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(struct inpcb *inp, struct inpcbinfo *pcbinfo)
320 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
321 KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0,
325 inp->inp_pcbinfo = NULL;
329 * Relink a pcb into a new pcbinfo.
332 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
334 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
335 KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0,
338 inp->inp_pcbinfo = pcbinfo;
343 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
345 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
346 struct inpcbportinfo *portinfo;
347 u_short first, last, lport, step;
350 int portinfo_first, portinfo_idx;
352 inp->inp_flags |= INP_ANONPORT;
354 step = pcbinfo->portinfo_mask + 1;
355 portinfo_first = mycpuid & pcbinfo->portinfo_mask;
356 portinfo_idx = portinfo_first;
358 portinfo = &pcbinfo->portinfo[portinfo_idx];
360 if (inp->inp_flags & INP_HIGHPORT) {
361 first = ipport_hifirstauto; /* sysctl */
362 last = ipport_hilastauto;
363 lastport = &portinfo->lasthi;
364 } else if (inp->inp_flags & INP_LOWPORT) {
367 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
368 inp->inp_laddr.s_addr = INADDR_ANY;
371 first = ipport_lowfirstauto; /* 1023 */
372 last = ipport_lowlastauto; /* 600 */
373 lastport = &portinfo->lastlow;
375 first = ipport_firstauto; /* sysctl */
376 last = ipport_lastauto;
377 lastport = &portinfo->lastport;
381 * This has to be atomic. If the porthash is shared across multiple
382 * protocol threads (aka tcp) then the token must be held.
384 GET_PORT_TOKEN(portinfo);
387 * Simple check to ensure all ports are not used up causing
390 * We split the two cases (up and down) so that the direction
391 * is not being tested on each round of the loop.
397 in_pcbportrange(&first, &last, portinfo->offset, step);
398 count = (first - last) / step;
401 if (count-- < 0) { /* completely used? */
402 error = EADDRNOTAVAIL;
406 if (*lastport > first || *lastport < last)
408 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
410 lport = htons(*lastport);
411 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
417 in_pcbportrange(&last, &first, portinfo->offset, step);
418 count = (last - first) / step;
421 if (count-- < 0) { /* completely used? */
422 error = EADDRNOTAVAIL;
426 if (*lastport < first || *lastport > last)
428 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
430 lport = htons(*lastport);
431 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
434 inp->inp_lport = lport;
435 in_pcbinsporthash(portinfo, inp);
438 REL_PORT_TOKEN(portinfo);
441 /* Try next portinfo */
443 portinfo_idx &= pcbinfo->portinfo_mask;
444 if (portinfo_idx != portinfo_first)
446 inp->inp_laddr.s_addr = INADDR_ANY;
452 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
454 struct socket *so = inp->inp_socket;
455 struct sockaddr_in jsin;
456 struct ucred *cred = NULL;
459 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
460 return (EADDRNOTAVAIL);
461 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
462 return (EINVAL); /* already bound */
464 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
465 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
467 cred = td->td_proc->p_ucred;
470 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
471 struct inpcbinfo *pcbinfo;
472 struct inpcbportinfo *portinfo;
474 u_short lport, lport_ho;
475 int reuseport = (so->so_options & SO_REUSEPORT);
478 if (nam->sa_len != sizeof *sin)
482 * We should check the family, but old programs
483 * incorrectly fail to initialize it.
485 if (sin->sin_family != AF_INET)
486 return (EAFNOSUPPORT);
488 if (!prison_replace_wildcards(td, nam))
491 lport = sin->sin_port;
492 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
494 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
495 * allow complete duplication of binding if
496 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
497 * and a multicast address is bound on both
498 * new and duplicated sockets.
500 if (so->so_options & SO_REUSEADDR)
501 reuseport = SO_REUSEADDR | SO_REUSEPORT;
502 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
503 sin->sin_port = 0; /* yech... */
504 bzero(&sin->sin_zero, sizeof sin->sin_zero);
505 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
506 return (EADDRNOTAVAIL);
509 inp->inp_laddr = sin->sin_addr;
511 jsin.sin_family = AF_INET;
512 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
513 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
514 inp->inp_laddr.s_addr = INADDR_ANY;
517 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
520 /* Auto-select local port */
521 return in_pcbsetlport(inp, wild, cred);
523 lport_ho = ntohs(lport);
526 if (lport_ho < IPPORT_RESERVED && cred &&
528 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
529 inp->inp_laddr.s_addr = INADDR_ANY;
534 * Locate the proper portinfo based on lport
536 pcbinfo = inp->inp_pcbinfo;
538 &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
539 KKASSERT((lport_ho & pcbinfo->portinfo_mask) ==
543 * This has to be atomic. If the porthash is shared across
544 * multiple protocol threads (aka tcp) then the token must
547 GET_PORT_TOKEN(portinfo);
549 if (so->so_cred->cr_uid != 0 &&
550 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
551 t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
552 INPLOOKUP_WILDCARD, cred);
554 (!in_nullhost(sin->sin_addr) ||
555 !in_nullhost(t->inp_laddr) ||
556 (t->inp_socket->so_options & SO_REUSEPORT) == 0) &&
557 (so->so_cred->cr_uid !=
558 t->inp_socket->so_cred->cr_uid)) {
560 if (!in_nullhost(sin->sin_addr) ||
561 !in_nullhost(t->inp_laddr) ||
562 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
565 inp->inp_laddr.s_addr = INADDR_ANY;
571 if (cred && !prison_replace_wildcards(td, nam)) {
572 inp->inp_laddr.s_addr = INADDR_ANY;
573 error = EADDRNOTAVAIL;
576 t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
578 if (t && !(reuseport & t->inp_socket->so_options)) {
580 if (!in_nullhost(sin->sin_addr) ||
581 !in_nullhost(t->inp_laddr) ||
582 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
585 inp->inp_laddr.s_addr = INADDR_ANY;
590 inp->inp_lport = lport;
591 in_pcbinsporthash(portinfo, inp);
594 REL_PORT_TOKEN(portinfo);
597 jsin.sin_family = AF_INET;
598 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
599 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
600 inp->inp_laddr.s_addr = INADDR_ANY;
603 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
605 return in_pcbsetlport(inp, wild, cred);
609 static struct inpcb *
610 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr,
611 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
614 struct inpcbporthead *porthash;
615 struct inpcbport *phd;
616 struct inpcb *match = NULL;
619 * If the porthashbase is shared across several cpus, it must
622 ASSERT_PORT_TOKEN_HELD(portinfo);
625 * Best fit PCB lookup.
