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>
89 #include <netinet/ip_var.h>
90 #include <netinet/tcp_var.h>
92 #include <netinet/tcp_hpts.h>
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
98 #include <netinet/in_var.h>
101 #include <netinet/ip6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #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 = *grpp;
293 for (; i + 1 < grp->il_inpcnt; ++i)
294 grp->il_inp[i] = grp->il_inp[i + 1];
297 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
298 grp->il_inpcnt <= (grp->il_inpsiz / 4)) {
299 /* Shrink this group. */
300 struct inpcblbgroup *new_grp =
301 in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
309 * Add PCB to load balance group for SO_REUSEPORT_LB option.
312 in_pcbinslbgrouphash(struct inpcb *inp)
314 const static struct timeval interval = { 60, 0 };
315 static struct timeval lastprint;
316 struct inpcbinfo *pcbinfo;
317 struct inpcblbgrouphead *hdr;
318 struct inpcblbgroup *grp;
319 uint16_t hashmask, lport;
320 uint32_t group_index;
323 pcbinfo = inp->inp_pcbinfo;
325 INP_WLOCK_ASSERT(inp);
326 INP_HASH_WLOCK_ASSERT(pcbinfo);
328 if (pcbinfo->ipi_lbgrouphashbase == NULL)
331 hashmask = pcbinfo->ipi_lbgrouphashmask;
332 lport = inp->inp_lport;
333 group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask);
334 hdr = &pcbinfo->ipi_lbgrouphashbase[group_index];
337 * Don't allow jailed socket to join local group.
339 if (inp->inp_socket != NULL)
340 cred = inp->inp_socket->so_cred;
343 if (cred != NULL && jailed(cred))
348 * Don't allow IPv4 mapped INET6 wild socket.
350 if ((inp->inp_vflag & INP_IPV4) &&
351 inp->inp_laddr.s_addr == INADDR_ANY &&
352 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
357 hdr = &pcbinfo->ipi_lbgrouphashbase[
358 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
359 pcbinfo->ipi_lbgrouphashmask)];
360 CK_LIST_FOREACH(grp, hdr, il_list) {
361 if (grp->il_vflag == inp->inp_vflag &&
362 grp->il_lport == inp->inp_lport &&
363 memcmp(&grp->il_dependladdr,
364 &inp->inp_inc.inc_ie.ie_dependladdr,
365 sizeof(grp->il_dependladdr)) == 0) {
370 /* Create new load balance group. */
371 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
372 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
373 INPCBLBGROUP_SIZMIN);
376 } else if (grp->il_inpcnt == grp->il_inpsiz) {
377 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
378 if (ratecheck(&lastprint, &interval))
379 printf("lb group port %d, limit reached\n",
380 ntohs(grp->il_lport));
384 /* Expand this local group. */
385 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
390 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
391 ("invalid local group size %d and count %d",
392 grp->il_inpsiz, grp->il_inpcnt));
394 grp->il_inp[grp->il_inpcnt] = inp;
400 * Remove PCB from load balance group.
403 in_pcbremlbgrouphash(struct inpcb *inp)
405 struct inpcbinfo *pcbinfo;
406 struct inpcblbgrouphead *hdr;
407 struct inpcblbgroup *grp;
410 pcbinfo = inp->inp_pcbinfo;
412 INP_WLOCK_ASSERT(inp);
413 INP_HASH_WLOCK_ASSERT(pcbinfo);
415 if (pcbinfo->ipi_lbgrouphashbase == NULL)
418 hdr = &pcbinfo->ipi_lbgrouphashbase[
419 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
420 pcbinfo->ipi_lbgrouphashmask)];
422 CK_LIST_FOREACH(grp, hdr, il_list) {
423 for (i = 0; i < grp->il_inpcnt; ++i) {
424 if (grp->il_inp[i] != inp)
427 if (grp->il_inpcnt == 1) {
428 /* We are the last, free this local group. */
429 in_pcblbgroup_free(grp);
431 /* Pull up inpcbs, shrink group if possible. */
432 in_pcblbgroup_reorder(hdr, &grp, i);
440 * Different protocols initialize their inpcbs differently - giving
441 * different name to the lock. But they all are disposed the same.
444 inpcb_fini(void *mem, int size)
446 struct inpcb *inp = mem;
448 INP_LOCK_DESTROY(inp);
452 * Initialize an inpcbinfo -- we should be able to reduce the number of
456 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
457 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
458 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
461 INP_INFO_LOCK_INIT(pcbinfo, name);
462 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
463 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
465 pcbinfo->ipi_vnet = curvnet;
467 pcbinfo->ipi_listhead = listhead;
468 CK_LIST_INIT(pcbinfo->ipi_listhead);
469 pcbinfo->ipi_count = 0;
470 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
471 &pcbinfo->ipi_hashmask);
472 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
473 &pcbinfo->ipi_porthashmask);
474 pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
475 &pcbinfo->ipi_lbgrouphashmask);
477 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
479 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
480 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
481 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
482 uma_zone_set_warning(pcbinfo->ipi_zone,
483 "kern.ipc.maxsockets limit reached");
487 * Destroy an inpcbinfo.
490 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
493 KASSERT(pcbinfo->ipi_count == 0,
494 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
496 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
497 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
498 pcbinfo->ipi_porthashmask);
499 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
500 pcbinfo->ipi_lbgrouphashmask);
502 in_pcbgroup_destroy(pcbinfo);
504 uma_zdestroy(pcbinfo->ipi_zone);
505 INP_LIST_LOCK_DESTROY(pcbinfo);
506 INP_HASH_LOCK_DESTROY(pcbinfo);
507 INP_INFO_LOCK_DESTROY(pcbinfo);
511 * Allocate a PCB and associate it with the socket.
512 * On success return with the PCB locked.
515 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
521 if (pcbinfo == &V_tcbinfo) {
522 INP_INFO_RLOCK_ASSERT(pcbinfo);
524 INP_INFO_WLOCK_ASSERT(pcbinfo);
529 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
532 bzero(&inp->inp_start_zero, inp_zero_size);
533 inp->inp_pcbinfo = pcbinfo;
534 inp->inp_socket = so;
535 inp->inp_cred = crhold(so->so_cred);
536 inp->inp_inc.inc_fibnum = so->so_fibnum;
538 error = mac_inpcb_init(inp, M_NOWAIT);
541 mac_inpcb_create(so, inp);
543 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
544 error = ipsec_init_pcbpolicy(inp);
547 mac_inpcb_destroy(inp);
553 if (INP_SOCKAF(so) == AF_INET6) {
554 inp->inp_vflag |= INP_IPV6PROTO;
556 inp->inp_flags |= IN6P_IPV6_V6ONLY;
560 INP_LIST_WLOCK(pcbinfo);
561 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
562 pcbinfo->ipi_count++;
563 so->so_pcb = (caddr_t)inp;
565 if (V_ip6_auto_flowlabel)
566 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
568 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
569 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
572 * Routes in inpcb's can cache L2 as well; they are guaranteed
575 inp->inp_route.ro_flags = RT_LLE_CACHE;
576 INP_LIST_WUNLOCK(pcbinfo);
577 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
580 crfree(inp->inp_cred);
581 uma_zfree(pcbinfo->ipi_zone, inp);
589 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
593 INP_WLOCK_ASSERT(inp);
594 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
596 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
598 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
599 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
600 &inp->inp_lport, cred);
603 if (in_pcbinshash(inp) != 0) {
604 inp->inp_laddr.s_addr = INADDR_ANY;
609 inp->inp_flags |= INP_ANONPORT;
615 * Select a local port (number) to use.
617 #if defined(INET) || defined(INET6)
619 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
620 struct ucred *cred, int lookupflags)
622 struct inpcbinfo *pcbinfo;
623 struct inpcb *tmpinp;
624 unsigned short *lastport;
625 int count, dorandom, error;
626 u_short aux, first, last, lport;
628 struct in_addr laddr;
631 pcbinfo = inp->inp_pcbinfo;
634 * Because no actual state changes occur here, a global write lock on
635 * the pcbinfo isn't required.
