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 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
273 in_pcblbgroup_free_deferred);
276 static struct inpcblbgroup *
277 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
278 struct inpcblbgroup *old_grp, int size)
280 struct inpcblbgroup *grp;
283 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
284 old_grp->il_lport, &old_grp->il_dependladdr, size);
288 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
289 ("invalid new local group size %d and old local group count %d",
290 grp->il_inpsiz, old_grp->il_inpcnt));
292 for (i = 0; i < old_grp->il_inpcnt; ++i)
293 grp->il_inp[i] = old_grp->il_inp[i];
294 grp->il_inpcnt = old_grp->il_inpcnt;
295 in_pcblbgroup_free(old_grp);
300 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
301 * and shrink group if possible.
304 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
307 struct inpcblbgroup *grp, *new_grp;
310 for (; i + 1 < grp->il_inpcnt; ++i)
311 grp->il_inp[i] = grp->il_inp[i + 1];
314 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
315 grp->il_inpcnt <= grp->il_inpsiz / 4) {
316 /* Shrink this group. */
317 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
324 * Add PCB to load balance group for SO_REUSEPORT_LB option.
327 in_pcbinslbgrouphash(struct inpcb *inp)
329 const static struct timeval interval = { 60, 0 };
330 static struct timeval lastprint;
331 struct inpcbinfo *pcbinfo;
332 struct inpcblbgrouphead *hdr;
333 struct inpcblbgroup *grp;
336 pcbinfo = inp->inp_pcbinfo;
338 INP_WLOCK_ASSERT(inp);
339 INP_HASH_WLOCK_ASSERT(pcbinfo);
342 * Don't allow jailed socket to join local group.
344 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
349 * Don't allow IPv4 mapped INET6 wild socket.
351 if ((inp->inp_vflag & INP_IPV4) &&
352 inp->inp_laddr.s_addr == INADDR_ANY &&
353 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
358 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
359 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
360 CK_LIST_FOREACH(grp, hdr, il_list) {
361 if (grp->il_vflag == inp->inp_vflag &&
362 grp->il_lport == inp->inp_lport &&
363 memcmp(&grp->il_dependladdr,
364 &inp->inp_inc.inc_ie.ie_dependladdr,
365 sizeof(grp->il_dependladdr)) == 0)
369 /* Create new load balance group. */
370 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
371 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
372 INPCBLBGROUP_SIZMIN);
375 } else if (grp->il_inpcnt == grp->il_inpsiz) {
376 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
377 if (ratecheck(&lastprint, &interval))
378 printf("lb group port %d, limit reached\n",
379 ntohs(grp->il_lport));
383 /* Expand this local group. */
384 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
389 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
390 ("invalid local group size %d and count %d", grp->il_inpsiz,
393 grp->il_inp[grp->il_inpcnt] = inp;
399 * Remove PCB from load balance group.
402 in_pcbremlbgrouphash(struct inpcb *inp)
404 struct inpcbinfo *pcbinfo;
405 struct inpcblbgrouphead *hdr;
406 struct inpcblbgroup *grp;
409 pcbinfo = inp->inp_pcbinfo;
411 INP_WLOCK_ASSERT(inp);
412 INP_HASH_WLOCK_ASSERT(pcbinfo);
414 hdr = &pcbinfo->ipi_lbgrouphashbase[
415 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
416 CK_LIST_FOREACH(grp, hdr, il_list) {
417 for (i = 0; i < grp->il_inpcnt; ++i) {
418 if (grp->il_inp[i] != inp)
421 if (grp->il_inpcnt == 1) {
422 /* We are the last, free this local group. */
423 in_pcblbgroup_free(grp);
425 /* Pull up inpcbs, shrink group if possible. */
426 in_pcblbgroup_reorder(hdr, &grp, i);
434 * Different protocols initialize their inpcbs differently - giving
435 * different name to the lock. But they all are disposed the same.
438 inpcb_fini(void *mem, int size)
440 struct inpcb *inp = mem;
442 INP_LOCK_DESTROY(inp);
446 * Initialize an inpcbinfo -- we should be able to reduce the number of
450 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
451 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
452 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
455 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
457 INP_INFO_LOCK_INIT(pcbinfo, name);
458 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
459 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
461 pcbinfo->ipi_vnet = curvnet;
463 pcbinfo->ipi_listhead = listhead;
464 CK_LIST_INIT(pcbinfo->ipi_listhead);
465 pcbinfo->ipi_count = 0;
466 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
467 &pcbinfo->ipi_hashmask);
468 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
469 &pcbinfo->ipi_porthashmask);
470 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
471 &pcbinfo->ipi_lbgrouphashmask);
473 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
475 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
476 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
477 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
478 uma_zone_set_warning(pcbinfo->ipi_zone,
479 "kern.ipc.maxsockets limit reached");
483 * Destroy an inpcbinfo.
