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");
216 * in_pcb.c: manage the Protocol Control Blocks.
218 * NOTE: It is assumed that most of these functions will be called with
219 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
220 * functions often modify hash chains or addresses in pcbs.
223 static struct inpcblbgroup *
224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
225 uint16_t port, const union in_dependaddr *addr, int size)
227 struct inpcblbgroup *grp;
230 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
231 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
234 grp->il_vflag = vflag;
235 grp->il_lport = port;
236 grp->il_dependladdr = *addr;
237 grp->il_inpsiz = size;
238 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
243 in_pcblbgroup_free_deferred(epoch_context_t ctx)
245 struct inpcblbgroup *grp;
247 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
252 in_pcblbgroup_free(struct inpcblbgroup *grp)
255 CK_LIST_REMOVE(grp, il_list);
256 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
257 in_pcblbgroup_free_deferred);
260 static struct inpcblbgroup *
261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
262 struct inpcblbgroup *old_grp, int size)
264 struct inpcblbgroup *grp;
267 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
268 old_grp->il_lport, &old_grp->il_dependladdr, size);
272 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
273 ("invalid new local group size %d and old local group count %d",
274 grp->il_inpsiz, old_grp->il_inpcnt));
276 for (i = 0; i < old_grp->il_inpcnt; ++i)
277 grp->il_inp[i] = old_grp->il_inp[i];
278 grp->il_inpcnt = old_grp->il_inpcnt;
279 in_pcblbgroup_free(old_grp);
284 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
285 * and shrink group if possible.
288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
291 struct inpcblbgroup *grp, *new_grp;
294 for (; i + 1 < grp->il_inpcnt; ++i)
295 grp->il_inp[i] = grp->il_inp[i + 1];
298 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
299 grp->il_inpcnt <= grp->il_inpsiz / 4) {
300 /* Shrink this group. */
301 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
308 * Add PCB to load balance group for SO_REUSEPORT_LB option.
311 in_pcbinslbgrouphash(struct inpcb *inp)
313 const static struct timeval interval = { 60, 0 };
314 static struct timeval lastprint;
315 struct inpcbinfo *pcbinfo;
316 struct inpcblbgrouphead *hdr;
317 struct inpcblbgroup *grp;
320 pcbinfo = inp->inp_pcbinfo;
322 INP_WLOCK_ASSERT(inp);
323 INP_HASH_WLOCK_ASSERT(pcbinfo);
326 * Don't allow jailed socket to join local group.
328 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
333 * Don't allow IPv4 mapped INET6 wild socket.
335 if ((inp->inp_vflag & INP_IPV4) &&
336 inp->inp_laddr.s_addr == INADDR_ANY &&
337 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
342 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
343 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
344 CK_LIST_FOREACH(grp, hdr, il_list) {
345 if (grp->il_vflag == inp->inp_vflag &&
346 grp->il_lport == inp->inp_lport &&
347 memcmp(&grp->il_dependladdr,
348 &inp->inp_inc.inc_ie.ie_dependladdr,
349 sizeof(grp->il_dependladdr)) == 0)
353 /* Create new load balance group. */
354 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
355 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
356 INPCBLBGROUP_SIZMIN);
359 } else if (grp->il_inpcnt == grp->il_inpsiz) {
360 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
361 if (ratecheck(&lastprint, &interval))
362 printf("lb group port %d, limit reached\n",
363 ntohs(grp->il_lport));
367 /* Expand this local group. */
368 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
373 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
374 ("invalid local group size %d and count %d", grp->il_inpsiz,
377 grp->il_inp[grp->il_inpcnt] = inp;
383 * Remove PCB from load balance group.
386 in_pcbremlbgrouphash(struct inpcb *inp)
388 struct inpcbinfo *pcbinfo;
389 struct inpcblbgrouphead *hdr;
390 struct inpcblbgroup *grp;
393 pcbinfo = inp->inp_pcbinfo;
395 INP_WLOCK_ASSERT(inp);
396 INP_HASH_WLOCK_ASSERT(pcbinfo);
398 hdr = &pcbinfo->ipi_lbgrouphashbase[
399 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
400 CK_LIST_FOREACH(grp, hdr, il_list) {
401 for (i = 0; i < grp->il_inpcnt; ++i) {
402 if (grp->il_inp[i] != inp)
405 if (grp->il_inpcnt == 1) {
406 /* We are the last, free this local group. */
407 in_pcblbgroup_free(grp);
409 /* Pull up inpcbs, shrink group if possible. */
410 in_pcblbgroup_reorder(hdr, &grp, i);
418 * Different protocols initialize their inpcbs differently - giving
419 * different name to the lock. But they all are disposed the same.
422 inpcb_fini(void *mem, int size)
424 struct inpcb *inp = mem;
426 INP_LOCK_DESTROY(inp);
430 * Initialize an inpcbinfo -- we should be able to reduce the number of
434 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
435 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
436 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
439 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
441 INP_INFO_LOCK_INIT(pcbinfo, name);
442 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
443 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
445 pcbinfo->ipi_vnet = curvnet;
447 pcbinfo->ipi_listhead = listhead;
448 CK_LIST_INIT(pcbinfo->ipi_listhead);
449 pcbinfo->ipi_count = 0;
450 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
451 &pcbinfo->ipi_hashmask);
452 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
453 &pcbinfo->ipi_porthashmask);
454 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
455 &pcbinfo->ipi_lbgrouphashmask);
457 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
459 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
460 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
461 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
462 uma_zone_set_warning(pcbinfo->ipi_zone,
463 "kern.ipc.maxsockets limit reached");
467 * Destroy an inpcbinfo.
470 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
473 KASSERT(pcbinfo->ipi_count == 0,
474 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
476 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
477 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
478 pcbinfo->ipi_porthashmask);
479 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
480 pcbinfo->ipi_lbgrouphashmask);
482 in_pcbgroup_destroy(pcbinfo);
484 uma_zdestroy(pcbinfo->ipi_zone);
485 INP_LIST_LOCK_DESTROY(pcbinfo);
486 INP_HASH_LOCK_DESTROY(pcbinfo);
487 INP_INFO_LOCK_DESTROY(pcbinfo);
491 * Allocate a PCB and associate it with the socket.
492 * On success return with the PCB locked.
495 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
501 if (pcbinfo == &V_tcbinfo) {
502 INP_INFO_RLOCK_ASSERT(pcbinfo);
504 INP_INFO_WLOCK_ASSERT(pcbinfo);
509 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
512 bzero(&inp->inp_start_zero, inp_zero_size);
513 inp->inp_pcbinfo = pcbinfo;
514 inp->inp_socket = so;
515 inp->inp_cred = crhold(so->so_cred);
516 inp->inp_inc.inc_fibnum = so->so_fibnum;
518 error = mac_inpcb_init(inp, M_NOWAIT);
521 mac_inpcb_create(so, inp);
523 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
524 error = ipsec_init_pcbpolicy(inp);
527 mac_inpcb_destroy(inp);
533 if (INP_SOCKAF(so) == AF_INET6) {
534 inp->inp_vflag |= INP_IPV6PROTO;
536 inp->inp_flags |= IN6P_IPV6_V6ONLY;
540 INP_LIST_WLOCK(pcbinfo);
541 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
542 pcbinfo->ipi_count++;
543 so->so_pcb = (caddr_t)inp;
545 if (V_ip6_auto_flowlabel)
546 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
548 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
549 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
552 * Routes in inpcb's can cache L2 as well; they are guaranteed
555 inp->inp_route.ro_flags = RT_LLE_CACHE;
556 INP_LIST_WUNLOCK(pcbinfo);
557 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
560 crfree(inp->inp_cred);
561 uma_zfree(pcbinfo->ipi_zone, inp);
569 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
573 INP_WLOCK_ASSERT(inp);
574 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
576 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
578 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
579 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
580 &inp->inp_lport, cred);
583 if (in_pcbinshash(inp) != 0) {
584 inp->inp_laddr.s_addr = INADDR_ANY;
589 inp->inp_flags |= INP_ANONPORT;
594 #if defined(INET) || defined(INET6)
596 * Assign a local port like in_pcb_lport(), but also used with connect()
597 * and a foreign address and port. If fsa is non-NULL, choose a local port
598 * that is unused with those, otherwise one that is completely unused.
599 * lsa can be NULL for IPv6.
