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,
176 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
180 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
181 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
184 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
185 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
189 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
191 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
192 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
193 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
195 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
196 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
197 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
199 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
200 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
201 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
203 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
204 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
205 &VNET_NAME(ipport_reservedhigh), 0, "");
206 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
207 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
209 CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
211 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
212 CTLFLAG_VNET | CTLFLAG_RW,
213 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
214 "allocations before switching to a sequental one");
215 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
216 CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(ipport_randomtime), 0,
218 "Minimum time to keep sequental port "
219 "allocation before switching to a random one");
222 counter_u64_t rate_limit_active;
223 counter_u64_t rate_limit_alloc_fail;
224 counter_u64_t rate_limit_set_ok;
226 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
228 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
229 &rate_limit_active, "Active rate limited connections");
230 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
231 &rate_limit_alloc_fail, "Rate limited connection failures");
232 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
233 &rate_limit_set_ok, "Rate limited setting succeeded");
234 #endif /* RATELIMIT */
239 * in_pcb.c: manage the Protocol Control Blocks.
241 * NOTE: It is assumed that most of these functions will be called with
242 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
243 * functions often modify hash chains or addresses in pcbs.
246 static struct inpcblbgroup *
247 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
248 uint16_t port, const union in_dependaddr *addr, int size)
250 struct inpcblbgroup *grp;
253 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
254 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
257 grp->il_vflag = vflag;
258 grp->il_lport = port;
259 grp->il_dependladdr = *addr;
260 grp->il_inpsiz = size;
261 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
266 in_pcblbgroup_free_deferred(epoch_context_t ctx)
268 struct inpcblbgroup *grp;
270 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
275 in_pcblbgroup_free(struct inpcblbgroup *grp)
278 CK_LIST_REMOVE(grp, il_list);
279 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
282 static struct inpcblbgroup *
283 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
284 struct inpcblbgroup *old_grp, int size)
286 struct inpcblbgroup *grp;
289 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
290 old_grp->il_lport, &old_grp->il_dependladdr, size);
294 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
295 ("invalid new local group size %d and old local group count %d",
296 grp->il_inpsiz, old_grp->il_inpcnt));
298 for (i = 0; i < old_grp->il_inpcnt; ++i)
299 grp->il_inp[i] = old_grp->il_inp[i];
300 grp->il_inpcnt = old_grp->il_inpcnt;
301 in_pcblbgroup_free(old_grp);
306 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
307 * and shrink group if possible.
310 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
313 struct inpcblbgroup *grp, *new_grp;
316 for (; i + 1 < grp->il_inpcnt; ++i)
317 grp->il_inp[i] = grp->il_inp[i + 1];
320 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
321 grp->il_inpcnt <= grp->il_inpsiz / 4) {
322 /* Shrink this group. */
323 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
330 * Add PCB to load balance group for SO_REUSEPORT_LB option.
333 in_pcbinslbgrouphash(struct inpcb *inp)
335 const static struct timeval interval = { 60, 0 };
336 static struct timeval lastprint;
337 struct inpcbinfo *pcbinfo;
338 struct inpcblbgrouphead *hdr;
339 struct inpcblbgroup *grp;
342 pcbinfo = inp->inp_pcbinfo;
344 INP_WLOCK_ASSERT(inp);
345 INP_HASH_WLOCK_ASSERT(pcbinfo);
348 * Don't allow jailed socket to join local group.
350 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
355 * Don't allow IPv4 mapped INET6 wild socket.
357 if ((inp->inp_vflag & INP_IPV4) &&
358 inp->inp_laddr.s_addr == INADDR_ANY &&
359 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
364 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
365 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
366 CK_LIST_FOREACH(grp, hdr, il_list) {
367 if (grp->il_vflag == inp->inp_vflag &&
368 grp->il_lport == inp->inp_lport &&
369 memcmp(&grp->il_dependladdr,
370 &inp->inp_inc.inc_ie.ie_dependladdr,
371 sizeof(grp->il_dependladdr)) == 0)
375 /* Create new load balance group. */
376 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
377 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
378 INPCBLBGROUP_SIZMIN);
381 } else if (grp->il_inpcnt == grp->il_inpsiz) {
382 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
383 if (ratecheck(&lastprint, &interval))
384 printf("lb group port %d, limit reached\n",
385 ntohs(grp->il_lport));
389 /* Expand this local group. */
390 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
395 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
396 ("invalid local group size %d and count %d", grp->il_inpsiz,
399 grp->il_inp[grp->il_inpcnt] = inp;
405 * Remove PCB from load balance group.
408 in_pcbremlbgrouphash(struct inpcb *inp)
410 struct inpcbinfo *pcbinfo;
411 struct inpcblbgrouphead *hdr;
412 struct inpcblbgroup *grp;
415 pcbinfo = inp->inp_pcbinfo;
417 INP_WLOCK_ASSERT(inp);
418 INP_HASH_WLOCK_ASSERT(pcbinfo);
420 hdr = &pcbinfo->ipi_lbgrouphashbase[
421 INP_PCBPORTHASH(inp->inp_lport, 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 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
463 INP_INFO_LOCK_INIT(pcbinfo, name);
464 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
465 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
467 pcbinfo->ipi_vnet = curvnet;
469 pcbinfo->ipi_listhead = listhead;
470 CK_LIST_INIT(pcbinfo->ipi_listhead);
471 pcbinfo->ipi_count = 0;
472 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
473 &pcbinfo->ipi_hashmask);
474 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
475 &pcbinfo->ipi_porthashmask);
476 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
477 &pcbinfo->ipi_lbgrouphashmask);
479 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
481 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
482 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
483 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
484 uma_zone_set_warning(pcbinfo->ipi_zone,
485 "kern.ipc.maxsockets limit reached");
489 * Destroy an inpcbinfo.
492 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
495 KASSERT(pcbinfo->ipi_count == 0,
496 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
498 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
499 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
500 pcbinfo->ipi_porthashmask);
501 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
502 pcbinfo->ipi_lbgrouphashmask);
504 in_pcbgroup_destroy(pcbinfo);
506 uma_zdestroy(pcbinfo->ipi_zone);
507 INP_LIST_LOCK_DESTROY(pcbinfo);
508 INP_HASH_LOCK_DESTROY(pcbinfo);
509 INP_INFO_LOCK_DESTROY(pcbinfo);
513 * Allocate a PCB and associate it with the socket.
514 * On success return with the PCB locked.
517 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
523 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
526 bzero(&inp->inp_start_zero, inp_zero_size);
528 inp->inp_numa_domain = M_NODOM;
530 inp->inp_pcbinfo = pcbinfo;
531 inp->inp_socket = so;
532 inp->inp_cred = crhold(so->so_cred);
533 inp->inp_inc.inc_fibnum = so->so_fibnum;
535 error = mac_inpcb_init(inp, M_NOWAIT);
538 mac_inpcb_create(so, inp);
540 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
541 error = ipsec_init_pcbpolicy(inp);
544 mac_inpcb_destroy(inp);
550 if (INP_SOCKAF(so) == AF_INET6) {
551 inp->inp_vflag |= INP_IPV6PROTO;
553 inp->inp_flags |= IN6P_IPV6_V6ONLY;
557 INP_LIST_WLOCK(pcbinfo);
558 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
559 pcbinfo->ipi_count++;
560 so->so_pcb = (caddr_t)inp;
562 if (V_ip6_auto_flowlabel)
563 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
565 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
566 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
569 * Routes in inpcb's can cache L2 as well; they are guaranteed
572 inp->inp_route.ro_flags = RT_LLE_CACHE;
573 INP_LIST_WUNLOCK(pcbinfo);
574 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
577 crfree(inp->inp_cred);
578 uma_zfree(pcbinfo->ipi_zone, inp);
586 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
590 INP_WLOCK_ASSERT(inp);
591 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
593 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
595 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
596 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
597 &inp->inp_lport, cred);
600 if (in_pcbinshash(inp) != 0) {
601 inp->inp_laddr.s_addr = INADDR_ANY;
606 inp->inp_flags |= INP_ANONPORT;
612 * Select a local port (number) to use.
