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 * 3. 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>
101 #include <netipsec/ipsec_support.h>
103 #include <security/mac/mac_framework.h>
105 static struct callout ipport_tick_callout;
108 * These configure the range of local port addresses assigned to
109 * "unspecified" outgoing connections/packets/whatever.
111 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
112 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
113 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
114 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
115 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
116 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
119 * Reserved ports accessible only to root. There are significant
120 * security considerations that must be accounted for when changing these,
121 * but the security benefits can be great. Please be careful.
123 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
124 VNET_DEFINE(int, ipport_reservedlow);
126 /* Variables dealing with random ephemeral port allocation. */
127 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
128 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
129 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
130 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
131 VNET_DEFINE(int, ipport_tcpallocs);
132 static VNET_DEFINE(int, ipport_tcplastcount);
134 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
136 static void in_pcbremlists(struct inpcb *inp);
138 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
139 struct in_addr faddr, u_int fport_arg,
140 struct in_addr laddr, u_int lport_arg,
141 int lookupflags, struct ifnet *ifp);
143 #define RANGECHK(var, min, max) \
144 if ((var) < (min)) { (var) = (min); } \
145 else if ((var) > (max)) { (var) = (max); }
148 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
152 error = sysctl_handle_int(oidp, arg1, arg2, req);
154 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
155 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
156 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
157 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
158 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
159 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
166 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
169 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
170 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
171 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
172 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
173 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
174 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
176 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
177 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
188 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
189 &VNET_NAME(ipport_reservedhigh), 0, "");
190 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
191 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
192 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
193 CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
195 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
196 CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
198 "allocations before switching to a sequental one");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
200 CTLFLAG_VNET | CTLFLAG_RW,
201 &VNET_NAME(ipport_randomtime), 0,
202 "Minimum time to keep sequental port "
203 "allocation before switching to a random one");
207 * in_pcb.c: manage the Protocol Control Blocks.
209 * NOTE: It is assumed that most of these functions will be called with
210 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
211 * functions often modify hash chains or addresses in pcbs.
215 * Different protocols initialize their inpcbs differently - giving
216 * different name to the lock. But they all are disposed the same.
219 inpcb_fini(void *mem, int size)
221 struct inpcb *inp = mem;
223 INP_LOCK_DESTROY(inp);
227 * Initialize an inpcbinfo -- we should be able to reduce the number of
231 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
232 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
233 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
236 INP_INFO_LOCK_INIT(pcbinfo, name);
237 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
238 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
240 pcbinfo->ipi_vnet = curvnet;
242 pcbinfo->ipi_listhead = listhead;
243 LIST_INIT(pcbinfo->ipi_listhead);
244 pcbinfo->ipi_count = 0;
245 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
246 &pcbinfo->ipi_hashmask);
247 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
248 &pcbinfo->ipi_porthashmask);
250 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
252 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
253 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
254 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
255 uma_zone_set_warning(pcbinfo->ipi_zone,
256 "kern.ipc.maxsockets limit reached");
260 * Destroy an inpcbinfo.
263 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
266 KASSERT(pcbinfo->ipi_count == 0,
267 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
269 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
270 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
271 pcbinfo->ipi_porthashmask);
273 in_pcbgroup_destroy(pcbinfo);
275 uma_zdestroy(pcbinfo->ipi_zone);
276 INP_LIST_LOCK_DESTROY(pcbinfo);
277 INP_HASH_LOCK_DESTROY(pcbinfo);
278 INP_INFO_LOCK_DESTROY(pcbinfo);
282 * Allocate a PCB and associate it with the socket.
283 * On success return with the PCB locked.
286 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
292 if (pcbinfo == &V_tcbinfo) {
293 INP_INFO_RLOCK_ASSERT(pcbinfo);
295 INP_INFO_WLOCK_ASSERT(pcbinfo);
300 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
303 bzero(&inp->inp_start_zero, inp_zero_size);
304 inp->inp_pcbinfo = pcbinfo;
305 inp->inp_socket = so;
306 inp->inp_cred = crhold(so->so_cred);
307 inp->inp_inc.inc_fibnum = so->so_fibnum;
309 error = mac_inpcb_init(inp, M_NOWAIT);
312 mac_inpcb_create(so, inp);
314 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
315 error = ipsec_init_pcbpolicy(inp);
318 mac_inpcb_destroy(inp);
324 if (INP_SOCKAF(so) == AF_INET6) {
325 inp->inp_vflag |= INP_IPV6PROTO;
327 inp->inp_flags |= IN6P_IPV6_V6ONLY;
331 INP_LIST_WLOCK(pcbinfo);
332 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
333 pcbinfo->ipi_count++;
334 so->so_pcb = (caddr_t)inp;
336 if (V_ip6_auto_flowlabel)
337 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
339 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
340 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
343 * Routes in inpcb's can cache L2 as well; they are guaranteed
346 inp->inp_route.ro_flags = RT_LLE_CACHE;
347 INP_LIST_WUNLOCK(pcbinfo);
348 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
351 crfree(inp->inp_cred);
352 uma_zfree(pcbinfo->ipi_zone, inp);
360 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
364 INP_WLOCK_ASSERT(inp);
365 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
367 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
369 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
370 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
371 &inp->inp_lport, cred);
374 if (in_pcbinshash(inp) != 0) {
375 inp->inp_laddr.s_addr = INADDR_ANY;
380 inp->inp_flags |= INP_ANONPORT;
386 * Select a local port (number) to use.
388 #if defined(INET) || defined(INET6)
390 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
391 struct ucred *cred, int lookupflags)
393 struct inpcbinfo *pcbinfo;
394 struct inpcb *tmpinp;
395 unsigned short *lastport;
396 int count, dorandom, error;
397 u_short aux, first, last, lport;
399 struct in_addr laddr;
402 pcbinfo = inp->inp_pcbinfo;
405 * Because no actual state changes occur here, a global write lock on
406 * the pcbinfo isn't required.
408 INP_LOCK_ASSERT(inp);
409 INP_HASH_LOCK_ASSERT(pcbinfo);
411 if (inp->inp_flags & INP_HIGHPORT) {
412 first = V_ipport_hifirstauto; /* sysctl */
413 last = V_ipport_hilastauto;
414 lastport = &pcbinfo->ipi_lasthi;
415 } else if (inp->inp_flags & INP_LOWPORT) {
416 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
419 first = V_ipport_lowfirstauto; /* 1023 */
420 last = V_ipport_lowlastauto; /* 600 */
421 lastport = &pcbinfo->ipi_lastlow;
423 first = V_ipport_firstauto; /* sysctl */
424 last = V_ipport_lastauto;
425 lastport = &pcbinfo->ipi_lastport;
428 * For UDP(-Lite), use random port allocation as long as the user
429 * allows it. For TCP (and as of yet unknown) connections,
430 * use random port allocation only if the user allows it AND
431 * ipport_tick() allows it.
433 if (V_ipport_randomized &&
434 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
435 pcbinfo == &V_ulitecbinfo))
440 * It makes no sense to do random port allocation if
441 * we have the only port available.
445 /* Make sure to not include UDP(-Lite) packets in the count. */
446 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
447 V_ipport_tcpallocs++;
449 * Instead of having two loops further down counting up or down
450 * make sure that first is always <= last and go with only one
451 * code path implementing all logic.