627 * First see if this local port is in use by looking on the
630 porthash = &portinfo->porthashbase[
631 INP_PCBPORTHASH(lport, portinfo->porthashmask)];
632 LIST_FOREACH(phd, porthash, phd_hash) {
633 if (phd->phd_port == lport)
637 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
639 if ((inp->inp_vflag & INP_IPV4) == 0)
642 if (inp->inp_laddr.s_addr != INADDR_ANY &&
643 inp->inp_laddr.s_addr != laddr.s_addr)
646 if (inp->inp_faddr.s_addr != INADDR_ANY &&
647 inp->inp_faddr.s_addr != faddr.s_addr)
650 if (inp->inp_fport != 0 && inp->inp_fport != fport)
655 inp->inp_socket->so_cred->cr_prison) {
665 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote,
668 struct proc *p = td->td_proc;
669 unsigned short *lastport;
670 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
671 struct sockaddr_in jsin;
672 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
673 struct inpcbportinfo *portinfo;
674 struct ucred *cred = NULL;
675 u_short first, last, lport, step;
676 int count, error, dup;
677 int portinfo_first, portinfo_idx;
679 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
680 return (EADDRNOTAVAIL);
682 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
683 if (inp->inp_lport != 0)
684 return (EINVAL); /* already bound */
689 jsin.sin_family = AF_INET;
690 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
691 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
692 inp->inp_laddr.s_addr = INADDR_ANY;
695 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
697 inp->inp_flags |= INP_ANONPORT;
699 step = pcbinfo->portinfo_mask + 1;
700 portinfo_first = mycpuid & pcbinfo->portinfo_mask;
701 portinfo_idx = portinfo_first;
703 portinfo = &pcbinfo->portinfo[portinfo_idx];
706 if (inp->inp_flags & INP_HIGHPORT) {
707 first = ipport_hifirstauto; /* sysctl */
708 last = ipport_hilastauto;
709 lastport = &portinfo->lasthi;
710 } else if (inp->inp_flags & INP_LOWPORT) {
713 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
714 inp->inp_laddr.s_addr = INADDR_ANY;
717 first = ipport_lowfirstauto; /* 1023 */
718 last = ipport_lowlastauto; /* 600 */
719 lastport = &portinfo->lastlow;
721 first = ipport_firstauto; /* sysctl */
722 last = ipport_lastauto;
723 lastport = &portinfo->lastport;
727 * This has to be atomic. If the porthash is shared across multiple
728 * protocol threads (aka tcp) then the token must be held.
730 GET_PORT_TOKEN(portinfo);
734 * Simple check to ensure all ports are not used up causing
737 * We split the two cases (up and down) so that the direction
738 * is not being tested on each round of the loop.
744 in_pcbportrange(&first, &last, portinfo->offset, step);
745 count = (first - last) / step;
748 if (count-- < 0) { /* completely used? */
749 error = EADDRNOTAVAIL;
753 if (*lastport > first || *lastport < last)
755 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
757 lport = htons(*lastport);
758 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
759 lport, sin->sin_addr, sin->sin_port, cred));
764 in_pcbportrange(&last, &first, portinfo->offset, step);
765 count = (last - first) / step;
768 if (count-- < 0) { /* completely used? */
769 error = EADDRNOTAVAIL;
773 if (*lastport < first || *lastport > last)
775 KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
777 lport = htons(*lastport);
778 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
779 lport, sin->sin_addr, sin->sin_port, cred));
782 /* This could happen on loopback interface */
783 if (sin->sin_port == lport &&
784 sin->sin_addr.s_addr == inp->inp_laddr.s_addr) {
787 * Duplicate again; give up
789 error = EADDRNOTAVAIL;
795 inp->inp_lport = lport;
796 in_pcbinsporthash(portinfo, inp);
799 REL_PORT_TOKEN(portinfo);
802 /* Try next portinfo */
804 portinfo_idx &= pcbinfo->portinfo_mask;
805 if (portinfo_idx != portinfo_first)
807 inp->inp_laddr.s_addr = INADDR_ANY;
813 * Transform old in_pcbconnect() into an inner subroutine for new
814 * in_pcbconnect(): Do some validity-checking on the remote
815 * address (in mbuf 'nam') and then determine local host address
816 * (i.e., which interface) to use to access that remote host.
818 * This preserves definition of in_pcbconnect(), while supporting a
819 * slightly different version for T/TCP. (This is more than
820 * a bit of a kludge, but cleaning up the internal interfaces would
821 * have forced minor changes in every protocol).
824 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
825 struct sockaddr_in **plocal_sin, struct thread *td, int find)
827 struct in_ifaddr *ia;
828 struct ucred *cred = NULL;
829 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
830 struct sockaddr *jsin;
831 int jailed = 0, alloc_route = 0;
833 if (nam->sa_len != sizeof *sin)
835 if (sin->sin_family != AF_INET)
836 return (EAFNOSUPPORT);
837 if (sin->sin_port == 0)
838 return (EADDRNOTAVAIL);
839 if (td && td->td_proc && td->td_proc->p_ucred)
840 cred = td->td_proc->p_ucred;
841 if (cred && cred->cr_prison)
843 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
844 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
846 * If the destination address is INADDR_ANY,
847 * use the primary local address.