637 INP_LOCK_ASSERT(inp);
638 INP_HASH_LOCK_ASSERT(pcbinfo);
640 if (inp->inp_flags & INP_HIGHPORT) {
641 first = V_ipport_hifirstauto; /* sysctl */
642 last = V_ipport_hilastauto;
643 lastport = &pcbinfo->ipi_lasthi;
644 } else if (inp->inp_flags & INP_LOWPORT) {
645 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
648 first = V_ipport_lowfirstauto; /* 1023 */
649 last = V_ipport_lowlastauto; /* 600 */
650 lastport = &pcbinfo->ipi_lastlow;
652 first = V_ipport_firstauto; /* sysctl */
653 last = V_ipport_lastauto;
654 lastport = &pcbinfo->ipi_lastport;
657 * For UDP(-Lite), use random port allocation as long as the user
658 * allows it. For TCP (and as of yet unknown) connections,
659 * use random port allocation only if the user allows it AND
660 * ipport_tick() allows it.
662 if (V_ipport_randomized &&
663 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
664 pcbinfo == &V_ulitecbinfo))
669 * It makes no sense to do random port allocation if
670 * we have the only port available.
674 /* Make sure to not include UDP(-Lite) packets in the count. */
675 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
676 V_ipport_tcpallocs++;
678 * Instead of having two loops further down counting up or down
679 * make sure that first is always <= last and go with only one
680 * code path implementing all logic.
689 /* Make the compiler happy. */
691 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
692 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
697 tmpinp = NULL; /* Make compiler happy. */
701 *lastport = first + (arc4random() % (last - first));
703 count = last - first;
706 if (count-- < 0) /* completely used? */
707 return (EADDRNOTAVAIL);
709 if (*lastport < first || *lastport > last)
711 lport = htons(*lastport);
714 if ((inp->inp_vflag & INP_IPV6) != 0)
715 tmpinp = in6_pcblookup_local(pcbinfo,
716 &inp->in6p_laddr, lport, lookupflags, cred);
718 #if defined(INET) && defined(INET6)
722 tmpinp = in_pcblookup_local(pcbinfo, laddr,
723 lport, lookupflags, cred);
725 } while (tmpinp != NULL);
728 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
729 laddrp->s_addr = laddr.s_addr;
737 * Return cached socket options.
740 inp_so_options(const struct inpcb *inp)
746 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
747 so_options |= SO_REUSEPORT_LB;
748 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
749 so_options |= SO_REUSEPORT;
750 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
751 so_options |= SO_REUSEADDR;
754 #endif /* INET || INET6 */
757 * Check if a new BINDMULTI socket is allowed to be created.
759 * ni points to the new inp.
760 * oi points to the exisitng inp.
762 * This checks whether the existing inp also has BINDMULTI and
763 * whether the credentials match.
766 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
768 /* Check permissions match */
769 if ((ni->inp_flags2 & INP_BINDMULTI) &&
770 (ni->inp_cred->cr_uid !=
771 oi->inp_cred->cr_uid))
774 /* Check the existing inp has BINDMULTI set */
775 if ((ni->inp_flags2 & INP_BINDMULTI) &&
776 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
780 * We're okay - either INP_BINDMULTI isn't set on ni, or
781 * it is and it matches the checks.
788 * Set up a bind operation on a PCB, performing port allocation
789 * as required, but do not actually modify the PCB. Callers can
790 * either complete the bind by setting inp_laddr/inp_lport and
791 * calling in_pcbinshash(), or they can just use the resulting
792 * port and address to authorise the sending of a once-off packet.
794 * On error, the values of *laddrp and *lportp are not changed.
797 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
798 u_short *lportp, struct ucred *cred)
800 struct socket *so = inp->inp_socket;
801 struct sockaddr_in *sin;
802 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
803 struct in_addr laddr;
805 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
809 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
810 * so that we don't have to add to the (already messy) code below.
812 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
815 * No state changes, so read locks are sufficient here.
817 INP_LOCK_ASSERT(inp);
818 INP_HASH_LOCK_ASSERT(pcbinfo);
820 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
821 return (EADDRNOTAVAIL);
822 laddr.s_addr = *laddrp;
823 if (nam != NULL && laddr.s_addr != INADDR_ANY)
825 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
826 lookupflags = INPLOOKUP_WILDCARD;
828 if ((error = prison_local_ip4(cred, &laddr)) != 0)
831 sin = (struct sockaddr_in *)nam;
832 if (nam->sa_len != sizeof (*sin))
836 * We should check the family, but old programs
837 * incorrectly fail to initialize it.
839 if (sin->sin_family != AF_INET)
840 return (EAFNOSUPPORT);
842 error = prison_local_ip4(cred, &sin->sin_addr);
845 if (sin->sin_port != *lportp) {
846 /* Don't allow the port to change. */
849 lport = sin->sin_port;
851 /* NB: lport is left as 0 if the port isn't being changed. */
852 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
854 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
855 * allow complete duplication of binding if
856 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
857 * and a multicast address is bound on both
858 * new and duplicated sockets.
860 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
861 reuseport = SO_REUSEADDR|SO_REUSEPORT;
863 * XXX: How to deal with SO_REUSEPORT_LB here?
864 * Treat same as SO_REUSEPORT for now.
866 if ((so->so_options &
867 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
868 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
869 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
870 sin->sin_port = 0; /* yech... */
871 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
873 * Is the address a local IP address?
874 * If INP_BINDANY is set, then the socket may be bound
875 * to any endpoint address, local or not.
877 if ((inp->inp_flags & INP_BINDANY) == 0 &&
878 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
879 return (EADDRNOTAVAIL);
881 laddr = sin->sin_addr;
887 if (ntohs(lport) <= V_ipport_reservedhigh &&
888 ntohs(lport) >= V_ipport_reservedlow &&
889 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
892 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
893 priv_check_cred(inp->inp_cred,
894 PRIV_NETINET_REUSEPORT, 0) != 0) {
895 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
896 lport, INPLOOKUP_WILDCARD, cred);
899 * This entire block sorely needs a rewrite.
902 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
903 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
904 (so->so_type != SOCK_STREAM ||
905 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
906 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
907 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
908 (t->inp_flags2 & INP_REUSEPORT) ||
909 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
910 (inp->inp_cred->cr_uid !=
911 t->inp_cred->cr_uid))
915 * If the socket is a BINDMULTI socket, then
916 * the credentials need to match and the
917 * original socket also has to have been bound
920 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
923 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
924 lport, lookupflags, cred);
925 if (t && (t->inp_flags & INP_TIMEWAIT)) {
927 * XXXRW: If an incpb has had its timewait
928 * state recycled, we treat the address as
929 * being in use (for now). This is better
930 * than a panic, but not desirable.
934 ((reuseport & tw->tw_so_options) == 0 &&
936 tw->tw_so_options) == 0)) {
940 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
941 (reuseport & inp_so_options(t)) == 0 &&
942 (reuseport_lb & inp_so_options(t)) == 0) {
944 if (ntohl(sin->sin_addr.s_addr) !=
946 ntohl(t->inp_laddr.s_addr) !=
948 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
949 (t->inp_vflag & INP_IPV6PROTO) == 0)
952 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
960 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
965 *laddrp = laddr.s_addr;
971 * Connect from a socket to a specified address.
972 * Both address and port must be specified in argument sin.
973 * If don't have a local address for this socket yet,
977 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
978 struct ucred *cred, struct mbuf *m)
980 u_short lport, fport;
981 in_addr_t laddr, faddr;
984 INP_WLOCK_ASSERT(inp);
985 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
987 lport = inp->inp_lport;
988 laddr = inp->inp_laddr.s_addr;
989 anonport = (lport == 0);
990 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
995 /* Do the initial binding of the local address if required. */
996 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
997 inp->inp_lport = lport;
998 inp->inp_laddr.s_addr = laddr;
999 if (in_pcbinshash(inp) != 0) {
1000 inp->inp_laddr.s_addr = INADDR_ANY;
1006 /* Commit the remaining changes. */
1007 inp->inp_lport = lport;
1008 inp->inp_laddr.s_addr = laddr;
1009 inp->inp_faddr.s_addr = faddr;
1010 inp->inp_fport = fport;
1011 in_pcbrehash_mbuf(inp, m);
1014 inp->inp_flags |= INP_ANONPORT;
1019 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1022 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1026 * Do proper source address selection on an unbound socket in case
1027 * of connect. Take jails into account as well.