486 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
489 KASSERT(pcbinfo->ipi_count == 0,
490 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
492 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
493 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
494 pcbinfo->ipi_porthashmask);
495 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
496 pcbinfo->ipi_lbgrouphashmask);
498 in_pcbgroup_destroy(pcbinfo);
500 uma_zdestroy(pcbinfo->ipi_zone);
501 INP_LIST_LOCK_DESTROY(pcbinfo);
502 INP_HASH_LOCK_DESTROY(pcbinfo);
503 INP_INFO_LOCK_DESTROY(pcbinfo);
507 * Allocate a PCB and associate it with the socket.
508 * On success return with the PCB locked.
511 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
517 if (pcbinfo == &V_tcbinfo) {
518 INP_INFO_RLOCK_ASSERT(pcbinfo);
520 INP_INFO_WLOCK_ASSERT(pcbinfo);
525 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
528 bzero(&inp->inp_start_zero, inp_zero_size);
530 inp->inp_numa_domain = M_NODOM;
532 inp->inp_pcbinfo = pcbinfo;
533 inp->inp_socket = so;
534 inp->inp_cred = crhold(so->so_cred);
535 inp->inp_inc.inc_fibnum = so->so_fibnum;
537 error = mac_inpcb_init(inp, M_NOWAIT);
540 mac_inpcb_create(so, inp);
542 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
543 error = ipsec_init_pcbpolicy(inp);
546 mac_inpcb_destroy(inp);
552 if (INP_SOCKAF(so) == AF_INET6) {
553 inp->inp_vflag |= INP_IPV6PROTO;
555 inp->inp_flags |= IN6P_IPV6_V6ONLY;
559 INP_LIST_WLOCK(pcbinfo);
560 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
561 pcbinfo->ipi_count++;
562 so->so_pcb = (caddr_t)inp;
564 if (V_ip6_auto_flowlabel)
565 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
567 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
568 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
571 * Routes in inpcb's can cache L2 as well; they are guaranteed
574 inp->inp_route.ro_flags = RT_LLE_CACHE;
575 INP_LIST_WUNLOCK(pcbinfo);
576 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
579 crfree(inp->inp_cred);
580 uma_zfree(pcbinfo->ipi_zone, inp);
588 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
592 INP_WLOCK_ASSERT(inp);
593 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
595 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
597 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
598 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
599 &inp->inp_lport, cred);
602 if (in_pcbinshash(inp) != 0) {
603 inp->inp_laddr.s_addr = INADDR_ANY;
608 inp->inp_flags |= INP_ANONPORT;
614 * Select a local port (number) to use.
616 #if defined(INET) || defined(INET6)
618 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
619 struct ucred *cred, int lookupflags)
621 struct inpcbinfo *pcbinfo;
622 struct inpcb *tmpinp;
623 unsigned short *lastport;
624 int count, dorandom, error;
625 u_short aux, first, last, lport;
627 struct in_addr laddr;
630 pcbinfo = inp->inp_pcbinfo;
633 * Because no actual state changes occur here, a global write lock on
634 * the pcbinfo isn't required.
636 INP_LOCK_ASSERT(inp);
637 INP_HASH_LOCK_ASSERT(pcbinfo);
639 if (inp->inp_flags & INP_HIGHPORT) {
640 first = V_ipport_hifirstauto; /* sysctl */
641 last = V_ipport_hilastauto;
642 lastport = &pcbinfo->ipi_lasthi;
643 } else if (inp->inp_flags & INP_LOWPORT) {
644 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
647 first = V_ipport_lowfirstauto; /* 1023 */
648 last = V_ipport_lowlastauto; /* 600 */
649 lastport = &pcbinfo->ipi_lastlow;
651 first = V_ipport_firstauto; /* sysctl */
652 last = V_ipport_lastauto;
653 lastport = &pcbinfo->ipi_lastport;
656 * For UDP(-Lite), use random port allocation as long as the user
657 * allows it. For TCP (and as of yet unknown) connections,
658 * use random port allocation only if the user allows it AND
659 * ipport_tick() allows it.
661 if (V_ipport_randomized &&
662 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
663 pcbinfo == &V_ulitecbinfo))
668 * It makes no sense to do random port allocation if
669 * we have the only port available.
