiple protocol threads, e.g. tcp and udp then the token
609 * must be held.
610 */
611 porthash = OBTAIN_LPORTHASH_TOKEN(inp->inp_pcbinfo, inp->inp_lport);
612
613 /*
614 * Restore the sin_port whacked by in_pcbbind_ladddr();
615 * sin->sin_port is checked by in_pcbbind_laddrport_check().
616 */
617 sin->sin_port = inp->inp_lport;
618
619 error = in_pcbbind_laddrport_check(so, sin, porthash, wild, cred, td);
620 if (error)
621 laddr->s_addr = INADDR_ANY;
622
623 REL_PORTHASH_TOKEN(porthash);
624 return (error);
625 }
626
627 int
in_pcbbind(struct inpcb * inp,struct sockaddr * nam,struct thread * td)628 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
629 {
630 const struct socket *so = inp->inp_socket;
631 struct ucred *cred = NULL;
632 int wild = 0;
633
634 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
635 return (EINVAL); /* already bound */
636
637 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
638 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
639 if (td->td_proc)
640 cred = td->td_proc->p_ucred;
641
642 if (nam != NULL) {
643 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
644 struct inpcbporthead *porthash;
645 u_short lport;
646 int error;
647
648 if (nam->sa_len != sizeof *sin)
649 return (EINVAL);
650 #ifdef notdef
651 /*
652 * We should check the family, but old programs
653 * incorrectly fail to initialize it.
654 */
655 if (sin->sin_family != AF_INET)
656 return (EAFNOSUPPORT);
657 #endif
658
659 /*
660 * Save sin_port for later use, since it will
661 * be whacked by in_pcbbind_laddr().
662 */
663 lport = sin->sin_port;
664
665 error = in_pcbbind_laddr(sin, &inp->inp_laddr, td);
666 if (error)
667 return (error);
668
669 if (lport == 0) {
670 /* Auto-select local port */
671 return in_pcbsetlport(inp, wild, cred);
672 }
673
674 /* GROSS */
675 if (ntohs(lport) < IPPORT_RESERVED && cred &&
676 (error = caps_priv_check(cred, SYSCAP_NONET_RESPORT)))
677 {
678 inp->inp_laddr.s_addr = INADDR_ANY;
679 return (error);
680 }
681
682 /*
683 * This has to be atomic. If the porthash is shared across
684 * multiple protocol threads, e.g. tcp and udp then the token
685 * must be held.
686 */
687 porthash = OBTAIN_LPORTHASH_TOKEN(inp->inp_pcbinfo, lport);
688
689 /*
690 * Restore the sin_port whacked by in_pcbbind_ladddr();
691 * sin->sin_port is checked by in_pcbbind_laddrport_check().
692 */
693 sin->sin_port = lport;
694
695 error = in_pcbbind_laddrport_check(so, sin, porthash,
696 wild, cred, td);
697 if (error) {
698 inp->inp_laddr.s_addr = INADDR_ANY;
699 goto done;
700 }
701
702 inp->inp_lport = lport;
703 in_pcbinsporthash(porthash, inp);
704 error = 0;
705 done:
706 REL_PORTHASH_TOKEN(porthash);
707 return (error);
708 } else {
709 struct sockaddr_in jsin;
710
711 jsin.sin_family = AF_INET;
712 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
713 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
714 inp->inp_laddr.s_addr = INADDR_ANY;
715 return (EINVAL);
716 }
717 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
718
719 return in_pcbsetlport(inp, wild, cred);
720 }
721 }
722
723 /*
724 * Lookup a PCB based on the local and remote address and port.
725 *
726 * This function is only used when scanning for a free port.
727 */
728 static struct inpcb *
in_pcblookup_localremote(struct inpcbporthead * porthash,struct in_addr laddr,u_short lport,struct in_addr faddr,u_short fport,struct ucred * cred)729 in_pcblookup_localremote(struct inpcbporthead *porthash, struct in_addr laddr,
730 u_short lport, struct in_addr faddr, u_short fport,
731 struct ucred *cred)
732 {
733 struct inpcb *inp;
734 struct inpcbport *phd;
735 struct inpcb *match = NULL;
736 struct prison *pscan;
737 struct prison *pr;
738
739 /*
740 * If the porthashbase is shared across several cpus, it must
741 * have been locked.
742 */
743 ASSERT_PORTHASH_TOKEN_HELD(porthash);
744
745 /*
746 * Best fit PCB lookup.
747 *
748 * First see if this local port is in use by looking on the
749 * port hash list.
750 */
751 LIST_FOREACH(phd, porthash, phd_hash) {
752 if (phd->phd_port == lport)
753 break;
754 }
755 if (phd != NULL) {
756 pr = cred ? cred->cr_prison : NULL;
757
758 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
759 #ifdef INET6
760 if (!INP_ISIPV4(inp))
761 continue;
762 #endif
763 if (inp->inp_laddr.s_addr == INADDR_ANY) {
764 if (inp->inp_socket && inp->inp_socket->so_cred)
765 pscan = inp->inp_socket->so_cred->cr_prison;
766 else
767 pscan = NULL;
768 if (pr != pscan)
769 continue;
770 } else {
771 if (inp->inp_laddr.s_addr != laddr.s_addr)
772 continue;
773 }
774
775 if (inp->inp_faddr.s_addr != INADDR_ANY &&
776 inp->inp_faddr.s_addr != faddr.s_addr)
777 continue;
778
779 if (inp->inp_fport != 0 && inp->inp_fport != fport)
780 continue;
781
782 match = inp;
783 break;
784 }
785 }
786 return (match);
787 }
788
789 static boolean_t
in_pcbporthash_update4(struct inpcbportinfo * portinfo,struct inpcb * inp,u_short lport,const struct sockaddr_in * sin,struct ucred * cred)790 in_pcbporthash_update4(struct inpcbportinfo *portinfo, struct inpcb *inp,
791 u_short lport, const struct sockaddr_in *sin,
792 struct ucred *cred)
793 {
794 struct inpcbporthead *porthash;
795
796 /*
797 * This has to be atomic. If the porthash is shared across multiple
798 * protocol threads, e.g. tcp and udp, then the token must be held.
799 */
800 porthash = in_pcbporthash_head(portinfo, lport);
801 GET_PORTHASH_TOKEN(porthash);
802
803 if (in_pcblookup_localremote(porthash, inp->inp_laddr, lport,
804 sin->sin_addr, sin->sin_port, cred)) {
805 REL_PORTHASH_TOKEN(porthash);
806 return FALSE;
807 }
808 inp->inp_lport = lport;
809 in_pcbinsporthash(porthash, inp);
810
811 REL_PORTHASH_TOKEN(porthash);
812 return TRUE;
813 }
814
815 int
in_pcbbind_remote(struct inpcb * inp,const struct sockaddr * remote,struct thread * td)816 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote,
817 struct thread *td)
818 {
819 struct proc *p = td->td_proc;
820 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
821 struct sockaddr_in jsin;
822 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
823 struct ucred *cred = NULL;
824 u_short first, last, lport;
825 int count, hash_count;
826 int error, selfconn = 0;
827 int cpuid = mycpuid;
828 uint32_t hash_base = 0, hash;
829
830 ASSERT_NETISR_NCPUS(cpuid);
831
832 if (TAILQ_EMPTY(&in_ifaddrheads[cpuid])) /* XXX broken! */
833 return (EADDRNOTAVAIL);
834
835 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
836 if (inp->inp_lport != 0)
837 return (EINVAL); /* already bound */
838
839 KKASSERT(p);
840 cred = p->p_ucred;
841
842 jsin.sin_family = AF_INET;
843 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
844 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
845 inp->inp_laddr.s_addr = INADDR_ANY;
846 return (EINVAL);
847 }
848 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
849
850 hash_count = ip_porthash_trycount;
851 if (hash_count > 0) {
852 hash_base = toeplitz_piecemeal_addr(sin->sin_addr.s_addr) ^
853 toeplitz_piecemeal_addr(inp->inp_laddr.s_addr) ^
854 toeplitz_piecemeal_port(sin->sin_port);
855 } else {
856 hash_count = 0;
857 }
858
859 inp->inp_flags |= INP_ANONPORT;
860
861 if (inp->inp_flags & INP_HIGHPORT) {
862 first = ipport_hifirstauto; /* sysctl */
863 last = ipport_hilastauto;
864 } else if (inp->inp_flags & INP_LOWPORT) {
865 if (cred &&
866 (error = caps_priv_check(cred, SYSCAP_NONET_RESPORT)))
867 {
868 inp->inp_laddr.s_addr = INADDR_ANY;
869 return (error);
870 }
871 first = ipport_lowfirstauto; /* 1023 */
872 last = ipport_lowlastauto; /* 600 */
873 } else {
874 first = ipport_firstauto; /* sysctl */
875 last = ipport_lastauto;
876 }
877 if (first > last) {
878 lport = last;
879 last = first;
880 first = lport;
881 }
882 KKASSERT(last >= first);
883
884 count = last - first;
885 lport = (karc4random() % count) + first;
886 count += hash_count;
887
888 /*
889 * Simple check to ensure all ports are not used up causing
890 * a deadlock here.
