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)
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 inp->inp_lport = lport;
1026 inp->inp_laddr.s_addr = laddr;
1027 if (in_pcbinshash(inp) != 0) {
1028 inp->inp_laddr.s_addr = INADDR_ANY;
1034 /* Commit the remaining changes. */
1035 inp->inp_lport = lport;
1036 inp->inp_laddr.s_addr = laddr;
1037 inp->inp_faddr.s_addr = faddr;
1038 inp->inp_fport = fport;
1039 in_pcbrehash_mbuf(inp, m);
1042 inp->inp_flags |= INP_ANONPORT;
1047 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1050 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1054 * Do proper source address selection on an unbound socket in case
1055 * of connect. Take jails into account as well.
1058 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1062 struct sockaddr *sa;
1063 struct sockaddr_in *sin;
1067 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1069 * Bypass source address selection and use the primary jail IP
1072 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1076 bzero(&sro, sizeof(sro));
1078 sin = (struct sockaddr_in *)&sro.ro_dst;
1079 sin->sin_family = AF_INET;
1080 sin->sin_len = sizeof(struct sockaddr_in);
1081 sin->sin_addr.s_addr = faddr->s_addr;
1084 * If route is known our src addr is taken from the i/f,
1087 * Find out route to destination.
1089 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1090 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1093 * If we found a route, use the address corresponding to
1094 * the outgoing interface.
1096 * Otherwise assume faddr is reachable on a directly connected
1097 * network and try to find a corresponding interface to take
1098 * the source address from.
1101 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1102 struct in_ifaddr *ia;
1105 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1106 inp->inp_socket->so_fibnum));
1108 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1109 inp->inp_socket->so_fibnum));
1113 error = ENETUNREACH;
1117 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1118 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1124 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1127 if (sa->sa_family != AF_INET)
1129 sin = (struct sockaddr_in *)sa;
1130 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1131 ia = (struct in_ifaddr *)ifa;
1136 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1140 /* 3. As a last resort return the 'default' jail address. */
1141 error = prison_get_ip4(cred, laddr);
1146 * If the outgoing interface on the route found is not
1147 * a loopback interface, use the address from that interface.
1148 * In case of jails do those three steps:
1149 * 1. check if the interface address belongs to the jail. If so use it.
1150 * 2. check if we have any address on the outgoing interface
1151 * belonging to this jail. If so use it.
1152 * 3. as a last resort return the 'default' jail address.
1154 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1155 struct in_ifaddr *ia;
1158 /* If not jailed, use the default returned. */
1159 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1160 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1161 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1166 /* 1. Check if the iface address belongs to the jail. */
1167 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1168 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1169 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1170 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1175 * 2. Check if we have any address on the outgoing interface
1176 * belonging to this jail.
1179 ifp = sro.ro_rt->rt_ifp;
1180 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1182 if (sa->sa_family != AF_INET)
1184 sin = (struct sockaddr_in *)sa;
1185 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1186 ia = (struct in_ifaddr *)ifa;
1191 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1195 /* 3. As a last resort return the 'default' jail address. */
1196 error = prison_get_ip4(cred, laddr);
1201 * The outgoing interface is marked with 'loopback net', so a route
1202 * to ourselves is here.
1203 * Try to find the interface of the destination address and then
1204 * take the address from there. That interface is not necessarily
1205 * a loopback interface.
1206 * In case of jails, check that it is an address of the jail
1207 * and if we cannot find, fall back to the 'default' jail address.
1209 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1210 struct sockaddr_in sain;
1211 struct in_ifaddr *ia;
1213 bzero(&sain, sizeof(struct sockaddr_in));
1214 sain.sin_family = AF_INET;
1215 sain.sin_len = sizeof(struct sockaddr_in);
1216 sain.sin_addr.s_addr = faddr->s_addr;
1218 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1219 inp->inp_socket->so_fibnum));
1221 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1222 inp->inp_socket->so_fibnum));
1224 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1226 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1228 error = ENETUNREACH;
1231 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1241 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1243 if (sa->sa_family != AF_INET)
1245 sin = (struct sockaddr_in *)sa;
1246 if (prison_check_ip4(cred,
1247 &sin->sin_addr) == 0) {
1248 ia = (struct in_ifaddr *)ifa;
1253 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1258 /* 3. As a last resort return the 'default' jail address. */
1259 error = prison_get_ip4(cred, laddr);
1265 if (sro.ro_rt != NULL)
1271 * Set up for a connect from a socket to the specified address.
1272 * On entry, *laddrp and *lportp should contain the current local
1273 * address and port for the PCB; these are updated to the values
1274 * that should be placed in inp_laddr and inp_lport to complete
1277 * On success, *faddrp and *fportp will be set to the remote address
1278 * and port. These are not updated in the error case.
1280 * If the operation fails because the connection already exists,
1281 * *oinpp will be set to the PCB of that connection so that the
1282 * caller can decide to override it. In all other cases, *oinpp
1286 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1287 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1288 struct inpcb **oinpp, struct ucred *cred)
1290 struct rm_priotracker in_ifa_tracker;
1291 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1292 struct in_ifaddr *ia;
1294 struct in_addr laddr, faddr;
1295 u_short lport, fport;
1299 * Because a global state change doesn't actually occur here, a read
1300 * lock is sufficient.
1302 INP_LOCK_ASSERT(inp);
1303 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1307 if (nam->sa_len != sizeof (*sin))
1309 if (sin->sin_family != AF_INET)
1310 return (EAFNOSUPPORT);
1311 if (sin->sin_port == 0)
1312 return (EADDRNOTAVAIL);
1313 laddr.s_addr = *laddrp;
1315 faddr = sin->sin_addr;
1316 fport = sin->sin_port;
1318 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1320 * If the destination address is INADDR_ANY,
1321 * use the primary local address.
1322 * If the supplied address is INADDR_BROADCAST,
1323 * and the primary interface supports broadcast,
1324 * choose the broadcast address for that interface.
1326 if (faddr.s_addr == INADDR_ANY) {
1327 IN_IFADDR_RLOCK(&in_ifa_tracker);
1329 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1330 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1332 (error = prison_get_ip4(cred, &faddr)) != 0)
1334 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1335 IN_IFADDR_RLOCK(&in_ifa_tracker);
1336 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1338 faddr = satosin(&CK_STAILQ_FIRST(
1339 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1340 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1343 if (laddr.s_addr == INADDR_ANY) {
1344 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1346 * If the destination address is multicast and an outgoing
1347 * interface has been set as a multicast option, prefer the
1348 * address of that interface as our source address.