614 #if defined(INET) || defined(INET6)
616 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
617 struct ucred *cred, int lookupflags)
619 struct inpcbinfo *pcbinfo;
620 struct inpcb *tmpinp;
621 unsigned short *lastport;
622 int count, dorandom, error;
623 u_short aux, first, last, lport;
625 struct in_addr laddr;
628 pcbinfo = inp->inp_pcbinfo;
631 * Because no actual state changes occur here, a global write lock on
632 * the pcbinfo isn't required.
634 INP_LOCK_ASSERT(inp);
635 INP_HASH_LOCK_ASSERT(pcbinfo);
637 if (inp->inp_flags & INP_HIGHPORT) {
638 first = V_ipport_hifirstauto; /* sysctl */
639 last = V_ipport_hilastauto;
640 lastport = &pcbinfo->ipi_lasthi;
641 } else if (inp->inp_flags & INP_LOWPORT) {
642 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
645 first = V_ipport_lowfirstauto; /* 1023 */
646 last = V_ipport_lowlastauto; /* 600 */
647 lastport = &pcbinfo->ipi_lastlow;
649 first = V_ipport_firstauto; /* sysctl */
650 last = V_ipport_lastauto;
651 lastport = &pcbinfo->ipi_lastport;
654 * For UDP(-Lite), use random port allocation as long as the user
655 * allows it. For TCP (and as of yet unknown) connections,
656 * use random port allocation only if the user allows it AND
657 * ipport_tick() allows it.
659 if (V_ipport_randomized &&
660 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
661 pcbinfo == &V_ulitecbinfo))
666 * It makes no sense to do random port allocation if
667 * we have the only port available.
671 /* Make sure to not include UDP(-Lite) packets in the count. */
672 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
673 V_ipport_tcpallocs++;
675 * Instead of having two loops further down counting up or down
676 * make sure that first is always <= last and go with only one
677 * code path implementing all logic.
686 /* Make the compiler happy. */
688 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
689 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
694 tmpinp = NULL; /* Make compiler happy. */
698 *lastport = first + (arc4random() % (last - first));
700 count = last - first;
703 if (count-- < 0) /* completely used? */
704 return (EADDRNOTAVAIL);
706 if (*lastport < first || *lastport > last)
708 lport = htons(*lastport);
711 if ((inp->inp_vflag & INP_IPV6) != 0)
712 tmpinp = in6_pcblookup_local(pcbinfo,
713 &inp->in6p_laddr, lport, lookupflags, cred);
715 #if defined(INET) && defined(INET6)
719 tmpinp = in_pcblookup_local(pcbinfo, laddr,
720 lport, lookupflags, cred);
722 } while (tmpinp != NULL);
725 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
726 laddrp->s_addr = laddr.s_addr;
734 * Return cached socket options.
737 inp_so_options(const struct inpcb *inp)
743 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
744 so_options |= SO_REUSEPORT_LB;
745 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
746 so_options |= SO_REUSEPORT;
747 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
748 so_options |= SO_REUSEADDR;
751 #endif /* INET || INET6 */
754 * Check if a new BINDMULTI socket is allowed to be created.
756 * ni points to the new inp.
757 * oi points to the exisitng inp.
759 * This checks whether the existing inp also has BINDMULTI and
760 * whether the credentials match.
763 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
765 /* Check permissions match */
766 if ((ni->inp_flags2 & INP_BINDMULTI) &&
767 (ni->inp_cred->cr_uid !=
768 oi->inp_cred->cr_uid))
771 /* Check the existing inp has BINDMULTI set */
772 if ((ni->inp_flags2 & INP_BINDMULTI) &&
773 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
777 * We're okay - either INP_BINDMULTI isn't set on ni, or
778 * it is and it matches the checks.
785 * Set up a bind operation on a PCB, performing port allocation
786 * as required, but do not actually modify the PCB. Callers can
787 * either complete the bind by setting inp_laddr/inp_lport and
788 * calling in_pcbinshash(), or they can just use the resulting
789 * port and address to authorise the sending of a once-off packet.
791 * On error, the values of *laddrp and *lportp are not changed.
794 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
795 u_short *lportp, struct ucred *cred)
797 struct socket *so = inp->inp_socket;
798 struct sockaddr_in *sin;
799 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
800 struct in_addr laddr;
802 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
806 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
807 * so that we don't have to add to the (already messy) code below.
809 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
812 * No state changes, so read locks are sufficient here.
814 INP_LOCK_ASSERT(inp);
815 INP_HASH_LOCK_ASSERT(pcbinfo);
817 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
818 return (EADDRNOTAVAIL);
819 laddr.s_addr = *laddrp;
820 if (nam != NULL && laddr.s_addr != INADDR_ANY)
822 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
823 lookupflags = INPLOOKUP_WILDCARD;
825 if ((error = prison_local_ip4(cred, &laddr)) != 0)
828 sin = (struct sockaddr_in *)nam;
829 if (nam->sa_len != sizeof (*sin))
833 * We should check the family, but old programs
834 * incorrectly fail to initialize it.
836 if (sin->sin_family != AF_INET)
837 return (EAFNOSUPPORT);
839 error = prison_local_ip4(cred, &sin->sin_addr);
842 if (sin->sin_port != *lportp) {
843 /* Don't allow the port to change. */
846 lport = sin->sin_port;
848 /* NB: lport is left as 0 if the port isn't being changed. */
849 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
851 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
852 * allow complete duplication of binding if
853 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
854 * and a multicast address is bound on both
855 * new and duplicated sockets.
857 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
858 reuseport = SO_REUSEADDR|SO_REUSEPORT;
860 * XXX: How to deal with SO_REUSEPORT_LB here?
861 * Treat same as SO_REUSEPORT for now.
863 if ((so->so_options &
864 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
865 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
866 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
867 sin->sin_port = 0; /* yech... */
868 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
870 * Is the address a local IP address?
871 * If INP_BINDANY is set, then the socket may be bound
872 * to any endpoint address, local or not.
874 if ((inp->inp_flags & INP_BINDANY) == 0 &&
875 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
876 return (EADDRNOTAVAIL);
878 laddr = sin->sin_addr;
884 if (ntohs(lport) <= V_ipport_reservedhigh &&
885 ntohs(lport) >= V_ipport_reservedlow &&
886 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
888 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
889 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
890 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
891 lport, INPLOOKUP_WILDCARD, cred);
894 * This entire block sorely needs a rewrite.
897 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
898 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
899 (so->so_type != SOCK_STREAM ||
900 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
901 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
902 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
903 (t->inp_flags2 & INP_REUSEPORT) ||
904 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
905 (inp->inp_cred->cr_uid !=
906 t->inp_cred->cr_uid))
910 * If the socket is a BINDMULTI socket, then
911 * the credentials need to match and the
912 * original socket also has to have been bound
915 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
918 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
919 lport, lookupflags, cred);
920 if (t && (t->inp_flags & INP_TIMEWAIT)) {
922 * XXXRW: If an incpb has had its timewait
923 * state recycled, we treat the address as
924 * being in use (for now). This is better
925 * than a panic, but not desirable.
929 ((reuseport & tw->tw_so_options) == 0 &&
931 tw->tw_so_options) == 0)) {
935 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
936 (reuseport & inp_so_options(t)) == 0 &&
937 (reuseport_lb & inp_so_options(t)) == 0) {
939 if (ntohl(sin->sin_addr.s_addr) !=
941 ntohl(t->inp_laddr.s_addr) !=
943 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
944 (t->inp_vflag & INP_IPV6PROTO) == 0)
947 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
955 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
960 *laddrp = laddr.s_addr;
966 * Connect from a socket to a specified address.
967 * Both address and port must be specified in argument sin.