891 */
892 for (;;) {
893 u_short lport_no;
894
895 if (count-- < 0) { /* completely used? */
896 error = EADDRNOTAVAIL;
897 break;
898 }
899
900 if (__predict_false(lport < first || lport > last))
901 lport = first;
902 lport_no = htons(lport);
903
904 /* This could happen on loopback interface */
905 if (__predict_false(sin->sin_port == lport_no &&
906 sin->sin_addr.s_addr == inp->inp_laddr.s_addr)) {
907 if (!selfconn) {
908 ++count; /* don't count this try */
909 selfconn = 1;
910 }
911 goto next;
912 }
913
914 if (hash_count) {
915 --hash_count;
916 hash = hash_base ^
917 toeplitz_piecemeal_port(lport_no);
918 if (netisr_hashcpu(hash) != cpuid && hash_count)
919 goto next;
920 }
921
922 if (in_pcbporthash_update4(
923 &pcbinfo->portinfo[lport % pcbinfo->portinfo_cnt],
924 inp, lport_no, sin, cred)) {
925 error = 0;
926 break;
927 }
928 next:
929 ++lport;
930 }
931
932 if (error)
933 inp->inp_laddr.s_addr = INADDR_ANY;
934 return (error);
935 }
936
937 /*
938 * Figure out the local interface address to pair against the requested
939 * target address, as well as validate the target address.
940 */
941 int
in_pcbladdr_find(struct inpcb * inp,struct sockaddr * nam,struct sockaddr_in ** plocal_sin,struct thread * td,int find)942 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
943 struct sockaddr_in **plocal_sin, struct thread *td, int find)
944 {
945 struct in_ifaddr_container *iac;
946 struct in_ifaddr *ia;
947 struct ucred *cred = NULL;
948 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
949 struct sockaddr *jsin;
950 struct prison *pr;
951 struct route *ro;
952 int alloc_route = 0;
953
954 if (nam->sa_len != sizeof *sin)
955 return (EINVAL);
956 if (sin->sin_family != AF_INET)
957 return (EAFNOSUPPORT);
958 if (sin->sin_port == 0)
959 return (EADDRNOTAVAIL);
960
961 /*
962 * Are we in a jail?
963 */
964 pr = NULL;
965 if (td && td->td_proc && td->td_proc->p_ucred)
966 cred = td->td_proc->p_ucred;
967 if (cred)
968 pr = cred->cr_prison;
969
970 /*
971 * If the destination address is INADDR_ANY then use the primary
972 * local address.
973 *
974 * If the supplied address is INADDR_BROADCAST, and the primary
975 * interface supports broadcast, choose the broadcast address for
976 * that interface.
977 *
978 * If jailed, locate an interface address acceptable to the jail.
979 */
980 if (sin->sin_addr.s_addr == INADDR_ANY) {
981 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) {
982 ia = iac->ia;
983 if (pr == NULL ||
984 jailed_ip(pr, sintosa(&ia->ia_addr))) {
985 sin->sin_addr = IA_SIN(ia)->sin_addr;
986 break;
987 }
988 }
989 } else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST) {
990 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) {
991 ia = iac->ia;
992 if ((pr == NULL ||
993 jailed_ip(pr, sintosa(&ia->ia_addr))) &&
994 (iac->ia->ia_ifp->if_flags & IFF_BROADCAST)) {
995 sin->sin_addr =
996 satosin(&ia->ia_broadaddr)->sin_addr;
997 break;
998 }
999 }
1000 }
1001
1002 /*
1003 * If asked to do a search, use the cached route or do a route table
1004 * lookup to try to find an acceptable local interface IP.
1005 */
1006 if (find == 0)
1007 return 0;
1008
1009 ia = NULL;
1010
1011 /*
1012 * If we have a cached route, check to see if it is acceptable.
1013 * If not, free it.
1014 */
1015 ro = &inp->inp_route;
1016 if (ro->ro_rt &&
1017 (!(ro->ro_rt->rt_flags & RTF_UP) ||
1018 ro->ro_dst.sa_family != AF_INET ||
1019 satosin(&ro->ro_dst)->sin_addr.s_addr !=
1020 sin->sin_addr.s_addr ||
1021 inp->inp_socket->so_options & SO_DONTROUTE)) {
1022 RTFREE(ro->ro_rt);
1023 ro->ro_rt = NULL;
1024 }
1025
1026 /*
1027 * If we do not have a route, construct one and do a lookup,
1028 * unless we are forbidden to do so.
1029 *
1030 * Note that we should check the address family of the cached
1031 * destination, in case of sharing the cache with IPv6.
1032 */
1033 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1034 (ro->ro_rt == NULL || ro->ro_rt->rt_ifp == NULL)) {
1035 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1036 ro->ro_dst.sa_family = AF_INET;
1037 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1038 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sin->sin_addr;
1039 rtalloc(ro);
1040 alloc_route = 1;
1041 }
1042
1043 /*
1044 * If we found a route, use the address corresponding to the
1045 * outgoing interface.
1046 *
1047 * If jailed, try to find a compatible address on the outgoing
1048 * interface.
1049 */
1050 if (ro->ro_rt) {
1051 ia = ifatoia(ro->ro_rt->rt_ifa);
1052 if (pr == NULL)
1053 goto skip;
1054 if (jailed_ip(pr, sintosa(&ia->ia_addr)))
1055 goto skip;
1056 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) {
1057 if (iac->ia->ia_ifp != ia->ia_ifp)
1058 continue;
1059 ia = iac->ia;
1060 if (jailed_ip(pr, sintosa(&ia->ia_addr)))
1061 goto skip;
1062 }
1063 ia = NULL;
1064 }
1065 skip:
1066
1067 /*
1068 * If the route didn't work or there was no route,
1069 * fall-back to the first address in in_ifaddrheads[].
1070 *
1071 * If jailed and this address is not available for
1072 * the jail, leave ia set to NULL.
1073 */
1074 if (ia == NULL) {
1075 u_short fport = sin->sin_port;
1076
1077 sin->sin_port = 0;
1078 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
1079 if (ia && pr && !jailed_ip(pr, sintosa(&ia->ia_addr)))
1080 ia = NULL;
1081
1082 if (ia == NULL)
1083 ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
1084 if (ia && pr && !jailed_ip(pr, sintosa(&ia->ia_addr)))
1085 ia = NULL;
1086
1087 sin->sin_port = fport;
1088 if (ia == NULL && !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1089 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
1090
1091 if (ia && pr && !jailed_ip(pr, sintosa(&ia->ia_addr)))
1092 ia = NULL;
1093
1094 if (pr == NULL && ia == NULL)
1095 goto fail;
1096 }
1097
1098 /*
1099 * If the destination address is multicast and an outgoing
1100 * interface has been set as a multicast option, use the
1101 * address of that interface as our source address.
1102 */
1103 if (pr == NULL && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
1104 inp->inp_moptions != NULL) {
1105 struct ip_moptions *imo;
1106 struct ifnet *ifp;
1107
1108 imo = inp->inp_moptions;
1109 if ((ifp = imo->imo_multicast_ifp) != NULL) {
1110 struct in_ifaddr_container *iac;
1111
1112 ia = NULL;
1113 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) {
1114 if (iac->ia->ia_ifp == ifp) {
1115 ia = iac->ia;
1116 break;
1117 }
1118 }
1119 if (ia == NULL)
1120 goto fail;
1121 }
1122 }
1123
1124 /*
1125 * If we still don't have a local address, and are jailed,
1126 * use the jail's first non-localhost IP. If there isn't
1127 * one, use the jail's first localhost IP.