1030 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1034 struct sockaddr *sa;
1035 struct sockaddr_in *sin;
1039 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1041 * Bypass source address selection and use the primary jail IP
1044 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1048 bzero(&sro, sizeof(sro));
1050 sin = (struct sockaddr_in *)&sro.ro_dst;
1051 sin->sin_family = AF_INET;
1052 sin->sin_len = sizeof(struct sockaddr_in);
1053 sin->sin_addr.s_addr = faddr->s_addr;
1056 * If route is known our src addr is taken from the i/f,
1059 * Find out route to destination.
1061 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1062 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1065 * If we found a route, use the address corresponding to
1066 * the outgoing interface.
1068 * Otherwise assume faddr is reachable on a directly connected
1069 * network and try to find a corresponding interface to take
1070 * the source address from.
1073 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1074 struct in_ifaddr *ia;
1077 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1078 inp->inp_socket->so_fibnum));
1080 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1081 inp->inp_socket->so_fibnum));
1085 error = ENETUNREACH;
1089 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1090 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1096 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1099 if (sa->sa_family != AF_INET)
1101 sin = (struct sockaddr_in *)sa;
1102 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1103 ia = (struct in_ifaddr *)ifa;
1108 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1112 /* 3. As a last resort return the 'default' jail address. */
1113 error = prison_get_ip4(cred, laddr);
1118 * If the outgoing interface on the route found is not
1119 * a loopback interface, use the address from that interface.
1120 * In case of jails do those three steps:
1121 * 1. check if the interface address belongs to the jail. If so use it.
1122 * 2. check if we have any address on the outgoing interface
1123 * belonging to this jail. If so use it.
1124 * 3. as a last resort return the 'default' jail address.
1126 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1127 struct in_ifaddr *ia;
1130 /* If not jailed, use the default returned. */
1131 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1132 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1133 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1138 /* 1. Check if the iface address belongs to the jail. */
1139 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1140 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1141 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1142 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1147 * 2. Check if we have any address on the outgoing interface
1148 * belonging to this jail.
1151 ifp = sro.ro_rt->rt_ifp;
1152 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1154 if (sa->sa_family != AF_INET)
1156 sin = (struct sockaddr_in *)sa;
1157 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1158 ia = (struct in_ifaddr *)ifa;
1163 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1167 /* 3. As a last resort return the 'default' jail address. */
1168 error = prison_get_ip4(cred, laddr);
1173 * The outgoing interface is marked with 'loopback net', so a route
1174 * to ourselves is here.
1175 * Try to find the interface of the destination address and then
1176 * take the address from there. That interface is not necessarily
1177 * a loopback interface.
1178 * In case of jails, check that it is an address of the jail
1179 * and if we cannot find, fall back to the 'default' jail address.
1181 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1182 struct sockaddr_in sain;
1183 struct in_ifaddr *ia;
1185 bzero(&sain, sizeof(struct sockaddr_in));
1186 sain.sin_family = AF_INET;
1187 sain.sin_len = sizeof(struct sockaddr_in);
1188 sain.sin_addr.s_addr = faddr->s_addr;
1190 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1191 inp->inp_socket->so_fibnum));
1193 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1194 inp->inp_socket->so_fibnum));
1196 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1198 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1200 error = ENETUNREACH;
1203 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1213 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1215 if (sa->sa_family != AF_INET)
1217 sin = (struct sockaddr_in *)sa;
1218 if (prison_check_ip4(cred,
1219 &sin->sin_addr) == 0) {
1220 ia = (struct in_ifaddr *)ifa;
1225 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1230 /* 3. As a last resort return the 'default' jail address. */
1231 error = prison_get_ip4(cred, laddr);
1237 if (sro.ro_rt != NULL)
1243 * Set up for a connect from a socket to the specified address.
1244 * On entry, *laddrp and *lportp should contain the current local
1245 * address and port for the PCB; these are updated to the values
1246 * that should be placed in inp_laddr and inp_lport to complete
1249 * On success, *faddrp and *fportp will be set to the remote address
1250 * and port. These are not updated in the error case.
1252 * If the operation fails because the connection already exists,
1253 * *oinpp will be set to the PCB of that connection so that the
1254 * caller can decide to override it. In all other cases, *oinpp
1258 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1259 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1260 struct inpcb **oinpp, struct ucred *cred)
1262 struct rm_priotracker in_ifa_tracker;
1263 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1264 struct in_ifaddr *ia;
1266 struct in_addr laddr, faddr;
1267 u_short lport, fport;
1271 * Because a global state change doesn't actually occur here, a read
1272 * lock is sufficient.
1274 INP_LOCK_ASSERT(inp);
1275 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1279 if (nam->sa_len != sizeof (*sin))
1281 if (sin->sin_family != AF_INET)
1282 return (EAFNOSUPPORT);
1283 if (sin->sin_port == 0)
1284 return (EADDRNOTAVAIL);
1285 laddr.s_addr = *laddrp;
1287 faddr = sin->sin_addr;
1288 fport = sin->sin_port;
1290 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1292 * If the destination address is INADDR_ANY,
1293 * use the primary local address.
1294 * If the supplied address is INADDR_BROADCAST,
1295 * and the primary interface supports broadcast,
1296 * choose the broadcast address for that interface.
1298 if (faddr.s_addr == INADDR_ANY) {
1299 IN_IFADDR_RLOCK(&in_ifa_tracker);
1301 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1302 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1304 (error = prison_get_ip4(cred, &faddr)) != 0)
1306 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1307 IN_IFADDR_RLOCK(&in_ifa_tracker);
1308 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1310 faddr = satosin(&CK_STAILQ_FIRST(
1311 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1312 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1315 if (laddr.s_addr == INADDR_ANY) {
1316 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1318 * If the destination address is multicast and an outgoing
1319 * interface has been set as a multicast option, prefer the
1320 * address of that interface as our source address.
1322 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1323 inp->inp_moptions != NULL) {
1324 struct ip_moptions *imo;
1327 imo = inp->inp_moptions;
1328 if (imo->imo_multicast_ifp != NULL) {
1329 ifp = imo->imo_multicast_ifp;
1330 IN_IFADDR_RLOCK(&in_ifa_tracker);
1331 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1332 if ((ia->ia_ifp == ifp) &&
1334 prison_check_ip4(cred,
1335 &ia->ia_addr.sin_addr) == 0))
1339 error = EADDRNOTAVAIL;
1341 laddr = ia->ia_addr.sin_addr;
1344 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1350 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1351 laddr, lport, 0, NULL);
1355 return (EADDRINUSE);
1358 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1363 *laddrp = laddr.s_addr;
1365 *faddrp = faddr.s_addr;
1371 in_pcbdisconnect(struct inpcb *inp)
1374 INP_WLOCK_ASSERT(inp);
1375 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1377 inp->inp_faddr.s_addr = INADDR_ANY;
1384 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1385 * For most protocols, this will be invoked immediately prior to calling
1386 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1387 * socket, in which case in_pcbfree() is deferred.
1390 in_pcbdetach(struct inpcb *inp)
1393 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1396 if (inp->inp_snd_tag != NULL)
1397 in_pcbdetach_txrtlmt(inp);
1399 inp->inp_socket->so_pcb = NULL;
1400 inp->inp_socket = NULL;
1404 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1405 * stability of an inpcb pointer despite the inpcb lock being released. This
1406 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1407 * but where the inpcb lock may already held, or when acquiring a reference
1410 * in_pcbref() should be used only to provide brief memory stability, and
1411 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1412 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1413 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1414 * lock and rele are the *only* safe operations that may be performed on the
1417 * While the inpcb will not be freed, releasing the inpcb lock means that the
1418 * connection's state may change, so the caller should be careful to
1419 * revalidate any cached state on reacquiring the lock. Drop the reference
1420 * using in_pcbrele().
1423 in_pcbref(struct inpcb *inp)
1426 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1428 refcount_acquire(&inp->inp_refcount);
1432 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1433 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1434 * return a flag indicating whether or not the inpcb remains valid. If it is
1435 * valid, we return with the inpcb lock held.
