usched: Allow process to change self cpu affinity
[dragonfly.git] / sys / netinet / in_pcb.c
blob7f8e68e39e109af001c4ed2103207bfe95695e54
1 /*
2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
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.
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.
35 * Copyright (c) 1982, 1986, 1991, 1993, 1995
36 * The Regents of the University of California. All rights reserved.
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.
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.
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 $
66 #include "opt_ipsec.h"
67 #include "opt_inet6.h"
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/malloc.h>
72 #include <sys/mbuf.h>
73 #include <sys/domain.h>
74 #include <sys/protosw.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.h>
77 #include <sys/proc.h>
78 #include <sys/priv.h>
79 #include <sys/jail.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
83 #include <sys/thread2.h>
84 #include <sys/socketvar2.h>
85 #include <sys/msgport2.h>
87 #include <machine/limits.h>
89 #include <net/if.h>
90 #include <net/if_types.h>
91 #include <net/route.h>
92 #include <net/netisr2.h>
93 #include <net/toeplitz2.h>
95 #include <netinet/in.h>
96 #include <netinet/in_pcb.h>
97 #include <netinet/in_var.h>
98 #include <netinet/ip_var.h>
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #include <netinet6/ip6_var.h>
102 #endif /* INET6 */
104 #ifdef IPSEC
105 #include <netinet6/ipsec.h>
106 #include <netproto/key/key.h>
107 #include <netproto/ipsec/esp_var.h>
108 #endif
110 #ifdef FAST_IPSEC
111 #if defined(IPSEC) || defined(IPSEC_ESP)
112 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!"
113 #endif
115 #include <netproto/ipsec/ipsec.h>
116 #include <netproto/ipsec/key.h>
117 #define IPSEC
118 #endif /* FAST_IPSEC */
120 #define INP_LOCALGROUP_SIZMIN 8
121 #define INP_LOCALGROUP_SIZMAX 256
123 static struct inpcb *in_pcblookup_local(struct inpcbporthead *porthash,
124 struct in_addr laddr, u_int lport_arg, int wild_okay,
125 struct ucred *cred);
127 struct in_addr zeroin_addr;
130 * These configure the range of local port addresses assigned to
131 * "unspecified" outgoing connections/packets/whatever.
133 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */
134 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */
136 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */
137 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */
139 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */
140 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */
142 #define RANGECHK(var, min, max) \
143 if ((var) < (min)) { (var) = (min); } \
144 else if ((var) > (max)) { (var) = (max); }
146 int udpencap_enable = 1; /* enabled by default */
147 int udpencap_port = 4500; /* triggers decapsulation */
150 * Per-netisr inpcb markers.
151 * NOTE: they should only be used in netisrs.
153 static struct inpcb *in_pcbmarkers;
154 static struct inpcontainer *in_pcbcontainer_markers;
156 static int
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
159 int error;
161 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
162 if (!error) {
163 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
166 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX);
167 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX);
169 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX);
170 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX);
172 return (error);
175 #undef RANGECHK
177 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW,
180 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW,
182 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", "");
183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW,
184 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", "");
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW,
186 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW,
188 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", "");
189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW,
190 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", "");
192 static int ip_porthash_trycount = 15;
193 SYSCTL_INT(_net_inet_ip, OID_AUTO, porthash_trycount, CTLFLAG_RW,
194 &ip_porthash_trycount, 0,
195 "Number of tries to find local port matching hash of 4-tuple");
198 * in_pcb.c: manage the Protocol Control Blocks.
200 * NOTE: It is assumed that most of these functions will be called from
201 * a critical section. XXX - There are, unfortunately, a few exceptions
202 * to this rule that should be fixed.
204 * NOTE: The caller should initialize the cpu field to the cpu running the
205 * protocol stack associated with this inpcbinfo.
208 void
209 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared)
211 KASSERT(cpu >= 0 && cpu < ncpus, ("invalid cpu%d", cpu));
212 pcbinfo->cpu = cpu;
214 LIST_INIT(&pcbinfo->pcblisthead);
215 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB,
216 M_WAITOK | M_ZERO);
218 if (shared) {
219 pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token),
220 M_PCB, M_WAITOK);
221 lwkt_token_init(pcbinfo->infotoken, "infotoken");
222 } else {
223 pcbinfo->infotoken = NULL;
227 void
228 in_pcbportinfo_set(struct inpcbinfo *pcbinfo, struct inpcbportinfo *portinfo,
229 int portinfo_cnt)
232 KASSERT(portinfo_cnt > 0, ("invalid portinfo_cnt %d", portinfo_cnt));
233 pcbinfo->portinfo = portinfo;
234 pcbinfo->portinfo_cnt = portinfo_cnt;
237 struct baddynamicports baddynamicports;
240 * Check if the specified port is invalid for dynamic allocation.
243 in_baddynamic(u_int16_t port, u_int16_t proto)
245 switch (proto) {
246 case IPPROTO_TCP:
247 return (DP_ISSET(baddynamicports.tcp, port));
248 case IPPROTO_UDP:
249 #ifdef IPSEC
250 /* Cannot preset this as it is a sysctl */
251 if (port == udpencap_port)
252 return (1);
253 #endif
254 return (DP_ISSET(baddynamicports.udp, port));
255 default:
256 return (0);
260 void
261 in_pcbonlist(struct inpcb *inp)
263 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
265 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
266 ("not in the correct netisr"));
267 KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist"));
268 inp->inp_flags |= INP_ONLIST;
270 GET_PCBINFO_TOKEN(pcbinfo);
271 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
272 pcbinfo->ipi_count++;
273 REL_PCBINFO_TOKEN(pcbinfo);
276 void
277 in_pcbofflist(struct inpcb *inp)
279 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
281 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
282 ("not in the correct netisr"));
283 KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist"));
284 inp->inp_flags &= ~INP_ONLIST;
286 GET_PCBINFO_TOKEN(pcbinfo);
287 LIST_REMOVE(inp, inp_list);
288 KASSERT(pcbinfo->ipi_count > 0,
289 ("invalid inpcb count %d", pcbinfo->ipi_count));
290 pcbinfo->ipi_count--;
291 REL_PCBINFO_TOKEN(pcbinfo);
295 * Allocate a PCB and associate it with the socket.
298 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
300 struct inpcb *inp;
301 #ifdef IPSEC
302 int error;
303 #endif
305 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK);
306 if (inp == NULL)
307 return (ENOMEM);
308 inp->inp_lgrpindex = -1;
309 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
310 inp->inp_pcbinfo = pcbinfo;
311 inp->inp_socket = so;
312 #ifdef IPSEC
313 error = ipsec_init_policy(so, &inp->inp_sp);
314 if (error != 0) {
315 kfree(inp, M_PCB);
316 return (error);
318 #endif
319 #ifdef INET6
320 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
321 if (ip6_auto_flowlabel)
322 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
323 inp->inp_af = AF_INET6;
324 } else
325 #endif
326 inp->inp_af = AF_INET;
327 soreference(so);
328 so->so_pcb = inp;
330 in_pcbonlist(inp);
331 return (0);
335 * Unlink a pcb with the intention of moving it to another cpu with a
336 * different pcbinfo. While unlinked nothing should attempt to dereference
337 * inp_pcbinfo, NULL it out so we assert if it does.
339 void
340 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
342 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
343 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
344 ("already linked"));
346 in_pcbofflist(inp);
347 inp->inp_pcbinfo = NULL;
350 void
351 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
353 in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD);
357 * Relink a pcb into a new pcbinfo.
359 void
360 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
362 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
363 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
364 ("already linked"));
366 inp->inp_pcbinfo = pcbinfo;
367 in_pcbonlist(inp);
370 void
371 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
373 return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD);
376 static boolean_t
377 in_pcbporthash_update(struct inpcbportinfo *portinfo,
378 struct inpcb *inp, u_short lport, struct ucred *cred, int wild)
380 struct inpcbporthead *porthash;
383 * This has to be atomic. If the porthash is shared across multiple
384 * protocol threads, e.g. tcp and udp, then the token must be held.
