2 * Copyright (c) 1982, 1986, 1991, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2007-2009 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Juniper Networks, Inc.
8 * Portions of this software were developed by Robert N. M. Watson under
9 * contract to Juniper Networks, Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
42 #include "opt_ipsec.h"
44 #include "opt_inet6.h"
45 #include "opt_ratelimit.h"
46 #include "opt_pcbgroup.h"
49 #include <sys/param.h>
50 #include <sys/systm.h>
52 #include <sys/malloc.h>
54 #include <sys/callout.h>
55 #include <sys/eventhandler.h>
56 #include <sys/domain.h>
57 #include <sys/protosw.h>
58 #include <sys/rmlock.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sockio.h>
64 #include <sys/refcount.h>
66 #include <sys/kernel.h>
67 #include <sys/sysctl.h>
76 #include <net/if_var.h>
77 #include <net/if_types.h>
78 #include <net/if_llatbl.h>
79 #include <net/route.h>
80 #include <net/rss_config.h>
83 #if defined(INET) || defined(INET6)
84 #include <netinet/in.h>
85 #include <netinet/in_pcb.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/udp.h>
89 #include <netinet/udp_var.h>
92 #include <netinet/in_var.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_pcb.h>
97 #include <netinet6/in6_var.h>
98 #include <netinet6/ip6_var.h>
103 #include <netipsec/ipsec.h>
104 #include <netipsec/key.h>
107 #include <security/mac/mac_framework.h>
109 static struct callout ipport_tick_callout;
112 * These configure the range of local port addresses assigned to
113 * "unspecified" outgoing connections/packets/whatever.
115 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
116 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
117 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
118 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
119 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
120 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
123 * Reserved ports accessible only to root. There are significant
124 * security considerations that must be accounted for when changing these,
125 * but the security benefits can be great. Please be careful.
127 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
128 VNET_DEFINE(int, ipport_reservedlow);
130 /* Variables dealing with random ephemeral port allocation. */
131 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
132 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
133 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
134 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
135 VNET_DEFINE(int, ipport_tcpallocs);
136 static VNET_DEFINE(int, ipport_tcplastcount);
138 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
140 static void in_pcbremlists(struct inpcb *inp);
142 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
143 struct in_addr faddr, u_int fport_arg,
144 struct in_addr laddr, u_int lport_arg,
145 int lookupflags, struct ifnet *ifp);
147 #define RANGECHK(var, min, max) \
148 if ((var) < (min)) { (var) = (min); } \
149 else if ((var) > (max)) { (var) = (max); }
152 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
156 error = sysctl_handle_int(oidp, arg1, arg2, req);
158 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
159 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
160 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
161 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
162 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
163 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
170 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
175 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
178 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
180 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
181 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
183 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
184 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
186 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
187 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
188 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
189 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
190 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
191 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
192 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
193 &VNET_NAME(ipport_reservedhigh), 0, "");
194 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
195 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
197 CTLFLAG_VNET | CTLFLAG_RW,
198 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
200 CTLFLAG_VNET | CTLFLAG_RW,
201 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
202 "allocations before switching to a sequental one");
203 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
204 CTLFLAG_VNET | CTLFLAG_RW,
205 &VNET_NAME(ipport_randomtime), 0,
206 "Minimum time to keep sequental port "
207 "allocation before switching to a random one");
211 * in_pcb.c: manage the Protocol Control Blocks.
213 * NOTE: It is assumed that most of these functions will be called with
214 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
215 * functions often modify hash chains or addresses in pcbs.
219 * Initialize an inpcbinfo -- we should be able to reduce the number of
223 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
224 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
225 char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
226 uint32_t inpcbzone_flags, u_int hashfields)
229 INP_INFO_LOCK_INIT(pcbinfo, name);
230 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
231 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
233 pcbinfo->ipi_vnet = curvnet;
235 pcbinfo->ipi_listhead = listhead;
236 LIST_INIT(pcbinfo->ipi_listhead);
237 pcbinfo->ipi_count = 0;
238 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
239 &pcbinfo->ipi_hashmask);
240 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
241 &pcbinfo->ipi_porthashmask);
243 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
245 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
246 NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
248 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
249 uma_zone_set_warning(pcbinfo->ipi_zone,
250 "kern.ipc.maxsockets limit reached");
254 * Destroy an inpcbinfo.
257 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
260 KASSERT(pcbinfo->ipi_count == 0,
261 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
263 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
264 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
265 pcbinfo->ipi_porthashmask);
267 in_pcbgroup_destroy(pcbinfo);
269 uma_zdestroy(pcbinfo->ipi_zone);
270 INP_LIST_LOCK_DESTROY(pcbinfo);
271 INP_HASH_LOCK_DESTROY(pcbinfo);
272 INP_INFO_LOCK_DESTROY(pcbinfo);
276 * Allocate a PCB and associate it with the socket.
277 * On success return with the PCB locked.
280 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
286 if (pcbinfo == &V_tcbinfo) {
287 INP_INFO_RLOCK_ASSERT(pcbinfo);
289 INP_INFO_WLOCK_ASSERT(pcbinfo);
294 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
297 bzero(inp, inp_zero_size);
298 inp->inp_pcbinfo = pcbinfo;
299 inp->inp_socket = so;
300 inp->inp_cred = crhold(so->so_cred);
301 inp->inp_inc.inc_fibnum = so->so_fibnum;
303 error = mac_inpcb_init(inp, M_NOWAIT);
306 mac_inpcb_create(so, inp);
309 error = ipsec_init_policy(so, &inp->inp_sp);
312 mac_inpcb_destroy(inp);
318 if (INP_SOCKAF(so) == AF_INET6) {
319 inp->inp_vflag |= INP_IPV6PROTO;
321 inp->inp_flags |= IN6P_IPV6_V6ONLY;
325 INP_LIST_WLOCK(pcbinfo);
326 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
327 pcbinfo->ipi_count++;
328 so->so_pcb = (caddr_t)inp;
330 if (V_ip6_auto_flowlabel)
331 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
333 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
334 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
335 INP_LIST_WUNLOCK(pcbinfo);
336 #if defined(IPSEC) || defined(MAC)
339 crfree(inp->inp_cred);
340 uma_zfree(pcbinfo->ipi_zone, inp);
348 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
352 INP_WLOCK_ASSERT(inp);
353 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
355 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
357 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
358 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
359 &inp->inp_lport, cred);
362 if (in_pcbinshash(inp) != 0) {
363 inp->inp_laddr.s_addr = INADDR_ANY;
368 inp->inp_flags |= INP_ANONPORT;
374 * Select a local port (number) to use.
376 #if defined(INET) || defined(INET6)
378 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
379 struct ucred *cred, int lookupflags)
381 struct inpcbinfo *pcbinfo;
382 struct inpcb *tmpinp;
383 unsigned short *lastport;
384 int count, dorandom, error;
385 u_short aux, first, last, lport;
387 struct in_addr laddr;
390 pcbinfo = inp->inp_pcbinfo;
393 * Because no actual state changes occur here, a global write lock on
394 * the pcbinfo isn't required.