1128 *
1129 * Don't do pcblookup call here; return interface in plocal_sin
1130 * and exit to caller, that will do the lookup.
1131 */
1132 if (ia == NULL && pr) {
1133 jsin = prison_get_nonlocal(cred->cr_prison, AF_INET, NULL);
1134 if (jsin == NULL)
1135 jsin = prison_get_local(cred->cr_prison, AF_INET, NULL);
1136 if (jsin)
1137 *plocal_sin = satosin(jsin);
1138 else
1139 goto fail;
1140 } else if (ia) {
1141 *plocal_sin = &ia->ia_addr;
1142 } else {
1143 goto fail;
1144 }
1145 return (0);
1146 fail:
1147 if (alloc_route)
1148 in_pcbresetroute(inp);
1149 return (EADDRNOTAVAIL);
1150 }
1151
1152 int
in_pcbladdr(struct inpcb * inp,struct sockaddr * nam,struct sockaddr_in ** plocal_sin,struct thread * td)1153 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
1154 struct sockaddr_in **plocal_sin, struct thread *td)
1155 {
1156 return in_pcbladdr_find(inp, nam, plocal_sin, td,
1157 (inp->inp_laddr.s_addr == INADDR_ANY));
1158 }
1159
1160 /*
1161 * Outer subroutine:
1162 * Connect from a socket to a specified address.
1163 * Both address and port must be specified in argument sin.
1164 * If don't have a local address for this socket yet,
1165 * then pick one.
1166 */
1167 int
in_pcbconnect(struct inpcb * inp,struct sockaddr * nam,struct thread * td)1168 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
1169 {
1170 struct sockaddr_in *if_sin;
1171 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1172 int error;
1173
1174 if_sin = NULL; /* avoid gcc warnings */
1175
1176 /* Call inner routine to assign local interface address. */
1177 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
1178 return (error);
1179
1180 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port,
1181 inp->inp_laddr.s_addr ?
1182 inp->inp_laddr : if_sin->sin_addr,
1183 inp->inp_lport, FALSE, NULL) != NULL) {
1184 return (EADDRINUSE);
1185 }
1186 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1187 if (inp->inp_lport == 0) {
1188 error = in_pcbbind(inp, NULL, td);
1189 if (error)
1190 return (error);
1191 }
1192 inp->inp_laddr = if_sin->sin_addr;
1193 }
1194 inp->inp_faddr = sin->sin_addr;
1195 inp->inp_fport = sin->sin_port;
1196 in_pcbinsconnhash(inp);
1197 return (0);
1198 }
1199
1200 void
in_pcbdisconnect(struct inpcb * inp)1201 in_pcbdisconnect(struct inpcb *inp)
1202 {
1203
1204 in_pcbremconnhash(inp);
1205 inp->inp_faddr.s_addr = INADDR_ANY;
1206 inp->inp_fport = 0;
1207 }
1208
1209 void
in_pcbdetach(struct inpcb * inp)1210 in_pcbdetach(struct inpcb *inp)
1211 {
1212 struct socket *so = inp->inp_socket;
1213 struct inpcbinfo *ipi = inp->inp_pcbinfo;
1214
1215 inp->inp_gencnt = ++ipi->ipi_gencnt;
1216 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
1217 in_pcbremlists(inp);
1218 so->so_pcb = NULL;
1219 sofree(so); /* remove pcb ref */
1220 if (inp->inp_options)
1221 m_free(inp->inp_options);
1222 if (inp->inp_route.ro_rt)
1223 rtfree(inp->inp_route.ro_rt);
1224 ip_freemoptions(inp->inp_moptions);
1225 kfree(inp, M_PCB);
1226 }
1227
1228 /*
1229 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP
1230 * socket received RST.
1231 */
1232 static int
in_setsockaddr(struct socket * so,struct sockaddr ** nam)1233 in_setsockaddr(struct socket *so, struct sockaddr **nam)
1234 {
1235 struct inpcb *inp;
1236 struct sockaddr_in *sin;
1237
1238 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
1239 inp = so->so_pcb;
1240 if (!inp)
1241 return (ECONNRESET);
1242
1243 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1244 sin->sin_family = AF_INET;
1245 sin->sin_len = sizeof *sin;
1246 sin->sin_port = inp->inp_lport;
1247 sin->sin_addr = inp->inp_laddr;
1248
1249 *nam = (struct sockaddr *)sin;
1250 return (0);
1251 }
1252
1253 void
in_setsockaddr_dispatch(netmsg_t msg)1254 in_setsockaddr_dispatch(netmsg_t msg)
1255 {
1256 int error;
1257
1258 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1259 lwkt_replymsg(&msg->lmsg, error);
1260 }
1261
1262 /*
1263 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP
1264 * socket received RST.
1265 */
1266 int
in_setpeeraddr(struct socket * so,struct sockaddr ** nam)1267 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1268 {
1269 struct inpcb *inp;
1270 struct sockaddr_in *sin;
1271
1272 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
1273 inp = so->so_pcb;
1274 if (!inp)
1275 return (ECONNRESET);
1276
1277 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1278 sin->sin_family = AF_INET;
1279 sin->sin_len = sizeof *sin;
1280 sin->sin_port = inp->inp_fport;
1281 sin->sin_addr = inp->inp_faddr;
1282
1283 *nam = (struct sockaddr *)sin;
1284 return (0);
1285 }
1286
1287 void
in_setpeeraddr_dispatch(netmsg_t msg)1288 in_setpeeraddr_dispatch(netmsg_t msg)
1289 {
1290 int error;
1291
1292 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1293 lwkt_replymsg(&msg->lmsg, error);
1294 }
1295
1296 void
in_pcbnotifyall(struct inpcbinfo * pcbinfo,struct in_addr faddr,int err,inp_notify_t notify)1297 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err,
1298 inp_notify_t notify)
1299 {
1300 struct inpcb *inp, *marker;
1301
1302 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1303 ("not in the correct netisr"));
1304 marker = in_pcbmarker();
1305
1306 /*
1307 * NOTE:
1308 * - If INP_PLACEMARKER is set we must ignore the rest of the
1309 * structure and skip it.
1310 * - It is safe to nuke inpcbs here, since we are in their own
1311 * netisr.
1312 */
1313 GET_PCBINFO_TOKEN(pcbinfo);
1314
1315 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1316 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1317 LIST_REMOVE(marker, inp_list);
1318 LIST_INSERT_AFTER(inp, marker, inp_list);
1319
1320 if (inp->inp_flags & INP_PLACEMARKER)
1321 continue;
1322 #ifdef INET6
1323 if (!INP_ISIPV4(inp))
1324 continue;
1325 #endif
1326 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1327 inp->inp_socket == NULL)
1328 continue;
1329 (*notify)(inp, err); /* can remove inp from list! */
1330 }
1331 LIST_REMOVE(marker, inp_list);
1332
1333 REL_PCBINFO_TOKEN(pcbinfo);
1334 }
1335
1336 void
in_pcbpurgeif0(struct inpcbinfo * pcbinfo,struct ifnet * ifp)1337 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1338 {
1339 struct inpcb *inp, *marker;
1340
1341 /*
1342 * We only need to make sure that we are in netisr0, where all
1343 * multicast operation happen. We could check inpcbinfo which
1344 * does not belong to netisr0 by holding the inpcbinfo's token.
1345 * In this case, the pcbinfo must be able to be shared, i.e.
1346 * pcbinfo->infotoken is not NULL.
1347 */
1348 ASSERT_NETISR0;
1349 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL,
1350 ("pcbinfo could not be shared"));
1351
1352 /*
1353 * Get a marker for the current netisr (netisr0).
1354 *
1355 * It is possible that the multicast address deletion blocks,
1356 * which could cause temporary token releasing. So we use
1357 * inpcb marker here to get a coherent view of the inpcb list.
1358 *
1359 * While, on the other hand, moptions are only added and deleted
1360 * in netisr0, so we would not see staled moption or miss moption
1361 * even if the token was released due to the blocking multicast
1362 * address deletion.
