2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1991, 1993, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2007-2009 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 #include "opt_ipsec.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/malloc.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
67 #include <sys/refcount.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
90 #include <netinet/in_var.h>
92 #include <netinet/ip_var.h>
93 #include <netinet/tcp_var.h>
95 #include <netinet/tcp_hpts.h>
97 #include <netinet/udp.h>
98 #include <netinet/udp_var.h>
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
107 #include <netipsec/ipsec_support.h>
109 #include <security/mac/mac_framework.h>
111 #define INPCBLBGROUP_SIZMIN 8
112 #define INPCBLBGROUP_SIZMAX 256
114 static struct callout ipport_tick_callout;
117 * These configure the range of local port addresses assigned to
118 * "unspecified" outgoing connections/packets/whatever.
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
128 * Reserved ports accessible only to root. There are significant
129 * security considerations that must be accounted for when changing these,
130 * but the security benefits can be great. Please be careful.
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
145 static void in_pcbremlists(struct inpcb *inp);
147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 struct in_addr faddr, u_int fport_arg,
149 struct in_addr laddr, u_int lport_arg,
150 int lookupflags, struct ifnet *ifp);
152 #define RANGECHK(var, min, max) \
153 if ((var) < (min)) { (var) = (min); } \
154 else if ((var) > (max)) { (var) = (max); }
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
161 error = sysctl_handle_int(oidp, arg1, arg2, req);
163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 &VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 CTLFLAG_VNET | CTLFLAG_RW,
206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 "allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(ipport_randomtime), 0,
211 "Minimum time to keep sequental port "
212 "allocation before switching to a random one");
215 counter_u64_t rate_limit_active;
216 counter_u64_t rate_limit_alloc_fail;
217 counter_u64_t rate_limit_set_ok;
219 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD, 0,
221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
222 &rate_limit_active, "Active rate limited connections");
223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
224 &rate_limit_alloc_fail, "Rate limited connection failures");
225 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
226 &rate_limit_set_ok, "Rate limited setting succeeded");
227 #endif /* RATELIMIT */
232 * in_pcb.c: manage the Protocol Control Blocks.
234 * NOTE: It is assumed that most of these functions will be called with
235 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
236 * functions often modify hash chains or addresses in pcbs.
239 static struct inpcblbgroup *
240 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
241 uint16_t port, const union in_dependaddr *addr, int size)
243 struct inpcblbgroup *grp;
246 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
247 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
250 grp->il_vflag = vflag;
251 grp->il_lport = port;
252 grp->il_dependladdr = *addr;
253 grp->il_inpsiz = size;
254 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
259 in_pcblbgroup_free_deferred(epoch_context_t ctx)
261 struct inpcblbgroup *grp;
263 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
268 in_pcblbgroup_free(struct inpcblbgroup *grp)
271 CK_LIST_REMOVE(grp, il_list);
272 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
275 static struct inpcblbgroup *
276 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
277 struct inpcblbgroup *old_grp, int size)
279 struct inpcblbgroup *grp;
282 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
283 old_grp->il_lport, &old_grp->il_dependladdr, size);
287 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
288 ("invalid new local group size %d and old local group count %d",
289 grp->il_inpsiz, old_grp->il_inpcnt));
291 for (i = 0; i < old_grp->il_inpcnt; ++i)
292 grp->il_inp[i] = old_grp->il_inp[i];
293 grp->il_inpcnt = old_grp->il_inpcnt;
294 in_pcblbgroup_free(old_grp);
299 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
300 * and shrink group if possible.
303 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
306 struct inpcblbgroup *grp, *new_grp;
309 for (; i + 1 < grp->il_inpcnt; ++i)
310 grp->il_inp[i] = grp->il_inp[i + 1];
313 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
314 grp->il_inpcnt <= grp->il_inpsiz / 4) {
315 /* Shrink this group. */
316 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
323 * Add PCB to load balance group for SO_REUSEPORT_LB option.
326 in_pcbinslbgrouphash(struct inpcb *inp)
328 const static struct timeval interval = { 60, 0 };
329 static struct timeval lastprint;
330 struct inpcbinfo *pcbinfo;
331 struct inpcblbgrouphead *hdr;
332 struct inpcblbgroup *grp;
335 pcbinfo = inp->inp_pcbinfo;
337 INP_WLOCK_ASSERT(inp);
338 INP_HASH_WLOCK_ASSERT(pcbinfo);
341 * Don't allow jailed socket to join local group.
343 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
348 * Don't allow IPv4 mapped INET6 wild socket.
350 if ((inp->inp_vflag & INP_IPV4) &&
351 inp->inp_laddr.s_addr == INADDR_ANY &&
352 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
357 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
358 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
359 CK_LIST_FOREACH(grp, hdr, il_list) {
360 if (grp->il_vflag == inp->inp_vflag &&
361 grp->il_lport == inp->inp_lport &&
362 memcmp(&grp->il_dependladdr,
363 &inp->inp_inc.inc_ie.ie_dependladdr,
364 sizeof(grp->il_dependladdr)) == 0)
368 /* Create new load balance group. */
369 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
370 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
371 INPCBLBGROUP_SIZMIN);
374 } else if (grp->il_inpcnt == grp->il_inpsiz) {
375 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
376 if (ratecheck(&lastprint, &interval))
377 printf("lb group port %d, limit reached\n",
378 ntohs(grp->il_lport));
382 /* Expand this local group. */
383 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
388 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
389 ("invalid local group size %d and count %d", grp->il_inpsiz,
392 grp->il_inp[grp->il_inpcnt] = inp;
398 * Remove PCB from load balance group.
401 in_pcbremlbgrouphash(struct inpcb *inp)
403 struct inpcbinfo *pcbinfo;
404 struct inpcblbgrouphead *hdr;
405 struct inpcblbgroup *grp;
408 pcbinfo = inp->inp_pcbinfo;
410 INP_WLOCK_ASSERT(inp);
411 INP_HASH_WLOCK_ASSERT(pcbinfo);
413 hdr = &pcbinfo->ipi_lbgrouphashbase[
414 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
415 CK_LIST_FOREACH(grp, hdr, il_list) {
416 for (i = 0; i < grp->il_inpcnt; ++i) {
417 if (grp->il_inp[i] != inp)
420 if (grp->il_inpcnt == 1) {
421 /* We are the last, free this local group. */
422 in_pcblbgroup_free(grp);
424 /* Pull up inpcbs, shrink group if possible. */
425 in_pcblbgroup_reorder(hdr, &grp, i);
433 * Different protocols initialize their inpcbs differently - giving
434 * different name to the lock. But they all are disposed the same.
437 inpcb_fini(void *mem, int size)
439 struct inpcb *inp = mem;
441 INP_LOCK_DESTROY(inp);
445 * Initialize an inpcbinfo -- we should be able to reduce the number of
449 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
450 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
451 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
454 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
456 INP_INFO_LOCK_INIT(pcbinfo, name);
457 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
458 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
460 pcbinfo->ipi_vnet = curvnet;
462 pcbinfo->ipi_listhead = listhead;
463 CK_LIST_INIT(pcbinfo->ipi_listhead);
464 pcbinfo->ipi_count = 0;
465 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
466 &pcbinfo->ipi_hashmask);
467 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
468 &pcbinfo->ipi_porthashmask);
469 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
470 &pcbinfo->ipi_lbgrouphashmask);
472 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
474 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
475 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
476 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
477 uma_zone_set_warning(pcbinfo->ipi_zone,
478 "kern.ipc.maxsockets limit reached");
482 * Destroy an inpcbinfo.
485 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
488 KASSERT(pcbinfo->ipi_count == 0,
489 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
491 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
492 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
493 pcbinfo->ipi_porthashmask);
494 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
495 pcbinfo->ipi_lbgrouphashmask);
497 in_pcbgroup_destroy(pcbinfo);
499 uma_zdestroy(pcbinfo->ipi_zone);
500 INP_LIST_LOCK_DESTROY(pcbinfo);
501 INP_HASH_LOCK_DESTROY(pcbinfo);
502 INP_INFO_LOCK_DESTROY(pcbinfo);
506 * Allocate a PCB and associate it with the socket.
507 * On success return with the PCB locked.