396 INP_LOCK_ASSERT(inp);
397 INP_HASH_LOCK_ASSERT(pcbinfo);
399 if (inp->inp_flags & INP_HIGHPORT) {
400 first = V_ipport_hifirstauto; /* sysctl */
401 last = V_ipport_hilastauto;
402 lastport = &pcbinfo->ipi_lasthi;
403 } else if (inp->inp_flags & INP_LOWPORT) {
404 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
407 first = V_ipport_lowfirstauto; /* 1023 */
408 last = V_ipport_lowlastauto; /* 600 */
409 lastport = &pcbinfo->ipi_lastlow;
411 first = V_ipport_firstauto; /* sysctl */
412 last = V_ipport_lastauto;
413 lastport = &pcbinfo->ipi_lastport;
416 * For UDP(-Lite), use random port allocation as long as the user
417 * allows it. For TCP (and as of yet unknown) connections,
418 * use random port allocation only if the user allows it AND
419 * ipport_tick() allows it.
421 if (V_ipport_randomized &&
422 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
423 pcbinfo == &V_ulitecbinfo))
428 * It makes no sense to do random port allocation if
429 * we have the only port available.
433 /* Make sure to not include UDP(-Lite) packets in the count. */
434 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
435 V_ipport_tcpallocs++;
437 * Instead of having two loops further down counting up or down
438 * make sure that first is always <= last and go with only one
439 * code path implementing all logic.
448 /* Make the compiler happy. */
450 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
451 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
456 tmpinp = NULL; /* Make compiler happy. */
460 *lastport = first + (arc4random() % (last - first));
462 count = last - first;
465 if (count-- < 0) /* completely used? */
466 return (EADDRNOTAVAIL);
468 if (*lastport < first || *lastport > last)
470 lport = htons(*lastport);
473 if ((inp->inp_vflag & INP_IPV6) != 0)
474 tmpinp = in6_pcblookup_local(pcbinfo,
475 &inp->in6p_laddr, lport, lookupflags, cred);
477 #if defined(INET) && defined(INET6)
481 tmpinp = in_pcblookup_local(pcbinfo, laddr,
482 lport, lookupflags, cred);
484 } while (tmpinp != NULL);
487 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
488 laddrp->s_addr = laddr.s_addr;
496 * Return cached socket options.
499 inp_so_options(const struct inpcb *inp)
505 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
506 so_options |= SO_REUSEPORT;
507 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
508 so_options |= SO_REUSEADDR;
511 #endif /* INET || INET6 */
514 * Check if a new BINDMULTI socket is allowed to be created.
516 * ni points to the new inp.
517 * oi points to the exisitng inp.
519 * This checks whether the existing inp also has BINDMULTI and
520 * whether the credentials match.
523 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
525 /* Check permissions match */
526 if ((ni->inp_flags2 & INP_BINDMULTI) &&
527 (ni->inp_cred->cr_uid !=
528 oi->inp_cred->cr_uid))
531 /* Check the existing inp has BINDMULTI set */
532 if ((ni->inp_flags2 & INP_BINDMULTI) &&
533 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
537 * We're okay - either INP_BINDMULTI isn't set on ni, or
538 * it is and it matches the checks.
545 * Set up a bind operation on a PCB, performing port allocation
546 * as required, but do not actually modify the PCB. Callers can
547 * either complete the bind by setting inp_laddr/inp_lport and
548 * calling in_pcbinshash(), or they can just use the resulting
549 * port and address to authorise the sending of a once-off packet.
551 * On error, the values of *laddrp and *lportp are not changed.
554 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
555 u_short *lportp, struct ucred *cred)
557 struct socket *so = inp->inp_socket;
558 struct sockaddr_in *sin;
559 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
560 struct in_addr laddr;
562 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
566 * No state changes, so read locks are sufficient here.
568 INP_LOCK_ASSERT(inp);
569 INP_HASH_LOCK_ASSERT(pcbinfo);
571 if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
572 return (EADDRNOTAVAIL);
573 laddr.s_addr = *laddrp;
574 if (nam != NULL && laddr.s_addr != INADDR_ANY)
576 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
577 lookupflags = INPLOOKUP_WILDCARD;
579 if ((error = prison_local_ip4(cred, &laddr)) != 0)
582 sin = (struct sockaddr_in *)nam;
583 if (nam->sa_len != sizeof (*sin))
587 * We should check the family, but old programs
588 * incorrectly fail to initialize it.
590 if (sin->sin_family != AF_INET)
591 return (EAFNOSUPPORT);
593 error = prison_local_ip4(cred, &sin->sin_addr);
596 if (sin->sin_port != *lportp) {
597 /* Don't allow the port to change. */
600 lport = sin->sin_port;
602 /* NB: lport is left as 0 if the port isn't being changed. */
603 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
605 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
606 * allow complete duplication of binding if
607 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
608 * and a multicast address is bound on both
609 * new and duplicated sockets.
611 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
612 reuseport = SO_REUSEADDR|SO_REUSEPORT;
613 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
614 sin->sin_port = 0; /* yech... */
615 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
617 * Is the address a local IP address?
618 * If INP_BINDANY is set, then the socket may be bound
619 * to any endpoint address, local or not.
621 if ((inp->inp_flags & INP_BINDANY) == 0 &&
622 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
623 return (EADDRNOTAVAIL);
625 laddr = sin->sin_addr;
631 if (ntohs(lport) <= V_ipport_reservedhigh &&
632 ntohs(lport) >= V_ipport_reservedlow &&
633 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
636 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
637 priv_check_cred(inp->inp_cred,
638 PRIV_NETINET_REUSEPORT, 0) != 0) {
639 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
640 lport, INPLOOKUP_WILDCARD, cred);
643 * This entire block sorely needs a rewrite.
646 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
647 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
648 (so->so_type != SOCK_STREAM ||
649 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
650 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
651 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
652 (t->inp_flags2 & INP_REUSEPORT) == 0) &&
653 (inp->inp_cred->cr_uid !=
654 t->inp_cred->cr_uid))
658 * If the socket is a BINDMULTI socket, then
659 * the credentials need to match and the
660 * original socket also has to have been bound
663 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
666 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
667 lport, lookupflags, cred);
668 if (t && (t->inp_flags & INP_TIMEWAIT)) {
670 * XXXRW: If an incpb has had its timewait
671 * state recycled, we treat the address as
672 * being in use (for now). This is better
673 * than a panic, but not desirable.
677 (reuseport & tw->tw_so_options) == 0)
680 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
681 (reuseport & inp_so_options(t)) == 0) {
683 if (ntohl(sin->sin_addr.s_addr) !=
685 ntohl(t->inp_laddr.s_addr) !=
687 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
688 (t->inp_vflag & INP_IPV6PROTO) == 0)
691 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
699 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
704 *laddrp = laddr.s_addr;
710 * Connect from a socket to a specified address.
711 * Both address and port must be specified in argument sin.
