2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1991, 1993, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2007-2009 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 #include "opt_ipsec.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/malloc.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
67 #include <sys/refcount.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
90 #include <netinet/in_var.h>
92 #include <netinet/ip_var.h>
93 #include <netinet/tcp_var.h>
95 #include <netinet/tcp_hpts.h>
97 #include <netinet/udp.h>
98 #include <netinet/udp_var.h>
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
107 #include <netipsec/ipsec_support.h>
109 #include <security/mac/mac_framework.h>
111 #define INPCBLBGROUP_SIZMIN 8
112 #define INPCBLBGROUP_SIZMAX 256
114 static struct callout ipport_tick_callout;
117 * These configure the range of local port addresses assigned to
118 * "unspecified" outgoing connections/packets/whatever.
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
128 * Reserved ports accessible only to root. There are significant
129 * security considerations that must be accounted for when changing these,
130 * but the security benefits can be great. Please be careful.
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
145 static void in_pcbremlists(struct inpcb *inp);
147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 struct in_addr faddr, u_int fport_arg,
149 struct in_addr laddr, u_int lport_arg,
150 int lookupflags, struct ifnet *ifp);
152 #define RANGECHK(var, min, max) \
153 if ((var) < (min)) { (var) = (min); } \
154 else if ((var) > (max)) { (var) = (max); }
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
161 error = sysctl_handle_int(oidp, arg1, arg2, req);
163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 &VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 CTLFLAG_VNET | CTLFLAG_RW,
206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 "allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(ipport_randomtime), 0,
211 "Minimum time to keep sequental port "
212 "allocation before switching to a random one");
215 counter_u64_t rate_limit_active;
216 counter_u64_t rate_limit_alloc_fail;
217 counter_u64_t rate_limit_set_ok;
219 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD, 0,
221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
222 &rate_limit_active, "Active rate limited connections");
223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
224 &rate_limit_alloc_fail, "Rate limited connection failures");
225 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
226 &rate_limit_set_ok, "Rate limited setting succeeded");
227 #endif /* RATELIMIT */
232 * in_pcb.c: manage the Protocol Control Blocks.
234 * NOTE: It is assumed that most of these functions will be called with
235 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
236 * functions often modify hash chains or addresses in pcbs.
239 static struct inpcblbgroup *
240 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
241 uint16_t port, const union in_dependaddr *addr, int size)
243 struct inpcblbgroup *grp;
246 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
247 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
250 grp->il_vflag = vflag;
251 grp->il_lport = port;
252 grp->il_dependladdr = *addr;
253 grp->il_inpsiz = size;
254 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
259 in_pcblbgroup_free_deferred(epoch_context_t ctx)
261 struct inpcblbgroup *grp;
263 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
268 in_pcblbgroup_free(struct inpcblbgroup *grp)
271 CK_LIST_REMOVE(grp, il_list);
272 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
275 static struct inpcblbgroup *
276 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
277 struct inpcblbgroup *old_grp, int size)
279 struct inpcblbgroup *grp;
282 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
283 old_grp->il_lport, &old_grp->il_dependladdr, size);
287 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
288 ("invalid new local group size %d and old local group count %d",
289 grp->il_inpsiz, old_grp->il_inpcnt));
291 for (i = 0; i < old_grp->il_inpcnt; ++i)
292 grp->il_inp[i] = old_grp->il_inp[i];
293 grp->il_inpcnt = old_grp->il_inpcnt;
294 in_pcblbgroup_free(old_grp);
299 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
300 * and shrink group if possible.
303 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
306 struct inpcblbgroup *grp, *new_grp;
309 for (; i + 1 < grp->il_inpcnt; ++i)
310 grp->il_inp[i] = grp->il_inp[i + 1];
313 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
314 grp->il_inpcnt <= grp->il_inpsiz / 4) {
315 /* Shrink this group. */
316 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
323 * Add PCB to load balance group for SO_REUSEPORT_LB option.
326 in_pcbinslbgrouphash(struct inpcb *inp)
328 const static struct timeval interval = { 60, 0 };
329 static struct timeval lastprint;
330 struct inpcbinfo *pcbinfo;
331 struct inpcblbgrouphead *hdr;
332 struct inpcblbgroup *grp;
335 pcbinfo = inp->inp_pcbinfo;
337 INP_WLOCK_ASSERT(inp);
338 INP_HASH_WLOCK_ASSERT(pcbinfo);
341 * Don't allow jailed socket to join local group.
343 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
348 * Don't allow IPv4 mapped INET6 wild socket.
350 if ((inp->inp_vflag & INP_IPV4) &&
351 inp->inp_laddr.s_addr == INADDR_ANY &&
352 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
357 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
358 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
359 CK_LIST_FOREACH(grp, hdr, il_list) {
360 if (grp->il_vflag == inp->inp_vflag &&
361 grp->il_lport == inp->inp_lport &&
362 memcmp(&grp->il_dependladdr,
363 &inp->inp_inc.inc_ie.ie_dependladdr,
364 sizeof(grp->il_dependladdr)) == 0)
368 /* Create new load balance group. */
369 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
370 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
371 INPCBLBGROUP_SIZMIN);
374 } else if (grp->il_inpcnt == grp->il_inpsiz) {
375 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
376 if (ratecheck(&lastprint, &interval))
377 printf("lb group port %d, limit reached\n",
378 ntohs(grp->il_lport));
382 /* Expand this local group. */
383 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
388 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
389 ("invalid local group size %d and count %d", grp->il_inpsiz,
392 grp->il_inp[grp->il_inpcnt] = inp;
398 * Remove PCB from load balance group.
401 in_pcbremlbgrouphash(struct inpcb *inp)
403 struct inpcbinfo *pcbinfo;
404 struct inpcblbgrouphead *hdr;
405 struct inpcblbgroup *grp;
408 pcbinfo = inp->inp_pcbinfo;
410 INP_WLOCK_ASSERT(inp);
411 INP_HASH_WLOCK_ASSERT(pcbinfo);
413 hdr = &pcbinfo->ipi_lbgrouphashbase[
414 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
415 CK_LIST_FOREACH(grp, hdr, il_list) {
416 for (i = 0; i < grp->il_inpcnt; ++i) {
417 if (grp->il_inp[i] != inp)
420 if (grp->il_inpcnt == 1) {
421 /* We are the last, free this local group. */
422 in_pcblbgroup_free(grp);
424 /* Pull up inpcbs, shrink group if possible. */
425 in_pcblbgroup_reorder(hdr, &grp, i);
433 * Different protocols initialize their inpcbs differently - giving
434 * different name to the lock. But they all are disposed the same.
437 inpcb_fini(void *mem, int size)
439 struct inpcb *inp = mem;
441 INP_LOCK_DESTROY(inp);
445 * Initialize an inpcbinfo -- we should be able to reduce the number of
449 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
450 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
451 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
454 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
456 INP_INFO_LOCK_INIT(pcbinfo, name);
457 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
458 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
460 pcbinfo->ipi_vnet = curvnet;
462 pcbinfo->ipi_listhead = listhead;
463 CK_LIST_INIT(pcbinfo->ipi_listhead);
464 pcbinfo->ipi_count = 0;
465 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
466 &pcbinfo->ipi_hashmask);
467 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
468 &pcbinfo->ipi_porthashmask);
469 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
470 &pcbinfo->ipi_lbgrouphashmask);
472 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
474 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
475 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
476 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
477 uma_zone_set_warning(pcbinfo->ipi_zone,
478 "kern.ipc.maxsockets limit reached");
482 * Destroy an inpcbinfo.
485 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
488 KASSERT(pcbinfo->ipi_count == 0,
489 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
491 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
492 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
493 pcbinfo->ipi_porthashmask);
494 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
495 pcbinfo->ipi_lbgrouphashmask);
497 in_pcbgroup_destroy(pcbinfo);
499 uma_zdestroy(pcbinfo->ipi_zone);
500 INP_LIST_LOCK_DESTROY(pcbinfo);
501 INP_HASH_LOCK_DESTROY(pcbinfo);
502 INP_INFO_LOCK_DESTROY(pcbinfo);
506 * Allocate a PCB and associate it with the socket.
507 * On success return with the PCB locked.