510 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
516 if (pcbinfo == &V_tcbinfo) {
519 INP_INFO_WLOCK_ASSERT(pcbinfo);
524 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
527 bzero(&inp->inp_start_zero, inp_zero_size);
529 inp->inp_numa_domain = M_NODOM;
531 inp->inp_pcbinfo = pcbinfo;
532 inp->inp_socket = so;
533 inp->inp_cred = crhold(so->so_cred);
534 inp->inp_inc.inc_fibnum = so->so_fibnum;
536 error = mac_inpcb_init(inp, M_NOWAIT);
539 mac_inpcb_create(so, inp);
541 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
542 error = ipsec_init_pcbpolicy(inp);
545 mac_inpcb_destroy(inp);
551 if (INP_SOCKAF(so) == AF_INET6) {
552 inp->inp_vflag |= INP_IPV6PROTO;
554 inp->inp_flags |= IN6P_IPV6_V6ONLY;
558 INP_LIST_WLOCK(pcbinfo);
559 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
560 pcbinfo->ipi_count++;
561 so->so_pcb = (caddr_t)inp;
563 if (V_ip6_auto_flowlabel)
564 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
566 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
567 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
570 * Routes in inpcb's can cache L2 as well; they are guaranteed
573 inp->inp_route.ro_flags = RT_LLE_CACHE;
574 INP_LIST_WUNLOCK(pcbinfo);
575 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
578 crfree(inp->inp_cred);
579 uma_zfree(pcbinfo->ipi_zone, inp);
587 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
591 INP_WLOCK_ASSERT(inp);
592 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
594 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
596 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
597 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
598 &inp->inp_lport, cred);
601 if (in_pcbinshash(inp) != 0) {
602 inp->inp_laddr.s_addr = INADDR_ANY;
607 inp->inp_flags |= INP_ANONPORT;
613 * Select a local port (number) to use.
615 #if defined(INET) || defined(INET6)
617 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
618 struct ucred *cred, int lookupflags)
620 struct inpcbinfo *pcbinfo;
621 struct inpcb *tmpinp;
622 unsigned short *lastport;
623 int count, dorandom, error;
624 u_short aux, first, last, lport;
626 struct in_addr laddr;
629 pcbinfo = inp->inp_pcbinfo;
632 * Because no actual state changes occur here, a global write lock on
633 * the pcbinfo isn't required.
635 INP_LOCK_ASSERT(inp);
636 INP_HASH_LOCK_ASSERT(pcbinfo);
638 if (inp->inp_flags & INP_HIGHPORT) {
639 first = V_ipport_hifirstauto; /* sysctl */
640 last = V_ipport_hilastauto;
641 lastport = &pcbinfo->ipi_lasthi;
642 } else if (inp->inp_flags & INP_LOWPORT) {
643 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
646 first = V_ipport_lowfirstauto; /* 1023 */
647 last = V_ipport_lowlastauto; /* 600 */
648 lastport = &pcbinfo->ipi_lastlow;
650 first = V_ipport_firstauto; /* sysctl */
651 last = V_ipport_lastauto;
652 lastport = &pcbinfo->ipi_lastport;
655 * For UDP(-Lite), use random port allocation as long as the user
656 * allows it. For TCP (and as of yet unknown) connections,
657 * use random port allocation only if the user allows it AND
658 * ipport_tick() allows it.
660 if (V_ipport_randomized &&
661 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
662 pcbinfo == &V_ulitecbinfo))
667 * It makes no sense to do random port allocation if
668 * we have the only port available.
672 /* Make sure to not include UDP(-Lite) packets in the count. */
673 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
674 V_ipport_tcpallocs++;
676 * Instead of having two loops further down counting up or down
677 * make sure that first is always <= last and go with only one
678 * code path implementing all logic.
687 /* Make the compiler happy. */
689 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
690 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
695 tmpinp = NULL; /* Make compiler happy. */
699 *lastport = first + (arc4random() % (last - first));
701 count = last - first;
704 if (count-- < 0) /* completely used? */
705 return (EADDRNOTAVAIL);
707 if (*lastport < first || *lastport > last)
709 lport = htons(*lastport);
712 if ((inp->inp_vflag & INP_IPV6) != 0)
713 tmpinp = in6_pcblookup_local(pcbinfo,
714 &inp->in6p_laddr, lport, lookupflags, cred);
716 #if defined(INET) && defined(INET6)
720 tmpinp = in_pcblookup_local(pcbinfo, laddr,
721 lport, lookupflags, cred);
723 } while (tmpinp != NULL);
726 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
727 laddrp->s_addr = laddr.s_addr;
735 * Return cached socket options.
738 inp_so_options(const struct inpcb *inp)
744 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
745 so_options |= SO_REUSEPORT_LB;
746 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
747 so_options |= SO_REUSEPORT;
748 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
749 so_options |= SO_REUSEADDR;
752 #endif /* INET || INET6 */
755 * Check if a new BINDMULTI socket is allowed to be created.
757 * ni points to the new inp.
758 * oi points to the exisitng inp.
760 * This checks whether the existing inp also has BINDMULTI and
761 * whether the credentials match.
764 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
766 /* Check permissions match */
767 if ((ni->inp_flags2 & INP_BINDMULTI) &&
768 (ni->inp_cred->cr_uid !=
769 oi->inp_cred->cr_uid))
772 /* Check the existing inp has BINDMULTI set */
773 if ((ni->inp_flags2 & INP_BINDMULTI) &&
774 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
778 * We're okay - either INP_BINDMULTI isn't set on ni, or
779 * it is and it matches the checks.
786 * Set up a bind operation on a PCB, performing port allocation
787 * as required, but do not actually modify the PCB. Callers can
788 * either complete the bind by setting inp_laddr/inp_lport and
789 * calling in_pcbinshash(), or they can just use the resulting
790 * port and address to authorise the sending of a once-off packet.
792 * On error, the values of *laddrp and *lportp are not changed.
795 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
796 u_short *lportp, struct ucred *cred)
798 struct socket *so = inp->inp_socket;
799 struct sockaddr_in *sin;
800 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
801 struct in_addr laddr;
803 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
807 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
808 * so that we don't have to add to the (already messy) code below.
810 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
813 * No state changes, so read locks are sufficient here.
815 INP_LOCK_ASSERT(inp);
816 INP_HASH_LOCK_ASSERT(pcbinfo);
818 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
819 return (EADDRNOTAVAIL);
820 laddr.s_addr = *laddrp;
821 if (nam != NULL && laddr.s_addr != INADDR_ANY)
823 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
824 lookupflags = INPLOOKUP_WILDCARD;
826 if ((error = prison_local_ip4(cred, &laddr)) != 0)
829 sin = (struct sockaddr_in *)nam;
830 if (nam->sa_len != sizeof (*sin))
834 * We should check the family, but old programs
835 * incorrectly fail to initialize it.
837 if (sin->sin_family != AF_INET)
838 return (EAFNOSUPPORT);
840 error = prison_local_ip4(cred, &sin->sin_addr);
843 if (sin->sin_port != *lportp) {
844 /* Don't allow the port to change. */
847 lport = sin->sin_port;
849 /* NB: lport is left as 0 if the port isn't being changed. */
850 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
852 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
853 * allow complete duplication of binding if
854 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
855 * and a multicast address is bound on both
856 * new and duplicated sockets.
858 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
859 reuseport = SO_REUSEADDR|SO_REUSEPORT;
861 * XXX: How to deal with SO_REUSEPORT_LB here?
862 * Treat same as SO_REUSEPORT for now.
864 if ((so->so_options &
865 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
866 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
867 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
868 sin->sin_port = 0; /* yech... */
869 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
871 * Is the address a local IP address?
872 * If INP_BINDANY is set, then the socket may be bound
873 * to any endpoint address, local or not.
875 if ((inp->inp_flags & INP_BINDANY) == 0 &&
876 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
877 return (EADDRNOTAVAIL);
879 laddr = sin->sin_addr;
885 if (ntohs(lport) <= V_ipport_reservedhigh &&
886 ntohs(lport) >= V_ipport_reservedlow &&
887 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
889 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
890 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
891 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
892 lport, INPLOOKUP_WILDCARD, cred);
895 * This entire block sorely needs a rewrite.
898 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
899 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
900 (so->so_type != SOCK_STREAM ||
901 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
902 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
903 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
904 (t->inp_flags2 & INP_REUSEPORT) ||
905 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
906 (inp->inp_cred->cr_uid !=
907 t->inp_cred->cr_uid))
911 * If the socket is a BINDMULTI socket, then
912 * the credentials need to match and the
913 * original socket also has to have been bound
916 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
919 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
920 lport, lookupflags, cred);
921 if (t && (t->inp_flags & INP_TIMEWAIT)) {
923 * XXXRW: If an incpb has had its timewait
924 * state recycled, we treat the address as
925 * being in use (for now). This is better
926 * than a panic, but not desirable.