602 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
603 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
605 struct inpcbinfo *pcbinfo;
606 struct inpcb *tmpinp;
607 unsigned short *lastport;
608 int count, dorandom, error;
609 u_short aux, first, last, lport;
611 struct in_addr laddr, faddr;
614 struct in6_addr *laddr6, *faddr6;
617 pcbinfo = inp->inp_pcbinfo;
620 * Because no actual state changes occur here, a global write lock on
621 * the pcbinfo isn't required.
623 INP_LOCK_ASSERT(inp);
624 INP_HASH_LOCK_ASSERT(pcbinfo);
626 if (inp->inp_flags & INP_HIGHPORT) {
627 first = V_ipport_hifirstauto; /* sysctl */
628 last = V_ipport_hilastauto;
629 lastport = &pcbinfo->ipi_lasthi;
630 } else if (inp->inp_flags & INP_LOWPORT) {
631 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
634 first = V_ipport_lowfirstauto; /* 1023 */
635 last = V_ipport_lowlastauto; /* 600 */
636 lastport = &pcbinfo->ipi_lastlow;
638 first = V_ipport_firstauto; /* sysctl */
639 last = V_ipport_lastauto;
640 lastport = &pcbinfo->ipi_lastport;
643 * For UDP(-Lite), use random port allocation as long as the user
644 * allows it. For TCP (and as of yet unknown) connections,
645 * use random port allocation only if the user allows it AND
646 * ipport_tick() allows it.
648 if (V_ipport_randomized &&
649 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
650 pcbinfo == &V_ulitecbinfo))
655 * It makes no sense to do random port allocation if
656 * we have the only port available.
660 /* Make sure to not include UDP(-Lite) packets in the count. */
661 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
662 V_ipport_tcpallocs++;
664 * Instead of having two loops further down counting up or down
665 * make sure that first is always <= last and go with only one
666 * code path implementing all logic.
675 laddr.s_addr = INADDR_ANY;
676 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
678 laddr = ((struct sockaddr_in *)lsa)->sin_addr;
680 faddr = ((struct sockaddr_in *)fsa)->sin_addr;
685 if ((inp->inp_vflag & INP_IPV6) != 0) {
687 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
689 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
697 *lastport = first + (arc4random() % (last - first));
699 count = last - first;
702 if (count-- < 0) /* completely used? */
703 return (EADDRNOTAVAIL);
705 if (*lastport < first || *lastport > last)
707 lport = htons(*lastport);
712 if (lsa->sa_family == AF_INET) {
713 tmpinp = in_pcblookup_hash_locked(pcbinfo,
714 faddr, fport, laddr, lport, lookupflags,
719 if (lsa->sa_family == AF_INET6) {
720 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
721 faddr6, fport, laddr6, lport, lookupflags,
727 if ((inp->inp_vflag & INP_IPV6) != 0)
728 tmpinp = in6_pcblookup_local(pcbinfo,
729 &inp->in6p_laddr, lport, lookupflags, cred);
731 #if defined(INET) && defined(INET6)
735 tmpinp = in_pcblookup_local(pcbinfo, laddr,
736 lport, lookupflags, cred);
739 } while (tmpinp != NULL);
747 * Select a local port (number) to use.
750 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
751 struct ucred *cred, int lookupflags)
753 struct sockaddr_in laddr;
756 bzero(&laddr, sizeof(laddr));
757 laddr.sin_family = AF_INET;
758 laddr.sin_addr = *laddrp;
760 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
761 NULL, lportp, NULL, 0, cred, lookupflags));
765 * Return cached socket options.
768 inp_so_options(const struct inpcb *inp)
774 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
775 so_options |= SO_REUSEPORT_LB;
776 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
777 so_options |= SO_REUSEPORT;
778 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
779 so_options |= SO_REUSEADDR;
782 #endif /* INET || INET6 */
785 * Check if a new BINDMULTI socket is allowed to be created.
787 * ni points to the new inp.
788 * oi points to the exisitng inp.
790 * This checks whether the existing inp also has BINDMULTI and
791 * whether the credentials match.
794 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
796 /* Check permissions match */
797 if ((ni->inp_flags2 & INP_BINDMULTI) &&
798 (ni->inp_cred->cr_uid !=
799 oi->inp_cred->cr_uid))
802 /* Check the existing inp has BINDMULTI set */
803 if ((ni->inp_flags2 & INP_BINDMULTI) &&
804 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
808 * We're okay - either INP_BINDMULTI isn't set on ni, or
809 * it is and it matches the checks.
816 * Set up a bind operation on a PCB, performing port allocation
817 * as required, but do not actually modify the PCB. Callers can
818 * either complete the bind by setting inp_laddr/inp_lport and
819 * calling in_pcbinshash(), or they can just use the resulting
820 * port and address to authorise the sending of a once-off packet.
822 * On error, the values of *laddrp and *lportp are not changed.
825 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
826 u_short *lportp, struct ucred *cred)
828 struct socket *so = inp->inp_socket;
829 struct sockaddr_in *sin;
830 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
831 struct in_addr laddr;
833 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
837 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
838 * so that we don't have to add to the (already messy) code below.
840 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
843 * No state changes, so read locks are sufficient here.
845 INP_LOCK_ASSERT(inp);
846 INP_HASH_LOCK_ASSERT(pcbinfo);
848 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
849 return (EADDRNOTAVAIL);
850 laddr.s_addr = *laddrp;
851 if (nam != NULL && laddr.s_addr != INADDR_ANY)
853 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
854 lookupflags = INPLOOKUP_WILDCARD;
856 if ((error = prison_local_ip4(cred, &laddr)) != 0)
859 sin = (struct sockaddr_in *)nam;
860 if (nam->sa_len != sizeof (*sin))
864 * We should check the family, but old programs
865 * incorrectly fail to initialize it.
867 if (sin->sin_family != AF_INET)
868 return (EAFNOSUPPORT);
870 error = prison_local_ip4(cred, &sin->sin_addr);
873 if (sin->sin_port != *lportp) {
874 /* Don't allow the port to change. */
877 lport = sin->sin_port;
879 /* NB: lport is left as 0 if the port isn't being changed. */
880 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
882 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
883 * allow complete duplication of binding if
884 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
885 * and a multicast address is bound on both
886 * new and duplicated sockets.
888 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
889 reuseport = SO_REUSEADDR|SO_REUSEPORT;
891 * XXX: How to deal with SO_REUSEPORT_LB here?
892 * Treat same as SO_REUSEPORT for now.
894 if ((so->so_options &
895 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
896 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
897 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
898 sin->sin_port = 0; /* yech... */
899 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
901 * Is the address a local IP address?
902 * If INP_BINDANY is set, then the socket may be bound
903 * to any endpoint address, local or not.
905 if ((inp->inp_flags & INP_BINDANY) == 0 &&
906 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
907 return (EADDRNOTAVAIL);
909 laddr = sin->sin_addr;
915 if (ntohs(lport) <= V_ipport_reservedhigh &&
916 ntohs(lport) >= V_ipport_reservedlow &&
917 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
920 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
921 priv_check_cred(inp->inp_cred,
922 PRIV_NETINET_REUSEPORT, 0) != 0) {
923 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
924 lport, INPLOOKUP_WILDCARD, cred);
927 * This entire block sorely needs a rewrite.
930 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
931 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
932 (so->so_type != SOCK_STREAM ||
933 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
934 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
935 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
936 (t->inp_flags2 & INP_REUSEPORT) ||
937 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
938 (inp->inp_cred->cr_uid !=
939 t->inp_cred->cr_uid))
943 * If the socket is a BINDMULTI socket, then
944 * the credentials need to match and the
945 * original socket also has to have been bound
948 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
951 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
952 lport, lookupflags, cred);
953 if (t && (t->inp_flags & INP_TIMEWAIT)) {
955 * XXXRW: If an incpb has had its timewait
956 * state recycled, we treat the address as
957 * being in use (for now). This is better
958 * than a panic, but not desirable.
962 ((reuseport & tw->tw_so_options) == 0 &&
964 tw->tw_so_options) == 0)) {
968 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
969 (reuseport & inp_so_options(t)) == 0 &&
970 (reuseport_lb & inp_so_options(t)) == 0) {
972 if (ntohl(sin->sin_addr.s_addr) !=
974 ntohl(t->inp_laddr.s_addr) !=
976 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
977 (t->inp_vflag & INP_IPV6PROTO) == 0)
980 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
988 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
993 *laddrp = laddr.s_addr;
999 * Connect from a socket to a specified address.