712 * If don't have a local address for this socket yet,
716 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
717 struct ucred *cred, struct mbuf *m)
719 u_short lport, fport;
720 in_addr_t laddr, faddr;
723 INP_WLOCK_ASSERT(inp);
724 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
726 lport = inp->inp_lport;
727 laddr = inp->inp_laddr.s_addr;
728 anonport = (lport == 0);
729 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
734 /* Do the initial binding of the local address if required. */
735 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
736 inp->inp_lport = lport;
737 inp->inp_laddr.s_addr = laddr;
738 if (in_pcbinshash(inp) != 0) {
739 inp->inp_laddr.s_addr = INADDR_ANY;
745 /* Commit the remaining changes. */
746 inp->inp_lport = lport;
747 inp->inp_laddr.s_addr = laddr;
748 inp->inp_faddr.s_addr = faddr;
749 inp->inp_fport = fport;
750 in_pcbrehash_mbuf(inp, m);
753 inp->inp_flags |= INP_ANONPORT;
758 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
761 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
765 * Do proper source address selection on an unbound socket in case
766 * of connect. Take jails into account as well.
769 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
774 struct sockaddr_in *sin;
778 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
781 * Bypass source address selection and use the primary jail IP
784 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
788 bzero(&sro, sizeof(sro));
790 sin = (struct sockaddr_in *)&sro.ro_dst;
791 sin->sin_family = AF_INET;
792 sin->sin_len = sizeof(struct sockaddr_in);
793 sin->sin_addr.s_addr = faddr->s_addr;
796 * If route is known our src addr is taken from the i/f,
799 * Find out route to destination.
801 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
802 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
805 * If we found a route, use the address corresponding to
806 * the outgoing interface.
808 * Otherwise assume faddr is reachable on a directly connected
809 * network and try to find a corresponding interface to take
810 * the source address from.
812 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
813 struct in_ifaddr *ia;
816 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
817 inp->inp_socket->so_fibnum));
819 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
820 inp->inp_socket->so_fibnum));
826 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
827 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
828 ifa_free(&ia->ia_ifa);
833 ifa_free(&ia->ia_ifa);
836 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
839 if (sa->sa_family != AF_INET)
841 sin = (struct sockaddr_in *)sa;
842 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
843 ia = (struct in_ifaddr *)ifa;
848 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
849 IF_ADDR_RUNLOCK(ifp);
852 IF_ADDR_RUNLOCK(ifp);
854 /* 3. As a last resort return the 'default' jail address. */
855 error = prison_get_ip4(cred, laddr);
860 * If the outgoing interface on the route found is not
861 * a loopback interface, use the address from that interface.
862 * In case of jails do those three steps:
863 * 1. check if the interface address belongs to the jail. If so use it.
864 * 2. check if we have any address on the outgoing interface
865 * belonging to this jail. If so use it.
866 * 3. as a last resort return the 'default' jail address.
868 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
869 struct in_ifaddr *ia;
872 /* If not jailed, use the default returned. */
873 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
874 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
875 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
880 /* 1. Check if the iface address belongs to the jail. */
881 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
882 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
883 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
884 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
889 * 2. Check if we have any address on the outgoing interface
890 * belonging to this jail.
893 ifp = sro.ro_rt->rt_ifp;
895 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
897 if (sa->sa_family != AF_INET)
899 sin = (struct sockaddr_in *)sa;
900 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
901 ia = (struct in_ifaddr *)ifa;
906 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
907 IF_ADDR_RUNLOCK(ifp);
910 IF_ADDR_RUNLOCK(ifp);
912 /* 3. As a last resort return the 'default' jail address. */
913 error = prison_get_ip4(cred, laddr);
918 * The outgoing interface is marked with 'loopback net', so a route
919 * to ourselves is here.
920 * Try to find the interface of the destination address and then
921 * take the address from there. That interface is not necessarily
922 * a loopback interface.
923 * In case of jails, check that it is an address of the jail
924 * and if we cannot find, fall back to the 'default' jail address.
926 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
927 struct sockaddr_in sain;
928 struct in_ifaddr *ia;
930 bzero(&sain, sizeof(struct sockaddr_in));
931 sain.sin_family = AF_INET;
932 sain.sin_len = sizeof(struct sockaddr_in);
933 sain.sin_addr.s_addr = faddr->s_addr;
935 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
936 inp->inp_socket->so_fibnum));
938 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
939 inp->inp_socket->so_fibnum));
941 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
943 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
948 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
949 ifa_free(&ia->ia_ifa);
958 ifa_free(&ia->ia_ifa);
961 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
964 if (sa->sa_family != AF_INET)
966 sin = (struct sockaddr_in *)sa;
967 if (prison_check_ip4(cred,
968 &sin->sin_addr) == 0) {
969 ia = (struct in_ifaddr *)ifa;
974 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
975 IF_ADDR_RUNLOCK(ifp);
978 IF_ADDR_RUNLOCK(ifp);
981 /* 3. As a last resort return the 'default' jail address. */
982 error = prison_get_ip4(cred, laddr);
987 if (sro.ro_rt != NULL)
993 * Set up for a connect from a socket to the specified address.
994 * On entry, *laddrp and *lportp should contain the current local
995 * address and port for the PCB; these are updated to the values
996 * that should be placed in inp_laddr and inp_lport to complete
999 * On success, *faddrp and *fportp will be set to the remote address
1000 * and port. These are not updated in the error case.
1002 * If the operation fails because the connection already exists,
1003 * *oinpp will be set to the PCB of that connection so that the
1004 * caller can decide to override it. In all other cases, *oinpp
1008 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1009 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1010 struct inpcb **oinpp, struct ucred *cred)
1012 struct rm_priotracker in_ifa_tracker;
1013 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1014 struct in_ifaddr *ia;
1016 struct in_addr laddr, faddr;
1017 u_short lport, fport;
1021 * Because a global state change doesn't actually occur here, a read
1022 * lock is sufficient.
1024 INP_LOCK_ASSERT(inp);
1025 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1029 if (nam->sa_len != sizeof (*sin))
1031 if (sin->sin_family != AF_INET)
1032 return (EAFNOSUPPORT);
1033 if (sin->sin_port == 0)
1034 return (EADDRNOTAVAIL);
1035 laddr.s_addr = *laddrp;
1037 faddr = sin->sin_addr;
1038 fport = sin->sin_port;
1040 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1042 * If the destination address is INADDR_ANY,
1043 * use the primary local address.
1044 * If the supplied address is INADDR_BROADCAST,
1045 * and the primary interface supports broadcast,
1046 * choose the broadcast address for that interface.
1048 if (faddr.s_addr == INADDR_ANY) {
1049 IN_IFADDR_RLOCK(&in_ifa_tracker);
1051 IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1052 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1054 (error = prison_get_ip4(cred, &faddr)) != 0)
1056 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1057 IN_IFADDR_RLOCK(&in_ifa_tracker);
1058 if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1060 faddr = satosin(&TAILQ_FIRST(
1061 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1062 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1065 if (laddr.s_addr == INADDR_ANY) {
1066 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1068 * If the destination address is multicast and an outgoing
1069 * interface has been set as a multicast option, prefer the
1070 * address of that interface as our source address.