460 /* Make the compiler happy. */
462 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
463 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
468 tmpinp = NULL; /* Make compiler happy. */
472 *lastport = first + (arc4random() % (last - first));
474 count = last - first;
477 if (count-- < 0) /* completely used? */
478 return (EADDRNOTAVAIL);
480 if (*lastport < first || *lastport > last)
482 lport = htons(*lastport);
485 if ((inp->inp_vflag & INP_IPV6) != 0)
486 tmpinp = in6_pcblookup_local(pcbinfo,
487 &inp->in6p_laddr, lport, lookupflags, cred);
489 #if defined(INET) && defined(INET6)
493 tmpinp = in_pcblookup_local(pcbinfo, laddr,
494 lport, lookupflags, cred);
496 } while (tmpinp != NULL);
499 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
500 laddrp->s_addr = laddr.s_addr;
508 * Return cached socket options.
511 inp_so_options(const struct inpcb *inp)
517 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
518 so_options |= SO_REUSEPORT;
519 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
520 so_options |= SO_REUSEADDR;
523 #endif /* INET || INET6 */
526 * Check if a new BINDMULTI socket is allowed to be created.
528 * ni points to the new inp.
529 * oi points to the exisitng inp.
531 * This checks whether the existing inp also has BINDMULTI and
532 * whether the credentials match.
535 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
537 /* Check permissions match */
538 if ((ni->inp_flags2 & INP_BINDMULTI) &&
539 (ni->inp_cred->cr_uid !=
540 oi->inp_cred->cr_uid))
543 /* Check the existing inp has BINDMULTI set */
544 if ((ni->inp_flags2 & INP_BINDMULTI) &&
545 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
549 * We're okay - either INP_BINDMULTI isn't set on ni, or
550 * it is and it matches the checks.
557 * Set up a bind operation on a PCB, performing port allocation
558 * as required, but do not actually modify the PCB. Callers can
559 * either complete the bind by setting inp_laddr/inp_lport and
560 * calling in_pcbinshash(), or they can just use the resulting
561 * port and address to authorise the sending of a once-off packet.
563 * On error, the values of *laddrp and *lportp are not changed.
566 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
567 u_short *lportp, struct ucred *cred)
569 struct socket *so = inp->inp_socket;
570 struct sockaddr_in *sin;
571 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
572 struct in_addr laddr;
574 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
578 * No state changes, so read locks are sufficient here.
580 INP_LOCK_ASSERT(inp);
581 INP_HASH_LOCK_ASSERT(pcbinfo);
583 if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
584 return (EADDRNOTAVAIL);
585 laddr.s_addr = *laddrp;
586 if (nam != NULL && laddr.s_addr != INADDR_ANY)
588 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
589 lookupflags = INPLOOKUP_WILDCARD;
591 if ((error = prison_local_ip4(cred, &laddr)) != 0)
594 sin = (struct sockaddr_in *)nam;
595 if (nam->sa_len != sizeof (*sin))
599 * We should check the family, but old programs
600 * incorrectly fail to initialize it.
602 if (sin->sin_family != AF_INET)
603 return (EAFNOSUPPORT);
605 error = prison_local_ip4(cred, &sin->sin_addr);
608 if (sin->sin_port != *lportp) {
609 /* Don't allow the port to change. */
612 lport = sin->sin_port;
614 /* NB: lport is left as 0 if the port isn't being changed. */
615 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
617 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
618 * allow complete duplication of binding if
619 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
620 * and a multicast address is bound on both
621 * new and duplicated sockets.
623 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
624 reuseport = SO_REUSEADDR|SO_REUSEPORT;
625 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
626 sin->sin_port = 0; /* yech... */
627 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
629 * Is the address a local IP address?
630 * If INP_BINDANY is set, then the socket may be bound
631 * to any endpoint address, local or not.
633 if ((inp->inp_flags & INP_BINDANY) == 0 &&
634 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
635 return (EADDRNOTAVAIL);
637 laddr = sin->sin_addr;
643 if (ntohs(lport) <= V_ipport_reservedhigh &&
644 ntohs(lport) >= V_ipport_reservedlow &&
645 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
648 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
649 priv_check_cred(inp->inp_cred,
650 PRIV_NETINET_REUSEPORT, 0) != 0) {
651 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
652 lport, INPLOOKUP_WILDCARD, cred);
655 * This entire block sorely needs a rewrite.
658 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
659 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
660 (so->so_type != SOCK_STREAM ||
661 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
662 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
663 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
664 (t->inp_flags2 & INP_REUSEPORT) == 0) &&
665 (inp->inp_cred->cr_uid !=
666 t->inp_cred->cr_uid))
670 * If the socket is a BINDMULTI socket, then
671 * the credentials need to match and the
672 * original socket also has to have been bound
675 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
678 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
679 lport, lookupflags, cred);
680 if (t && (t->inp_flags & INP_TIMEWAIT)) {
682 * XXXRW: If an incpb has had its timewait
683 * state recycled, we treat the address as
684 * being in use (for now). This is better
685 * than a panic, but not desirable.
689 (reuseport & tw->tw_so_options) == 0)
692 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
693 (reuseport & inp_so_options(t)) == 0) {
695 if (ntohl(sin->sin_addr.s_addr) !=
697 ntohl(t->inp_laddr.s_addr) !=
699 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
700 (t->inp_vflag & INP_IPV6PROTO) == 0)
703 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
711 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
716 *laddrp = laddr.s_addr;
722 * Connect from a socket to a specified address.
723 * Both address and port must be specified in argument sin.
724 * If don't have a local address for this socket yet,
728 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
729 struct ucred *cred, struct mbuf *m)
731 u_short lport, fport;
732 in_addr_t laddr, faddr;
735 INP_WLOCK_ASSERT(inp);
736 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
738 lport = inp->inp_lport;
739 laddr = inp->inp_laddr.s_addr;
740 anonport = (lport == 0);
741 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
746 /* Do the initial binding of the local address if required. */
747 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
748 inp->inp_lport = lport;
749 inp->inp_laddr.s_addr = laddr;
750 if (in_pcbinshash(inp) != 0) {
751 inp->inp_laddr.s_addr = INADDR_ANY;
757 /* Commit the remaining changes. */
758 inp->inp_lport = lport;
759 inp->inp_laddr.s_addr = laddr;
760 inp->inp_faddr.s_addr = faddr;
761 inp->inp_fport = fport;
762 in_pcbrehash_mbuf(inp, m);
765 inp->inp_flags |= INP_ANONPORT;
770 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
773 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
777 * Do proper source address selection on an unbound socket in case
778 * of connect. Take jails into account as well.
781 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
786 struct sockaddr_in *sin;
790 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
793 * Bypass source address selection and use the primary jail IP
796 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
800 bzero(&sro, sizeof(sro));
802 sin = (struct sockaddr_in *)&sro.ro_dst;
803 sin->sin_family = AF_INET;
804 sin->sin_len = sizeof(struct sockaddr_in);
805 sin->sin_addr.s_addr = faddr->s_addr;
808 * If route is known our src addr is taken from the i/f,
811 * Find out route to destination.
813 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
814 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
817 * If we found a route, use the address corresponding to
818 * the outgoing interface.
820 * Otherwise assume faddr is reachable on a directly connected
821 * network and try to find a corresponding interface to take
822 * the source address from.