1000 * Both address and port must be specified in argument sin.
1001 * If don't have a local address for this socket yet,
1005 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
1006 struct ucred *cred, struct mbuf *m, bool rehash)
1008 u_short lport, fport;
1009 in_addr_t laddr, faddr;
1010 int anonport, error;
1012 INP_WLOCK_ASSERT(inp);
1013 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1015 lport = inp->inp_lport;
1016 laddr = inp->inp_laddr.s_addr;
1017 anonport = (lport == 0);
1018 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
1023 /* Do the initial binding of the local address if required. */
1024 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1025 KASSERT(rehash == true,
1026 ("Rehashing required for unbound inps"));
1027 inp->inp_lport = lport;
1028 inp->inp_laddr.s_addr = laddr;
1029 if (in_pcbinshash(inp) != 0) {
1030 inp->inp_laddr.s_addr = INADDR_ANY;
1036 /* Commit the remaining changes. */
1037 inp->inp_lport = lport;
1038 inp->inp_laddr.s_addr = laddr;
1039 inp->inp_faddr.s_addr = faddr;
1040 inp->inp_fport = fport;
1042 in_pcbrehash_mbuf(inp, m);
1044 in_pcbinshash_mbuf(inp, m);
1048 inp->inp_flags |= INP_ANONPORT;
1053 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1056 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1060 * Do proper source address selection on an unbound socket in case
1061 * of connect. Take jails into account as well.
1064 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1068 struct sockaddr *sa;
1069 struct sockaddr_in *sin;
1073 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1075 * Bypass source address selection and use the primary jail IP
1078 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1082 bzero(&sro, sizeof(sro));
1084 sin = (struct sockaddr_in *)&sro.ro_dst;
1085 sin->sin_family = AF_INET;
1086 sin->sin_len = sizeof(struct sockaddr_in);
1087 sin->sin_addr.s_addr = faddr->s_addr;
1090 * If route is known our src addr is taken from the i/f,
1093 * Find out route to destination.
1095 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1096 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1099 * If we found a route, use the address corresponding to
1100 * the outgoing interface.
1102 * Otherwise assume faddr is reachable on a directly connected
1103 * network and try to find a corresponding interface to take
1104 * the source address from.
1107 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1108 struct in_ifaddr *ia;
1111 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1112 inp->inp_socket->so_fibnum));
1114 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1115 inp->inp_socket->so_fibnum));
1119 error = ENETUNREACH;
1123 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1124 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1130 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1133 if (sa->sa_family != AF_INET)
1135 sin = (struct sockaddr_in *)sa;
1136 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1137 ia = (struct in_ifaddr *)ifa;
1142 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1146 /* 3. As a last resort return the 'default' jail address. */
1147 error = prison_get_ip4(cred, laddr);
1152 * If the outgoing interface on the route found is not
1153 * a loopback interface, use the address from that interface.
1154 * In case of jails do those three steps:
1155 * 1. check if the interface address belongs to the jail. If so use it.
1156 * 2. check if we have any address on the outgoing interface
1157 * belonging to this jail. If so use it.
1158 * 3. as a last resort return the 'default' jail address.
1160 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1161 struct in_ifaddr *ia;
1164 /* If not jailed, use the default returned. */
1165 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1166 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1167 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1172 /* 1. Check if the iface address belongs to the jail. */
1173 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1174 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1175 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1176 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1181 * 2. Check if we have any address on the outgoing interface
1182 * belonging to this jail.
1185 ifp = sro.ro_rt->rt_ifp;
1186 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1188 if (sa->sa_family != AF_INET)
1190 sin = (struct sockaddr_in *)sa;
1191 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1192 ia = (struct in_ifaddr *)ifa;
1197 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1201 /* 3. As a last resort return the 'default' jail address. */
1202 error = prison_get_ip4(cred, laddr);
1207 * The outgoing interface is marked with 'loopback net', so a route
1208 * to ourselves is here.
1209 * Try to find the interface of the destination address and then
1210 * take the address from there. That interface is not necessarily
1211 * a loopback interface.
1212 * In case of jails, check that it is an address of the jail
1213 * and if we cannot find, fall back to the 'default' jail address.
1215 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1216 struct sockaddr_in sain;
1217 struct in_ifaddr *ia;
1219 bzero(&sain, sizeof(struct sockaddr_in));
1220 sain.sin_family = AF_INET;
1221 sain.sin_len = sizeof(struct sockaddr_in);
1222 sain.sin_addr.s_addr = faddr->s_addr;
1224 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1225 inp->inp_socket->so_fibnum));
1227 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1228 inp->inp_socket->so_fibnum));
1230 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1232 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1234 error = ENETUNREACH;
1237 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1247 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1249 if (sa->sa_family != AF_INET)
1251 sin = (struct sockaddr_in *)sa;
1252 if (prison_check_ip4(cred,
1253 &sin->sin_addr) == 0) {
1254 ia = (struct in_ifaddr *)ifa;
1259 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1264 /* 3. As a last resort return the 'default' jail address. */
1265 error = prison_get_ip4(cred, laddr);
1271 if (sro.ro_rt != NULL)
1277 * Set up for a connect from a socket to the specified address.
1278 * On entry, *laddrp and *lportp should contain the current local
1279 * address and port for the PCB; these are updated to the values
1280 * that should be placed in inp_laddr and inp_lport to complete
1283 * On success, *faddrp and *fportp will be set to the remote address
1284 * and port. These are not updated in the error case.
1286 * If the operation fails because the connection already exists,
1287 * *oinpp will be set to the PCB of that connection so that the
1288 * caller can decide to override it. In all other cases, *oinpp
1292 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1293 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1294 struct inpcb **oinpp, struct ucred *cred)
1296 struct rm_priotracker in_ifa_tracker;
1297 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1298 struct in_ifaddr *ia;
1300 struct in_addr laddr, faddr;
1301 u_short lport, fport;
1305 * Because a global state change doesn't actually occur here, a read
1306 * lock is sufficient.
1308 INP_LOCK_ASSERT(inp);
1309 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1313 if (nam->sa_len != sizeof (*sin))
1315 if (sin->sin_family != AF_INET)
1316 return (EAFNOSUPPORT);
1317 if (sin->sin_port == 0)
1318 return (EADDRNOTAVAIL);
1319 laddr.s_addr = *laddrp;
1321 faddr = sin->sin_addr;
1322 fport = sin->sin_port;
1324 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1326 * If the destination address is INADDR_ANY,
1327 * use the primary local address.
1328 * If the supplied address is INADDR_BROADCAST,
1329 * and the primary interface supports broadcast,
1330 * choose the broadcast address for that interface.
1332 if (faddr.s_addr == INADDR_ANY) {
1333 IN_IFADDR_RLOCK(&in_ifa_tracker);
1335 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1336 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1338 (error = prison_get_ip4(cred, &faddr)) != 0)
1340 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1341 IN_IFADDR_RLOCK(&in_ifa_tracker);
1342 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1344 faddr = satosin(&CK_STAILQ_FIRST(
1345 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1346 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1349 if (laddr.s_addr == INADDR_ANY) {
1350 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1352 * If the destination address is multicast and an outgoing
1353 * interface has been set as a multicast option, prefer the
1354 * address of that interface as our source address.
1356 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1357 inp->inp_moptions != NULL) {
1358 struct ip_moptions *imo;
1361 imo = inp->inp_moptions;
1362 if (imo->imo_multicast_ifp != NULL) {
1363 ifp = imo->imo_multicast_ifp;
1364 IN_IFADDR_RLOCK(&in_ifa_tracker);
1365 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1366 if ((ia->ia_ifp == ifp) &&
1368 prison_check_ip4(cred,
1369 &ia->ia_addr.sin_addr) == 0))
1373 error = EADDRNOTAVAIL;
1375 laddr = ia->ia_addr.sin_addr;
1378 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1385 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1386 fport, laddr, lport, 0, NULL);
1390 return (EADDRINUSE);
1393 struct sockaddr_in lsin, fsin;
1395 bzero(&lsin, sizeof(lsin));
1396 bzero(&fsin, sizeof(fsin));
1397 lsin.sin_family = AF_INET;
1398 lsin.sin_addr = laddr;
1399 fsin.sin_family = AF_INET;
1400 fsin.sin_addr = faddr;
1401 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1402 &lport, (struct sockaddr *)& fsin, fport, cred,
1403 INPLOOKUP_WILDCARD);
1407 *laddrp = laddr.s_addr;
1409 *faddrp = faddr.s_addr;
1415 in_pcbdisconnect(struct inpcb *inp)
1418 INP_WLOCK_ASSERT(inp);
1419 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1421 inp->inp_faddr.s_addr = INADDR_ANY;
1428 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1429 * For most protocols, this will be invoked immediately prior to calling
1430 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1431 * socket, in which case in_pcbfree() is deferred.