1363 */
1364 marker = in_pcbmarker();
1365
1366 GET_PCBINFO_TOKEN(pcbinfo);
1367
1368 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1369 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1370 struct ip_moptions *imo;
1371
1372 LIST_REMOVE(marker, inp_list);
1373 LIST_INSERT_AFTER(inp, marker, inp_list);
1374
1375 if (inp->inp_flags & INP_PLACEMARKER)
1376 continue;
1377 imo = inp->inp_moptions;
1378 if (INP_ISIPV4(inp) && imo != NULL) {
1379 int i, gap;
1380
1381 /*
1382 * Unselect the outgoing interface if it is being
1383 * detached.
1384 */
1385 if (imo->imo_multicast_ifp == ifp)
1386 imo->imo_multicast_ifp = NULL;
1387
1388 /*
1389 * Drop multicast group membership if we joined
1390 * through the interface being detached.
1391 */
1392 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1393 i++) {
1394 if (imo->imo_membership[i]->inm_ifp == ifp) {
1395 /*
1396 * NOTE:
1397 * This could block and the pcbinfo
1398 * token could be passively released.
1399 */
1400 in_delmulti(imo->imo_membership[i]);
1401 gap++;
1402 } else if (gap != 0)
1403 imo->imo_membership[i - gap] =
1404 imo->imo_membership[i];
1405 }
1406 imo->imo_num_memberships -= gap;
1407 }
1408 }
1409 LIST_REMOVE(marker, inp_list);
1410
1411 REL_PCBINFO_TOKEN(pcbinfo);
1412 }
1413
1414 /*
1415 * Check for alternatives when higher level complains
1416 * about service problems. For now, invalidate cached
1417 * routing information. If the route was created dynamically
1418 * (by a redirect), time to try a default gateway again.
1419 */
1420 void
in_losing(struct inpcb * inp)1421 in_losing(struct inpcb *inp)
1422 {
1423 struct rtentry *rt;
1424 struct rt_addrinfo rtinfo;
1425
1426 if ((rt = inp->inp_route.ro_rt)) {
1427 bzero(&rtinfo, sizeof(struct rt_addrinfo));
1428 rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1429 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1430 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1431 rtinfo.rti_flags = rt->rt_flags;
1432 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1433 if (rt->rt_flags & RTF_DYNAMIC) {
1434 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1435 rt_mask(rt), rt->rt_flags, NULL);
1436 }
1437 inp->inp_route.ro_rt = NULL;
1438 rtfree(rt);
1439 /*
1440 * A new route can be allocated
1441 * the next time output is attempted.
1442 */
1443 }
1444 }
1445
1446 /*
1447 * After a routing change, flush old routing
1448 * and allocate a (hopefully) better one.
1449 */
1450 void
in_rtchange(struct inpcb * inp,int err)1451 in_rtchange(struct inpcb *inp, int err)
1452 {
1453 if (inp->inp_route.ro_rt) {
1454 rtfree(inp->inp_route.ro_rt);
1455 inp->inp_route.ro_rt = NULL;
1456 /*
1457 * A new route can be allocated the next time
1458 * output is attempted.
1459 */
1460 }
1461 }
1462
1463 /*
1464 * Lookup a PCB based on the local address and port.
1465 *
1466 * This function is only used when scanning for a free port.
1467 */
1468 static struct inpcb *
in_pcblookup_local(struct inpcbporthead * porthash,struct in_addr laddr,u_int lport_arg,int wild_okay,struct ucred * cred)1469 in_pcblookup_local(struct inpcbporthead *porthash, struct in_addr laddr,
1470 u_int lport_arg, int wild_okay, struct ucred *cred)
1471 {
1472 struct prison *pscan;
1473 struct prison *pr;
1474 struct inpcb *inp;
1475 int matchwild = 3, wildcard;
1476 u_short lport = lport_arg;
1477 struct inpcbport *phd;
1478 struct inpcb *match = NULL;
1479
1480 /*
1481 * If the porthashbase is shared across several cpus, it must
1482 * have been locked.
1483 */
1484 ASSERT_PORTHASH_TOKEN_HELD(porthash);
1485
1486 /*
1487 * Best fit PCB lookup.
1488 *
1489 * First see if this local port is in use by looking on the
1490 * port hash list.
1491 */
1492 LIST_FOREACH(phd, porthash, phd_hash) {
1493 if (phd->phd_port == lport)
1494 break;
1495 }
1496 if (phd != NULL) {
1497 pr = cred ? cred->cr_prison : NULL;
1498
1499 /*
1500 * Port is in use by one or more PCBs. Look for best
1501 * fit.
1502 *
1503 * If in a prison we may wish to allow the jail to override
1504 * a wildcard listen on the host. Since the jail forces its
1505 * own wildcard listens to a specific set of jail IPs, this
1506 * override allows most services on the host to remain as
1507 * they were and still be 'jail friendly'.
1508 */
1509 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1510 wildcard = 0;
1511 #ifdef INET6
1512 if (!INP_ISIPV4(inp))
1513 continue;
1514 #endif
1515 if (inp->inp_faddr.s_addr != INADDR_ANY)
1516 wildcard++;
1517
1518 /*
1519 * Prison are independent of each other in terms
1520 * of allowing bindings. This can result in multiple
1521 * overloaded bindings which in_pcblookup_pkthash()
1522 * will have to sort out.
1523 *
1524 * Allow wildcarded entries to co-exist with specific
1525 * entries. Specific entries override wildcarded
1526 * entries.
1527 */
1528 if (inp->inp_socket && inp->inp_socket->so_cred)
1529 pscan = inp->inp_socket->so_cred->cr_prison;
1530 else
1531 pscan = NULL;
1532 if (pr != pscan)
1533 continue;
1534 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1535 if (laddr.s_addr != INADDR_ANY)
1536 wildcard++;
1537 } else {
1538 if (laddr.s_addr == INADDR_ANY)
1539 wildcard++;
1540 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1541 continue;
1542 }
1543 if (wildcard && !wild_okay)
1544 continue;
1545 if (wildcard < matchwild) {
1546 match = inp;
1547 matchwild = wildcard;
1548 if (matchwild == 0)
1549 break;
1550 }
1551 }
1552 }
1553 return (match);
1554 }
1555
1556 struct inpcb *
in_pcblocalgroup_last(const struct inpcbinfo * pcbinfo,const struct inpcb * inp)1557 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1558 const struct inpcb *inp)
1559 {
1560 const struct inp_localgrphead *hdr;
1561 const struct inp_localgroup *grp;
1562 int i;
1563
1564 if (pcbinfo->localgrphashbase == NULL)
1565 return NULL;
1566
1567 GET_PCBINFO_TOKEN(pcbinfo);
1568
1569 hdr = &pcbinfo->localgrphashbase[
1570 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1571
1572 LIST_FOREACH(grp, hdr, il_list) {
1573 if (grp->il_af == inp->inp_af &&
1574 grp->il_lport == inp->inp_lport &&
1575 memcmp(&grp->il_dependladdr,
1576 &inp->inp_inc.inc_ie.ie_dependladdr,
1577 sizeof(grp->il_dependladdr)) == 0) {
1578 break;
1579 }
1580 }
1581 if (grp == NULL || grp->il_inpcnt == 1) {
1582 REL_PCBINFO_TOKEN(pcbinfo);
1583 return NULL;
1584 }
1585
1586 KASSERT(grp->il_inpcnt >= 2,
1587 ("invalid localgroup inp count %d", grp->il_inpcnt));
1588 for (i = 0; i < grp->il_inpcnt; ++i) {
1589 if (grp->il_inp[i] == inp) {
1590 int last = grp->il_inpcnt - 1;
1591
1592 if (i == last)
1593 last = grp->il_inpcnt - 2;
1594 REL_PCBINFO_TOKEN(pcbinfo);
1595 return grp->il_inp[last];
1596 }
1597 }
1598 REL_PCBINFO_TOKEN(pcbinfo);
1599 return NULL;
1600 }
1601
1602 static struct inpcb *
inp_localgroup_lookup(const struct inpcbinfo * pcbinfo,struct in_addr laddr,uint16_t lport,uint32_t pkt_hash)1603 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1604 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1605 {
1606 struct inpcb *local_wild;
1607 struct inpcb *jinp;
1608 struct inpcb *jinp_wild;
1609 struct inpcb *inp;
1610 const struct inp_localgrphead *hdr;
1611 const struct inp_localgroup *grp;
1612 struct sockaddr_in jsin;
1613 struct prison *pr;
1614 struct ucred *cred;
1615 int idx;
1616 int net_listen_ov_local;
1617 int net_listen_ov_wild;
1618
1619 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1620
1621 hdr = &pcbinfo->localgrphashbase[
1622 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1623
1624 /*
1625 * Order of socket selection:
1626 * 1. non-wild.