1072 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1073 inp->inp_moptions != NULL) {
1074 struct ip_moptions *imo;
1077 imo = inp->inp_moptions;
1078 if (imo->imo_multicast_ifp != NULL) {
1079 ifp = imo->imo_multicast_ifp;
1080 IN_IFADDR_RLOCK(&in_ifa_tracker);
1081 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1082 if ((ia->ia_ifp == ifp) &&
1084 prison_check_ip4(cred,
1085 &ia->ia_addr.sin_addr) == 0))
1089 error = EADDRNOTAVAIL;
1091 laddr = ia->ia_addr.sin_addr;
1094 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1100 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1101 laddr, lport, 0, NULL);
1105 return (EADDRINUSE);
1108 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1113 *laddrp = laddr.s_addr;
1115 *faddrp = faddr.s_addr;
1121 in_pcbdisconnect(struct inpcb *inp)
1124 INP_WLOCK_ASSERT(inp);
1125 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1127 inp->inp_faddr.s_addr = INADDR_ANY;
1134 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1135 * For most protocols, this will be invoked immediately prior to calling
1136 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1137 * socket, in which case in_pcbfree() is deferred.
1140 in_pcbdetach(struct inpcb *inp)
1143 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1146 if (inp->inp_snd_tag != NULL)
1147 in_pcbdetach_txrtlmt(inp);
1149 inp->inp_socket->so_pcb = NULL;
1150 inp->inp_socket = NULL;
1154 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1155 * stability of an inpcb pointer despite the inpcb lock being released. This
1156 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1157 * but where the inpcb lock may already held, or when acquiring a reference
1160 * in_pcbref() should be used only to provide brief memory stability, and
1161 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1162 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1163 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1164 * lock and rele are the *only* safe operations that may be performed on the
1167 * While the inpcb will not be freed, releasing the inpcb lock means that the
1168 * connection's state may change, so the caller should be careful to
1169 * revalidate any cached state on reacquiring the lock. Drop the reference
1170 * using in_pcbrele().
1173 in_pcbref(struct inpcb *inp)
1176 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1178 refcount_acquire(&inp->inp_refcount);
1182 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1183 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1184 * return a flag indicating whether or not the inpcb remains valid. If it is
1185 * valid, we return with the inpcb lock held.
1187 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1188 * reference on an inpcb. Historically more work was done here (actually, in
1189 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1190 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1191 * about memory stability (and continued use of the write lock).
1194 in_pcbrele_rlocked(struct inpcb *inp)
1196 struct inpcbinfo *pcbinfo;
1198 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1200 INP_RLOCK_ASSERT(inp);
1202 if (refcount_release(&inp->inp_refcount) == 0) {
1204 * If the inpcb has been freed, let the caller know, even if
1205 * this isn't the last reference.
1207 if (inp->inp_flags2 & INP_FREED) {
1214 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1217 pcbinfo = inp->inp_pcbinfo;
1218 uma_zfree(pcbinfo->ipi_zone, inp);
1223 in_pcbrele_wlocked(struct inpcb *inp)
1225 struct inpcbinfo *pcbinfo;
1227 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1229 INP_WLOCK_ASSERT(inp);
1231 if (refcount_release(&inp->inp_refcount) == 0) {
1233 * If the inpcb has been freed, let the caller know, even if
1234 * this isn't the last reference.
1236 if (inp->inp_flags2 & INP_FREED) {
1243 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1246 pcbinfo = inp->inp_pcbinfo;
1247 uma_zfree(pcbinfo->ipi_zone, inp);
1252 * Temporary wrapper.
1255 in_pcbrele(struct inpcb *inp)
1258 return (in_pcbrele_wlocked(inp));
1262 * Unconditionally schedule an inpcb to be freed by decrementing its
1263 * reference count, which should occur only after the inpcb has been detached
1264 * from its socket. If another thread holds a temporary reference (acquired
1265 * using in_pcbref()) then the free is deferred until that reference is
1266 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1267 * work, including removal from global lists, is done in this context, where
1268 * the pcbinfo lock is held.
1271 in_pcbfree(struct inpcb *inp)
1273 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1275 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1278 if (pcbinfo == &V_tcbinfo) {
1279 INP_INFO_LOCK_ASSERT(pcbinfo);
1281 INP_INFO_WLOCK_ASSERT(pcbinfo);
1284 INP_WLOCK_ASSERT(inp);
1286 /* XXXRW: Do as much as possible here. */
1288 if (inp->inp_sp != NULL)
1289 ipsec_delete_pcbpolicy(inp);
1291 INP_LIST_WLOCK(pcbinfo);
1292 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1293 in_pcbremlists(inp);
1294 INP_LIST_WUNLOCK(pcbinfo);
1296 if (inp->inp_vflag & INP_IPV6PROTO) {
1297 ip6_freepcbopts(inp->in6p_outputopts);
1298 if (inp->in6p_moptions != NULL)
1299 ip6_freemoptions(inp->in6p_moptions);
1302 if (inp->inp_options)
1303 (void)m_free(inp->inp_options);
1305 if (inp->inp_moptions != NULL)
1306 inp_freemoptions(inp->inp_moptions);
1308 RO_RTFREE(&inp->inp_route);
1309 if (inp->inp_route.ro_lle)
1310 LLE_FREE(inp->inp_route.ro_lle); /* zeros ro_lle */
1313 inp->inp_flags2 |= INP_FREED;
1314 crfree(inp->inp_cred);
1316 mac_inpcb_destroy(inp);
1318 if (!in_pcbrele_wlocked(inp))
1323 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1324 * port reservation, and preventing it from being returned by inpcb lookups.
1326 * It is used by TCP to mark an inpcb as unused and avoid future packet
1327 * delivery or event notification when a socket remains open but TCP has
1328 * closed. This might occur as a result of a shutdown()-initiated TCP close
1329 * or a RST on the wire, and allows the port binding to be reused while still
1330 * maintaining the invariant that so_pcb always points to a valid inpcb until
1333 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1334 * in_pcbnotifyall() and in_pcbpurgeif0()?
1337 in_pcbdrop(struct inpcb *inp)
1340 INP_WLOCK_ASSERT(inp);
1343 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1346 inp->inp_flags |= INP_DROPPED;
1347 if (inp->inp_flags & INP_INHASHLIST) {
1348 struct inpcbport *phd = inp->inp_phd;
1350 INP_HASH_WLOCK(inp->inp_pcbinfo);
1351 LIST_REMOVE(inp, inp_hash);
1352 LIST_REMOVE(inp, inp_portlist);
1353 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1354 LIST_REMOVE(phd, phd_hash);
1357 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1358 inp->inp_flags &= ~INP_INHASHLIST;
1360 in_pcbgroup_remove(inp);
1367 * Common routines to return the socket addresses associated with inpcbs.
1370 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1372 struct sockaddr_in *sin;
1374 sin = malloc(sizeof *sin, M_SONAME,
1376 sin->sin_family = AF_INET;
1377 sin->sin_len = sizeof(*sin);
1378 sin->sin_addr = *addr_p;
1379 sin->sin_port = port;
1381 return (struct sockaddr *)sin;
1385 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1388 struct in_addr addr;
1391 inp = sotoinpcb(so);
1392 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1395 port = inp->inp_lport;
1396 addr = inp->inp_laddr;
1399 *nam = in_sockaddr(port, &addr);
1404 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1407 struct in_addr addr;
1410 inp = sotoinpcb(so);
1411 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1414 port = inp->inp_fport;
1415 addr = inp->inp_faddr;
1418 *nam = in_sockaddr(port, &addr);
1423 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1424 struct inpcb *(*notify)(struct inpcb *, int))
1426 struct inpcb *inp, *inp_temp;
1428 INP_INFO_WLOCK(pcbinfo);
1429 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1432 if ((inp->inp_vflag & INP_IPV4) == 0) {
1437 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1438 inp->inp_socket == NULL) {
1442 if ((*notify)(inp, errno))
1445 INP_INFO_WUNLOCK(pcbinfo);
1449 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1452 struct ip_moptions *imo;
1455 INP_INFO_WLOCK(pcbinfo);
1456 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1458 imo = inp->inp_moptions;
1459 if ((inp->inp_vflag & INP_IPV4) &&
1462 * Unselect the outgoing interface if it is being
1465 if (imo->imo_multicast_ifp == ifp)
1466 imo->imo_multicast_ifp = NULL;
1469 * Drop multicast group membership if we joined
1470 * through the interface being detached.