1437 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1438 * reference on an inpcb. Historically more work was done here (actually, in
1439 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1440 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1441 * about memory stability (and continued use of the write lock).
1444 in_pcbrele_rlocked(struct inpcb *inp)
1446 struct inpcbinfo *pcbinfo;
1448 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1450 INP_RLOCK_ASSERT(inp);
1452 if (refcount_release(&inp->inp_refcount) == 0) {
1454 * If the inpcb has been freed, let the caller know, even if
1455 * this isn't the last reference.
1457 if (inp->inp_flags2 & INP_FREED) {
1464 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1466 if (inp->inp_in_hpts || inp->inp_in_input) {
1467 struct tcp_hpts_entry *hpts;
1469 * We should not be on the hpts at
1470 * this point in any form. we must
1471 * get the lock to be sure.
1473 hpts = tcp_hpts_lock(inp);
1474 if (inp->inp_in_hpts)
1475 panic("Hpts:%p inp:%p at free still on hpts",
1477 mtx_unlock(&hpts->p_mtx);
1478 hpts = tcp_input_lock(inp);
1479 if (inp->inp_in_input)
1480 panic("Hpts:%p inp:%p at free still on input hpts",
1482 mtx_unlock(&hpts->p_mtx);
1486 pcbinfo = inp->inp_pcbinfo;
1487 uma_zfree(pcbinfo->ipi_zone, inp);
1492 in_pcbrele_wlocked(struct inpcb *inp)
1494 struct inpcbinfo *pcbinfo;
1496 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1498 INP_WLOCK_ASSERT(inp);
1500 if (refcount_release(&inp->inp_refcount) == 0) {
1502 * If the inpcb has been freed, let the caller know, even if
1503 * this isn't the last reference.
1505 if (inp->inp_flags2 & INP_FREED) {
1512 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1514 if (inp->inp_in_hpts || inp->inp_in_input) {
1515 struct tcp_hpts_entry *hpts;
1517 * We should not be on the hpts at
1518 * this point in any form. we must
1519 * get the lock to be sure.
1521 hpts = tcp_hpts_lock(inp);
1522 if (inp->inp_in_hpts)
1523 panic("Hpts:%p inp:%p at free still on hpts",
1525 mtx_unlock(&hpts->p_mtx);
1526 hpts = tcp_input_lock(inp);
1527 if (inp->inp_in_input)
1528 panic("Hpts:%p inp:%p at free still on input hpts",
1530 mtx_unlock(&hpts->p_mtx);
1534 pcbinfo = inp->inp_pcbinfo;
1535 uma_zfree(pcbinfo->ipi_zone, inp);
1540 * Temporary wrapper.
1543 in_pcbrele(struct inpcb *inp)
1546 return (in_pcbrele_wlocked(inp));
1550 in_pcblist_rele_rlocked(epoch_context_t ctx)
1552 struct in_pcblist *il;
1554 struct inpcbinfo *pcbinfo;
1557 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1558 pcbinfo = il->il_pcbinfo;
1560 INP_INFO_WLOCK(pcbinfo);
1561 for (i = 0; i < n; i++) {
1562 inp = il->il_inp_list[i];
1564 if (!in_pcbrele_rlocked(inp))
1567 INP_INFO_WUNLOCK(pcbinfo);
1572 inpcbport_free(epoch_context_t ctx)
1574 struct inpcbport *phd;
1576 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1581 in_pcbfree_deferred(epoch_context_t ctx)
1584 int released __unused;
1586 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1590 struct ip_moptions *imo = inp->inp_moptions;
1591 inp->inp_moptions = NULL;
1593 /* XXXRW: Do as much as possible here. */
1594 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1595 if (inp->inp_sp != NULL)
1596 ipsec_delete_pcbpolicy(inp);
1599 struct ip6_moptions *im6o = NULL;
1600 if (inp->inp_vflag & INP_IPV6PROTO) {
1601 ip6_freepcbopts(inp->in6p_outputopts);
1602 im6o = inp->in6p_moptions;
1603 inp->in6p_moptions = NULL;
1606 if (inp->inp_options)
1607 (void)m_free(inp->inp_options);
1609 crfree(inp->inp_cred);
1611 mac_inpcb_destroy(inp);
1613 released = in_pcbrele_wlocked(inp);
1616 ip6_freemoptions(im6o);
1619 inp_freemoptions(imo);
1624 * Unconditionally schedule an inpcb to be freed by decrementing its
1625 * reference count, which should occur only after the inpcb has been detached
1626 * from its socket. If another thread holds a temporary reference (acquired
1627 * using in_pcbref()) then the free is deferred until that reference is
1628 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1629 * work, including removal from global lists, is done in this context, where
1630 * the pcbinfo lock is held.
1633 in_pcbfree(struct inpcb *inp)
1635 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1637 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1638 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1639 ("%s: called twice for pcb %p", __func__, inp));
1640 if (inp->inp_flags2 & INP_FREED) {
1646 if (pcbinfo == &V_tcbinfo) {
1647 INP_INFO_LOCK_ASSERT(pcbinfo);
1649 INP_INFO_WLOCK_ASSERT(pcbinfo);
1652 INP_WLOCK_ASSERT(inp);
1653 INP_LIST_WLOCK(pcbinfo);
1654 in_pcbremlists(inp);
1655 INP_LIST_WUNLOCK(pcbinfo);
1656 RO_INVALIDATE_CACHE(&inp->inp_route);
1657 /* mark as destruction in progress */
1658 inp->inp_flags2 |= INP_FREED;
1660 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1664 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1665 * port reservation, and preventing it from being returned by inpcb lookups.
1667 * It is used by TCP to mark an inpcb as unused and avoid future packet
1668 * delivery or event notification when a socket remains open but TCP has
1669 * closed. This might occur as a result of a shutdown()-initiated TCP close
1670 * or a RST on the wire, and allows the port binding to be reused while still
1671 * maintaining the invariant that so_pcb always points to a valid inpcb until
1674 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1675 * in_pcbnotifyall() and in_pcbpurgeif0()?
1678 in_pcbdrop(struct inpcb *inp)
1681 INP_WLOCK_ASSERT(inp);
1683 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1684 MPASS(inp->inp_refcount > 1);
1688 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1691 inp->inp_flags |= INP_DROPPED;
1692 if (inp->inp_flags & INP_INHASHLIST) {
1693 struct inpcbport *phd = inp->inp_phd;
1695 INP_HASH_WLOCK(inp->inp_pcbinfo);
1696 in_pcbremlbgrouphash(inp);
1697 CK_LIST_REMOVE(inp, inp_hash);
1698 CK_LIST_REMOVE(inp, inp_portlist);
1699 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1700 CK_LIST_REMOVE(phd, phd_hash);
1701 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1703 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1704 inp->inp_flags &= ~INP_INHASHLIST;
1706 in_pcbgroup_remove(inp);
1713 * Common routines to return the socket addresses associated with inpcbs.
1716 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1718 struct sockaddr_in *sin;
1720 sin = malloc(sizeof *sin, M_SONAME,
1722 sin->sin_family = AF_INET;
1723 sin->sin_len = sizeof(*sin);
1724 sin->sin_addr = *addr_p;
1725 sin->sin_port = port;
1727 return (struct sockaddr *)sin;
1731 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1734 struct in_addr addr;
1737 inp = sotoinpcb(so);
1738 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1741 port = inp->inp_lport;
1742 addr = inp->inp_laddr;
1745 *nam = in_sockaddr(port, &addr);
1750 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1753 struct in_addr addr;
1756 inp = sotoinpcb(so);
1757 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1760 port = inp->inp_fport;
1761 addr = inp->inp_faddr;
1764 *nam = in_sockaddr(port, &addr);
1769 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1770 struct inpcb *(*notify)(struct inpcb *, int))
1772 struct inpcb *inp, *inp_temp;
1774 INP_INFO_WLOCK(pcbinfo);
1775 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1778 if ((inp->inp_vflag & INP_IPV4) == 0) {
1783 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1784 inp->inp_socket == NULL) {
1788 if ((*notify)(inp, errno))
1791 INP_INFO_WUNLOCK(pcbinfo);
1795 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1798 struct ip_moptions *imo;
1801 INP_INFO_WLOCK(pcbinfo);
1802 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1804 imo = inp->inp_moptions;
1805 if ((inp->inp_vflag & INP_IPV4) &&
1808 * Unselect the outgoing interface if it is being
1811 if (imo->imo_multicast_ifp == ifp)
1812 imo->imo_multicast_ifp = NULL;
1815 * Drop multicast group membership if we joined
1816 * through the interface being detached.