510 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
516 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
519 bzero(&inp->inp_start_zero, inp_zero_size);
521 inp->inp_numa_domain = M_NODOM;
523 inp->inp_pcbinfo = pcbinfo;
524 inp->inp_socket = so;
525 inp->inp_cred = crhold(so->so_cred);
526 inp->inp_inc.inc_fibnum = so->so_fibnum;
528 error = mac_inpcb_init(inp, M_NOWAIT);
531 mac_inpcb_create(so, inp);
533 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
534 error = ipsec_init_pcbpolicy(inp);
537 mac_inpcb_destroy(inp);
543 if (INP_SOCKAF(so) == AF_INET6) {
544 inp->inp_vflag |= INP_IPV6PROTO;
546 inp->inp_flags |= IN6P_IPV6_V6ONLY;
550 INP_LIST_WLOCK(pcbinfo);
551 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
552 pcbinfo->ipi_count++;
553 so->so_pcb = (caddr_t)inp;
555 if (V_ip6_auto_flowlabel)
556 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
558 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
559 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
562 * Routes in inpcb's can cache L2 as well; they are guaranteed
565 inp->inp_route.ro_flags = RT_LLE_CACHE;
566 INP_LIST_WUNLOCK(pcbinfo);
567 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
570 crfree(inp->inp_cred);
571 uma_zfree(pcbinfo->ipi_zone, inp);
579 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
583 INP_WLOCK_ASSERT(inp);
584 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
586 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
588 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
589 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
590 &inp->inp_lport, cred);
593 if (in_pcbinshash(inp) != 0) {
594 inp->inp_laddr.s_addr = INADDR_ANY;
599 inp->inp_flags |= INP_ANONPORT;
605 * Select a local port (number) to use.
607 #if defined(INET) || defined(INET6)
609 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
610 struct ucred *cred, int lookupflags)
612 struct inpcbinfo *pcbinfo;
613 struct inpcb *tmpinp;
614 unsigned short *lastport;
615 int count, dorandom, error;
616 u_short aux, first, last, lport;
618 struct in_addr laddr;
621 pcbinfo = inp->inp_pcbinfo;
624 * Because no actual state changes occur here, a global write lock on
625 * the pcbinfo isn't required.
627 INP_LOCK_ASSERT(inp);
628 INP_HASH_LOCK_ASSERT(pcbinfo);
630 if (inp->inp_flags & INP_HIGHPORT) {
631 first = V_ipport_hifirstauto; /* sysctl */
632 last = V_ipport_hilastauto;
633 lastport = &pcbinfo->ipi_lasthi;
634 } else if (inp->inp_flags & INP_LOWPORT) {
635 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
638 first = V_ipport_lowfirstauto; /* 1023 */
639 last = V_ipport_lowlastauto; /* 600 */
640 lastport = &pcbinfo->ipi_lastlow;
642 first = V_ipport_firstauto; /* sysctl */
643 last = V_ipport_lastauto;
644 lastport = &pcbinfo->ipi_lastport;
647 * For UDP(-Lite), use random port allocation as long as the user
648 * allows it. For TCP (and as of yet unknown) connections,
649 * use random port allocation only if the user allows it AND
650 * ipport_tick() allows it.
652 if (V_ipport_randomized &&
653 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
654 pcbinfo == &V_ulitecbinfo))
659 * It makes no sense to do random port allocation if
660 * we have the only port available.
664 /* Make sure to not include UDP(-Lite) packets in the count. */
665 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
666 V_ipport_tcpallocs++;
668 * Instead of having two loops further down counting up or down
669 * make sure that first is always <= last and go with only one
670 * code path implementing all logic.
679 /* Make the compiler happy. */
681 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
682 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
687 tmpinp = NULL; /* Make compiler happy. */
691 *lastport = first + (arc4random() % (last - first));
693 count = last - first;
696 if (count-- < 0) /* completely used? */
697 return (EADDRNOTAVAIL);
699 if (*lastport < first || *lastport > last)
701 lport = htons(*lastport);
704 if ((inp->inp_vflag & INP_IPV6) != 0)
705 tmpinp = in6_pcblookup_local(pcbinfo,
706 &inp->in6p_laddr, lport, lookupflags, cred);
708 #if defined(INET) && defined(INET6)
712 tmpinp = in_pcblookup_local(pcbinfo, laddr,
713 lport, lookupflags, cred);
715 } while (tmpinp != NULL);
718 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
719 laddrp->s_addr = laddr.s_addr;
727 * Return cached socket options.
730 inp_so_options(const struct inpcb *inp)
736 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
737 so_options |= SO_REUSEPORT_LB;
738 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
739 so_options |= SO_REUSEPORT;
740 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
741 so_options |= SO_REUSEADDR;
744 #endif /* INET || INET6 */
747 * Check if a new BINDMULTI socket is allowed to be created.
749 * ni points to the new inp.
750 * oi points to the exisitng inp.
752 * This checks whether the existing inp also has BINDMULTI and
753 * whether the credentials match.
756 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
758 /* Check permissions match */
759 if ((ni->inp_flags2 & INP_BINDMULTI) &&
760 (ni->inp_cred->cr_uid !=
761 oi->inp_cred->cr_uid))
764 /* Check the existing inp has BINDMULTI set */
765 if ((ni->inp_flags2 & INP_BINDMULTI) &&
766 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
770 * We're okay - either INP_BINDMULTI isn't set on ni, or
771 * it is and it matches the checks.
778 * Set up a bind operation on a PCB, performing port allocation
779 * as required, but do not actually modify the PCB. Callers can
780 * either complete the bind by setting inp_laddr/inp_lport and
781 * calling in_pcbinshash(), or they can just use the resulting
782 * port and address to authorise the sending of a once-off packet.
784 * On error, the values of *laddrp and *lportp are not changed.
787 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
788 u_short *lportp, struct ucred *cred)
790 struct socket *so = inp->inp_socket;
791 struct sockaddr_in *sin;
792 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
793 struct in_addr laddr;
795 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
799 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
800 * so that we don't have to add to the (already messy) code below.
802 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
805 * No state changes, so read locks are sufficient here.
807 INP_LOCK_ASSERT(inp);
808 INP_HASH_LOCK_ASSERT(pcbinfo);
810 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
811 return (EADDRNOTAVAIL);
812 laddr.s_addr = *laddrp;
813 if (nam != NULL && laddr.s_addr != INADDR_ANY)
815 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
816 lookupflags = INPLOOKUP_WILDCARD;
818 if ((error = prison_local_ip4(cred, &laddr)) != 0)
821 sin = (struct sockaddr_in *)nam;
822 if (nam->sa_len != sizeof (*sin))
826 * We should check the family, but old programs
827 * incorrectly fail to initialize it.
829 if (sin->sin_family != AF_INET)
830 return (EAFNOSUPPORT);
832 error = prison_local_ip4(cred, &sin->sin_addr);
835 if (sin->sin_port != *lportp) {
836 /* Don't allow the port to change. */
839 lport = sin->sin_port;
841 /* NB: lport is left as 0 if the port isn't being changed. */
842 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
844 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
845 * allow complete duplication of binding if
846 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
847 * and a multicast address is bound on both
848 * new and duplicated sockets.
850 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
851 reuseport = SO_REUSEADDR|SO_REUSEPORT;
853 * XXX: How to deal with SO_REUSEPORT_LB here?
854 * Treat same as SO_REUSEPORT for now.
856 if ((so->so_options &
857 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
858 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
859 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
860 sin->sin_port = 0; /* yech... */
861 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
863 * Is the address a local IP address?
864 * If INP_BINDANY is set, then the socket may be bound
865 * to any endpoint address, local or not.
867 if ((inp->inp_flags & INP_BINDANY) == 0 &&
868 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
869 return (EADDRNOTAVAIL);
871 laddr = sin->sin_addr;
877 if (ntohs(lport) <= V_ipport_reservedhigh &&
878 ntohs(lport) >= V_ipport_reservedlow &&
879 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
881 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
882 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
883 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
884 lport, INPLOOKUP_WILDCARD, cred);
887 * This entire block sorely needs a rewrite.
890 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
891 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
892 (so->so_type != SOCK_STREAM ||
893 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
894 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
895 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
896 (t->inp_flags2 & INP_REUSEPORT) ||
897 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
898 (inp->inp_cred->cr_uid !=
899 t->inp_cred->cr_uid))
903 * If the socket is a BINDMULTI socket, then
904 * the credentials need to match and the
905 * original socket also has to have been bound
908 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
911 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
912 lport, lookupflags, cred);
913 if (t && (t->inp_flags & INP_TIMEWAIT)) {
915 * XXXRW: If an incpb has had its timewait
916 * state recycled, we treat the address as
917 * being in use (for now). This is better
918 * than a panic, but not desirable.