824 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
825 struct in_ifaddr *ia;
828 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
829 inp->inp_socket->so_fibnum));
831 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
832 inp->inp_socket->so_fibnum));
838 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
839 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
840 ifa_free(&ia->ia_ifa);
845 ifa_free(&ia->ia_ifa);
848 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
851 if (sa->sa_family != AF_INET)
853 sin = (struct sockaddr_in *)sa;
854 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
855 ia = (struct in_ifaddr *)ifa;
860 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
861 IF_ADDR_RUNLOCK(ifp);
864 IF_ADDR_RUNLOCK(ifp);
866 /* 3. As a last resort return the 'default' jail address. */
867 error = prison_get_ip4(cred, laddr);
872 * If the outgoing interface on the route found is not
873 * a loopback interface, use the address from that interface.
874 * In case of jails do those three steps:
875 * 1. check if the interface address belongs to the jail. If so use it.
876 * 2. check if we have any address on the outgoing interface
877 * belonging to this jail. If so use it.
878 * 3. as a last resort return the 'default' jail address.
880 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
881 struct in_ifaddr *ia;
884 /* If not jailed, use the default returned. */
885 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
886 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
887 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
892 /* 1. Check if the iface address belongs to the jail. */
893 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
894 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
895 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
896 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
901 * 2. Check if we have any address on the outgoing interface
902 * belonging to this jail.
905 ifp = sro.ro_rt->rt_ifp;
907 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
909 if (sa->sa_family != AF_INET)
911 sin = (struct sockaddr_in *)sa;
912 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
913 ia = (struct in_ifaddr *)ifa;
918 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
919 IF_ADDR_RUNLOCK(ifp);
922 IF_ADDR_RUNLOCK(ifp);
924 /* 3. As a last resort return the 'default' jail address. */
925 error = prison_get_ip4(cred, laddr);
930 * The outgoing interface is marked with 'loopback net', so a route
931 * to ourselves is here.
932 * Try to find the interface of the destination address and then
933 * take the address from there. That interface is not necessarily
934 * a loopback interface.
935 * In case of jails, check that it is an address of the jail
936 * and if we cannot find, fall back to the 'default' jail address.
938 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
939 struct sockaddr_in sain;
940 struct in_ifaddr *ia;
942 bzero(&sain, sizeof(struct sockaddr_in));
943 sain.sin_family = AF_INET;
944 sain.sin_len = sizeof(struct sockaddr_in);
945 sain.sin_addr.s_addr = faddr->s_addr;
947 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
948 inp->inp_socket->so_fibnum));
950 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
951 inp->inp_socket->so_fibnum));
953 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
955 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
960 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
961 ifa_free(&ia->ia_ifa);
970 ifa_free(&ia->ia_ifa);
973 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
976 if (sa->sa_family != AF_INET)
978 sin = (struct sockaddr_in *)sa;
979 if (prison_check_ip4(cred,
980 &sin->sin_addr) == 0) {
981 ia = (struct in_ifaddr *)ifa;
986 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
987 IF_ADDR_RUNLOCK(ifp);
990 IF_ADDR_RUNLOCK(ifp);
993 /* 3. As a last resort return the 'default' jail address. */
994 error = prison_get_ip4(cred, laddr);
999 if (sro.ro_rt != NULL)
1005 * Set up for a connect from a socket to the specified address.
1006 * On entry, *laddrp and *lportp should contain the current local
1007 * address and port for the PCB; these are updated to the values
1008 * that should be placed in inp_laddr and inp_lport to complete
1011 * On success, *faddrp and *fportp will be set to the remote address
1012 * and port. These are not updated in the error case.
1014 * If the operation fails because the connection already exists,
1015 * *oinpp will be set to the PCB of that connection so that the
1016 * caller can decide to override it. In all other cases, *oinpp
1020 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1021 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1022 struct inpcb **oinpp, struct ucred *cred)
1024 struct rm_priotracker in_ifa_tracker;
1025 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1026 struct in_ifaddr *ia;
1028 struct in_addr laddr, faddr;
1029 u_short lport, fport;
1033 * Because a global state change doesn't actually occur here, a read
1034 * lock is sufficient.
1036 INP_LOCK_ASSERT(inp);
1037 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1041 if (nam->sa_len != sizeof (*sin))
1043 if (sin->sin_family != AF_INET)
1044 return (EAFNOSUPPORT);
1045 if (sin->sin_port == 0)
1046 return (EADDRNOTAVAIL);
1047 laddr.s_addr = *laddrp;
1049 faddr = sin->sin_addr;
1050 fport = sin->sin_port;
1052 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1054 * If the destination address is INADDR_ANY,
1055 * use the primary local address.
1056 * If the supplied address is INADDR_BROADCAST,
1057 * and the primary interface supports broadcast,
1058 * choose the broadcast address for that interface.
1060 if (faddr.s_addr == INADDR_ANY) {
1061 IN_IFADDR_RLOCK(&in_ifa_tracker);
1063 IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1064 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1066 (error = prison_get_ip4(cred, &faddr)) != 0)
1068 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1069 IN_IFADDR_RLOCK(&in_ifa_tracker);
1070 if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1072 faddr = satosin(&TAILQ_FIRST(
1073 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1074 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1077 if (laddr.s_addr == INADDR_ANY) {
1078 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1080 * If the destination address is multicast and an outgoing
1081 * interface has been set as a multicast option, prefer the
1082 * address of that interface as our source address.
1084 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1085 inp->inp_moptions != NULL) {
1086 struct ip_moptions *imo;
1089 imo = inp->inp_moptions;
1090 if (imo->imo_multicast_ifp != NULL) {
1091 ifp = imo->imo_multicast_ifp;
1092 IN_IFADDR_RLOCK(&in_ifa_tracker);
1093 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1094 if ((ia->ia_ifp == ifp) &&
1096 prison_check_ip4(cred,
1097 &ia->ia_addr.sin_addr) == 0))
1101 error = EADDRNOTAVAIL;
1103 laddr = ia->ia_addr.sin_addr;
1106 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1112 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1113 laddr, lport, 0, NULL);
1117 return (EADDRINUSE);
1120 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1125 *laddrp = laddr.s_addr;
1127 *faddrp = faddr.s_addr;
1133 in_pcbdisconnect(struct inpcb *inp)
1136 INP_WLOCK_ASSERT(inp);
1137 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1139 inp->inp_faddr.s_addr = INADDR_ANY;
1146 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1147 * For most protocols, this will be invoked immediately prior to calling
1148 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1149 * socket, in which case in_pcbfree() is deferred.
1152 in_pcbdetach(struct inpcb *inp)
1155 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1158 if (inp->inp_snd_tag != NULL)
1159 in_pcbdetach_txrtlmt(inp);
1161 inp->inp_socket->so_pcb = NULL;
1162 inp->inp_socket = NULL;
1166 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1167 * stability of an inpcb pointer despite the inpcb lock being released. This
1168 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1169 * but where the inpcb lock may already held, or when acquiring a reference
1172 * in_pcbref() should be used only to provide brief memory stability, and
1173 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1174 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1175 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1176 * lock and rele are the *only* safe operations that may be performed on the
1179 * While the inpcb will not be freed, releasing the inpcb lock means that the
1180 * connection's state may change, so the caller should be careful to
1181 * revalidate any cached state on reacquiring the lock. Drop the reference
1182 * using in_pcbrele().