1434 in_pcbdetach(struct inpcb *inp)
1437 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1440 if (inp->inp_snd_tag != NULL)
1441 in_pcbdetach_txrtlmt(inp);
1443 inp->inp_socket->so_pcb = NULL;
1444 inp->inp_socket = NULL;
1448 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1449 * stability of an inpcb pointer despite the inpcb lock being released. This
1450 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1451 * but where the inpcb lock may already held, or when acquiring a reference
1454 * in_pcbref() should be used only to provide brief memory stability, and
1455 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1456 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1457 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1458 * lock and rele are the *only* safe operations that may be performed on the
1461 * While the inpcb will not be freed, releasing the inpcb lock means that the
1462 * connection's state may change, so the caller should be careful to
1463 * revalidate any cached state on reacquiring the lock. Drop the reference
1464 * using in_pcbrele().
1467 in_pcbref(struct inpcb *inp)
1470 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1472 refcount_acquire(&inp->inp_refcount);
1476 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1477 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1478 * return a flag indicating whether or not the inpcb remains valid. If it is
1479 * valid, we return with the inpcb lock held.
1481 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1482 * reference on an inpcb. Historically more work was done here (actually, in
1483 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1484 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1485 * about memory stability (and continued use of the write lock).
1488 in_pcbrele_rlocked(struct inpcb *inp)
1490 struct inpcbinfo *pcbinfo;
1492 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1494 INP_RLOCK_ASSERT(inp);
1496 if (refcount_release(&inp->inp_refcount) == 0) {
1498 * If the inpcb has been freed, let the caller know, even if
1499 * this isn't the last reference.
1501 if (inp->inp_flags2 & INP_FREED) {
1508 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1510 if (inp->inp_in_hpts || inp->inp_in_input) {
1511 struct tcp_hpts_entry *hpts;
1513 * We should not be on the hpts at
1514 * this point in any form. we must
1515 * get the lock to be sure.
1517 hpts = tcp_hpts_lock(inp);
1518 if (inp->inp_in_hpts)
1519 panic("Hpts:%p inp:%p at free still on hpts",
1521 mtx_unlock(&hpts->p_mtx);
1522 hpts = tcp_input_lock(inp);
1523 if (inp->inp_in_input)
1524 panic("Hpts:%p inp:%p at free still on input hpts",
1526 mtx_unlock(&hpts->p_mtx);
1530 pcbinfo = inp->inp_pcbinfo;
1531 uma_zfree(pcbinfo->ipi_zone, inp);
1536 in_pcbrele_wlocked(struct inpcb *inp)
1538 struct inpcbinfo *pcbinfo;
1540 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1542 INP_WLOCK_ASSERT(inp);
1544 if (refcount_release(&inp->inp_refcount) == 0) {
1546 * If the inpcb has been freed, let the caller know, even if
1547 * this isn't the last reference.
1549 if (inp->inp_flags2 & INP_FREED) {
1556 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1558 if (inp->inp_in_hpts || inp->inp_in_input) {
1559 struct tcp_hpts_entry *hpts;
1561 * We should not be on the hpts at
1562 * this point in any form. we must
1563 * get the lock to be sure.
1565 hpts = tcp_hpts_lock(inp);
1566 if (inp->inp_in_hpts)
1567 panic("Hpts:%p inp:%p at free still on hpts",
1569 mtx_unlock(&hpts->p_mtx);
1570 hpts = tcp_input_lock(inp);
1571 if (inp->inp_in_input)
1572 panic("Hpts:%p inp:%p at free still on input hpts",
1574 mtx_unlock(&hpts->p_mtx);
1578 pcbinfo = inp->inp_pcbinfo;
1579 uma_zfree(pcbinfo->ipi_zone, inp);
1584 * Temporary wrapper.
1587 in_pcbrele(struct inpcb *inp)
1590 return (in_pcbrele_wlocked(inp));
1594 in_pcblist_rele_rlocked(epoch_context_t ctx)
1596 struct in_pcblist *il;
1598 struct inpcbinfo *pcbinfo;
1601 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1602 pcbinfo = il->il_pcbinfo;
1604 INP_INFO_WLOCK(pcbinfo);
1605 for (i = 0; i < n; i++) {
1606 inp = il->il_inp_list[i];
1608 if (!in_pcbrele_rlocked(inp))
1611 INP_INFO_WUNLOCK(pcbinfo);
1616 inpcbport_free(epoch_context_t ctx)
1618 struct inpcbport *phd;
1620 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1625 in_pcbfree_deferred(epoch_context_t ctx)
1628 int released __unused;
1630 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1633 CURVNET_SET(inp->inp_vnet);
1635 struct ip_moptions *imo = inp->inp_moptions;
1636 inp->inp_moptions = NULL;
1638 /* XXXRW: Do as much as possible here. */
1639 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1640 if (inp->inp_sp != NULL)
1641 ipsec_delete_pcbpolicy(inp);
1644 struct ip6_moptions *im6o = NULL;
1645 if (inp->inp_vflag & INP_IPV6PROTO) {
1646 ip6_freepcbopts(inp->in6p_outputopts);
1647 im6o = inp->in6p_moptions;
1648 inp->in6p_moptions = NULL;
1651 if (inp->inp_options)
1652 (void)m_free(inp->inp_options);
1654 crfree(inp->inp_cred);
1656 mac_inpcb_destroy(inp);
1658 released = in_pcbrele_wlocked(inp);
1661 ip6_freemoptions(im6o);
1664 inp_freemoptions(imo);
1670 * Unconditionally schedule an inpcb to be freed by decrementing its
1671 * reference count, which should occur only after the inpcb has been detached
1672 * from its socket. If another thread holds a temporary reference (acquired
1673 * using in_pcbref()) then the free is deferred until that reference is
1674 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1675 * work, including removal from global lists, is done in this context, where
1676 * the pcbinfo lock is held.
1679 in_pcbfree(struct inpcb *inp)
1681 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1683 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1684 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1685 ("%s: called twice for pcb %p", __func__, inp));
1686 if (inp->inp_flags2 & INP_FREED) {
1692 if (pcbinfo == &V_tcbinfo) {
1693 INP_INFO_LOCK_ASSERT(pcbinfo);
1695 INP_INFO_WLOCK_ASSERT(pcbinfo);
1698 INP_WLOCK_ASSERT(inp);
1699 INP_LIST_WLOCK(pcbinfo);
1700 in_pcbremlists(inp);
1701 INP_LIST_WUNLOCK(pcbinfo);
1702 RO_INVALIDATE_CACHE(&inp->inp_route);
1703 /* mark as destruction in progress */
1704 inp->inp_flags2 |= INP_FREED;
1706 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1710 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1711 * port reservation, and preventing it from being returned by inpcb lookups.
1713 * It is used by TCP to mark an inpcb as unused and avoid future packet
1714 * delivery or event notification when a socket remains open but TCP has
1715 * closed. This might occur as a result of a shutdown()-initiated TCP close
1716 * or a RST on the wire, and allows the port binding to be reused while still
1717 * maintaining the invariant that so_pcb always points to a valid inpcb until
1720 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1721 * in_pcbnotifyall() and in_pcbpurgeif0()?
1724 in_pcbdrop(struct inpcb *inp)
1727 INP_WLOCK_ASSERT(inp);
1729 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1730 MPASS(inp->inp_refcount > 1);
1734 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1737 inp->inp_flags |= INP_DROPPED;
1738 if (inp->inp_flags & INP_INHASHLIST) {
1739 struct inpcbport *phd = inp->inp_phd;
1741 INP_HASH_WLOCK(inp->inp_pcbinfo);
1742 in_pcbremlbgrouphash(inp);
1743 CK_LIST_REMOVE(inp, inp_hash);
1744 CK_LIST_REMOVE(inp, inp_portlist);
1745 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1746 CK_LIST_REMOVE(phd, phd_hash);
1747 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1749 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1750 inp->inp_flags &= ~INP_INHASHLIST;
1752 in_pcbgroup_remove(inp);
1759 * Common routines to return the socket addresses associated with inpcbs.