386 porthash = in_pcbporthash_head(portinfo, lport);
387 GET_PORTHASH_TOKEN(porthash);
389 if (in_pcblookup_local(porthash, inp->inp_laddr, lport,
390 wild, cred) != NULL) {
391 REL_PORTHASH_TOKEN(porthash);
392 return FALSE;
394 inp->inp_lport = lport;
395 in_pcbinsporthash(porthash, inp);
397 REL_PORTHASH_TOKEN(porthash);
398 return TRUE;
401 static int
402 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
404 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
405 struct inpcbportinfo *portinfo;
406 u_short first, last, lport, step;
407 u_short *lastport;
408 int count, error;
409 int portinfo_first, portinfo_idx;
411 inp->inp_flags |= INP_ANONPORT;
413 step = pcbinfo->portinfo_cnt;
414 portinfo_first = mycpuid % pcbinfo->portinfo_cnt;
415 portinfo_idx = portinfo_first;
416 loop:
417 portinfo = &pcbinfo->portinfo[portinfo_idx];
419 if (inp->inp_flags & INP_HIGHPORT) {
420 first = ipport_hifirstauto; /* sysctl */
421 last = ipport_hilastauto;
422 lastport = &portinfo->lasthi;
423 } else if (inp->inp_flags & INP_LOWPORT) {
424 if (cred &&
425 (error =
426 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
427 inp->inp_laddr.s_addr = INADDR_ANY;
428 return error;
430 first = ipport_lowfirstauto; /* 1023 */
431 last = ipport_lowlastauto; /* 600 */
432 lastport = &portinfo->lastlow;
433 } else {
434 first = ipport_firstauto; /* sysctl */
435 last = ipport_lastauto;
436 lastport = &portinfo->lastport;
440 * Simple check to ensure all ports are not used up causing
441 * a deadlock here.
443 * We split the two cases (up and down) so that the direction
444 * is not being tested on each round of the loop.
446 if (first > last) {
448 * counting down
450 in_pcbportrange(&first, &last, portinfo->offset, step);
451 count = (first - last) / step;
453 for (;;) {
454 if (count-- < 0) { /* completely used? */
455 error = EADDRNOTAVAIL;
456 break;
458 lport = in_pcblastport_down(lastport,
459 first, last, step);
460 KKASSERT((lport % pcbinfo->portinfo_cnt) ==
461 portinfo->offset);
462 lport = htons(lport);
464 if (in_pcbporthash_update(portinfo, inp, lport,
465 cred, wild)) {
466 error = 0;
467 break;
470 } else {
472 * counting up
474 in_pcbportrange(&last, &first, portinfo->offset, step);
475 count = (last - first) / step;
477 for (;;) {
478 if (count-- < 0) { /* completely used? */
479 error = EADDRNOTAVAIL;
480 break;
482 lport = in_pcblastport_up(lastport, first, last, step);
483 KKASSERT((lport % pcbinfo->portinfo_cnt) ==
484 portinfo->offset);
485 lport = htons(lport);
487 if (in_pcbporthash_update(portinfo, inp, lport,
488 cred, wild)) {
489 error = 0;
490 break;
495 if (error) {
496 /* Try next portinfo */
497 portinfo_idx++;
498 portinfo_idx %= pcbinfo->portinfo_cnt;
499 if (portinfo_idx != portinfo_first)
500 goto loop;
501 inp->inp_laddr.s_addr = INADDR_ANY;
503 return error;
507 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
509 struct socket *so = inp->inp_socket;
510 struct sockaddr_in jsin;
511 struct ucred *cred = NULL;
512 int wild = 0;
514 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
515 return (EADDRNOTAVAIL);
516 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
517 return (EINVAL); /* already bound */
519 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
520 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
521 if (td->td_proc)
522 cred = td->td_proc->p_ucred;
524 if (nam != NULL) {
525 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
526 struct inpcbinfo *pcbinfo;
527 struct inpcbportinfo *portinfo;
528 struct inpcbporthead *porthash;
529 struct inpcb *t;
530 u_short lport, lport_ho;
531 int reuseport = (so->so_options & SO_REUSEPORT);
532 int error;
534 if (nam->sa_len != sizeof *sin)
535 return (EINVAL);
536 #ifdef notdef
538 * We should check the family, but old programs
539 * incorrectly fail to initialize it.
541 if (sin->sin_family != AF_INET)
542 return (EAFNOSUPPORT);
543 #endif
544 if (!prison_replace_wildcards(td, nam))
545 return (EINVAL);
547 lport = sin->sin_port;
548 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
550 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
551 * allow complete duplication of binding if
552 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
553 * and a multicast address is bound on both
554 * new and duplicated sockets.
556 if (so->so_options & SO_REUSEADDR)
557 reuseport = SO_REUSEADDR | SO_REUSEPORT;
558 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
559 sin->sin_port = 0; /* yech... */
560 bzero(&sin->sin_zero, sizeof sin->sin_zero);
561 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
562 return (EADDRNOTAVAIL);
565 inp->inp_laddr = sin->sin_addr;
567 jsin.sin_family = AF_INET;
568 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
569 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
570 inp->inp_laddr.s_addr = INADDR_ANY;
571 return (EINVAL);
573 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
575 if (lport == 0) {
576 /* Auto-select local port */
577 return in_pcbsetlport(inp, wild, cred);
579 lport_ho = ntohs(lport);
581 /* GROSS */
582 if (lport_ho < IPPORT_RESERVED && cred &&
583 (error =
584 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
585 inp->inp_laddr.s_addr = INADDR_ANY;
586 return (error);
590 * Locate the proper portinfo based on lport
592 pcbinfo = inp->inp_pcbinfo;
593 portinfo =
594 &pcbinfo->portinfo[lport_ho % pcbinfo->portinfo_cnt];
595 KKASSERT((lport_ho % pcbinfo->portinfo_cnt) ==
596 portinfo->offset);
599 * This has to be atomic. If the porthash is shared across
600 * multiple protocol threads, e.g. tcp and udp then the token
601 * must be held.
603 porthash = in_pcbporthash_head(portinfo, lport);
604 GET_PORTHASH_TOKEN(porthash);
606 if (so->so_cred->cr_uid != 0 &&
607 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
608 t = in_pcblookup_local(porthash, sin->sin_addr, lport,
609 INPLOOKUP_WILDCARD, cred);
610 if (t &&
611 (so->so_cred->cr_uid !=
612 t->inp_socket->so_cred->cr_uid)) {
613 inp->inp_laddr.s_addr = INADDR_ANY;
614 error = EADDRINUSE;
615 goto done;
618 if (cred && !prison_replace_wildcards(td, nam)) {
619 inp->inp_laddr.s_addr = INADDR_ANY;
620 error = EADDRNOTAVAIL;
621 goto done;
623 t = in_pcblookup_local(porthash, sin->sin_addr, lport,
624 wild, cred);
625 if (t && !(reuseport & t->inp_socket->so_options)) {
626 inp->inp_laddr.s_addr = INADDR_ANY;
627 error = EADDRINUSE;
628 goto done;
630 inp->inp_lport = lport;
631 in_pcbinsporthash(porthash, inp);
632 error = 0;
633 done:
634 REL_PORTHASH_TOKEN(porthash);
635 return (error);
636 } else {
637 jsin.sin_family = AF_INET;
638 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
639 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
640 inp->inp_laddr.s_addr = INADDR_ANY;
641 return (EINVAL);
643 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
645 return in_pcbsetlport(inp, wild, cred);
649 static struct inpcb *
650 in_pcblookup_localremote(struct inpcbporthead *porthash, struct in_addr laddr,
651 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
653 struct inpcb *inp;
654 struct inpcbport *phd;
655 struct inpcb *match = NULL;
658 * If the porthashbase is shared across several cpus, it must
659 * have been locked.
661 ASSERT_PORTHASH_TOKEN_HELD(porthash);
664 * Best fit PCB lookup.
666 * First see if this local port is in use by looking on the
667 * port hash list.