848 * If the supplied address is INADDR_BROADCAST,
849 * and the primary interface supports broadcast,
850 * choose the broadcast address for that interface.
852 if (sin->sin_addr.s_addr == INADDR_ANY)
853 sin->sin_addr = IA_SIN(ia)->sin_addr;
854 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
855 (ia->ia_ifp->if_flags & IFF_BROADCAST))
856 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
863 * If route is known or can be allocated now,
864 * our src addr is taken from the i/f, else punt.
865 * Note that we should check the address family of the cached
866 * destination, in case of sharing the cache with IPv6.
868 ro = &inp->inp_route;
870 (!(ro->ro_rt->rt_flags & RTF_UP) ||
871 ro->ro_dst.sa_family != AF_INET ||
872 satosin(&ro->ro_dst)->sin_addr.s_addr !=
873 sin->sin_addr.s_addr ||
874 inp->inp_socket->so_options & SO_DONTROUTE)) {
878 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
879 (ro->ro_rt == NULL ||
880 ro->ro_rt->rt_ifp == NULL)) {
881 /* No route yet, so try to acquire one */
882 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
883 ro->ro_dst.sa_family = AF_INET;
884 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
885 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
891 * If we found a route, use the address
892 * corresponding to the outgoing interface
893 * unless it is the loopback (in case a route
894 * to our address on another net goes to loopback).
897 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
899 if (jailed_ip(cred->cr_prison,
900 ro->ro_rt->rt_ifa->ifa_addr)) {
901 ia = ifatoia(ro->ro_rt->rt_ifa);
904 ia = ifatoia(ro->ro_rt->rt_ifa);
908 u_short fport = sin->sin_port;
911 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
912 if (ia && jailed && !jailed_ip(cred->cr_prison,
913 sintosa(&ia->ia_addr)))
916 ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
917 if (ia && jailed && !jailed_ip(cred->cr_prison,
918 sintosa(&ia->ia_addr)))
920 sin->sin_port = fport;
922 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
923 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
924 if (ia && jailed && !jailed_ip(cred->cr_prison,
925 sintosa(&ia->ia_addr)))
928 if (!jailed && ia == NULL)
932 * If the destination address is multicast and an outgoing
933 * interface has been set as a multicast option, use the
934 * address of that interface as our source address.
936 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
937 inp->inp_moptions != NULL) {
938 struct ip_moptions *imo;
941 imo = inp->inp_moptions;
942 if (imo->imo_multicast_ifp != NULL) {
943 struct in_ifaddr_container *iac;
945 ifp = imo->imo_multicast_ifp;
948 &in_ifaddrheads[mycpuid], ia_link) {
949 if (iac->ia->ia_ifp == ifp) {
959 * Don't do pcblookup call here; return interface in plocal_sin
960 * and exit to caller, that will do the lookup.
962 if (ia == NULL && jailed) {
963 if ((jsin = prison_get_nonlocal(
964 cred->cr_prison, AF_INET, NULL)) != NULL ||
965 (jsin = prison_get_local(
966 cred->cr_prison, AF_INET, NULL)) != NULL) {
967 *plocal_sin = satosin(jsin);
973 *plocal_sin = &ia->ia_addr;
979 struct route *ro = &inp->inp_route;
981 if (ro->ro_rt != NULL)
983 bzero(ro, sizeof(*ro));
985 return (EADDRNOTAVAIL);
989 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
990 struct sockaddr_in **plocal_sin, struct thread *td)
992 return in_pcbladdr_find(inp, nam, plocal_sin, td,
993 (inp->inp_laddr.s_addr == INADDR_ANY));
998 * Connect from a socket to a specified address.
999 * Both address and port must be specified in argument sin.
1000 * If don't have a local address for this socket yet,
1004 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
1006 struct sockaddr_in *if_sin;
1007 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1010 /* Call inner routine to assign local interface address. */
1011 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
1014 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port,
1015 inp->inp_laddr.s_addr ?
1016 inp->inp_laddr : if_sin->sin_addr,
1017 inp->inp_lport, FALSE, NULL) != NULL) {
1018 return (EADDRINUSE);
1020 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1021 if (inp->inp_lport == 0) {
1022 error = in_pcbbind(inp, NULL, td);
1026 inp->inp_laddr = if_sin->sin_addr;
1028 inp->inp_faddr = sin->sin_addr;
1029 inp->inp_fport = sin->sin_port;
1030 in_pcbinsconnhash(inp);
1035 in_pcbdisconnect(struct inpcb *inp)
1038 in_pcbremconnhash(inp);
1039 inp->inp_faddr.s_addr = INADDR_ANY;
1044 in_pcbdetach(struct inpcb *inp)
1046 struct socket *so = inp->inp_socket;
1047 struct inpcbinfo *ipi = inp->inp_pcbinfo;
1050 ipsec4_delete_pcbpolicy(inp);
1052 inp->inp_gencnt = ++ipi->ipi_gencnt;
1053 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
1054 in_pcbremlists(inp);
1056 sofree(so); /* remove pcb ref */
1057 if (inp->inp_options)
1058 m_free(inp->inp_options);
1059 if (inp->inp_route.ro_rt)
1060 rtfree(inp->inp_route.ro_rt);
1061 ip_freemoptions(inp->inp_moptions);
1067 * The calling convention of in_setsockaddr() and in_setpeeraddr() was
1068 * modified to match the pru_sockaddr() and pru_peeraddr() entry points
1069 * in struct pr_usrreqs, so that protocols can just reference then directly
1070 * without the need for a wrapper function. The socket must have a valid
1071 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one
1072 * except through a kernel programming error, so it is acceptable to panic
1073 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap
1074 * because there actually /is/ a programming error somewhere... XXX)
1077 in_setsockaddr(struct socket *so, struct sockaddr **nam)
1080 struct sockaddr_in *sin;
1083 * Do the malloc first in case it blocks.