930 ((reuseport & tw->tw_so_options) == 0 &&
932 tw->tw_so_options) == 0)) {
936 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
937 (reuseport & inp_so_options(t)) == 0 &&
938 (reuseport_lb & inp_so_options(t)) == 0) {
940 if (ntohl(sin->sin_addr.s_addr) !=
942 ntohl(t->inp_laddr.s_addr) !=
944 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
945 (t->inp_vflag & INP_IPV6PROTO) == 0)
948 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
956 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
961 *laddrp = laddr.s_addr;
967 * Connect from a socket to a specified address.
968 * Both address and port must be specified in argument sin.
969 * If don't have a local address for this socket yet,
973 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
974 struct ucred *cred, struct mbuf *m, bool rehash)
976 u_short lport, fport;
977 in_addr_t laddr, faddr;
980 INP_WLOCK_ASSERT(inp);
981 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
983 lport = inp->inp_lport;
984 laddr = inp->inp_laddr.s_addr;
985 anonport = (lport == 0);
986 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
991 /* Do the initial binding of the local address if required. */
992 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
993 KASSERT(rehash == true,
994 ("Rehashing required for unbound inps"));
995 inp->inp_lport = lport;
996 inp->inp_laddr.s_addr = laddr;
997 if (in_pcbinshash(inp) != 0) {
998 inp->inp_laddr.s_addr = INADDR_ANY;
1004 /* Commit the remaining changes. */
1005 inp->inp_lport = lport;
1006 inp->inp_laddr.s_addr = laddr;
1007 inp->inp_faddr.s_addr = faddr;
1008 inp->inp_fport = fport;
1010 in_pcbrehash_mbuf(inp, m);
1012 in_pcbinshash_mbuf(inp, m);
1016 inp->inp_flags |= INP_ANONPORT;
1021 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1024 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1028 * Do proper source address selection on an unbound socket in case
1029 * of connect. Take jails into account as well.
1032 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1036 struct sockaddr *sa;
1037 struct sockaddr_in *sin;
1039 struct epoch_tracker et;
1042 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1044 * Bypass source address selection and use the primary jail IP
1047 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1051 bzero(&sro, sizeof(sro));
1053 sin = (struct sockaddr_in *)&sro.ro_dst;
1054 sin->sin_family = AF_INET;
1055 sin->sin_len = sizeof(struct sockaddr_in);
1056 sin->sin_addr.s_addr = faddr->s_addr;
1059 * If route is known our src addr is taken from the i/f,
1062 * Find out route to destination.
1064 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1065 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1068 * If we found a route, use the address corresponding to
1069 * the outgoing interface.
1071 * Otherwise assume faddr is reachable on a directly connected
1072 * network and try to find a corresponding interface to take
1073 * the source address from.
1075 NET_EPOCH_ENTER(et);
1076 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1077 struct in_ifaddr *ia;
1080 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1081 inp->inp_socket->so_fibnum));
1083 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1084 inp->inp_socket->so_fibnum));
1088 error = ENETUNREACH;
1092 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1093 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1099 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1102 if (sa->sa_family != AF_INET)
1104 sin = (struct sockaddr_in *)sa;
1105 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1106 ia = (struct in_ifaddr *)ifa;
1111 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1115 /* 3. As a last resort return the 'default' jail address. */
1116 error = prison_get_ip4(cred, laddr);
1121 * If the outgoing interface on the route found is not
1122 * a loopback interface, use the address from that interface.
1123 * In case of jails do those three steps:
1124 * 1. check if the interface address belongs to the jail. If so use it.
1125 * 2. check if we have any address on the outgoing interface
1126 * belonging to this jail. If so use it.
1127 * 3. as a last resort return the 'default' jail address.
1129 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1130 struct in_ifaddr *ia;
1133 /* If not jailed, use the default returned. */
1134 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1135 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1136 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1141 /* 1. Check if the iface address belongs to the jail. */
1142 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1143 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1144 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1145 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1150 * 2. Check if we have any address on the outgoing interface
1151 * belonging to this jail.
1154 ifp = sro.ro_rt->rt_ifp;
1155 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1157 if (sa->sa_family != AF_INET)
1159 sin = (struct sockaddr_in *)sa;
1160 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1161 ia = (struct in_ifaddr *)ifa;
1166 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1170 /* 3. As a last resort return the 'default' jail address. */
1171 error = prison_get_ip4(cred, laddr);
1176 * The outgoing interface is marked with 'loopback net', so a route
1177 * to ourselves is here.
1178 * Try to find the interface of the destination address and then
1179 * take the address from there. That interface is not necessarily
1180 * a loopback interface.
1181 * In case of jails, check that it is an address of the jail
1182 * and if we cannot find, fall back to the 'default' jail address.
1184 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1185 struct sockaddr_in sain;
1186 struct in_ifaddr *ia;
1188 bzero(&sain, sizeof(struct sockaddr_in));
1189 sain.sin_family = AF_INET;
1190 sain.sin_len = sizeof(struct sockaddr_in);
1191 sain.sin_addr.s_addr = faddr->s_addr;
1193 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1194 inp->inp_socket->so_fibnum));
1196 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1197 inp->inp_socket->so_fibnum));
1199 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1201 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1203 error = ENETUNREACH;
1206 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1216 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1218 if (sa->sa_family != AF_INET)
1220 sin = (struct sockaddr_in *)sa;
1221 if (prison_check_ip4(cred,
1222 &sin->sin_addr) == 0) {
1223 ia = (struct in_ifaddr *)ifa;
1228 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1233 /* 3. As a last resort return the 'default' jail address. */
1234 error = prison_get_ip4(cred, laddr);
1240 if (sro.ro_rt != NULL)
1246 * Set up for a connect from a socket to the specified address.
1247 * On entry, *laddrp and *lportp should contain the current local
1248 * address and port for the PCB; these are updated to the values
1249 * that should be placed in inp_laddr and inp_lport to complete
1252 * On success, *faddrp and *fportp will be set to the remote address
1253 * and port. These are not updated in the error case.
1255 * If the operation fails because the connection already exists,
1256 * *oinpp will be set to the PCB of that connection so that the
1257 * caller can decide to override it. In all other cases, *oinpp
1261 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1262 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1263 struct inpcb **oinpp, struct ucred *cred)
1265 struct rm_priotracker in_ifa_tracker;
1266 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1267 struct in_ifaddr *ia;
1269 struct in_addr laddr, faddr;
1270 u_short lport, fport;
1274 * Because a global state change doesn't actually occur here, a read
1275 * lock is sufficient.
1277 INP_LOCK_ASSERT(inp);
1278 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1282 if (nam->sa_len != sizeof (*sin))
1284 if (sin->sin_family != AF_INET)
1285 return (EAFNOSUPPORT);
1286 if (sin->sin_port == 0)
1287 return (EADDRNOTAVAIL);
1288 laddr.s_addr = *laddrp;
1290 faddr = sin->sin_addr;
1291 fport = sin->sin_port;
1293 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1295 * If the destination address is INADDR_ANY,
1296 * use the primary local address.
1297 * If the supplied address is INADDR_BROADCAST,
1298 * and the primary interface supports broadcast,
1299 * choose the broadcast address for that interface.
1301 if (faddr.s_addr == INADDR_ANY) {
1302 IN_IFADDR_RLOCK(&in_ifa_tracker);
1304 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1305 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1307 (error = prison_get_ip4(cred, &faddr)) != 0)
1309 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1310 IN_IFADDR_RLOCK(&in_ifa_tracker);
1311 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1313 faddr = satosin(&CK_STAILQ_FIRST(
1314 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1315 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1318 if (laddr.s_addr == INADDR_ANY) {
1319 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1321 * If the destination address is multicast and an outgoing
1322 * interface has been set as a multicast option, prefer the
1323 * address of that interface as our source address.