669 LIST_FOREACH(phd, porthash, phd_hash) {
670 if (phd->phd_port == lport)
671 break;
673 if (phd != NULL) {
674 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
675 #ifdef INET6
676 if (!INP_ISIPV4(inp))
677 continue;
678 #endif
679 if (inp->inp_laddr.s_addr != INADDR_ANY &&
680 inp->inp_laddr.s_addr != laddr.s_addr)
681 continue;
683 if (inp->inp_faddr.s_addr != INADDR_ANY &&
684 inp->inp_faddr.s_addr != faddr.s_addr)
685 continue;
687 if (inp->inp_fport != 0 && inp->inp_fport != fport)
688 continue;
690 if (cred == NULL ||
691 cred->cr_prison ==
692 inp->inp_socket->so_cred->cr_prison) {
693 match = inp;
694 break;
698 return (match);
701 static boolean_t
702 in_pcbporthash_update4(struct inpcbportinfo *portinfo,
703 struct inpcb *inp, u_short lport, const struct sockaddr_in *sin,
704 struct ucred *cred)
706 struct inpcbporthead *porthash;
709 * This has to be atomic. If the porthash is shared across multiple
710 * protocol threads, e.g. tcp and udp, then the token must be held.
712 porthash = in_pcbporthash_head(portinfo, lport);
713 GET_PORTHASH_TOKEN(porthash);
715 if (in_pcblookup_localremote(porthash, inp->inp_laddr,
716 lport, sin->sin_addr, sin->sin_port, cred) != NULL) {
717 REL_PORTHASH_TOKEN(porthash);
718 return FALSE;
720 inp->inp_lport = lport;
721 in_pcbinsporthash(porthash, inp);
723 REL_PORTHASH_TOKEN(porthash);
724 return TRUE;
728 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote,
729 struct thread *td)
731 struct proc *p = td->td_proc;
732 u_short *lastport;
733 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
734 struct sockaddr_in jsin;
735 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
736 struct inpcbportinfo *portinfo;
737 struct ucred *cred = NULL;
738 u_short first, last, lport, step;
739 int count, error, selfconn;
740 int portinfo_first, portinfo_idx;
741 int hash_cpu, hash_count, hash_count0;
742 uint32_t hash_base = 0, hash;
744 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
745 return (EADDRNOTAVAIL);
747 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
748 if (inp->inp_lport != 0)
749 return (EINVAL); /* already bound */
751 KKASSERT(p);
752 cred = p->p_ucred;
754 jsin.sin_family = AF_INET;
755 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
756 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
757 inp->inp_laddr.s_addr = INADDR_ANY;
758 return (EINVAL);
760 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
762 hash_count0 = ip_porthash_trycount;
763 if (hash_count0 > 0) {
764 hash_base = toeplitz_piecemeal_addr(sin->sin_addr.s_addr) ^
765 toeplitz_piecemeal_addr(inp->inp_laddr.s_addr) ^
766 toeplitz_piecemeal_port(sin->sin_port);
767 } else {
768 hash_count0 = 0;
771 inp->inp_flags |= INP_ANONPORT;
773 step = pcbinfo->portinfo_cnt;
774 portinfo_first = mycpuid % pcbinfo->portinfo_cnt;
775 portinfo_idx = portinfo_first;
776 loop:
777 hash_cpu = portinfo_idx % netisr_ncpus;
778 portinfo = &pcbinfo->portinfo[portinfo_idx];
779 selfconn = 0;
781 if (inp->inp_flags & INP_HIGHPORT) {
782 first = ipport_hifirstauto; /* sysctl */
783 last = ipport_hilastauto;
784 lastport = &portinfo->lasthi;
785 } else if (inp->inp_flags & INP_LOWPORT) {
786 if (cred &&
787 (error =
788 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
789 inp->inp_laddr.s_addr = INADDR_ANY;
790 return (error);
792 first = ipport_lowfirstauto; /* 1023 */
793 last = ipport_lowlastauto; /* 600 */
794 lastport = &portinfo->lastlow;
795 } else {
796 first = ipport_firstauto; /* sysctl */
797 last = ipport_lastauto;
798 lastport = &portinfo->lastport;
801 /* This could happen on loopback interface */
802 #define IS_SELFCONNECT(inp, lport, sin) \
803 (__predict_false((sin)->sin_port == (lport) && \
804 (sin)->sin_addr.s_addr == (inp)->inp_laddr.s_addr))
807 * Simple check to ensure all ports are not used up causing
808 * a deadlock here.
810 * We split the two cases (up and down) so that the direction
811 * is not being tested on each round of the loop.
813 hash_count = hash_count0;
814 if (first > last) {
816 * counting down
818 in_pcbportrange(&first, &last, portinfo->offset, step);
819 count = ((first - last) / step) + hash_count;
821 for (;;) {
822 if (count-- < 0) { /* completely used? */
823 error = EADDRNOTAVAIL;
824 break;
827 lport = in_pcblastport_down(lastport, first, last,
828 step);
829 KKASSERT((lport % pcbinfo->portinfo_cnt) ==
830 portinfo->offset);
831 lport = htons(lport);
832 if (IS_SELFCONNECT(inp, lport, sin)) {
833 if (!selfconn) {
834 ++count; /* don't count this try */
835 selfconn = 1;
837 continue;
840 if (hash_count) {
841 --hash_count;
842 hash = hash_base ^
843 toeplitz_piecemeal_port(lport);
844 if (netisr_hashcpu(hash) != hash_cpu &&
845 hash_count)
846 continue;
849 if (in_pcbporthash_update4(portinfo,
850 inp, lport, sin, cred)) {
851 error = 0;
852 break;
855 } else {
857 * counting up
859 in_pcbportrange(&last, &first, portinfo->offset, step);
860 count = ((last - first) / step) + hash_count;
862 for (;;) {
863 if (count-- < 0) { /* completely used? */
864 error = EADDRNOTAVAIL;
865 break;
868 lport = in_pcblastport_up(lastport, first, last, step);
869 KKASSERT((lport % pcbinfo->portinfo_cnt) ==
870 portinfo->offset);
871 lport = htons(lport);
872 if (IS_SELFCONNECT(inp, lport, sin)) {
873 if (!selfconn) {
874 ++count; /* don't count this try */
875 selfconn = 1;
877 continue;
880 if (hash_count) {
881 --hash_count;
882 hash = hash_base ^
883 toeplitz_piecemeal_port(lport);
884 if (netisr_hashcpu(hash) != hash_cpu &&
885 hash_count)
886 continue;
889 if (in_pcbporthash_update4(portinfo,
890 inp, lport, sin, cred)) {
891 error = 0;
892 break;
897 #undef IS_SELFCONNECT
899 if (error) {
900 /* Try next portinfo */
901 portinfo_idx++;
902 portinfo_idx %= pcbinfo->portinfo_cnt;
903 if (portinfo_idx != portinfo_first)
904 goto loop;
905 inp->inp_laddr.s_addr = INADDR_ANY;
907 return error;
911 * Transform old in_pcbconnect() into an inner subroutine for new
912 * in_pcbconnect(): Do some validity-checking on the remote
913 * address (in mbuf 'nam') and then determine local host address
914 * (i.e., which interface) to use to access that remote host.
916 * This preserves definition of in_pcbconnect(), while supporting a
917 * slightly different version for T/TCP. (This is more than
918 * a bit of a kludge, but cleaning up the internal interfaces would
919 * have forced minor changes in every protocol).
922 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
923 struct sockaddr_in **plocal_sin, struct thread *td, int find)
925 struct in_ifaddr *ia;
926 struct ucred *cred = NULL;
927 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
928 struct sockaddr *jsin;
929 int jailed = 0, alloc_route = 0;
931 if (nam->sa_len != sizeof *sin)
932 return (EINVAL);
933 if (sin->sin_family != AF_INET)
934 return (EAFNOSUPPORT);
935 if (sin->sin_port == 0)
936 return (EADDRNOTAVAIL);
937 if (td && td->td_proc && td->td_proc->p_ucred)
938 cred = td->td_proc->p_ucred;
939 if (cred && cred->cr_prison)
940 jailed = 1;
941 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
942 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
944 * If the destination address is INADDR_ANY,
945 * use the primary local address.
946 * If the supplied address is INADDR_BROADCAST,
947 * and the primary interface supports broadcast,
948 * choose the broadcast address for that interface.
950 if (sin->sin_addr.s_addr == INADDR_ANY)
951 sin->sin_addr = IA_SIN(ia)->sin_addr;
952 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
953 (ia->ia_ifp->if_flags & IFF_BROADCAST))
954 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
956 if (find) {
957 struct route *ro;
959 ia = NULL;
961 * If route is known or can be allocated now,
962 * our src addr is taken from the i/f, else punt.