1818 * XXX This can all be deferred to an epoch_call
1820 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1822 if (imo->imo_membership[i]->inm_ifp == ifp) {
1823 IN_MULTI_LOCK_ASSERT();
1824 in_leavegroup_locked(imo->imo_membership[i], NULL);
1826 } else if (gap != 0)
1827 imo->imo_membership[i - gap] =
1828 imo->imo_membership[i];
1830 imo->imo_num_memberships -= gap;
1834 INP_INFO_WUNLOCK(pcbinfo);
1838 * Lookup a PCB based on the local address and port. Caller must hold the
1839 * hash lock. No inpcb locks or references are acquired.
1841 #define INP_LOOKUP_MAPPED_PCB_COST 3
1843 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1844 u_short lport, int lookupflags, struct ucred *cred)
1848 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1854 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1855 ("%s: invalid lookup flags %d", __func__, lookupflags));
1857 INP_HASH_LOCK_ASSERT(pcbinfo);
1859 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1860 struct inpcbhead *head;
1862 * Look for an unconnected (wildcard foreign addr) PCB that
1863 * matches the local address and port we're looking for.
1865 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1866 0, pcbinfo->ipi_hashmask)];
1867 CK_LIST_FOREACH(inp, head, inp_hash) {
1869 /* XXX inp locking */
1870 if ((inp->inp_vflag & INP_IPV4) == 0)
1873 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1874 inp->inp_laddr.s_addr == laddr.s_addr &&
1875 inp->inp_lport == lport) {
1880 prison_equal_ip4(cred->cr_prison,
1881 inp->inp_cred->cr_prison))
1890 struct inpcbporthead *porthash;
1891 struct inpcbport *phd;
1892 struct inpcb *match = NULL;
1894 * Best fit PCB lookup.
1896 * First see if this local port is in use by looking on the
1899 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1900 pcbinfo->ipi_porthashmask)];
1901 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1902 if (phd->phd_port == lport)
1907 * Port is in use by one or more PCBs. Look for best
1910 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1913 !prison_equal_ip4(inp->inp_cred->cr_prison,
1917 /* XXX inp locking */
1918 if ((inp->inp_vflag & INP_IPV4) == 0)
1921 * We never select the PCB that has
1922 * INP_IPV6 flag and is bound to :: if
1923 * we have another PCB which is bound
1924 * to 0.0.0.0. If a PCB has the
1925 * INP_IPV6 flag, then we set its cost
1926 * higher than IPv4 only PCBs.
1928 * Note that the case only happens
1929 * when a socket is bound to ::, under
1930 * the condition that the use of the
1931 * mapped address is allowed.
1933 if ((inp->inp_vflag & INP_IPV6) != 0)
1934 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1936 if (inp->inp_faddr.s_addr != INADDR_ANY)
1938 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1939 if (laddr.s_addr == INADDR_ANY)
1941 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1944 if (laddr.s_addr != INADDR_ANY)
1947 if (wildcard < matchwild) {
1949 matchwild = wildcard;
1958 #undef INP_LOOKUP_MAPPED_PCB_COST
1960 static struct inpcb *
1961 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1962 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1963 uint16_t fport, int lookupflags)
1965 struct inpcb *local_wild;
1966 const struct inpcblbgrouphead *hdr;
1967 struct inpcblbgroup *grp;
1970 INP_HASH_LOCK_ASSERT(pcbinfo);
1972 hdr = &pcbinfo->ipi_lbgrouphashbase[INP_PCBLBGROUP_PORTHASH(lport,
1973 pcbinfo->ipi_lbgrouphashmask)];
1976 * Order of socket selection:
1978 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1981 * - Load balanced group does not contain jailed sockets
1982 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1985 CK_LIST_FOREACH(grp, hdr, il_list) {
1987 if (!(grp->il_vflag & INP_IPV4))
1990 if (grp->il_lport != lport)
1993 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1995 if (grp->il_laddr.s_addr == laddr->s_addr)
1996 return (grp->il_inp[idx]);
1997 if (grp->il_laddr.s_addr == INADDR_ANY &&
1998 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1999 local_wild = grp->il_inp[idx];
2001 return (local_wild);
2006 * Lookup PCB in hash list, using pcbgroup tables.
2008 static struct inpcb *
2009 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2010 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2011 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2013 struct inpcbhead *head;
2014 struct inpcb *inp, *tmpinp;
2015 u_short fport = fport_arg, lport = lport_arg;
2019 * First look for an exact match.
2022 INP_GROUP_LOCK(pcbgroup);
2023 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2024 pcbgroup->ipg_hashmask)];
2025 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2027 /* XXX inp locking */
2028 if ((inp->inp_vflag & INP_IPV4) == 0)
2031 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2032 inp->inp_laddr.s_addr == laddr.s_addr &&
2033 inp->inp_fport == fport &&
2034 inp->inp_lport == lport) {
2036 * XXX We should be able to directly return
2037 * the inp here, without any checks.
2038 * Well unless both bound with SO_REUSEPORT?
2040 if (prison_flag(inp->inp_cred, PR_IP4))
2046 if (tmpinp != NULL) {
2053 * For incoming connections, we may wish to do a wildcard
2054 * match for an RSS-local socket.
2056 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2057 struct inpcb *local_wild = NULL, *local_exact = NULL;
2059 struct inpcb *local_wild_mapped = NULL;
2061 struct inpcb *jail_wild = NULL;
2062 struct inpcbhead *head;
2066 * Order of socket selection - we always prefer jails.
2067 * 1. jailed, non-wild.
2069 * 3. non-jailed, non-wild.
2070 * 4. non-jailed, wild.
2073 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2074 lport, 0, pcbgroup->ipg_hashmask)];
2075 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2077 /* XXX inp locking */
2078 if ((inp->inp_vflag & INP_IPV4) == 0)
2081 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2082 inp->inp_lport != lport)
2085 injail = prison_flag(inp->inp_cred, PR_IP4);
2087 if (prison_check_ip4(inp->inp_cred,
2091 if (local_exact != NULL)
2095 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2100 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2102 /* XXX inp locking, NULL check */
2103 if (inp->inp_vflag & INP_IPV6PROTO)
2104 local_wild_mapped = inp;
2112 } /* LIST_FOREACH */
2121 inp = local_wild_mapped;
2129 * Then look for a wildcard match, if requested.
2131 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2132 struct inpcb *local_wild = NULL, *local_exact = NULL;
2134 struct inpcb *local_wild_mapped = NULL;
2136 struct inpcb *jail_wild = NULL;
2137 struct inpcbhead *head;
2141 * Order of socket selection - we always prefer jails.
2142 * 1. jailed, non-wild.
2144 * 3. non-jailed, non-wild.
2145 * 4. non-jailed, wild.
2147 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2148 0, pcbinfo->ipi_wildmask)];
2149 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2151 /* XXX inp locking */
2152 if ((inp->inp_vflag & INP_IPV4) == 0)
2155 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2156 inp->inp_lport != lport)
2159 injail = prison_flag(inp->inp_cred, PR_IP4);
2161 if (prison_check_ip4(inp->inp_cred,
2165 if (local_exact != NULL)
2169 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2174 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2176 /* XXX inp locking, NULL check */
2177 if (inp->inp_vflag & INP_IPV6PROTO)
2178 local_wild_mapped = inp;
2186 } /* LIST_FOREACH */
2194 inp = local_wild_mapped;
2198 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2199 INP_GROUP_UNLOCK(pcbgroup);
2203 if (lookupflags & INPLOOKUP_WLOCKPCB)
2204 locked = INP_TRY_WLOCK(inp);
2205 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2206 locked = INP_TRY_RLOCK(inp);
2208 panic("%s: locking bug", __func__);
2209 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2210 if (lookupflags & INPLOOKUP_WLOCKPCB)
2217 INP_GROUP_UNLOCK(pcbgroup);
2219 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2221 if (in_pcbrele_wlocked(inp))
2225 if (in_pcbrele_rlocked(inp))
2230 if (lookupflags & INPLOOKUP_WLOCKPCB)
2231 INP_WLOCK_ASSERT(inp);
2233 INP_RLOCK_ASSERT(inp);
2237 #endif /* PCBGROUP */
2240 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2241 * that the caller has locked the hash list, and will not perform any further
2242 * locking or reference operations on either the hash list or the connection.