673 /* Make sure to not include UDP(-Lite) packets in the count. */
674 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
675 V_ipport_tcpallocs++;
677 * Instead of having two loops further down counting up or down
678 * make sure that first is always <= last and go with only one
679 * code path implementing all logic.
688 /* Make the compiler happy. */
690 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
691 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
696 tmpinp = NULL; /* Make compiler happy. */
700 *lastport = first + (arc4random() % (last - first));
702 count = last - first;
705 if (count-- < 0) /* completely used? */
706 return (EADDRNOTAVAIL);
708 if (*lastport < first || *lastport > last)
710 lport = htons(*lastport);
713 if ((inp->inp_vflag & INP_IPV6) != 0)
714 tmpinp = in6_pcblookup_local(pcbinfo,
715 &inp->in6p_laddr, lport, lookupflags, cred);
717 #if defined(INET) && defined(INET6)
721 tmpinp = in_pcblookup_local(pcbinfo, laddr,
722 lport, lookupflags, cred);
724 } while (tmpinp != NULL);
727 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
728 laddrp->s_addr = laddr.s_addr;
736 * Return cached socket options.
739 inp_so_options(const struct inpcb *inp)
745 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
746 so_options |= SO_REUSEPORT_LB;
747 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
748 so_options |= SO_REUSEPORT;
749 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
750 so_options |= SO_REUSEADDR;
753 #endif /* INET || INET6 */
756 * Check if a new BINDMULTI socket is allowed to be created.
758 * ni points to the new inp.
759 * oi points to the exisitng inp.
761 * This checks whether the existing inp also has BINDMULTI and
762 * whether the credentials match.
765 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
767 /* Check permissions match */
768 if ((ni->inp_flags2 & INP_BINDMULTI) &&
769 (ni->inp_cred->cr_uid !=
770 oi->inp_cred->cr_uid))
773 /* Check the existing inp has BINDMULTI set */
774 if ((ni->inp_flags2 & INP_BINDMULTI) &&
775 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
779 * We're okay - either INP_BINDMULTI isn't set on ni, or
780 * it is and it matches the checks.
787 * Set up a bind operation on a PCB, performing port allocation
788 * as required, but do not actually modify the PCB. Callers can
789 * either complete the bind by setting inp_laddr/inp_lport and
790 * calling in_pcbinshash(), or they can just use the resulting
791 * port and address to authorise the sending of a once-off packet.
793 * On error, the values of *laddrp and *lportp are not changed.
796 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
797 u_short *lportp, struct ucred *cred)
799 struct socket *so = inp->inp_socket;
800 struct sockaddr_in *sin;
801 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
802 struct in_addr laddr;
804 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
808 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
809 * so that we don't have to add to the (already messy) code below.
811 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
814 * No state changes, so read locks are sufficient here.
816 INP_LOCK_ASSERT(inp);
817 INP_HASH_LOCK_ASSERT(pcbinfo);
819 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
820 return (EADDRNOTAVAIL);
821 laddr.s_addr = *laddrp;
822 if (nam != NULL && laddr.s_addr != INADDR_ANY)
824 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
825 lookupflags = INPLOOKUP_WILDCARD;
827 if ((error = prison_local_ip4(cred, &laddr)) != 0)
830 sin = (struct sockaddr_in *)nam;
831 if (nam->sa_len != sizeof (*sin))
835 * We should check the family, but old programs
836 * incorrectly fail to initialize it.
838 if (sin->sin_family != AF_INET)
839 return (EAFNOSUPPORT);
841 error = prison_local_ip4(cred, &sin->sin_addr);
844 if (sin->sin_port != *lportp) {
845 /* Don't allow the port to change. */
848 lport = sin->sin_port;
850 /* NB: lport is left as 0 if the port isn't being changed. */
851 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
853 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
854 * allow complete duplication of binding if
855 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
856 * and a multicast address is bound on both
857 * new and duplicated sockets.
859 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
860 reuseport = SO_REUSEADDR|SO_REUSEPORT;
862 * XXX: How to deal with SO_REUSEPORT_LB here?
863 * Treat same as SO_REUSEPORT for now.
865 if ((so->so_options &
866 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
867 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
868 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
869 sin->sin_port = 0; /* yech... */
870 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
872 * Is the address a local IP address?
873 * If INP_BINDANY is set, then the socket may be bound
874 * to any endpoint address, local or not.