1085 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1086 sin->sin_family = AF_INET;
1087 sin->sin_len = sizeof *sin;
1093 kfree(sin, M_SONAME);
1094 return (ECONNRESET);
1096 sin->sin_port = inp->inp_lport;
1097 sin->sin_addr = inp->inp_laddr;
1100 *nam = (struct sockaddr *)sin;
1105 in_setsockaddr_dispatch(netmsg_t msg)
1109 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1110 lwkt_replymsg(&msg->lmsg, error);
1114 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1117 struct sockaddr_in *sin;
1120 * Do the malloc first in case it blocks.
1122 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1123 sin->sin_family = AF_INET;
1124 sin->sin_len = sizeof *sin;
1130 kfree(sin, M_SONAME);
1131 return (ECONNRESET);
1133 sin->sin_port = inp->inp_fport;
1134 sin->sin_addr = inp->inp_faddr;
1137 *nam = (struct sockaddr *)sin;
1142 in_setpeeraddr_dispatch(netmsg_t msg)
1146 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1147 lwkt_replymsg(&msg->lmsg, error);
1151 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err,
1152 void (*notify)(struct inpcb *, int))
1154 struct inpcb *inp, *marker;
1156 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1157 ("not in the correct netisr"));
1158 marker = &in_pcbmarkers[mycpuid];
1162 * - If INP_PLACEMARKER is set we must ignore the rest of the
1163 * structure and skip it.
1164 * - It is safe to nuke inpcbs here, since we are in their own
1167 GET_PCBINFO_TOKEN(pcbinfo);
1169 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1170 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1171 LIST_REMOVE(marker, inp_list);
1172 LIST_INSERT_AFTER(inp, marker, inp_list);
1174 if (inp->inp_flags & INP_PLACEMARKER)
1177 if (!(inp->inp_vflag & INP_IPV4))
1180 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1181 inp->inp_socket == NULL)
1183 (*notify)(inp, err); /* can remove inp from list! */
1185 LIST_REMOVE(marker, inp_list);
1187 REL_PCBINFO_TOKEN(pcbinfo);
1191 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1193 struct inpcb *inp, *marker;
1196 * We only need to make sure that we are in netisr0, where all
1197 * multicast operation happen. We could check inpcbinfo which
1198 * does not belong to netisr0 by holding the inpcbinfo's token.
1199 * In this case, the pcbinfo must be able to be shared, i.e.
1200 * pcbinfo->infotoken is not NULL.
1202 KASSERT(&curthread->td_msgport == netisr_cpuport(0),
1203 ("not in netisr0"));
1204 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL,
1205 ("pcbinfo could not be shared"));
1208 * Get a marker for the current netisr (netisr0).
1210 * It is possible that the multicast address deletion blocks,
1211 * which could cause temporary token releasing. So we use
1212 * inpcb marker here to get a coherent view of the inpcb list.
1214 * While, on the other hand, moptions are only added and deleted
1215 * in netisr0, so we would not see staled moption or miss moption
1216 * even if the token was released due to the blocking multicast
1219 marker = &in_pcbmarkers[mycpuid];
1221 GET_PCBINFO_TOKEN(pcbinfo);
1223 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1224 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1225 struct ip_moptions *imo;
1227 LIST_REMOVE(marker, inp_list);
1228 LIST_INSERT_AFTER(inp, marker, inp_list);
1230 if (inp->inp_flags & INP_PLACEMARKER)
1232 imo = inp->inp_moptions;
1233 if ((inp->inp_vflag & INP_IPV4) && imo != NULL) {
1237 * Unselect the outgoing interface if it is being
1240 if (imo->imo_multicast_ifp == ifp)
1241 imo->imo_multicast_ifp = NULL;
1244 * Drop multicast group membership if we joined
1245 * through the interface being detached.
1247 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1249 if (imo->imo_membership[i]->inm_ifp == ifp) {
1252 * This could block and the pcbinfo
1253 * token could be passively released.
1255 in_delmulti(imo->imo_membership[i]);
1257 } else if (gap != 0)
1258 imo->imo_membership[i - gap] =
1259 imo->imo_membership[i];
1261 imo->imo_num_memberships -= gap;
1264 LIST_REMOVE(marker, inp_list);
1266 REL_PCBINFO_TOKEN(pcbinfo);
1270 * Check for alternatives when higher level complains
1271 * about service problems. For now, invalidate cached
1272 * routing information. If the route was created dynamically
1273 * (by a redirect), time to try a default gateway again.
1276 in_losing(struct inpcb *inp)
1279 struct rt_addrinfo rtinfo;
1281 if ((rt = inp->inp_route.ro_rt)) {
1282 bzero(&rtinfo, sizeof(struct rt_addrinfo));
1283 rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1284 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1285 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1286 rtinfo.rti_flags = rt->rt_flags;
1287 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1288 if (rt->rt_flags & RTF_DYNAMIC) {
1289 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1290 rt_mask(rt), rt->rt_flags, NULL);
1292 inp->inp_route.ro_rt = NULL;
1295 * A new route can be allocated
1296 * the next time output is attempted.
1302 * After a routing change, flush old routing
1303 * and allocate a (hopefully) better one.
1306 in_rtchange(struct inpcb *inp, int err)
1308 if (inp->inp_route.ro_rt) {
1309 rtfree(inp->inp_route.ro_rt);
1310 inp->inp_route.ro_rt = NULL;
1312 * A new route can be allocated the next time
1313 * output is attempted.
1319 * Lookup a PCB based on the local address and port.
1322 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr,
1323 u_int lport_arg, int wild_okay, struct ucred *cred)
1326 int matchwild = 3, wildcard;
1327 u_short lport = lport_arg;
1328 struct inpcbporthead *porthash;
1329 struct inpcbport *phd;
1330 struct inpcb *match = NULL;
1333 * If the porthashbase is shared across several cpus, it must
1336 ASSERT_PORT_TOKEN_HELD(portinfo);
1339 * Best fit PCB lookup.