1325 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1326 inp->inp_moptions != NULL) {
1327 struct ip_moptions *imo;
1330 imo = inp->inp_moptions;
1331 if (imo->imo_multicast_ifp != NULL) {
1332 ifp = imo->imo_multicast_ifp;
1333 IN_IFADDR_RLOCK(&in_ifa_tracker);
1334 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1335 if ((ia->ia_ifp == ifp) &&
1337 prison_check_ip4(cred,
1338 &ia->ia_addr.sin_addr) == 0))
1342 error = EADDRNOTAVAIL;
1344 laddr = ia->ia_addr.sin_addr;
1347 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1353 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1354 laddr, lport, 0, NULL);
1358 return (EADDRINUSE);
1361 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1366 *laddrp = laddr.s_addr;
1368 *faddrp = faddr.s_addr;
1374 in_pcbdisconnect(struct inpcb *inp)
1377 INP_WLOCK_ASSERT(inp);
1378 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1380 inp->inp_faddr.s_addr = INADDR_ANY;
1387 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1388 * For most protocols, this will be invoked immediately prior to calling
1389 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1390 * socket, in which case in_pcbfree() is deferred.
1393 in_pcbdetach(struct inpcb *inp)
1396 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1399 if (inp->inp_snd_tag != NULL)
1400 in_pcbdetach_txrtlmt(inp);
1402 inp->inp_socket->so_pcb = NULL;
1403 inp->inp_socket = NULL;
1407 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1408 * stability of an inpcb pointer despite the inpcb lock being released. This
1409 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1410 * but where the inpcb lock may already held, or when acquiring a reference
1413 * in_pcbref() should be used only to provide brief memory stability, and
1414 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1415 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1416 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1417 * lock and rele are the *only* safe operations that may be performed on the
1420 * While the inpcb will not be freed, releasing the inpcb lock means that the
1421 * connection's state may change, so the caller should be careful to
1422 * revalidate any cached state on reacquiring the lock. Drop the reference
1423 * using in_pcbrele().
1426 in_pcbref(struct inpcb *inp)
1429 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1431 refcount_acquire(&inp->inp_refcount);
1435 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1436 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1437 * return a flag indicating whether or not the inpcb remains valid. If it is
1438 * valid, we return with the inpcb lock held.
1440 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1441 * reference on an inpcb. Historically more work was done here (actually, in
1442 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1443 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1444 * about memory stability (and continued use of the write lock).
1447 in_pcbrele_rlocked(struct inpcb *inp)
1449 struct inpcbinfo *pcbinfo;
1451 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1453 INP_RLOCK_ASSERT(inp);
1455 if (refcount_release(&inp->inp_refcount) == 0) {
1457 * If the inpcb has been freed, let the caller know, even if
1458 * this isn't the last reference.
1460 if (inp->inp_flags2 & INP_FREED) {
1467 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1469 if (inp->inp_in_hpts || inp->inp_in_input) {
1470 struct tcp_hpts_entry *hpts;
1472 * We should not be on the hpts at
1473 * this point in any form. we must
1474 * get the lock to be sure.
1476 hpts = tcp_hpts_lock(inp);
1477 if (inp->inp_in_hpts)
1478 panic("Hpts:%p inp:%p at free still on hpts",
1480 mtx_unlock(&hpts->p_mtx);
1481 hpts = tcp_input_lock(inp);
1482 if (inp->inp_in_input)
1483 panic("Hpts:%p inp:%p at free still on input hpts",
1485 mtx_unlock(&hpts->p_mtx);
1489 pcbinfo = inp->inp_pcbinfo;
1490 uma_zfree(pcbinfo->ipi_zone, inp);
1495 in_pcbrele_wlocked(struct inpcb *inp)
1497 struct inpcbinfo *pcbinfo;
1499 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1501 INP_WLOCK_ASSERT(inp);
1503 if (refcount_release(&inp->inp_refcount) == 0) {
1505 * If the inpcb has been freed, let the caller know, even if
1506 * this isn't the last reference.
1508 if (inp->inp_flags2 & INP_FREED) {
1515 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1517 if (inp->inp_in_hpts || inp->inp_in_input) {
1518 struct tcp_hpts_entry *hpts;
1520 * We should not be on the hpts at
1521 * this point in any form. we must
1522 * get the lock to be sure.
1524 hpts = tcp_hpts_lock(inp);
1525 if (inp->inp_in_hpts)
1526 panic("Hpts:%p inp:%p at free still on hpts",
1528 mtx_unlock(&hpts->p_mtx);
1529 hpts = tcp_input_lock(inp);
1530 if (inp->inp_in_input)
1531 panic("Hpts:%p inp:%p at free still on input hpts",
1533 mtx_unlock(&hpts->p_mtx);
1537 pcbinfo = inp->inp_pcbinfo;
1538 uma_zfree(pcbinfo->ipi_zone, inp);
1543 * Temporary wrapper.
1546 in_pcbrele(struct inpcb *inp)
1549 return (in_pcbrele_wlocked(inp));
1553 in_pcblist_rele_rlocked(epoch_context_t ctx)
1555 struct in_pcblist *il;
1557 struct inpcbinfo *pcbinfo;
1560 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1561 pcbinfo = il->il_pcbinfo;
1563 INP_INFO_WLOCK(pcbinfo);
1564 for (i = 0; i < n; i++) {
1565 inp = il->il_inp_list[i];
1567 if (!in_pcbrele_rlocked(inp))
1570 INP_INFO_WUNLOCK(pcbinfo);
1575 inpcbport_free(epoch_context_t ctx)
1577 struct inpcbport *phd;
1579 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1584 in_pcbfree_deferred(epoch_context_t ctx)
1587 int released __unused;
1589 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1592 CURVNET_SET(inp->inp_vnet);
1594 struct ip_moptions *imo = inp->inp_moptions;
1595 inp->inp_moptions = NULL;
1597 /* XXXRW: Do as much as possible here. */
1598 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1599 if (inp->inp_sp != NULL)
1600 ipsec_delete_pcbpolicy(inp);
1603 struct ip6_moptions *im6o = NULL;
1604 if (inp->inp_vflag & INP_IPV6PROTO) {
1605 ip6_freepcbopts(inp->in6p_outputopts);
1606 im6o = inp->in6p_moptions;
1607 inp->in6p_moptions = NULL;
1610 if (inp->inp_options)
1611 (void)m_free(inp->inp_options);
1613 crfree(inp->inp_cred);
1615 mac_inpcb_destroy(inp);
1617 released = in_pcbrele_wlocked(inp);
1620 ip6_freemoptions(im6o);
1623 inp_freemoptions(imo);
1629 * Unconditionally schedule an inpcb to be freed by decrementing its
1630 * reference count, which should occur only after the inpcb has been detached
1631 * from its socket. If another thread holds a temporary reference (acquired
1632 * using in_pcbref()) then the free is deferred until that reference is
1633 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1634 * work, including removal from global lists, is done in this context, where
1635 * the pcbinfo lock is held.
1638 in_pcbfree(struct inpcb *inp)
1640 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1642 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1643 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1644 ("%s: called twice for pcb %p", __func__, inp));
1645 if (inp->inp_flags2 & INP_FREED) {
1651 if (pcbinfo == &V_tcbinfo) {
1652 INP_INFO_LOCK_ASSERT(pcbinfo);
1654 INP_INFO_WLOCK_ASSERT(pcbinfo);
1657 INP_WLOCK_ASSERT(inp);
1658 INP_LIST_WLOCK(pcbinfo);
1659 in_pcbremlists(inp);
1660 INP_LIST_WUNLOCK(pcbinfo);
1661 RO_INVALIDATE_CACHE(&inp->inp_route);
1662 /* mark as destruction in progress */
1663 inp->inp_flags2 |= INP_FREED;
1665 NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx);
1669 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1670 * port reservation, and preventing it from being returned by inpcb lookups.
1672 * It is used by TCP to mark an inpcb as unused and avoid future packet
1673 * delivery or event notification when a socket remains open but TCP has
1674 * closed. This might occur as a result of a shutdown()-initiated TCP close
1675 * or a RST on the wire, and allows the port binding to be reused while still
1676 * maintaining the invariant that so_pcb always points to a valid inpcb until
1679 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1680 * in_pcbnotifyall() and in_pcbpurgeif0()?
1683 in_pcbdrop(struct inpcb *inp)
1686 INP_WLOCK_ASSERT(inp);
1688 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1689 MPASS(inp->inp_refcount > 1);
1693 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1696 inp->inp_flags |= INP_DROPPED;
1697 if (inp->inp_flags & INP_INHASHLIST) {
1698 struct inpcbport *phd = inp->inp_phd;
1700 INP_HASH_WLOCK(inp->inp_pcbinfo);
1701 in_pcbremlbgrouphash(inp);
1702 CK_LIST_REMOVE(inp, inp_hash);
1703 CK_LIST_REMOVE(inp, inp_portlist);
1704 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1705 CK_LIST_REMOVE(phd, phd_hash);
1706 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
1708 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1709 inp->inp_flags &= ~INP_INHASHLIST;
1711 in_pcbgroup_remove(inp);
1718 * Common routines to return the socket addresses associated with inpcbs.
