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
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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
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
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
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>
73 #include <sys/domain.h>
74 #include <sys/protosw.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.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>
90 #include <net/if_types.h>
91 #include <net/route.h>
92 #include <net/netisr2.h>
94 #include <netinet/in.h>
95 #include <netinet/in_pcb.h>
96 #include <netinet/in_var.h>
97 #include <netinet/ip_var.h>
99 #include <netinet/ip6.h>
100 #include <netinet6/ip6_var.h>
104 #include <netinet6/ipsec.h>
105 #include <netproto/key/key.h>
106 #include <netproto/ipsec/esp_var.h>
110 #if defined(IPSEC) || defined(IPSEC_ESP)
111 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!"
114 #include <netproto/ipsec/ipsec.h>
115 #include <netproto/ipsec/key.h>
117 #endif /* FAST_IPSEC */
119 #define INP_LOCALGROUP_SIZMIN 8
120 #define INP_LOCALGROUP_SIZMAX 256
122 struct in_addr zeroin_addr;
125 * These configure the range of local port addresses assigned to
126 * "unspecified" outgoing connections/packets/whatever.
128 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */
129 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */
131 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */
132 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */
134 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */
135 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */
137 #define RANGECHK(var, min, max) \
138 if ((var) < (min)) { (var) = (min); } \
139 else if ((var) > (max)) { (var) = (max); }
141 int udpencap_enable = 1; /* enabled by default */
142 int udpencap_port = 4500; /* triggers decapsulation */
145 * Per-netisr inpcb markers.
146 * NOTE: they should only be used in netisrs.
148 static struct inpcb *in_pcbmarkers;
149 static struct inpcontainer *in_pcbcontainer_markers;
152 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
156 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
158 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
159 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
161 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX);
162 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX);
164 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX);
165 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX);
172 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW,
175 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW,
177 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", "");
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW,
179 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", "");
180 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW,
181 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", "");
182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW,
183 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW,
185 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", "");
188 * in_pcb.c: manage the Protocol Control Blocks.
190 * NOTE: It is assumed that most of these functions will be called from
191 * a critical section. XXX - There are, unfortunately, a few exceptions
192 * to this rule that should be fixed.
194 * NOTE: The caller should initialize the cpu field to the cpu running the
195 * protocol stack associated with this inpcbinfo.
199 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared)
201 KASSERT(cpu >= 0 && cpu < ncpus, ("invalid cpu%d", cpu));
204 LIST_INIT(&pcbinfo->pcblisthead);
205 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB,
209 pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token),
211 lwkt_token_init(pcbinfo->infotoken, "infotoken");
213 pcbinfo->infotoken = NULL;
217 struct baddynamicports baddynamicports;
220 * Check if the specified port is invalid for dynamic allocation.
223 in_baddynamic(u_int16_t port, u_int16_t proto)
227 return (DP_ISSET(baddynamicports.tcp, port));
230 /* Cannot preset this as it is a sysctl */
231 if (port == udpencap_port)
234 return (DP_ISSET(baddynamicports.udp, port));
241 in_pcbonlist(struct inpcb *inp)
243 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
245 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
246 ("not in the correct netisr"));
247 KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist"));
248 inp->inp_flags |= INP_ONLIST;
250 GET_PCBINFO_TOKEN(pcbinfo);
251 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
252 pcbinfo->ipi_count++;
253 REL_PCBINFO_TOKEN(pcbinfo);
257 in_pcbofflist(struct inpcb *inp)
259 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
261 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
262 ("not in the correct netisr"));
263 KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist"));
264 inp->inp_flags &= ~INP_ONLIST;
266 GET_PCBINFO_TOKEN(pcbinfo);
267 LIST_REMOVE(inp, inp_list);
268 KASSERT(pcbinfo->ipi_count > 0,
269 ("invalid inpcb count %d", pcbinfo->ipi_count));
270 pcbinfo->ipi_count--;
271 REL_PCBINFO_TOKEN(pcbinfo);
275 * Allocate a PCB and associate it with the socket.
278 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
285 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK);
288 inp->inp_lgrpindex = -1;
289 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
290 inp->inp_pcbinfo = pcbinfo;
291 inp->inp_socket = so;
293 error = ipsec_init_policy(so, &inp->inp_sp);
300 if (INP_SOCKAF(so) == AF_INET6 && ip6_v6only)
301 inp->inp_flags |= IN6P_IPV6_V6ONLY;
302 if (ip6_auto_flowlabel)
303 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
313 * Unlink a pcb with the intention of moving it to another cpu with a
314 * different pcbinfo. While unlinked nothing should attempt to dereference
315 * inp_pcbinfo, NULL it out so we assert if it does.
318 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
320 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
321 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
325 inp->inp_pcbinfo = NULL;
329 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
331 in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD);
335 * Relink a pcb into a new pcbinfo.
338 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
340 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
341 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
344 inp->inp_pcbinfo = pcbinfo;
349 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
351 return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD);
355 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
357 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
358 struct inpcbportinfo *portinfo;
359 u_short first, last, lport, step;
362 int portinfo_first, portinfo_idx;
364 inp->inp_flags |= INP_ANONPORT;
366 step = pcbinfo->portinfo_mask + 1;
367 portinfo_first = mycpuid & pcbinfo->portinfo_mask;
368 portinfo_idx = portinfo_first;
370 portinfo = &pcbinfo->portinfo[portinfo_idx];
372 if (inp->inp_flags & INP_HIGHPORT) {
373 first = ipport_hifirstauto; /* sysctl */
374 last = ipport_hilastauto;
375 lastport = &portinfo->lasthi;
376 } else if (inp->inp_flags & INP_LOWPORT) {
379 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
380 inp->inp_laddr.s_addr = INADDR_ANY;
383 first = ipport_lowfirstauto; /* 1023 */
384 last = ipport_lowlastauto; /* 600 */
385 lastport = &portinfo->lastlow;
387 first = ipport_firstauto; /* sysctl */
388 last = ipport_lastauto;
389 lastport = &portinfo->lastport;
393 * This has to be atomic. If the porthash is shared across multiple
394 * protocol threads (aka tcp) then the token must be held.
396 GET_PORT_TOKEN(portinfo);
399 * Simple check to ensure all ports are not used up causing
402 * We split the two cases (up and down) so that the direction
403 * is not being tested on each round of the loop.