1350 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1351 inp->inp_moptions != NULL) {
1352 struct ip_moptions *imo;
1355 imo = inp->inp_moptions;
1356 if (imo->imo_multicast_ifp != NULL) {
1357 ifp = imo->imo_multicast_ifp;
1358 IN_IFADDR_RLOCK(&in_ifa_tracker);
1359 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1360 if ((ia->ia_ifp == ifp) &&
1362 prison_check_ip4(cred,
1363 &ia->ia_addr.sin_addr) == 0))
1367 error = EADDRNOTAVAIL;
1369 laddr = ia->ia_addr.sin_addr;
1372 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1379 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1380 fport, laddr, lport, 0, NULL);
1384 return (EADDRINUSE);
1387 struct sockaddr_in lsin, fsin;
1389 bzero(&lsin, sizeof(lsin));
1390 bzero(&fsin, sizeof(fsin));
1391 lsin.sin_family = AF_INET;
1392 lsin.sin_addr = laddr;
1393 fsin.sin_family = AF_INET;
1394 fsin.sin_addr = faddr;
1395 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1396 &lport, (struct sockaddr *)& fsin, fport, cred,
1397 INPLOOKUP_WILDCARD);
1401 *laddrp = laddr.s_addr;
1403 *faddrp = faddr.s_addr;
1409 in_pcbdisconnect(struct inpcb *inp)
1412 INP_WLOCK_ASSERT(inp);
1413 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1415 inp->inp_faddr.s_addr = INADDR_ANY;
1422 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1423 * For most protocols, this will be invoked immediately prior to calling
1424 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1425 * socket, in which case in_pcbfree() is deferred.
1428 in_pcbdetach(struct inpcb *inp)
1431 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1434 if (inp->inp_snd_tag != NULL)
1435 in_pcbdetach_txrtlmt(inp);
1437 inp->inp_socket->so_pcb = NULL;
1438 inp->inp_socket = NULL;
1442 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1443 * stability of an inpcb pointer despite the inpcb lock being released. This
1444 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1445 * but where the inpcb lock may already held, or when acquiring a reference
1448 * in_pcbref() should be used only to provide brief memory stability, and
1449 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1450 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1451 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1452 * lock and rele are the *only* safe operations that may be performed on the
1455 * While the inpcb will not be freed, releasing the inpcb lock means that the
1456 * connection's state may change, so the caller should be careful to
1457 * revalidate any cached state on reacquiring the lock. Drop the reference
1458 * using in_pcbrele().
1461 in_pcbref(struct inpcb *inp)
1464 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1466 refcount_acquire(&inp->inp_refcount);
1470 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1471 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1472 * return a flag indicating whether or not the inpcb remains valid. If it is
1473 * valid, we return with the inpcb lock held.
1475 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1476 * reference on an inpcb. Historically more work was done here (actually, in
1477 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1478 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1479 * about memory stability (and continued use of the write lock).
1482 in_pcbrele_rlocked(struct inpcb *inp)
1484 struct inpcbinfo *pcbinfo;
1486 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1488 INP_RLOCK_ASSERT(inp);
1490 if (refcount_release(&inp->inp_refcount) == 0) {
1492 * If the inpcb has been freed, let the caller know, even if
1493 * this isn't the last reference.
1495 if (inp->inp_flags2 & INP_FREED) {
1502 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1504 if (inp->inp_in_hpts || inp->inp_in_input) {
1505 struct tcp_hpts_entry *hpts;
1507 * We should not be on the hpts at
1508 * this point in any form. we must
1509 * get the lock to be sure.
1511 hpts = tcp_hpts_lock(inp);
1512 if (inp->inp_in_hpts)
1513 panic("Hpts:%p inp:%p at free still on hpts",
1515 mtx_unlock(&hpts->p_mtx);
1516 hpts = tcp_input_lock(inp);
1517 if (inp->inp_in_input)
1518 panic("Hpts:%p inp:%p at free still on input hpts",
1520 mtx_unlock(&hpts->p_mtx);
1524 pcbinfo = inp->inp_pcbinfo;
1525 uma_zfree(pcbinfo->ipi_zone, inp);
1530 in_pcbrele_wlocked(struct inpcb *inp)
1532 struct inpcbinfo *pcbinfo;
1534 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1536 INP_WLOCK_ASSERT(inp);
1538 if (refcount_release(&inp->inp_refcount) == 0) {
1540 * If the inpcb has been freed, let the caller know, even if
1541 * this isn't the last reference.
1543 if (inp->inp_flags2 & INP_FREED) {
1550 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1552 if (inp->inp_in_hpts || inp->inp_in_input) {
1553 struct tcp_hpts_entry *hpts;
1555 * We should not be on the hpts at
1556 * this point in any form. we must
1557 * get the lock to be sure.
1559 hpts = tcp_hpts_lock(inp);
1560 if (inp->inp_in_hpts)
1561 panic("Hpts:%p inp:%p at free still on hpts",
1563 mtx_unlock(&hpts->p_mtx);
1564 hpts = tcp_input_lock(inp);
1565 if (inp->inp_in_input)
1566 panic("Hpts:%p inp:%p at free still on input hpts",
1568 mtx_unlock(&hpts->p_mtx);
1572 pcbinfo = inp->inp_pcbinfo;
1573 uma_zfree(pcbinfo->ipi_zone, inp);
1578 * Temporary wrapper.