2244 static struct inpcb *
2245 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2246 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2249 struct inpcbhead *head;
2250 struct inpcb *inp, *tmpinp;
2251 u_short fport = fport_arg, lport = lport_arg;
2254 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2255 ("%s: invalid lookup flags %d", __func__, lookupflags));
2256 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2257 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2260 * First look for an exact match.
2263 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2264 pcbinfo->ipi_hashmask)];
2265 CK_LIST_FOREACH(inp, head, inp_hash) {
2267 /* XXX inp locking */
2268 if ((inp->inp_vflag & INP_IPV4) == 0)
2271 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2272 inp->inp_laddr.s_addr == laddr.s_addr &&
2273 inp->inp_fport == fport &&
2274 inp->inp_lport == lport) {
2276 * XXX We should be able to directly return
2277 * the inp here, without any checks.
2278 * Well unless both bound with SO_REUSEPORT?
2280 if (prison_flag(inp->inp_cred, PR_IP4))
2290 * Then look in lb group (for wildcard match).
2292 if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2293 (lookupflags & INPLOOKUP_WILDCARD)) {
2294 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2295 fport, lookupflags);
2302 * Then look for a wildcard match, if requested.
2304 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2305 struct inpcb *local_wild = NULL, *local_exact = NULL;
2307 struct inpcb *local_wild_mapped = NULL;
2309 struct inpcb *jail_wild = NULL;
2313 * Order of socket selection - we always prefer jails.
2314 * 1. jailed, non-wild.
2316 * 3. non-jailed, non-wild.
2317 * 4. non-jailed, wild.
2320 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2321 0, pcbinfo->ipi_hashmask)];
2322 CK_LIST_FOREACH(inp, head, inp_hash) {
2324 /* XXX inp locking */
2325 if ((inp->inp_vflag & INP_IPV4) == 0)
2328 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2329 inp->inp_lport != lport)
2332 injail = prison_flag(inp->inp_cred, PR_IP4);
2334 if (prison_check_ip4(inp->inp_cred,
2338 if (local_exact != NULL)
2342 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2347 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2349 /* XXX inp locking, NULL check */
2350 if (inp->inp_vflag & INP_IPV6PROTO)
2351 local_wild_mapped = inp;
2359 } /* LIST_FOREACH */
2360 if (jail_wild != NULL)
2362 if (local_exact != NULL)
2363 return (local_exact);
2364 if (local_wild != NULL)
2365 return (local_wild);
2367 if (local_wild_mapped != NULL)
2368 return (local_wild_mapped);
2370 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2376 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2377 * hash list lock, and will return the inpcb locked (i.e., requires
2378 * INPLOOKUP_LOCKPCB).
2380 static struct inpcb *
2381 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2382 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2387 INP_HASH_RLOCK(pcbinfo);
2388 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2389 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2391 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2393 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2397 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2399 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2404 panic("%s: locking bug", __func__);
2407 if (lookupflags & INPLOOKUP_WLOCKPCB)
2408 INP_WLOCK_ASSERT(inp);
2410 INP_RLOCK_ASSERT(inp);
2414 INP_HASH_RUNLOCK(pcbinfo);
2419 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2420 * from which a pre-calculated hash value may be extracted.
2422 * Possibly more of this logic should be in in_pcbgroup.c.
2425 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2426 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2428 #if defined(PCBGROUP) && !defined(RSS)
2429 struct inpcbgroup *pcbgroup;
2432 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2433 ("%s: invalid lookup flags %d", __func__, lookupflags));
2434 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2435 ("%s: LOCKPCB not set", __func__));
2438 * When not using RSS, use connection groups in preference to the
2439 * reservation table when looking up 4-tuples. When using RSS, just
2440 * use the reservation table, due to the cost of the Toeplitz hash
2443 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2444 * we could be doing RSS with a non-Toeplitz hash that is affordable
2447 #if defined(PCBGROUP) && !defined(RSS)
2448 if (in_pcbgroup_enabled(pcbinfo)) {
2449 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2451 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2452 laddr, lport, lookupflags, ifp));
2455 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2460 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2461 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2462 struct ifnet *ifp, struct mbuf *m)
2465 struct inpcbgroup *pcbgroup;
2468 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2469 ("%s: invalid lookup flags %d", __func__, lookupflags));
2470 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2471 ("%s: LOCKPCB not set", __func__));
2475 * If we can use a hardware-generated hash to look up the connection
2476 * group, use that connection group to find the inpcb. Otherwise
2477 * fall back on a software hash -- or the reservation table if we're
2480 * XXXRW: As above, that policy belongs in the pcbgroup code.
2482 if (in_pcbgroup_enabled(pcbinfo) &&
2483 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2484 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2485 m->m_pkthdr.flowid);
2486 if (pcbgroup != NULL)
2487 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2488 fport, laddr, lport, lookupflags, ifp));
2490 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2492 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2493 laddr, lport, lookupflags, ifp));
2497 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2503 * Insert PCB onto various hash lists.
2506 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2508 struct inpcbhead *pcbhash;
2509 struct inpcbporthead *pcbporthash;
2510 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2511 struct inpcbport *phd;
2512 u_int32_t hashkey_faddr;
2515 INP_WLOCK_ASSERT(inp);
2516 INP_HASH_WLOCK_ASSERT(pcbinfo);
2518 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2519 ("in_pcbinshash: INP_INHASHLIST"));
2522 if (inp->inp_vflag & INP_IPV6)
2523 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2526 hashkey_faddr = inp->inp_faddr.s_addr;
2528 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2529 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2531 pcbporthash = &pcbinfo->ipi_porthashbase[
2532 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2535 * Add entry to load balance group.
2536 * Only do this if SO_REUSEPORT_LB is set.
2538 so_options = inp_so_options(inp);
2539 if (so_options & SO_REUSEPORT_LB) {
2540 int ret = in_pcbinslbgrouphash(inp);
2542 /* pcb lb group malloc fail (ret=ENOBUFS). */
2548 * Go through port list and look for a head for this lport.
2550 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2551 if (phd->phd_port == inp->inp_lport)
2555 * If none exists, malloc one and tack it on.
2558 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2560 return (ENOBUFS); /* XXX */
2562 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2563 phd->phd_port = inp->inp_lport;
2564 CK_LIST_INIT(&phd->phd_pcblist);
2565 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2568 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2569 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2570 inp->inp_flags |= INP_INHASHLIST;
2572 if (do_pcbgroup_update)
2573 in_pcbgroup_update(inp);
2579 * For now, there are two public interfaces to insert an inpcb into the hash
2580 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2581 * is used only in the TCP syncache, where in_pcbinshash is called before the
2582 * full 4-tuple is set for the inpcb, and we don't want to install in the
2583 * pcbgroup until later.
2585 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2586 * connection groups, and partially initialised inpcbs should not be exposed
2587 * to either reservation hash tables or pcbgroups.
2590 in_pcbinshash(struct inpcb *inp)
2593 return (in_pcbinshash_internal(inp, 1));
2597 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2600 return (in_pcbinshash_internal(inp, 0));
2604 * Move PCB to the proper hash bucket when { faddr, fport } have been
2605 * changed. NOTE: This does not handle the case of the lport changing (the
2606 * hashed port list would have to be updated as well), so the lport must
2607 * not change after in_pcbinshash() has been called.