1627 * 2. wild.
1628 *
1629 * NOTE: Local group does not contain jailed sockets
1630 */
1631 jsin.sin_family = AF_INET;
1632 jsin.sin_addr.s_addr = laddr.s_addr;
1633
1634 jinp = NULL;
1635 jinp_wild = NULL;
1636 local_wild = NULL;
1637 net_listen_ov_local = 0;
1638 net_listen_ov_wild = 0;
1639
1640 LIST_FOREACH(grp, hdr, il_list) {
1641 #ifdef INET6
1642 if (grp->il_af != AF_INET)
1643 continue;
1644 #endif
1645 if (grp->il_lport != lport)
1646 continue;
1647
1648 /*
1649 * look for a match
1650 */
1651 idx = netisr_hashlsb(pkt_hash) % grp->il_inpcnt;
1652 inp = grp->il_inp[idx];
1653
1654 /*
1655 * Modulo-N is used here, which greatly reduces
1656 * completion queue token contention, thus more
1657 * cpu time is saved.
1658 */
1659 if (grp->il_jailed) {
1660 if (inp->inp_socket == NULL)
1661 continue;
1662 cred = inp->inp_socket->so_cred;
1663 if (cred == NULL)
1664 continue;
1665 pr = cred->cr_prison;
1666 if (pr == NULL)
1667 continue;
1668 if (!jailed_ip(pr, (struct sockaddr *)&jsin))
1669 continue;
1670 if (grp->il_laddr.s_addr == laddr.s_addr) {
1671 jinp = inp;
1672 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE))
1673 net_listen_ov_local = 1;
1674
1675 } else if (grp->il_laddr.s_addr == INADDR_ANY &&
1676 jinp_wild == NULL) {
1677 jinp_wild = inp;
1678 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE))
1679 net_listen_ov_wild = 1;
1680 }
1681 } else {
1682 if (grp->il_laddr.s_addr == laddr.s_addr) {
1683 return inp;
1684 } else if (grp->il_laddr.s_addr == INADDR_ANY) {
1685 local_wild = inp;
1686 }
1687 }
1688 }
1689
1690 if (net_listen_ov_local)
1691 return jinp;
1692 if (net_listen_ov_wild)
1693 return jinp_wild;
1694 if (local_wild)
1695 return (local_wild);
1696 if (jinp)
1697 return (jinp);
1698 return (jinp_wild);
1699 }
1700
1701 /*
1702 * Lookup PCB in hash list.
1703 *
1704 * This is used to match incoming packets to a pcb
1705 */
1706 struct inpcb *
in_pcblookup_pkthash(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport_arg,struct in_addr laddr,u_int lport_arg,boolean_t wildcard,struct ifnet * ifp,const struct mbuf * m)1707 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1708 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1709 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1710 {
1711 struct inpcbhead *head;
1712 struct inpcb *inp, *jinp=NULL;
1713 u_short fport = fport_arg, lport = lport_arg;
1714
1715 /*
1716 * First look for an exact match.
1717 */
1718 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1719 laddr.s_addr, lport,
1720 pcbinfo->hashmask)];
1721 LIST_FOREACH(inp, head, inp_hash) {
1722 #ifdef INET6
1723 if (!INP_ISIPV4(inp))
1724 continue;
1725 #endif
1726 if (in_hosteq(inp->inp_faddr, faddr) &&
1727 in_hosteq(inp->inp_laddr, laddr) &&
1728 inp->inp_fport == fport && inp->inp_lport == lport) {
1729 /*
1730 * Found specific address, host overrides jailed
1731 * inpcb.
1732 */
1733 if (inp->inp_socket == NULL ||
1734 inp->inp_socket->so_cred->cr_prison == NULL) {
1735 return (inp);
1736 }
1737 if (jinp == NULL)
1738 jinp = inp;
1739 }
1740 }
1741 if (jinp != NULL)
1742 return (jinp);
1743
1744 /*
1745 * We generally get here for connections to wildcarded listeners.
1746 * Any wildcarded listeners in jails must be restricted to the
1747 * jailed IPs only.
1748 */
1749 if (wildcard) {
1750 struct inpcb *local_wild = NULL;
1751 struct inpcb *jinp_wild = NULL;
1752 struct inpcontainer *ic;
1753 struct inpcontainerhead *chead;
1754 struct sockaddr_in jsin;
1755 struct ucred *cred;
1756 struct prison *pr;
1757 int net_listen_ov_local = 0;
1758 int net_listen_ov_wild = 0;
1759
1760 GET_PCBINFO_TOKEN(pcbinfo);
1761
1762 /*
1763 * Check local group first. When present, the localgroup
1764 * hash utilizes the same non-jailed-vs/jailed priortization
1765 * that the normal wildcardhash does.
1766 */
1767 if (pcbinfo->localgrphashbase != NULL &&
1768 m != NULL && (m->m_flags & M_HASH)) {
1769 inp = inp_localgroup_lookup(pcbinfo, laddr, lport,
1770 m->m_pkthdr.hash);
1771 if (inp != NULL) {
1772 REL_PCBINFO_TOKEN(pcbinfo);
1773 return inp;
1774 }
1775 }
1776
1777 /*
1778 * Order of socket selection:
1779 *
1780 * 1. non-jailed, non-wild.
1781 * 2. non-jailed, wild. (allow_listen_override on)
1782 * 3. jailed, non-wild.
1783 * 4. jailed, wild.
1784 * 5. non-jailed, wild. (allow_listen_override off)
1785 *
1786 * NOTE: jailed wildcards are still restricted to the jail
1787 * IPs.
1788 *
1789 * NOTE: (1) and (3) already handled above.
1790 */
1791 jsin.sin_family = AF_INET;
1792 chead = &pcbinfo->wildcardhashbase[
1793 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1794
1795 LIST_FOREACH(ic, chead, ic_list) {
1796 inp = ic->ic_inp;
1797 if (inp->inp_flags & INP_PLACEMARKER)
1798 continue;
1799
1800 /*
1801 * Basic validation
1802 */
1803 #ifdef INET6
1804 if (!INP_ISIPV4(inp))
1805 continue;
1806 #endif
1807 if (inp->inp_lport != lport)
1808 continue;
1809
1810 /*
1811 * Calculate prison, setup jsin for jailed_ip()
1812 * check.
1813 */
1814 jsin.sin_addr.s_addr = laddr.s_addr;
1815 pr = NULL;
1816 cred = NULL;
1817 if (inp->inp_socket) {
1818 cred = inp->inp_socket->so_cred;
1819 if (cred)
1820 pr = cred->cr_prison;
1821 }
1822
1823 /*
1824 * Assign jinp, jinp_wild, and local_wild as
1825 * appropriate, track whether the jail supports
1826 * listen overrides.
1827 */
1828 if (pr) {
1829 if (!jailed_ip(pr, (struct sockaddr *)&jsin))
1830 continue;
1831 if (inp->inp_laddr.s_addr == laddr.s_addr &&
1832 jinp == NULL) {
1833 jinp = inp;
1834 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE))
1835 net_listen_ov_local = 1;
1836 }
1837 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1838 jinp_wild == NULL) {
1839 jinp_wild = inp;
1840 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE))
1841 net_listen_ov_wild = 1;
1842 }
1843 } else {
1844 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1845 REL_PCBINFO_TOKEN(pcbinfo);
1846 return (inp);
1847 }
1848 if (inp->inp_laddr.s_addr == INADDR_ANY)
1849 local_wild = inp;
1850 }
1851 }
1852
1853 REL_PCBINFO_TOKEN(pcbinfo);
1854
1855 if (net_listen_ov_local)
1856 return jinp;
1857 if (net_listen_ov_wild)
1858 return jinp_wild;
1859 if (local_wild)
1860 return (local_wild);
1861 if (jinp)
1862 return (jinp);
1863 return (jinp_wild);
1864 }
1865
1866 /*
1867 * Not found.