1341 * First see if this local port is in use by looking on the
1344 porthash = &portinfo->porthashbase[
1345 INP_PCBPORTHASH(lport, portinfo->porthashmask)];
1346 LIST_FOREACH(phd, porthash, phd_hash) {
1347 if (phd->phd_port == lport)
1352 * Port is in use by one or more PCBs. Look for best
1355 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1358 if ((inp->inp_vflag & INP_IPV4) == 0)
1361 if (inp->inp_faddr.s_addr != INADDR_ANY)
1363 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1364 if (laddr.s_addr == INADDR_ANY)
1366 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1369 if (laddr.s_addr != INADDR_ANY)
1372 if (wildcard && !wild_okay)
1374 if (wildcard < matchwild &&
1377 inp->inp_socket->so_cred->cr_prison)) {
1379 matchwild = wildcard;
1380 if (matchwild == 0) {
1390 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1391 const struct inpcb *inp)
1393 const struct inp_localgrphead *hdr;
1394 const struct inp_localgroup *grp;
1397 if (pcbinfo->localgrphashbase == NULL)
1400 GET_PCBINFO_TOKEN(pcbinfo);
1402 hdr = &pcbinfo->localgrphashbase[
1403 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1405 LIST_FOREACH(grp, hdr, il_list) {
1406 if (grp->il_vflag == inp->inp_vflag &&
1407 grp->il_lport == inp->inp_lport &&
1408 memcmp(&grp->il_dependladdr,
1409 &inp->inp_inc.inc_ie.ie_dependladdr,
1410 sizeof(grp->il_dependladdr)) == 0) {
1414 if (grp == NULL || grp->il_inpcnt == 1) {
1415 REL_PCBINFO_TOKEN(pcbinfo);
1419 KASSERT(grp->il_inpcnt >= 2,
1420 ("invalid localgroup inp count %d", grp->il_inpcnt));
1421 for (i = 0; i < grp->il_inpcnt; ++i) {
1422 if (grp->il_inp[i] == inp) {
1423 int last = grp->il_inpcnt - 1;
1426 last = grp->il_inpcnt - 2;
1427 REL_PCBINFO_TOKEN(pcbinfo);
1428 return grp->il_inp[last];
1431 REL_PCBINFO_TOKEN(pcbinfo);
1435 static struct inpcb *
1436 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1437 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1439 struct inpcb *local_wild = NULL;
1440 const struct inp_localgrphead *hdr;
1441 const struct inp_localgroup *grp;
1443 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1445 hdr = &pcbinfo->localgrphashbase[
1446 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1449 * Order of socket selection:
1454 * - Local group does not contain jailed sockets
1455 * - Local group does not contain IPv4 mapped INET6 wild sockets
1457 LIST_FOREACH(grp, hdr, il_list) {
1459 if (!(grp->il_vflag & INP_IPV4))
1462 if (grp->il_lport == lport) {
1466 * Modulo-N is used here, which greatly reduces
1467 * completion queue token contention, thus more
1468 * cpu time is saved.
1470 idx = pkt_hash % grp->il_inpcnt;
1471 if (grp->il_laddr.s_addr == laddr.s_addr)
1472 return grp->il_inp[idx];
1473 else if (grp->il_laddr.s_addr == INADDR_ANY)
1474 local_wild = grp->il_inp[idx];
1477 if (local_wild != NULL)
1483 * Lookup PCB in hash list.
1486 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1487 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1488 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1490 struct inpcbhead *head;
1491 struct inpcb *inp, *jinp=NULL;
1492 u_short fport = fport_arg, lport = lport_arg;
1495 * First look for an exact match.
1497 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1498 laddr.s_addr, lport, pcbinfo->hashmask)];
1499 LIST_FOREACH(inp, head, inp_hash) {
1501 if (!(inp->inp_vflag & INP_IPV4))
1504 if (in_hosteq(inp->inp_faddr, faddr) &&
1505 in_hosteq(inp->inp_laddr, laddr) &&
1506 inp->inp_fport == fport && inp->inp_lport == lport) {
1508 if (inp->inp_socket == NULL ||
1509 inp->inp_socket->so_cred->cr_prison == NULL) {
1521 struct inpcb *local_wild = NULL;
1522 struct inpcb *jinp_wild = NULL;
1524 struct inpcb *local_wild_mapped = NULL;
1526 struct inpcontainer *ic;
1527 struct inpcontainerhead *chead;
1528 struct sockaddr_in jsin;
1531 GET_PCBINFO_TOKEN(pcbinfo);
1534 * Check local group first
1536 if (pcbinfo->localgrphashbase != NULL &&
1537 m != NULL && (m->m_flags & M_HASH) &&
1538 !(ifp && ifp->if_type == IFT_FAITH)) {
1539 inp = inp_localgroup_lookup(pcbinfo,
1540 laddr, lport, m->m_pkthdr.hash);
1542 REL_PCBINFO_TOKEN(pcbinfo);
1548 * Order of socket selection:
1549 * 1. non-jailed, non-wild.
1550 * 2. non-jailed, wild.
1551 * 3. jailed, non-wild.