968 * If don't have a local address for this socket yet,
972 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
973 struct ucred *cred, struct mbuf *m, bool rehash)
975 u_short lport, fport;
976 in_addr_t laddr, faddr;
979 INP_WLOCK_ASSERT(inp);
980 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
982 lport = inp->inp_lport;
983 laddr = inp->inp_laddr.s_addr;
984 anonport = (lport == 0);
985 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
990 /* Do the initial binding of the local address if required. */
991 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
992 KASSERT(rehash == true,
993 ("Rehashing required for unbound inps"));
994 inp->inp_lport = lport;
995 inp->inp_laddr.s_addr = laddr;
996 if (in_pcbinshash(inp) != 0) {
997 inp->inp_laddr.s_addr = INADDR_ANY;
1003 /* Commit the remaining changes. */
1004 inp->inp_lport = lport;
1005 inp->inp_laddr.s_addr = laddr;
1006 inp->inp_faddr.s_addr = faddr;
1007 inp->inp_fport = fport;
1009 in_pcbrehash_mbuf(inp, m);
1011 in_pcbinshash_mbuf(inp, m);
1015 inp->inp_flags |= INP_ANONPORT;
1020 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1023 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1027 * Do proper source address selection on an unbound socket in case
1028 * of connect. Take jails into account as well.
1031 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1035 struct sockaddr *sa;
1036 struct sockaddr_in *sin;
1041 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1043 * Bypass source address selection and use the primary jail IP
1046 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1050 bzero(&sro, sizeof(sro));
1052 sin = (struct sockaddr_in *)&sro.ro_dst;
1053 sin->sin_family = AF_INET;
1054 sin->sin_len = sizeof(struct sockaddr_in);
1055 sin->sin_addr.s_addr = faddr->s_addr;
1058 * If route is known our src addr is taken from the i/f,
1061 * Find out route to destination.
1063 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1064 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1067 * If we found a route, use the address corresponding to
1068 * the outgoing interface.
1070 * Otherwise assume faddr is reachable on a directly connected
1071 * network and try to find a corresponding interface to take
1072 * the source address from.
1074 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1075 struct in_ifaddr *ia;
1078 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1079 inp->inp_socket->so_fibnum));
1081 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1082 inp->inp_socket->so_fibnum));
1086 error = ENETUNREACH;
1090 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1091 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1097 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1100 if (sa->sa_family != AF_INET)
1102 sin = (struct sockaddr_in *)sa;
1103 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1104 ia = (struct in_ifaddr *)ifa;
1109 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1113 /* 3. As a last resort return the 'default' jail address. */
1114 error = prison_get_ip4(cred, laddr);
1119 * If the outgoing interface on the route found is not
1120 * a loopback interface, use the address from that interface.
1121 * In case of jails do those three steps:
1122 * 1. check if the interface address belongs to the jail. If so use it.
1123 * 2. check if we have any address on the outgoing interface
1124 * belonging to this jail. If so use it.
1125 * 3. as a last resort return the 'default' jail address.
1127 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1128 struct in_ifaddr *ia;
1131 /* If not jailed, use the default returned. */
1132 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1133 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1134 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1139 /* 1. Check if the iface address belongs to the jail. */
1140 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1141 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1142 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1143 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1148 * 2. Check if we have any address on the outgoing interface
1149 * belonging to this jail.
1152 ifp = sro.ro_rt->rt_ifp;
1153 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1155 if (sa->sa_family != AF_INET)
1157 sin = (struct sockaddr_in *)sa;
1158 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1159 ia = (struct in_ifaddr *)ifa;
1164 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1168 /* 3. As a last resort return the 'default' jail address. */
1169 error = prison_get_ip4(cred, laddr);
1174 * The outgoing interface is marked with 'loopback net', so a route
1175 * to ourselves is here.
1176 * Try to find the interface of the destination address and then
1177 * take the address from there. That interface is not necessarily
1178 * a loopback interface.
1179 * In case of jails, check that it is an address of the jail
1180 * and if we cannot find, fall back to the 'default' jail address.
1182 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1183 struct sockaddr_in sain;
1184 struct in_ifaddr *ia;
1186 bzero(&sain, sizeof(struct sockaddr_in));
1187 sain.sin_family = AF_INET;
1188 sain.sin_len = sizeof(struct sockaddr_in);
1189 sain.sin_addr.s_addr = faddr->s_addr;
1191 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1192 inp->inp_socket->so_fibnum));
1194 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1195 inp->inp_socket->so_fibnum));
1197 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1199 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1201 error = ENETUNREACH;
1204 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1214 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1216 if (sa->sa_family != AF_INET)
1218 sin = (struct sockaddr_in *)sa;
1219 if (prison_check_ip4(cred,
1220 &sin->sin_addr) == 0) {
1221 ia = (struct in_ifaddr *)ifa;
1226 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1231 /* 3. As a last resort return the 'default' jail address. */
1232 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.
1275 INP_LOCK_ASSERT(inp);
1276 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1280 if (nam->sa_len != sizeof (*sin))
1282 if (sin->sin_family != AF_INET)
1283 return (EAFNOSUPPORT);
1284 if (sin->sin_port == 0)
1285 return (EADDRNOTAVAIL);
1286 laddr.s_addr = *laddrp;
1288 faddr = sin->sin_addr;
1289 fport = sin->sin_port;
1291 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1293 * If the destination address is INADDR_ANY,
1294 * use the primary local address.
1295 * If the supplied address is INADDR_BROADCAST,
1296 * and the primary interface supports broadcast,
1297 * choose the broadcast address for that interface.
1299 if (faddr.s_addr == INADDR_ANY) {
1300 IN_IFADDR_RLOCK(&in_ifa_tracker);
1302 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1303 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1305 (error = prison_get_ip4(cred, &faddr)) != 0)
1307 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1308 IN_IFADDR_RLOCK(&in_ifa_tracker);
1309 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1311 faddr = satosin(&CK_STAILQ_FIRST(
1312 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1313 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1316 if (laddr.s_addr == INADDR_ANY) {
1317 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1319 * If the destination address is multicast and an outgoing
1320 * interface has been set as a multicast option, prefer the
1321 * address of that interface as our source address.
1323 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1324 inp->inp_moptions != NULL) {
1325 struct ip_moptions *imo;
1328 imo = inp->inp_moptions;
1329 if (imo->imo_multicast_ifp != NULL) {
1330 ifp = imo->imo_multicast_ifp;
1331 IN_IFADDR_RLOCK(&in_ifa_tracker);
1332 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1333 if ((ia->ia_ifp == ifp) &&
1335 prison_check_ip4(cred,
1336 &ia->ia_addr.sin_addr) == 0))
1340 error = EADDRNOTAVAIL;
1342 laddr = ia->ia_addr.sin_addr;
1345 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1351 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1352 laddr, lport, 0, NULL);
1356 return (EADDRINUSE);
1359 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1364 *laddrp = laddr.s_addr;
1366 *faddrp = faddr.s_addr;
1372 in_pcbdisconnect(struct inpcb *inp)
1375 INP_WLOCK_ASSERT(inp);
1376 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1378 inp->inp_faddr.s_addr = INADDR_ANY;
1385 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1386 * For most protocols, this will be invoked immediately prior to calling
1387 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1388 * socket, in which case in_pcbfree() is deferred.
1391 in_pcbdetach(struct inpcb *inp)
1394 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1397 if (inp->inp_snd_tag != NULL)
1398 in_pcbdetach_txrtlmt(inp);
1400 inp->inp_socket->so_pcb = NULL;
1401 inp->inp_socket = NULL;
1405 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1406 * stability of an inpcb pointer despite the inpcb lock being released. This
1407 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1408 * but where the inpcb lock may already held, or when acquiring a reference
1411 * in_pcbref() should be used only to provide brief memory stability, and
1412 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1413 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1414 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1415 * lock and rele are the *only* safe operations that may be performed on the
1418 * While the inpcb will not be freed, releasing the inpcb lock means that the
1419 * connection's state may change, so the caller should be careful to
1420 * revalidate any cached state on reacquiring the lock. Drop the reference
1421 * using in_pcbrele().
1424 in_pcbref(struct inpcb *inp)
1427 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1429 refcount_acquire(&inp->inp_refcount);
1433 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1434 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1435 * return a flag indicating whether or not the inpcb remains valid. If it is
1436 * valid, we return with the inpcb lock held.