1185 in_pcbref(struct inpcb *inp)
1188 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1190 refcount_acquire(&inp->inp_refcount);
1194 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1195 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1196 * return a flag indicating whether or not the inpcb remains valid. If it is
1197 * valid, we return with the inpcb lock held.
1199 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1200 * reference on an inpcb. Historically more work was done here (actually, in
1201 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1202 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1203 * about memory stability (and continued use of the write lock).
1206 in_pcbrele_rlocked(struct inpcb *inp)
1208 struct inpcbinfo *pcbinfo;
1210 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1212 INP_RLOCK_ASSERT(inp);
1214 if (refcount_release(&inp->inp_refcount) == 0) {
1216 * If the inpcb has been freed, let the caller know, even if
1217 * this isn't the last reference.
1219 if (inp->inp_flags2 & INP_FREED) {
1226 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1229 pcbinfo = inp->inp_pcbinfo;
1230 uma_zfree(pcbinfo->ipi_zone, inp);
1235 in_pcbrele_wlocked(struct inpcb *inp)
1237 struct inpcbinfo *pcbinfo;
1239 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1241 INP_WLOCK_ASSERT(inp);
1243 if (refcount_release(&inp->inp_refcount) == 0) {
1245 * If the inpcb has been freed, let the caller know, even if
1246 * this isn't the last reference.
1248 if (inp->inp_flags2 & INP_FREED) {
1255 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1258 pcbinfo = inp->inp_pcbinfo;
1259 uma_zfree(pcbinfo->ipi_zone, inp);
1264 * Temporary wrapper.
1267 in_pcbrele(struct inpcb *inp)
1270 return (in_pcbrele_wlocked(inp));
1274 * Unconditionally schedule an inpcb to be freed by decrementing its
1275 * reference count, which should occur only after the inpcb has been detached
1276 * from its socket. If another thread holds a temporary reference (acquired
1277 * using in_pcbref()) then the free is deferred until that reference is
1278 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1279 * work, including removal from global lists, is done in this context, where
1280 * the pcbinfo lock is held.
1283 in_pcbfree(struct inpcb *inp)
1285 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1287 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1290 if (pcbinfo == &V_tcbinfo) {
1291 INP_INFO_LOCK_ASSERT(pcbinfo);
1293 INP_INFO_WLOCK_ASSERT(pcbinfo);
1296 INP_WLOCK_ASSERT(inp);
1298 /* XXXRW: Do as much as possible here. */
1299 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1300 if (inp->inp_sp != NULL)
1301 ipsec_delete_pcbpolicy(inp);
1303 INP_LIST_WLOCK(pcbinfo);
1304 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1305 in_pcbremlists(inp);
1306 INP_LIST_WUNLOCK(pcbinfo);
1308 if (inp->inp_vflag & INP_IPV6PROTO) {
1309 ip6_freepcbopts(inp->in6p_outputopts);
1310 if (inp->in6p_moptions != NULL)
1311 ip6_freemoptions(inp->in6p_moptions);
1314 if (inp->inp_options)
1315 (void)m_free(inp->inp_options);
1317 if (inp->inp_moptions != NULL)
1318 inp_freemoptions(inp->inp_moptions);
1320 RO_RTFREE(&inp->inp_route);
1321 if (inp->inp_route.ro_lle)
1322 LLE_FREE(inp->inp_route.ro_lle); /* zeros ro_lle */
1325 inp->inp_flags2 |= INP_FREED;
1326 crfree(inp->inp_cred);
1328 mac_inpcb_destroy(inp);
1330 if (!in_pcbrele_wlocked(inp))
1335 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1336 * port reservation, and preventing it from being returned by inpcb lookups.
1338 * It is used by TCP to mark an inpcb as unused and avoid future packet
1339 * delivery or event notification when a socket remains open but TCP has
1340 * closed. This might occur as a result of a shutdown()-initiated TCP close
1341 * or a RST on the wire, and allows the port binding to be reused while still
1342 * maintaining the invariant that so_pcb always points to a valid inpcb until
1345 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1346 * in_pcbnotifyall() and in_pcbpurgeif0()?
1349 in_pcbdrop(struct inpcb *inp)
1352 INP_WLOCK_ASSERT(inp);
1355 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1358 inp->inp_flags |= INP_DROPPED;
1359 if (inp->inp_flags & INP_INHASHLIST) {
1360 struct inpcbport *phd = inp->inp_phd;
1362 INP_HASH_WLOCK(inp->inp_pcbinfo);
1363 LIST_REMOVE(inp, inp_hash);
1364 LIST_REMOVE(inp, inp_portlist);
1365 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1366 LIST_REMOVE(phd, phd_hash);
1369 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1370 inp->inp_flags &= ~INP_INHASHLIST;
1372 in_pcbgroup_remove(inp);
1379 * Common routines to return the socket addresses associated with inpcbs.
1382 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1384 struct sockaddr_in *sin;
1386 sin = malloc(sizeof *sin, M_SONAME,
1388 sin->sin_family = AF_INET;
1389 sin->sin_len = sizeof(*sin);
1390 sin->sin_addr = *addr_p;
1391 sin->sin_port = port;
1393 return (struct sockaddr *)sin;
1397 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1400 struct in_addr addr;
1403 inp = sotoinpcb(so);
1404 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1407 port = inp->inp_lport;
1408 addr = inp->inp_laddr;
1411 *nam = in_sockaddr(port, &addr);
1416 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1419 struct in_addr addr;
1422 inp = sotoinpcb(so);
1423 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1426 port = inp->inp_fport;
1427 addr = inp->inp_faddr;
1430 *nam = in_sockaddr(port, &addr);
1435 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1436 struct inpcb *(*notify)(struct inpcb *, int))
1438 struct inpcb *inp, *inp_temp;
1440 INP_INFO_WLOCK(pcbinfo);
1441 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1444 if ((inp->inp_vflag & INP_IPV4) == 0) {
1449 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1450 inp->inp_socket == NULL) {
1454 if ((*notify)(inp, errno))
1457 INP_INFO_WUNLOCK(pcbinfo);
1461 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1464 struct ip_moptions *imo;
1467 INP_INFO_WLOCK(pcbinfo);
1468 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1470 imo = inp->inp_moptions;
1471 if ((inp->inp_vflag & INP_IPV4) &&
1474 * Unselect the outgoing interface if it is being
1477 if (imo->imo_multicast_ifp == ifp)
1478 imo->imo_multicast_ifp = NULL;
1481 * Drop multicast group membership if we joined
1482 * through the interface being detached.
1484 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1486 if (imo->imo_membership[i]->inm_ifp == ifp) {
1487 in_delmulti(imo->imo_membership[i]);
1489 } else if (gap != 0)
1490 imo->imo_membership[i - gap] =
1491 imo->imo_membership[i];
1493 imo->imo_num_memberships -= gap;
1497 INP_INFO_WUNLOCK(pcbinfo);
1501 * Lookup a PCB based on the local address and port. Caller must hold the
1502 * hash lock. No inpcb locks or references are acquired.