1581 in_pcbrele(struct inpcb *inp)
1584 return (in_pcbrele_wlocked(inp));
1588 in_pcblist_rele_rlocked(epoch_context_t ctx)
1590 struct in_pcblist *il;
1592 struct inpcbinfo *pcbinfo;
1595 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1596 pcbinfo = il->il_pcbinfo;
1598 INP_INFO_WLOCK(pcbinfo);
1599 for (i = 0; i < n; i++) {
1600 inp = il->il_inp_list[i];
1602 if (!in_pcbrele_rlocked(inp))
1605 INP_INFO_WUNLOCK(pcbinfo);
1610 inpcbport_free(epoch_context_t ctx)
1612 struct inpcbport *phd;
1614 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1619 in_pcbfree_deferred(epoch_context_t ctx)
1622 int released __unused;
1624 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1627 CURVNET_SET(inp->inp_vnet);
1629 struct ip_moptions *imo = inp->inp_moptions;
1630 inp->inp_moptions = NULL;
1632 /* XXXRW: Do as much as possible here. */
1633 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1634 if (inp->inp_sp != NULL)
1635 ipsec_delete_pcbpolicy(inp);
1638 struct ip6_moptions *im6o = NULL;
1639 if (inp->inp_vflag & INP_IPV6PROTO) {
1640 ip6_freepcbopts(inp->in6p_outputopts);
1641 im6o = inp->in6p_moptions;
1642 inp->in6p_moptions = NULL;
1645 if (inp->inp_options)
1646 (void)m_free(inp->inp_options);
1648 crfree(inp->inp_cred);
1650 mac_inpcb_destroy(inp);
1652 released = in_pcbrele_wlocked(inp);
1655 ip6_freemoptions(im6o);
1658 inp_freemoptions(imo);
1664 * Unconditionally schedule an inpcb to be freed by decrementing its
1665 * reference count, which should occur only after the inpcb has been detached
1666 * from its socket. If another thread holds a temporary reference (acquired
1667 * using in_pcbref()) then the free is deferred until that reference is
1668 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1669 * work, including removal from global lists, is done in this context, where
1670 * the pcbinfo lock is held.
1673 in_pcbfree(struct inpcb *inp)
1675 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1677 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1678 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1679 ("%s: called twice for pcb %p", __func__, inp));
1680 if (inp->inp_flags2 & INP_FREED) {
1686 if (pcbinfo == &V_tcbinfo) {
1687 INP_INFO_LOCK_ASSERT(pcbinfo);
1689 INP_INFO_WLOCK_ASSERT(pcbinfo);
1692 INP_WLOCK_ASSERT(inp);
1693 INP_LIST_WLOCK(pcbinfo);
1694 in_pcbremlists(inp);
1695 INP_LIST_WUNLOCK(pcbinfo);
1696 RO_INVALIDATE_CACHE(&inp->inp_route);
1697 /* mark as destruction in progress */
1698 inp->inp_flags2 |= INP_FREED;
1700 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1704 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1705 * port reservation, and preventing it from being returned by inpcb lookups.
1707 * It is used by TCP to mark an inpcb as unused and avoid future packet
1708 * delivery or event notification when a socket remains open but TCP has
1709 * closed. This might occur as a result of a shutdown()-initiated TCP close
1710 * or a RST on the wire, and allows the port binding to be reused while still
1711 * maintaining the invariant that so_pcb always points to a valid inpcb until
1714 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1715 * in_pcbnotifyall() and in_pcbpurgeif0()?
1718 in_pcbdrop(struct inpcb *inp)
1721 INP_WLOCK_ASSERT(inp);
1723 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1724 MPASS(inp->inp_refcount > 1);
1728 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1731 inp->inp_flags |= INP_DROPPED;
1732 if (inp->inp_flags & INP_INHASHLIST) {
1733 struct inpcbport *phd = inp->inp_phd;
1735 INP_HASH_WLOCK(inp->inp_pcbinfo);
1736 in_pcbremlbgrouphash(inp);
1737 CK_LIST_REMOVE(inp, inp_hash);
1738 CK_LIST_REMOVE(inp, inp_portlist);
1739 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1740 CK_LIST_REMOVE(phd, phd_hash);
1741 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1743 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1744 inp->inp_flags &= ~INP_INHASHLIST;
1746 in_pcbgroup_remove(inp);
1753 * Common routines to return the socket addresses associated with inpcbs.
1756 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1758 struct sockaddr_in *sin;
1760 sin = malloc(sizeof *sin, M_SONAME,
1762 sin->sin_family = AF_INET;
1763 sin->sin_len = sizeof(*sin);
1764 sin->sin_addr = *addr_p;
1765 sin->sin_port = port;
1767 return (struct sockaddr *)sin;
1771 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1774 struct in_addr addr;
1777 inp = sotoinpcb(so);
1778 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1781 port = inp->inp_lport;
1782 addr = inp->inp_laddr;
1785 *nam = in_sockaddr(port, &addr);
1790 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1793 struct in_addr addr;
1796 inp = sotoinpcb(so);
1797 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1800 port = inp->inp_fport;
1801 addr = inp->inp_faddr;
1804 *nam = in_sockaddr(port, &addr);
1809 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1810 struct inpcb *(*notify)(struct inpcb *, int))
1812 struct inpcb *inp, *inp_temp;
1814 INP_INFO_WLOCK(pcbinfo);
1815 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1818 if ((inp->inp_vflag & INP_IPV4) == 0) {
1823 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1824 inp->inp_socket == NULL) {
1828 if ((*notify)(inp, errno))
1831 INP_INFO_WUNLOCK(pcbinfo);
1835 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1838 struct in_multi *inm;
1839 struct in_mfilter *imf;
1840 struct ip_moptions *imo;
1842 INP_INFO_WLOCK(pcbinfo);
1843 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1845 imo = inp->inp_moptions;
1846 if ((inp->inp_vflag & INP_IPV4) &&
1849 * Unselect the outgoing interface if it is being
1852 if (imo->imo_multicast_ifp == ifp)
1853 imo->imo_multicast_ifp = NULL;
1856 * Drop multicast group membership if we joined
1857 * through the interface being detached.
1859 * XXX This can all be deferred to an epoch_call
1862 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1863 if ((inm = imf->imf_inm) == NULL)
1865 if (inm->inm_ifp != ifp)
1867 ip_mfilter_remove(&imo->imo_head, imf);
1868 IN_MULTI_LOCK_ASSERT();
1869 in_leavegroup_locked(inm, NULL);
1870 ip_mfilter_free(imf);
1876 INP_INFO_WUNLOCK(pcbinfo);
1880 * Lookup a PCB based on the local address and port. Caller must hold the
1881 * hash lock. No inpcb locks or references are acquired.