1438 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1439 * reference on an inpcb. Historically more work was done here (actually, in
1440 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1441 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1442 * about memory stability (and continued use of the write lock).
1445 in_pcbrele_rlocked(struct inpcb *inp)
1447 struct inpcbinfo *pcbinfo;
1449 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1451 INP_RLOCK_ASSERT(inp);
1453 if (refcount_release(&inp->inp_refcount) == 0) {
1455 * If the inpcb has been freed, let the caller know, even if
1456 * this isn't the last reference.
1458 if (inp->inp_flags2 & INP_FREED) {
1465 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1467 if (inp->inp_in_hpts || inp->inp_in_input) {
1468 struct tcp_hpts_entry *hpts;
1470 * We should not be on the hpts at
1471 * this point in any form. we must
1472 * get the lock to be sure.
1474 hpts = tcp_hpts_lock(inp);
1475 if (inp->inp_in_hpts)
1476 panic("Hpts:%p inp:%p at free still on hpts",
1478 mtx_unlock(&hpts->p_mtx);
1479 hpts = tcp_input_lock(inp);
1480 if (inp->inp_in_input)
1481 panic("Hpts:%p inp:%p at free still on input hpts",
1483 mtx_unlock(&hpts->p_mtx);
1487 pcbinfo = inp->inp_pcbinfo;
1488 uma_zfree(pcbinfo->ipi_zone, inp);
1493 in_pcbrele_wlocked(struct inpcb *inp)
1495 struct inpcbinfo *pcbinfo;
1497 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1499 INP_WLOCK_ASSERT(inp);
1501 if (refcount_release(&inp->inp_refcount) == 0) {
1503 * If the inpcb has been freed, let the caller know, even if
1504 * this isn't the last reference.
1506 if (inp->inp_flags2 & INP_FREED) {
1513 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1515 if (inp->inp_in_hpts || inp->inp_in_input) {
1516 struct tcp_hpts_entry *hpts;
1518 * We should not be on the hpts at
1519 * this point in any form. we must
1520 * get the lock to be sure.
1522 hpts = tcp_hpts_lock(inp);
1523 if (inp->inp_in_hpts)
1524 panic("Hpts:%p inp:%p at free still on hpts",
1526 mtx_unlock(&hpts->p_mtx);
1527 hpts = tcp_input_lock(inp);
1528 if (inp->inp_in_input)
1529 panic("Hpts:%p inp:%p at free still on input hpts",
1531 mtx_unlock(&hpts->p_mtx);
1535 pcbinfo = inp->inp_pcbinfo;
1536 uma_zfree(pcbinfo->ipi_zone, inp);
1541 * Temporary wrapper.
1544 in_pcbrele(struct inpcb *inp)
1547 return (in_pcbrele_wlocked(inp));
1551 in_pcblist_rele_rlocked(epoch_context_t ctx)
1553 struct in_pcblist *il;
1555 struct inpcbinfo *pcbinfo;
1558 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1559 pcbinfo = il->il_pcbinfo;
1561 INP_INFO_WLOCK(pcbinfo);
1562 for (i = 0; i < n; i++) {
1563 inp = il->il_inp_list[i];
1565 if (!in_pcbrele_rlocked(inp))
1568 INP_INFO_WUNLOCK(pcbinfo);
1573 inpcbport_free(epoch_context_t ctx)
1575 struct inpcbport *phd;
1577 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1582 in_pcbfree_deferred(epoch_context_t ctx)
1585 int released __unused;
1587 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1590 CURVNET_SET(inp->inp_vnet);
1592 struct ip_moptions *imo = inp->inp_moptions;
1593 inp->inp_moptions = NULL;
1595 /* XXXRW: Do as much as possible here. */
1596 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1597 if (inp->inp_sp != NULL)
1598 ipsec_delete_pcbpolicy(inp);
1601 struct ip6_moptions *im6o = NULL;
1602 if (inp->inp_vflag & INP_IPV6PROTO) {
1603 ip6_freepcbopts(inp->in6p_outputopts);
1604 im6o = inp->in6p_moptions;
1605 inp->in6p_moptions = NULL;
1608 if (inp->inp_options)
1609 (void)m_free(inp->inp_options);
1611 crfree(inp->inp_cred);
1613 mac_inpcb_destroy(inp);
1615 released = in_pcbrele_wlocked(inp);
1618 ip6_freemoptions(im6o);
1621 inp_freemoptions(imo);
1627 * Unconditionally schedule an inpcb to be freed by decrementing its
1628 * reference count, which should occur only after the inpcb has been detached
1629 * from its socket. If another thread holds a temporary reference (acquired
1630 * using in_pcbref()) then the free is deferred until that reference is
1631 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1632 * work, including removal from global lists, is done in this context, where
1633 * the pcbinfo lock is held.
1636 in_pcbfree(struct inpcb *inp)
1638 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1640 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1641 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1642 ("%s: called twice for pcb %p", __func__, inp));
1643 if (inp->inp_flags2 & INP_FREED) {
1648 INP_WLOCK_ASSERT(inp);
1649 INP_LIST_WLOCK(pcbinfo);
1650 in_pcbremlists(inp);
1651 INP_LIST_WUNLOCK(pcbinfo);
1652 RO_INVALIDATE_CACHE(&inp->inp_route);
1653 /* mark as destruction in progress */
1654 inp->inp_flags2 |= INP_FREED;
1656 NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx);
1660 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1661 * port reservation, and preventing it from being returned by inpcb lookups.
1663 * It is used by TCP to mark an inpcb as unused and avoid future packet
1664 * delivery or event notification when a socket remains open but TCP has
1665 * closed. This might occur as a result of a shutdown()-initiated TCP close
1666 * or a RST on the wire, and allows the port binding to be reused while still
1667 * maintaining the invariant that so_pcb always points to a valid inpcb until
1670 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1671 * in_pcbnotifyall() and in_pcbpurgeif0()?
1674 in_pcbdrop(struct inpcb *inp)
1677 INP_WLOCK_ASSERT(inp);
1679 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1680 MPASS(inp->inp_refcount > 1);
1684 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1687 inp->inp_flags |= INP_DROPPED;
1688 if (inp->inp_flags & INP_INHASHLIST) {
1689 struct inpcbport *phd = inp->inp_phd;
1691 INP_HASH_WLOCK(inp->inp_pcbinfo);
1692 in_pcbremlbgrouphash(inp);
1693 CK_LIST_REMOVE(inp, inp_hash);
1694 CK_LIST_REMOVE(inp, inp_portlist);
1695 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1696 CK_LIST_REMOVE(phd, phd_hash);
1697 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
1699 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1700 inp->inp_flags &= ~INP_INHASHLIST;
1702 in_pcbgroup_remove(inp);
1709 * Common routines to return the socket addresses associated with inpcbs.
1712 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1714 struct sockaddr_in *sin;
1716 sin = malloc(sizeof *sin, M_SONAME,
1718 sin->sin_family = AF_INET;
1719 sin->sin_len = sizeof(*sin);
1720 sin->sin_addr = *addr_p;
1721 sin->sin_port = port;
1723 return (struct sockaddr *)sin;
1727 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1730 struct in_addr addr;
1733 inp = sotoinpcb(so);
1734 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1737 port = inp->inp_lport;
1738 addr = inp->inp_laddr;
1741 *nam = in_sockaddr(port, &addr);
1746 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1749 struct in_addr addr;
1752 inp = sotoinpcb(so);
1753 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1756 port = inp->inp_fport;
1757 addr = inp->inp_faddr;
1760 *nam = in_sockaddr(port, &addr);
1765 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1766 struct inpcb *(*notify)(struct inpcb *, int))
1768 struct inpcb *inp, *inp_temp;
1770 INP_INFO_WLOCK(pcbinfo);
1771 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1774 if ((inp->inp_vflag & INP_IPV4) == 0) {
1779 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1780 inp->inp_socket == NULL) {
1784 if ((*notify)(inp, errno))
1787 INP_INFO_WUNLOCK(pcbinfo);
1791 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1794 struct in_multi *inm;
1795 struct in_mfilter *imf;
1796 struct ip_moptions *imo;
1798 INP_INFO_WLOCK(pcbinfo);
1799 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1801 imo = inp->inp_moptions;
1802 if ((inp->inp_vflag & INP_IPV4) &&
1805 * Unselect the outgoing interface if it is being
1808 if (imo->imo_multicast_ifp == ifp)
1809 imo->imo_multicast_ifp = NULL;
1812 * Drop multicast group membership if we joined
1813 * through the interface being detached.