1721 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1723 struct sockaddr_in *sin;
1725 sin = malloc(sizeof *sin, M_SONAME,
1727 sin->sin_family = AF_INET;
1728 sin->sin_len = sizeof(*sin);
1729 sin->sin_addr = *addr_p;
1730 sin->sin_port = port;
1732 return (struct sockaddr *)sin;
1736 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1739 struct in_addr addr;
1742 inp = sotoinpcb(so);
1743 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1746 port = inp->inp_lport;
1747 addr = inp->inp_laddr;
1750 *nam = in_sockaddr(port, &addr);
1755 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1758 struct in_addr addr;
1761 inp = sotoinpcb(so);
1762 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1765 port = inp->inp_fport;
1766 addr = inp->inp_faddr;
1769 *nam = in_sockaddr(port, &addr);
1774 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1775 struct inpcb *(*notify)(struct inpcb *, int))
1777 struct inpcb *inp, *inp_temp;
1779 INP_INFO_WLOCK(pcbinfo);
1780 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1783 if ((inp->inp_vflag & INP_IPV4) == 0) {
1788 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1789 inp->inp_socket == NULL) {
1793 if ((*notify)(inp, errno))
1796 INP_INFO_WUNLOCK(pcbinfo);
1800 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1803 struct in_multi *inm;
1804 struct in_mfilter *imf;
1805 struct ip_moptions *imo;
1807 INP_INFO_WLOCK(pcbinfo);
1808 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1810 imo = inp->inp_moptions;
1811 if ((inp->inp_vflag & INP_IPV4) &&
1814 * Unselect the outgoing interface if it is being
1817 if (imo->imo_multicast_ifp == ifp)
1818 imo->imo_multicast_ifp = NULL;
1821 * Drop multicast group membership if we joined
1822 * through the interface being detached.
1824 * XXX This can all be deferred to an epoch_call
1827 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1828 if ((inm = imf->imf_inm) == NULL)
1830 if (inm->inm_ifp != ifp)
1832 ip_mfilter_remove(&imo->imo_head, imf);
1833 IN_MULTI_LOCK_ASSERT();
1834 in_leavegroup_locked(inm, NULL);
1835 ip_mfilter_free(imf);
1841 INP_INFO_WUNLOCK(pcbinfo);
1845 * Lookup a PCB based on the local address and port. Caller must hold the
1846 * hash lock. No inpcb locks or references are acquired.
1848 #define INP_LOOKUP_MAPPED_PCB_COST 3
1850 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1851 u_short lport, int lookupflags, struct ucred *cred)
1855 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1861 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1862 ("%s: invalid lookup flags %d", __func__, lookupflags));
1864 INP_HASH_LOCK_ASSERT(pcbinfo);
1866 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1867 struct inpcbhead *head;
1869 * Look for an unconnected (wildcard foreign addr) PCB that
1870 * matches the local address and port we're looking for.
1872 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1873 0, pcbinfo->ipi_hashmask)];
1874 CK_LIST_FOREACH(inp, head, inp_hash) {
1876 /* XXX inp locking */
1877 if ((inp->inp_vflag & INP_IPV4) == 0)
1880 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1881 inp->inp_laddr.s_addr == laddr.s_addr &&
1882 inp->inp_lport == lport) {
1887 prison_equal_ip4(cred->cr_prison,
1888 inp->inp_cred->cr_prison))
1897 struct inpcbporthead *porthash;
1898 struct inpcbport *phd;
1899 struct inpcb *match = NULL;
1901 * Best fit PCB lookup.
1903 * First see if this local port is in use by looking on the
1906 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1907 pcbinfo->ipi_porthashmask)];
1908 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1909 if (phd->phd_port == lport)
1914 * Port is in use by one or more PCBs. Look for best
1917 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1920 !prison_equal_ip4(inp->inp_cred->cr_prison,
1924 /* XXX inp locking */
1925 if ((inp->inp_vflag & INP_IPV4) == 0)
1928 * We never select the PCB that has
1929 * INP_IPV6 flag and is bound to :: if
1930 * we have another PCB which is bound
1931 * to 0.0.0.0. If a PCB has the
1932 * INP_IPV6 flag, then we set its cost
1933 * higher than IPv4 only PCBs.
1935 * Note that the case only happens
1936 * when a socket is bound to ::, under
1937 * the condition that the use of the
1938 * mapped address is allowed.
1940 if ((inp->inp_vflag & INP_IPV6) != 0)
1941 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1943 if (inp->inp_faddr.s_addr != INADDR_ANY)
1945 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1946 if (laddr.s_addr == INADDR_ANY)
1948 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1951 if (laddr.s_addr != INADDR_ANY)
1954 if (wildcard < matchwild) {
1956 matchwild = wildcard;
1965 #undef INP_LOOKUP_MAPPED_PCB_COST
1967 static struct inpcb *
1968 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1969 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1970 uint16_t fport, int lookupflags)
1972 struct inpcb *local_wild;
1973 const struct inpcblbgrouphead *hdr;
1974 struct inpcblbgroup *grp;
1977 INP_HASH_LOCK_ASSERT(pcbinfo);
1979 hdr = &pcbinfo->ipi_lbgrouphashbase[
1980 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1983 * Order of socket selection:
1985 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1988 * - Load balanced group does not contain jailed sockets
1989 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1992 CK_LIST_FOREACH(grp, hdr, il_list) {
1994 if (!(grp->il_vflag & INP_IPV4))
1997 if (grp->il_lport != lport)
2000 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
2002 if (grp->il_laddr.s_addr == laddr->s_addr)
2003 return (grp->il_inp[idx]);
2004 if (grp->il_laddr.s_addr == INADDR_ANY &&
2005 (lookupflags & INPLOOKUP_WILDCARD) != 0)
2006 local_wild = grp->il_inp[idx];
2008 return (local_wild);
2013 * Lookup PCB in hash list, using pcbgroup tables.
2015 static struct inpcb *
2016 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2017 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2018 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2020 struct inpcbhead *head;
2021 struct inpcb *inp, *tmpinp;
2022 u_short fport = fport_arg, lport = lport_arg;
2026 * First look for an exact match.
2029 INP_GROUP_LOCK(pcbgroup);
2030 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2031 pcbgroup->ipg_hashmask)];
2032 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2034 /* XXX inp locking */
2035 if ((inp->inp_vflag & INP_IPV4) == 0)
2038 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2039 inp->inp_laddr.s_addr == laddr.s_addr &&
2040 inp->inp_fport == fport &&
2041 inp->inp_lport == lport) {
2043 * XXX We should be able to directly return
2044 * the inp here, without any checks.
2045 * Well unless both bound with SO_REUSEPORT?
2047 if (prison_flag(inp->inp_cred, PR_IP4))
2053 if (tmpinp != NULL) {
2060 * For incoming connections, we may wish to do a wildcard
2061 * match for an RSS-local socket.
2063 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2064 struct inpcb *local_wild = NULL, *local_exact = NULL;
2066 struct inpcb *local_wild_mapped = NULL;
2068 struct inpcb *jail_wild = NULL;
2069 struct inpcbhead *head;
2073 * Order of socket selection - we always prefer jails.
2074 * 1. jailed, non-wild.
2076 * 3. non-jailed, non-wild.
2077 * 4. non-jailed, wild.
2080 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2081 lport, 0, pcbgroup->ipg_hashmask)];
2082 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2084 /* XXX inp locking */
2085 if ((inp->inp_vflag & INP_IPV4) == 0)
2088 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2089 inp->inp_lport != lport)
2092 injail = prison_flag(inp->inp_cred, PR_IP4);
2094 if (prison_check_ip4(inp->inp_cred,
2098 if (local_exact != NULL)
2102 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2107 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2109 /* XXX inp locking, NULL check */
2110 if (inp->inp_vflag & INP_IPV6PROTO)
2111 local_wild_mapped = inp;
2119 } /* LIST_FOREACH */
2128 inp = local_wild_mapped;
2136 * Then look for a wildcard match, if requested.
2138 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2139 struct inpcb *local_wild = NULL, *local_exact = NULL;
2141 struct inpcb *local_wild_mapped = NULL;
2143 struct inpcb *jail_wild = NULL;
2144 struct inpcbhead *head;
2148 * Order of socket selection - we always prefer jails.