1554 jsin.sin_family = AF_INET;
1555 chead = &pcbinfo->wildcardhashbase[
1556 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1557 LIST_FOREACH(ic, chead, ic_list) {
1559 if (inp->inp_flags & INP_PLACEMARKER)
1562 jsin.sin_addr.s_addr = laddr.s_addr;
1564 if (!(inp->inp_vflag & INP_IPV4))
1567 if (inp->inp_socket != NULL)
1568 cred = inp->inp_socket->so_cred;
1571 if (cred != NULL && jailed(cred)) {
1575 if (!jailed_ip(cred->cr_prison,
1576 (struct sockaddr *)&jsin))
1579 if (inp->inp_lport == lport) {
1580 if (ifp && ifp->if_type == IFT_FAITH &&
1581 !(inp->inp_flags & INP_FAITH))
1583 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1584 if (cred != NULL && jailed(cred)) {
1587 REL_PCBINFO_TOKEN(pcbinfo);
1591 if (inp->inp_laddr.s_addr == INADDR_ANY) {
1593 if (INP_CHECK_SOCKAF(inp->inp_socket,
1595 local_wild_mapped = inp;
1607 REL_PCBINFO_TOKEN(pcbinfo);
1609 if (local_wild != NULL)
1610 return (local_wild);
1612 if (local_wild_mapped != NULL)
1613 return (local_wild_mapped);
1627 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1628 u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1629 boolean_t wildcard, struct ifnet *ifp)
1631 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1632 laddr, lport_arg, wildcard, ifp, NULL);
1636 * Insert PCB into connection hash table.
1639 in_pcbinsconnhash(struct inpcb *inp)
1641 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1642 struct inpcbhead *bucket;
1643 u_int32_t hashkey_faddr, hashkey_laddr;
1646 if (inp->inp_vflag & INP_IPV6) {
1647 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1648 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1651 hashkey_faddr = inp->inp_faddr.s_addr;
1652 hashkey_laddr = inp->inp_laddr.s_addr;
1657 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1658 ("not in the correct netisr"));
1659 KASSERT(!(inp->inp_flags & INP_WILDCARD), ("already on wildcardhash"));
1660 KASSERT(!(inp->inp_flags & INP_CONNECTED), ("already on connhash"));
1661 inp->inp_flags |= INP_CONNECTED;
1664 * Insert into the connection hash table.
1666 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1667 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1668 LIST_INSERT_HEAD(bucket, inp, inp_hash);
1672 * Remove PCB from connection hash table.
1675 in_pcbremconnhash(struct inpcb *inp)
1677 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo;
1679 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1680 ("not in the correct netisr"));
1681 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1683 LIST_REMOVE(inp, inp_hash);
1684 inp->inp_flags &= ~INP_CONNECTED;
1688 * Insert PCB into port hash table.
1691 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp)
1693 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1694 struct inpcbporthead *pcbporthash;
1695 struct inpcbport *phd;
1698 * If the porthashbase is shared across several cpus, it must
1701 ASSERT_PORT_TOKEN_HELD(portinfo);
1704 * Insert into the port hash table.
1706 pcbporthash = &portinfo->porthashbase[
1707 INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)];
1709 /* Go through port list and look for a head for this lport. */
1710 LIST_FOREACH(phd, pcbporthash, phd_hash) {
1711 if (phd->phd_port == inp->inp_lport)
1715 /* If none exists, use saved one and tack it on. */
1717 KKASSERT(pcbinfo->portsave != NULL);
1718 phd = pcbinfo->portsave;
1719 pcbinfo->portsave = NULL;
1720 phd->phd_port = inp->inp_lport;
1721 LIST_INIT(&phd->phd_pcblist);
1722 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1725 inp->inp_portinfo = portinfo;
1727 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1730 * Malloc one inpcbport for later use. It is safe to use
1731 * "wait" malloc here (port token would be released, if
1732 * malloc ever blocked), since all changes to the porthash
1735 if (pcbinfo->portsave == NULL) {
1736 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1737 M_PCB, M_INTWAIT | M_ZERO);
1742 in_pcbinsporthash_lport(struct inpcb *inp)
1744 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1745 struct inpcbportinfo *portinfo;
1748 /* Locate the proper portinfo based on lport */
1749 lport_ho = ntohs(inp->inp_lport);
1750 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
1751 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset);
1753 GET_PORT_TOKEN(portinfo);
1754 in_pcbinsporthash(portinfo, inp);
1755 REL_PORT_TOKEN(portinfo);
1758 static struct inp_localgroup *
1759 inp_localgroup_alloc(u_char vflag,
1760 uint16_t port, const union in_dependaddr *addr, int size)
1762 struct inp_localgroup *grp;
1764 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
1765 M_TEMP, M_INTWAIT | M_ZERO);
1766 grp->il_vflag = vflag;
1767 grp->il_lport = port;
1768 grp->il_dependladdr = *addr;
1769 grp->il_inpsiz = size;
1775 inp_localgroup_free(struct inp_localgroup *grp)
1781 inp_localgroup_destroy(struct inp_localgroup *grp)
1783 LIST_REMOVE(grp, il_list);
1784 inp_localgroup_free(grp);
1788 inp_localgroup_copy(struct inp_localgroup *grp,
1789 const struct inp_localgroup *old_grp)
1793 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
1794 ("invalid new local group size %d and old local group count %d",
1795 grp->il_inpsiz, old_grp->il_inpcnt));
1796 for (i = 0; i < old_grp->il_inpcnt; ++i)
1797 grp->il_inp[i] = old_grp->il_inp[i];
1798 grp->il_inpcnt = old_grp->il_inpcnt;
1802 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1804 struct inp_localgrphead *hdr;
1805 struct inp_localgroup *grp, *grp_alloc = NULL;
1809 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1811 if (pcbinfo->localgrphashbase == NULL)
1815 * XXX don't allow jailed socket to join local group
1817 if (inp->inp_socket != NULL)
1818 cred = inp->inp_socket->so_cred;
1821 if (cred != NULL && jailed(cred))
1826 * XXX don't allow IPv4 mapped INET6 wild socket
1828 if ((inp->inp_vflag & INP_IPV4) &&
1829 inp->inp_laddr.s_addr == INADDR_ANY &&
1830 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6))
1834 hdr = &pcbinfo->localgrphashbase[
1835 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1838 LIST_FOREACH(grp, hdr, il_list) {
1839 if (grp->il_vflag == inp->inp_vflag &&
1840 grp->il_lport == inp->inp_lport &&
1841 memcmp(&grp->il_dependladdr,
1842 &inp->inp_inc.inc_ie.ie_dependladdr,
1843 sizeof(grp->il_dependladdr)) == 0) {
1849 * Create a new local group
1851 if (grp_alloc == NULL) {
1852 grp_alloc = inp_localgroup_alloc(inp->inp_vflag,
1853 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
1854 INP_LOCALGROUP_SIZMIN);
1856 * Local group allocation could block and the
1857 * local group w/ the same property might have
1858 * been added by others when we were blocked;
1863 /* Local group has been allocated; link it */
1866 LIST_INSERT_HEAD(hdr, grp, il_list);
1868 } else if (grp->il_inpcnt == grp->il_inpsiz) {
1869 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
1870 static int limit_logged = 0;
1872 if (!limit_logged) {
1874 kprintf("local group port %d, "
1875 "limit reached\n", ntohs(grp->il_lport));
1877 if (grp_alloc != NULL) {
1879 * This would happen if the local group
1880 * w/ the same property was expanded when
1881 * our local group allocation blocked.