1815 * XXX This can all be deferred to an epoch_call
1818 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1819 if ((inm = imf->imf_inm) == NULL)
1821 if (inm->inm_ifp != ifp)
1823 ip_mfilter_remove(&imo->imo_head, imf);
1824 IN_MULTI_LOCK_ASSERT();
1825 in_leavegroup_locked(inm, NULL);
1826 ip_mfilter_free(imf);
1832 INP_INFO_WUNLOCK(pcbinfo);
1836 * Lookup a PCB based on the local address and port. Caller must hold the
1837 * hash lock. No inpcb locks or references are acquired.
1839 #define INP_LOOKUP_MAPPED_PCB_COST 3
1841 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1842 u_short lport, int lookupflags, struct ucred *cred)
1846 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1852 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1853 ("%s: invalid lookup flags %d", __func__, lookupflags));
1855 INP_HASH_LOCK_ASSERT(pcbinfo);
1857 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1858 struct inpcbhead *head;
1860 * Look for an unconnected (wildcard foreign addr) PCB that
1861 * matches the local address and port we're looking for.
1863 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1864 0, pcbinfo->ipi_hashmask)];
1865 CK_LIST_FOREACH(inp, head, inp_hash) {
1867 /* XXX inp locking */
1868 if ((inp->inp_vflag & INP_IPV4) == 0)
1871 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1872 inp->inp_laddr.s_addr == laddr.s_addr &&
1873 inp->inp_lport == lport) {
1878 prison_equal_ip4(cred->cr_prison,
1879 inp->inp_cred->cr_prison))
1888 struct inpcbporthead *porthash;
1889 struct inpcbport *phd;
1890 struct inpcb *match = NULL;
1892 * Best fit PCB lookup.
1894 * First see if this local port is in use by looking on the
1897 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1898 pcbinfo->ipi_porthashmask)];
1899 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1900 if (phd->phd_port == lport)
1905 * Port is in use by one or more PCBs. Look for best
1908 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1911 !prison_equal_ip4(inp->inp_cred->cr_prison,
1915 /* XXX inp locking */
1916 if ((inp->inp_vflag & INP_IPV4) == 0)
1919 * We never select the PCB that has
1920 * INP_IPV6 flag and is bound to :: if
1921 * we have another PCB which is bound
1922 * to 0.0.0.0. If a PCB has the
1923 * INP_IPV6 flag, then we set its cost
1924 * higher than IPv4 only PCBs.
1926 * Note that the case only happens
1927 * when a socket is bound to ::, under
1928 * the condition that the use of the
1929 * mapped address is allowed.
1931 if ((inp->inp_vflag & INP_IPV6) != 0)
1932 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1934 if (inp->inp_faddr.s_addr != INADDR_ANY)
1936 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1937 if (laddr.s_addr == INADDR_ANY)
1939 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1942 if (laddr.s_addr != INADDR_ANY)
1945 if (wildcard < matchwild) {
1947 matchwild = wildcard;
1956 #undef INP_LOOKUP_MAPPED_PCB_COST
1958 static struct inpcb *
1959 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1960 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1961 uint16_t fport, int lookupflags)
1963 struct inpcb *local_wild;
1964 const struct inpcblbgrouphead *hdr;
1965 struct inpcblbgroup *grp;
1968 INP_HASH_LOCK_ASSERT(pcbinfo);
1970 hdr = &pcbinfo->ipi_lbgrouphashbase[
1971 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1974 * Order of socket selection:
1976 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1979 * - Load balanced group does not contain jailed sockets
1980 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1983 CK_LIST_FOREACH(grp, hdr, il_list) {
1985 if (!(grp->il_vflag & INP_IPV4))
1988 if (grp->il_lport != lport)
1991 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1993 if (grp->il_laddr.s_addr == laddr->s_addr)
1994 return (grp->il_inp[idx]);
1995 if (grp->il_laddr.s_addr == INADDR_ANY &&
1996 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1997 local_wild = grp->il_inp[idx];
1999 return (local_wild);
2004 * Lookup PCB in hash list, using pcbgroup tables.
2006 static struct inpcb *
2007 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2008 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2009 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2011 struct inpcbhead *head;
2012 struct inpcb *inp, *tmpinp;
2013 u_short fport = fport_arg, lport = lport_arg;
2017 * First look for an exact match.
2020 INP_GROUP_LOCK(pcbgroup);
2021 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2022 pcbgroup->ipg_hashmask)];
2023 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2025 /* XXX inp locking */
2026 if ((inp->inp_vflag & INP_IPV4) == 0)
2029 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2030 inp->inp_laddr.s_addr == laddr.s_addr &&
2031 inp->inp_fport == fport &&
2032 inp->inp_lport == lport) {
2034 * XXX We should be able to directly return
2035 * the inp here, without any checks.
2036 * Well unless both bound with SO_REUSEPORT?
2038 if (prison_flag(inp->inp_cred, PR_IP4))
2044 if (tmpinp != NULL) {
2051 * For incoming connections, we may wish to do a wildcard
2052 * match for an RSS-local socket.
2054 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2055 struct inpcb *local_wild = NULL, *local_exact = NULL;
2057 struct inpcb *local_wild_mapped = NULL;
2059 struct inpcb *jail_wild = NULL;
2060 struct inpcbhead *head;
2064 * Order of socket selection - we always prefer jails.
2065 * 1. jailed, non-wild.
2067 * 3. non-jailed, non-wild.
2068 * 4. non-jailed, wild.
2071 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2072 lport, 0, pcbgroup->ipg_hashmask)];
2073 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2075 /* XXX inp locking */
2076 if ((inp->inp_vflag & INP_IPV4) == 0)
2079 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2080 inp->inp_lport != lport)
2083 injail = prison_flag(inp->inp_cred, PR_IP4);
2085 if (prison_check_ip4(inp->inp_cred,
2089 if (local_exact != NULL)
2093 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2098 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2100 /* XXX inp locking, NULL check */
2101 if (inp->inp_vflag & INP_IPV6PROTO)
2102 local_wild_mapped = inp;
2110 } /* LIST_FOREACH */
2119 inp = local_wild_mapped;
2127 * Then look for a wildcard match, if requested.
2129 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2130 struct inpcb *local_wild = NULL, *local_exact = NULL;
2132 struct inpcb *local_wild_mapped = NULL;
2134 struct inpcb *jail_wild = NULL;
2135 struct inpcbhead *head;
2139 * Order of socket selection - we always prefer jails.
2140 * 1. jailed, non-wild.
2142 * 3. non-jailed, non-wild.
2143 * 4. non-jailed, wild.
2145 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2146 0, pcbinfo->ipi_wildmask)];
2147 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2149 /* XXX inp locking */
2150 if ((inp->inp_vflag & INP_IPV4) == 0)
2153 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2154 inp->inp_lport != lport)
2157 injail = prison_flag(inp->inp_cred, PR_IP4);
2159 if (prison_check_ip4(inp->inp_cred,
2163 if (local_exact != NULL)
2167 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2172 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2174 /* XXX inp locking, NULL check */
2175 if (inp->inp_vflag & INP_IPV6PROTO)
2176 local_wild_mapped = inp;
2184 } /* LIST_FOREACH */
2192 inp = local_wild_mapped;
2196 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2197 INP_GROUP_UNLOCK(pcbgroup);
2201 if (lookupflags & INPLOOKUP_WLOCKPCB)
2202 locked = INP_TRY_WLOCK(inp);
2203 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2204 locked = INP_TRY_RLOCK(inp);
2206 panic("%s: locking bug", __func__);
2207 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2208 if (lookupflags & INPLOOKUP_WLOCKPCB)
2215 INP_GROUP_UNLOCK(pcbgroup);
2217 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2219 if (in_pcbrele_wlocked(inp))
2223 if (in_pcbrele_rlocked(inp))
2228 if (lookupflags & INPLOOKUP_WLOCKPCB)
2229 INP_WLOCK_ASSERT(inp);
2231 INP_RLOCK_ASSERT(inp);
2235 #endif /* PCBGROUP */
2238 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2239 * that the caller has locked the hash list, and will not perform any further
2240 * locking or reference operations on either the hash list or the connection.