2149 * 1. jailed, non-wild.
2151 * 3. non-jailed, non-wild.
2152 * 4. non-jailed, wild.
2154 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2155 0, pcbinfo->ipi_wildmask)];
2156 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2158 /* XXX inp locking */
2159 if ((inp->inp_vflag & INP_IPV4) == 0)
2162 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2163 inp->inp_lport != lport)
2166 injail = prison_flag(inp->inp_cred, PR_IP4);
2168 if (prison_check_ip4(inp->inp_cred,
2172 if (local_exact != NULL)
2176 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2181 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2183 /* XXX inp locking, NULL check */
2184 if (inp->inp_vflag & INP_IPV6PROTO)
2185 local_wild_mapped = inp;
2193 } /* LIST_FOREACH */
2201 inp = local_wild_mapped;
2205 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2206 INP_GROUP_UNLOCK(pcbgroup);
2210 if (lookupflags & INPLOOKUP_WLOCKPCB)
2211 locked = INP_TRY_WLOCK(inp);
2212 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2213 locked = INP_TRY_RLOCK(inp);
2215 panic("%s: locking bug", __func__);
2216 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2217 if (lookupflags & INPLOOKUP_WLOCKPCB)
2224 INP_GROUP_UNLOCK(pcbgroup);
2226 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2228 if (in_pcbrele_wlocked(inp))
2232 if (in_pcbrele_rlocked(inp))
2237 if (lookupflags & INPLOOKUP_WLOCKPCB)
2238 INP_WLOCK_ASSERT(inp);
2240 INP_RLOCK_ASSERT(inp);
2244 #endif /* PCBGROUP */
2247 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2248 * that the caller has locked the hash list, and will not perform any further
2249 * locking or reference operations on either the hash list or the connection.
2251 static struct inpcb *
2252 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2253 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2256 struct inpcbhead *head;
2257 struct inpcb *inp, *tmpinp;
2258 u_short fport = fport_arg, lport = lport_arg;
2260 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2261 ("%s: invalid lookup flags %d", __func__, lookupflags));
2262 INP_HASH_LOCK_ASSERT(pcbinfo);
2265 * First look for an exact match.
2268 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2269 pcbinfo->ipi_hashmask)];
2270 CK_LIST_FOREACH(inp, head, inp_hash) {
2272 /* XXX inp locking */
2273 if ((inp->inp_vflag & INP_IPV4) == 0)
2276 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2277 inp->inp_laddr.s_addr == laddr.s_addr &&
2278 inp->inp_fport == fport &&
2279 inp->inp_lport == lport) {
2281 * XXX We should be able to directly return
2282 * the inp here, without any checks.
2283 * Well unless both bound with SO_REUSEPORT?
2285 if (prison_flag(inp->inp_cred, PR_IP4))
2295 * Then look in lb group (for wildcard match).
2297 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2298 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2299 fport, lookupflags);
2305 * Then look for a wildcard match, if requested.
2307 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2308 struct inpcb *local_wild = NULL, *local_exact = NULL;
2310 struct inpcb *local_wild_mapped = NULL;
2312 struct inpcb *jail_wild = NULL;
2316 * Order of socket selection - we always prefer jails.
2317 * 1. jailed, non-wild.
2319 * 3. non-jailed, non-wild.
2320 * 4. non-jailed, wild.
2323 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2324 0, pcbinfo->ipi_hashmask)];
2325 CK_LIST_FOREACH(inp, head, inp_hash) {
2327 /* XXX inp locking */
2328 if ((inp->inp_vflag & INP_IPV4) == 0)
2331 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2332 inp->inp_lport != lport)
2335 injail = prison_flag(inp->inp_cred, PR_IP4);
2337 if (prison_check_ip4(inp->inp_cred,
2341 if (local_exact != NULL)
2345 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2350 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2352 /* XXX inp locking, NULL check */
2353 if (inp->inp_vflag & INP_IPV6PROTO)
2354 local_wild_mapped = inp;
2362 } /* LIST_FOREACH */
2363 if (jail_wild != NULL)
2365 if (local_exact != NULL)
2366 return (local_exact);
2367 if (local_wild != NULL)
2368 return (local_wild);
2370 if (local_wild_mapped != NULL)
2371 return (local_wild_mapped);
2373 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2379 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2380 * hash list lock, and will return the inpcb locked (i.e., requires
2381 * INPLOOKUP_LOCKPCB).
2383 static struct inpcb *
2384 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2385 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2390 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2391 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2393 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2395 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2399 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2401 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2406 panic("%s: locking bug", __func__);
2409 if (lookupflags & INPLOOKUP_WLOCKPCB)
2410 INP_WLOCK_ASSERT(inp);
2412 INP_RLOCK_ASSERT(inp);
2421 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2422 * from which a pre-calculated hash value may be extracted.
2424 * Possibly more of this logic should be in in_pcbgroup.c.
2427 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2428 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2430 #if defined(PCBGROUP) && !defined(RSS)
2431 struct inpcbgroup *pcbgroup;
2434 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2435 ("%s: invalid lookup flags %d", __func__, lookupflags));
2436 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2437 ("%s: LOCKPCB not set", __func__));
2440 * When not using RSS, use connection groups in preference to the
2441 * reservation table when looking up 4-tuples. When using RSS, just
2442 * use the reservation table, due to the cost of the Toeplitz hash
2445 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2446 * we could be doing RSS with a non-Toeplitz hash that is affordable
2449 #if defined(PCBGROUP) && !defined(RSS)
2450 if (in_pcbgroup_enabled(pcbinfo)) {
2451 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2453 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2454 laddr, lport, lookupflags, ifp));
2457 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2462 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2463 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2464 struct ifnet *ifp, struct mbuf *m)
2467 struct inpcbgroup *pcbgroup;
2470 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2471 ("%s: invalid lookup flags %d", __func__, lookupflags));
2472 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2473 ("%s: LOCKPCB not set", __func__));
2477 * If we can use a hardware-generated hash to look up the connection
2478 * group, use that connection group to find the inpcb. Otherwise
2479 * fall back on a software hash -- or the reservation table if we're
2482 * XXXRW: As above, that policy belongs in the pcbgroup code.
2484 if (in_pcbgroup_enabled(pcbinfo) &&
2485 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2486 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2487 m->m_pkthdr.flowid);
2488 if (pcbgroup != NULL)
2489 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2490 fport, laddr, lport, lookupflags, ifp));
2492 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2494 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2495 laddr, lport, lookupflags, ifp));
2499 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2505 * Insert PCB onto various hash lists.
2508 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2510 struct inpcbhead *pcbhash;
2511 struct inpcbporthead *pcbporthash;
2512 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2513 struct inpcbport *phd;
2514 u_int32_t hashkey_faddr;
2517 INP_WLOCK_ASSERT(inp);
2518 INP_HASH_WLOCK_ASSERT(pcbinfo);
2520 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2521 ("in_pcbinshash: INP_INHASHLIST"));
2524 if (inp->inp_vflag & INP_IPV6)
2525 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2528 hashkey_faddr = inp->inp_faddr.s_addr;
2530 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2531 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2533 pcbporthash = &pcbinfo->ipi_porthashbase[
2534 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2537 * Add entry to load balance group.
2538 * Only do this if SO_REUSEPORT_LB is set.
2540 so_options = inp_so_options(inp);
2541 if (so_options & SO_REUSEPORT_LB) {
2542 int ret = in_pcbinslbgrouphash(inp);
2544 /* pcb lb group malloc fail (ret=ENOBUFS). */
2550 * Go through port list and look for a head for this lport.
2552 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2553 if (phd->phd_port == inp->inp_lport)
2557 * If none exists, malloc one and tack it on.
2560 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2562 return (ENOBUFS); /* XXX */
2564 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2565 phd->phd_port = inp->inp_lport;
2566 CK_LIST_INIT(&phd->phd_pcblist);
2567 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2570 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2571 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2572 inp->inp_flags |= INP_INHASHLIST;
2575 in_pcbgroup_update_mbuf(inp, m);
2577 in_pcbgroup_update(inp);
2584 in_pcbinshash(struct inpcb *inp)
2587 return (in_pcbinshash_internal(inp, NULL));
2591 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2594 return (in_pcbinshash_internal(inp, m));
2598 * Move PCB to the proper hash bucket when { faddr, fport } have been
2599 * changed. NOTE: This does not handle the case of the lport changing (the
2600 * hashed port list would have to be updated as well), so the lport must
2601 * not change after in_pcbinshash() has been called.