1883 #define INP_LOOKUP_MAPPED_PCB_COST 3
1885 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1886 u_short lport, int lookupflags, struct ucred *cred)
1890 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1896 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1897 ("%s: invalid lookup flags %d", __func__, lookupflags));
1899 INP_HASH_LOCK_ASSERT(pcbinfo);
1901 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1902 struct inpcbhead *head;
1904 * Look for an unconnected (wildcard foreign addr) PCB that
1905 * matches the local address and port we're looking for.
1907 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1908 0, pcbinfo->ipi_hashmask)];
1909 CK_LIST_FOREACH(inp, head, inp_hash) {
1911 /* XXX inp locking */
1912 if ((inp->inp_vflag & INP_IPV4) == 0)
1915 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1916 inp->inp_laddr.s_addr == laddr.s_addr &&
1917 inp->inp_lport == lport) {
1922 prison_equal_ip4(cred->cr_prison,
1923 inp->inp_cred->cr_prison))
1932 struct inpcbporthead *porthash;
1933 struct inpcbport *phd;
1934 struct inpcb *match = NULL;
1936 * Best fit PCB lookup.
1938 * First see if this local port is in use by looking on the
1941 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1942 pcbinfo->ipi_porthashmask)];
1943 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1944 if (phd->phd_port == lport)
1949 * Port is in use by one or more PCBs. Look for best
1952 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1955 !prison_equal_ip4(inp->inp_cred->cr_prison,
1959 /* XXX inp locking */
1960 if ((inp->inp_vflag & INP_IPV4) == 0)
1963 * We never select the PCB that has
1964 * INP_IPV6 flag and is bound to :: if
1965 * we have another PCB which is bound
1966 * to 0.0.0.0. If a PCB has the
1967 * INP_IPV6 flag, then we set its cost
1968 * higher than IPv4 only PCBs.
1970 * Note that the case only happens
1971 * when a socket is bound to ::, under
1972 * the condition that the use of the
1973 * mapped address is allowed.
1975 if ((inp->inp_vflag & INP_IPV6) != 0)
1976 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1978 if (inp->inp_faddr.s_addr != INADDR_ANY)
1980 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1981 if (laddr.s_addr == INADDR_ANY)
1983 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1986 if (laddr.s_addr != INADDR_ANY)
1989 if (wildcard < matchwild) {
1991 matchwild = wildcard;
2000 #undef INP_LOOKUP_MAPPED_PCB_COST
2002 static struct inpcb *
2003 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2004 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
2005 uint16_t fport, int lookupflags)
2007 struct inpcb *local_wild;
2008 const struct inpcblbgrouphead *hdr;
2009 struct inpcblbgroup *grp;
2012 INP_HASH_LOCK_ASSERT(pcbinfo);
2014 hdr = &pcbinfo->ipi_lbgrouphashbase[
2015 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2018 * Order of socket selection:
2020 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
2023 * - Load balanced group does not contain jailed sockets
2024 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
2027 CK_LIST_FOREACH(grp, hdr, il_list) {
2029 if (!(grp->il_vflag & INP_IPV4))
2032 if (grp->il_lport != lport)
2035 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
2037 if (grp->il_laddr.s_addr == laddr->s_addr)
2038 return (grp->il_inp[idx]);
2039 if (grp->il_laddr.s_addr == INADDR_ANY &&
2040 (lookupflags & INPLOOKUP_WILDCARD) != 0)
2041 local_wild = grp->il_inp[idx];
2043 return (local_wild);
2048 * Lookup PCB in hash list, using pcbgroup tables.
2050 static struct inpcb *
2051 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2052 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2053 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2055 struct inpcbhead *head;
2056 struct inpcb *inp, *tmpinp;
2057 u_short fport = fport_arg, lport = lport_arg;
2061 * First look for an exact match.
2064 INP_GROUP_LOCK(pcbgroup);
2065 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2066 pcbgroup->ipg_hashmask)];
2067 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2069 /* XXX inp locking */
2070 if ((inp->inp_vflag & INP_IPV4) == 0)
2073 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2074 inp->inp_laddr.s_addr == laddr.s_addr &&
2075 inp->inp_fport == fport &&
2076 inp->inp_lport == lport) {
2078 * XXX We should be able to directly return
2079 * the inp here, without any checks.
2080 * Well unless both bound with SO_REUSEPORT?
2082 if (prison_flag(inp->inp_cred, PR_IP4))
2088 if (tmpinp != NULL) {
2095 * For incoming connections, we may wish to do a wildcard
2096 * match for an RSS-local socket.
2098 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2099 struct inpcb *local_wild = NULL, *local_exact = NULL;
2101 struct inpcb *local_wild_mapped = NULL;
2103 struct inpcb *jail_wild = NULL;
2104 struct inpcbhead *head;
2108 * Order of socket selection - we always prefer jails.
2109 * 1. jailed, non-wild.
2111 * 3. non-jailed, non-wild.
2112 * 4. non-jailed, wild.
2115 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2116 lport, 0, pcbgroup->ipg_hashmask)];
2117 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2119 /* XXX inp locking */
2120 if ((inp->inp_vflag & INP_IPV4) == 0)
2123 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2124 inp->inp_lport != lport)
2127 injail = prison_flag(inp->inp_cred, PR_IP4);
2129 if (prison_check_ip4(inp->inp_cred,
2133 if (local_exact != NULL)
2137 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2142 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2144 /* XXX inp locking, NULL check */
2145 if (inp->inp_vflag & INP_IPV6PROTO)
2146 local_wild_mapped = inp;
2154 } /* LIST_FOREACH */
2163 inp = local_wild_mapped;
2171 * Then look for a wildcard match, if requested.
2173 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2174 struct inpcb *local_wild = NULL, *local_exact = NULL;
2176 struct inpcb *local_wild_mapped = NULL;
2178 struct inpcb *jail_wild = NULL;
2179 struct inpcbhead *head;
2183 * Order of socket selection - we always prefer jails.
2184 * 1. jailed, non-wild.
2186 * 3. non-jailed, non-wild.
2187 * 4. non-jailed, wild.