922 ((reuseport & tw->tw_so_options) == 0 &&
924 tw->tw_so_options) == 0)) {
928 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
929 (reuseport & inp_so_options(t)) == 0 &&
930 (reuseport_lb & inp_so_options(t)) == 0) {
932 if (ntohl(sin->sin_addr.s_addr) !=
934 ntohl(t->inp_laddr.s_addr) !=
936 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
937 (t->inp_vflag & INP_IPV6PROTO) == 0)
940 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
948 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
953 *laddrp = laddr.s_addr;
959 * Connect from a socket to a specified address.
960 * Both address and port must be specified in argument sin.
961 * If don't have a local address for this socket yet,
965 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
966 struct ucred *cred, struct mbuf *m, bool rehash)
968 u_short lport, fport;
969 in_addr_t laddr, faddr;
972 INP_WLOCK_ASSERT(inp);
973 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
975 lport = inp->inp_lport;
976 laddr = inp->inp_laddr.s_addr;
977 anonport = (lport == 0);
978 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
983 /* Do the initial binding of the local address if required. */
984 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
985 KASSERT(rehash == true,
986 ("Rehashing required for unbound inps"));
987 inp->inp_lport = lport;
988 inp->inp_laddr.s_addr = laddr;
989 if (in_pcbinshash(inp) != 0) {
990 inp->inp_laddr.s_addr = INADDR_ANY;
996 /* Commit the remaining changes. */
997 inp->inp_lport = lport;
998 inp->inp_laddr.s_addr = laddr;
999 inp->inp_faddr.s_addr = faddr;
1000 inp->inp_fport = fport;
1002 in_pcbrehash_mbuf(inp, m);
1004 in_pcbinshash_mbuf(inp, m);
1008 inp->inp_flags |= INP_ANONPORT;
1013 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1016 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1020 * Do proper source address selection on an unbound socket in case
1021 * of connect. Take jails into account as well.
1024 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1028 struct sockaddr *sa;
1029 struct sockaddr_in *sin;
1034 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1036 * Bypass source address selection and use the primary jail IP
1039 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1043 bzero(&sro, sizeof(sro));
1045 sin = (struct sockaddr_in *)&sro.ro_dst;
1046 sin->sin_family = AF_INET;
1047 sin->sin_len = sizeof(struct sockaddr_in);
1048 sin->sin_addr.s_addr = faddr->s_addr;
1051 * If route is known our src addr is taken from the i/f,
1054 * Find out route to destination.
1056 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1057 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1060 * If we found a route, use the address corresponding to
1061 * the outgoing interface.
1063 * Otherwise assume faddr is reachable on a directly connected
1064 * network and try to find a corresponding interface to take
1065 * the source address from.
1067 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1068 struct in_ifaddr *ia;
1071 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1072 inp->inp_socket->so_fibnum));
1074 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1075 inp->inp_socket->so_fibnum));
1079 error = ENETUNREACH;
1083 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1084 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1090 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1093 if (sa->sa_family != AF_INET)
1095 sin = (struct sockaddr_in *)sa;
1096 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1097 ia = (struct in_ifaddr *)ifa;
1102 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1106 /* 3. As a last resort return the 'default' jail address. */
1107 error = prison_get_ip4(cred, laddr);
1112 * If the outgoing interface on the route found is not
1113 * a loopback interface, use the address from that interface.
1114 * In case of jails do those three steps:
1115 * 1. check if the interface address belongs to the jail. If so use it.
1116 * 2. check if we have any address on the outgoing interface
1117 * belonging to this jail. If so use it.
1118 * 3. as a last resort return the 'default' jail address.
1120 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1121 struct in_ifaddr *ia;
1124 /* If not jailed, use the default returned. */
1125 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1126 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1127 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1132 /* 1. Check if the iface address belongs to the jail. */
1133 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1134 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1135 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1136 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1141 * 2. Check if we have any address on the outgoing interface
1142 * belonging to this jail.
1145 ifp = sro.ro_rt->rt_ifp;
1146 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1148 if (sa->sa_family != AF_INET)
1150 sin = (struct sockaddr_in *)sa;
1151 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1152 ia = (struct in_ifaddr *)ifa;
1157 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1161 /* 3. As a last resort return the 'default' jail address. */
1162 error = prison_get_ip4(cred, laddr);
1167 * The outgoing interface is marked with 'loopback net', so a route
1168 * to ourselves is here.
1169 * Try to find the interface of the destination address and then
1170 * take the address from there. That interface is not necessarily
1171 * a loopback interface.
1172 * In case of jails, check that it is an address of the jail
1173 * and if we cannot find, fall back to the 'default' jail address.
1175 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1176 struct sockaddr_in sain;
1177 struct in_ifaddr *ia;
1179 bzero(&sain, sizeof(struct sockaddr_in));
1180 sain.sin_family = AF_INET;
1181 sain.sin_len = sizeof(struct sockaddr_in);
1182 sain.sin_addr.s_addr = faddr->s_addr;
1184 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1185 inp->inp_socket->so_fibnum));
1187 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1188 inp->inp_socket->so_fibnum));
1190 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1192 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1194 error = ENETUNREACH;
1197 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1207 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1209 if (sa->sa_family != AF_INET)
1211 sin = (struct sockaddr_in *)sa;
1212 if (prison_check_ip4(cred,
1213 &sin->sin_addr) == 0) {
1214 ia = (struct in_ifaddr *)ifa;
1219 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1224 /* 3. As a last resort return the 'default' jail address. */
1225 error = prison_get_ip4(cred, laddr);
1230 if (sro.ro_rt != NULL)
1236 * Set up for a connect from a socket to the specified address.
1237 * On entry, *laddrp and *lportp should contain the current local
1238 * address and port for the PCB; these are updated to the values
1239 * that should be placed in inp_laddr and inp_lport to complete
1242 * On success, *faddrp and *fportp will be set to the remote address
1243 * and port. These are not updated in the error case.
1245 * If the operation fails because the connection already exists,
1246 * *oinpp will be set to the PCB of that connection so that the
1247 * caller can decide to override it. In all other cases, *oinpp
1251 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1252 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1253 struct inpcb **oinpp, struct ucred *cred)
1255 struct rm_priotracker in_ifa_tracker;
1256 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1257 struct in_ifaddr *ia;
1259 struct in_addr laddr, faddr;
1260 u_short lport, fport;
1264 * Because a global state change doesn't actually occur here, a read
1265 * lock is sufficient.
1268 INP_LOCK_ASSERT(inp);
1269 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1273 if (nam->sa_len != sizeof (*sin))
1275 if (sin->sin_family != AF_INET)
1276 return (EAFNOSUPPORT);
1277 if (sin->sin_port == 0)
1278 return (EADDRNOTAVAIL);
1279 laddr.s_addr = *laddrp;
1281 faddr = sin->sin_addr;
1282 fport = sin->sin_port;
1284 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1286 * If the destination address is INADDR_ANY,
1287 * use the primary local address.
1288 * If the supplied address is INADDR_BROADCAST,
1289 * and the primary interface supports broadcast,
1290 * choose the broadcast address for that interface.
1292 if (faddr.s_addr == INADDR_ANY) {
1293 IN_IFADDR_RLOCK(&in_ifa_tracker);
1295 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1296 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1298 (error = prison_get_ip4(cred, &faddr)) != 0)
1300 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1301 IN_IFADDR_RLOCK(&in_ifa_tracker);
1302 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1304 faddr = satosin(&CK_STAILQ_FIRST(
1305 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1306 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1309 if (laddr.s_addr == INADDR_ANY) {
1310 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1312 * If the destination address is multicast and an outgoing
1313 * interface has been set as a multicast option, prefer the
1314 * address of that interface as our source address.