1504 #define INP_LOOKUP_MAPPED_PCB_COST 3
1506 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1507 u_short lport, int lookupflags, struct ucred *cred)
1511 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1517 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1518 ("%s: invalid lookup flags %d", __func__, lookupflags));
1520 INP_HASH_LOCK_ASSERT(pcbinfo);
1522 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1523 struct inpcbhead *head;
1525 * Look for an unconnected (wildcard foreign addr) PCB that
1526 * matches the local address and port we're looking for.
1528 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1529 0, pcbinfo->ipi_hashmask)];
1530 LIST_FOREACH(inp, head, inp_hash) {
1532 /* XXX inp locking */
1533 if ((inp->inp_vflag & INP_IPV4) == 0)
1536 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1537 inp->inp_laddr.s_addr == laddr.s_addr &&
1538 inp->inp_lport == lport) {
1543 prison_equal_ip4(cred->cr_prison,
1544 inp->inp_cred->cr_prison))
1553 struct inpcbporthead *porthash;
1554 struct inpcbport *phd;
1555 struct inpcb *match = NULL;
1557 * Best fit PCB lookup.
1559 * First see if this local port is in use by looking on the
1562 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1563 pcbinfo->ipi_porthashmask)];
1564 LIST_FOREACH(phd, porthash, phd_hash) {
1565 if (phd->phd_port == lport)
1570 * Port is in use by one or more PCBs. Look for best
1573 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1576 !prison_equal_ip4(inp->inp_cred->cr_prison,
1580 /* XXX inp locking */
1581 if ((inp->inp_vflag & INP_IPV4) == 0)
1584 * We never select the PCB that has
1585 * INP_IPV6 flag and is bound to :: if
1586 * we have another PCB which is bound
1587 * to 0.0.0.0. If a PCB has the
1588 * INP_IPV6 flag, then we set its cost
1589 * higher than IPv4 only PCBs.
1591 * Note that the case only happens
1592 * when a socket is bound to ::, under
1593 * the condition that the use of the
1594 * mapped address is allowed.
1596 if ((inp->inp_vflag & INP_IPV6) != 0)
1597 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1599 if (inp->inp_faddr.s_addr != INADDR_ANY)
1601 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1602 if (laddr.s_addr == INADDR_ANY)
1604 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1607 if (laddr.s_addr != INADDR_ANY)
1610 if (wildcard < matchwild) {
1612 matchwild = wildcard;
1621 #undef INP_LOOKUP_MAPPED_PCB_COST
1625 * Lookup PCB in hash list, using pcbgroup tables.
1627 static struct inpcb *
1628 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1629 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1630 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1632 struct inpcbhead *head;
1633 struct inpcb *inp, *tmpinp;
1634 u_short fport = fport_arg, lport = lport_arg;
1637 * First look for an exact match.
1640 INP_GROUP_LOCK(pcbgroup);
1641 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1642 pcbgroup->ipg_hashmask)];
1643 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1645 /* XXX inp locking */
1646 if ((inp->inp_vflag & INP_IPV4) == 0)
1649 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1650 inp->inp_laddr.s_addr == laddr.s_addr &&
1651 inp->inp_fport == fport &&
1652 inp->inp_lport == lport) {
1654 * XXX We should be able to directly return
1655 * the inp here, without any checks.
1656 * Well unless both bound with SO_REUSEPORT?
1658 if (prison_flag(inp->inp_cred, PR_IP4))
1664 if (tmpinp != NULL) {
1671 * For incoming connections, we may wish to do a wildcard
1672 * match for an RSS-local socket.
1674 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1675 struct inpcb *local_wild = NULL, *local_exact = NULL;
1677 struct inpcb *local_wild_mapped = NULL;
1679 struct inpcb *jail_wild = NULL;
1680 struct inpcbhead *head;
1684 * Order of socket selection - we always prefer jails.
1685 * 1. jailed, non-wild.
1687 * 3. non-jailed, non-wild.
1688 * 4. non-jailed, wild.
1691 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1692 lport, 0, pcbgroup->ipg_hashmask)];
1693 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1695 /* XXX inp locking */
1696 if ((inp->inp_vflag & INP_IPV4) == 0)
1699 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1700 inp->inp_lport != lport)
1703 injail = prison_flag(inp->inp_cred, PR_IP4);
1705 if (prison_check_ip4(inp->inp_cred,
1709 if (local_exact != NULL)
1713 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1718 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1720 /* XXX inp locking, NULL check */
1721 if (inp->inp_vflag & INP_IPV6PROTO)
1722 local_wild_mapped = inp;
1730 } /* LIST_FOREACH */
1739 inp = local_wild_mapped;
1747 * Then look for a wildcard match, if requested.
1749 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1750 struct inpcb *local_wild = NULL, *local_exact = NULL;
1752 struct inpcb *local_wild_mapped = NULL;
1754 struct inpcb *jail_wild = NULL;
1755 struct inpcbhead *head;
1759 * Order of socket selection - we always prefer jails.
1760 * 1. jailed, non-wild.
1762 * 3. non-jailed, non-wild.
1763 * 4. non-jailed, wild.
1765 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1766 0, pcbinfo->ipi_wildmask)];
1767 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1769 /* XXX inp locking */
1770 if ((inp->inp_vflag & INP_IPV4) == 0)
1773 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1774 inp->inp_lport != lport)
1777 injail = prison_flag(inp->inp_cred, PR_IP4);
1779 if (prison_check_ip4(inp->inp_cred,
1783 if (local_exact != NULL)
1787 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1792 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1794 /* XXX inp locking, NULL check */
1795 if (inp->inp_vflag & INP_IPV6PROTO)
1796 local_wild_mapped = inp;
1804 } /* LIST_FOREACH */
1812 inp = local_wild_mapped;
1816 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1817 INP_GROUP_UNLOCK(pcbgroup);
1822 INP_GROUP_UNLOCK(pcbgroup);
1823 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1825 if (in_pcbrele_wlocked(inp))
1827 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1829 if (in_pcbrele_rlocked(inp))
1832 panic("%s: locking bug", __func__);
1835 #endif /* PCBGROUP */
1838 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
1839 * that the caller has locked the hash list, and will not perform any further
1840 * locking or reference operations on either the hash list or the connection.
1842 static struct inpcb *
1843 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1844 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1847 struct inpcbhead *head;
1848 struct inpcb *inp, *tmpinp;
1849 u_short fport = fport_arg, lport = lport_arg;
1851 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1852 ("%s: invalid lookup flags %d", __func__, lookupflags));
1854 INP_HASH_LOCK_ASSERT(pcbinfo);
1857 * First look for an exact match.
1860 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1861 pcbinfo->ipi_hashmask)];
1862 LIST_FOREACH(inp, head, inp_hash) {
1864 /* XXX inp locking */
1865 if ((inp->inp_vflag & INP_IPV4) == 0)
1868 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1869 inp->inp_laddr.s_addr == laddr.s_addr &&
1870 inp->inp_fport == fport &&
1871 inp->inp_lport == lport) {
1873 * XXX We should be able to directly return
1874 * the inp here, without any checks.
1875 * Well unless both bound with SO_REUSEPORT?
1877 if (prison_flag(inp->inp_cred, PR_IP4))
1887 * Then look for a wildcard match, if requested.
1889 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1890 struct inpcb *local_wild = NULL, *local_exact = NULL;
1892 struct inpcb *local_wild_mapped = NULL;
1894 struct inpcb *jail_wild = NULL;
1898 * Order of socket selection - we always prefer jails.