409 in_pcbportrange(&first, &last, portinfo->offset, step);
410 count = (first - last) / step;
413 if (count-- < 0) { /* completely used? */
414 error = EADDRNOTAVAIL;
418 if (*lastport > first || *lastport < last)
420 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
422 lport = htons(*lastport);
423 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
429 in_pcbportrange(&last, &first, portinfo->offset, step);
430 count = (last - first) / step;
433 if (count-- < 0) { /* completely used? */
434 error = EADDRNOTAVAIL;
438 if (*lastport < first || *lastport > last)
440 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
442 lport = htons(*lastport);
443 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
446 inp->inp_lport = lport;
447 in_pcbinsporthash(portinfo, inp);
450 REL_PORT_TOKEN(portinfo);
453 /* Try next portinfo */
455 portinfo_idx &= pcbinfo->portinfo_mask;
456 if (portinfo_idx != portinfo_first)
458 inp->inp_laddr.s_addr = INADDR_ANY;
464 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
466 struct socket *so = inp->inp_socket;
467 struct sockaddr_in jsin;
468 struct ucred *cred = NULL;
471 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
472 return (EADDRNOTAVAIL);
473 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
474 return (EINVAL); /* already bound */
476 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
477 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
479 cred = td->td_proc->p_ucred;
482 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
483 struct inpcbinfo *pcbinfo;
484 struct inpcbportinfo *portinfo;
486 u_short lport, lport_ho;
487 int reuseport = (so->so_options & SO_REUSEPORT);
490 if (nam->sa_len != sizeof *sin)
494 * We should check the family, but old programs
495 * incorrectly fail to initialize it.
497 if (sin->sin_family != AF_INET)
498 return (EAFNOSUPPORT);
500 if (!prison_replace_wildcards(td, nam))
503 lport = sin->sin_port;
504 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
506 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
507 * allow complete duplication of binding if
508 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
509 * and a multicast address is bound on both
510 * new and duplicated sockets.
512 if (so->so_options & SO_REUSEADDR)
513 reuseport = SO_REUSEADDR | SO_REUSEPORT;
514 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
515 sin->sin_port = 0; /* yech... */
516 bzero(&sin->sin_zero, sizeof sin->sin_zero);
517 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
518 return (EADDRNOTAVAIL);
521 inp->inp_laddr = sin->sin_addr;
523 jsin.sin_family = AF_INET;
524 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
525 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
526 inp->inp_laddr.s_addr = INADDR_ANY;
529 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
532 /* Auto-select local port */
533 return in_pcbsetlport(inp, wild, cred);
535 lport_ho = ntohs(lport);
538 if (lport_ho < IPPORT_RESERVED && cred &&
540 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
541 inp->inp_laddr.s_addr = INADDR_ANY;
546 * Locate the proper portinfo based on lport
548 pcbinfo = inp->inp_pcbinfo;
550 &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
551 KKASSERT((lport_ho & pcbinfo->portinfo_mask) ==
555 * This has to be atomic. If the porthash is shared across
556 * multiple protocol threads (aka tcp) then the token must
559 GET_PORT_TOKEN(portinfo);
561 if (so->so_cred->cr_uid != 0 &&
562 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
563 t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
564 INPLOOKUP_WILDCARD, cred);
566 (!in_nullhost(sin->sin_addr) ||
567 !in_nullhost(t->inp_laddr) ||
568 (t->inp_socket->so_options & SO_REUSEPORT) == 0) &&
569 (so->so_cred->cr_uid !=
570 t->inp_socket->so_cred->cr_uid)) {
572 if (!in_nullhost(sin->sin_addr) ||
573 !in_nullhost(t->inp_laddr) ||
574 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
577 inp->inp_laddr.s_addr = INADDR_ANY;
583 if (cred && !prison_replace_wildcards(td, nam)) {
584 inp->inp_laddr.s_addr = INADDR_ANY;
585 error = EADDRNOTAVAIL;
588 t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
590 if (t && !(reuseport & t->inp_socket->so_options)) {
592 if (!in_nullhost(sin->sin_addr) ||
593 !in_nullhost(t->inp_laddr) ||
594 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
597 inp->inp_laddr.s_addr = INADDR_ANY;
602 inp->inp_lport = lport;
603 in_pcbinsporthash(portinfo, inp);
606 REL_PORT_TOKEN(portinfo);
609 jsin.sin_family = AF_INET;
610 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
611 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
612 inp->inp_laddr.s_addr = INADDR_ANY;
615 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
617 return in_pcbsetlport(inp, wild, cred);
621 static struct inpcb *
622 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr,
623 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
626 struct inpcbporthead *porthash;
627 struct inpcbport *phd;
628 struct inpcb *match = NULL;
631 * If the porthashbase is shared across several cpus, it must
634 ASSERT_PORT_TOKEN_HELD(portinfo);
637 * Best fit PCB lookup.
639 * First see if this local port is in use by looking on the
642 porthash = &portinfo->porthashbase[
643 INP_PCBPORTHASH(lport, portinfo->porthashmask)];
644 LIST_FOREACH(phd, porthash, phd_hash) {
645 if (phd->phd_port == lport)
649 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
651 if ((inp->inp_vflag & INP_IPV4) == 0)
654 if (inp->inp_laddr.s_addr != INADDR_ANY &&
655 inp->inp_laddr.s_addr != laddr.s_addr)
658 if (inp->inp_faddr.s_addr != INADDR_ANY &&
659 inp->inp_faddr.s_addr != faddr.s_addr)
662 if (inp->inp_fport != 0 && inp->inp_fport != fport)
667 inp->inp_socket->so_cred->cr_prison) {
677 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote,
680 struct proc *p = td->td_proc;
681 unsigned short *lastport;
682 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
683 struct sockaddr_in jsin;
684 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
685 struct inpcbportinfo *portinfo;
686 struct ucred *cred = NULL;
687 u_short first, last, lport, step;
688 int count, error, dup;
689 int portinfo_first, portinfo_idx;
691 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
692 return (EADDRNOTAVAIL);
694 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
695 if (inp->inp_lport != 0)
696 return (EINVAL); /* already bound */
701 jsin.sin_family = AF_INET;
702 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
703 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
704 inp->inp_laddr.s_addr = INADDR_ANY;
707 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
709 inp->inp_flags |= INP_ANONPORT;
711 step = pcbinfo->portinfo_mask + 1;
712 portinfo_first = mycpuid & pcbinfo->portinfo_mask;
713 portinfo_idx = portinfo_first;
715 portinfo = &pcbinfo->portinfo[portinfo_idx];
718 if (inp->inp_flags & INP_HIGHPORT) {
719 first = ipport_hifirstauto; /* sysctl */
720 last = ipport_hilastauto;
721 lastport = &portinfo->lasthi;
722 } else if (inp->inp_flags & INP_LOWPORT) {
725 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
726 inp->inp_laddr.s_addr = INADDR_ANY;
729 first = ipport_lowfirstauto; /* 1023 */
730 last = ipport_lowlastauto; /* 600 */
731 lastport = &portinfo->lastlow;
733 first = ipport_firstauto; /* sysctl */
734 last = ipport_lastauto;
735 lastport = &portinfo->lastport;
739 * This has to be atomic. If the porthash is shared across multiple
740 * protocol threads (aka tcp) then the token must be held.