1316 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1317 inp->inp_moptions != NULL) {
1318 struct ip_moptions *imo;
1321 imo = inp->inp_moptions;
1322 if (imo->imo_multicast_ifp != NULL) {
1323 ifp = imo->imo_multicast_ifp;
1324 IN_IFADDR_RLOCK(&in_ifa_tracker);
1325 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1326 if ((ia->ia_ifp == ifp) &&
1328 prison_check_ip4(cred,
1329 &ia->ia_addr.sin_addr) == 0))
1333 error = EADDRNOTAVAIL;
1335 laddr = ia->ia_addr.sin_addr;
1338 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1344 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1345 laddr, lport, 0, NULL);
1349 return (EADDRINUSE);
1352 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1357 *laddrp = laddr.s_addr;
1359 *faddrp = faddr.s_addr;
1365 in_pcbdisconnect(struct inpcb *inp)
1368 INP_WLOCK_ASSERT(inp);
1369 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1371 inp->inp_faddr.s_addr = INADDR_ANY;
1378 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1379 * For most protocols, this will be invoked immediately prior to calling
1380 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1381 * socket, in which case in_pcbfree() is deferred.
1384 in_pcbdetach(struct inpcb *inp)
1387 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1390 if (inp->inp_snd_tag != NULL)
1391 in_pcbdetach_txrtlmt(inp);
1393 inp->inp_socket->so_pcb = NULL;
1394 inp->inp_socket = NULL;
1398 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1399 * stability of an inpcb pointer despite the inpcb lock being released. This
1400 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1401 * but where the inpcb lock may already held, or when acquiring a reference
1404 * in_pcbref() should be used only to provide brief memory stability, and
1405 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1406 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1407 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1408 * lock and rele are the *only* safe operations that may be performed on the
1411 * While the inpcb will not be freed, releasing the inpcb lock means that the
1412 * connection's state may change, so the caller should be careful to
1413 * revalidate any cached state on reacquiring the lock. Drop the reference
1414 * using in_pcbrele().
1417 in_pcbref(struct inpcb *inp)
1420 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1422 refcount_acquire(&inp->inp_refcount);
1426 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1427 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1428 * return a flag indicating whether or not the inpcb remains valid. If it is
1429 * valid, we return with the inpcb lock held.
1431 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1432 * reference on an inpcb. Historically more work was done here (actually, in
1433 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1434 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1435 * about memory stability (and continued use of the write lock).
1438 in_pcbrele_rlocked(struct inpcb *inp)
1440 struct inpcbinfo *pcbinfo;
1442 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1444 INP_RLOCK_ASSERT(inp);
1446 if (refcount_release(&inp->inp_refcount) == 0) {
1448 * If the inpcb has been freed, let the caller know, even if
1449 * this isn't the last reference.
1451 if (inp->inp_flags2 & INP_FREED) {
1458 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1460 if (inp->inp_in_hpts || inp->inp_in_input) {
1461 struct tcp_hpts_entry *hpts;
1463 * We should not be on the hpts at
1464 * this point in any form. we must
1465 * get the lock to be sure.
1467 hpts = tcp_hpts_lock(inp);
1468 if (inp->inp_in_hpts)
1469 panic("Hpts:%p inp:%p at free still on hpts",
1471 mtx_unlock(&hpts->p_mtx);
1472 hpts = tcp_input_lock(inp);
1473 if (inp->inp_in_input)
1474 panic("Hpts:%p inp:%p at free still on input hpts",
1476 mtx_unlock(&hpts->p_mtx);
1480 pcbinfo = inp->inp_pcbinfo;
1481 uma_zfree(pcbinfo->ipi_zone, inp);
1486 in_pcbrele_wlocked(struct inpcb *inp)
1488 struct inpcbinfo *pcbinfo;
1490 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1492 INP_WLOCK_ASSERT(inp);
1494 if (refcount_release(&inp->inp_refcount) == 0) {
1496 * If the inpcb has been freed, let the caller know, even if
1497 * this isn't the last reference.
1499 if (inp->inp_flags2 & INP_FREED) {
1506 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1508 if (inp->inp_in_hpts || inp->inp_in_input) {
1509 struct tcp_hpts_entry *hpts;
1511 * We should not be on the hpts at
1512 * this point in any form. we must
1513 * get the lock to be sure.
1515 hpts = tcp_hpts_lock(inp);
1516 if (inp->inp_in_hpts)
1517 panic("Hpts:%p inp:%p at free still on hpts",
1519 mtx_unlock(&hpts->p_mtx);
1520 hpts = tcp_input_lock(inp);
1521 if (inp->inp_in_input)
1522 panic("Hpts:%p inp:%p at free still on input hpts",
1524 mtx_unlock(&hpts->p_mtx);
1528 pcbinfo = inp->inp_pcbinfo;
1529 uma_zfree(pcbinfo->ipi_zone, inp);
1534 * Temporary wrapper.
1537 in_pcbrele(struct inpcb *inp)
1540 return (in_pcbrele_wlocked(inp));
1544 in_pcblist_rele_rlocked(epoch_context_t ctx)
1546 struct in_pcblist *il;
1548 struct inpcbinfo *pcbinfo;
1551 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1552 pcbinfo = il->il_pcbinfo;
1554 INP_INFO_WLOCK(pcbinfo);
1555 for (i = 0; i < n; i++) {
1556 inp = il->il_inp_list[i];
1558 if (!in_pcbrele_rlocked(inp))
1561 INP_INFO_WUNLOCK(pcbinfo);
1566 inpcbport_free(epoch_context_t ctx)
1568 struct inpcbport *phd;
1570 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1575 in_pcbfree_deferred(epoch_context_t ctx)
1578 int released __unused;
1580 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1583 CURVNET_SET(inp->inp_vnet);
1585 struct ip_moptions *imo = inp->inp_moptions;
1586 inp->inp_moptions = NULL;
1588 /* XXXRW: Do as much as possible here. */
1589 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1590 if (inp->inp_sp != NULL)
1591 ipsec_delete_pcbpolicy(inp);
1594 struct ip6_moptions *im6o = NULL;
1595 if (inp->inp_vflag & INP_IPV6PROTO) {
1596 ip6_freepcbopts(inp->in6p_outputopts);
1597 im6o = inp->in6p_moptions;
1598 inp->in6p_moptions = NULL;
1601 if (inp->inp_options)
1602 (void)m_free(inp->inp_options);
1604 crfree(inp->inp_cred);
1606 mac_inpcb_destroy(inp);
1608 released = in_pcbrele_wlocked(inp);
1611 ip6_freemoptions(im6o);
1614 inp_freemoptions(imo);
1620 * Unconditionally schedule an inpcb to be freed by decrementing its
1621 * reference count, which should occur only after the inpcb has been detached
1622 * from its socket. If another thread holds a temporary reference (acquired
1623 * using in_pcbref()) then the free is deferred until that reference is
1624 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1625 * work, including removal from global lists, is done in this context, where
1626 * the pcbinfo lock is held.
1629 in_pcbfree(struct inpcb *inp)
1631 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1633 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1634 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1635 ("%s: called twice for pcb %p", __func__, inp));
1636 if (inp->inp_flags2 & INP_FREED) {
1641 INP_WLOCK_ASSERT(inp);
1642 INP_LIST_WLOCK(pcbinfo);
1643 in_pcbremlists(inp);
1644 INP_LIST_WUNLOCK(pcbinfo);
1645 RO_INVALIDATE_CACHE(&inp->inp_route);
1646 /* mark as destruction in progress */
1647 inp->inp_flags2 |= INP_FREED;
1649 NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx);
1653 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1654 * port reservation, and preventing it from being returned by inpcb lookups.
1656 * It is used by TCP to mark an inpcb as unused and avoid future packet
1657 * delivery or event notification when a socket remains open but TCP has
1658 * closed. This might occur as a result of a shutdown()-initiated TCP close
1659 * or a RST on the wire, and allows the port binding to be reused while still
1660 * maintaining the invariant that so_pcb always points to a valid inpcb until
1663 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1664 * in_pcbnotifyall() and in_pcbpurgeif0()?
1667 in_pcbdrop(struct inpcb *inp)
1670 INP_WLOCK_ASSERT(inp);
1672 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1673 MPASS(inp->inp_refcount > 1);
1677 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1680 inp->inp_flags |= INP_DROPPED;
1681 if (inp->inp_flags & INP_INHASHLIST) {
1682 struct inpcbport *phd = inp->inp_phd;
1684 INP_HASH_WLOCK(inp->inp_pcbinfo);
1685 in_pcbremlbgrouphash(inp);
1686 CK_LIST_REMOVE(inp, inp_hash);
1687 CK_LIST_REMOVE(inp, inp_portlist);
1688 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1689 CK_LIST_REMOVE(phd, phd_hash);
1690 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
1692 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1693 inp->inp_flags &= ~INP_INHASHLIST;
1695 in_pcbgroup_remove(inp);
1702 * Common routines to return the socket addresses associated with inpcbs.