1899 * 1. jailed, non-wild.
1901 * 3. non-jailed, non-wild.
1902 * 4. non-jailed, wild.
1905 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1906 0, pcbinfo->ipi_hashmask)];
1907 LIST_FOREACH(inp, head, inp_hash) {
1909 /* XXX inp locking */
1910 if ((inp->inp_vflag & INP_IPV4) == 0)
1913 if (inp->inp_faddr.s_addr != INADDR_ANY ||
1914 inp->inp_lport != lport)
1917 injail = prison_flag(inp->inp_cred, PR_IP4);
1919 if (prison_check_ip4(inp->inp_cred,
1923 if (local_exact != NULL)
1927 if (inp->inp_laddr.s_addr == laddr.s_addr) {
1932 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1934 /* XXX inp locking, NULL check */
1935 if (inp->inp_vflag & INP_IPV6PROTO)
1936 local_wild_mapped = inp;
1944 } /* LIST_FOREACH */
1945 if (jail_wild != NULL)
1947 if (local_exact != NULL)
1948 return (local_exact);
1949 if (local_wild != NULL)
1950 return (local_wild);
1952 if (local_wild_mapped != NULL)
1953 return (local_wild_mapped);
1955 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1961 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
1962 * hash list lock, and will return the inpcb locked (i.e., requires
1963 * INPLOOKUP_LOCKPCB).
1965 static struct inpcb *
1966 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1967 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1972 INP_HASH_RLOCK(pcbinfo);
1973 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1974 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1977 INP_HASH_RUNLOCK(pcbinfo);
1978 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1980 if (in_pcbrele_wlocked(inp))
1982 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1984 if (in_pcbrele_rlocked(inp))
1987 panic("%s: locking bug", __func__);
1989 INP_HASH_RUNLOCK(pcbinfo);
1994 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1995 * from which a pre-calculated hash value may be extracted.
1997 * Possibly more of this logic should be in in_pcbgroup.c.
2000 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2001 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2003 #if defined(PCBGROUP) && !defined(RSS)
2004 struct inpcbgroup *pcbgroup;
2007 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2008 ("%s: invalid lookup flags %d", __func__, lookupflags));
2009 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2010 ("%s: LOCKPCB not set", __func__));
2013 * When not using RSS, use connection groups in preference to the
2014 * reservation table when looking up 4-tuples. When using RSS, just
2015 * use the reservation table, due to the cost of the Toeplitz hash
2018 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2019 * we could be doing RSS with a non-Toeplitz hash that is affordable
2022 #if defined(PCBGROUP) && !defined(RSS)
2023 if (in_pcbgroup_enabled(pcbinfo)) {
2024 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2026 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2027 laddr, lport, lookupflags, ifp));
2030 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2035 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2036 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2037 struct ifnet *ifp, struct mbuf *m)
2040 struct inpcbgroup *pcbgroup;
2043 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2044 ("%s: invalid lookup flags %d", __func__, lookupflags));
2045 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2046 ("%s: LOCKPCB not set", __func__));
2050 * If we can use a hardware-generated hash to look up the connection
2051 * group, use that connection group to find the inpcb. Otherwise
2052 * fall back on a software hash -- or the reservation table if we're
2055 * XXXRW: As above, that policy belongs in the pcbgroup code.
2057 if (in_pcbgroup_enabled(pcbinfo) &&
2058 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2059 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2060 m->m_pkthdr.flowid);
2061 if (pcbgroup != NULL)
2062 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2063 fport, laddr, lport, lookupflags, ifp));
2065 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2067 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2068 laddr, lport, lookupflags, ifp));
2072 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2078 * Insert PCB onto various hash lists.
2081 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2083 struct inpcbhead *pcbhash;
2084 struct inpcbporthead *pcbporthash;
2085 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2086 struct inpcbport *phd;
2087 u_int32_t hashkey_faddr;
2089 INP_WLOCK_ASSERT(inp);
2090 INP_HASH_WLOCK_ASSERT(pcbinfo);
2092 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2093 ("in_pcbinshash: INP_INHASHLIST"));
2096 if (inp->inp_vflag & INP_IPV6)
2097 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2100 hashkey_faddr = inp->inp_faddr.s_addr;
2102 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2103 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2105 pcbporthash = &pcbinfo->ipi_porthashbase[
2106 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2109 * Go through port list and look for a head for this lport.
2111 LIST_FOREACH(phd, pcbporthash, phd_hash) {
2112 if (phd->phd_port == inp->inp_lport)
2116 * If none exists, malloc one and tack it on.
2119 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2121 return (ENOBUFS); /* XXX */
2123 phd->phd_port = inp->inp_lport;
2124 LIST_INIT(&phd->phd_pcblist);
2125 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2128 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2129 LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2130 inp->inp_flags |= INP_INHASHLIST;
2132 if (do_pcbgroup_update)
2133 in_pcbgroup_update(inp);
2139 * For now, there are two public interfaces to insert an inpcb into the hash
2140 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2141 * is used only in the TCP syncache, where in_pcbinshash is called before the
2142 * full 4-tuple is set for the inpcb, and we don't want to install in the
2143 * pcbgroup until later.
2145 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2146 * connection groups, and partially initialised inpcbs should not be exposed
2147 * to either reservation hash tables or pcbgroups.
2150 in_pcbinshash(struct inpcb *inp)
2153 return (in_pcbinshash_internal(inp, 1));
2157 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2160 return (in_pcbinshash_internal(inp, 0));
2164 * Move PCB to the proper hash bucket when { faddr, fport } have been
2165 * changed. NOTE: This does not handle the case of the lport changing (the
2166 * hashed port list would have to be updated as well), so the lport must
2167 * not change after in_pcbinshash() has been called.
2170 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2172 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2173 struct inpcbhead *head;
2174 u_int32_t hashkey_faddr;
2176 INP_WLOCK_ASSERT(inp);
2177 INP_HASH_WLOCK_ASSERT(pcbinfo);
2179 KASSERT(inp->inp_flags & INP_INHASHLIST,
2180 ("in_pcbrehash: !INP_INHASHLIST"));
2183 if (inp->inp_vflag & INP_IPV6)
2184 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2187 hashkey_faddr = inp->inp_faddr.s_addr;
2189 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2190 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2192 LIST_REMOVE(inp, inp_hash);
2193 LIST_INSERT_HEAD(head, inp, inp_hash);
2197 in_pcbgroup_update_mbuf(inp, m);
2199 in_pcbgroup_update(inp);
2204 in_pcbrehash(struct inpcb *inp)
2207 in_pcbrehash_mbuf(inp, NULL);
2211 * Remove PCB from various lists.
2214 in_pcbremlists(struct inpcb *inp)
2216 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2219 if (pcbinfo == &V_tcbinfo) {
2220 INP_INFO_RLOCK_ASSERT(pcbinfo);
2222 INP_INFO_WLOCK_ASSERT(pcbinfo);
2226 INP_WLOCK_ASSERT(inp);
2227 INP_LIST_WLOCK_ASSERT(pcbinfo);
2229 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2230 if (inp->inp_flags & INP_INHASHLIST) {
2231 struct inpcbport *phd = inp->inp_phd;
2233 INP_HASH_WLOCK(pcbinfo);
2234 LIST_REMOVE(inp, inp_hash);
2235 LIST_REMOVE(inp, inp_portlist);
2236 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2237 LIST_REMOVE(phd, phd_hash);
2240 INP_HASH_WUNLOCK(pcbinfo);
2241 inp->inp_flags &= ~INP_INHASHLIST;
2243 LIST_REMOVE(inp, inp_list);
2244 pcbinfo->ipi_count--;
2246 in_pcbgroup_remove(inp);
2251 * Check for alternatives when higher level complains
2252 * about service problems. For now, invalidate cached
2253 * routing information. If the route was created dynamically
2254 * (by a redirect), time to try a default gateway again.