742 GET_PORT_TOKEN(portinfo);
746 * Simple check to ensure all ports are not used up causing
749 * We split the two cases (up and down) so that the direction
750 * is not being tested on each round of the loop.
756 in_pcbportrange(&first, &last, portinfo->offset, step);
757 count = (first - last) / step;
760 if (count-- < 0) { /* completely used? */
761 error = EADDRNOTAVAIL;
765 if (*lastport > first || *lastport < last)
767 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
769 lport = htons(*lastport);
770 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
771 lport, sin->sin_addr, sin->sin_port, cred));
776 in_pcbportrange(&last, &first, portinfo->offset, step);
777 count = (last - first) / step;
780 if (count-- < 0) { /* completely used? */
781 error = EADDRNOTAVAIL;
785 if (*lastport < first || *lastport > last)
787 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
789 lport = htons(*lastport);
790 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
791 lport, sin->sin_addr, sin->sin_port, cred));
794 /* This could happen on loopback interface */
795 if (sin->sin_port == lport &&
796 sin->sin_addr.s_addr == inp->inp_laddr.s_addr) {
799 * Duplicate again; give up
801 error = EADDRNOTAVAIL;
807 inp->inp_lport = lport;
808 in_pcbinsporthash(portinfo, inp);
811 REL_PORT_TOKEN(portinfo);
814 /* Try next portinfo */
816 portinfo_idx &= pcbinfo->portinfo_mask;
817 if (portinfo_idx != portinfo_first)
819 inp->inp_laddr.s_addr = INADDR_ANY;
825 * Transform old in_pcbconnect() into an inner subroutine for new
826 * in_pcbconnect(): Do some validity-checking on the remote
827 * address (in mbuf 'nam') and then determine local host address
828 * (i.e., which interface) to use to access that remote host.
830 * This preserves definition of in_pcbconnect(), while supporting a
831 * slightly different version for T/TCP. (This is more than
832 * a bit of a kludge, but cleaning up the internal interfaces would
833 * have forced minor changes in every protocol).
836 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
837 struct sockaddr_in **plocal_sin, struct thread *td, int find)
839 struct in_ifaddr *ia;
840 struct ucred *cred = NULL;
841 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
842 struct sockaddr *jsin;
843 int jailed = 0, alloc_route = 0;
845 if (nam->sa_len != sizeof *sin)
847 if (sin->sin_family != AF_INET)
848 return (EAFNOSUPPORT);
849 if (sin->sin_port == 0)
850 return (EADDRNOTAVAIL);
851 if (td && td->td_proc && td->td_proc->p_ucred)
852 cred = td->td_proc->p_ucred;
853 if (cred && cred->cr_prison)
855 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
856 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
858 * If the destination address is INADDR_ANY,
859 * use the primary local address.
860 * If the supplied address is INADDR_BROADCAST,
861 * and the primary interface supports broadcast,
862 * choose the broadcast address for that interface.
864 if (sin->sin_addr.s_addr == INADDR_ANY)
865 sin->sin_addr = IA_SIN(ia)->sin_addr;
866 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
867 (ia->ia_ifp->if_flags & IFF_BROADCAST))
868 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
875 * If route is known or can be allocated now,
876 * our src addr is taken from the i/f, else punt.
877 * Note that we should check the address family of the cached
878 * destination, in case of sharing the cache with IPv6.
880 ro = &inp->inp_route;
882 (!(ro->ro_rt->rt_flags & RTF_UP) ||
883 ro->ro_dst.sa_family != AF_INET ||
884 satosin(&ro->ro_dst)->sin_addr.s_addr !=
885 sin->sin_addr.s_addr ||
886 inp->inp_socket->so_options & SO_DONTROUTE)) {
890 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
891 (ro->ro_rt == NULL ||
892 ro->ro_rt->rt_ifp == NULL)) {
893 /* No route yet, so try to acquire one */
894 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
895 ro->ro_dst.sa_family = AF_INET;
896 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
897 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
903 * If we found a route, use the address
904 * corresponding to the outgoing interface
905 * unless it is the loopback (in case a route
906 * to our address on another net goes to loopback).
909 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
911 if (jailed_ip(cred->cr_prison,
912 ro->ro_rt->rt_ifa->ifa_addr)) {
913 ia = ifatoia(ro->ro_rt->rt_ifa);
916 ia = ifatoia(ro->ro_rt->rt_ifa);
920 u_short fport = sin->sin_port;
923 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
924 if (ia && jailed && !jailed_ip(cred->cr_prison,
925 sintosa(&ia->ia_addr)))
928 ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
929 if (ia && jailed && !jailed_ip(cred->cr_prison,
930 sintosa(&ia->ia_addr)))
932 sin->sin_port = fport;
934 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
935 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
936 if (ia && jailed && !jailed_ip(cred->cr_prison,
937 sintosa(&ia->ia_addr)))
940 if (!jailed && ia == NULL)
944 * If the destination address is multicast and an outgoing
945 * interface has been set as a multicast option, use the
946 * address of that interface as our source address.
948 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
949 inp->inp_moptions != NULL) {
950 struct ip_moptions *imo;
953 imo = inp->inp_moptions;
954 if (imo->imo_multicast_ifp != NULL) {
955 struct in_ifaddr_container *iac;
957 ifp = imo->imo_multicast_ifp;
960 &in_ifaddrheads[mycpuid], ia_link) {
961 if (iac->ia->ia_ifp == ifp) {
971 * Don't do pcblookup call here; return interface in plocal_sin
972 * and exit to caller, that will do the lookup.