1762 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1764 struct sockaddr_in *sin;
1766 sin = malloc(sizeof *sin, M_SONAME,
1768 sin->sin_family = AF_INET;
1769 sin->sin_len = sizeof(*sin);
1770 sin->sin_addr = *addr_p;
1771 sin->sin_port = port;
1773 return (struct sockaddr *)sin;
1777 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1780 struct in_addr addr;
1783 inp = sotoinpcb(so);
1784 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1787 port = inp->inp_lport;
1788 addr = inp->inp_laddr;
1791 *nam = in_sockaddr(port, &addr);
1796 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1799 struct in_addr addr;
1802 inp = sotoinpcb(so);
1803 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1806 port = inp->inp_fport;
1807 addr = inp->inp_faddr;
1810 *nam = in_sockaddr(port, &addr);
1815 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1816 struct inpcb *(*notify)(struct inpcb *, int))
1818 struct inpcb *inp, *inp_temp;
1820 INP_INFO_WLOCK(pcbinfo);
1821 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1824 if ((inp->inp_vflag & INP_IPV4) == 0) {
1829 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1830 inp->inp_socket == NULL) {
1834 if ((*notify)(inp, errno))
1837 INP_INFO_WUNLOCK(pcbinfo);
1841 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1844 struct in_multi *inm;
1845 struct in_mfilter *imf;
1846 struct ip_moptions *imo;
1848 INP_INFO_WLOCK(pcbinfo);
1849 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1851 imo = inp->inp_moptions;
1852 if ((inp->inp_vflag & INP_IPV4) &&
1855 * Unselect the outgoing interface if it is being
1858 if (imo->imo_multicast_ifp == ifp)
1859 imo->imo_multicast_ifp = NULL;
1862 * Drop multicast group membership if we joined
1863 * through the interface being detached.
1865 * XXX This can all be deferred to an epoch_call
1868 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1869 if ((inm = imf->imf_inm) == NULL)
1871 if (inm->inm_ifp != ifp)
1873 ip_mfilter_remove(&imo->imo_head, imf);
1874 IN_MULTI_LOCK_ASSERT();
1875 in_leavegroup_locked(inm, NULL);
1876 ip_mfilter_free(imf);
1882 INP_INFO_WUNLOCK(pcbinfo);
1886 * Lookup a PCB based on the local address and port. Caller must hold the
1887 * hash lock. No inpcb locks or references are acquired.
1889 #define INP_LOOKUP_MAPPED_PCB_COST 3
1891 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1892 u_short lport, int lookupflags, struct ucred *cred)
1896 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1902 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1903 ("%s: invalid lookup flags %d", __func__, lookupflags));
1905 INP_HASH_LOCK_ASSERT(pcbinfo);
1907 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1908 struct inpcbhead *head;
1910 * Look for an unconnected (wildcard foreign addr) PCB that
1911 * matches the local address and port we're looking for.
1913 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1914 0, pcbinfo->ipi_hashmask)];
1915 CK_LIST_FOREACH(inp, head, inp_hash) {
1917 /* XXX inp locking */
1918 if ((inp->inp_vflag & INP_IPV4) == 0)
1921 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1922 inp->inp_laddr.s_addr == laddr.s_addr &&
1923 inp->inp_lport == lport) {
1928 prison_equal_ip4(cred->cr_prison,
1929 inp->inp_cred->cr_prison))
1938 struct inpcbporthead *porthash;
1939 struct inpcbport *phd;
1940 struct inpcb *match = NULL;
1942 * Best fit PCB lookup.
1944 * First see if this local port is in use by looking on the
1947 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1948 pcbinfo->ipi_porthashmask)];
1949 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1950 if (phd->phd_port == lport)
1955 * Port is in use by one or more PCBs. Look for best
1958 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1961 !prison_equal_ip4(inp->inp_cred->cr_prison,
1965 /* XXX inp locking */
1966 if ((inp->inp_vflag & INP_IPV4) == 0)
1969 * We never select the PCB that has
1970 * INP_IPV6 flag and is bound to :: if
1971 * we have another PCB which is bound
1972 * to 0.0.0.0. If a PCB has the
1973 * INP_IPV6 flag, then we set its cost
1974 * higher than IPv4 only PCBs.
1976 * Note that the case only happens
1977 * when a socket is bound to ::, under
1978 * the condition that the use of the
1979 * mapped address is allowed.
1981 if ((inp->inp_vflag & INP_IPV6) != 0)
1982 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1984 if (inp->inp_faddr.s_addr != INADDR_ANY)
1986 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1987 if (laddr.s_addr == INADDR_ANY)
1989 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1992 if (laddr.s_addr != INADDR_ANY)
1995 if (wildcard < matchwild) {
1997 matchwild = wildcard;
2006 #undef INP_LOOKUP_MAPPED_PCB_COST
2008 static struct inpcb *
2009 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2010 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
2011 uint16_t fport, int lookupflags)
2013 struct inpcb *local_wild;
2014 const struct inpcblbgrouphead *hdr;
2015 struct inpcblbgroup *grp;
2018 INP_HASH_LOCK_ASSERT(pcbinfo);
2020 hdr = &pcbinfo->ipi_lbgrouphashbase[
2021 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2024 * Order of socket selection:
2026 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
2029 * - Load balanced group does not contain jailed sockets
2030 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
2033 CK_LIST_FOREACH(grp, hdr, il_list) {
2035 if (!(grp->il_vflag & INP_IPV4))
2038 if (grp->il_lport != lport)
2041 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
2043 if (grp->il_laddr.s_addr == laddr->s_addr)
2044 return (grp->il_inp[idx]);
2045 if (grp->il_laddr.s_addr == INADDR_ANY &&
2046 (lookupflags & INPLOOKUP_WILDCARD) != 0)
2047 local_wild = grp->il_inp[idx];
2049 return (local_wild);
2054 * Lookup PCB in hash list, using pcbgroup tables.
2056 static struct inpcb *
2057 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2058 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2059 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2061 struct inpcbhead *head;
2062 struct inpcb *inp, *tmpinp;
2063 u_short fport = fport_arg, lport = lport_arg;
2067 * First look for an exact match.
2070 INP_GROUP_LOCK(pcbgroup);
2071 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2072 pcbgroup->ipg_hashmask)];
2073 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2075 /* XXX inp locking */
2076 if ((inp->inp_vflag & INP_IPV4) == 0)
2079 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2080 inp->inp_laddr.s_addr == laddr.s_addr &&
2081 inp->inp_fport == fport &&
2082 inp->inp_lport == lport) {
2084 * XXX We should be able to directly return
2085 * the inp here, without any checks.
2086 * Well unless both bound with SO_REUSEPORT?
2088 if (prison_flag(inp->inp_cred, PR_IP4))
2094 if (tmpinp != NULL) {
2101 * For incoming connections, we may wish to do a wildcard
2102 * match for an RSS-local socket.
2104 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2105 struct inpcb *local_wild = NULL, *local_exact = NULL;
2107 struct inpcb *local_wild_mapped = NULL;
2109 struct inpcb *jail_wild = NULL;
2110 struct inpcbhead *head;
2114 * Order of socket selection - we always prefer jails.
2115 * 1. jailed, non-wild.
2117 * 3. non-jailed, non-wild.
2118 * 4. non-jailed, wild.
2121 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2122 lport, 0, pcbgroup->ipg_hashmask)];
2123 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2125 /* XXX inp locking */
2126 if ((inp->inp_vflag & INP_IPV4) == 0)
2129 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2130 inp->inp_lport != lport)
2133 injail = prison_flag(inp->inp_cred, PR_IP4);
2135 if (prison_check_ip4(inp->inp_cred,
2139 if (local_exact != NULL)
2143 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2148 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2150 /* XXX inp locking, NULL check */
2151 if (inp->inp_vflag & INP_IPV6PROTO)
2152 local_wild_mapped = inp;
2160 } /* LIST_FOREACH */
2169 inp = local_wild_mapped;
2177 * Then look for a wildcard match, if requested.
2179 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2180 struct inpcb *local_wild = NULL, *local_exact = NULL;
2182 struct inpcb *local_wild_mapped = NULL;
2184 struct inpcb *jail_wild = NULL;
2185 struct inpcbhead *head;
2189 * Order of socket selection - we always prefer jails.