1868 */
1869 return (NULL);
1870 }
1871
1872 struct inpcb *
in_pcblookup_hash(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport_arg,struct in_addr laddr,u_int lport_arg,boolean_t wildcard,struct ifnet * ifp)1873 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1874 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1875 boolean_t wildcard, struct ifnet *ifp)
1876 {
1877 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1878 laddr, lport_arg, wildcard, ifp, NULL);
1879 }
1880
1881 /*
1882 * Insert PCB into connection hash table.
1883 */
1884 void
in_pcbinsconnhash(struct inpcb * inp)1885 in_pcbinsconnhash(struct inpcb *inp)
1886 {
1887 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1888 struct inpcbhead *bucket;
1889 u_int32_t hashkey_faddr, hashkey_laddr;
1890
1891 #ifdef INET6
1892 if (INP_ISIPV6(inp)) {
1893 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1894 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1895 } else {
1896 #endif
1897 hashkey_faddr = inp->inp_faddr.s_addr;
1898 hashkey_laddr = inp->inp_laddr.s_addr;
1899 #ifdef INET6
1900 }
1901 #endif
1902
1903 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1904 ("not in the correct netisr"));
1905 ASSERT_INP_NOTINHASH(inp);
1906 inp->inp_flags |= INP_CONNECTED;
1907
1908 /*
1909 * Insert into the connection hash table.
1910 */
1911 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1912 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1913 LIST_INSERT_HEAD(bucket, inp, inp_hash);
1914 }
1915
1916 /*
1917 * Remove PCB from connection hash table.
1918 */
1919 void
in_pcbremconnhash(struct inpcb * inp)1920 in_pcbremconnhash(struct inpcb *inp)
1921 {
1922 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo;
1923
1924 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1925 ("not in the correct netisr"));
1926 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1927
1928 LIST_REMOVE(inp, inp_hash);
1929 inp->inp_flags &= ~INP_CONNECTED;
1930 }
1931
1932 /*
1933 * Insert PCB into port hash table.
1934 */
1935 void
in_pcbinsporthash(struct inpcbporthead * pcbporthash,struct inpcb * inp)1936 in_pcbinsporthash(struct inpcbporthead *pcbporthash, struct inpcb *inp)
1937 {
1938 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1939 struct inpcbport *phd;
1940
1941 /*
1942 * If the porthashbase is shared across several cpus, it must
1943 * have been locked.
1944 */
1945 ASSERT_PORTHASH_TOKEN_HELD(pcbporthash);
1946
1947 /*
1948 * Insert into the port hash table.
1949 */
1950
1951 /* Go through port list and look for a head for this lport. */
1952 LIST_FOREACH(phd, pcbporthash, phd_hash) {
1953 if (phd->phd_port == inp->inp_lport)
1954 break;
1955 }
1956
1957 /* If none exists, use saved one and tack it on. */
1958 if (phd == NULL) {
1959 KKASSERT(pcbinfo->portsave != NULL);
1960 phd = pcbinfo->portsave;
1961 pcbinfo->portsave = NULL;
1962 phd->phd_port = inp->inp_lport;
1963 LIST_INIT(&phd->phd_pcblist);
1964 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1965 }
1966
1967 inp->inp_porthash = pcbporthash;
1968 inp->inp_phd = phd;
1969 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1970
1971 /*
1972 * Malloc one inpcbport for later use. It is safe to use
1973 * "wait" malloc here (port token would be released, if
1974 * malloc ever blocked), since all changes to the porthash
1975 * are done.
1976 */
1977 if (pcbinfo->portsave == NULL) {
1978 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1979 M_PCB, M_INTWAIT | M_ZERO);
1980 }
1981 }
1982
1983 void
in_pcbinsporthash_lport(struct inpcb * inp)1984 in_pcbinsporthash_lport(struct inpcb *inp)
1985 {
1986 struct inpcbporthead *porthash;
1987
1988 porthash = OBTAIN_LPORTHASH_TOKEN(inp->inp_pcbinfo, inp->inp_lport);
1989 in_pcbinsporthash(porthash, inp);
1990 REL_PORTHASH_TOKEN(porthash);
1991 }
1992
1993 void
in_pcbremporthash(struct inpcb * inp)1994 in_pcbremporthash(struct inpcb *inp)
1995 {
1996 struct inpcbporthead *porthash;
1997 struct inpcbport *phd;
1998
1999 if (inp->inp_phd == NULL)
2000 return;
2001 KASSERT(inp->inp_lport != 0, ("inpcb has no lport"));
2002
2003 porthash = inp->inp_porthash;
2004 KASSERT(porthash != NULL, ("no porthash"));
2005
2006 GET_PORTHASH_TOKEN(porthash);
2007
2008 phd = inp->inp_phd;
2009 LIST_REMOVE(inp, inp_portlist);
2010 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2011 LIST_REMOVE(phd, phd_hash);
2012 kfree(phd, M_PCB);
2013 }
2014
2015 REL_PORTHASH_TOKEN(porthash);
2016
2017 inp->inp_phd = NULL;
2018 /* NOTE: Don't whack inp_lport, which may be used later */
2019 }
2020
2021 static struct inp_localgroup *
inp_localgroup_alloc(u_char af,uint16_t port,const union in_dependaddr * addr,int size)2022 inp_localgroup_alloc(u_char af, uint16_t port,
2023 const union in_dependaddr *addr, int size)
2024 {
2025 struct inp_localgroup *grp;
2026
2027 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
2028 M_TEMP, M_INTWAIT | M_ZERO);
2029 grp->il_af = af;
2030 grp->il_lport = port;
2031 grp->il_dependladdr = *addr;
2032 grp->il_inpsiz = size;
2033
2034 return grp;
2035 }
2036
2037 static void
inp_localgroup_free(struct inp_localgroup * grp)2038 inp_localgroup_free(struct inp_localgroup *grp)
2039 {
2040 kfree(grp, M_TEMP);
2041 }
2042
2043 static void
inp_localgroup_destroy(struct inp_localgroup * grp)2044 inp_localgroup_destroy(struct inp_localgroup *grp)
2045 {
2046 LIST_REMOVE(grp, il_list);
2047 inp_localgroup_free(grp);
2048 }
2049
2050 static void
inp_localgroup_copy(struct inp_localgroup * grp,const struct inp_localgroup * old_grp)2051 inp_localgroup_copy(struct inp_localgroup *grp,
2052 const struct inp_localgroup *old_grp)
2053 {
2054 int i;
2055
2056 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
2057 ("invalid new local group size %d and old local group count %d",
2058 grp->il_inpsiz, old_grp->il_inpcnt));
2059 for (i = 0; i < old_grp->il_inpcnt; ++i)
2060 grp->il_inp[i] = old_grp->il_inp[i];
2061 grp->il_inpcnt = old_grp->il_inpcnt;
2062 }
2063
2064 static void
in_pcbinslocalgrphash_oncpu(struct inpcb * inp,struct inpcbinfo * pcbinfo)2065 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2066 {
2067 struct inp_localgrphead *hdr;
2068 struct inp_localgroup *grp, *grp_alloc = NULL;
2069 u_char isjailed;
2070 int i, idx;
2071
2072 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
2073
2074 if (pcbinfo->localgrphashbase == NULL)
2075 return;
2076
2077 /*
2078 * Further separate groups by whether the inp is jailed or not.
2079 * This allows the inp_localgroup_lookup() code to manage port
2080 * overloading between jails and non-jails.
2081 *
2082 * XXX all jails are collected into one group, which works fine
2083 * as we expect the jails to be listening on different addresses.
2084 * If this changes in the future we may have to break the groups
2085 * up by prison pointer as well.
2086 */
2087 if (inp->inp_socket && inp->inp_socket->so_cred)
2088 isjailed = jailed(inp->inp_socket->so_cred);
2089 else
2090 isjailed = 0;
2091
2092 hdr = &pcbinfo->localgrphashbase[
2093 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
2094
2095 again:
2096 LIST_FOREACH(grp, hdr, il_list) {
2097 if (grp->il_af == inp->inp_af &&
2098 grp->il_lport == inp->inp_lport &&
2099 grp->il_jailed == isjailed &&
2100 memcmp(&grp->il_dependladdr,
2101 &inp->inp_inc.inc_ie.ie_dependladdr,
2102 sizeof(grp->il_dependladdr)) == 0) {
2103 break;
2104 }
2105 }
2106 if (grp == NULL) {
2107 /*
2108 * Create a new local group
2109 */
2110 if (grp_alloc == NULL) {
2111 grp_alloc = inp_localgroup_alloc(inp->inp_af,
2112 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
2113 INP_LOCALGROUP_SIZMIN);
2114 /*
2115 * Local group allocation could block and the
2116 * local group w/ the same property might have
2117 * been added by others when we were blocked;
2118 * check again.