2189 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2190 0, pcbinfo->ipi_wildmask)];
2191 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2193 /* XXX inp locking */
2194 if ((inp->inp_vflag & INP_IPV4) == 0)
2197 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2198 inp->inp_lport != lport)
2201 injail = prison_flag(inp->inp_cred, PR_IP4);
2203 if (prison_check_ip4(inp->inp_cred,
2207 if (local_exact != NULL)
2211 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2216 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2218 /* XXX inp locking, NULL check */
2219 if (inp->inp_vflag & INP_IPV6PROTO)
2220 local_wild_mapped = inp;
2228 } /* LIST_FOREACH */
2236 inp = local_wild_mapped;
2240 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2241 INP_GROUP_UNLOCK(pcbgroup);
2245 if (lookupflags & INPLOOKUP_WLOCKPCB)
2246 locked = INP_TRY_WLOCK(inp);
2247 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2248 locked = INP_TRY_RLOCK(inp);
2250 panic("%s: locking bug", __func__);
2251 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2252 if (lookupflags & INPLOOKUP_WLOCKPCB)
2259 INP_GROUP_UNLOCK(pcbgroup);
2261 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2263 if (in_pcbrele_wlocked(inp))
2267 if (in_pcbrele_rlocked(inp))
2272 if (lookupflags & INPLOOKUP_WLOCKPCB)
2273 INP_WLOCK_ASSERT(inp);
2275 INP_RLOCK_ASSERT(inp);
2279 #endif /* PCBGROUP */
2282 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2283 * that the caller has locked the hash list, and will not perform any further
2284 * locking or reference operations on either the hash list or the connection.
2286 static struct inpcb *
2287 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2288 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2291 struct inpcbhead *head;
2292 struct inpcb *inp, *tmpinp;
2293 u_short fport = fport_arg, lport = lport_arg;
2296 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2297 ("%s: invalid lookup flags %d", __func__, lookupflags));
2298 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2299 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2302 * First look for an exact match.
2305 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2306 pcbinfo->ipi_hashmask)];
2307 CK_LIST_FOREACH(inp, head, inp_hash) {
2309 /* XXX inp locking */
2310 if ((inp->inp_vflag & INP_IPV4) == 0)
2313 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2314 inp->inp_laddr.s_addr == laddr.s_addr &&
2315 inp->inp_fport == fport &&
2316 inp->inp_lport == lport) {
2318 * XXX We should be able to directly return
2319 * the inp here, without any checks.
2320 * Well unless both bound with SO_REUSEPORT?
2322 if (prison_flag(inp->inp_cred, PR_IP4))
2332 * Then look in lb group (for wildcard match).
2334 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2335 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2336 fport, lookupflags);
2342 * Then look for a wildcard match, if requested.
2344 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2345 struct inpcb *local_wild = NULL, *local_exact = NULL;
2347 struct inpcb *local_wild_mapped = NULL;
2349 struct inpcb *jail_wild = NULL;
2353 * Order of socket selection - we always prefer jails.
2354 * 1. jailed, non-wild.
2356 * 3. non-jailed, non-wild.
2357 * 4. non-jailed, wild.
2360 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2361 0, pcbinfo->ipi_hashmask)];
2362 CK_LIST_FOREACH(inp, head, inp_hash) {
2364 /* XXX inp locking */
2365 if ((inp->inp_vflag & INP_IPV4) == 0)
2368 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2369 inp->inp_lport != lport)
2372 injail = prison_flag(inp->inp_cred, PR_IP4);
2374 if (prison_check_ip4(inp->inp_cred,
2378 if (local_exact != NULL)
2382 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2387 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2389 /* XXX inp locking, NULL check */
2390 if (inp->inp_vflag & INP_IPV6PROTO)
2391 local_wild_mapped = inp;
2399 } /* LIST_FOREACH */
2400 if (jail_wild != NULL)
2402 if (local_exact != NULL)
2403 return (local_exact);
2404 if (local_wild != NULL)
2405 return (local_wild);
2407 if (local_wild_mapped != NULL)
2408 return (local_wild_mapped);
2410 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2416 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2417 * hash list lock, and will return the inpcb locked (i.e., requires
2418 * INPLOOKUP_LOCKPCB).
2420 static struct inpcb *
2421 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2422 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2427 INP_HASH_RLOCK(pcbinfo);
2428 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2429 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2431 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2433 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2437 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2439 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2444 panic("%s: locking bug", __func__);
2447 if (lookupflags & INPLOOKUP_WLOCKPCB)
2448 INP_WLOCK_ASSERT(inp);
2450 INP_RLOCK_ASSERT(inp);
2454 INP_HASH_RUNLOCK(pcbinfo);
2459 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2460 * from which a pre-calculated hash value may be extracted.
2462 * Possibly more of this logic should be in in_pcbgroup.c.
2465 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2466 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2468 #if defined(PCBGROUP) && !defined(RSS)
2469 struct inpcbgroup *pcbgroup;
2472 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2473 ("%s: invalid lookup flags %d", __func__, lookupflags));
2474 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2475 ("%s: LOCKPCB not set", __func__));
2478 * When not using RSS, use connection groups in preference to the
2479 * reservation table when looking up 4-tuples. When using RSS, just
2480 * use the reservation table, due to the cost of the Toeplitz hash
2483 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2484 * we could be doing RSS with a non-Toeplitz hash that is affordable
2487 #if defined(PCBGROUP) && !defined(RSS)
2488 if (in_pcbgroup_enabled(pcbinfo)) {
2489 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2491 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2492 laddr, lport, lookupflags, ifp));
2495 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2500 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2501 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2502 struct ifnet *ifp, struct mbuf *m)
2505 struct inpcbgroup *pcbgroup;
2508 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2509 ("%s: invalid lookup flags %d", __func__, lookupflags));
2510 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2511 ("%s: LOCKPCB not set", __func__));
2515 * If we can use a hardware-generated hash to look up the connection
2516 * group, use that connection group to find the inpcb. Otherwise
2517 * fall back on a software hash -- or the reservation table if we're
2520 * XXXRW: As above, that policy belongs in the pcbgroup code.
2522 if (in_pcbgroup_enabled(pcbinfo) &&
2523 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2524 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2525 m->m_pkthdr.flowid);
2526 if (pcbgroup != NULL)
2527 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2528 fport, laddr, lport, lookupflags, ifp));
2530 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2532 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2533 laddr, lport, lookupflags, ifp));
2537 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2543 * Insert PCB onto various hash lists.