2604 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2606 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2607 struct inpcbhead *head;
2608 u_int32_t hashkey_faddr;
2610 INP_WLOCK_ASSERT(inp);
2611 INP_HASH_WLOCK_ASSERT(pcbinfo);
2613 KASSERT(inp->inp_flags & INP_INHASHLIST,
2614 ("in_pcbrehash: !INP_INHASHLIST"));
2617 if (inp->inp_vflag & INP_IPV6)
2618 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2621 hashkey_faddr = inp->inp_faddr.s_addr;
2623 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2624 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2626 CK_LIST_REMOVE(inp, inp_hash);
2627 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2631 in_pcbgroup_update_mbuf(inp, m);
2633 in_pcbgroup_update(inp);
2638 in_pcbrehash(struct inpcb *inp)
2641 in_pcbrehash_mbuf(inp, NULL);
2645 * Remove PCB from various lists.
2648 in_pcbremlists(struct inpcb *inp)
2650 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2653 if (pcbinfo == &V_tcbinfo) {
2656 INP_INFO_WLOCK_ASSERT(pcbinfo);
2660 INP_WLOCK_ASSERT(inp);
2661 INP_LIST_WLOCK_ASSERT(pcbinfo);
2663 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2664 if (inp->inp_flags & INP_INHASHLIST) {
2665 struct inpcbport *phd = inp->inp_phd;
2667 INP_HASH_WLOCK(pcbinfo);
2669 /* XXX: Only do if SO_REUSEPORT_LB set? */
2670 in_pcbremlbgrouphash(inp);
2672 CK_LIST_REMOVE(inp, inp_hash);
2673 CK_LIST_REMOVE(inp, inp_portlist);
2674 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2675 CK_LIST_REMOVE(phd, phd_hash);
2676 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
2678 INP_HASH_WUNLOCK(pcbinfo);
2679 inp->inp_flags &= ~INP_INHASHLIST;
2681 CK_LIST_REMOVE(inp, inp_list);
2682 pcbinfo->ipi_count--;
2684 in_pcbgroup_remove(inp);
2689 * Check for alternatives when higher level complains
2690 * about service problems. For now, invalidate cached
2691 * routing information. If the route was created dynamically
2692 * (by a redirect), time to try a default gateway again.
2695 in_losing(struct inpcb *inp)
2698 RO_INVALIDATE_CACHE(&inp->inp_route);
2703 * A set label operation has occurred at the socket layer, propagate the
2704 * label change into the in_pcb for the socket.
2707 in_pcbsosetlabel(struct socket *so)
2712 inp = sotoinpcb(so);
2713 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2717 mac_inpcb_sosetlabel(so, inp);
2724 * ipport_tick runs once per second, determining if random port allocation
2725 * should be continued. If more than ipport_randomcps ports have been
2726 * allocated in the last second, then we return to sequential port
2727 * allocation. We return to random allocation only once we drop below
2728 * ipport_randomcps for at least ipport_randomtime seconds.
2731 ipport_tick(void *xtp)
2733 VNET_ITERATOR_DECL(vnet_iter);
2735 VNET_LIST_RLOCK_NOSLEEP();
2736 VNET_FOREACH(vnet_iter) {
2737 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2738 if (V_ipport_tcpallocs <=
2739 V_ipport_tcplastcount + V_ipport_randomcps) {
2740 if (V_ipport_stoprandom > 0)
2741 V_ipport_stoprandom--;
2743 V_ipport_stoprandom = V_ipport_randomtime;
2744 V_ipport_tcplastcount = V_ipport_tcpallocs;
2747 VNET_LIST_RUNLOCK_NOSLEEP();
2748 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2755 callout_stop(&ipport_tick_callout);
2759 * The ipport_callout should start running at about the time we attach the
2760 * inet or inet6 domains.
2763 ipport_tick_init(const void *unused __unused)
2766 /* Start ipport_tick. */
2767 callout_init(&ipport_tick_callout, 1);
2768 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2769 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2770 SHUTDOWN_PRI_DEFAULT);
2772 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2773 ipport_tick_init, NULL);
2776 inp_wlock(struct inpcb *inp)
2783 inp_wunlock(struct inpcb *inp)
2790 inp_rlock(struct inpcb *inp)
2797 inp_runlock(struct inpcb *inp)
2803 #ifdef INVARIANT_SUPPORT
2805 inp_lock_assert(struct inpcb *inp)
2808 INP_WLOCK_ASSERT(inp);
2812 inp_unlock_assert(struct inpcb *inp)
2815 INP_UNLOCK_ASSERT(inp);
2820 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2824 INP_INFO_WLOCK(&V_tcbinfo);
2825 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2830 INP_INFO_WUNLOCK(&V_tcbinfo);
2834 inp_inpcbtosocket(struct inpcb *inp)
2837 INP_WLOCK_ASSERT(inp);
2838 return (inp->inp_socket);
2842 inp_inpcbtotcpcb(struct inpcb *inp)
2845 INP_WLOCK_ASSERT(inp);
2846 return ((struct tcpcb *)inp->inp_ppcb);
2850 inp_ip_tos_get(const struct inpcb *inp)
2853 return (inp->inp_ip_tos);
2857 inp_ip_tos_set(struct inpcb *inp, int val)
2860 inp->inp_ip_tos = val;
2864 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2865 uint32_t *faddr, uint16_t *fp)
2868 INP_LOCK_ASSERT(inp);
2869 *laddr = inp->inp_laddr.s_addr;
2870 *faddr = inp->inp_faddr.s_addr;
2871 *lp = inp->inp_lport;
2872 *fp = inp->inp_fport;
2876 so_sotoinpcb(struct socket *so)
2879 return (sotoinpcb(so));
2883 so_sototcpcb(struct socket *so)
2886 return (sototcpcb(so));
2890 * Create an external-format (``xinpcb'') structure using the information in
2891 * the kernel-format in_pcb structure pointed to by inp. This is done to
2892 * reduce the spew of irrelevant information over this interface, to isolate
2893 * user code from changes in the kernel structure, and potentially to provide
2894 * information-hiding if we decide that some of this information should be
2895 * hidden from users.