1705 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1707 struct sockaddr_in *sin;
1709 sin = malloc(sizeof *sin, M_SONAME,
1711 sin->sin_family = AF_INET;
1712 sin->sin_len = sizeof(*sin);
1713 sin->sin_addr = *addr_p;
1714 sin->sin_port = port;
1716 return (struct sockaddr *)sin;
1720 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1723 struct in_addr addr;
1726 inp = sotoinpcb(so);
1727 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1730 port = inp->inp_lport;
1731 addr = inp->inp_laddr;
1734 *nam = in_sockaddr(port, &addr);
1739 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1742 struct in_addr addr;
1745 inp = sotoinpcb(so);
1746 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1749 port = inp->inp_fport;
1750 addr = inp->inp_faddr;
1753 *nam = in_sockaddr(port, &addr);
1758 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1759 struct inpcb *(*notify)(struct inpcb *, int))
1761 struct inpcb *inp, *inp_temp;
1763 INP_INFO_WLOCK(pcbinfo);
1764 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1767 if ((inp->inp_vflag & INP_IPV4) == 0) {
1772 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1773 inp->inp_socket == NULL) {
1777 if ((*notify)(inp, errno))
1780 INP_INFO_WUNLOCK(pcbinfo);
1784 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1787 struct in_multi *inm;
1788 struct in_mfilter *imf;
1789 struct ip_moptions *imo;
1791 INP_INFO_WLOCK(pcbinfo);
1792 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1794 imo = inp->inp_moptions;
1795 if ((inp->inp_vflag & INP_IPV4) &&
1798 * Unselect the outgoing interface if it is being
1801 if (imo->imo_multicast_ifp == ifp)
1802 imo->imo_multicast_ifp = NULL;
1805 * Drop multicast group membership if we joined
1806 * through the interface being detached.
1808 * XXX This can all be deferred to an epoch_call
1811 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1812 if ((inm = imf->imf_inm) == NULL)
1814 if (inm->inm_ifp != ifp)
1816 ip_mfilter_remove(&imo->imo_head, imf);
1817 IN_MULTI_LOCK_ASSERT();
1818 in_leavegroup_locked(inm, NULL);
1819 ip_mfilter_free(imf);
1825 INP_INFO_WUNLOCK(pcbinfo);
1829 * Lookup a PCB based on the local address and port. Caller must hold the
1830 * hash lock. No inpcb locks or references are acquired.
1832 #define INP_LOOKUP_MAPPED_PCB_COST 3
1834 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1835 u_short lport, int lookupflags, struct ucred *cred)
1839 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1845 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1846 ("%s: invalid lookup flags %d", __func__, lookupflags));
1848 INP_HASH_LOCK_ASSERT(pcbinfo);
1850 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1851 struct inpcbhead *head;
1853 * Look for an unconnected (wildcard foreign addr) PCB that
1854 * matches the local address and port we're looking for.
1856 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1857 0, pcbinfo->ipi_hashmask)];
1858 CK_LIST_FOREACH(inp, head, inp_hash) {
1860 /* XXX inp locking */
1861 if ((inp->inp_vflag & INP_IPV4) == 0)
1864 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1865 inp->inp_laddr.s_addr == laddr.s_addr &&
1866 inp->inp_lport == lport) {
1871 prison_equal_ip4(cred->cr_prison,
1872 inp->inp_cred->cr_prison))
1881 struct inpcbporthead *porthash;
1882 struct inpcbport *phd;
1883 struct inpcb *match = NULL;
1885 * Best fit PCB lookup.
1887 * First see if this local port is in use by looking on the
1890 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1891 pcbinfo->ipi_porthashmask)];
1892 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1893 if (phd->phd_port == lport)
1898 * Port is in use by one or more PCBs. Look for best
1901 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1904 !prison_equal_ip4(inp->inp_cred->cr_prison,
1908 /* XXX inp locking */
1909 if ((inp->inp_vflag & INP_IPV4) == 0)
1912 * We never select the PCB that has
1913 * INP_IPV6 flag and is bound to :: if
1914 * we have another PCB which is bound
1915 * to 0.0.0.0. If a PCB has the
1916 * INP_IPV6 flag, then we set its cost
1917 * higher than IPv4 only PCBs.
1919 * Note that the case only happens
1920 * when a socket is bound to ::, under
1921 * the condition that the use of the
1922 * mapped address is allowed.
1924 if ((inp->inp_vflag & INP_IPV6) != 0)
1925 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1927 if (inp->inp_faddr.s_addr != INADDR_ANY)
1929 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1930 if (laddr.s_addr == INADDR_ANY)
1932 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1935 if (laddr.s_addr != INADDR_ANY)
1938 if (wildcard < matchwild) {
1940 matchwild = wildcard;
1949 #undef INP_LOOKUP_MAPPED_PCB_COST
1951 static struct inpcb *
1952 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1953 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1954 uint16_t fport, int lookupflags)
1956 struct inpcb *local_wild;
1957 const struct inpcblbgrouphead *hdr;
1958 struct inpcblbgroup *grp;
1961 INP_HASH_LOCK_ASSERT(pcbinfo);
1963 hdr = &pcbinfo->ipi_lbgrouphashbase[
1964 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1967 * Order of socket selection:
1969 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1972 * - Load balanced group does not contain jailed sockets
1973 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1976 CK_LIST_FOREACH(grp, hdr, il_list) {
1978 if (!(grp->il_vflag & INP_IPV4))
1981 if (grp->il_lport != lport)
1984 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1986 if (grp->il_laddr.s_addr == laddr->s_addr)
1987 return (grp->il_inp[idx]);
1988 if (grp->il_laddr.s_addr == INADDR_ANY &&
1989 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1990 local_wild = grp->il_inp[idx];
1992 return (local_wild);
1997 * Lookup PCB in hash list, using pcbgroup tables.
1999 static struct inpcb *
2000 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2001 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2002 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2004 struct inpcbhead *head;
2005 struct inpcb *inp, *tmpinp;
2006 u_short fport = fport_arg, lport = lport_arg;
2010 * First look for an exact match.
2013 INP_GROUP_LOCK(pcbgroup);
2014 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2015 pcbgroup->ipg_hashmask)];
2016 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2018 /* XXX inp locking */
2019 if ((inp->inp_vflag & INP_IPV4) == 0)
2022 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2023 inp->inp_laddr.s_addr == laddr.s_addr &&
2024 inp->inp_fport == fport &&
2025 inp->inp_lport == lport) {
2027 * XXX We should be able to directly return
2028 * the inp here, without any checks.
2029 * Well unless both bound with SO_REUSEPORT?
2031 if (prison_flag(inp->inp_cred, PR_IP4))
2037 if (tmpinp != NULL) {
2044 * For incoming connections, we may wish to do a wildcard
2045 * match for an RSS-local socket.
2047 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2048 struct inpcb *local_wild = NULL, *local_exact = NULL;
2050 struct inpcb *local_wild_mapped = NULL;
2052 struct inpcb *jail_wild = NULL;
2053 struct inpcbhead *head;
2057 * Order of socket selection - we always prefer jails.
2058 * 1. jailed, non-wild.
2060 * 3. non-jailed, non-wild.
2061 * 4. non-jailed, wild.
2064 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2065 lport, 0, pcbgroup->ipg_hashmask)];
2066 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2068 /* XXX inp locking */
2069 if ((inp->inp_vflag & INP_IPV4) == 0)
2072 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2073 inp->inp_lport != lport)
2076 injail = prison_flag(inp->inp_cred, PR_IP4);
2078 if (prison_check_ip4(inp->inp_cred,
2082 if (local_exact != NULL)
2086 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2091 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2093 /* XXX inp locking, NULL check */
2094 if (inp->inp_vflag & INP_IPV6PROTO)
2095 local_wild_mapped = inp;
2103 } /* LIST_FOREACH */
2112 inp = local_wild_mapped;
2120 * Then look for a wildcard match, if requested.
2122 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2123 struct inpcb *local_wild = NULL, *local_exact = NULL;
2125 struct inpcb *local_wild_mapped = NULL;
2127 struct inpcb *jail_wild = NULL;
2128 struct inpcbhead *head;
2132 * Order of socket selection - we always prefer jails.