2190 * 1. jailed, non-wild.
2192 * 3. non-jailed, non-wild.
2193 * 4. non-jailed, wild.
2195 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2196 0, pcbinfo->ipi_wildmask)];
2197 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2199 /* XXX inp locking */
2200 if ((inp->inp_vflag & INP_IPV4) == 0)
2203 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2204 inp->inp_lport != lport)
2207 injail = prison_flag(inp->inp_cred, PR_IP4);
2209 if (prison_check_ip4(inp->inp_cred,
2213 if (local_exact != NULL)
2217 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2222 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2224 /* XXX inp locking, NULL check */
2225 if (inp->inp_vflag & INP_IPV6PROTO)
2226 local_wild_mapped = inp;
2234 } /* LIST_FOREACH */
2242 inp = local_wild_mapped;
2246 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2247 INP_GROUP_UNLOCK(pcbgroup);
2251 if (lookupflags & INPLOOKUP_WLOCKPCB)
2252 locked = INP_TRY_WLOCK(inp);
2253 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2254 locked = INP_TRY_RLOCK(inp);
2256 panic("%s: locking bug", __func__);
2257 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2258 if (lookupflags & INPLOOKUP_WLOCKPCB)
2265 INP_GROUP_UNLOCK(pcbgroup);
2267 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2269 if (in_pcbrele_wlocked(inp))
2273 if (in_pcbrele_rlocked(inp))
2278 if (lookupflags & INPLOOKUP_WLOCKPCB)
2279 INP_WLOCK_ASSERT(inp);
2281 INP_RLOCK_ASSERT(inp);
2285 #endif /* PCBGROUP */
2288 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2289 * that the caller has locked the hash list, and will not perform any further
2290 * locking or reference operations on either the hash list or the connection.
2292 static struct inpcb *
2293 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2294 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2297 struct inpcbhead *head;
2298 struct inpcb *inp, *tmpinp;
2299 u_short fport = fport_arg, lport = lport_arg;
2302 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2303 ("%s: invalid lookup flags %d", __func__, lookupflags));
2304 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2305 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2308 * First look for an exact match.
2311 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2312 pcbinfo->ipi_hashmask)];
2313 CK_LIST_FOREACH(inp, head, inp_hash) {
2315 /* XXX inp locking */
2316 if ((inp->inp_vflag & INP_IPV4) == 0)
2319 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2320 inp->inp_laddr.s_addr == laddr.s_addr &&
2321 inp->inp_fport == fport &&
2322 inp->inp_lport == lport) {
2324 * XXX We should be able to directly return
2325 * the inp here, without any checks.
2326 * Well unless both bound with SO_REUSEPORT?
2328 if (prison_flag(inp->inp_cred, PR_IP4))
2338 * Then look in lb group (for wildcard match).
2340 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2341 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2342 fport, lookupflags);
2348 * Then look for a wildcard match, if requested.
2350 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2351 struct inpcb *local_wild = NULL, *local_exact = NULL;
2353 struct inpcb *local_wild_mapped = NULL;
2355 struct inpcb *jail_wild = NULL;
2359 * Order of socket selection - we always prefer jails.
2360 * 1. jailed, non-wild.
2362 * 3. non-jailed, non-wild.
2363 * 4. non-jailed, wild.
2366 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2367 0, pcbinfo->ipi_hashmask)];
2368 CK_LIST_FOREACH(inp, head, inp_hash) {
2370 /* XXX inp locking */
2371 if ((inp->inp_vflag & INP_IPV4) == 0)
2374 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2375 inp->inp_lport != lport)
2378 injail = prison_flag(inp->inp_cred, PR_IP4);
2380 if (prison_check_ip4(inp->inp_cred,
2384 if (local_exact != NULL)
2388 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2393 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2395 /* XXX inp locking, NULL check */
2396 if (inp->inp_vflag & INP_IPV6PROTO)
2397 local_wild_mapped = inp;
2405 } /* LIST_FOREACH */
2406 if (jail_wild != NULL)
2408 if (local_exact != NULL)
2409 return (local_exact);
2410 if (local_wild != NULL)
2411 return (local_wild);
2413 if (local_wild_mapped != NULL)
2414 return (local_wild_mapped);
2416 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2422 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2423 * hash list lock, and will return the inpcb locked (i.e., requires
2424 * INPLOOKUP_LOCKPCB).
2426 static struct inpcb *
2427 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2428 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2433 INP_HASH_RLOCK(pcbinfo);
2434 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2435 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2437 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2439 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2443 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2445 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2450 panic("%s: locking bug", __func__);
2453 if (lookupflags & INPLOOKUP_WLOCKPCB)
2454 INP_WLOCK_ASSERT(inp);
2456 INP_RLOCK_ASSERT(inp);
2460 INP_HASH_RUNLOCK(pcbinfo);
2465 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2466 * from which a pre-calculated hash value may be extracted.
2468 * Possibly more of this logic should be in in_pcbgroup.c.
2471 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2472 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2474 #if defined(PCBGROUP) && !defined(RSS)
2475 struct inpcbgroup *pcbgroup;
2478 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2479 ("%s: invalid lookup flags %d", __func__, lookupflags));
2480 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2481 ("%s: LOCKPCB not set", __func__));
2484 * When not using RSS, use connection groups in preference to the
2485 * reservation table when looking up 4-tuples. When using RSS, just
2486 * use the reservation table, due to the cost of the Toeplitz hash
2489 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2490 * we could be doing RSS with a non-Toeplitz hash that is affordable
2493 #if defined(PCBGROUP) && !defined(RSS)
2494 if (in_pcbgroup_enabled(pcbinfo)) {
2495 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2497 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2498 laddr, lport, lookupflags, ifp));
2501 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2506 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2507 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2508 struct ifnet *ifp, struct mbuf *m)
2511 struct inpcbgroup *pcbgroup;
2514 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2515 ("%s: invalid lookup flags %d", __func__, lookupflags));
2516 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2517 ("%s: LOCKPCB not set", __func__));
2521 * If we can use a hardware-generated hash to look up the connection
2522 * group, use that connection group to find the inpcb. Otherwise
2523 * fall back on a software hash -- or the reservation table if we're
2526 * XXXRW: As above, that policy belongs in the pcbgroup code.
2528 if (in_pcbgroup_enabled(pcbinfo) &&
2529 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2530 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2531 m->m_pkthdr.flowid);
2532 if (pcbgroup != NULL)
2533 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2534 fport, laddr, lport, lookupflags, ifp));
2536 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2538 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2539 laddr, lport, lookupflags, ifp));
2543 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2549 * Insert PCB onto various hash lists.
2552 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2554 struct inpcbhead *pcbhash;
2555 struct inpcbporthead *pcbporthash;
2556 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2557 struct inpcbport *phd;
2558 u_int32_t hashkey_faddr;
2561 INP_WLOCK_ASSERT(inp);
2562 INP_HASH_WLOCK_ASSERT(pcbinfo);
2564 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2565 ("in_pcbinshash: INP_INHASHLIST"));
2568 if (inp->inp_vflag & INP_IPV6)
2569 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2572 hashkey_faddr = inp->inp_faddr.s_addr;
2574 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2575 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2577 pcbporthash = &pcbinfo->ipi_porthashbase[
2578 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2581 * Add entry to load balance group.
2582 * Only do this if SO_REUSEPORT_LB is set.
2584 so_options = inp_so_options(inp);
2585 if (so_options & SO_REUSEPORT_LB) {
2586 int ret = in_pcbinslbgrouphash(inp);
2588 /* pcb lb group malloc fail (ret=ENOBUFS). */
2594 * Go through port list and look for a head for this lport.
2596 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2597 if (phd->phd_port == inp->inp_lport)
2601 * If none exists, malloc one and tack it on.
2604 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2606 return (ENOBUFS); /* XXX */
2608 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2609 phd->phd_port = inp->inp_lport;
2610 CK_LIST_INIT(&phd->phd_pcblist);
2611 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2614 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2615 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2616 inp->inp_flags |= INP_INHASHLIST;
2619 in_pcbgroup_update_mbuf(inp, m);
2621 in_pcbgroup_update(inp);
2628 in_pcbinshash(struct inpcb *inp)
2631 return (in_pcbinshash_internal(inp, NULL));
2635 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2638 return (in_pcbinshash_internal(inp, m));
2642 * Move PCB to the proper hash bucket when { faddr, fport } have been
2643 * changed. NOTE: This does not handle the case of the lport changing (the
2644 * hashed port list would have to be updated as well), so the lport must
2645 * not change after in_pcbinshash() has been called.