2119 */
2120 goto again;
2121 } else {
2122 /* Local group has been allocated; link it */
2123 grp = grp_alloc;
2124 grp->il_jailed = isjailed;
2125 grp_alloc = NULL;
2126 LIST_INSERT_HEAD(hdr, grp, il_list);
2127 }
2128 } else if (grp->il_inpcnt == grp->il_inpsiz) {
2129 #if 0
2130 /*
2131 * REMOVED - Ensure that all entries are placed in the
2132 * localgroup so jail operations can be
2133 * deterministic on a il_lport basis.
2134 */
2135 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
2136 static int limit_logged = 0;
2137
2138 if (!limit_logged) {
2139 limit_logged = 1;
2140 kprintf("local group port %d, "
2141 "limit reached\n", ntohs(grp->il_lport));
2142 }
2143 if (grp_alloc != NULL) {
2144 /*
2145 * This would happen if the local group
2146 * w/ the same property was expanded when
2147 * our local group allocation blocked.
2148 */
2149 inp_localgroup_free(grp_alloc);
2150 }
2151 return;
2152 }
2153 #endif
2154
2155 /*
2156 * Expand this local group
2157 */
2158 if (grp_alloc == NULL ||
2159 grp->il_inpcnt >= grp_alloc->il_inpsiz) {
2160 if (grp_alloc != NULL)
2161 inp_localgroup_free(grp_alloc);
2162 grp_alloc = inp_localgroup_alloc(grp->il_af,
2163 grp->il_lport, &grp->il_dependladdr,
2164 grp->il_inpsiz * 2);
2165 /*
2166 * Local group allocation could block and the
2167 * local group w/ the same property might have
2168 * been expanded by others when we were blocked;
2169 * check again.
2170 */
2171 goto again;
2172 }
2173
2174 /*
2175 * Save the old local group, link the new one, and then
2176 * destroy the old local group
2177 */
2178 inp_localgroup_copy(grp_alloc, grp);
2179 LIST_INSERT_HEAD(hdr, grp_alloc, il_list);
2180 inp_localgroup_destroy(grp);
2181
2182 grp = grp_alloc;
2183 grp->il_jailed = isjailed;
2184 grp_alloc = NULL;
2185 } else {
2186 /*
2187 * Found the local group
2188 */
2189 if (grp_alloc != NULL) {
2190 /*
2191 * This would happen if the local group w/ the
2192 * same property was added or expanded when our
2193 * local group allocation blocked.
2194 */
2195 inp_localgroup_free(grp_alloc);
2196 grp_alloc = NULL;
2197 }
2198 }
2199
2200 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
2201 ("invalid local group size %d and count %d",
2202 grp->il_inpsiz, grp->il_inpcnt));
2203
2204 /*
2205 * Keep the local group sorted by the inpcb local group index
2206 * in ascending order.
2207 *
2208 * This eases the multi-process userland application which uses
2209 * SO_REUSEPORT sockets and binds process to the owner cpu of
2210 * the SO_REUSEPORT socket:
2211 * If we didn't sort the local group by the inpcb local group
2212 * index and one of the process owning an inpcb in this local
2213 * group restarted, e.g. crashed and restarted by watchdog,
2214 * other processes owning a inpcb in this local group would have
2215 * to detect that event, refetch its socket's owner cpu, and
2216 * re-bind.
2217 */
2218 idx = grp->il_inpcnt;
2219 for (i = 0; i < idx; ++i) {
2220 struct inpcb *oinp = grp->il_inp[i];
2221
2222 if (oinp->inp_lgrpindex > i) {
2223 if (inp->inp_lgrpindex < 0) {
2224 inp->inp_lgrpindex = i;
2225 } else if (inp->inp_lgrpindex != i) {
2226 if (bootverbose) {
2227 kprintf("inp %p: grpidx %d, "
2228 "assigned to %d, cpu%d\n",
2229 inp, inp->inp_lgrpindex, i,
2230 mycpuid);
2231 }
2232 }
2233 grp->il_inp[i] = inp;
2234
2235 /* Pull down inpcbs */
2236 for (; i < grp->il_inpcnt; ++i) {
2237 struct inpcb *oinp1 = grp->il_inp[i + 1];
2238
2239 grp->il_inp[i + 1] = oinp;
2240 oinp = oinp1;
2241 }
2242 grp->il_inpcnt++;
2243 return;
2244 }
2245 }
2246
2247 if (inp->inp_lgrpindex < 0) {
2248 inp->inp_lgrpindex = idx;
2249 } else if (inp->inp_lgrpindex != idx) {
2250 if (bootverbose) {
2251 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n",
2252 inp, inp->inp_lgrpindex, idx, mycpuid);
2253 }
2254 }
2255 grp->il_inp[idx] = inp;
2256 grp->il_inpcnt++;
2257 }
2258
2259 void
in_pcbinswildcardhash_oncpu(struct inpcb * inp,struct inpcbinfo * pcbinfo)2260 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2261 {
2262 struct inpcontainer *ic;
2263 struct inpcontainerhead *bucket;
2264
2265 GET_PCBINFO_TOKEN(pcbinfo);
2266
2267 in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
2268
2269 bucket = &pcbinfo->wildcardhashbase[
2270 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2271
2272 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
2273 ic->ic_inp = inp;
2274 LIST_INSERT_HEAD(bucket, ic, ic_list);
2275
2276 REL_PCBINFO_TOKEN(pcbinfo);
2277 }
2278
2279 /*
2280 * Insert PCB into wildcard hash table.
2281 */
2282 void
in_pcbinswildcardhash(struct inpcb * inp)2283 in_pcbinswildcardhash(struct inpcb *inp)
2284 {
2285 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2286
2287 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2288 ("not in correct netisr"));
2289 ASSERT_INP_NOTINHASH(inp);
2290 inp->inp_flags |= INP_WILDCARD;
2291
2292 in_pcbinswildcardhash_oncpu(inp, pcbinfo);
2293 }
2294
2295 static void
in_pcbremlocalgrphash_oncpu(struct inpcb * inp,struct inpcbinfo * pcbinfo)2296 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2297 {
2298 struct inp_localgrphead *hdr;
2299 struct inp_localgroup *grp;
2300
2301 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
2302
2303 if (pcbinfo->localgrphashbase == NULL)
2304 return;
2305
2306 hdr = &pcbinfo->localgrphashbase[
2307 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
2308
2309 LIST_FOREACH(grp, hdr, il_list) {
2310 int i;
2311
2312 for (i = 0; i < grp->il_inpcnt; ++i) {
2313 if (grp->il_inp[i] != inp)
2314 continue;
2315
2316 if (grp->il_inpcnt == 1) {
2317 /* Destroy this local group */
2318 inp_localgroup_destroy(grp);
2319 } else {
2320 /* Pull up inpcbs */
2321 for (; i + 1 < grp->il_inpcnt; ++i)
2322 grp->il_inp[i] = grp->il_inp[i + 1];
2323 grp->il_inpcnt--;
2324 }
2325 return;
2326 }
2327 }
2328 }
2329
2330 void
in_pcbremwildcardhash_oncpu(struct inpcb * inp,struct inpcbinfo * pcbinfo)2331 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2332 {
2333 struct inpcontainer *ic;
2334 struct inpcontainerhead *head;
2335
2336 GET_PCBINFO_TOKEN(pcbinfo);
2337
2338 in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
2339
2340 /* find bucket */
2341 head = &pcbinfo->wildcardhashbase[
2342 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2343
2344 LIST_FOREACH(ic, head, ic_list) {
2345 if (ic->ic_inp == inp)
2346 goto found;
2347 }
2348 REL_PCBINFO_TOKEN(pcbinfo);
2349 return; /* not found! */
2350
2351 found:
2352 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */
2353 REL_PCBINFO_TOKEN(pcbinfo);
2354 kfree(ic, M_TEMP); /* deallocate container */
2355 }
2356
2357 /*
2358 * Remove PCB from wildcard hash table.