2242 static struct inpcb *
2243 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2244 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2247 struct inpcbhead *head;
2248 struct inpcb *inp, *tmpinp;
2249 u_short fport = fport_arg, lport = lport_arg;
2251 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2252 ("%s: invalid lookup flags %d", __func__, lookupflags));
2253 INP_HASH_LOCK_ASSERT(pcbinfo);
2256 * First look for an exact match.
2259 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2260 pcbinfo->ipi_hashmask)];
2261 CK_LIST_FOREACH(inp, head, inp_hash) {
2263 /* XXX inp locking */
2264 if ((inp->inp_vflag & INP_IPV4) == 0)
2267 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2268 inp->inp_laddr.s_addr == laddr.s_addr &&
2269 inp->inp_fport == fport &&
2270 inp->inp_lport == lport) {
2272 * XXX We should be able to directly return
2273 * the inp here, without any checks.
2274 * Well unless both bound with SO_REUSEPORT?
2276 if (prison_flag(inp->inp_cred, PR_IP4))
2286 * Then look in lb group (for wildcard match).
2288 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2289 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2290 fport, lookupflags);
2296 * Then look for a wildcard match, if requested.
2298 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2299 struct inpcb *local_wild = NULL, *local_exact = NULL;
2301 struct inpcb *local_wild_mapped = NULL;
2303 struct inpcb *jail_wild = NULL;
2307 * Order of socket selection - we always prefer jails.
2308 * 1. jailed, non-wild.
2310 * 3. non-jailed, non-wild.
2311 * 4. non-jailed, wild.
2314 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2315 0, pcbinfo->ipi_hashmask)];
2316 CK_LIST_FOREACH(inp, head, inp_hash) {
2318 /* XXX inp locking */
2319 if ((inp->inp_vflag & INP_IPV4) == 0)
2322 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2323 inp->inp_lport != lport)
2326 injail = prison_flag(inp->inp_cred, PR_IP4);
2328 if (prison_check_ip4(inp->inp_cred,
2332 if (local_exact != NULL)
2336 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2341 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2343 /* XXX inp locking, NULL check */
2344 if (inp->inp_vflag & INP_IPV6PROTO)
2345 local_wild_mapped = inp;
2353 } /* LIST_FOREACH */
2354 if (jail_wild != NULL)
2356 if (local_exact != NULL)
2357 return (local_exact);
2358 if (local_wild != NULL)
2359 return (local_wild);
2361 if (local_wild_mapped != NULL)
2362 return (local_wild_mapped);
2364 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2370 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2371 * hash list lock, and will return the inpcb locked (i.e., requires
2372 * INPLOOKUP_LOCKPCB).
2374 static struct inpcb *
2375 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2376 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2381 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2382 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2384 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2386 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2390 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2392 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2397 panic("%s: locking bug", __func__);
2400 if (lookupflags & INPLOOKUP_WLOCKPCB)
2401 INP_WLOCK_ASSERT(inp);
2403 INP_RLOCK_ASSERT(inp);
2412 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2413 * from which a pre-calculated hash value may be extracted.
2415 * Possibly more of this logic should be in in_pcbgroup.c.
2418 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2419 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2421 #if defined(PCBGROUP) && !defined(RSS)
2422 struct inpcbgroup *pcbgroup;
2425 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2426 ("%s: invalid lookup flags %d", __func__, lookupflags));
2427 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2428 ("%s: LOCKPCB not set", __func__));
2431 * When not using RSS, use connection groups in preference to the
2432 * reservation table when looking up 4-tuples. When using RSS, just
2433 * use the reservation table, due to the cost of the Toeplitz hash
2436 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2437 * we could be doing RSS with a non-Toeplitz hash that is affordable
2440 #if defined(PCBGROUP) && !defined(RSS)
2441 if (in_pcbgroup_enabled(pcbinfo)) {
2442 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2444 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2445 laddr, lport, lookupflags, ifp));
2448 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2453 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2454 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2455 struct ifnet *ifp, struct mbuf *m)
2458 struct inpcbgroup *pcbgroup;
2461 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2462 ("%s: invalid lookup flags %d", __func__, lookupflags));
2463 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2464 ("%s: LOCKPCB not set", __func__));
2468 * If we can use a hardware-generated hash to look up the connection
2469 * group, use that connection group to find the inpcb. Otherwise
2470 * fall back on a software hash -- or the reservation table if we're
2473 * XXXRW: As above, that policy belongs in the pcbgroup code.
2475 if (in_pcbgroup_enabled(pcbinfo) &&
2476 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2477 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2478 m->m_pkthdr.flowid);
2479 if (pcbgroup != NULL)
2480 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2481 fport, laddr, lport, lookupflags, ifp));
2483 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2485 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2486 laddr, lport, lookupflags, ifp));
2490 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2496 * Insert PCB onto various hash lists.
2499 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2501 struct inpcbhead *pcbhash;
2502 struct inpcbporthead *pcbporthash;
2503 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2504 struct inpcbport *phd;
2505 u_int32_t hashkey_faddr;
2508 INP_WLOCK_ASSERT(inp);
2509 INP_HASH_WLOCK_ASSERT(pcbinfo);
2511 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2512 ("in_pcbinshash: INP_INHASHLIST"));
2515 if (inp->inp_vflag & INP_IPV6)
2516 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2519 hashkey_faddr = inp->inp_faddr.s_addr;
2521 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2522 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2524 pcbporthash = &pcbinfo->ipi_porthashbase[
2525 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2528 * Add entry to load balance group.
2529 * Only do this if SO_REUSEPORT_LB is set.
2531 so_options = inp_so_options(inp);
2532 if (so_options & SO_REUSEPORT_LB) {
2533 int ret = in_pcbinslbgrouphash(inp);
2535 /* pcb lb group malloc fail (ret=ENOBUFS). */
2541 * Go through port list and look for a head for this lport.
2543 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2544 if (phd->phd_port == inp->inp_lport)
2548 * If none exists, malloc one and tack it on.
2551 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2553 return (ENOBUFS); /* XXX */
2555 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2556 phd->phd_port = inp->inp_lport;
2557 CK_LIST_INIT(&phd->phd_pcblist);
2558 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2561 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2562 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2563 inp->inp_flags |= INP_INHASHLIST;
2566 in_pcbgroup_update_mbuf(inp, m);
2568 in_pcbgroup_update(inp);
2575 in_pcbinshash(struct inpcb *inp)
2578 return (in_pcbinshash_internal(inp, NULL));
2582 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2585 return (in_pcbinshash_internal(inp, m));
2589 * Move PCB to the proper hash bucket when { faddr, fport } have been
2590 * changed. NOTE: This does not handle the case of the lport changing (the
2591 * hashed port list would have to be updated as well), so the lport must
2592 * not change after in_pcbinshash() has been called.
2595 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2597 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2598 struct inpcbhead *head;
2599 u_int32_t hashkey_faddr;
2601 INP_WLOCK_ASSERT(inp);
2602 INP_HASH_WLOCK_ASSERT(pcbinfo);
2604 KASSERT(inp->inp_flags & INP_INHASHLIST,
2605 ("in_pcbrehash: !INP_INHASHLIST"));
2608 if (inp->inp_vflag & INP_IPV6)
2609 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2612 hashkey_faddr = inp->inp_faddr.s_addr;
2614 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2615 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2617 CK_LIST_REMOVE(inp, inp_hash);
2618 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2622 in_pcbgroup_update_mbuf(inp, m);
2624 in_pcbgroup_update(inp);
2629 in_pcbrehash(struct inpcb *inp)
2632 in_pcbrehash_mbuf(inp, NULL);
2636 * Remove PCB from various lists.