2133 * 1. jailed, non-wild.
2135 * 3. non-jailed, non-wild.
2136 * 4. non-jailed, wild.
2138 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2139 0, pcbinfo->ipi_wildmask)];
2140 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2142 /* XXX inp locking */
2143 if ((inp->inp_vflag & INP_IPV4) == 0)
2146 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2147 inp->inp_lport != lport)
2150 injail = prison_flag(inp->inp_cred, PR_IP4);
2152 if (prison_check_ip4(inp->inp_cred,
2156 if (local_exact != NULL)
2160 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2165 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2167 /* XXX inp locking, NULL check */
2168 if (inp->inp_vflag & INP_IPV6PROTO)
2169 local_wild_mapped = inp;
2177 } /* LIST_FOREACH */
2185 inp = local_wild_mapped;
2189 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2190 INP_GROUP_UNLOCK(pcbgroup);
2194 if (lookupflags & INPLOOKUP_WLOCKPCB)
2195 locked = INP_TRY_WLOCK(inp);
2196 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2197 locked = INP_TRY_RLOCK(inp);
2199 panic("%s: locking bug", __func__);
2200 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2201 if (lookupflags & INPLOOKUP_WLOCKPCB)
2208 INP_GROUP_UNLOCK(pcbgroup);
2210 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2212 if (in_pcbrele_wlocked(inp))
2216 if (in_pcbrele_rlocked(inp))
2221 if (lookupflags & INPLOOKUP_WLOCKPCB)
2222 INP_WLOCK_ASSERT(inp);
2224 INP_RLOCK_ASSERT(inp);
2228 #endif /* PCBGROUP */
2231 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2232 * that the caller has locked the hash list, and will not perform any further
2233 * locking or reference operations on either the hash list or the connection.
2235 static struct inpcb *
2236 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2237 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2240 struct inpcbhead *head;
2241 struct inpcb *inp, *tmpinp;
2242 u_short fport = fport_arg, lport = lport_arg;
2244 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2245 ("%s: invalid lookup flags %d", __func__, lookupflags));
2246 INP_HASH_LOCK_ASSERT(pcbinfo);
2249 * First look for an exact match.
2252 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2253 pcbinfo->ipi_hashmask)];
2254 CK_LIST_FOREACH(inp, head, inp_hash) {
2256 /* XXX inp locking */
2257 if ((inp->inp_vflag & INP_IPV4) == 0)
2260 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2261 inp->inp_laddr.s_addr == laddr.s_addr &&
2262 inp->inp_fport == fport &&
2263 inp->inp_lport == lport) {
2265 * XXX We should be able to directly return
2266 * the inp here, without any checks.
2267 * Well unless both bound with SO_REUSEPORT?
2269 if (prison_flag(inp->inp_cred, PR_IP4))
2279 * Then look in lb group (for wildcard match).
2281 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2282 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2283 fport, lookupflags);
2289 * Then look for a wildcard match, if requested.
2291 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2292 struct inpcb *local_wild = NULL, *local_exact = NULL;
2294 struct inpcb *local_wild_mapped = NULL;
2296 struct inpcb *jail_wild = NULL;
2300 * Order of socket selection - we always prefer jails.
2301 * 1. jailed, non-wild.
2303 * 3. non-jailed, non-wild.
2304 * 4. non-jailed, wild.
2307 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2308 0, pcbinfo->ipi_hashmask)];
2309 CK_LIST_FOREACH(inp, head, inp_hash) {
2311 /* XXX inp locking */
2312 if ((inp->inp_vflag & INP_IPV4) == 0)
2315 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2316 inp->inp_lport != lport)
2319 injail = prison_flag(inp->inp_cred, PR_IP4);
2321 if (prison_check_ip4(inp->inp_cred,
2325 if (local_exact != NULL)
2329 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2334 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2336 /* XXX inp locking, NULL check */
2337 if (inp->inp_vflag & INP_IPV6PROTO)
2338 local_wild_mapped = inp;
2346 } /* LIST_FOREACH */
2347 if (jail_wild != NULL)
2349 if (local_exact != NULL)
2350 return (local_exact);
2351 if (local_wild != NULL)
2352 return (local_wild);
2354 if (local_wild_mapped != NULL)
2355 return (local_wild_mapped);
2357 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2363 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2364 * hash list lock, and will return the inpcb locked (i.e., requires
2365 * INPLOOKUP_LOCKPCB).
2367 static struct inpcb *
2368 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2369 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2374 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2375 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2377 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2379 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2383 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2385 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2390 panic("%s: locking bug", __func__);
2393 if (lookupflags & INPLOOKUP_WLOCKPCB)
2394 INP_WLOCK_ASSERT(inp);
2396 INP_RLOCK_ASSERT(inp);
2405 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2406 * from which a pre-calculated hash value may be extracted.
2408 * Possibly more of this logic should be in in_pcbgroup.c.
2411 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2412 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2414 #if defined(PCBGROUP) && !defined(RSS)
2415 struct inpcbgroup *pcbgroup;
2418 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2419 ("%s: invalid lookup flags %d", __func__, lookupflags));
2420 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2421 ("%s: LOCKPCB not set", __func__));
2424 * When not using RSS, use connection groups in preference to the
2425 * reservation table when looking up 4-tuples. When using RSS, just
2426 * use the reservation table, due to the cost of the Toeplitz hash
2429 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2430 * we could be doing RSS with a non-Toeplitz hash that is affordable
2433 #if defined(PCBGROUP) && !defined(RSS)
2434 if (in_pcbgroup_enabled(pcbinfo)) {
2435 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2437 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2438 laddr, lport, lookupflags, ifp));
2441 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2446 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2447 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2448 struct ifnet *ifp, struct mbuf *m)
2451 struct inpcbgroup *pcbgroup;
2454 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2455 ("%s: invalid lookup flags %d", __func__, lookupflags));
2456 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2457 ("%s: LOCKPCB not set", __func__));
2461 * If we can use a hardware-generated hash to look up the connection
2462 * group, use that connection group to find the inpcb. Otherwise
2463 * fall back on a software hash -- or the reservation table if we're
2466 * XXXRW: As above, that policy belongs in the pcbgroup code.
2468 if (in_pcbgroup_enabled(pcbinfo) &&
2469 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2470 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2471 m->m_pkthdr.flowid);
2472 if (pcbgroup != NULL)
2473 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2474 fport, laddr, lport, lookupflags, ifp));
2476 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2478 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2479 laddr, lport, lookupflags, ifp));
2483 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2489 * Insert PCB onto various hash lists.
2492 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2494 struct inpcbhead *pcbhash;
2495 struct inpcbporthead *pcbporthash;
2496 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2497 struct inpcbport *phd;
2498 u_int32_t hashkey_faddr;
2501 INP_WLOCK_ASSERT(inp);
2502 INP_HASH_WLOCK_ASSERT(pcbinfo);
2504 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2505 ("in_pcbinshash: INP_INHASHLIST"));
2508 if (inp->inp_vflag & INP_IPV6)
2509 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2512 hashkey_faddr = inp->inp_faddr.s_addr;
2514 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2515 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2517 pcbporthash = &pcbinfo->ipi_porthashbase[
2518 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2521 * Add entry to load balance group.
2522 * Only do this if SO_REUSEPORT_LB is set.
2524 so_options = inp_so_options(inp);
2525 if (so_options & SO_REUSEPORT_LB) {
2526 int ret = in_pcbinslbgrouphash(inp);
2528 /* pcb lb group malloc fail (ret=ENOBUFS). */
2534 * Go through port list and look for a head for this lport.
2536 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2537 if (phd->phd_port == inp->inp_lport)
2541 * If none exists, malloc one and tack it on.
2544 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2546 return (ENOBUFS); /* XXX */
2548 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2549 phd->phd_port = inp->inp_lport;
2550 CK_LIST_INIT(&phd->phd_pcblist);
2551 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2554 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2555 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2556 inp->inp_flags |= INP_INHASHLIST;
2559 in_pcbgroup_update_mbuf(inp, m);
2561 in_pcbgroup_update(inp);
2568 in_pcbinshash(struct inpcb *inp)
2571 return (in_pcbinshash_internal(inp, NULL));
2575 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2578 return (in_pcbinshash_internal(inp, m));
2582 * Move PCB to the proper hash bucket when { faddr, fport } have been
2583 * changed. NOTE: This does not handle the case of the lport changing (the
2584 * hashed port list would have to be updated as well), so the lport must
2585 * not change after in_pcbinshash() has been called.