1883 inp_localgroup_free(grp_alloc);
1889 * Expand this local group
1891 if (grp_alloc == NULL ||
1892 grp->il_inpcnt >= grp_alloc->il_inpsiz) {
1893 if (grp_alloc != NULL)
1894 inp_localgroup_free(grp_alloc);
1895 grp_alloc = inp_localgroup_alloc(grp->il_vflag,
1896 grp->il_lport, &grp->il_dependladdr,
1897 grp->il_inpsiz * 2);
1899 * Local group allocation could block and the
1900 * local group w/ the same property might have
1901 * been expanded by others when we were blocked;
1908 * Save the old local group, link the new one, and then
1909 * destroy the old local group
1911 inp_localgroup_copy(grp_alloc, grp);
1912 LIST_INSERT_HEAD(hdr, grp_alloc, il_list);
1913 inp_localgroup_destroy(grp);
1919 * Found the local group
1921 if (grp_alloc != NULL) {
1923 * This would happen if the local group w/ the
1924 * same property was added or expanded when our
1925 * local group allocation blocked.
1927 inp_localgroup_free(grp_alloc);
1932 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
1933 ("invalid local group size %d and count %d",
1934 grp->il_inpsiz, grp->il_inpcnt));
1937 * Keep the local group sorted by the inpcb local group index
1938 * in ascending order.
1940 * This eases the multi-process userland application which uses
1941 * SO_REUSEPORT sockets and binds process to the owner cpu of
1942 * the SO_REUSEPORT socket:
1943 * If we didn't sort the local group by the inpcb local group
1944 * index and one of the process owning an inpcb in this local
1945 * group restarted, e.g. crashed and restarted by watchdog,
1946 * other processes owning a inpcb in this local group would have
1947 * to detect that event, refetch its socket's owner cpu, and
1950 idx = grp->il_inpcnt;
1951 for (i = 0; i < idx; ++i) {
1952 struct inpcb *oinp = grp->il_inp[i];
1954 if (oinp->inp_lgrpindex > i) {
1955 if (inp->inp_lgrpindex < 0) {
1956 inp->inp_lgrpindex = i;
1957 } else if (inp->inp_lgrpindex != i) {
1959 kprintf("inp %p: grpidx %d, "
1960 "assigned to %d, cpu%d\n",
1961 inp, inp->inp_lgrpindex, i,
1965 grp->il_inp[i] = inp;
1967 /* Pull down inpcbs */
1968 for (; i < grp->il_inpcnt; ++i) {
1969 struct inpcb *oinp1 = grp->il_inp[i + 1];
1971 grp->il_inp[i + 1] = oinp;
1979 if (inp->inp_lgrpindex < 0) {
1980 inp->inp_lgrpindex = idx;
1981 } else if (inp->inp_lgrpindex != idx) {
1983 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n",
1984 inp, inp->inp_lgrpindex, idx, mycpuid);
1987 grp->il_inp[idx] = inp;
1992 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1994 struct inpcontainer *ic;
1995 struct inpcontainerhead *bucket;
1997 GET_PCBINFO_TOKEN(pcbinfo);
1999 in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
2001 bucket = &pcbinfo->wildcardhashbase[
2002 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2004 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
2006 LIST_INSERT_HEAD(bucket, ic, ic_list);
2008 REL_PCBINFO_TOKEN(pcbinfo);
2012 * Insert PCB into wildcard hash table.
2015 in_pcbinswildcardhash(struct inpcb *inp)
2017 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2019 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2020 ("not in correct netisr"));
2021 KASSERT(!(inp->inp_flags & INP_CONNECTED),
2022 ("already on connhash"));
2023 KASSERT(!(inp->inp_flags & INP_WILDCARD),
2024 ("already on wildcardhash"));
2025 inp->inp_flags |= INP_WILDCARD;
2027 in_pcbinswildcardhash_oncpu(inp, pcbinfo);
2031 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2033 struct inp_localgrphead *hdr;
2034 struct inp_localgroup *grp;
2036 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
2038 if (pcbinfo->localgrphashbase == NULL)
2041 hdr = &pcbinfo->localgrphashbase[
2042 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
2044 LIST_FOREACH(grp, hdr, il_list) {
2047 for (i = 0; i < grp->il_inpcnt; ++i) {
2048 if (grp->il_inp[i] != inp)
2051 if (grp->il_inpcnt == 1) {
2052 /* Destroy this local group */
2053 inp_localgroup_destroy(grp);
2055 /* Pull up inpcbs */
2056 for (; i + 1 < grp->il_inpcnt; ++i)
2057 grp->il_inp[i] = grp->il_inp[i + 1];
2066 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2068 struct inpcontainer *ic;
2069 struct inpcontainerhead *head;
2071 GET_PCBINFO_TOKEN(pcbinfo);
2073 in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
2076 head = &pcbinfo->wildcardhashbase[
2077 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2079 LIST_FOREACH(ic, head, ic_list) {
2080 if (ic->ic_inp == inp)
2083 REL_PCBINFO_TOKEN(pcbinfo);
2084 return; /* not found! */
2087 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */
2088 REL_PCBINFO_TOKEN(pcbinfo);
2089 kfree(ic, M_TEMP); /* deallocate container */
2093 * Remove PCB from wildcard hash table.