2639 in_pcbremlists(struct inpcb *inp)
2641 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2643 INP_WLOCK_ASSERT(inp);
2644 INP_LIST_WLOCK_ASSERT(pcbinfo);
2646 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2647 if (inp->inp_flags & INP_INHASHLIST) {
2648 struct inpcbport *phd = inp->inp_phd;
2650 INP_HASH_WLOCK(pcbinfo);
2652 /* XXX: Only do if SO_REUSEPORT_LB set? */
2653 in_pcbremlbgrouphash(inp);
2655 CK_LIST_REMOVE(inp, inp_hash);
2656 CK_LIST_REMOVE(inp, inp_portlist);
2657 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2658 CK_LIST_REMOVE(phd, phd_hash);
2659 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
2661 INP_HASH_WUNLOCK(pcbinfo);
2662 inp->inp_flags &= ~INP_INHASHLIST;
2664 CK_LIST_REMOVE(inp, inp_list);
2665 pcbinfo->ipi_count--;
2667 in_pcbgroup_remove(inp);
2672 * Check for alternatives when higher level complains
2673 * about service problems. For now, invalidate cached
2674 * routing information. If the route was created dynamically
2675 * (by a redirect), time to try a default gateway again.
2678 in_losing(struct inpcb *inp)
2681 RO_INVALIDATE_CACHE(&inp->inp_route);
2686 * A set label operation has occurred at the socket layer, propagate the
2687 * label change into the in_pcb for the socket.
2690 in_pcbsosetlabel(struct socket *so)
2695 inp = sotoinpcb(so);
2696 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2700 mac_inpcb_sosetlabel(so, inp);
2707 * ipport_tick runs once per second, determining if random port allocation
2708 * should be continued. If more than ipport_randomcps ports have been
2709 * allocated in the last second, then we return to sequential port
2710 * allocation. We return to random allocation only once we drop below
2711 * ipport_randomcps for at least ipport_randomtime seconds.
2714 ipport_tick(void *xtp)
2716 VNET_ITERATOR_DECL(vnet_iter);
2718 VNET_LIST_RLOCK_NOSLEEP();
2719 VNET_FOREACH(vnet_iter) {
2720 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2721 if (V_ipport_tcpallocs <=
2722 V_ipport_tcplastcount + V_ipport_randomcps) {
2723 if (V_ipport_stoprandom > 0)
2724 V_ipport_stoprandom--;
2726 V_ipport_stoprandom = V_ipport_randomtime;
2727 V_ipport_tcplastcount = V_ipport_tcpallocs;
2730 VNET_LIST_RUNLOCK_NOSLEEP();
2731 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2738 callout_stop(&ipport_tick_callout);
2742 * The ipport_callout should start running at about the time we attach the
2743 * inet or inet6 domains.
2746 ipport_tick_init(const void *unused __unused)
2749 /* Start ipport_tick. */
2750 callout_init(&ipport_tick_callout, 1);
2751 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2752 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2753 SHUTDOWN_PRI_DEFAULT);
2755 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2756 ipport_tick_init, NULL);
2759 inp_wlock(struct inpcb *inp)
2766 inp_wunlock(struct inpcb *inp)
2773 inp_rlock(struct inpcb *inp)
2780 inp_runlock(struct inpcb *inp)
2786 #ifdef INVARIANT_SUPPORT
2788 inp_lock_assert(struct inpcb *inp)
2791 INP_WLOCK_ASSERT(inp);
2795 inp_unlock_assert(struct inpcb *inp)
2798 INP_UNLOCK_ASSERT(inp);
2803 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2807 INP_INFO_WLOCK(&V_tcbinfo);
2808 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2813 INP_INFO_WUNLOCK(&V_tcbinfo);
2817 inp_inpcbtosocket(struct inpcb *inp)
2820 INP_WLOCK_ASSERT(inp);
2821 return (inp->inp_socket);
2825 inp_inpcbtotcpcb(struct inpcb *inp)
2828 INP_WLOCK_ASSERT(inp);
2829 return ((struct tcpcb *)inp->inp_ppcb);
2833 inp_ip_tos_get(const struct inpcb *inp)
2836 return (inp->inp_ip_tos);
2840 inp_ip_tos_set(struct inpcb *inp, int val)
2843 inp->inp_ip_tos = val;
2847 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2848 uint32_t *faddr, uint16_t *fp)
2851 INP_LOCK_ASSERT(inp);
2852 *laddr = inp->inp_laddr.s_addr;
2853 *faddr = inp->inp_faddr.s_addr;
2854 *lp = inp->inp_lport;
2855 *fp = inp->inp_fport;
2859 so_sotoinpcb(struct socket *so)
2862 return (sotoinpcb(so));
2866 so_sototcpcb(struct socket *so)
2869 return (sototcpcb(so));
2873 * Create an external-format (``xinpcb'') structure using the information in
2874 * the kernel-format in_pcb structure pointed to by inp. This is done to
2875 * reduce the spew of irrelevant information over this interface, to isolate
2876 * user code from changes in the kernel structure, and potentially to provide
2877 * information-hiding if we decide that some of this information should be
2878 * hidden from users.
2881 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2884 bzero(xi, sizeof(*xi));
2885 xi->xi_len = sizeof(struct xinpcb);
2886 if (inp->inp_socket)
2887 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2888 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2889 xi->inp_gencnt = inp->inp_gencnt;
2890 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2891 xi->inp_flow = inp->inp_flow;
2892 xi->inp_flowid = inp->inp_flowid;
2893 xi->inp_flowtype = inp->inp_flowtype;
2894 xi->inp_flags = inp->inp_flags;
2895 xi->inp_flags2 = inp->inp_flags2;
2896 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2897 xi->in6p_cksum = inp->in6p_cksum;
2898 xi->in6p_hops = inp->in6p_hops;
2899 xi->inp_ip_tos = inp->inp_ip_tos;
2900 xi->inp_vflag = inp->inp_vflag;
2901 xi->inp_ip_ttl = inp->inp_ip_ttl;
2902 xi->inp_ip_p = inp->inp_ip_p;
2903 xi->inp_ip_minttl = inp->inp_ip_minttl;
2908 db_print_indent(int indent)
2912 for (i = 0; i < indent; i++)
2917 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2919 char faddr_str[48], laddr_str[48];
2921 db_print_indent(indent);
2922 db_printf("%s at %p\n", name, inc);
2927 if (inc->inc_flags & INC_ISIPV6) {
2929 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2930 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2935 inet_ntoa_r(inc->inc_laddr, laddr_str);
2936 inet_ntoa_r(inc->inc_faddr, faddr_str);
2938 db_print_indent(indent);
2939 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2940 ntohs(inc->inc_lport));
2941 db_print_indent(indent);
2942 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2943 ntohs(inc->inc_fport));
2947 db_print_inpflags(int inp_flags)
2952 if (inp_flags & INP_RECVOPTS) {
2953 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2956 if (inp_flags & INP_RECVRETOPTS) {
2957 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2960 if (inp_flags & INP_RECVDSTADDR) {
2961 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2964 if (inp_flags & INP_ORIGDSTADDR) {
2965 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2968 if (inp_flags & INP_HDRINCL) {
2969 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2972 if (inp_flags & INP_HIGHPORT) {
2973 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2976 if (inp_flags & INP_LOWPORT) {
2977 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2980 if (inp_flags & INP_ANONPORT) {
2981 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2984 if (inp_flags & INP_RECVIF) {
2985 db_printf("%sINP_RECVIF", comma ? ", " : "");
2988 if (inp_flags & INP_MTUDISC) {
2989 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2992 if (inp_flags & INP_RECVTTL) {
2993 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2996 if (inp_flags & INP_DONTFRAG) {
2997 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3000 if (inp_flags & INP_RECVTOS) {
3001 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3004 if (inp_flags & IN6P_IPV6_V6ONLY) {
3005 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3008 if (inp_flags & IN6P_PKTINFO) {
3009 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3012 if (inp_flags & IN6P_HOPLIMIT) {
3013 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3016 if (inp_flags & IN6P_HOPOPTS) {
3017 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3020 if (inp_flags & IN6P_DSTOPTS) {
3021 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3024 if (inp_flags & IN6P_RTHDR) {
3025 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3028 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3029 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3032 if (inp_flags & IN6P_TCLASS) {
3033 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3036 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3037 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3040 if (inp_flags & INP_TIMEWAIT) {