974 if (ia == NULL && jailed) {
975 if ((jsin = prison_get_nonlocal(
976 cred->cr_prison, AF_INET, NULL)) != NULL ||
977 (jsin = prison_get_local(
978 cred->cr_prison, AF_INET, NULL)) != NULL) {
979 *plocal_sin = satosin(jsin);
985 *plocal_sin = &ia->ia_addr;
991 struct route *ro = &inp->inp_route;
993 if (ro->ro_rt != NULL)
995 bzero(ro, sizeof(*ro));
997 return (EADDRNOTAVAIL);
1001 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
1002 struct sockaddr_in **plocal_sin, struct thread *td)
1004 return in_pcbladdr_find(inp, nam, plocal_sin, td,
1005 (inp->inp_laddr.s_addr == INADDR_ANY));
1010 * Connect from a socket to a specified address.
1011 * Both address and port must be specified in argument sin.
1012 * If don't have a local address for this socket yet,
1016 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
1018 struct sockaddr_in *if_sin;
1019 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1022 /* Call inner routine to assign local interface address. */
1023 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
1026 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port,
1027 inp->inp_laddr.s_addr ?
1028 inp->inp_laddr : if_sin->sin_addr,
1029 inp->inp_lport, FALSE, NULL) != NULL) {
1030 return (EADDRINUSE);
1032 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1033 if (inp->inp_lport == 0) {
1034 error = in_pcbbind(inp, NULL, td);
1038 inp->inp_laddr = if_sin->sin_addr;
1040 inp->inp_faddr = sin->sin_addr;
1041 inp->inp_fport = sin->sin_port;
1042 in_pcbinsconnhash(inp);
1047 in_pcbdisconnect(struct inpcb *inp)
1050 in_pcbremconnhash(inp);
1051 inp->inp_faddr.s_addr = INADDR_ANY;
1056 in_pcbdetach(struct inpcb *inp)
1058 struct socket *so = inp->inp_socket;
1059 struct inpcbinfo *ipi = inp->inp_pcbinfo;
1062 ipsec4_delete_pcbpolicy(inp);
1064 inp->inp_gencnt = ++ipi->ipi_gencnt;
1065 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
1066 in_pcbremlists(inp);
1068 sofree(so); /* remove pcb ref */
1069 if (inp->inp_options)
1070 m_free(inp->inp_options);
1071 if (inp->inp_route.ro_rt)
1072 rtfree(inp->inp_route.ro_rt);
1073 ip_freemoptions(inp->inp_moptions);
1079 * The calling convention of in_setsockaddr() and in_setpeeraddr() was
1080 * modified to match the pru_sockaddr() and pru_peeraddr() entry points
1081 * in struct pr_usrreqs, so that protocols can just reference then directly
1082 * without the need for a wrapper function. The socket must have a valid
1083 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one
1084 * except through a kernel programming error, so it is acceptable to panic
1085 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap
1086 * because there actually /is/ a programming error somewhere... XXX)
1089 in_setsockaddr(struct socket *so, struct sockaddr **nam)
1092 struct sockaddr_in *sin;
1095 * Do the malloc first in case it blocks.
1097 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1098 sin->sin_family = AF_INET;
1099 sin->sin_len = sizeof *sin;
1105 kfree(sin, M_SONAME);
1106 return (ECONNRESET);
1108 sin->sin_port = inp->inp_lport;
1109 sin->sin_addr = inp->inp_laddr;
1112 *nam = (struct sockaddr *)sin;
1117 in_setsockaddr_dispatch(netmsg_t msg)
1121 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1122 lwkt_replymsg(&msg->lmsg, error);
1126 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1129 struct sockaddr_in *sin;
1132 * Do the malloc first in case it blocks.
1134 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1135 sin->sin_family = AF_INET;
1136 sin->sin_len = sizeof *sin;
1142 kfree(sin, M_SONAME);
1143 return (ECONNRESET);
1145 sin->sin_port = inp->inp_fport;
1146 sin->sin_addr = inp->inp_faddr;
1149 *nam = (struct sockaddr *)sin;
1154 in_setpeeraddr_dispatch(netmsg_t msg)
1158 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1159 lwkt_replymsg(&msg->lmsg, error);
1163 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err,
1164 void (*notify)(struct inpcb *, int))
1166 struct inpcb *inp, *marker;
1168 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1169 ("not in the correct netisr"));
1170 marker = &in_pcbmarkers[mycpuid];
1174 * - If INP_PLACEMARKER is set we must ignore the rest of the
1175 * structure and skip it.
1176 * - It is safe to nuke inpcbs here, since we are in their own
1179 GET_PCBINFO_TOKEN(pcbinfo);
1181 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1182 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1183 LIST_REMOVE(marker, inp_list);
1184 LIST_INSERT_AFTER(inp, marker, inp_list);
1186 if (inp->inp_flags & INP_PLACEMARKER)
1189 if (!(inp->inp_vflag & INP_IPV4))
1192 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1193 inp->inp_socket == NULL)
1195 (*notify)(inp, err); /* can remove inp from list! */
1197 LIST_REMOVE(marker, inp_list);
1199 REL_PCBINFO_TOKEN(pcbinfo);
1203 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1205 struct inpcb *inp, *marker;
1208 * We only need to make sure that we are in netisr0, where all
1209 * multicast operation happen. We could check inpcbinfo which
1210 * does not belong to netisr0 by holding the inpcbinfo's token.
1211 * In this case, the pcbinfo must be able to be shared, i.e.
1212 * pcbinfo->infotoken is not NULL.
1214 KASSERT(&curthread->td_msgport == netisr_cpuport(0),
1215 ("not in netisr0"));
1216 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL,
1217 ("pcbinfo could not be shared"));
1220 * Get a marker for the current netisr (netisr0).
1222 * It is possible that the multicast address deletion blocks,
1223 * which could cause temporary token releasing. So we use
1224 * inpcb marker here to get a coherent view of the inpcb list.
1226 * While, on the other hand, moptions are only added and deleted
1227 * in netisr0, so we would not see staled moption or miss moption
1228 * even if the token was released due to the blocking multicast
1231 marker = &in_pcbmarkers[mycpuid];
1233 GET_PCBINFO_TOKEN(pcbinfo);
1235 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1236 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1237 struct ip_moptions *imo;
1239 LIST_REMOVE(marker, inp_list);
1240 LIST_INSERT_AFTER(inp, marker, inp_list);
1242 if (inp->inp_flags & INP_PLACEMARKER)
1244 imo = inp->inp_moptions;
1245 if ((inp->inp_vflag & INP_IPV4) && imo != NULL) {
1249 * Unselect the outgoing interface if it is being
1252 if (imo->imo_multicast_ifp == ifp)
1253 imo->imo_multicast_ifp = NULL;
1256 * Drop multicast group membership if we joined
1257 * through the interface being detached.