2648 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2650 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2651 struct inpcbhead *head;
2652 u_int32_t hashkey_faddr;
2654 INP_WLOCK_ASSERT(inp);
2655 INP_HASH_WLOCK_ASSERT(pcbinfo);
2657 KASSERT(inp->inp_flags & INP_INHASHLIST,
2658 ("in_pcbrehash: !INP_INHASHLIST"));
2661 if (inp->inp_vflag & INP_IPV6)
2662 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2665 hashkey_faddr = inp->inp_faddr.s_addr;
2667 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2668 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2670 CK_LIST_REMOVE(inp, inp_hash);
2671 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2675 in_pcbgroup_update_mbuf(inp, m);
2677 in_pcbgroup_update(inp);
2682 in_pcbrehash(struct inpcb *inp)
2685 in_pcbrehash_mbuf(inp, NULL);
2689 * Remove PCB from various lists.
2692 in_pcbremlists(struct inpcb *inp)
2694 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2697 if (pcbinfo == &V_tcbinfo) {
2698 INP_INFO_RLOCK_ASSERT(pcbinfo);
2700 INP_INFO_WLOCK_ASSERT(pcbinfo);
2704 INP_WLOCK_ASSERT(inp);
2705 INP_LIST_WLOCK_ASSERT(pcbinfo);
2707 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2708 if (inp->inp_flags & INP_INHASHLIST) {
2709 struct inpcbport *phd = inp->inp_phd;
2711 INP_HASH_WLOCK(pcbinfo);
2713 /* XXX: Only do if SO_REUSEPORT_LB set? */
2714 in_pcbremlbgrouphash(inp);
2716 CK_LIST_REMOVE(inp, inp_hash);
2717 CK_LIST_REMOVE(inp, inp_portlist);
2718 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2719 CK_LIST_REMOVE(phd, phd_hash);
2720 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2722 INP_HASH_WUNLOCK(pcbinfo);
2723 inp->inp_flags &= ~INP_INHASHLIST;
2725 CK_LIST_REMOVE(inp, inp_list);
2726 pcbinfo->ipi_count--;
2728 in_pcbgroup_remove(inp);
2733 * Check for alternatives when higher level complains
2734 * about service problems. For now, invalidate cached
2735 * routing information. If the route was created dynamically
2736 * (by a redirect), time to try a default gateway again.
2739 in_losing(struct inpcb *inp)
2742 RO_INVALIDATE_CACHE(&inp->inp_route);
2747 * A set label operation has occurred at the socket layer, propagate the
2748 * label change into the in_pcb for the socket.
2751 in_pcbsosetlabel(struct socket *so)
2756 inp = sotoinpcb(so);
2757 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2761 mac_inpcb_sosetlabel(so, inp);
2768 * ipport_tick runs once per second, determining if random port allocation
2769 * should be continued. If more than ipport_randomcps ports have been
2770 * allocated in the last second, then we return to sequential port
2771 * allocation. We return to random allocation only once we drop below
2772 * ipport_randomcps for at least ipport_randomtime seconds.
2775 ipport_tick(void *xtp)
2777 VNET_ITERATOR_DECL(vnet_iter);
2779 VNET_LIST_RLOCK_NOSLEEP();
2780 VNET_FOREACH(vnet_iter) {
2781 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2782 if (V_ipport_tcpallocs <=
2783 V_ipport_tcplastcount + V_ipport_randomcps) {
2784 if (V_ipport_stoprandom > 0)
2785 V_ipport_stoprandom--;
2787 V_ipport_stoprandom = V_ipport_randomtime;
2788 V_ipport_tcplastcount = V_ipport_tcpallocs;
2791 VNET_LIST_RUNLOCK_NOSLEEP();
2792 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2799 callout_stop(&ipport_tick_callout);
2803 * The ipport_callout should start running at about the time we attach the
2804 * inet or inet6 domains.
2807 ipport_tick_init(const void *unused __unused)
2810 /* Start ipport_tick. */
2811 callout_init(&ipport_tick_callout, 1);
2812 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2813 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2814 SHUTDOWN_PRI_DEFAULT);
2816 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2817 ipport_tick_init, NULL);
2820 inp_wlock(struct inpcb *inp)
2827 inp_wunlock(struct inpcb *inp)
2834 inp_rlock(struct inpcb *inp)
2841 inp_runlock(struct inpcb *inp)
2847 #ifdef INVARIANT_SUPPORT
2849 inp_lock_assert(struct inpcb *inp)
2852 INP_WLOCK_ASSERT(inp);
2856 inp_unlock_assert(struct inpcb *inp)
2859 INP_UNLOCK_ASSERT(inp);
2864 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2868 INP_INFO_WLOCK(&V_tcbinfo);
2869 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2874 INP_INFO_WUNLOCK(&V_tcbinfo);
2878 inp_inpcbtosocket(struct inpcb *inp)
2881 INP_WLOCK_ASSERT(inp);
2882 return (inp->inp_socket);
2886 inp_inpcbtotcpcb(struct inpcb *inp)
2889 INP_WLOCK_ASSERT(inp);
2890 return ((struct tcpcb *)inp->inp_ppcb);
2894 inp_ip_tos_get(const struct inpcb *inp)
2897 return (inp->inp_ip_tos);
2901 inp_ip_tos_set(struct inpcb *inp, int val)
2904 inp->inp_ip_tos = val;
2908 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2909 uint32_t *faddr, uint16_t *fp)
2912 INP_LOCK_ASSERT(inp);
2913 *laddr = inp->inp_laddr.s_addr;
2914 *faddr = inp->inp_faddr.s_addr;
2915 *lp = inp->inp_lport;
2916 *fp = inp->inp_fport;
2920 so_sotoinpcb(struct socket *so)
2923 return (sotoinpcb(so));
2927 so_sototcpcb(struct socket *so)
2930 return (sototcpcb(so));
2934 * Create an external-format (``xinpcb'') structure using the information in
2935 * the kernel-format in_pcb structure pointed to by inp. This is done to
2936 * reduce the spew of irrelevant information over this interface, to isolate
2937 * user code from changes in the kernel structure, and potentially to provide
2938 * information-hiding if we decide that some of this information should be
2939 * hidden from users.