963 * Note that we should check the address family of the cached
964 * destination, in case of sharing the cache with IPv6.
966 ro = &inp->inp_route;
967 if (ro->ro_rt &&
968 (!(ro->ro_rt->rt_flags & RTF_UP) ||
969 ro->ro_dst.sa_family != AF_INET ||
970 satosin(&ro->ro_dst)->sin_addr.s_addr !=
971 sin->sin_addr.s_addr ||
972 inp->inp_socket->so_options & SO_DONTROUTE)) {
973 RTFREE(ro->ro_rt);
974 ro->ro_rt = NULL;
976 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
977 (ro->ro_rt == NULL ||
978 ro->ro_rt->rt_ifp == NULL)) {
979 /* No route yet, so try to acquire one */
980 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
981 ro->ro_dst.sa_family = AF_INET;
982 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
983 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
984 sin->sin_addr;
985 rtalloc(ro);
986 alloc_route = 1;
989 * If we found a route, use the address
990 * corresponding to the outgoing interface
991 * unless it is the loopback (in case a route
992 * to our address on another net goes to loopback).
994 if (ro->ro_rt &&
995 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
996 if (jailed) {
997 if (jailed_ip(cred->cr_prison,
998 ro->ro_rt->rt_ifa->ifa_addr)) {
999 ia = ifatoia(ro->ro_rt->rt_ifa);
1001 } else {
1002 ia = ifatoia(ro->ro_rt->rt_ifa);
1005 if (ia == NULL) {
1006 u_short fport = sin->sin_port;
1008 sin->sin_port = 0;
1009 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
1010 if (ia && jailed && !jailed_ip(cred->cr_prison,
1011 sintosa(&ia->ia_addr)))
1012 ia = NULL;
1013 if (ia == NULL)
1014 ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
1015 if (ia && jailed && !jailed_ip(cred->cr_prison,
1016 sintosa(&ia->ia_addr)))
1017 ia = NULL;
1018 sin->sin_port = fport;
1019 if (ia == NULL &&
1020 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1021 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
1022 if (ia && jailed && !jailed_ip(cred->cr_prison,
1023 sintosa(&ia->ia_addr)))
1024 ia = NULL;
1026 if (!jailed && ia == NULL)
1027 goto fail;
1030 * If the destination address is multicast and an outgoing
1031 * interface has been set as a multicast option, use the
1032 * address of that interface as our source address.
1034 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
1035 inp->inp_moptions != NULL) {
1036 struct ip_moptions *imo;
1037 struct ifnet *ifp;
1039 imo = inp->inp_moptions;
1040 if ((ifp = imo->imo_multicast_ifp) != NULL) {
1041 struct in_ifaddr_container *iac;
1043 ia = NULL;
1044 TAILQ_FOREACH(iac,
1045 &in_ifaddrheads[mycpuid], ia_link) {
1046 if (iac->ia->ia_ifp == ifp) {
1047 ia = iac->ia;
1048 break;
1051 if (ia == NULL)
1052 goto fail;
1056 * Don't do pcblookup call here; return interface in plocal_sin
1057 * and exit to caller, that will do the lookup.
1059 if (ia == NULL && jailed) {
1060 if ((jsin = prison_get_nonlocal(
1061 cred->cr_prison, AF_INET, NULL)) != NULL ||
1062 (jsin = prison_get_local(
1063 cred->cr_prison, AF_INET, NULL)) != NULL) {
1064 *plocal_sin = satosin(jsin);
1065 } else {
1066 /* IPv6 only Jail */
1067 goto fail;
1069 } else {
1070 *plocal_sin = &ia->ia_addr;
1073 return (0);
1074 fail:
1075 if (alloc_route)
1076 in_pcbresetroute(inp);
1077 return (EADDRNOTAVAIL);
1081 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
1082 struct sockaddr_in **plocal_sin, struct thread *td)
1084 return in_pcbladdr_find(inp, nam, plocal_sin, td,
1085 (inp->inp_laddr.s_addr == INADDR_ANY));
1089 * Outer subroutine:
1090 * Connect from a socket to a specified address.
1091 * Both address and port must be specified in argument sin.
1092 * If don't have a local address for this socket yet,
1093 * then pick one.
1096 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
1098 struct sockaddr_in *if_sin;
1099 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1100 int error;
1102 if_sin = NULL; /* avoid gcc warnings */
1104 /* Call inner routine to assign local interface address. */
1105 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
1106 return (error);
1108 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port,
1109 inp->inp_laddr.s_addr ?
1110 inp->inp_laddr : if_sin->sin_addr,
1111 inp->inp_lport, FALSE, NULL) != NULL) {
1112 return (EADDRINUSE);
1114 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1115 if (inp->inp_lport == 0) {
1116 error = in_pcbbind(inp, NULL, td);
1117 if (error)
1118 return (error);
1120 inp->inp_laddr = if_sin->sin_addr;
1122 inp->inp_faddr = sin->sin_addr;
1123 inp->inp_fport = sin->sin_port;
1124 in_pcbinsconnhash(inp);
1125 return (0);
1128 void
1129 in_pcbdisconnect(struct inpcb *inp)
1132 in_pcbremconnhash(inp);
1133 inp->inp_faddr.s_addr = INADDR_ANY;
1134 inp->inp_fport = 0;
1137 void
1138 in_pcbdetach(struct inpcb *inp)
1140 struct socket *so = inp->inp_socket;
1141 struct inpcbinfo *ipi = inp->inp_pcbinfo;
1143 #ifdef IPSEC
1144 ipsec4_delete_pcbpolicy(inp);
1145 #endif /*IPSEC*/
1146 inp->inp_gencnt = ++ipi->ipi_gencnt;
1147 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
1148 in_pcbremlists(inp);
1149 so->so_pcb = NULL;
1150 sofree(so); /* remove pcb ref */
1151 if (inp->inp_options)
1152 m_free(inp->inp_options);
1153 if (inp->inp_route.ro_rt)
1154 rtfree(inp->inp_route.ro_rt);
1155 ip_freemoptions(inp->inp_moptions);
1156 kfree(inp, M_PCB);
1160 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP
1161 * socket received RST.
1163 static int
1164 in_setsockaddr(struct socket *so, struct sockaddr **nam)
1166 struct inpcb *inp;
1167 struct sockaddr_in *sin;
1169 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
1170 inp = so->so_pcb;
1171 if (!inp)
1172 return (ECONNRESET);
1174 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1175 sin->sin_family = AF_INET;
1176 sin->sin_len = sizeof *sin;
1177 sin->sin_port = inp->inp_lport;
1178 sin->sin_addr = inp->inp_laddr;
1180 *nam = (struct sockaddr *)sin;
1181 return (0);
1184 void
1185 in_setsockaddr_dispatch(netmsg_t msg)
1187 int error;
1189 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1190 lwkt_replymsg(&msg->lmsg, error);
1194 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP
1195 * socket received RST.
1198 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1200 struct inpcb *inp;
1201 struct sockaddr_in *sin;
1203 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
1204 inp = so->so_pcb;
1205 if (!inp)
1206 return (ECONNRESET);
1208 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1209 sin->sin_family = AF_INET;
1210 sin->sin_len = sizeof *sin;
1211 sin->sin_port = inp->inp_fport;
1212 sin->sin_addr = inp->inp_faddr;
1214 *nam = (struct sockaddr *)sin;
1215 return (0);
1218 void
1219 in_setpeeraddr_dispatch(netmsg_t msg)
1221 int error;
1223 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1224 lwkt_replymsg(&msg->lmsg, error);
1227 void
1228 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err,
1229 inp_notify_t notify)
1231 struct inpcb *inp, *marker;
1233 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1234 ("not in the correct netisr"));
1235 marker = &in_pcbmarkers[mycpuid];
1238 * NOTE:
1239 * - If INP_PLACEMARKER is set we must ignore the rest of the
1240 * structure and skip it.
1241 * - It is safe to nuke inpcbs here, since we are in their own
1242 * netisr.