1259 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1261 if (imo->imo_membership[i]->inm_ifp == ifp) {
1264 * This could block and the pcbinfo
1265 * token could be passively released.
1267 in_delmulti(imo->imo_membership[i]);
1269 } else if (gap != 0)
1270 imo->imo_membership[i - gap] =
1271 imo->imo_membership[i];
1273 imo->imo_num_memberships -= gap;
1276 LIST_REMOVE(marker, inp_list);
1278 REL_PCBINFO_TOKEN(pcbinfo);
1282 * Check for alternatives when higher level complains
1283 * about service problems. For now, invalidate cached
1284 * routing information. If the route was created dynamically
1285 * (by a redirect), time to try a default gateway again.
1288 in_losing(struct inpcb *inp)
1291 struct rt_addrinfo rtinfo;
1293 if ((rt = inp->inp_route.ro_rt)) {
1294 bzero(&rtinfo, sizeof(struct rt_addrinfo));
1295 rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1296 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1297 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1298 rtinfo.rti_flags = rt->rt_flags;
1299 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1300 if (rt->rt_flags & RTF_DYNAMIC) {
1301 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1302 rt_mask(rt), rt->rt_flags, NULL);
1304 inp->inp_route.ro_rt = NULL;
1307 * A new route can be allocated
1308 * the next time output is attempted.
1314 * After a routing change, flush old routing
1315 * and allocate a (hopefully) better one.
1318 in_rtchange(struct inpcb *inp, int err)
1320 if (inp->inp_route.ro_rt) {
1321 rtfree(inp->inp_route.ro_rt);
1322 inp->inp_route.ro_rt = NULL;
1324 * A new route can be allocated the next time
1325 * output is attempted.
1331 * Lookup a PCB based on the local address and port.
1334 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr,
1335 u_int lport_arg, int wild_okay, struct ucred *cred)
1338 int matchwild = 3, wildcard;
1339 u_short lport = lport_arg;
1340 struct inpcbporthead *porthash;
1341 struct inpcbport *phd;
1342 struct inpcb *match = NULL;
1345 * If the porthashbase is shared across several cpus, it must
1348 ASSERT_PORT_TOKEN_HELD(portinfo);
1351 * Best fit PCB lookup.
1353 * First see if this local port is in use by looking on the
1356 porthash = &portinfo->porthashbase[
1357 INP_PCBPORTHASH(lport, portinfo->porthashmask)];
1358 LIST_FOREACH(phd, porthash, phd_hash) {
1359 if (phd->phd_port == lport)
1364 * Port is in use by one or more PCBs. Look for best
1367 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1370 if ((inp->inp_vflag & INP_IPV4) == 0)
1373 if (inp->inp_faddr.s_addr != INADDR_ANY)
1375 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1376 if (laddr.s_addr == INADDR_ANY)
1378 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1381 if (laddr.s_addr != INADDR_ANY)
1384 if (wildcard && !wild_okay)
1386 if (wildcard < matchwild &&
1389 inp->inp_socket->so_cred->cr_prison)) {
1391 matchwild = wildcard;
1392 if (matchwild == 0) {
1402 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1403 const struct inpcb *inp)
1405 const struct inp_localgrphead *hdr;
1406 const struct inp_localgroup *grp;
1409 if (pcbinfo->localgrphashbase == NULL)
1412 GET_PCBINFO_TOKEN(pcbinfo);
1414 hdr = &pcbinfo->localgrphashbase[
1415 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1417 LIST_FOREACH(grp, hdr, il_list) {
1418 if (grp->il_vflag == inp->inp_vflag &&
1419 grp->il_lport == inp->inp_lport &&
1420 memcmp(&grp->il_dependladdr,
1421 &inp->inp_inc.inc_ie.ie_dependladdr,
1422 sizeof(grp->il_dependladdr)) == 0) {
1426 if (grp == NULL || grp->il_inpcnt == 1) {
1427 REL_PCBINFO_TOKEN(pcbinfo);
1431 KASSERT(grp->il_inpcnt >= 2,
1432 ("invalid localgroup inp count %d", grp->il_inpcnt));
1433 for (i = 0; i < grp->il_inpcnt; ++i) {
1434 if (grp->il_inp[i] == inp) {
1435 int last = grp->il_inpcnt - 1;
1438 last = grp->il_inpcnt - 2;
1439 REL_PCBINFO_TOKEN(pcbinfo);
1440 return grp->il_inp[last];
1443 REL_PCBINFO_TOKEN(pcbinfo);
1447 static struct inpcb *
1448 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1449 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1451 struct inpcb *local_wild = NULL;
1452 const struct inp_localgrphead *hdr;
1453 const struct inp_localgroup *grp;
1455 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1457 hdr = &pcbinfo->localgrphashbase[
1458 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1461 * Order of socket selection:
1466 * - Local group does not contain jailed sockets
1467 * - Local group does not contain IPv4 mapped INET6 wild sockets
1469 LIST_FOREACH(grp, hdr, il_list) {
1471 if (!(grp->il_vflag & INP_IPV4))
1474 if (grp->il_lport == lport) {
1478 * Modulo-N is used here, which greatly reduces
1479 * completion queue token contention, thus more
1480 * cpu time is saved.
1482 idx = pkt_hash % grp->il_inpcnt;
1483 if (grp->il_laddr.s_addr == laddr.s_addr)
1484 return grp->il_inp[idx];
1485 else if (grp->il_laddr.s_addr == INADDR_ANY)
1486 local_wild = grp->il_inp[idx];
1489 if (local_wild != NULL)
1495 * Lookup PCB in hash list.
1498 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1499 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1500 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1502 struct inpcbhead *head;
1503 struct inpcb *inp, *jinp=NULL;
1504 u_short fport = fport_arg, lport = lport_arg;
1507 * First look for an exact match.