1472 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1474 if (imo->imo_membership[i]->inm_ifp == ifp) {
1475 in_delmulti(imo->imo_membership[i]);
1477 } else if (gap != 0)
1478 imo->imo_membership[i - gap] =
1479 imo->imo_membership[i];
1481 imo->imo_num_memberships -= gap;
1485 INP_INFO_WUNLOCK(pcbinfo);
1489 * Lookup a PCB based on the local address and port. Caller must hold the
1490 * hash lock. No inpcb locks or references are acquired.
1492 #define INP_LOOKUP_MAPPED_PCB_COST 3
1494 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1495 u_short lport, int lookupflags, struct ucred *cred)
1499 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1505 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1506 ("%s: invalid lookup flags %d", __func__, lookupflags));
1508 INP_HASH_LOCK_ASSERT(pcbinfo);
1510 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1511 struct inpcbhead *head;
1513 * Look for an unconnected (wildcard foreign addr) PCB that
1514 * matches the local address and port we're looking for.
1516 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1517 0, pcbinfo->ipi_hashmask)];
1518 LIST_FOREACH(inp, head, inp_hash) {
1520 /* XXX inp locking */
1521 if ((inp->inp_vflag & INP_IPV4) == 0)
1524 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1525 inp->inp_laddr.s_addr == laddr.s_addr &&
1526 inp->inp_lport == lport) {
1531 prison_equal_ip4(cred->cr_prison,
1532 inp->inp_cred->cr_prison))
1541 struct inpcbporthead *porthash;
1542 struct inpcbport *phd;
1543 struct inpcb *match = NULL;
1545 * Best fit PCB lookup.
1547 * First see if this local port is in use by looking on the
1550 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1551 pcbinfo->ipi_porthashmask)];
1552 LIST_FOREACH(phd, porthash, phd_hash) {
1553 if (phd->phd_port == lport)
1558 * Port is in use by one or more PCBs. Look for best
1561 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1564 !prison_equal_ip4(inp->inp_cred->cr_prison,
1568 /* XXX inp locking */
1569 if ((inp->inp_vflag & INP_IPV4) == 0)
1572 * We never select the PCB that has
1573 * INP_IPV6 flag and is bound to :: if
1574 * we have another PCB which is bound
1575 * to 0.0.0.0. If a PCB has the
1576 * INP_IPV6 flag, then we set its cost
1577 * higher than IPv4 only PCBs.
1579 * Note that the case only happens
1580 * when a socket is bound to ::, under
1581 * the condition that the use of the
1582 * mapped address is allowed.
1584 if ((inp->inp_vflag & INP_IPV6) != 0)
1585 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1587 if (inp->inp_faddr.s_addr != INADDR_ANY)
1589 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1590 if (laddr.s_addr == INADDR_ANY)
1592 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1595 if (laddr.s_addr != INADDR_ANY)
1598 if (wildcard < matchwild) {
1600 matchwild = wildcard;
1609 #undef INP_LOOKUP_MAPPED_PCB_COST
1613 * Lookup PCB in hash list, using pcbgroup tables.
1615 static struct inpcb *
1616 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1617 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1618 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1620 struct inpcbhead *head;
1621 struct inpcb *inp, *tmpinp;
1622 u_short fport = fport_arg, lport = lport_arg;
1625 * First look for an exact match.
1628 INP_GROUP_LOCK(pcbgroup);
1629 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1630 pcbgroup->ipg_hashmask)];
1631 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1633 /* XXX inp locking */
1634 if ((inp->inp_vflag & INP_IPV4) == 0)
1637 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1638 inp->inp_laddr.s_addr == laddr.s_addr &&
1639 inp->inp_fport == fport &&
1640 inp->inp_lport == lport) {
1642 * XXX We should be able to directly return
1643 * the inp here, without any checks.
1644 * Well unless both bound with SO_REUSEPORT?
1646 if (prison_flag(inp->inp_cred, PR_IP4))
1652 if (tmpinp != NULL) {
1659 * For incoming connections, we may wish to do a wildcard
1660 * match for an RSS-local socket.
1662 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1663 struct inpcb *local_wild = NULL, *local_exact = NULL;
1665 struct inpcb *local_wild_mapped = NULL;
1667 struct inpcb *jail_wild = NULL;
1668 struct inpcbhead *head;
1672 * Order of socket selection - we always prefer jails.
1673 * 1. jailed, non-wild.
1675 * 3. non-jailed, non-wild.
1676 * 4. non-jailed, wild.
1679 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1680 lport, 0, pcbgroup->ipg_hashmask)];
1681 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1683 /* XXX inp locking */
1684 if ((inp->inp_vflag & INP_IPV4) == 0)
1687 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1688 inp->inp_lport != lport)
1691 injail = prison_flag(inp->inp_cred, PR_IP4);
1693 if (prison_check_ip4(inp->inp_cred,
1697 if (local_exact != NULL)
1701 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1706 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1708 /* XXX inp locking, NULL check */
1709 if (inp->inp_vflag & INP_IPV6PROTO)
1710 local_wild_mapped = inp;
1718 } /* LIST_FOREACH */
1727 inp = local_wild_mapped;
1735 * Then look for a wildcard match, if requested.
1737 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1738 struct inpcb *local_wild = NULL, *local_exact = NULL;
1740 struct inpcb *local_wild_mapped = NULL;
1742 struct inpcb *jail_wild = NULL;
1743 struct inpcbhead *head;
1747 * Order of socket selection - we always prefer jails.
1748 * 1. jailed, non-wild.
1750 * 3. non-jailed, non-wild.
1751 * 4. non-jailed, wild.
1753 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1754 0, pcbinfo->ipi_wildmask)];
1755 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1757 /* XXX inp locking */
1758 if ((inp->inp_vflag & INP_IPV4) == 0)
1761 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1762 inp->inp_lport != lport)
1765 injail = prison_flag(inp->inp_cred, PR_IP4);
1767 if (prison_check_ip4(inp->inp_cred,
1771 if (local_exact != NULL)
1775 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1780 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1782 /* XXX inp locking, NULL check */
1783 if (inp->inp_vflag & INP_IPV6PROTO)
1784 local_wild_mapped = inp;
1792 } /* LIST_FOREACH */
1800 inp = local_wild_mapped;
1804 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1805 INP_GROUP_UNLOCK(pcbgroup);
1810 INP_GROUP_UNLOCK(pcbgroup);
1811 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1813 if (in_pcbrele_wlocked(inp))
1815 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1817 if (in_pcbrele_rlocked(inp))
1820 panic("%s: locking bug", __func__);
1823 #endif /* PCBGROUP */
1826 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
1827 * that the caller has locked the hash list, and will not perform any further
1828 * locking or reference operations on either the hash list or the connection.