876 if ((inp->inp_flags & INP_BINDANY) == 0 &&
877 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
878 return (EADDRNOTAVAIL);
880 laddr = sin->sin_addr;
886 if (ntohs(lport) <= V_ipport_reservedhigh &&
887 ntohs(lport) >= V_ipport_reservedlow &&
888 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
890 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
891 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
892 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
893 lport, INPLOOKUP_WILDCARD, cred);
896 * This entire block sorely needs a rewrite.
899 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
900 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
901 (so->so_type != SOCK_STREAM ||
902 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
903 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
904 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
905 (t->inp_flags2 & INP_REUSEPORT) ||
906 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
907 (inp->inp_cred->cr_uid !=
908 t->inp_cred->cr_uid))
912 * If the socket is a BINDMULTI socket, then
913 * the credentials need to match and the
914 * original socket also has to have been bound
917 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
920 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
921 lport, lookupflags, cred);
922 if (t && (t->inp_flags & INP_TIMEWAIT)) {
924 * XXXRW: If an incpb has had its timewait
925 * state recycled, we treat the address as
926 * being in use (for now). This is better
927 * than a panic, but not desirable.
931 ((reuseport & tw->tw_so_options) == 0 &&
933 tw->tw_so_options) == 0)) {
937 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
938 (reuseport & inp_so_options(t)) == 0 &&
939 (reuseport_lb & inp_so_options(t)) == 0) {
941 if (ntohl(sin->sin_addr.s_addr) !=
943 ntohl(t->inp_laddr.s_addr) !=
945 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
946 (t->inp_vflag & INP_IPV6PROTO) == 0)
949 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
957 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
962 *laddrp = laddr.s_addr;
968 * Connect from a socket to a specified address.
969 * Both address and port must be specified in argument sin.
970 * If don't have a local address for this socket yet,
974 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
975 struct ucred *cred, struct mbuf *m)
977 u_short lport, fport;
978 in_addr_t laddr, faddr;
981 INP_WLOCK_ASSERT(inp);
982 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
984 lport = inp->inp_lport;
985 laddr = inp->inp_laddr.s_addr;
986 anonport = (lport == 0);
987 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
992 /* Do the initial binding of the local address if required. */
993 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
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;
1008 in_pcbrehash_mbuf(inp, m);
1011 inp->inp_flags |= INP_ANONPORT;
1016 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1019 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1023 * Do proper source address selection on an unbound socket in case
1024 * of connect. Take jails into account as well.
1027 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1031 struct sockaddr *sa;
1032 struct sockaddr_in *sin;
1034 struct epoch_tracker et;
1037 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1039 * Bypass source address selection and use the primary jail IP
1042 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1046 bzero(&sro, sizeof(sro));
1048 sin = (struct sockaddr_in *)&sro.ro_dst;
1049 sin->sin_family = AF_INET;
1050 sin->sin_len = sizeof(struct sockaddr_in);
1051 sin->sin_addr.s_addr = faddr->s_addr;
1054 * If route is known our src addr is taken from the i/f,
1057 * Find out route to destination.
1059 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1060 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1063 * If we found a route, use the address corresponding to
1064 * the outgoing interface.
1066 * Otherwise assume faddr is reachable on a directly connected
1067 * network and try to find a corresponding interface to take
1068 * the source address from.
1070 NET_EPOCH_ENTER(et);
1071 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1072 struct in_ifaddr *ia;
1075 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1076 inp->inp_socket->so_fibnum));
1078 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1079 inp->inp_socket->so_fibnum));
1083 error = ENETUNREACH;
1087 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1088 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1094 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1097 if (sa->sa_family != AF_INET)
1099 sin = (struct sockaddr_in *)sa;
1100 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1101 ia = (struct in_ifaddr *)ifa;
1106 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1110 /* 3. As a last resort return the 'default' jail address. */
1111 error = prison_get_ip4(cred, laddr);
1116 * If the outgoing interface on the route found is not
1117 * a loopback interface, use the address from that interface.
1118 * In case of jails do those three steps:
1119 * 1. check if the interface address belongs to the jail. If so use it.
1120 * 2. check if we have any address on the outgoing interface
1121 * belonging to this jail. If so use it.
1122 * 3. as a last resort return the 'default' jail address.
1124 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1125 struct in_ifaddr *ia;
1128 /* If not jailed, use the default returned. */
1129 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1130 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1131 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1136 /* 1. Check if the iface address belongs to the jail. */
1137 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1138 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1139 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1140 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1145 * 2. Check if we have any address on the outgoing interface
1146 * belonging to this jail.