1509 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1510 laddr.s_addr, lport, pcbinfo->hashmask)];
1511 LIST_FOREACH(inp, head, inp_hash) {
1513 if (!(inp->inp_vflag & INP_IPV4))
1516 if (in_hosteq(inp->inp_faddr, faddr) &&
1517 in_hosteq(inp->inp_laddr, laddr) &&
1518 inp->inp_fport == fport && inp->inp_lport == lport) {
1520 if (inp->inp_socket == NULL ||
1521 inp->inp_socket->so_cred->cr_prison == NULL) {
1533 struct inpcb *local_wild = NULL;
1534 struct inpcb *jinp_wild = NULL;
1536 struct inpcb *local_wild_mapped = NULL;
1538 struct inpcontainer *ic;
1539 struct inpcontainerhead *chead;
1540 struct sockaddr_in jsin;
1543 GET_PCBINFO_TOKEN(pcbinfo);
1546 * Check local group first
1548 if (pcbinfo->localgrphashbase != NULL &&
1549 m != NULL && (m->m_flags & M_HASH) &&
1550 !(ifp && ifp->if_type == IFT_FAITH)) {
1551 inp = inp_localgroup_lookup(pcbinfo,
1552 laddr, lport, m->m_pkthdr.hash);
1554 REL_PCBINFO_TOKEN(pcbinfo);
1560 * Order of socket selection:
1561 * 1. non-jailed, non-wild.
1562 * 2. non-jailed, wild.
1563 * 3. jailed, non-wild.
1566 jsin.sin_family = AF_INET;
1567 chead = &pcbinfo->wildcardhashbase[
1568 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1569 LIST_FOREACH(ic, chead, ic_list) {
1571 if (inp->inp_flags & INP_PLACEMARKER)
1574 jsin.sin_addr.s_addr = laddr.s_addr;
1576 if (!(inp->inp_vflag & INP_IPV4))
1579 if (inp->inp_socket != NULL)
1580 cred = inp->inp_socket->so_cred;
1583 if (cred != NULL && jailed(cred)) {
1587 if (!jailed_ip(cred->cr_prison,
1588 (struct sockaddr *)&jsin))
1591 if (inp->inp_lport == lport) {
1592 if (ifp && ifp->if_type == IFT_FAITH &&
1593 !(inp->inp_flags & INP_FAITH))
1595 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1596 if (cred != NULL && jailed(cred)) {
1599 REL_PCBINFO_TOKEN(pcbinfo);
1603 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1605 if (INP_CHECK_SOCKAF(inp->inp_socket,
1607 local_wild_mapped = inp;
1619 REL_PCBINFO_TOKEN(pcbinfo);
1621 if (local_wild != NULL)
1622 return (local_wild);
1624 if (local_wild_mapped != NULL)
1625 return (local_wild_mapped);
1639 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1640 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1641 boolean_t wildcard, struct ifnet *ifp)
1643 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1644 laddr, lport_arg, wildcard, ifp, NULL);
1648 * Insert PCB into connection hash table.
1651 in_pcbinsconnhash(struct inpcb *inp)
1653 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1654 struct inpcbhead *bucket;
1655 u_int32_t hashkey_faddr, hashkey_laddr;
1658 if (inp->inp_vflag & INP_IPV6) {
1659 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1660 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1663 hashkey_faddr = inp->inp_faddr.s_addr;
1664 hashkey_laddr = inp->inp_laddr.s_addr;
1669 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1670 ("not in the correct netisr"));
1671 KASSERT(!(inp->inp_flags & INP_WILDCARD), ("already on wildcardhash"));
1672 KASSERT(!(inp->inp_flags & INP_CONNECTED), ("already on connhash"));
1673 inp->inp_flags |= INP_CONNECTED;
1676 * Insert into the connection hash table.
1678 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1679 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1680 LIST_INSERT_HEAD(bucket, inp, inp_hash);
1684 * Remove PCB from connection hash table.
1687 in_pcbremconnhash(struct inpcb *inp)
1689 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo;
1691 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1692 ("not in the correct netisr"));
1693 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1695 LIST_REMOVE(inp, inp_hash);
1696 inp->inp_flags &= ~INP_CONNECTED;
1700 * Insert PCB into port hash table.
1703 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp)
1705 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1706 struct inpcbporthead *pcbporthash;
1707 struct inpcbport *phd;
1710 * If the porthashbase is shared across several cpus, it must
1713 ASSERT_PORT_TOKEN_HELD(portinfo);
1716 * Insert into the port hash table.
1718 pcbporthash = &portinfo->porthashbase[
1719 INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)];
1721 /* Go through port list and look for a head for this lport. */
1722 LIST_FOREACH(phd, pcbporthash, phd_hash) {
1723 if (phd->phd_port == inp->inp_lport)
1727 /* If none exists, use saved one and tack it on. */
1729 KKASSERT(pcbinfo->portsave != NULL);
1730 phd = pcbinfo->portsave;
1731 pcbinfo->portsave = NULL;
1732 phd->phd_port = inp->inp_lport;
1733 LIST_INIT(&phd->phd_pcblist);
1734 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1737 inp->inp_portinfo = portinfo;
1739 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1742 * Malloc one inpcbport for later use. It is safe to use
1743 * "wait" malloc here (port token would be released, if
1744 * malloc ever blocked), since all changes to the porthash
1747 if (pcbinfo->portsave == NULL) {
1748 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1749 M_PCB, M_INTWAIT | M_ZERO);
1754 in_pcbinsporthash_lport(struct inpcb *inp)
1756 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1757 struct inpcbportinfo *portinfo;
1760 /* Locate the proper portinfo based on lport */
1761 lport_ho = ntohs(inp->inp_lport);
1762 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
1763 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset);
1765 GET_PORT_TOKEN(portinfo);
1766 in_pcbinsporthash(portinfo, inp);
1767 REL_PORT_TOKEN(portinfo);
1770 static struct inp_localgroup *
1771 inp_localgroup_alloc(u_char vflag,
1772 uint16_t port, const union in_dependaddr *addr, int size)
1774 struct inp_localgroup *grp;
1776 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
1777 M_TEMP, M_INTWAIT | M_ZERO);
1778 grp->il_vflag = vflag;
1779 grp->il_lport = port;
1780 grp->il_dependladdr = *addr;
1781 grp->il_inpsiz = size;
1787 inp_localgroup_free(struct inp_localgroup *grp)
1793 inp_localgroup_destroy(struct inp_localgroup *grp)
1795 LIST_REMOVE(grp, il_list);
1796 inp_localgroup_free(grp);
1800 inp_localgroup_copy(struct inp_localgroup *grp,
1801 const struct inp_localgroup *old_grp)
1805 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
1806 ("invalid new local group size %d and old local group count %d",
1807 grp->il_inpsiz, old_grp->il_inpcnt));
1808 for (i = 0; i < old_grp->il_inpcnt; ++i)
1809 grp->il_inp[i] = old_grp->il_inp[i];
1810 grp->il_inpcnt = old_grp->il_inpcnt;