1244 GET_PCBINFO_TOKEN(pcbinfo);
1246 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1247 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1248 LIST_REMOVE(marker, inp_list);
1249 LIST_INSERT_AFTER(inp, marker, inp_list);
1251 if (inp->inp_flags & INP_PLACEMARKER)
1252 continue;
1253 #ifdef INET6
1254 if (!INP_ISIPV4(inp))
1255 continue;
1256 #endif
1257 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1258 inp->inp_socket == NULL)
1259 continue;
1260 (*notify)(inp, err); /* can remove inp from list! */
1262 LIST_REMOVE(marker, inp_list);
1264 REL_PCBINFO_TOKEN(pcbinfo);
1267 void
1268 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1270 struct inpcb *inp, *marker;
1273 * We only need to make sure that we are in netisr0, where all
1274 * multicast operation happen. We could check inpcbinfo which
1275 * does not belong to netisr0 by holding the inpcbinfo's token.
1276 * In this case, the pcbinfo must be able to be shared, i.e.
1277 * pcbinfo->infotoken is not NULL.
1279 ASSERT_IN_NETISR(0);
1280 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL,
1281 ("pcbinfo could not be shared"));
1284 * Get a marker for the current netisr (netisr0).
1286 * It is possible that the multicast address deletion blocks,
1287 * which could cause temporary token releasing. So we use
1288 * inpcb marker here to get a coherent view of the inpcb list.
1290 * While, on the other hand, moptions are only added and deleted
1291 * in netisr0, so we would not see staled moption or miss moption
1292 * even if the token was released due to the blocking multicast
1293 * address deletion.
1295 marker = &in_pcbmarkers[mycpuid];
1297 GET_PCBINFO_TOKEN(pcbinfo);
1299 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1300 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1301 struct ip_moptions *imo;
1303 LIST_REMOVE(marker, inp_list);
1304 LIST_INSERT_AFTER(inp, marker, inp_list);
1306 if (inp->inp_flags & INP_PLACEMARKER)
1307 continue;
1308 imo = inp->inp_moptions;
1309 if (INP_ISIPV4(inp) && imo != NULL) {
1310 int i, gap;
1313 * Unselect the outgoing interface if it is being
1314 * detached.
1316 if (imo->imo_multicast_ifp == ifp)
1317 imo->imo_multicast_ifp = NULL;
1320 * Drop multicast group membership if we joined
1321 * through the interface being detached.
1323 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1324 i++) {
1325 if (imo->imo_membership[i]->inm_ifp == ifp) {
1327 * NOTE:
1328 * This could block and the pcbinfo
1329 * token could be passively released.
1331 in_delmulti(imo->imo_membership[i]);
1332 gap++;
1333 } else if (gap != 0)
1334 imo->imo_membership[i - gap] =
1335 imo->imo_membership[i];
1337 imo->imo_num_memberships -= gap;
1340 LIST_REMOVE(marker, inp_list);
1342 REL_PCBINFO_TOKEN(pcbinfo);
1346 * Check for alternatives when higher level complains
1347 * about service problems. For now, invalidate cached
1348 * routing information. If the route was created dynamically
1349 * (by a redirect), time to try a default gateway again.
1351 void
1352 in_losing(struct inpcb *inp)
1354 struct rtentry *rt;
1355 struct rt_addrinfo rtinfo;
1357 if ((rt = inp->inp_route.ro_rt)) {
1358 bzero(&rtinfo, sizeof(struct rt_addrinfo));
1359 rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1360 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1361 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1362 rtinfo.rti_flags = rt->rt_flags;
1363 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1364 if (rt->rt_flags & RTF_DYNAMIC) {
1365 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1366 rt_mask(rt), rt->rt_flags, NULL);
1368 inp->inp_route.ro_rt = NULL;
1369 rtfree(rt);
1371 * A new route can be allocated
1372 * the next time output is attempted.
1378 * After a routing change, flush old routing
1379 * and allocate a (hopefully) better one.
1381 void
1382 in_rtchange(struct inpcb *inp, int err)
1384 if (inp->inp_route.ro_rt) {
1385 rtfree(inp->inp_route.ro_rt);
1386 inp->inp_route.ro_rt = NULL;
1388 * A new route can be allocated the next time
1389 * output is attempted.
1395 * Lookup a PCB based on the local address and port.
1397 static struct inpcb *
1398 in_pcblookup_local(struct inpcbporthead *porthash, struct in_addr laddr,
1399 u_int lport_arg, int wild_okay, struct ucred *cred)
1401 struct inpcb *inp;
1402 int matchwild = 3, wildcard;
1403 u_short lport = lport_arg;
1404 struct inpcbport *phd;
1405 struct inpcb *match = NULL;
1408 * If the porthashbase is shared across several cpus, it must
1409 * have been locked.
1411 ASSERT_PORTHASH_TOKEN_HELD(porthash);
1414 * Best fit PCB lookup.
1416 * First see if this local port is in use by looking on the
1417 * port hash list.
1419 LIST_FOREACH(phd, porthash, phd_hash) {
1420 if (phd->phd_port == lport)
1421 break;
1423 if (phd != NULL) {
1425 * Port is in use by one or more PCBs. Look for best
1426 * fit.
1428 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1429 wildcard = 0;
1430 #ifdef INET6
1431 if (!INP_ISIPV4(inp))
1432 continue;
1433 #endif
1434 if (inp->inp_faddr.s_addr != INADDR_ANY)
1435 wildcard++;
1436 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1437 if (laddr.s_addr == INADDR_ANY)
1438 wildcard++;
1439 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1440 continue;
1441 } else {
1442 if (laddr.s_addr != INADDR_ANY)
1443 wildcard++;
1445 if (wildcard && !wild_okay)
1446 continue;
1447 if (wildcard < matchwild &&
1448 (cred == NULL ||
1449 cred->cr_prison ==
1450 inp->inp_socket->so_cred->cr_prison)) {
1451 match = inp;
1452 matchwild = wildcard;
1453 if (matchwild == 0) {
1454 break;
1459 return (match);
1462 struct inpcb *
1463 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1464 const struct inpcb *inp)
1466 const struct inp_localgrphead *hdr;
1467 const struct inp_localgroup *grp;
1468 int i;
1470 if (pcbinfo->localgrphashbase == NULL)
1471 return NULL;
1473 GET_PCBINFO_TOKEN(pcbinfo);
1475 hdr = &pcbinfo->localgrphashbase[
1476 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1478 LIST_FOREACH(grp, hdr, il_list) {
1479 if (grp->il_af == inp->inp_af &&
1480 grp->il_lport == inp->inp_lport &&
1481 memcmp(&grp->il_dependladdr,
1482 &inp->inp_inc.inc_ie.ie_dependladdr,
1483 sizeof(grp->il_dependladdr)) == 0) {
1484 break;
1487 if (grp == NULL || grp->il_inpcnt == 1) {
1488 REL_PCBINFO_TOKEN(pcbinfo);
1489 return NULL;
1492 KASSERT(grp->il_inpcnt >= 2,
1493 ("invalid localgroup inp count %d", grp->il_inpcnt));
1494 for (i = 0; i < grp->il_inpcnt; ++i) {
1495 if (grp->il_inp[i] == inp) {
1496 int last = grp->il_inpcnt - 1;
1498 if (i == last)
1499 last = grp->il_inpcnt - 2;
1500 REL_PCBINFO_TOKEN(pcbinfo);
1501 return grp->il_inp[last];
1504 REL_PCBINFO_TOKEN(pcbinfo);
1505 return NULL;
1508 static struct inpcb *
1509 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1510 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1512 struct inpcb *local_wild = NULL;
1513 const struct inp_localgrphead *hdr;
1514 const struct inp_localgroup *grp;
1516 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1518 hdr = &pcbinfo->localgrphashbase[
1519 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1522 * Order of socket selection:
1523 * 1. non-wild.
1524 * 2. wild.
1526 * NOTE: Local group does not contain jailed sockets
1528 LIST_FOREACH(grp, hdr, il_list) {
1529 #ifdef INET6
1530 if (grp->il_af != AF_INET)
1531 continue;
1532 #endif
1533 if (grp->il_lport == lport) {
1534 int idx;
1537 * Modulo-N is used here, which greatly reduces
1538 * completion queue token contention, thus more
1539 * cpu time is saved.