2942 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2945 bzero(xi, sizeof(*xi));
2946 xi->xi_len = sizeof(struct xinpcb);
2947 if (inp->inp_socket)
2948 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2949 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2950 xi->inp_gencnt = inp->inp_gencnt;
2951 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2952 xi->inp_flow = inp->inp_flow;
2953 xi->inp_flowid = inp->inp_flowid;
2954 xi->inp_flowtype = inp->inp_flowtype;
2955 xi->inp_flags = inp->inp_flags;
2956 xi->inp_flags2 = inp->inp_flags2;
2957 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2958 xi->in6p_cksum = inp->in6p_cksum;
2959 xi->in6p_hops = inp->in6p_hops;
2960 xi->inp_ip_tos = inp->inp_ip_tos;
2961 xi->inp_vflag = inp->inp_vflag;
2962 xi->inp_ip_ttl = inp->inp_ip_ttl;
2963 xi->inp_ip_p = inp->inp_ip_p;
2964 xi->inp_ip_minttl = inp->inp_ip_minttl;
2969 db_print_indent(int indent)
2973 for (i = 0; i < indent; i++)
2978 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2980 char faddr_str[48], laddr_str[48];
2982 db_print_indent(indent);
2983 db_printf("%s at %p\n", name, inc);
2988 if (inc->inc_flags & INC_ISIPV6) {
2990 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2991 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2996 inet_ntoa_r(inc->inc_laddr, laddr_str);
2997 inet_ntoa_r(inc->inc_faddr, faddr_str);
2999 db_print_indent(indent);
3000 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
3001 ntohs(inc->inc_lport));
3002 db_print_indent(indent);
3003 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
3004 ntohs(inc->inc_fport));
3008 db_print_inpflags(int inp_flags)
3013 if (inp_flags & INP_RECVOPTS) {
3014 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
3017 if (inp_flags & INP_RECVRETOPTS) {
3018 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
3021 if (inp_flags & INP_RECVDSTADDR) {
3022 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
3025 if (inp_flags & INP_ORIGDSTADDR) {
3026 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
3029 if (inp_flags & INP_HDRINCL) {
3030 db_printf("%sINP_HDRINCL", comma ? ", " : "");
3033 if (inp_flags & INP_HIGHPORT) {
3034 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3037 if (inp_flags & INP_LOWPORT) {
3038 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3041 if (inp_flags & INP_ANONPORT) {
3042 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3045 if (inp_flags & INP_RECVIF) {
3046 db_printf("%sINP_RECVIF", comma ? ", " : "");
3049 if (inp_flags & INP_MTUDISC) {
3050 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3053 if (inp_flags & INP_RECVTTL) {
3054 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3057 if (inp_flags & INP_DONTFRAG) {
3058 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3061 if (inp_flags & INP_RECVTOS) {
3062 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3065 if (inp_flags & IN6P_IPV6_V6ONLY) {
3066 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3069 if (inp_flags & IN6P_PKTINFO) {
3070 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3073 if (inp_flags & IN6P_HOPLIMIT) {
3074 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3077 if (inp_flags & IN6P_HOPOPTS) {
3078 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3081 if (inp_flags & IN6P_DSTOPTS) {
3082 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3085 if (inp_flags & IN6P_RTHDR) {
3086 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3089 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3090 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3093 if (inp_flags & IN6P_TCLASS) {
3094 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3097 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3098 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3101 if (inp_flags & INP_TIMEWAIT) {
3102 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3105 if (inp_flags & INP_ONESBCAST) {
3106 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3109 if (inp_flags & INP_DROPPED) {
3110 db_printf("%sINP_DROPPED", comma ? ", " : "");
3113 if (inp_flags & INP_SOCKREF) {
3114 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3117 if (inp_flags & IN6P_RFC2292) {
3118 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3121 if (inp_flags & IN6P_MTU) {
3122 db_printf("IN6P_MTU%s", comma ? ", " : "");
3128 db_print_inpvflag(u_char inp_vflag)
3133 if (inp_vflag & INP_IPV4) {
3134 db_printf("%sINP_IPV4", comma ? ", " : "");
3137 if (inp_vflag & INP_IPV6) {
3138 db_printf("%sINP_IPV6", comma ? ", " : "");
3141 if (inp_vflag & INP_IPV6PROTO) {
3142 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3148 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3151 db_print_indent(indent);
3152 db_printf("%s at %p\n", name, inp);
3156 db_print_indent(indent);
3157 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3159 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3161 db_print_indent(indent);
3162 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3163 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3165 db_print_indent(indent);
3166 db_printf("inp_label: %p inp_flags: 0x%x (",
3167 inp->inp_label, inp->inp_flags);
3168 db_print_inpflags(inp->inp_flags);
3171 db_print_indent(indent);
3172 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3174 db_print_inpvflag(inp->inp_vflag);
3177 db_print_indent(indent);
3178 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3179 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3181 db_print_indent(indent);
3183 if (inp->inp_vflag & INP_IPV6) {
3184 db_printf("in6p_options: %p in6p_outputopts: %p "
3185 "in6p_moptions: %p\n", inp->in6p_options,
3186 inp->in6p_outputopts, inp->in6p_moptions);
3187 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3188 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3193 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3194 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3195 inp->inp_options, inp->inp_moptions);
3198 db_print_indent(indent);
3199 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3200 (uintmax_t)inp->inp_gencnt);
3203 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3208 db_printf("usage: show inpcb <addr>\n");
3211 inp = (struct inpcb *)addr;
3213 db_print_inpcb(inp, "inpcb", 0);
3219 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3223 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3225 union if_snd_tag_modify_params params = {
3226 .rate_limit.max_rate = max_pacing_rate,
3228 struct m_snd_tag *mst;
3232 mst = inp->inp_snd_tag;
3240 if (ifp->if_snd_tag_modify == NULL) {
3243 error = ifp->if_snd_tag_modify(mst, ¶ms);
3249 * Query existing TX rate limit based on the existing
3250 * "inp->inp_snd_tag", if any.
3253 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3255 union if_snd_tag_query_params params = { };
3256 struct m_snd_tag *mst;
3260 mst = inp->inp_snd_tag;
3268 if (ifp->if_snd_tag_query == NULL) {
3271 error = ifp->if_snd_tag_query(mst, ¶ms);
3272 if (error == 0 && p_max_pacing_rate != NULL)
3273 *p_max_pacing_rate = params.rate_limit.max_rate;
3279 * Query existing TX queue level based on the existing
3280 * "inp->inp_snd_tag", if any.
3283 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3285 union if_snd_tag_query_params params = { };
3286 struct m_snd_tag *mst;
3290 mst = inp->inp_snd_tag;
3298 if (ifp->if_snd_tag_query == NULL)
3299 return (EOPNOTSUPP);
3301 error = ifp->if_snd_tag_query(mst, ¶ms);
3302 if (error == 0 && p_txqueue_level != NULL)
3303 *p_txqueue_level = params.rate_limit.queue_level;
3308 * Allocate a new TX rate limit send tag from the network interface
3309 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3312 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3313 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3315 union if_snd_tag_alloc_params params = {
3316 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3317 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3318 .rate_limit.hdr.flowid = flowid,
3319 .rate_limit.hdr.flowtype = flowtype,
3320 .rate_limit.max_rate = max_pacing_rate,
3324 INP_WLOCK_ASSERT(inp);
3327 * If there is already a send tag, or the INP is being torn
3328 * down, allocating a new send tag is not allowed. Else send
3331 if (inp->inp_snd_tag != NULL || (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0)
3334 if (ifp->if_snd_tag_alloc == NULL) {
3337 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3340 * At success increment the refcount on
3341 * the send tag's network interface:
3344 if_ref(inp->inp_snd_tag->ifp);
3350 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3354 in_pcbdetach_txrtlmt(struct inpcb *inp)
3356 struct m_snd_tag *mst;
3359 INP_WLOCK_ASSERT(inp);
3361 mst = inp->inp_snd_tag;
3362 inp->inp_snd_tag = NULL;
3372 * If the device was detached while we still had reference(s)
3373 * on the ifp, we assume if_snd_tag_free() was replaced with
3376 ifp->if_snd_tag_free(mst);
3378 /* release reference count on network interface */
3383 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3384 * is set in the fast path and will attach/detach/modify the TX rate
3385 * limit send tag based on the socket's so_max_pacing_rate value.
3388 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3390 struct socket *socket;
3391 uint32_t max_pacing_rate;
3398 socket = inp->inp_socket;
3402 if (!INP_WLOCKED(inp)) {
3404 * NOTE: If the write locking fails, we need to bail
3405 * out and use the non-ratelimited ring for the
3406 * transmit until there is a new chance to get the
3409 if (!INP_TRY_UPGRADE(inp))
3417 * NOTE: The so_max_pacing_rate value is read unlocked,
3418 * because atomic updates are not required since the variable
3419 * is checked at every mbuf we send. It is assumed that the
3420 * variable read itself will be atomic.
3422 max_pacing_rate = socket->so_max_pacing_rate;
3425 * NOTE: When attaching to a network interface a reference is
3426 * made to ensure the network interface doesn't go away until
3427 * all ratelimit connections are gone. The network interface
3428 * pointers compared below represent valid network interfaces,
3429 * except when comparing towards NULL.
3431 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3433 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3434 if (inp->inp_snd_tag != NULL)
3435 in_pcbdetach_txrtlmt(inp);
3437 } else if (inp->inp_snd_tag == NULL) {
3439 * In order to utilize packet pacing with RSS, we need
3440 * to wait until there is a valid RSS hash before we
3443 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3446 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3447 mb->m_pkthdr.flowid, max_pacing_rate);
3450 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3452 if (error == 0 || error == EOPNOTSUPP)
3453 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3459 * Track route changes for TX rate limiting.
3462 in_pcboutput_eagain(struct inpcb *inp)
3469 if (inp->inp_snd_tag == NULL)
3472 if (!INP_WLOCKED(inp)) {
3474 * NOTE: If the write locking fails, we need to bail
3475 * out and use the non-ratelimited ring for the
3476 * transmit until there is a new chance to get the
3479 if (!INP_TRY_UPGRADE(inp))
3486 /* detach rate limiting */
3487 in_pcbdetach_txrtlmt(inp);
3489 /* make sure new mbuf send tag allocation is made */
3490 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3495 #endif /* RATELIMIT */