2546 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2548 struct inpcbhead *pcbhash;
2549 struct inpcbporthead *pcbporthash;
2550 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2551 struct inpcbport *phd;
2552 u_int32_t hashkey_faddr;
2555 INP_WLOCK_ASSERT(inp);
2556 INP_HASH_WLOCK_ASSERT(pcbinfo);
2558 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2559 ("in_pcbinshash: INP_INHASHLIST"));
2562 if (inp->inp_vflag & INP_IPV6)
2563 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2566 hashkey_faddr = inp->inp_faddr.s_addr;
2568 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2569 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2571 pcbporthash = &pcbinfo->ipi_porthashbase[
2572 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2575 * Add entry to load balance group.
2576 * Only do this if SO_REUSEPORT_LB is set.
2578 so_options = inp_so_options(inp);
2579 if (so_options & SO_REUSEPORT_LB) {
2580 int ret = in_pcbinslbgrouphash(inp);
2582 /* pcb lb group malloc fail (ret=ENOBUFS). */
2588 * Go through port list and look for a head for this lport.
2590 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2591 if (phd->phd_port == inp->inp_lport)
2595 * If none exists, malloc one and tack it on.
2598 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2600 return (ENOBUFS); /* XXX */
2602 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2603 phd->phd_port = inp->inp_lport;
2604 CK_LIST_INIT(&phd->phd_pcblist);
2605 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2608 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2609 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2610 inp->inp_flags |= INP_INHASHLIST;
2612 if (do_pcbgroup_update)
2613 in_pcbgroup_update(inp);
2619 * For now, there are two public interfaces to insert an inpcb into the hash
2620 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2621 * is used only in the TCP syncache, where in_pcbinshash is called before the
2622 * full 4-tuple is set for the inpcb, and we don't want to install in the
2623 * pcbgroup until later.
2625 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2626 * connection groups, and partially initialised inpcbs should not be exposed
2627 * to either reservation hash tables or pcbgroups.
2630 in_pcbinshash(struct inpcb *inp)
2633 return (in_pcbinshash_internal(inp, 1));
2637 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2640 return (in_pcbinshash_internal(inp, 0));
2644 * Move PCB to the proper hash bucket when { faddr, fport } have been
2645 * changed. NOTE: This does not handle the case of the lport changing (the
2646 * hashed port list would have to be updated as well), so the lport must
2647 * not change after in_pcbinshash() has been called.
2650 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2652 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2653 struct inpcbhead *head;
2654 u_int32_t hashkey_faddr;
2656 INP_WLOCK_ASSERT(inp);
2657 INP_HASH_WLOCK_ASSERT(pcbinfo);
2659 KASSERT(inp->inp_flags & INP_INHASHLIST,
2660 ("in_pcbrehash: !INP_INHASHLIST"));
2663 if (inp->inp_vflag & INP_IPV6)
2664 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2667 hashkey_faddr = inp->inp_faddr.s_addr;
2669 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2670 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2672 CK_LIST_REMOVE(inp, inp_hash);
2673 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2677 in_pcbgroup_update_mbuf(inp, m);
2679 in_pcbgroup_update(inp);
2684 in_pcbrehash(struct inpcb *inp)
2687 in_pcbrehash_mbuf(inp, NULL);
2691 * Remove PCB from various lists.
2694 in_pcbremlists(struct inpcb *inp)
2696 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2699 if (pcbinfo == &V_tcbinfo) {
2700 INP_INFO_RLOCK_ASSERT(pcbinfo);
2702 INP_INFO_WLOCK_ASSERT(pcbinfo);
2706 INP_WLOCK_ASSERT(inp);
2707 INP_LIST_WLOCK_ASSERT(pcbinfo);
2709 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2710 if (inp->inp_flags & INP_INHASHLIST) {
2711 struct inpcbport *phd = inp->inp_phd;
2713 INP_HASH_WLOCK(pcbinfo);
2715 /* XXX: Only do if SO_REUSEPORT_LB set? */
2716 in_pcbremlbgrouphash(inp);
2718 CK_LIST_REMOVE(inp, inp_hash);
2719 CK_LIST_REMOVE(inp, inp_portlist);
2720 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2721 CK_LIST_REMOVE(phd, phd_hash);
2722 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2724 INP_HASH_WUNLOCK(pcbinfo);
2725 inp->inp_flags &= ~INP_INHASHLIST;
2727 CK_LIST_REMOVE(inp, inp_list);
2728 pcbinfo->ipi_count--;
2730 in_pcbgroup_remove(inp);
2735 * Check for alternatives when higher level complains
2736 * about service problems. For now, invalidate cached
2737 * routing information. If the route was created dynamically
2738 * (by a redirect), time to try a default gateway again.
2741 in_losing(struct inpcb *inp)
2744 RO_INVALIDATE_CACHE(&inp->inp_route);
2749 * A set label operation has occurred at the socket layer, propagate the
2750 * label change into the in_pcb for the socket.
2753 in_pcbsosetlabel(struct socket *so)
2758 inp = sotoinpcb(so);
2759 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2763 mac_inpcb_sosetlabel(so, inp);
2770 * ipport_tick runs once per second, determining if random port allocation
2771 * should be continued. If more than ipport_randomcps ports have been
2772 * allocated in the last second, then we return to sequential port
2773 * allocation. We return to random allocation only once we drop below
2774 * ipport_randomcps for at least ipport_randomtime seconds.
2777 ipport_tick(void *xtp)
2779 VNET_ITERATOR_DECL(vnet_iter);
2781 VNET_LIST_RLOCK_NOSLEEP();
2782 VNET_FOREACH(vnet_iter) {
2783 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2784 if (V_ipport_tcpallocs <=
2785 V_ipport_tcplastcount + V_ipport_randomcps) {
2786 if (V_ipport_stoprandom > 0)
2787 V_ipport_stoprandom--;
2789 V_ipport_stoprandom = V_ipport_randomtime;
2790 V_ipport_tcplastcount = V_ipport_tcpallocs;
2793 VNET_LIST_RUNLOCK_NOSLEEP();
2794 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2801 callout_stop(&ipport_tick_callout);
2805 * The ipport_callout should start running at about the time we attach the
2806 * inet or inet6 domains.