2257 in_losing(struct inpcb *inp)
2260 RO_RTFREE(&inp->inp_route);
2261 if (inp->inp_route.ro_lle)
2262 LLE_FREE(inp->inp_route.ro_lle); /* zeros ro_lle */
2267 * A set label operation has occurred at the socket layer, propagate the
2268 * label change into the in_pcb for the socket.
2271 in_pcbsosetlabel(struct socket *so)
2276 inp = sotoinpcb(so);
2277 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2281 mac_inpcb_sosetlabel(so, inp);
2288 * ipport_tick runs once per second, determining if random port allocation
2289 * should be continued. If more than ipport_randomcps ports have been
2290 * allocated in the last second, then we return to sequential port
2291 * allocation. We return to random allocation only once we drop below
2292 * ipport_randomcps for at least ipport_randomtime seconds.
2295 ipport_tick(void *xtp)
2297 VNET_ITERATOR_DECL(vnet_iter);
2299 VNET_LIST_RLOCK_NOSLEEP();
2300 VNET_FOREACH(vnet_iter) {
2301 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2302 if (V_ipport_tcpallocs <=
2303 V_ipport_tcplastcount + V_ipport_randomcps) {
2304 if (V_ipport_stoprandom > 0)
2305 V_ipport_stoprandom--;
2307 V_ipport_stoprandom = V_ipport_randomtime;
2308 V_ipport_tcplastcount = V_ipport_tcpallocs;
2311 VNET_LIST_RUNLOCK_NOSLEEP();
2312 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2319 callout_stop(&ipport_tick_callout);
2323 * The ipport_callout should start running at about the time we attach the
2324 * inet or inet6 domains.
2327 ipport_tick_init(const void *unused __unused)
2330 /* Start ipport_tick. */
2331 callout_init(&ipport_tick_callout, 1);
2332 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2333 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2334 SHUTDOWN_PRI_DEFAULT);
2336 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2337 ipport_tick_init, NULL);
2340 inp_wlock(struct inpcb *inp)
2347 inp_wunlock(struct inpcb *inp)
2354 inp_rlock(struct inpcb *inp)
2361 inp_runlock(struct inpcb *inp)
2367 #ifdef INVARIANT_SUPPORT
2369 inp_lock_assert(struct inpcb *inp)
2372 INP_WLOCK_ASSERT(inp);
2376 inp_unlock_assert(struct inpcb *inp)
2379 INP_UNLOCK_ASSERT(inp);
2384 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2388 INP_INFO_WLOCK(&V_tcbinfo);
2389 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2394 INP_INFO_WUNLOCK(&V_tcbinfo);
2398 inp_inpcbtosocket(struct inpcb *inp)
2401 INP_WLOCK_ASSERT(inp);
2402 return (inp->inp_socket);
2406 inp_inpcbtotcpcb(struct inpcb *inp)
2409 INP_WLOCK_ASSERT(inp);
2410 return ((struct tcpcb *)inp->inp_ppcb);
2414 inp_ip_tos_get(const struct inpcb *inp)
2417 return (inp->inp_ip_tos);
2421 inp_ip_tos_set(struct inpcb *inp, int val)
2424 inp->inp_ip_tos = val;
2428 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2429 uint32_t *faddr, uint16_t *fp)
2432 INP_LOCK_ASSERT(inp);
2433 *laddr = inp->inp_laddr.s_addr;
2434 *faddr = inp->inp_faddr.s_addr;
2435 *lp = inp->inp_lport;
2436 *fp = inp->inp_fport;
2440 so_sotoinpcb(struct socket *so)
2443 return (sotoinpcb(so));
2447 so_sototcpcb(struct socket *so)
2450 return (sototcpcb(so));
2454 * Create an external-format (``xinpcb'') structure using the information in
2455 * the kernel-format in_pcb structure pointed to by inp. This is done to
2456 * reduce the spew of irrelevant information over this interface, to isolate
2457 * user code from changes in the kernel structure, and potentially to provide
2458 * information-hiding if we decide that some of this information should be
2459 * hidden from users.
2462 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2465 xi->xi_len = sizeof(struct xinpcb);
2466 if (inp->inp_socket)
2467 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2469 bzero(&xi->xi_socket, sizeof(struct xsocket));
2470 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2471 xi->inp_gencnt = inp->inp_gencnt;
2472 xi->inp_ppcb = inp->inp_ppcb;
2473 xi->inp_flow = inp->inp_flow;
2474 xi->inp_flowid = inp->inp_flowid;
2475 xi->inp_flowtype = inp->inp_flowtype;
2476 xi->inp_flags = inp->inp_flags;
2477 xi->inp_flags2 = inp->inp_flags2;
2478 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2479 xi->in6p_cksum = inp->in6p_cksum;
2480 xi->in6p_hops = inp->in6p_hops;
2481 xi->inp_ip_tos = inp->inp_ip_tos;
2482 xi->inp_vflag = inp->inp_vflag;
2483 xi->inp_ip_ttl = inp->inp_ip_ttl;
2484 xi->inp_ip_p = inp->inp_ip_p;
2485 xi->inp_ip_minttl = inp->inp_ip_minttl;
2490 db_print_indent(int indent)
2494 for (i = 0; i < indent; i++)
2499 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2501 char faddr_str[48], laddr_str[48];
2503 db_print_indent(indent);
2504 db_printf("%s at %p\n", name, inc);
2509 if (inc->inc_flags & INC_ISIPV6) {
2511 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2512 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2517 inet_ntoa_r(inc->inc_laddr, laddr_str);
2518 inet_ntoa_r(inc->inc_faddr, faddr_str);
2520 db_print_indent(indent);
2521 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2522 ntohs(inc->inc_lport));
2523 db_print_indent(indent);
2524 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2525 ntohs(inc->inc_fport));
2529 db_print_inpflags(int inp_flags)
2534 if (inp_flags & INP_RECVOPTS) {
2535 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2538 if (inp_flags & INP_RECVRETOPTS) {
2539 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2542 if (inp_flags & INP_RECVDSTADDR) {
2543 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2546 if (inp_flags & INP_ORIGDSTADDR) {
2547 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2550 if (inp_flags & INP_HDRINCL) {
2551 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2554 if (inp_flags & INP_HIGHPORT) {
2555 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2558 if (inp_flags & INP_LOWPORT) {
2559 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2562 if (inp_flags & INP_ANONPORT) {
2563 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2566 if (inp_flags & INP_RECVIF) {
2567 db_printf("%sINP_RECVIF", comma ? ", " : "");
2570 if (inp_flags & INP_MTUDISC) {
2571 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2574 if (inp_flags & INP_RECVTTL) {
2575 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2578 if (inp_flags & INP_DONTFRAG) {
2579 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2582 if (inp_flags & INP_RECVTOS) {
2583 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2586 if (inp_flags & IN6P_IPV6_V6ONLY) {
2587 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2590 if (inp_flags & IN6P_PKTINFO) {
2591 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2594 if (inp_flags & IN6P_HOPLIMIT) {
2595 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2598 if (inp_flags & IN6P_HOPOPTS) {
2599 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2602 if (inp_flags & IN6P_DSTOPTS) {
2603 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2606 if (inp_flags & IN6P_RTHDR) {
2607 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2610 if (inp_flags & IN6P_RTHDRDSTOPTS) {
2611 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2614 if (inp_flags & IN6P_TCLASS) {