1830 static struct inpcb *
1831 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1832 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1835 struct inpcbhead *head;
1836 struct inpcb *inp, *tmpinp;
1837 u_short fport = fport_arg, lport = lport_arg;
1839 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1840 ("%s: invalid lookup flags %d", __func__, lookupflags));
1842 INP_HASH_LOCK_ASSERT(pcbinfo);
1845 * First look for an exact match.
1848 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1849 pcbinfo->ipi_hashmask)];
1850 LIST_FOREACH(inp, head, inp_hash) {
1852 /* XXX inp locking */
1853 if ((inp->inp_vflag & INP_IPV4) == 0)
1856 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1857 inp->inp_laddr.s_addr == laddr.s_addr &&
1858 inp->inp_fport == fport &&
1859 inp->inp_lport == lport) {
1861 * XXX We should be able to directly return
1862 * the inp here, without any checks.
1863 * Well unless both bound with SO_REUSEPORT?
1865 if (prison_flag(inp->inp_cred, PR_IP4))
1875 * Then look for a wildcard match, if requested.
1877 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1878 struct inpcb *local_wild = NULL, *local_exact = NULL;
1880 struct inpcb *local_wild_mapped = NULL;
1882 struct inpcb *jail_wild = NULL;
1886 * Order of socket selection - we always prefer jails.
1887 * 1. jailed, non-wild.
1889 * 3. non-jailed, non-wild.
1890 * 4. non-jailed, wild.
1893 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1894 0, pcbinfo->ipi_hashmask)];
1895 LIST_FOREACH(inp, head, inp_hash) {
1897 /* XXX inp locking */
1898 if ((inp->inp_vflag & INP_IPV4) == 0)
1901 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1902 inp->inp_lport != lport)
1905 injail = prison_flag(inp->inp_cred, PR_IP4);
1907 if (prison_check_ip4(inp->inp_cred,
1911 if (local_exact != NULL)
1915 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1920 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1922 /* XXX inp locking, NULL check */
1923 if (inp->inp_vflag & INP_IPV6PROTO)
1924 local_wild_mapped = inp;
1932 } /* LIST_FOREACH */
1933 if (jail_wild != NULL)
1935 if (local_exact != NULL)
1936 return (local_exact);
1937 if (local_wild != NULL)
1938 return (local_wild);
1940 if (local_wild_mapped != NULL)
1941 return (local_wild_mapped);
1943 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1949 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
1950 * hash list lock, and will return the inpcb locked (i.e., requires
1951 * INPLOOKUP_LOCKPCB).
1953 static struct inpcb *
1954 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1955 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1960 INP_HASH_RLOCK(pcbinfo);
1961 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1962 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1965 INP_HASH_RUNLOCK(pcbinfo);
1966 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1968 if (in_pcbrele_wlocked(inp))
1970 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1972 if (in_pcbrele_rlocked(inp))
1975 panic("%s: locking bug", __func__);
1977 INP_HASH_RUNLOCK(pcbinfo);
1982 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1983 * from which a pre-calculated hash value may be extracted.
1985 * Possibly more of this logic should be in in_pcbgroup.c.
1988 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1989 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1991 #if defined(PCBGROUP) && !defined(RSS)
1992 struct inpcbgroup *pcbgroup;
1995 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1996 ("%s: invalid lookup flags %d", __func__, lookupflags));
1997 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1998 ("%s: LOCKPCB not set", __func__));
2001 * When not using RSS, use connection groups in preference to the
2002 * reservation table when looking up 4-tuples. When using RSS, just
2003 * use the reservation table, due to the cost of the Toeplitz hash
2006 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2007 * we could be doing RSS with a non-Toeplitz hash that is affordable
2010 #if defined(PCBGROUP) && !defined(RSS)
2011 if (in_pcbgroup_enabled(pcbinfo)) {
2012 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2014 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2015 laddr, lport, lookupflags, ifp));
2018 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2023 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2024 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2025 struct ifnet *ifp, struct mbuf *m)
2028 struct inpcbgroup *pcbgroup;
2031 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2032 ("%s: invalid lookup flags %d", __func__, lookupflags));
2033 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2034 ("%s: LOCKPCB not set", __func__));
2038 * If we can use a hardware-generated hash to look up the connection
2039 * group, use that connection group to find the inpcb. Otherwise
2040 * fall back on a software hash -- or the reservation table if we're
2043 * XXXRW: As above, that policy belongs in the pcbgroup code.
2045 if (in_pcbgroup_enabled(pcbinfo) &&
2046 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2047 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2048 m->m_pkthdr.flowid);
2049 if (pcbgroup != NULL)
2050 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2051 fport, laddr, lport, lookupflags, ifp));
2053 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2055 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2056 laddr, lport, lookupflags, ifp));
2060 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2066 * Insert PCB onto various hash lists.
2069 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2071 struct inpcbhead *pcbhash;
2072 struct inpcbporthead *pcbporthash;
2073 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2074 struct inpcbport *phd;
2075 u_int32_t hashkey_faddr;
2077 INP_WLOCK_ASSERT(inp);
2078 INP_HASH_WLOCK_ASSERT(pcbinfo);
2080 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2081 ("in_pcbinshash: INP_INHASHLIST"));
2084 if (inp->inp_vflag & INP_IPV6)
2085 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2088 hashkey_faddr = inp->inp_faddr.s_addr;
2090 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2091 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2093 pcbporthash = &pcbinfo->ipi_porthashbase[
2094 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2097 * Go through port list and look for a head for this lport.
2099 LIST_FOREACH(phd, pcbporthash, phd_hash) {
2100 if (phd->phd_port == inp->inp_lport)
2104 * If none exists, malloc one and tack it on.
2107 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2109 return (ENOBUFS); /* XXX */
2111 phd->phd_port = inp->inp_lport;
2112 LIST_INIT(&phd->phd_pcblist);
2113 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2116 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2117 LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2118 inp->inp_flags |= INP_INHASHLIST;
2120 if (do_pcbgroup_update)
2121 in_pcbgroup_update(inp);
2127 * For now, there are two public interfaces to insert an inpcb into the hash
2128 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2129 * is used only in the TCP syncache, where in_pcbinshash is called before the
2130 * full 4-tuple is set for the inpcb, and we don't want to install in the
2131 * pcbgroup until later.
2133 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2134 * connection groups, and partially initialised inpcbs should not be exposed
2135 * to either reservation hash tables or pcbgroups.
2138 in_pcbinshash(struct inpcb *inp)
2141 return (in_pcbinshash_internal(inp, 1));
2145 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2148 return (in_pcbinshash_internal(inp, 0));
2152 * Move PCB to the proper hash bucket when { faddr, fport } have been
2153 * changed. NOTE: This does not handle the case of the lport changing (the
2154 * hashed port list would have to be updated as well), so the lport must
2155 * not change after in_pcbinshash() has been called.