2809 ipport_tick_init(const void *unused __unused)
2812 /* Start ipport_tick. */
2813 callout_init(&ipport_tick_callout, 1);
2814 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2815 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2816 SHUTDOWN_PRI_DEFAULT);
2818 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2819 ipport_tick_init, NULL);
2822 inp_wlock(struct inpcb *inp)
2829 inp_wunlock(struct inpcb *inp)
2836 inp_rlock(struct inpcb *inp)
2843 inp_runlock(struct inpcb *inp)
2849 #ifdef INVARIANT_SUPPORT
2851 inp_lock_assert(struct inpcb *inp)
2854 INP_WLOCK_ASSERT(inp);
2858 inp_unlock_assert(struct inpcb *inp)
2861 INP_UNLOCK_ASSERT(inp);
2866 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2870 INP_INFO_WLOCK(&V_tcbinfo);
2871 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2876 INP_INFO_WUNLOCK(&V_tcbinfo);
2880 inp_inpcbtosocket(struct inpcb *inp)
2883 INP_WLOCK_ASSERT(inp);
2884 return (inp->inp_socket);
2888 inp_inpcbtotcpcb(struct inpcb *inp)
2891 INP_WLOCK_ASSERT(inp);
2892 return ((struct tcpcb *)inp->inp_ppcb);
2896 inp_ip_tos_get(const struct inpcb *inp)
2899 return (inp->inp_ip_tos);
2903 inp_ip_tos_set(struct inpcb *inp, int val)
2906 inp->inp_ip_tos = val;
2910 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2911 uint32_t *faddr, uint16_t *fp)
2914 INP_LOCK_ASSERT(inp);
2915 *laddr = inp->inp_laddr.s_addr;
2916 *faddr = inp->inp_faddr.s_addr;
2917 *lp = inp->inp_lport;
2918 *fp = inp->inp_fport;
2922 so_sotoinpcb(struct socket *so)
2925 return (sotoinpcb(so));
2929 so_sototcpcb(struct socket *so)
2932 return (sototcpcb(so));
2936 * Create an external-format (``xinpcb'') structure using the information in
2937 * the kernel-format in_pcb structure pointed to by inp. This is done to
2938 * reduce the spew of irrelevant information over this interface, to isolate
2939 * user code from changes in the kernel structure, and potentially to provide
2940 * information-hiding if we decide that some of this information should be
2941 * hidden from users.
2944 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2947 bzero(xi, sizeof(*xi));
2948 xi->xi_len = sizeof(struct xinpcb);
2949 if (inp->inp_socket)
2950 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2951 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2952 xi->inp_gencnt = inp->inp_gencnt;
2953 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2954 xi->inp_flow = inp->inp_flow;
2955 xi->inp_flowid = inp->inp_flowid;
2956 xi->inp_flowtype = inp->inp_flowtype;
2957 xi->inp_flags = inp->inp_flags;
2958 xi->inp_flags2 = inp->inp_flags2;
2959 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2960 xi->in6p_cksum = inp->in6p_cksum;
2961 xi->in6p_hops = inp->in6p_hops;
2962 xi->inp_ip_tos = inp->inp_ip_tos;
2963 xi->inp_vflag = inp->inp_vflag;
2964 xi->inp_ip_ttl = inp->inp_ip_ttl;
2965 xi->inp_ip_p = inp->inp_ip_p;
2966 xi->inp_ip_minttl = inp->inp_ip_minttl;
2971 db_print_indent(int indent)
2975 for (i = 0; i < indent; i++)
2980 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2982 char faddr_str[48], laddr_str[48];
2984 db_print_indent(indent);
2985 db_printf("%s at %p\n", name, inc);
2990 if (inc->inc_flags & INC_ISIPV6) {
2992 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2993 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2998 inet_ntoa_r(inc->inc_laddr, laddr_str);
2999 inet_ntoa_r(inc->inc_faddr, faddr_str);
3001 db_print_indent(indent);
3002 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
3003 ntohs(inc->inc_lport));
3004 db_print_indent(indent);
3005 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
3006 ntohs(inc->inc_fport));
3010 db_print_inpflags(int inp_flags)
3015 if (inp_flags & INP_RECVOPTS) {
3016 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
3019 if (inp_flags & INP_RECVRETOPTS) {
3020 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
3023 if (inp_flags & INP_RECVDSTADDR) {
3024 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
3027 if (inp_flags & INP_ORIGDSTADDR) {
3028 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
3031 if (inp_flags & INP_HDRINCL) {
3032 db_printf("%sINP_HDRINCL", comma ? ", " : "");
3035 if (inp_flags & INP_HIGHPORT) {
3036 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3039 if (inp_flags & INP_LOWPORT) {
3040 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3043 if (inp_flags & INP_ANONPORT) {
3044 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3047 if (inp_flags & INP_RECVIF) {
3048 db_printf("%sINP_RECVIF", comma ? ", " : "");
3051 if (inp_flags & INP_MTUDISC) {
3052 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3055 if (inp_flags & INP_RECVTTL) {
3056 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3059 if (inp_flags & INP_DONTFRAG) {
3060 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3063 if (inp_flags & INP_RECVTOS) {
3064 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3067 if (inp_flags & IN6P_IPV6_V6ONLY) {
3068 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3071 if (inp_flags & IN6P_PKTINFO) {
3072 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3075 if (inp_flags & IN6P_HOPLIMIT) {
3076 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3079 if (inp_flags & IN6P_HOPOPTS) {
3080 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3083 if (inp_flags & IN6P_DSTOPTS) {
3084 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3087 if (inp_flags & IN6P_RTHDR) {
3088 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3091 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3092 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3095 if (inp_flags & IN6P_TCLASS) {
3096 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3099 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3100 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3103 if (inp_flags & INP_TIMEWAIT) {
3104 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3107 if (inp_flags & INP_ONESBCAST) {
3108 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3111 if (inp_flags & INP_DROPPED) {
3112 db_printf("%sINP_DROPPED", comma ? ", " : "");
3115 if (inp_flags & INP_SOCKREF) {
3116 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3119 if (inp_flags & IN6P_RFC2292) {
3120 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3123 if (inp_flags & IN6P_MTU) {
3124 db_printf("IN6P_MTU%s", comma ? ", " : "");
3130 db_print_inpvflag(u_char inp_vflag)
3135 if (inp_vflag & INP_IPV4) {
3136 db_printf("%sINP_IPV4", comma ? ", " : "");
3139 if (inp_vflag & INP_IPV6) {
3140 db_printf("%sINP_IPV6", comma ? ", " : "");
3143 if (inp_vflag & INP_IPV6PROTO) {