2588 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2590 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2591 struct inpcbhead *head;
2592 u_int32_t hashkey_faddr;
2594 INP_WLOCK_ASSERT(inp);
2595 INP_HASH_WLOCK_ASSERT(pcbinfo);
2597 KASSERT(inp->inp_flags & INP_INHASHLIST,
2598 ("in_pcbrehash: !INP_INHASHLIST"));
2601 if (inp->inp_vflag & INP_IPV6)
2602 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2605 hashkey_faddr = inp->inp_faddr.s_addr;
2607 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2608 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2610 CK_LIST_REMOVE(inp, inp_hash);
2611 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2615 in_pcbgroup_update_mbuf(inp, m);
2617 in_pcbgroup_update(inp);
2622 in_pcbrehash(struct inpcb *inp)
2625 in_pcbrehash_mbuf(inp, NULL);
2629 * Remove PCB from various lists.
2632 in_pcbremlists(struct inpcb *inp)
2634 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2636 INP_WLOCK_ASSERT(inp);
2637 INP_LIST_WLOCK_ASSERT(pcbinfo);
2639 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2640 if (inp->inp_flags & INP_INHASHLIST) {
2641 struct inpcbport *phd = inp->inp_phd;
2643 INP_HASH_WLOCK(pcbinfo);
2645 /* XXX: Only do if SO_REUSEPORT_LB set? */
2646 in_pcbremlbgrouphash(inp);
2648 CK_LIST_REMOVE(inp, inp_hash);
2649 CK_LIST_REMOVE(inp, inp_portlist);
2650 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2651 CK_LIST_REMOVE(phd, phd_hash);
2652 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
2654 INP_HASH_WUNLOCK(pcbinfo);
2655 inp->inp_flags &= ~INP_INHASHLIST;
2657 CK_LIST_REMOVE(inp, inp_list);
2658 pcbinfo->ipi_count--;
2660 in_pcbgroup_remove(inp);
2665 * Check for alternatives when higher level complains
2666 * about service problems. For now, invalidate cached
2667 * routing information. If the route was created dynamically
2668 * (by a redirect), time to try a default gateway again.
2671 in_losing(struct inpcb *inp)
2674 RO_INVALIDATE_CACHE(&inp->inp_route);
2679 * A set label operation has occurred at the socket layer, propagate the
2680 * label change into the in_pcb for the socket.
2683 in_pcbsosetlabel(struct socket *so)
2688 inp = sotoinpcb(so);
2689 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2693 mac_inpcb_sosetlabel(so, inp);
2700 * ipport_tick runs once per second, determining if random port allocation
2701 * should be continued. If more than ipport_randomcps ports have been
2702 * allocated in the last second, then we return to sequential port
2703 * allocation. We return to random allocation only once we drop below
2704 * ipport_randomcps for at least ipport_randomtime seconds.
2707 ipport_tick(void *xtp)
2709 VNET_ITERATOR_DECL(vnet_iter);
2711 VNET_LIST_RLOCK_NOSLEEP();
2712 VNET_FOREACH(vnet_iter) {
2713 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2714 if (V_ipport_tcpallocs <=
2715 V_ipport_tcplastcount + V_ipport_randomcps) {
2716 if (V_ipport_stoprandom > 0)
2717 V_ipport_stoprandom--;
2719 V_ipport_stoprandom = V_ipport_randomtime;
2720 V_ipport_tcplastcount = V_ipport_tcpallocs;
2723 VNET_LIST_RUNLOCK_NOSLEEP();
2724 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2731 callout_stop(&ipport_tick_callout);
2735 * The ipport_callout should start running at about the time we attach the
2736 * inet or inet6 domains.
2739 ipport_tick_init(const void *unused __unused)
2742 /* Start ipport_tick. */
2743 callout_init(&ipport_tick_callout, 1);
2744 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2745 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2746 SHUTDOWN_PRI_DEFAULT);
2748 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2749 ipport_tick_init, NULL);
2752 inp_wlock(struct inpcb *inp)
2759 inp_wunlock(struct inpcb *inp)
2766 inp_rlock(struct inpcb *inp)
2773 inp_runlock(struct inpcb *inp)
2779 #ifdef INVARIANT_SUPPORT
2781 inp_lock_assert(struct inpcb *inp)
2784 INP_WLOCK_ASSERT(inp);
2788 inp_unlock_assert(struct inpcb *inp)
2791 INP_UNLOCK_ASSERT(inp);
2796 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2800 INP_INFO_WLOCK(&V_tcbinfo);
2801 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2806 INP_INFO_WUNLOCK(&V_tcbinfo);
2810 inp_inpcbtosocket(struct inpcb *inp)
2813 INP_WLOCK_ASSERT(inp);
2814 return (inp->inp_socket);
2818 inp_inpcbtotcpcb(struct inpcb *inp)
2821 INP_WLOCK_ASSERT(inp);
2822 return ((struct tcpcb *)inp->inp_ppcb);
2826 inp_ip_tos_get(const struct inpcb *inp)
2829 return (inp->inp_ip_tos);
2833 inp_ip_tos_set(struct inpcb *inp, int val)
2836 inp->inp_ip_tos = val;
2840 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2841 uint32_t *faddr, uint16_t *fp)
2844 INP_LOCK_ASSERT(inp);
2845 *laddr = inp->inp_laddr.s_addr;
2846 *faddr = inp->inp_faddr.s_addr;
2847 *lp = inp->inp_lport;
2848 *fp = inp->inp_fport;
2852 so_sotoinpcb(struct socket *so)
2855 return (sotoinpcb(so));
2859 so_sototcpcb(struct socket *so)
2862 return (sototcpcb(so));
2866 * Create an external-format (``xinpcb'') structure using the information in
2867 * the kernel-format in_pcb structure pointed to by inp. This is done to
2868 * reduce the spew of irrelevant information over this interface, to isolate
2869 * user code from changes in the kernel structure, and potentially to provide
2870 * information-hiding if we decide that some of this information should be
2871 * hidden from users.