1541 idx = netisr_hashlsb(pkt_hash) % grp->il_inpcnt;
1542 if (grp->il_laddr.s_addr == laddr.s_addr)
1543 return grp->il_inp[idx];
1544 else if (grp->il_laddr.s_addr == INADDR_ANY)
1545 local_wild = grp->il_inp[idx];
1548 if (local_wild != NULL)
1549 return local_wild;
1550 return NULL;
1554 * Lookup PCB in hash list.
1556 struct inpcb *
1557 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1558 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1559 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1561 struct inpcbhead *head;
1562 struct inpcb *inp, *jinp=NULL;
1563 u_short fport = fport_arg, lport = lport_arg;
1566 * First look for an exact match.
1568 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1569 laddr.s_addr, lport, pcbinfo->hashmask)];
1570 LIST_FOREACH(inp, head, inp_hash) {
1571 #ifdef INET6
1572 if (!INP_ISIPV4(inp))
1573 continue;
1574 #endif
1575 if (in_hosteq(inp->inp_faddr, faddr) &&
1576 in_hosteq(inp->inp_laddr, laddr) &&
1577 inp->inp_fport == fport && inp->inp_lport == lport) {
1578 /* found */
1579 if (inp->inp_socket == NULL ||
1580 inp->inp_socket->so_cred->cr_prison == NULL) {
1581 return (inp);
1582 } else {
1583 if (jinp == NULL)
1584 jinp = inp;
1588 if (jinp != NULL)
1589 return (jinp);
1591 if (wildcard) {
1592 struct inpcb *local_wild = NULL;
1593 struct inpcb *jinp_wild = NULL;
1594 struct inpcontainer *ic;
1595 struct inpcontainerhead *chead;
1596 struct sockaddr_in jsin;
1597 struct ucred *cred;
1599 GET_PCBINFO_TOKEN(pcbinfo);
1602 * Check local group first
1604 if (pcbinfo->localgrphashbase != NULL &&
1605 m != NULL && (m->m_flags & M_HASH) &&
1606 !(ifp && ifp->if_type == IFT_FAITH)) {
1607 inp = inp_localgroup_lookup(pcbinfo,
1608 laddr, lport, m->m_pkthdr.hash);
1609 if (inp != NULL) {
1610 REL_PCBINFO_TOKEN(pcbinfo);
1611 return inp;
1616 * Order of socket selection:
1617 * 1. non-jailed, non-wild.
1618 * 2. non-jailed, wild.
1619 * 3. jailed, non-wild.
1620 * 4. jailed, wild.
1622 jsin.sin_family = AF_INET;
1623 chead = &pcbinfo->wildcardhashbase[
1624 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1625 LIST_FOREACH(ic, chead, ic_list) {
1626 inp = ic->ic_inp;
1627 if (inp->inp_flags & INP_PLACEMARKER)
1628 continue;
1630 jsin.sin_addr.s_addr = laddr.s_addr;
1631 #ifdef INET6
1632 if (!INP_ISIPV4(inp))
1633 continue;
1634 #endif
1635 if (inp->inp_socket != NULL)
1636 cred = inp->inp_socket->so_cred;
1637 else
1638 cred = NULL;
1639 if (cred != NULL && jailed(cred)) {
1640 if (jinp != NULL)
1641 continue;
1642 else
1643 if (!jailed_ip(cred->cr_prison,
1644 (struct sockaddr *)&jsin))
1645 continue;
1647 if (inp->inp_lport == lport) {
1648 if (ifp && ifp->if_type == IFT_FAITH &&
1649 !(inp->inp_flags & INP_FAITH))
1650 continue;
1651 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1652 if (cred != NULL && jailed(cred)) {
1653 jinp = inp;
1654 } else {
1655 REL_PCBINFO_TOKEN(pcbinfo);
1656 return (inp);
1659 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1660 if (cred != NULL && jailed(cred))
1661 jinp_wild = inp;
1662 else
1663 local_wild = inp;
1668 REL_PCBINFO_TOKEN(pcbinfo);
1670 if (local_wild != NULL)
1671 return (local_wild);
1672 if (jinp != NULL)
1673 return (jinp);
1674 return (jinp_wild);
1678 * Not found.
1680 return (NULL);
1683 struct inpcb *
1684 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1685 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1686 boolean_t wildcard, struct ifnet *ifp)
1688 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1689 laddr, lport_arg, wildcard, ifp, NULL);
1693 * Insert PCB into connection hash table.
1695 void
1696 in_pcbinsconnhash(struct inpcb *inp)
1698 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1699 struct inpcbhead *bucket;
1700 u_int32_t hashkey_faddr, hashkey_laddr;
1702 #ifdef INET6
1703 if (INP_ISIPV6(inp)) {
1704 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1705 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1706 } else {
1707 #endif
1708 hashkey_faddr = inp->inp_faddr.s_addr;
1709 hashkey_laddr = inp->inp_laddr.s_addr;
1710 #ifdef INET6
1712 #endif
1714 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1715 ("not in the correct netisr"));
1716 ASSERT_INP_NOTINHASH(inp);
1717 inp->inp_flags |= INP_CONNECTED;
1720 * Insert into the connection hash table.
1722 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1723 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1724 LIST_INSERT_HEAD(bucket, inp, inp_hash);
1728 * Remove PCB from connection hash table.
1730 void
1731 in_pcbremconnhash(struct inpcb *inp)
1733 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo;
1735 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1736 ("not in the correct netisr"));
1737 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1739 LIST_REMOVE(inp, inp_hash);
1740 inp->inp_flags &= ~INP_CONNECTED;
1744 * Insert PCB into port hash table.
1746 void
1747 in_pcbinsporthash(struct inpcbporthead *pcbporthash, struct inpcb *inp)
1749 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1750 struct inpcbport *phd;
1753 * If the porthashbase is shared across several cpus, it must
1754 * have been locked.
1756 ASSERT_PORTHASH_TOKEN_HELD(pcbporthash);
1759 * Insert into the port hash table.
1762 /* Go through port list and look for a head for this lport. */
1763 LIST_FOREACH(phd, pcbporthash, phd_hash) {
1764 if (phd->phd_port == inp->inp_lport)
1765 break;
1768 /* If none exists, use saved one and tack it on. */
1769 if (phd == NULL) {
1770 KKASSERT(pcbinfo->portsave != NULL);
1771 phd = pcbinfo->portsave;
1772 pcbinfo->portsave = NULL;
1773 phd->phd_port = inp->inp_lport;
1774 LIST_INIT(&phd->phd_pcblist);
1775 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1778 inp->inp_porthash = pcbporthash;
1779 inp->inp_phd = phd;
1780 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1783 * Malloc one inpcbport for later use. It is safe to use
1784 * "wait" malloc here (port token would be released, if
1785 * malloc ever blocked), since all changes to the porthash
1786 * are done.