3041 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3044 if (inp_flags & INP_ONESBCAST) {
3045 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3048 if (inp_flags & INP_DROPPED) {
3049 db_printf("%sINP_DROPPED", comma ? ", " : "");
3052 if (inp_flags & INP_SOCKREF) {
3053 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3056 if (inp_flags & IN6P_RFC2292) {
3057 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3060 if (inp_flags & IN6P_MTU) {
3061 db_printf("IN6P_MTU%s", comma ? ", " : "");
3067 db_print_inpvflag(u_char inp_vflag)
3072 if (inp_vflag & INP_IPV4) {
3073 db_printf("%sINP_IPV4", comma ? ", " : "");
3076 if (inp_vflag & INP_IPV6) {
3077 db_printf("%sINP_IPV6", comma ? ", " : "");
3080 if (inp_vflag & INP_IPV6PROTO) {
3081 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3087 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3090 db_print_indent(indent);
3091 db_printf("%s at %p\n", name, inp);
3095 db_print_indent(indent);
3096 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3098 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3100 db_print_indent(indent);
3101 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3102 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3104 db_print_indent(indent);
3105 db_printf("inp_label: %p inp_flags: 0x%x (",
3106 inp->inp_label, inp->inp_flags);
3107 db_print_inpflags(inp->inp_flags);
3110 db_print_indent(indent);
3111 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3113 db_print_inpvflag(inp->inp_vflag);
3116 db_print_indent(indent);
3117 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3118 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3120 db_print_indent(indent);
3122 if (inp->inp_vflag & INP_IPV6) {
3123 db_printf("in6p_options: %p in6p_outputopts: %p "
3124 "in6p_moptions: %p\n", inp->in6p_options,
3125 inp->in6p_outputopts, inp->in6p_moptions);
3126 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3127 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3132 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3133 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3134 inp->inp_options, inp->inp_moptions);
3137 db_print_indent(indent);
3138 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3139 (uintmax_t)inp->inp_gencnt);
3142 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3147 db_printf("usage: show inpcb <addr>\n");
3150 inp = (struct inpcb *)addr;
3152 db_print_inpcb(inp, "inpcb", 0);
3158 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3162 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3164 union if_snd_tag_modify_params params = {
3165 .rate_limit.max_rate = max_pacing_rate,
3166 .rate_limit.flags = M_NOWAIT,
3168 struct m_snd_tag *mst;
3172 mst = inp->inp_snd_tag;
3180 if (ifp->if_snd_tag_modify == NULL) {
3183 error = ifp->if_snd_tag_modify(mst, ¶ms);
3189 * Query existing TX rate limit based on the existing
3190 * "inp->inp_snd_tag", if any.
3193 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3195 union if_snd_tag_query_params params = { };
3196 struct m_snd_tag *mst;
3200 mst = inp->inp_snd_tag;
3208 if (ifp->if_snd_tag_query == NULL) {
3211 error = ifp->if_snd_tag_query(mst, ¶ms);
3212 if (error == 0 && p_max_pacing_rate != NULL)
3213 *p_max_pacing_rate = params.rate_limit.max_rate;
3219 * Query existing TX queue level based on the existing
3220 * "inp->inp_snd_tag", if any.
3223 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3225 union if_snd_tag_query_params params = { };
3226 struct m_snd_tag *mst;
3230 mst = inp->inp_snd_tag;
3238 if (ifp->if_snd_tag_query == NULL)
3239 return (EOPNOTSUPP);
3241 error = ifp->if_snd_tag_query(mst, ¶ms);
3242 if (error == 0 && p_txqueue_level != NULL)
3243 *p_txqueue_level = params.rate_limit.queue_level;
3248 * Allocate a new TX rate limit send tag from the network interface
3249 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3252 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3253 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3256 union if_snd_tag_alloc_params params = {
3257 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3258 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3259 .rate_limit.hdr.flowid = flowid,
3260 .rate_limit.hdr.flowtype = flowtype,
3261 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3262 .rate_limit.max_rate = max_pacing_rate,
3263 .rate_limit.flags = M_NOWAIT,
3267 INP_WLOCK_ASSERT(inp);
3272 if (ifp->if_snd_tag_alloc == NULL) {
3275 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3279 counter_u64_add(rate_limit_set_ok, 1);
3280 counter_u64_add(rate_limit_active, 1);
3282 counter_u64_add(rate_limit_alloc_fail, 1);
3289 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3295 * If the device was detached while we still had reference(s)
3296 * on the ifp, we assume if_snd_tag_free() was replaced with
3299 ifp->if_snd_tag_free(mst);
3301 /* release reference count on network interface */
3304 counter_u64_add(rate_limit_active, -1);
3309 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3313 in_pcbdetach_txrtlmt(struct inpcb *inp)
3315 struct m_snd_tag *mst;
3317 INP_WLOCK_ASSERT(inp);
3319 mst = inp->inp_snd_tag;
3320 inp->inp_snd_tag = NULL;
3325 m_snd_tag_rele(mst);
3329 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3334 * If the existing send tag is for the wrong interface due to
3335 * a route change, first drop the existing tag. Set the
3336 * CHANGED flag so that we will keep trying to allocate a new
3337 * tag if we fail to allocate one this time.
3339 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3340 in_pcbdetach_txrtlmt(inp);
3341 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3345 * NOTE: When attaching to a network interface a reference is
3346 * made to ensure the network interface doesn't go away until
3347 * all ratelimit connections are gone. The network interface
3348 * pointers compared below represent valid network interfaces,
3349 * except when comparing towards NULL.
3351 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3353 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3354 if (inp->inp_snd_tag != NULL)
3355 in_pcbdetach_txrtlmt(inp);
3357 } else if (inp->inp_snd_tag == NULL) {
3359 * In order to utilize packet pacing with RSS, we need
3360 * to wait until there is a valid RSS hash before we
3363 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3366 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3367 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3370 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3372 if (error == 0 || error == EOPNOTSUPP)
3373 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3379 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3380 * is set in the fast path and will attach/detach/modify the TX rate
3381 * limit send tag based on the socket's so_max_pacing_rate value.
3384 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3386 struct socket *socket;
3387 uint32_t max_pacing_rate;
3394 socket = inp->inp_socket;
3398 if (!INP_WLOCKED(inp)) {
3400 * NOTE: If the write locking fails, we need to bail
3401 * out and use the non-ratelimited ring for the
3402 * transmit until there is a new chance to get the
3405 if (!INP_TRY_UPGRADE(inp))
3413 * NOTE: The so_max_pacing_rate value is read unlocked,
3414 * because atomic updates are not required since the variable
3415 * is checked at every mbuf we send. It is assumed that the
3416 * variable read itself will be atomic.
3418 max_pacing_rate = socket->so_max_pacing_rate;
3420 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3427 * Track route changes for TX rate limiting.
3430 in_pcboutput_eagain(struct inpcb *inp)
3437 if (inp->inp_snd_tag == NULL)
3440 if (!INP_WLOCKED(inp)) {
3442 * NOTE: If the write locking fails, we need to bail
3443 * out and use the non-ratelimited ring for the
3444 * transmit until there is a new chance to get the
3447 if (!INP_TRY_UPGRADE(inp))
3454 /* detach rate limiting */
3455 in_pcbdetach_txrtlmt(inp);
3457 /* make sure new mbuf send tag allocation is made */
3458 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3468 rate_limit_active = counter_u64_alloc(M_WAITOK);
3469 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3470 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3473 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3475 #endif /* RATELIMIT */