2359 */
2360 void
in_pcbremwildcardhash(struct inpcb * inp)2361 in_pcbremwildcardhash(struct inpcb *inp)
2362 {
2363 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2364
2365 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2366 ("not in correct netisr"));
2367 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
2368
2369 in_pcbremwildcardhash_oncpu(inp, pcbinfo);
2370 inp->inp_lgrpindex = -1;
2371 inp->inp_flags &= ~INP_WILDCARD;
2372 }
2373
2374 /*
2375 * Remove PCB from various lists.
2376 */
2377 void
in_pcbremlists(struct inpcb * inp)2378 in_pcbremlists(struct inpcb *inp)
2379 {
2380 in_pcbremporthash(inp);
2381 if (inp->inp_flags & INP_WILDCARD) {
2382 in_pcbremwildcardhash(inp);
2383 } else if (inp->inp_flags & INP_CONNECTED) {
2384 in_pcbremconnhash(inp);
2385 }
2386
2387 if (inp->inp_flags & INP_ONLIST)
2388 in_pcbofflist(inp);
2389 }
2390
2391 int
prison_xinpcb(struct thread * td,struct inpcb * inp)2392 prison_xinpcb(struct thread *td, struct inpcb *inp)
2393 {
2394 struct ucred *cr;
2395
2396 if (td->td_proc == NULL)
2397 return (0);
2398 cr = td->td_proc->p_ucred;
2399 if (cr->cr_prison == NULL)
2400 return (0);
2401 if (inp->inp_socket && inp->inp_socket->so_cred &&
2402 inp->inp_socket->so_cred->cr_prison &&
2403 cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
2404 return (0);
2405 return (1);
2406 }
2407
2408 int
in_pcblist_range(SYSCTL_HANDLER_ARGS)2409 in_pcblist_range(SYSCTL_HANDLER_ARGS)
2410 {
2411 struct inpcbinfo *pcbinfo_arr = arg1;
2412 int pcbinfo_arrlen = arg2;
2413 struct inpcb *marker;
2414 int cpu, origcpu;
2415 int error, n;
2416
2417 KASSERT(pcbinfo_arrlen <= netisr_ncpus && pcbinfo_arrlen >= 1,
2418 ("invalid pcbinfo count %d", pcbinfo_arrlen));
2419
2420 /*
2421 * The process of preparing the TCB list is too time-consuming and
2422 * resource-intensive to repeat twice on every request.
2423 */
2424 n = 0;
2425 if (req->oldptr == NULL) {
2426 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu)
2427 n += pcbinfo_arr[cpu].ipi_count;
2428 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
2429 return 0;
2430 }
2431
2432 if (req->newptr != NULL)
2433 return EPERM;
2434
2435 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
2436 marker->inp_flags |= INP_PLACEMARKER;
2437
2438 /*
2439 * OK, now we're committed to doing something. Re-fetch ipi_count
2440 * after obtaining the generation count.
2441 */
2442 error = 0;
2443 origcpu = mycpuid;
2444 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) {
2445 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu];
2446 struct inpcb *inp;
2447 struct xinpcb xi;
2448 int i;
2449
2450 lwkt_migratecpu(cpu);
2451
2452 GET_PCBINFO_TOKEN(pcbinfo);
2453
2454 n = pcbinfo->ipi_count;
2455
2456 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
2457 i = 0;
2458 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
2459 LIST_REMOVE(marker, inp_list);
2460 LIST_INSERT_AFTER(inp, marker, inp_list);
2461
2462 if (inp->inp_flags & INP_PLACEMARKER)
2463 continue;
2464 if (prison_xinpcb(req->td, inp))
2465 continue;
2466
2467 bzero(&xi, sizeof xi);
2468 xi.xi_len = sizeof xi;
2469 bcopy(inp, &xi.xi_inp, sizeof *inp);
2470 if (inp->inp_socket)
2471 sotoxsocket(inp->inp_socket, &xi.xi_socket);
2472 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
2473 break;
2474 ++i;
2475 }
2476 LIST_REMOVE(marker, inp_list);
2477
2478 REL_PCBINFO_TOKEN(pcbinfo);
2479
2480 if (error == 0 && i < n) {
2481 bzero(&xi, sizeof xi);
2482 xi.xi_len = sizeof xi;
2483 while (i < n) {
2484 error = SYSCTL_OUT(req, &xi, sizeof xi);
2485 if (error)
2486 break;
2487 ++i;
2488 }
2489 }
2490 }
2491
2492 lwkt_migratecpu(origcpu);
2493 kfree(marker, M_TEMP);
2494 return error;
2495 }
2496
2497 int
in_pcblist_ncpus(SYSCTL_HANDLER_ARGS)2498 in_pcblist_ncpus(SYSCTL_HANDLER_ARGS)
2499 {
2500
2501 return (in_pcblist_range(oidp, arg1, netisr_ncpus, req));
2502 }
2503
2504 void
in_savefaddr(struct socket * so,const struct sockaddr * faddr)2505 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2506 {
2507 struct sockaddr_in *sin;
2508
2509 KASSERT(faddr->sa_family == AF_INET,
2510 ("not AF_INET faddr %d", faddr->sa_family));
2511
2512 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2513 sin->sin_family = AF_INET;
2514 sin->sin_len = sizeof(*sin);
2515 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2516 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2517
2518 so->so_faddr = (struct sockaddr *)sin;
2519 }
2520
2521 void
in_pcbportinfo_init(struct inpcbportinfo * portinfo,int hashsize,u_short offset)2522 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2523 u_short offset)
2524 {
2525 memset(portinfo, 0, sizeof(*portinfo));
2526
2527 portinfo->offset = offset;
2528 portinfo->porthashbase = phashinit(hashsize, M_PCB,
2529 &portinfo->porthashcnt);
2530 }
2531
2532 void
in_pcbportrange(u_short * hi0,u_short * lo0,u_short ofs,u_short step)2533 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step)
2534 {
2535 int hi, lo;
2536
2537 if (step == 1)
2538 return;
2539
2540 hi = *hi0;
2541 lo = *lo0;
2542
2543 hi = rounddown(hi, step);
2544 hi += ofs;
2545 if (hi > (int)*hi0)
2546 hi -= step;
2547
2548 lo = roundup(lo, step);
2549 lo -= (step - ofs);
2550 if (lo < (int)*lo0)
2551 lo += step;
2552
2553 *hi0 = hi;
2554 *lo0 = lo;
2555 }
2556
2557 void
in_pcbglobalinit(void)2558 in_pcbglobalinit(void)
2559 {
2560 int cpu;
2561
2562 in_pcbmarkers = kmalloc(netisr_ncpus * sizeof(struct inpcb), M_PCB,
2563 M_WAITOK | M_ZERO);
2564 in_pcbcontainer_markers =
2565 kmalloc(netisr_ncpus * sizeof(struct inpcontainer), M_PCB,
2566 M_WAITOK | M_ZERO);
2567
2568 for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
2569 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu];
2570 struct inpcb *marker = &in_pcbmarkers[cpu];
2571
2572 marker->inp_flags |= INP_PLACEMARKER;
2573 ic->ic_inp = marker;
2574 }
2575 }
2576
2577 struct inpcb *
in_pcbmarker(void)2578 in_pcbmarker(void)
2579 {
2580
2581 ASSERT_NETISR_NCPUS(mycpuid);
2582 return &in_pcbmarkers[mycpuid];
2583 }
2584
2585 struct inpcontainer *
in_pcbcontainer_marker(void)2586 in_pcbcontainer_marker(void)
2587 {
2588
2589 ASSERT_NETISR_NCPUS(mycpuid);
2590 return &in_pcbcontainer_markers[mycpuid];
2591 }
2592
2593 void
in_pcbresetroute(struct inpcb * inp)2594 in_pcbresetroute(struct inpcb *inp)
2595 {
2596 struct route *ro = &inp->inp_route;
2597
2598 if (ro->ro_rt != NULL)
2599 RTFREE(ro->ro_rt);
2600 bzero(ro, sizeof(*ro));
2601 }
2602