1788 if (pcbinfo->portsave == NULL) {
1789 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1790 M_PCB, M_INTWAIT | M_ZERO);
1794 void
1795 in_pcbinsporthash_lport(struct inpcb *inp)
1797 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1798 struct inpcbportinfo *portinfo;
1799 struct inpcbporthead *porthash;
1800 u_short lport_ho;
1802 /* Locate the proper portinfo based on lport */
1803 lport_ho = ntohs(inp->inp_lport);
1804 portinfo = &pcbinfo->portinfo[lport_ho % pcbinfo->portinfo_cnt];
1805 KKASSERT((lport_ho % pcbinfo->portinfo_cnt) == portinfo->offset);
1807 porthash = in_pcbporthash_head(portinfo, inp->inp_lport);
1808 GET_PORTHASH_TOKEN(porthash);
1809 in_pcbinsporthash(porthash, inp);
1810 REL_PORTHASH_TOKEN(porthash);
1813 void
1814 in_pcbremporthash(struct inpcb *inp)
1816 struct inpcbporthead *porthash;
1817 struct inpcbport *phd;
1819 if (inp->inp_phd == NULL)
1820 return;
1821 KASSERT(inp->inp_lport != 0, ("inpcb has no lport"));
1823 porthash = inp->inp_porthash;
1824 KASSERT(porthash != NULL, ("no porthash"));
1826 GET_PORTHASH_TOKEN(porthash);
1828 phd = inp->inp_phd;
1829 LIST_REMOVE(inp, inp_portlist);
1830 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1831 LIST_REMOVE(phd, phd_hash);
1832 kfree(phd, M_PCB);
1835 REL_PORTHASH_TOKEN(porthash);
1837 inp->inp_phd = NULL;
1838 /* NOTE: Don't whack inp_lport, which may be used later */
1841 static struct inp_localgroup *
1842 inp_localgroup_alloc(u_char af, uint16_t port,
1843 const union in_dependaddr *addr, int size)
1845 struct inp_localgroup *grp;
1847 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
1848 M_TEMP, M_INTWAIT | M_ZERO);
1849 grp->il_af = af;
1850 grp->il_lport = port;
1851 grp->il_dependladdr = *addr;
1852 grp->il_inpsiz = size;
1854 return grp;
1857 static void
1858 inp_localgroup_free(struct inp_localgroup *grp)
1860 kfree(grp, M_TEMP);
1863 static void
1864 inp_localgroup_destroy(struct inp_localgroup *grp)
1866 LIST_REMOVE(grp, il_list);
1867 inp_localgroup_free(grp);
1870 static void
1871 inp_localgroup_copy(struct inp_localgroup *grp,
1872 const struct inp_localgroup *old_grp)
1874 int i;
1876 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
1877 ("invalid new local group size %d and old local group count %d",
1878 grp->il_inpsiz, old_grp->il_inpcnt));
1879 for (i = 0; i < old_grp->il_inpcnt; ++i)
1880 grp->il_inp[i] = old_grp->il_inp[i];
1881 grp->il_inpcnt = old_grp->il_inpcnt;
1884 static void
1885 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1887 struct inp_localgrphead *hdr;
1888 struct inp_localgroup *grp, *grp_alloc = NULL;
1889 struct ucred *cred;
1890 int i, idx;
1892 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1894 if (pcbinfo->localgrphashbase == NULL)
1895 return;
1898 * XXX don't allow jailed socket to join local group
1900 if (inp->inp_socket != NULL)
1901 cred = inp->inp_socket->so_cred;
1902 else
1903 cred = NULL;
1904 if (cred != NULL && jailed(cred))
1905 return;
1907 hdr = &pcbinfo->localgrphashbase[
1908 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1910 again:
1911 LIST_FOREACH(grp, hdr, il_list) {
1912 if (grp->il_af == inp->inp_af &&
1913 grp->il_lport == inp->inp_lport &&
1914 memcmp(&grp->il_dependladdr,
1915 &inp->inp_inc.inc_ie.ie_dependladdr,
1916 sizeof(grp->il_dependladdr)) == 0) {
1917 break;
1920 if (grp == NULL) {
1922 * Create a new local group
1924 if (grp_alloc == NULL) {
1925 grp_alloc = inp_localgroup_alloc(inp->inp_af,
1926 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
1927 INP_LOCALGROUP_SIZMIN);
1929 * Local group allocation could block and the
1930 * local group w/ the same property might have
1931 * been added by others when we were blocked;
1932 * check again.
1934 goto again;
1935 } else {
1936 /* Local group has been allocated; link it */
1937 grp = grp_alloc;
1938 grp_alloc = NULL;
1939 LIST_INSERT_HEAD(hdr, grp, il_list);
1941 } else if (grp->il_inpcnt == grp->il_inpsiz) {
1942 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
1943 static int limit_logged = 0;
1945 if (!limit_logged) {
1946 limit_logged = 1;
1947 kprintf("local group port %d, "
1948 "limit reached\n", ntohs(grp->il_lport));
1950 if (grp_alloc != NULL) {
1952 * This would happen if the local group
1953 * w/ the same property was expanded when
1954 * our local group allocation blocked.
1956 inp_localgroup_free(grp_alloc);
1958 return;
1962 * Expand this local group
1964 if (grp_alloc == NULL ||
1965 grp->il_inpcnt >= grp_alloc->il_inpsiz) {
1966 if (grp_alloc != NULL)
1967 inp_localgroup_free(grp_alloc);
1968 grp_alloc = inp_localgroup_alloc(grp->il_af,
1969 grp->il_lport, &grp->il_dependladdr,
1970 grp->il_inpsiz * 2);
1972 * Local group allocation could block and the
1973 * local group w/ the same property might have
1974 * been expanded by others when we were blocked;
1975 * check again.
1977 goto again;
1981 * Save the old local group, link the new one, and then
1982 * destroy the old local group
1984 inp_localgroup_copy(grp_alloc, grp);
1985 LIST_INSERT_HEAD(hdr, grp_alloc, il_list);
1986 inp_localgroup_destroy(grp);
1988 grp = grp_alloc;
1989 grp_alloc = NULL;
1990 } else {
1992 * Found the local group
1994 if (grp_alloc != NULL) {
1996 * This would happen if the local group w/ the
1997 * same property was added or expanded when our
1998 * local group allocation blocked.
2000 inp_localgroup_free(grp_alloc);
2001 grp_alloc = NULL;
2005 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
2006 ("invalid local group size %d and count %d",
2007 grp->il_inpsiz, grp->il_inpcnt));
2010 * Keep the local group sorted by the inpcb local group index
2011 * in ascending order.
2013 * This eases the multi-process userland application which uses
2014 * SO_REUSEPORT sockets and binds process to the owner cpu of
2015 * the SO_REUSEPORT socket:
2016 * If we didn't sort the local group by the inpcb local group
2017 * index and one of the process owning an inpcb in this local
2018 * group restarted, e.g. crashed and restarted by watchdog,
2019 * other processes owning a inpcb in this local group would have
2020 * to detect that event, refetch its socket's owner cpu, and
2021 * re-bind.
2023 idx = grp->il_inpcnt;
2024 for (i = 0; i < idx; ++i) {
2025 struct inpcb *oinp = grp->il_inp[i];
2027 if (oinp->inp_lgrpindex > i) {
2028 if (inp->inp_lgrpindex < 0) {
2029 inp->inp_lgrpindex = i;
2030 } else if (inp->inp_lgrpindex != i) {
2031 if (bootverbose) {
2032 kprintf("inp %p: grpidx %d, "
2033 "assigned to %d, cpu%d\n",
2034 inp, inp->inp_lgrpindex, i,
2035 mycpuid);
2038 grp->il_inp[i] = inp;
2040 /* Pull down inpcbs */
2041 for (; i < grp->il_inpcnt; ++i) {
2042 struct inpcb *oinp1 = grp->il_inp[i + 1];
2044 grp->il_inp[i + 1] = oinp;
2045 oinp = oinp1;
2047 grp->il_inpcnt++;
2048 return;
2052 if (inp->inp_lgrpindex < 0) {
2053 inp->inp_lgrpindex = idx;
2054 } else if (inp->inp_lgrpindex != idx) {
2055 if (bootverbose) {
2056 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n",
2057 inp, inp->inp_lgrpindex, idx, mycpuid);
2060 grp->il_inp[idx] = inp;
2061 grp->il_inpcnt++;
2064 void
2065 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2067 struct inpcontainer *ic;
2068 struct inpcontainerhead *bucket;
2070 GET_PCBINFO_TOKEN(pcbinfo);
2072 in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
2074 bucket = &pcbinfo->wildcardhashbase[
2075 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2077 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
2078 ic->ic_inp = inp;
2079 LIST_INSERT_HEAD(bucket, ic, ic_list);
2081 REL_PCBINFO_TOKEN(pcbinfo);
2085 * Insert PCB into wildcard hash table.