2096 in_pcbremwildcardhash(struct inpcb *inp)
2098 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2100 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2101 ("not in correct netisr"));
2102 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
2104 in_pcbremwildcardhash_oncpu(inp, pcbinfo);
2105 inp->inp_lgrpindex = -1;
2106 inp->inp_flags &= ~INP_WILDCARD;
2110 * Remove PCB from various lists.
2113 in_pcbremlists(struct inpcb *inp)
2115 if (inp->inp_lport) {
2116 struct inpcbportinfo *portinfo;
2117 struct inpcbport *phd;
2121 * inp->inp_portinfo is _not_ necessary same as
2122 * inp->inp_pcbinfo->portinfo.
2124 portinfo = inp->inp_portinfo;
2125 GET_PORT_TOKEN(portinfo);
2128 LIST_REMOVE(inp, inp_portlist);
2129 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2130 LIST_REMOVE(phd, phd_hash);
2134 REL_PORT_TOKEN(portinfo);
2136 if (inp->inp_flags & INP_WILDCARD) {
2137 in_pcbremwildcardhash(inp);
2138 } else if (inp->inp_flags & INP_CONNECTED) {
2139 in_pcbremconnhash(inp);
2142 if (inp->inp_flags & INP_ONLIST)
2147 prison_xinpcb(struct thread *td, struct inpcb *inp)
2151 if (td->td_proc == NULL)
2153 cr = td->td_proc->p_ucred;
2154 if (cr->cr_prison == NULL)
2156 if (inp->inp_socket && inp->inp_socket->so_cred &&
2157 inp->inp_socket->so_cred->cr_prison &&
2158 cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
2164 in_pcblist_global(SYSCTL_HANDLER_ARGS)
2166 struct inpcbinfo *pcbinfo_arr = arg1;
2167 int pcbinfo_arrlen = arg2;
2168 struct inpcb *marker;
2172 KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1,
2173 ("invalid pcbinfo count %d", pcbinfo_arrlen));
2176 * The process of preparing the TCB list is too time-consuming and
2177 * resource-intensive to repeat twice on every request.
2180 if (req->oldptr == NULL) {
2181 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu)
2182 n += pcbinfo_arr[cpu].ipi_count;
2183 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
2187 if (req->newptr != NULL)
2190 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
2191 marker->inp_flags |= INP_PLACEMARKER;
2194 * OK, now we're committed to doing something. Re-fetch ipi_count
2195 * after obtaining the generation count.
2199 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) {
2200 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu];
2205 lwkt_migratecpu(cpu);
2207 GET_PCBINFO_TOKEN(pcbinfo);
2209 n = pcbinfo->ipi_count;
2211 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
2213 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
2214 LIST_REMOVE(marker, inp_list);
2215 LIST_INSERT_AFTER(inp, marker, inp_list);
2217 if (inp->inp_flags & INP_PLACEMARKER)
2219 if (prison_xinpcb(req->td, inp))
2222 bzero(&xi, sizeof xi);
2223 xi.xi_len = sizeof xi;
2224 bcopy(inp, &xi.xi_inp, sizeof *inp);
2225 if (inp->inp_socket)
2226 sotoxsocket(inp->inp_socket, &xi.xi_socket);
2227 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
2231 LIST_REMOVE(marker, inp_list);
2233 REL_PCBINFO_TOKEN(pcbinfo);
2235 if (error == 0 && i < n) {
2236 bzero(&xi, sizeof xi);
2237 xi.xi_len = sizeof xi;
2239 error = SYSCTL_OUT(req, &xi, sizeof xi);
2247 lwkt_migratecpu(origcpu);
2248 kfree(marker, M_TEMP);
2253 in_pcblist_global_ncpus2(SYSCTL_HANDLER_ARGS)
2255 return in_pcblist_global(oidp, arg1, ncpus2, req);
2259 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2261 struct sockaddr_in *sin;
2263 KASSERT(faddr->sa_family == AF_INET,
2264 ("not AF_INET faddr %d", faddr->sa_family));
2266 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2267 sin->sin_family = AF_INET;
2268 sin->sin_len = sizeof(*sin);
2269 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2270 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2272 so->so_faddr = (struct sockaddr *)sin;
2276 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2277 boolean_t shared, u_short offset)
2279 memset(portinfo, 0, sizeof(*portinfo));
2281 portinfo->offset = offset;
2282 portinfo->lastport = offset;
2283 portinfo->lastlow = offset;
2284 portinfo->lasthi = offset;
2286 portinfo->porthashbase = hashinit(hashsize, M_PCB,
2287 &portinfo->porthashmask);
2290 portinfo->porttoken = kmalloc(sizeof(struct lwkt_token),
2292 lwkt_token_init(portinfo->porttoken, "porttoken");
2297 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step)
2307 hi = rounddown2(hi, step);
2312 lo = roundup2(lo, step);
2322 in_pcbglobalinit(void)
2326 in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB,
2328 in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer),
2329 M_PCB, M_WAITOK | M_ZERO);
2331 for (cpu = 0; cpu < ncpus; ++cpu) {
2332 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu];
2333 struct inpcb *marker = &in_pcbmarkers[cpu];
2335 marker->inp_flags |= INP_PLACEMARKER;
2336 ic->ic_inp = marker;
2341 in_pcbmarker(int cpuid)
2343 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2344 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2346 return &in_pcbmarkers[cpuid];
2349 struct inpcontainer *
2350 in_pcbcontainer_marker(int cpuid)
2352 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2353 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2355 return &in_pcbcontainer_markers[cpuid];