1814 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1816 struct inp_localgrphead *hdr;
1817 struct inp_localgroup *grp, *grp_alloc = NULL;
1821 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1823 if (pcbinfo->localgrphashbase == NULL)
1827 * XXX don't allow jailed socket to join local group
1829 if (inp->inp_socket != NULL)
1830 cred = inp->inp_socket->so_cred;
1833 if (cred != NULL && jailed(cred))
1838 * XXX don't allow IPv4 mapped INET6 wild socket
1840 if ((inp->inp_vflag & INP_IPV4) &&
1841 inp->inp_laddr.s_addr == INADDR_ANY &&
1842 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6))
1846 hdr = &pcbinfo->localgrphashbase[
1847 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1850 LIST_FOREACH(grp, hdr, il_list) {
1851 if (grp->il_vflag == inp->inp_vflag &&
1852 grp->il_lport == inp->inp_lport &&
1853 memcmp(&grp->il_dependladdr,
1854 &inp->inp_inc.inc_ie.ie_dependladdr,
1855 sizeof(grp->il_dependladdr)) == 0) {
1861 * Create a new local group
1863 if (grp_alloc == NULL) {
1864 grp_alloc = inp_localgroup_alloc(inp->inp_vflag,
1865 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
1866 INP_LOCALGROUP_SIZMIN);
1868 * Local group allocation could block and the
1869 * local group w/ the same property might have
1870 * been added by others when we were blocked;
1875 /* Local group has been allocated; link it */
1878 LIST_INSERT_HEAD(hdr, grp, il_list);
1880 } else if (grp->il_inpcnt == grp->il_inpsiz) {
1881 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
1882 static int limit_logged = 0;
1884 if (!limit_logged) {
1886 kprintf("local group port %d, "
1887 "limit reached\n", ntohs(grp->il_lport));
1889 if (grp_alloc != NULL) {
1891 * This would happen if the local group
1892 * w/ the same property was expanded when
1893 * our local group allocation blocked.
1895 inp_localgroup_free(grp_alloc);
1901 * Expand this local group
1903 if (grp_alloc == NULL ||
1904 grp->il_inpcnt >= grp_alloc->il_inpsiz) {
1905 if (grp_alloc != NULL)
1906 inp_localgroup_free(grp_alloc);
1907 grp_alloc = inp_localgroup_alloc(grp->il_vflag,
1908 grp->il_lport, &grp->il_dependladdr,
1909 grp->il_inpsiz * 2);
1911 * Local group allocation could block and the
1912 * local group w/ the same property might have
1913 * been expanded by others when we were blocked;
1920 * Save the old local group, link the new one, and then
1921 * destroy the old local group
1923 inp_localgroup_copy(grp_alloc, grp);
1924 LIST_INSERT_HEAD(hdr, grp_alloc, il_list);
1925 inp_localgroup_destroy(grp);
1931 * Found the local group
1933 if (grp_alloc != NULL) {
1935 * This would happen if the local group w/ the
1936 * same property was added or expanded when our
1937 * local group allocation blocked.
1939 inp_localgroup_free(grp_alloc);
1944 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
1945 ("invalid local group size %d and count %d",
1946 grp->il_inpsiz, grp->il_inpcnt));
1949 * Keep the local group sorted by the inpcb local group index
1950 * in ascending order.
1952 * This eases the multi-process userland application which uses
1953 * SO_REUSEPORT sockets and binds process to the owner cpu of
1954 * the SO_REUSEPORT socket:
1955 * If we didn't sort the local group by the inpcb local group
1956 * index and one of the process owning an inpcb in this local
1957 * group restarted, e.g. crashed and restarted by watchdog,
1958 * other processes owning a inpcb in this local group would have
1959 * to detect that event, refetch its socket's owner cpu, and
1962 idx = grp->il_inpcnt;
1963 for (i = 0; i < idx; ++i) {
1964 struct inpcb *oinp = grp->il_inp[i];
1966 if (oinp->inp_lgrpindex > i) {
1967 if (inp->inp_lgrpindex < 0) {
1968 inp->inp_lgrpindex = i;
1969 } else if (inp->inp_lgrpindex != i) {
1971 kprintf("inp %p: grpidx %d, "
1972 "assigned to %d, cpu%d\n",
1973 inp, inp->inp_lgrpindex, i,
1977 grp->il_inp[i] = inp;
1979 /* Pull down inpcbs */
1980 for (; i < grp->il_inpcnt; ++i) {
1981 struct inpcb *oinp1 = grp->il_inp[i + 1];
1983 grp->il_inp[i + 1] = oinp;
1991 if (inp->inp_lgrpindex < 0) {
1992 inp->inp_lgrpindex = idx;
1993 } else if (inp->inp_lgrpindex != idx) {
1995 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n",
1996 inp, inp->inp_lgrpindex, idx, mycpuid);
1999 grp->il_inp[idx] = inp;
2004 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2006 struct inpcontainer *ic;
2007 struct inpcontainerhead *bucket;
2009 GET_PCBINFO_TOKEN(pcbinfo);
2011 in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
2013 bucket = &pcbinfo->wildcardhashbase[
2014 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2016 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
2018 LIST_INSERT_HEAD(bucket, ic, ic_list);
2020 REL_PCBINFO_TOKEN(pcbinfo);
2024 * Insert PCB into wildcard hash table.
2027 in_pcbinswildcardhash(struct inpcb *inp)
2029 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2031 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2032 ("not in correct netisr"));
2033 KASSERT(!(inp->inp_flags & INP_CONNECTED),
2034 ("already on connhash"));
2035 KASSERT(!(inp->inp_flags & INP_WILDCARD),
2036 ("already on wildcardhash"));
2037 inp->inp_flags |= INP_WILDCARD;
2039 in_pcbinswildcardhash_oncpu(inp, pcbinfo);
2043 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2045 struct inp_localgrphead *hdr;
2046 struct inp_localgroup *grp;
2048 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
2050 if (pcbinfo->localgrphashbase == NULL)
2053 hdr = &pcbinfo->localgrphashbase[
2054 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
2056 LIST_FOREACH(grp, hdr, il_list) {
2059 for (i = 0; i < grp->il_inpcnt; ++i) {
2060 if (grp->il_inp[i] != inp)
2063 if (grp->il_inpcnt == 1) {
2064 /* Destroy this local group */
2065 inp_localgroup_destroy(grp);
2067 /* Pull up inpcbs */
2068 for (; i + 1 < grp->il_inpcnt; ++i)
2069 grp->il_inp[i] = grp->il_inp[i + 1];
2078 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2080 struct inpcontainer *ic;
2081 struct inpcontainerhead *head;
2083 GET_PCBINFO_TOKEN(pcbinfo);
2085 in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
2088 head = &pcbinfo->wildcardhashbase[
2089 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2091 LIST_FOREACH(ic, head, ic_list) {
2092 if (ic->ic_inp == inp)
2095 REL_PCBINFO_TOKEN(pcbinfo);
2096 return; /* not found! */
2099 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */
2100 REL_PCBINFO_TOKEN(pcbinfo);
2101 kfree(ic, M_TEMP); /* deallocate container */
2105 * Remove PCB from wildcard hash table.