2874 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2877 bzero(xi, sizeof(*xi));
2878 xi->xi_len = sizeof(struct xinpcb);
2879 if (inp->inp_socket)
2880 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2881 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2882 xi->inp_gencnt = inp->inp_gencnt;
2883 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2884 xi->inp_flow = inp->inp_flow;
2885 xi->inp_flowid = inp->inp_flowid;
2886 xi->inp_flowtype = inp->inp_flowtype;
2887 xi->inp_flags = inp->inp_flags;
2888 xi->inp_flags2 = inp->inp_flags2;
2889 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2890 xi->in6p_cksum = inp->in6p_cksum;
2891 xi->in6p_hops = inp->in6p_hops;
2892 xi->inp_ip_tos = inp->inp_ip_tos;
2893 xi->inp_vflag = inp->inp_vflag;
2894 xi->inp_ip_ttl = inp->inp_ip_ttl;
2895 xi->inp_ip_p = inp->inp_ip_p;
2896 xi->inp_ip_minttl = inp->inp_ip_minttl;
2901 db_print_indent(int indent)
2905 for (i = 0; i < indent; i++)
2910 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2912 char faddr_str[48], laddr_str[48];
2914 db_print_indent(indent);
2915 db_printf("%s at %p\n", name, inc);
2920 if (inc->inc_flags & INC_ISIPV6) {
2922 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2923 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2928 inet_ntoa_r(inc->inc_laddr, laddr_str);
2929 inet_ntoa_r(inc->inc_faddr, faddr_str);
2931 db_print_indent(indent);
2932 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2933 ntohs(inc->inc_lport));
2934 db_print_indent(indent);
2935 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2936 ntohs(inc->inc_fport));
2940 db_print_inpflags(int inp_flags)
2945 if (inp_flags & INP_RECVOPTS) {
2946 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2949 if (inp_flags & INP_RECVRETOPTS) {
2950 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2953 if (inp_flags & INP_RECVDSTADDR) {
2954 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2957 if (inp_flags & INP_ORIGDSTADDR) {
2958 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2961 if (inp_flags & INP_HDRINCL) {
2962 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2965 if (inp_flags & INP_HIGHPORT) {
2966 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2969 if (inp_flags & INP_LOWPORT) {
2970 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2973 if (inp_flags & INP_ANONPORT) {
2974 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2977 if (inp_flags & INP_RECVIF) {
2978 db_printf("%sINP_RECVIF", comma ? ", " : "");
2981 if (inp_flags & INP_MTUDISC) {
2982 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2985 if (inp_flags & INP_RECVTTL) {
2986 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2989 if (inp_flags & INP_DONTFRAG) {
2990 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2993 if (inp_flags & INP_RECVTOS) {
2994 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2997 if (inp_flags & IN6P_IPV6_V6ONLY) {
2998 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3001 if (inp_flags & IN6P_PKTINFO) {
3002 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3005 if (inp_flags & IN6P_HOPLIMIT) {
3006 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3009 if (inp_flags & IN6P_HOPOPTS) {
3010 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3013 if (inp_flags & IN6P_DSTOPTS) {
3014 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3017 if (inp_flags & IN6P_RTHDR) {
3018 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3021 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3022 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3025 if (inp_flags & IN6P_TCLASS) {
3026 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3029 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3030 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3033 if (inp_flags & INP_TIMEWAIT) {
3034 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3037 if (inp_flags & INP_ONESBCAST) {
3038 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3041 if (inp_flags & INP_DROPPED) {
3042 db_printf("%sINP_DROPPED", comma ? ", " : "");
3045 if (inp_flags & INP_SOCKREF) {
3046 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3049 if (inp_flags & IN6P_RFC2292) {
3050 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3053 if (inp_flags & IN6P_MTU) {
3054 db_printf("IN6P_MTU%s", comma ? ", " : "");
3060 db_print_inpvflag(u_char inp_vflag)
3065 if (inp_vflag & INP_IPV4) {
3066 db_printf("%sINP_IPV4", comma ? ", " : "");
3069 if (inp_vflag & INP_IPV6) {
3070 db_printf("%sINP_IPV6", comma ? ", " : "");
3073 if (inp_vflag & INP_IPV6PROTO) {
3074 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3080 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3083 db_print_indent(indent);
3084 db_printf("%s at %p\n", name, inp);
3088 db_print_indent(indent);
3089 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3091 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3093 db_print_indent(indent);
3094 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3095 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3097 db_print_indent(indent);
3098 db_printf("inp_label: %p inp_flags: 0x%x (",
3099 inp->inp_label, inp->inp_flags);
3100 db_print_inpflags(inp->inp_flags);
3103 db_print_indent(indent);
3104 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3106 db_print_inpvflag(inp->inp_vflag);
3109 db_print_indent(indent);
3110 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3111 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3113 db_print_indent(indent);
3115 if (inp->inp_vflag & INP_IPV6) {
3116 db_printf("in6p_options: %p in6p_outputopts: %p "
3117 "in6p_moptions: %p\n", inp->in6p_options,
3118 inp->in6p_outputopts, inp->in6p_moptions);
3119 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3120 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3125 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3126 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3127 inp->inp_options, inp->inp_moptions);
3130 db_print_indent(indent);
3131 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3132 (uintmax_t)inp->inp_gencnt);
3135 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3140 db_printf("usage: show inpcb <addr>\n");
3143 inp = (struct inpcb *)addr;
3145 db_print_inpcb(inp, "inpcb", 0);
3151 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3155 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3157 union if_snd_tag_modify_params params = {
3158 .rate_limit.max_rate = max_pacing_rate,
3159 .rate_limit.flags = M_NOWAIT,
3161 struct m_snd_tag *mst;
3165 mst = inp->inp_snd_tag;
3173 if (ifp->if_snd_tag_modify == NULL) {
3176 error = ifp->if_snd_tag_modify(mst, ¶ms);
3182 * Query existing TX rate limit based on the existing
3183 * "inp->inp_snd_tag", if any.
3186 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3188 union if_snd_tag_query_params params = { };
3189 struct m_snd_tag *mst;
3193 mst = inp->inp_snd_tag;
3201 if (ifp->if_snd_tag_query == NULL) {
3204 error = ifp->if_snd_tag_query(mst, ¶ms);
3205 if (error == 0 && p_max_pacing_rate != NULL)
3206 *p_max_pacing_rate = params.rate_limit.max_rate;
3212 * Query existing TX queue level based on the existing
3213 * "inp->inp_snd_tag", if any.
3216 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
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)
3232 return (EOPNOTSUPP);
3234 error = ifp->if_snd_tag_query(mst, ¶ms);
3235 if (error == 0 && p_txqueue_level != NULL)
3236 *p_txqueue_level = params.rate_limit.queue_level;
3241 * Allocate a new TX rate limit send tag from the network interface
3242 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3245 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3246 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3249 union if_snd_tag_alloc_params params = {
3250 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3251 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3252 .rate_limit.hdr.flowid = flowid,
3253 .rate_limit.hdr.flowtype = flowtype,
3254 .rate_limit.max_rate = max_pacing_rate,
3255 .rate_limit.flags = M_NOWAIT,
3259 INP_WLOCK_ASSERT(inp);
3264 if (ifp->if_snd_tag_alloc == NULL) {
3267 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3271 counter_u64_add(rate_limit_set_ok, 1);
3272 counter_u64_add(rate_limit_active, 1);
3274 counter_u64_add(rate_limit_alloc_fail, 1);
3281 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3287 * If the device was detached while we still had reference(s)
3288 * on the ifp, we assume if_snd_tag_free() was replaced with
3291 ifp->if_snd_tag_free(mst);
3293 /* release reference count on network interface */
3296 counter_u64_add(rate_limit_active, -1);
3301 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3305 in_pcbdetach_txrtlmt(struct inpcb *inp)
3307 struct m_snd_tag *mst;
3309 INP_WLOCK_ASSERT(inp);
3311 mst = inp->inp_snd_tag;
3312 inp->inp_snd_tag = NULL;
3317 m_snd_tag_rele(mst);
3321 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3326 * If the existing send tag is for the wrong interface due to
3327 * a route change, first drop the existing tag. Set the
3328 * CHANGED flag so that we will keep trying to allocate a new
3329 * tag if we fail to allocate one this time.
3331 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3332 in_pcbdetach_txrtlmt(inp);
3333 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3337 * NOTE: When attaching to a network interface a reference is
3338 * made to ensure the network interface doesn't go away until
3339 * all ratelimit connections are gone. The network interface
3340 * pointers compared below represent valid network interfaces,
3341 * except when comparing towards NULL.
3343 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3345 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3346 if (inp->inp_snd_tag != NULL)
3347 in_pcbdetach_txrtlmt(inp);
3349 } else if (inp->inp_snd_tag == NULL) {
3351 * In order to utilize packet pacing with RSS, we need
3352 * to wait until there is a valid RSS hash before we
3355 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3358 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3359 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3362 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3364 if (error == 0 || error == EOPNOTSUPP)
3365 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3371 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3372 * is set in the fast path and will attach/detach/modify the TX rate
3373 * limit send tag based on the socket's so_max_pacing_rate value.
3376 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3378 struct socket *socket;
3379 uint32_t max_pacing_rate;
3386 socket = inp->inp_socket;
3390 if (!INP_WLOCKED(inp)) {
3392 * NOTE: If the write locking fails, we need to bail
3393 * out and use the non-ratelimited ring for the
3394 * transmit until there is a new chance to get the
3397 if (!INP_TRY_UPGRADE(inp))
3405 * NOTE: The so_max_pacing_rate value is read unlocked,
3406 * because atomic updates are not required since the variable
3407 * is checked at every mbuf we send. It is assumed that the
3408 * variable read itself will be atomic.
3410 max_pacing_rate = socket->so_max_pacing_rate;
3412 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3419 * Track route changes for TX rate limiting.
3422 in_pcboutput_eagain(struct inpcb *inp)
3429 if (inp->inp_snd_tag == NULL)
3432 if (!INP_WLOCKED(inp)) {
3434 * NOTE: If the write locking fails, we need to bail
3435 * out and use the non-ratelimited ring for the
3436 * transmit until there is a new chance to get the
3439 if (!INP_TRY_UPGRADE(inp))
3446 /* detach rate limiting */
3447 in_pcbdetach_txrtlmt(inp);
3449 /* make sure new mbuf send tag allocation is made */
3450 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3460 rate_limit_active = counter_u64_alloc(M_WAITOK);
3461 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3462 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3465 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3467 #endif /* RATELIMIT */