2158 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2160 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2161 struct inpcbhead *head;
2162 u_int32_t hashkey_faddr;
2164 INP_WLOCK_ASSERT(inp);
2165 INP_HASH_WLOCK_ASSERT(pcbinfo);
2167 KASSERT(inp->inp_flags & INP_INHASHLIST,
2168 ("in_pcbrehash: !INP_INHASHLIST"));
2171 if (inp->inp_vflag & INP_IPV6)
2172 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2175 hashkey_faddr = inp->inp_faddr.s_addr;
2177 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2178 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2180 LIST_REMOVE(inp, inp_hash);
2181 LIST_INSERT_HEAD(head, inp, inp_hash);
2185 in_pcbgroup_update_mbuf(inp, m);
2187 in_pcbgroup_update(inp);
2192 in_pcbrehash(struct inpcb *inp)
2195 in_pcbrehash_mbuf(inp, NULL);
2199 * Remove PCB from various lists.
2202 in_pcbremlists(struct inpcb *inp)
2204 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2207 if (pcbinfo == &V_tcbinfo) {
2208 INP_INFO_RLOCK_ASSERT(pcbinfo);
2210 INP_INFO_WLOCK_ASSERT(pcbinfo);
2214 INP_WLOCK_ASSERT(inp);
2215 INP_LIST_WLOCK_ASSERT(pcbinfo);
2217 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2218 if (inp->inp_flags & INP_INHASHLIST) {
2219 struct inpcbport *phd = inp->inp_phd;
2221 INP_HASH_WLOCK(pcbinfo);
2222 LIST_REMOVE(inp, inp_hash);
2223 LIST_REMOVE(inp, inp_portlist);
2224 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2225 LIST_REMOVE(phd, phd_hash);
2228 INP_HASH_WUNLOCK(pcbinfo);
2229 inp->inp_flags &= ~INP_INHASHLIST;
2231 LIST_REMOVE(inp, inp_list);
2232 pcbinfo->ipi_count--;
2234 in_pcbgroup_remove(inp);
2239 * Check for alternatives when higher level complains
2240 * about service problems. For now, invalidate cached
2241 * routing information. If the route was created dynamically
2242 * (by a redirect), time to try a default gateway again.
2245 in_losing(struct inpcb *inp)
2248 RO_RTFREE(&inp->inp_route);
2249 if (inp->inp_route.ro_lle)
2250 LLE_FREE(inp->inp_route.ro_lle); /* zeros ro_lle */
2255 * A set label operation has occurred at the socket layer, propagate the
2256 * label change into the in_pcb for the socket.
2259 in_pcbsosetlabel(struct socket *so)
2264 inp = sotoinpcb(so);
2265 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2269 mac_inpcb_sosetlabel(so, inp);
2276 * ipport_tick runs once per second, determining if random port allocation
2277 * should be continued. If more than ipport_randomcps ports have been
2278 * allocated in the last second, then we return to sequential port
2279 * allocation. We return to random allocation only once we drop below
2280 * ipport_randomcps for at least ipport_randomtime seconds.
2283 ipport_tick(void *xtp)
2285 VNET_ITERATOR_DECL(vnet_iter);
2287 VNET_LIST_RLOCK_NOSLEEP();
2288 VNET_FOREACH(vnet_iter) {
2289 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2290 if (V_ipport_tcpallocs <=
2291 V_ipport_tcplastcount + V_ipport_randomcps) {
2292 if (V_ipport_stoprandom > 0)
2293 V_ipport_stoprandom--;
2295 V_ipport_stoprandom = V_ipport_randomtime;
2296 V_ipport_tcplastcount = V_ipport_tcpallocs;
2299 VNET_LIST_RUNLOCK_NOSLEEP();
2300 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2307 callout_stop(&ipport_tick_callout);
2311 * The ipport_callout should start running at about the time we attach the
2312 * inet or inet6 domains.
2315 ipport_tick_init(const void *unused __unused)
2318 /* Start ipport_tick. */
2319 callout_init(&ipport_tick_callout, 1);
2320 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2321 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2322 SHUTDOWN_PRI_DEFAULT);
2324 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2325 ipport_tick_init, NULL);
2328 inp_wlock(struct inpcb *inp)
2335 inp_wunlock(struct inpcb *inp)
2342 inp_rlock(struct inpcb *inp)
2349 inp_runlock(struct inpcb *inp)
2357 inp_lock_assert(struct inpcb *inp)
2360 INP_WLOCK_ASSERT(inp);
2364 inp_unlock_assert(struct inpcb *inp)
2367 INP_UNLOCK_ASSERT(inp);
2372 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2376 INP_INFO_WLOCK(&V_tcbinfo);
2377 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2382 INP_INFO_WUNLOCK(&V_tcbinfo);
2386 inp_inpcbtosocket(struct inpcb *inp)
2389 INP_WLOCK_ASSERT(inp);
2390 return (inp->inp_socket);
2394 inp_inpcbtotcpcb(struct inpcb *inp)
2397 INP_WLOCK_ASSERT(inp);
2398 return ((struct tcpcb *)inp->inp_ppcb);
2402 inp_ip_tos_get(const struct inpcb *inp)
2405 return (inp->inp_ip_tos);
2409 inp_ip_tos_set(struct inpcb *inp, int val)
2412 inp->inp_ip_tos = val;
2416 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2417 uint32_t *faddr, uint16_t *fp)
2420 INP_LOCK_ASSERT(inp);
2421 *laddr = inp->inp_laddr.s_addr;
2422 *faddr = inp->inp_faddr.s_addr;
2423 *lp = inp->inp_lport;
2424 *fp = inp->inp_fport;
2428 so_sotoinpcb(struct socket *so)
2431 return (sotoinpcb(so));
2435 so_sototcpcb(struct socket *so)
2438 return (sototcpcb(so));
2443 db_print_indent(int indent)
2447 for (i = 0; i < indent; i++)
2452 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2454 char faddr_str[48], laddr_str[48];
2456 db_print_indent(indent);
2457 db_printf("%s at %p\n", name, inc);
2462 if (inc->inc_flags & INC_ISIPV6) {
2464 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2465 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2470 inet_ntoa_r(inc->inc_laddr, laddr_str);
2471 inet_ntoa_r(inc->inc_faddr, faddr_str);
2473 db_print_indent(indent);
2474 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2475 ntohs(inc->inc_lport));
2476 db_print_indent(indent);
2477 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2478 ntohs(inc->inc_fport));
2482 db_print_inpflags(int inp_flags)
2487 if (inp_flags & INP_RECVOPTS) {
2488 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2491 if (inp_flags & INP_RECVRETOPTS) {
2492 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2495 if (inp_flags & INP_RECVDSTADDR) {
2496 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2499 if (inp_flags & INP_HDRINCL) {
2500 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2503 if (inp_flags & INP_HIGHPORT) {
2504 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2507 if (inp_flags & INP_LOWPORT) {
2508 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2511 if (inp_flags & INP_ANONPORT) {
2512 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2515 if (inp_flags & INP_RECVIF) {
2516 db_printf("%sINP_RECVIF", comma ? ", " : "");
2519 if (inp_flags & INP_MTUDISC) {
2520 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2523 if (inp_flags & INP_RECVTTL) {
2524 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2527 if (inp_flags & INP_DONTFRAG) {
2528 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2531 if (inp_flags & INP_RECVTOS) {
2532 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2535 if (inp_flags & IN6P_IPV6_V6ONLY) {
2536 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2539 if (inp_flags & IN6P_PKTINFO) {
2540 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2543 if (inp_flags & IN6P_HOPLIMIT) {
2544 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2547 if (inp_flags & IN6P_HOPOPTS) {
2548 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2551 if (inp_flags & IN6P_DSTOPTS) {
2552 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2555 if (inp_flags & IN6P_RTHDR) {