2108 in_pcbremwildcardhash(struct inpcb *inp)
2110 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2112 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2113 ("not in correct netisr"));
2114 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
2116 in_pcbremwildcardhash_oncpu(inp, pcbinfo);
2117 inp->inp_lgrpindex = -1;
2118 inp->inp_flags &= ~INP_WILDCARD;
2122 * Remove PCB from various lists.
2125 in_pcbremlists(struct inpcb *inp)
2127 if (inp->inp_lport) {
2128 struct inpcbportinfo *portinfo;
2129 struct inpcbport *phd;
2133 * inp->inp_portinfo is _not_ necessary same as
2134 * inp->inp_pcbinfo->portinfo.
2136 portinfo = inp->inp_portinfo;
2137 GET_PORT_TOKEN(portinfo);
2140 LIST_REMOVE(inp, inp_portlist);
2141 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2142 LIST_REMOVE(phd, phd_hash);
2146 REL_PORT_TOKEN(portinfo);
2148 if (inp->inp_flags & INP_WILDCARD) {
2149 in_pcbremwildcardhash(inp);
2150 } else if (inp->inp_flags & INP_CONNECTED) {
2151 in_pcbremconnhash(inp);
2154 if (inp->inp_flags & INP_ONLIST)
2159 prison_xinpcb(struct thread *td, struct inpcb *inp)
2163 if (td->td_proc == NULL)
2165 cr = td->td_proc->p_ucred;
2166 if (cr->cr_prison == NULL)
2168 if (inp->inp_socket && inp->inp_socket->so_cred &&
2169 inp->inp_socket->so_cred->cr_prison &&
2170 cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
2176 in_pcblist_global(SYSCTL_HANDLER_ARGS)
2178 struct inpcbinfo *pcbinfo_arr = arg1;
2179 int pcbinfo_arrlen = arg2;
2180 struct inpcb *marker;
2184 KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1,
2185 ("invalid pcbinfo count %d", pcbinfo_arrlen));
2188 * The process of preparing the TCB list is too time-consuming and
2189 * resource-intensive to repeat twice on every request.
2192 if (req->oldptr == NULL) {
2193 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu)
2194 n += pcbinfo_arr[cpu].ipi_count;
2195 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
2199 if (req->newptr != NULL)
2202 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
2203 marker->inp_flags |= INP_PLACEMARKER;
2206 * OK, now we're committed to doing something. Re-fetch ipi_count
2207 * after obtaining the generation count.
2211 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) {
2212 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu];
2217 lwkt_migratecpu(cpu);
2219 GET_PCBINFO_TOKEN(pcbinfo);
2221 n = pcbinfo->ipi_count;
2223 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
2225 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
2226 LIST_REMOVE(marker, inp_list);
2227 LIST_INSERT_AFTER(inp, marker, inp_list);
2229 if (inp->inp_flags & INP_PLACEMARKER)
2231 if (prison_xinpcb(req->td, inp))
2234 bzero(&xi, sizeof xi);
2235 xi.xi_len = sizeof xi;
2236 bcopy(inp, &xi.xi_inp, sizeof *inp);
2237 if (inp->inp_socket)
2238 sotoxsocket(inp->inp_socket, &xi.xi_socket);
2239 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
2243 LIST_REMOVE(marker, inp_list);
2245 REL_PCBINFO_TOKEN(pcbinfo);
2247 if (error == 0 && i < n) {
2248 bzero(&xi, sizeof xi);
2249 xi.xi_len = sizeof xi;
2251 error = SYSCTL_OUT(req, &xi, sizeof xi);
2259 lwkt_migratecpu(origcpu);
2260 kfree(marker, M_TEMP);
2265 in_pcblist_global_ncpus2(SYSCTL_HANDLER_ARGS)
2267 return in_pcblist_global(oidp, arg1, ncpus2, req);
2271 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2273 struct sockaddr_in *sin;
2275 KASSERT(faddr->sa_family == AF_INET,
2276 ("not AF_INET faddr %d", faddr->sa_family));
2278 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2279 sin->sin_family = AF_INET;
2280 sin->sin_len = sizeof(*sin);
2281 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2282 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2284 so->so_faddr = (struct sockaddr *)sin;
2288 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2289 boolean_t shared, u_short offset)
2291 memset(portinfo, 0, sizeof(*portinfo));
2293 portinfo->offset = offset;
2294 portinfo->lastport = offset;
2295 portinfo->lastlow = offset;
2296 portinfo->lasthi = offset;
2298 portinfo->porthashbase = hashinit(hashsize, M_PCB,
2299 &portinfo->porthashmask);
2302 portinfo->porttoken = kmalloc(sizeof(struct lwkt_token),
2304 lwkt_token_init(portinfo->porttoken, "porttoken");
2309 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step)
2319 hi = rounddown2(hi, step);
2324 lo = roundup2(lo, step);
2334 in_pcbglobalinit(void)
2338 in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB,
2340 in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer),
2341 M_PCB, M_WAITOK | M_ZERO);
2343 for (cpu = 0; cpu < ncpus; ++cpu) {
2344 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu];
2345 struct inpcb *marker = &in_pcbmarkers[cpu];
2347 marker->inp_flags |= INP_PLACEMARKER;
2348 ic->ic_inp = marker;
2353 in_pcbmarker(int cpuid)
2355 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2356 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2358 return &in_pcbmarkers[cpuid];
2361 struct inpcontainer *
2362 in_pcbcontainer_marker(int cpuid)
2364 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2365 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2367 return &in_pcbcontainer_markers[cpuid];