3144 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3150 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3153 db_print_indent(indent);
3154 db_printf("%s at %p\n", name, inp);
3158 db_print_indent(indent);
3159 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3161 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3163 db_print_indent(indent);
3164 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3165 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3167 db_print_indent(indent);
3168 db_printf("inp_label: %p inp_flags: 0x%x (",
3169 inp->inp_label, inp->inp_flags);
3170 db_print_inpflags(inp->inp_flags);
3173 db_print_indent(indent);
3174 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3176 db_print_inpvflag(inp->inp_vflag);
3179 db_print_indent(indent);
3180 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3181 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3183 db_print_indent(indent);
3185 if (inp->inp_vflag & INP_IPV6) {
3186 db_printf("in6p_options: %p in6p_outputopts: %p "
3187 "in6p_moptions: %p\n", inp->in6p_options,
3188 inp->in6p_outputopts, inp->in6p_moptions);
3189 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3190 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3195 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3196 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3197 inp->inp_options, inp->inp_moptions);
3200 db_print_indent(indent);
3201 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3202 (uintmax_t)inp->inp_gencnt);
3205 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3210 db_printf("usage: show inpcb <addr>\n");
3213 inp = (struct inpcb *)addr;
3215 db_print_inpcb(inp, "inpcb", 0);
3221 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3225 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3227 union if_snd_tag_modify_params params = {
3228 .rate_limit.max_rate = max_pacing_rate,
3230 struct m_snd_tag *mst;
3234 mst = inp->inp_snd_tag;
3242 if (ifp->if_snd_tag_modify == NULL) {
3245 error = ifp->if_snd_tag_modify(mst, ¶ms);
3251 * Query existing TX rate limit based on the existing
3252 * "inp->inp_snd_tag", if any.
3255 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3257 union if_snd_tag_query_params params = { };
3258 struct m_snd_tag *mst;
3262 mst = inp->inp_snd_tag;
3270 if (ifp->if_snd_tag_query == NULL) {
3273 error = ifp->if_snd_tag_query(mst, ¶ms);
3274 if (error == 0 && p_max_pacing_rate != NULL)
3275 *p_max_pacing_rate = params.rate_limit.max_rate;
3281 * Query existing TX queue level based on the existing
3282 * "inp->inp_snd_tag", if any.
3285 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3287 union if_snd_tag_query_params params = { };
3288 struct m_snd_tag *mst;
3292 mst = inp->inp_snd_tag;
3300 if (ifp->if_snd_tag_query == NULL)
3301 return (EOPNOTSUPP);
3303 error = ifp->if_snd_tag_query(mst, ¶ms);
3304 if (error == 0 && p_txqueue_level != NULL)
3305 *p_txqueue_level = params.rate_limit.queue_level;
3310 * Allocate a new TX rate limit send tag from the network interface
3311 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3314 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3315 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3317 union if_snd_tag_alloc_params params = {
3318 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3319 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3320 .rate_limit.hdr.flowid = flowid,
3321 .rate_limit.hdr.flowtype = flowtype,
3322 .rate_limit.max_rate = max_pacing_rate,
3326 INP_WLOCK_ASSERT(inp);
3328 if (inp->inp_snd_tag != NULL)
3331 if (ifp->if_snd_tag_alloc == NULL) {
3334 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3337 * At success increment the refcount on
3338 * the send tag's network interface:
3341 if_ref(inp->inp_snd_tag->ifp);
3347 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3351 in_pcbdetach_txrtlmt(struct inpcb *inp)
3353 struct m_snd_tag *mst;
3356 INP_WLOCK_ASSERT(inp);
3358 mst = inp->inp_snd_tag;
3359 inp->inp_snd_tag = NULL;
3369 * If the device was detached while we still had reference(s)
3370 * on the ifp, we assume if_snd_tag_free() was replaced with
3373 ifp->if_snd_tag_free(mst);
3375 /* release reference count on network interface */
3380 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3381 * is set in the fast path and will attach/detach/modify the TX rate
3382 * limit send tag based on the socket's so_max_pacing_rate value.
3385 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3387 struct socket *socket;
3388 uint32_t max_pacing_rate;
3395 socket = inp->inp_socket;
3399 if (!INP_WLOCKED(inp)) {
3401 * NOTE: If the write locking fails, we need to bail
3402 * out and use the non-ratelimited ring for the
3403 * transmit until there is a new chance to get the
3406 if (!INP_TRY_UPGRADE(inp))
3414 * NOTE: The so_max_pacing_rate value is read unlocked,
3415 * because atomic updates are not required since the variable
3416 * is checked at every mbuf we send. It is assumed that the
3417 * variable read itself will be atomic.
3419 max_pacing_rate = socket->so_max_pacing_rate;
3422 * NOTE: When attaching to a network interface a reference is
3423 * made to ensure the network interface doesn't go away until
3424 * all ratelimit connections are gone. The network interface
3425 * pointers compared below represent valid network interfaces,
3426 * except when comparing towards NULL.
3428 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3430 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3431 if (inp->inp_snd_tag != NULL)
3432 in_pcbdetach_txrtlmt(inp);
3434 } else if (inp->inp_snd_tag == NULL) {
3436 * In order to utilize packet pacing with RSS, we need
3437 * to wait until there is a valid RSS hash before we
3440 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3443 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3444 mb->m_pkthdr.flowid, max_pacing_rate);
3447 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3449 if (error == 0 || error == EOPNOTSUPP)
3450 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3456 * Track route changes for TX rate limiting.
3459 in_pcboutput_eagain(struct inpcb *inp)
3466 if (inp->inp_snd_tag == NULL)
3469 if (!INP_WLOCKED(inp)) {
3471 * NOTE: If the write locking fails, we need to bail
3472 * out and use the non-ratelimited ring for the
3473 * transmit until there is a new chance to get the
3476 if (!INP_TRY_UPGRADE(inp))
3483 /* detach rate limiting */
3484 in_pcbdetach_txrtlmt(inp);
3486 /* make sure new mbuf send tag allocation is made */
3487 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3492 #endif /* RATELIMIT */