2087 void
2088 in_pcbinswildcardhash(struct inpcb *inp)
2090 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2092 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2093 ("not in correct netisr"));
2094 ASSERT_INP_NOTINHASH(inp);
2095 inp->inp_flags |= INP_WILDCARD;
2097 in_pcbinswildcardhash_oncpu(inp, pcbinfo);
2100 static void
2101 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2103 struct inp_localgrphead *hdr;
2104 struct inp_localgroup *grp;
2106 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
2108 if (pcbinfo->localgrphashbase == NULL)
2109 return;
2111 hdr = &pcbinfo->localgrphashbase[
2112 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
2114 LIST_FOREACH(grp, hdr, il_list) {
2115 int i;
2117 for (i = 0; i < grp->il_inpcnt; ++i) {
2118 if (grp->il_inp[i] != inp)
2119 continue;
2121 if (grp->il_inpcnt == 1) {
2122 /* Destroy this local group */
2123 inp_localgroup_destroy(grp);
2124 } else {
2125 /* Pull up inpcbs */
2126 for (; i + 1 < grp->il_inpcnt; ++i)
2127 grp->il_inp[i] = grp->il_inp[i + 1];
2128 grp->il_inpcnt--;
2130 return;
2135 void
2136 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2138 struct inpcontainer *ic;
2139 struct inpcontainerhead *head;
2141 GET_PCBINFO_TOKEN(pcbinfo);
2143 in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
2145 /* find bucket */
2146 head = &pcbinfo->wildcardhashbase[
2147 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2149 LIST_FOREACH(ic, head, ic_list) {
2150 if (ic->ic_inp == inp)
2151 goto found;
2153 REL_PCBINFO_TOKEN(pcbinfo);
2154 return; /* not found! */
2156 found:
2157 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */
2158 REL_PCBINFO_TOKEN(pcbinfo);
2159 kfree(ic, M_TEMP); /* deallocate container */
2163 * Remove PCB from wildcard hash table.
2165 void
2166 in_pcbremwildcardhash(struct inpcb *inp)
2168 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2170 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2171 ("not in correct netisr"));
2172 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
2174 in_pcbremwildcardhash_oncpu(inp, pcbinfo);
2175 inp->inp_lgrpindex = -1;
2176 inp->inp_flags &= ~INP_WILDCARD;
2180 * Remove PCB from various lists.
2182 void
2183 in_pcbremlists(struct inpcb *inp)
2185 in_pcbremporthash(inp);
2186 if (inp->inp_flags & INP_WILDCARD) {
2187 in_pcbremwildcardhash(inp);
2188 } else if (inp->inp_flags & INP_CONNECTED) {
2189 in_pcbremconnhash(inp);
2192 if (inp->inp_flags & INP_ONLIST)
2193 in_pcbofflist(inp);
2197 prison_xinpcb(struct thread *td, struct inpcb *inp)
2199 struct ucred *cr;
2201 if (td->td_proc == NULL)
2202 return (0);
2203 cr = td->td_proc->p_ucred;
2204 if (cr->cr_prison == NULL)
2205 return (0);
2206 if (inp->inp_socket && inp->inp_socket->so_cred &&
2207 inp->inp_socket->so_cred->cr_prison &&
2208 cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
2209 return (0);
2210 return (1);
2214 in_pcblist_range(SYSCTL_HANDLER_ARGS)
2216 struct inpcbinfo *pcbinfo_arr = arg1;
2217 int pcbinfo_arrlen = arg2;
2218 struct inpcb *marker;
2219 int cpu, origcpu;
2220 int error, n;
2222 KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1,
2223 ("invalid pcbinfo count %d", pcbinfo_arrlen));
2226 * The process of preparing the TCB list is too time-consuming and
2227 * resource-intensive to repeat twice on every request.
2229 n = 0;
2230 if (req->oldptr == NULL) {
2231 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu)
2232 n += pcbinfo_arr[cpu].ipi_count;
2233 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
2234 return 0;
2237 if (req->newptr != NULL)
2238 return EPERM;
2240 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
2241 marker->inp_flags |= INP_PLACEMARKER;
2244 * OK, now we're committed to doing something. Re-fetch ipi_count
2245 * after obtaining the generation count.
2247 error = 0;
2248 origcpu = mycpuid;
2249 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) {
2250 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu];
2251 struct inpcb *inp;
2252 struct xinpcb xi;
2253 int i;
2255 lwkt_migratecpu(cpu);
2257 GET_PCBINFO_TOKEN(pcbinfo);
2259 n = pcbinfo->ipi_count;
2261 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
2262 i = 0;
2263 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
2264 LIST_REMOVE(marker, inp_list);
2265 LIST_INSERT_AFTER(inp, marker, inp_list);
2267 if (inp->inp_flags & INP_PLACEMARKER)
2268 continue;
2269 if (prison_xinpcb(req->td, inp))
2270 continue;
2272 bzero(&xi, sizeof xi);
2273 xi.xi_len = sizeof xi;
2274 bcopy(inp, &xi.xi_inp, sizeof *inp);
2275 if (inp->inp_socket)
2276 sotoxsocket(inp->inp_socket, &xi.xi_socket);
2277 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
2278 break;
2279 ++i;
2281 LIST_REMOVE(marker, inp_list);
2283 REL_PCBINFO_TOKEN(pcbinfo);
2285 if (error == 0 && i < n) {
2286 bzero(&xi, sizeof xi);
2287 xi.xi_len = sizeof xi;
2288 while (i < n) {
2289 error = SYSCTL_OUT(req, &xi, sizeof xi);
2290 if (error)
2291 break;
2292 ++i;
2297 lwkt_migratecpu(origcpu);
2298 kfree(marker, M_TEMP);
2299 return error;
2303 in_pcblist_ncpus(SYSCTL_HANDLER_ARGS)
2306 return (in_pcblist_range(oidp, arg1, netisr_ncpus, req));
2309 void
2310 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2312 struct sockaddr_in *sin;
2314 KASSERT(faddr->sa_family == AF_INET,
2315 ("not AF_INET faddr %d", faddr->sa_family));
2317 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2318 sin->sin_family = AF_INET;
2319 sin->sin_len = sizeof(*sin);
2320 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2321 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2323 so->so_faddr = (struct sockaddr *)sin;
2326 void
2327 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2328 u_short offset)
2330 memset(portinfo, 0, sizeof(*portinfo));
2332 portinfo->offset = offset;
2333 portinfo->lastport = offset;
2334 portinfo->lastlow = offset;
2335 portinfo->lasthi = offset;
2337 portinfo->porthashbase = hashinit(hashsize, M_PCB,
2338 &portinfo->porthashmask);
2341 void
2342 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step)
2344 int hi, lo;
2346 if (step == 1)
2347 return;
2349 hi = *hi0;
2350 lo = *lo0;
2352 hi = rounddown(hi, step);
2353 hi += ofs;
2354 if (hi > (int)*hi0)
2355 hi -= step;
2357 lo = roundup(lo, step);
2358 lo -= (step - ofs);
2359 if (lo < (int)*lo0)
2360 lo += step;
2362 *hi0 = hi;
2363 *lo0 = lo;
2366 void
2367 in_pcbglobalinit(void)
2369 int cpu;
2371 in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB,
2372 M_WAITOK | M_ZERO);
2373 in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer),
2374 M_PCB, M_WAITOK | M_ZERO);
2376 for (cpu = 0; cpu < ncpus; ++cpu) {
2377 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu];
2378 struct inpcb *marker = &in_pcbmarkers[cpu];
2380 marker->inp_flags |= INP_PLACEMARKER;
2381 ic->ic_inp = marker;
2385 struct inpcb *
2386 in_pcbmarker(int cpuid)
2388 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2389 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2391 return &in_pcbmarkers[cpuid];
2394 struct inpcontainer *
2395 in_pcbcontainer_marker(int cpuid)
2397 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2398 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2400 return &in_pcbcontainer_markers[cpuid];
2403 void
2404 in_pcbresetroute(struct inpcb *inp)
2406 struct route *ro = &inp->inp_route;
2408 if (ro->ro_rt != NULL)
2409 RTFREE(ro->ro_rt);
2410 bzero(ro, sizeof(*ro));
2413 u_short
2414 in_pcblastport_down(volatile u_short *lastport, u_short first, u_short last,
2415 u_short step)
2417 u_short lport;
2419 for (;;) {
2420 u_short olport;
2422 olport = *lastport;
2423 lport = olport - step;
2424 if (__predict_false(lport > first || lport < last))
2425 lport = first;
2426 if (atomic_cmpset_short(lastport, olport, lport))
2427 break;
2429 return lport;
2432 u_short
2433 in_pcblastport_up(volatile u_short *lastport, u_short first, u_short last,
2434 u_short step)
2436 u_short lport;
2438 for (;;) {
2439 u_short olport;
2441 olport = *lastport;
2442 lport = olport + step;
2443 if (__predict_false(lport < first || lport > last))
2444 lport = first;
2445 if (atomic_cmpset_short(lastport, olport, lport))
2446 break;
2448 return lport;