2610 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2612 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2613 struct inpcbhead *head;
2614 u_int32_t hashkey_faddr;
2616 INP_WLOCK_ASSERT(inp);
2617 INP_HASH_WLOCK_ASSERT(pcbinfo);
2619 KASSERT(inp->inp_flags & INP_INHASHLIST,
2620 ("in_pcbrehash: !INP_INHASHLIST"));
2623 if (inp->inp_vflag & INP_IPV6)
2624 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2627 hashkey_faddr = inp->inp_faddr.s_addr;
2629 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2630 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2632 CK_LIST_REMOVE(inp, inp_hash);
2633 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2637 in_pcbgroup_update_mbuf(inp, m);
2639 in_pcbgroup_update(inp);
2644 in_pcbrehash(struct inpcb *inp)
2647 in_pcbrehash_mbuf(inp, NULL);
2651 * Remove PCB from various lists.
2654 in_pcbremlists(struct inpcb *inp)
2656 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2659 if (pcbinfo == &V_tcbinfo) {
2660 INP_INFO_RLOCK_ASSERT(pcbinfo);
2662 INP_INFO_WLOCK_ASSERT(pcbinfo);
2666 INP_WLOCK_ASSERT(inp);
2667 INP_LIST_WLOCK_ASSERT(pcbinfo);
2669 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2670 if (inp->inp_flags & INP_INHASHLIST) {
2671 struct inpcbport *phd = inp->inp_phd;
2673 INP_HASH_WLOCK(pcbinfo);
2675 /* XXX: Only do if SO_REUSEPORT_LB set? */
2676 in_pcbremlbgrouphash(inp);
2678 CK_LIST_REMOVE(inp, inp_hash);
2679 CK_LIST_REMOVE(inp, inp_portlist);
2680 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2681 CK_LIST_REMOVE(phd, phd_hash);
2682 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2684 INP_HASH_WUNLOCK(pcbinfo);
2685 inp->inp_flags &= ~INP_INHASHLIST;
2687 CK_LIST_REMOVE(inp, inp_list);
2688 pcbinfo->ipi_count--;
2690 in_pcbgroup_remove(inp);
2695 * Check for alternatives when higher level complains
2696 * about service problems. For now, invalidate cached
2697 * routing information. If the route was created dynamically
2698 * (by a redirect), time to try a default gateway again.
2701 in_losing(struct inpcb *inp)
2704 RO_INVALIDATE_CACHE(&inp->inp_route);
2709 * A set label operation has occurred at the socket layer, propagate the
2710 * label change into the in_pcb for the socket.
2713 in_pcbsosetlabel(struct socket *so)
2718 inp = sotoinpcb(so);
2719 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2723 mac_inpcb_sosetlabel(so, inp);
2730 * ipport_tick runs once per second, determining if random port allocation
2731 * should be continued. If more than ipport_randomcps ports have been
2732 * allocated in the last second, then we return to sequential port
2733 * allocation. We return to random allocation only once we drop below
2734 * ipport_randomcps for at least ipport_randomtime seconds.
2737 ipport_tick(void *xtp)
2739 VNET_ITERATOR_DECL(vnet_iter);
2741 VNET_LIST_RLOCK_NOSLEEP();
2742 VNET_FOREACH(vnet_iter) {
2743 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2744 if (V_ipport_tcpallocs <=
2745 V_ipport_tcplastcount + V_ipport_randomcps) {
2746 if (V_ipport_stoprandom > 0)
2747 V_ipport_stoprandom--;
2749 V_ipport_stoprandom = V_ipport_randomtime;
2750 V_ipport_tcplastcount = V_ipport_tcpallocs;
2753 VNET_LIST_RUNLOCK_NOSLEEP();
2754 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2761 callout_stop(&ipport_tick_callout);
2765 * The ipport_callout should start running at about the time we attach the
2766 * inet or inet6 domains.
2769 ipport_tick_init(const void *unused __unused)
2772 /* Start ipport_tick. */
2773 callout_init(&ipport_tick_callout, 1);
2774 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2775 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2776 SHUTDOWN_PRI_DEFAULT);
2778 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2779 ipport_tick_init, NULL);
2782 inp_wlock(struct inpcb *inp)
2789 inp_wunlock(struct inpcb *inp)
2796 inp_rlock(struct inpcb *inp)
2803 inp_runlock(struct inpcb *inp)
2809 #ifdef INVARIANT_SUPPORT
2811 inp_lock_assert(struct inpcb *inp)
2814 INP_WLOCK_ASSERT(inp);
2818 inp_unlock_assert(struct inpcb *inp)
2821 INP_UNLOCK_ASSERT(inp);
2826 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2830 INP_INFO_WLOCK(&V_tcbinfo);
2831 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2836 INP_INFO_WUNLOCK(&V_tcbinfo);
2840 inp_inpcbtosocket(struct inpcb *inp)
2843 INP_WLOCK_ASSERT(inp);
2844 return (inp->inp_socket);
2848 inp_inpcbtotcpcb(struct inpcb *inp)
2851 INP_WLOCK_ASSERT(inp);
2852 return ((struct tcpcb *)inp->inp_ppcb);
2856 inp_ip_tos_get(const struct inpcb *inp)
2859 return (inp->inp_ip_tos);
2863 inp_ip_tos_set(struct inpcb *inp, int val)
2866 inp->inp_ip_tos = val;
2870 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2871 uint32_t *faddr, uint16_t *fp)
2874 INP_LOCK_ASSERT(inp);
2875 *laddr = inp->inp_laddr.s_addr;
2876 *faddr = inp->inp_faddr.s_addr;
2877 *lp = inp->inp_lport;
2878 *fp = inp->inp_fport;
2882 so_sotoinpcb(struct socket *so)
2885 return (sotoinpcb(so));
2889 so_sototcpcb(struct socket *so)
2892 return (sototcpcb(so));
2896 * Create an external-format (``xinpcb'') structure using the information in
2897 * the kernel-format in_pcb structure pointed to by inp. This is done to
2898 * reduce the spew of irrelevant information over this interface, to isolate
2899 * user code from changes in the kernel structure, and potentially to provide
2900 * information-hiding if we decide that some of this information should be
2901 * hidden from users.