2615 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2618 if (inp_flags & IN6P_AUTOFLOWLABEL) {
2619 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2622 if (inp_flags & INP_TIMEWAIT) {
2623 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2626 if (inp_flags & INP_ONESBCAST) {
2627 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2630 if (inp_flags & INP_DROPPED) {
2631 db_printf("%sINP_DROPPED", comma ? ", " : "");
2634 if (inp_flags & INP_SOCKREF) {
2635 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2638 if (inp_flags & IN6P_RFC2292) {
2639 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2642 if (inp_flags & IN6P_MTU) {
2643 db_printf("IN6P_MTU%s", comma ? ", " : "");
2649 db_print_inpvflag(u_char inp_vflag)
2654 if (inp_vflag & INP_IPV4) {
2655 db_printf("%sINP_IPV4", comma ? ", " : "");
2658 if (inp_vflag & INP_IPV6) {
2659 db_printf("%sINP_IPV6", comma ? ", " : "");
2662 if (inp_vflag & INP_IPV6PROTO) {
2663 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2669 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2672 db_print_indent(indent);
2673 db_printf("%s at %p\n", name, inp);
2677 db_print_indent(indent);
2678 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2680 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2682 db_print_indent(indent);
2683 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
2684 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2686 db_print_indent(indent);
2687 db_printf("inp_label: %p inp_flags: 0x%x (",
2688 inp->inp_label, inp->inp_flags);
2689 db_print_inpflags(inp->inp_flags);
2692 db_print_indent(indent);
2693 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
2695 db_print_inpvflag(inp->inp_vflag);
2698 db_print_indent(indent);
2699 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
2700 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2702 db_print_indent(indent);
2704 if (inp->inp_vflag & INP_IPV6) {
2705 db_printf("in6p_options: %p in6p_outputopts: %p "
2706 "in6p_moptions: %p\n", inp->in6p_options,
2707 inp->in6p_outputopts, inp->in6p_moptions);
2708 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
2709 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2714 db_printf("inp_ip_tos: %d inp_ip_options: %p "
2715 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2716 inp->inp_options, inp->inp_moptions);
2719 db_print_indent(indent);
2720 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
2721 (uintmax_t)inp->inp_gencnt);
2724 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2729 db_printf("usage: show inpcb <addr>\n");
2732 inp = (struct inpcb *)addr;
2734 db_print_inpcb(inp, "inpcb", 0);
2740 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
2744 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
2746 union if_snd_tag_modify_params params = {
2747 .rate_limit.max_rate = max_pacing_rate,
2749 struct m_snd_tag *mst;
2753 mst = inp->inp_snd_tag;
2761 if (ifp->if_snd_tag_modify == NULL) {
2764 error = ifp->if_snd_tag_modify(mst, ¶ms);
2770 * Query existing TX rate limit based on the existing
2771 * "inp->inp_snd_tag", if any.
2774 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
2776 union if_snd_tag_query_params params = { };
2777 struct m_snd_tag *mst;
2781 mst = inp->inp_snd_tag;
2789 if (ifp->if_snd_tag_query == NULL) {
2792 error = ifp->if_snd_tag_query(mst, ¶ms);
2793 if (error == 0 && p_max_pacing_rate != NULL)
2794 *p_max_pacing_rate = params.rate_limit.max_rate;
2800 * Allocate a new TX rate limit send tag from the network interface
2801 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
2804 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
2805 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
2807 union if_snd_tag_alloc_params params = {
2808 .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT,
2809 .rate_limit.hdr.flowid = flowid,
2810 .rate_limit.hdr.flowtype = flowtype,
2811 .rate_limit.max_rate = max_pacing_rate,
2815 INP_WLOCK_ASSERT(inp);
2817 if (inp->inp_snd_tag != NULL)
2820 if (ifp->if_snd_tag_alloc == NULL) {
2823 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
2826 * At success increment the refcount on
2827 * the send tag's network interface:
2830 if_ref(inp->inp_snd_tag->ifp);
2836 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
2840 in_pcbdetach_txrtlmt(struct inpcb *inp)
2842 struct m_snd_tag *mst;
2845 INP_WLOCK_ASSERT(inp);
2847 mst = inp->inp_snd_tag;
2848 inp->inp_snd_tag = NULL;
2858 * If the device was detached while we still had reference(s)
2859 * on the ifp, we assume if_snd_tag_free() was replaced with
2862 ifp->if_snd_tag_free(mst);
2864 /* release reference count on network interface */
2869 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
2870 * is set in the fast path and will attach/detach/modify the TX rate
2871 * limit send tag based on the socket's so_max_pacing_rate value.
2874 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
2876 struct socket *socket;
2877 uint32_t max_pacing_rate;
2884 socket = inp->inp_socket;
2888 if (!INP_WLOCKED(inp)) {
2890 * NOTE: If the write locking fails, we need to bail
2891 * out and use the non-ratelimited ring for the
2892 * transmit until there is a new chance to get the
2895 if (!INP_TRY_UPGRADE(inp))
2903 * NOTE: The so_max_pacing_rate value is read unlocked,
2904 * because atomic updates are not required since the variable
2905 * is checked at every mbuf we send. It is assumed that the
2906 * variable read itself will be atomic.
2908 max_pacing_rate = socket->so_max_pacing_rate;
2911 * NOTE: When attaching to a network interface a reference is
2912 * made to ensure the network interface doesn't go away until
2913 * all ratelimit connections are gone. The network interface
2914 * pointers compared below represent valid network interfaces,
2915 * except when comparing towards NULL.
2917 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
2919 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
2920 if (inp->inp_snd_tag != NULL)
2921 in_pcbdetach_txrtlmt(inp);
2923 } else if (inp->inp_snd_tag == NULL) {
2925 * In order to utilize packet pacing with RSS, we need
2926 * to wait until there is a valid RSS hash before we
2929 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
2932 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
2933 mb->m_pkthdr.flowid, max_pacing_rate);
2936 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
2938 if (error == 0 || error == EOPNOTSUPP)
2939 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
2945 * Track route changes for TX rate limiting.
2948 in_pcboutput_eagain(struct inpcb *inp)
2950 struct socket *socket;
2956 socket = inp->inp_socket;
2960 if (inp->inp_snd_tag == NULL)
2963 if (!INP_WLOCKED(inp)) {
2965 * NOTE: If the write locking fails, we need to bail
2966 * out and use the non-ratelimited ring for the
2967 * transmit until there is a new chance to get the
2970 if (!INP_TRY_UPGRADE(inp))
2977 /* detach rate limiting */
2978 in_pcbdetach_txrtlmt(inp);
2980 /* make sure new mbuf send tag allocation is made */
2981 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
2986 #endif /* RATELIMIT */