2556 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2559 if (inp_flags & IN6P_RTHDRDSTOPTS) {
2560 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2563 if (inp_flags & IN6P_TCLASS) {
2564 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2567 if (inp_flags & IN6P_AUTOFLOWLABEL) {
2568 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2571 if (inp_flags & INP_TIMEWAIT) {
2572 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2575 if (inp_flags & INP_ONESBCAST) {
2576 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2579 if (inp_flags & INP_DROPPED) {
2580 db_printf("%sINP_DROPPED", comma ? ", " : "");
2583 if (inp_flags & INP_SOCKREF) {
2584 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2587 if (inp_flags & IN6P_RFC2292) {
2588 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2591 if (inp_flags & IN6P_MTU) {
2592 db_printf("IN6P_MTU%s", comma ? ", " : "");
2598 db_print_inpvflag(u_char inp_vflag)
2603 if (inp_vflag & INP_IPV4) {
2604 db_printf("%sINP_IPV4", comma ? ", " : "");
2607 if (inp_vflag & INP_IPV6) {
2608 db_printf("%sINP_IPV6", comma ? ", " : "");
2611 if (inp_vflag & INP_IPV6PROTO) {
2612 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2618 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2621 db_print_indent(indent);
2622 db_printf("%s at %p\n", name, inp);
2626 db_print_indent(indent);
2627 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2629 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2631 db_print_indent(indent);
2632 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
2633 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2635 db_print_indent(indent);
2636 db_printf("inp_label: %p inp_flags: 0x%x (",
2637 inp->inp_label, inp->inp_flags);
2638 db_print_inpflags(inp->inp_flags);
2641 db_print_indent(indent);
2642 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
2644 db_print_inpvflag(inp->inp_vflag);
2647 db_print_indent(indent);
2648 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
2649 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2651 db_print_indent(indent);
2653 if (inp->inp_vflag & INP_IPV6) {
2654 db_printf("in6p_options: %p in6p_outputopts: %p "
2655 "in6p_moptions: %p\n", inp->in6p_options,
2656 inp->in6p_outputopts, inp->in6p_moptions);
2657 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
2658 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2663 db_printf("inp_ip_tos: %d inp_ip_options: %p "
2664 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2665 inp->inp_options, inp->inp_moptions);
2668 db_print_indent(indent);
2669 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
2670 (uintmax_t)inp->inp_gencnt);
2673 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2678 db_printf("usage: show inpcb <addr>\n");
2681 inp = (struct inpcb *)addr;
2683 db_print_inpcb(inp, "inpcb", 0);
2689 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
2693 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
2695 union if_snd_tag_modify_params params = {
2696 .rate_limit.max_rate = max_pacing_rate,
2698 struct m_snd_tag *mst;
2702 mst = inp->inp_snd_tag;
2710 if (ifp->if_snd_tag_modify == NULL) {
2713 error = ifp->if_snd_tag_modify(mst, ¶ms);
2719 * Query existing TX rate limit based on the existing
2720 * "inp->inp_snd_tag", if any.
2723 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
2725 union if_snd_tag_query_params params = { };
2726 struct m_snd_tag *mst;
2730 mst = inp->inp_snd_tag;
2738 if (ifp->if_snd_tag_query == NULL) {
2741 error = ifp->if_snd_tag_query(mst, ¶ms);
2742 if (error == 0 && p_max_pacing_rate != NULL)
2743 *p_max_pacing_rate = params.rate_limit.max_rate;
2749 * Allocate a new TX rate limit send tag from the network interface
2750 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
2753 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
2754 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
2756 union if_snd_tag_alloc_params params = {
2757 .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT,
2758 .rate_limit.hdr.flowid = flowid,
2759 .rate_limit.hdr.flowtype = flowtype,
2760 .rate_limit.max_rate = max_pacing_rate,
2764 INP_WLOCK_ASSERT(inp);
2766 if (inp->inp_snd_tag != NULL)
2769 if (ifp->if_snd_tag_alloc == NULL) {
2772 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
2775 * At success increment the refcount on
2776 * the send tag's network interface:
2779 if_ref(inp->inp_snd_tag->ifp);
2785 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
2789 in_pcbdetach_txrtlmt(struct inpcb *inp)
2791 struct m_snd_tag *mst;
2794 INP_WLOCK_ASSERT(inp);
2796 mst = inp->inp_snd_tag;
2797 inp->inp_snd_tag = NULL;
2807 * If the device was detached while we still had reference(s)
2808 * on the ifp, we assume if_snd_tag_free() was replaced with
2811 ifp->if_snd_tag_free(mst);
2813 /* release reference count on network interface */
2818 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
2819 * is set in the fast path and will attach/detach/modify the TX rate
2820 * limit send tag based on the socket's so_max_pacing_rate value.
2823 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
2825 struct socket *socket;
2826 uint32_t max_pacing_rate;
2833 socket = inp->inp_socket;
2837 if (!INP_WLOCKED(inp)) {
2839 * NOTE: If the write locking fails, we need to bail
2840 * out and use the non-ratelimited ring for the
2841 * transmit until there is a new chance to get the
2844 if (!INP_TRY_UPGRADE(inp))
2852 * NOTE: The so_max_pacing_rate value is read unlocked,
2853 * because atomic updates are not required since the variable
2854 * is checked at every mbuf we send. It is assumed that the
2855 * variable read itself will be atomic.
2857 max_pacing_rate = socket->so_max_pacing_rate;
2860 * NOTE: When attaching to a network interface a reference is
2861 * made to ensure the network interface doesn't go away until
2862 * all ratelimit connections are gone. The network interface
2863 * pointers compared below represent valid network interfaces,
2864 * except when comparing towards NULL.
2866 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
2868 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
2869 if (inp->inp_snd_tag != NULL)
2870 in_pcbdetach_txrtlmt(inp);
2872 } else if (inp->inp_snd_tag == NULL) {
2874 * In order to utilize packet pacing with RSS, we need
2875 * to wait until there is a valid RSS hash before we
2878 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
2881 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
2882 mb->m_pkthdr.flowid, max_pacing_rate);
2885 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
2887 if (error == 0 || error == EOPNOTSUPP)
2888 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
2894 * Track route changes for TX rate limiting.
2897 in_pcboutput_eagain(struct inpcb *inp)
2899 struct socket *socket;
2905 socket = inp->inp_socket;
2909 if (inp->inp_snd_tag == NULL)
2912 if (!INP_WLOCKED(inp)) {
2914 * NOTE: If the write locking fails, we need to bail
2915 * out and use the non-ratelimited ring for the
2916 * transmit until there is a new chance to get the
2919 if (!INP_TRY_UPGRADE(inp))
2926 /* detach rate limiting */
2927 in_pcbdetach_txrtlmt(inp);
2929 /* make sure new mbuf send tag allocation is made */
2930 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
2935 #endif /* RATELIMIT */