2904 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2907 xi->xi_len = sizeof(struct xinpcb);
2908 if (inp->inp_socket)
2909 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2911 bzero(&xi->xi_socket, sizeof(struct xsocket));
2912 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2913 xi->inp_gencnt = inp->inp_gencnt;
2914 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2915 xi->inp_flow = inp->inp_flow;
2916 xi->inp_flowid = inp->inp_flowid;
2917 xi->inp_flowtype = inp->inp_flowtype;
2918 xi->inp_flags = inp->inp_flags;
2919 xi->inp_flags2 = inp->inp_flags2;
2920 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2921 xi->in6p_cksum = inp->in6p_cksum;
2922 xi->in6p_hops = inp->in6p_hops;
2923 xi->inp_ip_tos = inp->inp_ip_tos;
2924 xi->inp_vflag = inp->inp_vflag;
2925 xi->inp_ip_ttl = inp->inp_ip_ttl;
2926 xi->inp_ip_p = inp->inp_ip_p;
2927 xi->inp_ip_minttl = inp->inp_ip_minttl;
2932 db_print_indent(int indent)
2936 for (i = 0; i < indent; i++)
2941 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2943 char faddr_str[48], laddr_str[48];
2945 db_print_indent(indent);
2946 db_printf("%s at %p\n", name, inc);
2951 if (inc->inc_flags & INC_ISIPV6) {
2953 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2954 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2959 inet_ntoa_r(inc->inc_laddr, laddr_str);
2960 inet_ntoa_r(inc->inc_faddr, faddr_str);
2962 db_print_indent(indent);
2963 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2964 ntohs(inc->inc_lport));
2965 db_print_indent(indent);
2966 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2967 ntohs(inc->inc_fport));
2971 db_print_inpflags(int inp_flags)
2976 if (inp_flags & INP_RECVOPTS) {
2977 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2980 if (inp_flags & INP_RECVRETOPTS) {
2981 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2984 if (inp_flags & INP_RECVDSTADDR) {
2985 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2988 if (inp_flags & INP_ORIGDSTADDR) {
2989 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2992 if (inp_flags & INP_HDRINCL) {
2993 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2996 if (inp_flags & INP_HIGHPORT) {
2997 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3000 if (inp_flags & INP_LOWPORT) {
3001 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3004 if (inp_flags & INP_ANONPORT) {
3005 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3008 if (inp_flags & INP_RECVIF) {
3009 db_printf("%sINP_RECVIF", comma ? ", " : "");
3012 if (inp_flags & INP_MTUDISC) {
3013 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3016 if (inp_flags & INP_RECVTTL) {
3017 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3020 if (inp_flags & INP_DONTFRAG) {
3021 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3024 if (inp_flags & INP_RECVTOS) {
3025 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3028 if (inp_flags & IN6P_IPV6_V6ONLY) {
3029 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3032 if (inp_flags & IN6P_PKTINFO) {
3033 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3036 if (inp_flags & IN6P_HOPLIMIT) {
3037 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3040 if (inp_flags & IN6P_HOPOPTS) {
3041 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3044 if (inp_flags & IN6P_DSTOPTS) {
3045 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3048 if (inp_flags & IN6P_RTHDR) {
3049 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3052 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3053 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3056 if (inp_flags & IN6P_TCLASS) {
3057 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3060 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3061 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3064 if (inp_flags & INP_TIMEWAIT) {
3065 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3068 if (inp_flags & INP_ONESBCAST) {
3069 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3072 if (inp_flags & INP_DROPPED) {
3073 db_printf("%sINP_DROPPED", comma ? ", " : "");
3076 if (inp_flags & INP_SOCKREF) {
3077 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3080 if (inp_flags & IN6P_RFC2292) {
3081 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3084 if (inp_flags & IN6P_MTU) {
3085 db_printf("IN6P_MTU%s", comma ? ", " : "");
3091 db_print_inpvflag(u_char inp_vflag)
3096 if (inp_vflag & INP_IPV4) {
3097 db_printf("%sINP_IPV4", comma ? ", " : "");
3100 if (inp_vflag & INP_IPV6) {
3101 db_printf("%sINP_IPV6", comma ? ", " : "");
3104 if (inp_vflag & INP_IPV6PROTO) {
3105 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3111 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3114 db_print_indent(indent);
3115 db_printf("%s at %p\n", name, inp);
3119 db_print_indent(indent);
3120 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3122 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3124 db_print_indent(indent);
3125 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3126 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3128 db_print_indent(indent);
3129 db_printf("inp_label: %p inp_flags: 0x%x (",
3130 inp->inp_label, inp->inp_flags);
3131 db_print_inpflags(inp->inp_flags);
3134 db_print_indent(indent);
3135 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3137 db_print_inpvflag(inp->inp_vflag);
3140 db_print_indent(indent);
3141 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3142 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3144 db_print_indent(indent);
3146 if (inp->inp_vflag & INP_IPV6) {
3147 db_printf("in6p_options: %p in6p_outputopts: %p "
3148 "in6p_moptions: %p\n", inp->in6p_options,
3149 inp->in6p_outputopts, inp->in6p_moptions);
3150 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3151 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3156 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3157 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3158 inp->inp_options, inp->inp_moptions);
3161 db_print_indent(indent);
3162 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3163 (uintmax_t)inp->inp_gencnt);
3166 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3171 db_printf("usage: show inpcb <addr>\n");
3174 inp = (struct inpcb *)addr;
3176 db_print_inpcb(inp, "inpcb", 0);
3182 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3186 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3188 union if_snd_tag_modify_params params = {
3189 .rate_limit.max_rate = max_pacing_rate,
3191 struct m_snd_tag *mst;
3195 mst = inp->inp_snd_tag;
3203 if (ifp->if_snd_tag_modify == NULL) {
3206 error = ifp->if_snd_tag_modify(mst, ¶ms);
3212 * Query existing TX rate limit based on the existing
3213 * "inp->inp_snd_tag", if any.
3216 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3218 union if_snd_tag_query_params params = { };
3219 struct m_snd_tag *mst;
3223 mst = inp->inp_snd_tag;
3231 if (ifp->if_snd_tag_query == NULL) {
3234 error = ifp->if_snd_tag_query(mst, ¶ms);
3235 if (error == 0 && p_max_pacing_rate != NULL)
3236 *p_max_pacing_rate = params.rate_limit.max_rate;
3242 * Query existing TX queue level based on the existing
3243 * "inp->inp_snd_tag", if any.
3246 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3248 union if_snd_tag_query_params params = { };
3249 struct m_snd_tag *mst;
3253 mst = inp->inp_snd_tag;
3261 if (ifp->if_snd_tag_query == NULL)
3262 return (EOPNOTSUPP);
3264 error = ifp->if_snd_tag_query(mst, ¶ms);
3265 if (error == 0 && p_txqueue_level != NULL)
3266 *p_txqueue_level = params.rate_limit.queue_level;
3271 * Allocate a new TX rate limit send tag from the network interface
3272 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3275 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3276 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3278 union if_snd_tag_alloc_params params = {
3279 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3280 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3281 .rate_limit.hdr.flowid = flowid,
3282 .rate_limit.hdr.flowtype = flowtype,
3283 .rate_limit.max_rate = max_pacing_rate,
3287 INP_WLOCK_ASSERT(inp);
3289 if (inp->inp_snd_tag != NULL)
3292 if (ifp->if_snd_tag_alloc == NULL) {
3295 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3298 * At success increment the refcount on
3299 * the send tag's network interface:
3302 if_ref(inp->inp_snd_tag->ifp);
3308 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3312 in_pcbdetach_txrtlmt(struct inpcb *inp)
3314 struct m_snd_tag *mst;
3317 INP_WLOCK_ASSERT(inp);
3319 mst = inp->inp_snd_tag;
3320 inp->inp_snd_tag = NULL;
3330 * If the device was detached while we still had reference(s)
3331 * on the ifp, we assume if_snd_tag_free() was replaced with
3334 ifp->if_snd_tag_free(mst);
3336 /* release reference count on network interface */
3341 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3342 * is set in the fast path and will attach/detach/modify the TX rate
3343 * limit send tag based on the socket's so_max_pacing_rate value.
3346 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3348 struct socket *socket;
3349 uint32_t max_pacing_rate;
3356 socket = inp->inp_socket;
3360 if (!INP_WLOCKED(inp)) {
3362 * NOTE: If the write locking fails, we need to bail
3363 * out and use the non-ratelimited ring for the
3364 * transmit until there is a new chance to get the
3367 if (!INP_TRY_UPGRADE(inp))
3375 * NOTE: The so_max_pacing_rate value is read unlocked,
3376 * because atomic updates are not required since the variable
3377 * is checked at every mbuf we send. It is assumed that the
3378 * variable read itself will be atomic.
3380 max_pacing_rate = socket->so_max_pacing_rate;
3383 * NOTE: When attaching to a network interface a reference is
3384 * made to ensure the network interface doesn't go away until
3385 * all ratelimit connections are gone. The network interface
3386 * pointers compared below represent valid network interfaces,
3387 * except when comparing towards NULL.
3389 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3391 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3392 if (inp->inp_snd_tag != NULL)
3393 in_pcbdetach_txrtlmt(inp);
3395 } else if (inp->inp_snd_tag == NULL) {
3397 * In order to utilize packet pacing with RSS, we need
3398 * to wait until there is a valid RSS hash before we
3401 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3404 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3405 mb->m_pkthdr.flowid, max_pacing_rate);
3408 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3410 if (error == 0 || error == EOPNOTSUPP)
3411 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3417 * Track route changes for TX rate limiting.
3420 in_pcboutput_eagain(struct inpcb *inp)
3422 struct socket *socket;
3428 socket = inp->inp_socket;
3432 if (inp->inp_snd_tag == NULL)
3435 if (!INP_WLOCKED(inp)) {
3437 * NOTE: If the write locking fails, we need to bail
3438 * out and use the non-ratelimited ring for the
3439 * transmit until there is a new chance to get the
3442 if (!INP_TRY_UPGRADE(inp))
3449 /* detach rate limiting */
3450 in_pcbdetach_txrtlmt(inp);
3452 /* make sure new mbuf send tag allocation is made */
3453 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3458 #endif /* RATELIMIT */