2898 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2901 bzero(xi, sizeof(*xi));
2902 xi->xi_len = sizeof(struct xinpcb);
2903 if (inp->inp_socket)
2904 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2905 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2906 xi->inp_gencnt = inp->inp_gencnt;
2907 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2908 xi->inp_flow = inp->inp_flow;
2909 xi->inp_flowid = inp->inp_flowid;
2910 xi->inp_flowtype = inp->inp_flowtype;
2911 xi->inp_flags = inp->inp_flags;
2912 xi->inp_flags2 = inp->inp_flags2;
2913 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2914 xi->in6p_cksum = inp->in6p_cksum;
2915 xi->in6p_hops = inp->in6p_hops;
2916 xi->inp_ip_tos = inp->inp_ip_tos;
2917 xi->inp_vflag = inp->inp_vflag;
2918 xi->inp_ip_ttl = inp->inp_ip_ttl;
2919 xi->inp_ip_p = inp->inp_ip_p;
2920 xi->inp_ip_minttl = inp->inp_ip_minttl;
2925 db_print_indent(int indent)
2929 for (i = 0; i < indent; i++)
2934 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2936 char faddr_str[48], laddr_str[48];
2938 db_print_indent(indent);
2939 db_printf("%s at %p\n", name, inc);
2944 if (inc->inc_flags & INC_ISIPV6) {
2946 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2947 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2952 inet_ntoa_r(inc->inc_laddr, laddr_str);
2953 inet_ntoa_r(inc->inc_faddr, faddr_str);
2955 db_print_indent(indent);
2956 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2957 ntohs(inc->inc_lport));
2958 db_print_indent(indent);
2959 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2960 ntohs(inc->inc_fport));
2964 db_print_inpflags(int inp_flags)
2969 if (inp_flags & INP_RECVOPTS) {
2970 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2973 if (inp_flags & INP_RECVRETOPTS) {
2974 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2977 if (inp_flags & INP_RECVDSTADDR) {
2978 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2981 if (inp_flags & INP_ORIGDSTADDR) {
2982 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2985 if (inp_flags & INP_HDRINCL) {
2986 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2989 if (inp_flags & INP_HIGHPORT) {
2990 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2993 if (inp_flags & INP_LOWPORT) {
2994 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2997 if (inp_flags & INP_ANONPORT) {
2998 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3001 if (inp_flags & INP_RECVIF) {
3002 db_printf("%sINP_RECVIF", comma ? ", " : "");
3005 if (inp_flags & INP_MTUDISC) {
3006 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3009 if (inp_flags & INP_RECVTTL) {
3010 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3013 if (inp_flags & INP_DONTFRAG) {
3014 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3017 if (inp_flags & INP_RECVTOS) {
3018 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3021 if (inp_flags & IN6P_IPV6_V6ONLY) {
3022 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3025 if (inp_flags & IN6P_PKTINFO) {
3026 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3029 if (inp_flags & IN6P_HOPLIMIT) {
3030 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3033 if (inp_flags & IN6P_HOPOPTS) {
3034 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3037 if (inp_flags & IN6P_DSTOPTS) {
3038 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3041 if (inp_flags & IN6P_RTHDR) {
3042 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3045 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3046 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3049 if (inp_flags & IN6P_TCLASS) {
3050 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3053 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3054 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3057 if (inp_flags & INP_TIMEWAIT) {
3058 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3061 if (inp_flags & INP_ONESBCAST) {
3062 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3065 if (inp_flags & INP_DROPPED) {
3066 db_printf("%sINP_DROPPED", comma ? ", " : "");
3069 if (inp_flags & INP_SOCKREF) {
3070 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3073 if (inp_flags & IN6P_RFC2292) {
3074 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3077 if (inp_flags & IN6P_MTU) {
3078 db_printf("IN6P_MTU%s", comma ? ", " : "");
3084 db_print_inpvflag(u_char inp_vflag)
3089 if (inp_vflag & INP_IPV4) {
3090 db_printf("%sINP_IPV4", comma ? ", " : "");
3093 if (inp_vflag & INP_IPV6) {
3094 db_printf("%sINP_IPV6", comma ? ", " : "");
3097 if (inp_vflag & INP_IPV6PROTO) {
3098 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3104 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3107 db_print_indent(indent);
3108 db_printf("%s at %p\n", name, inp);
3112 db_print_indent(indent);
3113 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3115 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3117 db_print_indent(indent);
3118 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3119 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3121 db_print_indent(indent);
3122 db_printf("inp_label: %p inp_flags: 0x%x (",
3123 inp->inp_label, inp->inp_flags);
3124 db_print_inpflags(inp->inp_flags);
3127 db_print_indent(indent);
3128 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3130 db_print_inpvflag(inp->inp_vflag);
3133 db_print_indent(indent);
3134 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3135 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3137 db_print_indent(indent);
3139 if (inp->inp_vflag & INP_IPV6) {
3140 db_printf("in6p_options: %p in6p_outputopts: %p "
3141 "in6p_moptions: %p\n", inp->in6p_options,
3142 inp->in6p_outputopts, inp->in6p_moptions);
3143 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3144 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3149 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3150 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3151 inp->inp_options, inp->inp_moptions);
3154 db_print_indent(indent);
3155 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3156 (uintmax_t)inp->inp_gencnt);
3159 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3164 db_printf("usage: show inpcb <addr>\n");
3167 inp = (struct inpcb *)addr;
3169 db_print_inpcb(inp, "inpcb", 0);
3175 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3179 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3181 union if_snd_tag_modify_params params = {
3182 .rate_limit.max_rate = max_pacing_rate,
3183 .rate_limit.flags = M_NOWAIT,
3185 struct m_snd_tag *mst;
3189 mst = inp->inp_snd_tag;
3197 if (ifp->if_snd_tag_modify == NULL) {
3200 error = ifp->if_snd_tag_modify(mst, ¶ms);
3206 * Query existing TX rate limit based on the existing
3207 * "inp->inp_snd_tag", if any.
3210 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3212 union if_snd_tag_query_params params = { };
3213 struct m_snd_tag *mst;
3217 mst = inp->inp_snd_tag;
3225 if (ifp->if_snd_tag_query == NULL) {
3228 error = ifp->if_snd_tag_query(mst, ¶ms);
3229 if (error == 0 && p_max_pacing_rate != NULL)
3230 *p_max_pacing_rate = params.rate_limit.max_rate;
3236 * Query existing TX queue level based on the existing
3237 * "inp->inp_snd_tag", if any.
3240 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3242 union if_snd_tag_query_params params = { };
3243 struct m_snd_tag *mst;
3247 mst = inp->inp_snd_tag;
3255 if (ifp->if_snd_tag_query == NULL)
3256 return (EOPNOTSUPP);
3258 error = ifp->if_snd_tag_query(mst, ¶ms);
3259 if (error == 0 && p_txqueue_level != NULL)
3260 *p_txqueue_level = params.rate_limit.queue_level;
3265 * Allocate a new TX rate limit send tag from the network interface
3266 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3269 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3270 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3273 union if_snd_tag_alloc_params params = {
3274 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3275 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3276 .rate_limit.hdr.flowid = flowid,
3277 .rate_limit.hdr.flowtype = flowtype,
3278 .rate_limit.max_rate = max_pacing_rate,
3279 .rate_limit.flags = M_NOWAIT,
3283 INP_WLOCK_ASSERT(inp);
3288 if (ifp->if_snd_tag_alloc == NULL) {
3291 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3295 counter_u64_add(rate_limit_set_ok, 1);
3296 counter_u64_add(rate_limit_active, 1);
3298 counter_u64_add(rate_limit_alloc_fail, 1);
3305 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3311 * If the device was detached while we still had reference(s)
3312 * on the ifp, we assume if_snd_tag_free() was replaced with
3315 ifp->if_snd_tag_free(mst);
3317 /* release reference count on network interface */
3320 counter_u64_add(rate_limit_active, -1);
3325 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3329 in_pcbdetach_txrtlmt(struct inpcb *inp)
3331 struct m_snd_tag *mst;
3333 INP_WLOCK_ASSERT(inp);
3335 mst = inp->inp_snd_tag;
3336 inp->inp_snd_tag = NULL;
3341 m_snd_tag_rele(mst);
3345 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3350 * If the existing send tag is for the wrong interface due to
3351 * a route change, first drop the existing tag. Set the
3352 * CHANGED flag so that we will keep trying to allocate a new
3353 * tag if we fail to allocate one this time.
3355 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3356 in_pcbdetach_txrtlmt(inp);
3357 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3361 * NOTE: When attaching to a network interface a reference is
3362 * made to ensure the network interface doesn't go away until
3363 * all ratelimit connections are gone. The network interface
3364 * pointers compared below represent valid network interfaces,
3365 * except when comparing towards NULL.
3367 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3369 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3370 if (inp->inp_snd_tag != NULL)
3371 in_pcbdetach_txrtlmt(inp);
3373 } else if (inp->inp_snd_tag == NULL) {
3375 * In order to utilize packet pacing with RSS, we need
3376 * to wait until there is a valid RSS hash before we
3379 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3382 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3383 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3386 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3388 if (error == 0 || error == EOPNOTSUPP)
3389 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3395 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3396 * is set in the fast path and will attach/detach/modify the TX rate
3397 * limit send tag based on the socket's so_max_pacing_rate value.
3400 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3402 struct socket *socket;
3403 uint32_t max_pacing_rate;
3410 socket = inp->inp_socket;
3414 if (!INP_WLOCKED(inp)) {
3416 * NOTE: If the write locking fails, we need to bail
3417 * out and use the non-ratelimited ring for the
3418 * transmit until there is a new chance to get the
3421 if (!INP_TRY_UPGRADE(inp))
3429 * NOTE: The so_max_pacing_rate value is read unlocked,
3430 * because atomic updates are not required since the variable
3431 * is checked at every mbuf we send. It is assumed that the
3432 * variable read itself will be atomic.
3434 max_pacing_rate = socket->so_max_pacing_rate;
3436 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3443 * Track route changes for TX rate limiting.
3446 in_pcboutput_eagain(struct inpcb *inp)
3453 if (inp->inp_snd_tag == NULL)
3456 if (!INP_WLOCKED(inp)) {
3458 * NOTE: If the write locking fails, we need to bail
3459 * out and use the non-ratelimited ring for the
3460 * transmit until there is a new chance to get the
3463 if (!INP_TRY_UPGRADE(inp))
3470 /* detach rate limiting */
3471 in_pcbdetach_txrtlmt(inp);
3473 /* make sure new mbuf send tag allocation is made */
3474 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3484 rate_limit_active = counter_u64_alloc(M_WAITOK);
3485 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3486 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3489 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3491 #endif /* RATELIMIT */