1149 ifp = sro.ro_rt->rt_ifp;
1150 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1152 if (sa->sa_family != AF_INET)
1154 sin = (struct sockaddr_in *)sa;
1155 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1156 ia = (struct in_ifaddr *)ifa;
1161 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1165 /* 3. As a last resort return the 'default' jail address. */
1166 error = prison_get_ip4(cred, laddr);
1171 * The outgoing interface is marked with 'loopback net', so a route
1172 * to ourselves is here.
1173 * Try to find the interface of the destination address and then
1174 * take the address from there. That interface is not necessarily
1175 * a loopback interface.
1176 * In case of jails, check that it is an address of the jail
1177 * and if we cannot find, fall back to the 'default' jail address.
1179 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1180 struct sockaddr_in sain;
1181 struct in_ifaddr *ia;
1183 bzero(&sain, sizeof(struct sockaddr_in));
1184 sain.sin_family = AF_INET;
1185 sain.sin_len = sizeof(struct sockaddr_in);
1186 sain.sin_addr.s_addr = faddr->s_addr;
1188 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1189 inp->inp_socket->so_fibnum));
1191 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1192 inp->inp_socket->so_fibnum));
1194 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1196 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1198 error = ENETUNREACH;
1201 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1211 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1213 if (sa->sa_family != AF_INET)
1215 sin = (struct sockaddr_in *)sa;
1216 if (prison_check_ip4(cred,
1217 &sin->sin_addr) == 0) {
1218 ia = (struct in_ifaddr *)ifa;
1223 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1228 /* 3. As a last resort return the 'default' jail address. */
1229 error = prison_get_ip4(cred, laddr);
1235 if (sro.ro_rt != NULL)
1241 * Set up for a connect from a socket to the specified address.
1242 * On entry, *laddrp and *lportp should contain the current local
1243 * address and port for the PCB; these are updated to the values
1244 * that should be placed in inp_laddr and inp_lport to complete
1247 * On success, *faddrp and *fportp will be set to the remote address
1248 * and port. These are not updated in the error case.
1250 * If the operation fails because the connection already exists,
1251 * *oinpp will be set to the PCB of that connection so that the
1252 * caller can decide to override it. In all other cases, *oinpp
1256 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1257 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1258 struct inpcb **oinpp, struct ucred *cred)
1260 struct rm_priotracker in_ifa_tracker;
1261 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1262 struct in_ifaddr *ia;
1264 struct in_addr laddr, faddr;
1265 u_short lport, fport;
1269 * Because a global state change doesn't actually occur here, a read
1270 * lock is sufficient.
1272 INP_LOCK_ASSERT(inp);
1273 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1277 if (nam->sa_len != sizeof (*sin))
1279 if (sin->sin_family != AF_INET)
1280 return (EAFNOSUPPORT);
1281 if (sin->sin_port == 0)
1282 return (EADDRNOTAVAIL);
1283 laddr.s_addr = *laddrp;
1285 faddr = sin->sin_addr;
1286 fport = sin->sin_port;
1288 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1290 * If the destination address is INADDR_ANY,
1291 * use the primary local address.
1292 * If the supplied address is INADDR_BROADCAST,
1293 * and the primary interface supports broadcast,
1294 * choose the broadcast address for that interface.
1296 if (faddr.s_addr == INADDR_ANY) {
1297 IN_IFADDR_RLOCK(&in_ifa_tracker);
1299 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1300 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1302 (error = prison_get_ip4(cred, &faddr)) != 0)
1304 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1305 IN_IFADDR_RLOCK(&in_ifa_tracker);
1306 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1308 faddr = satosin(&CK_STAILQ_FIRST(
1309 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1310 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1313 if (laddr.s_addr == INADDR_ANY) {
1314 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1316 * If the destination address is multicast and an outgoing
1317 * interface has been set as a multicast option, prefer the
1318 * address of that interface as our source address.
1320 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1321 inp->inp_moptions != NULL) {
1322 struct ip_moptions *imo;
1325 imo = inp->inp_moptions;
1326 if (imo->imo_multicast_ifp != NULL) {
1327 ifp = imo->imo_multicast_ifp;
1328 IN_IFADDR_RLOCK(&in_ifa_tracker);
1329 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1330 if ((ia->ia_ifp == ifp) &&
1332 prison_check_ip4(cred,
1333 &ia->ia_addr.sin_addr) == 0))
1337 error = EADDRNOTAVAIL;
1339 laddr = ia->ia_addr.sin_addr;
1342 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1348 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1349 laddr, lport, 0, NULL);
1353 return (EADDRINUSE);
1356 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1361 *laddrp = laddr.s_addr;
1363 *faddrp = faddr.s_addr;
1369 in_pcbdisconnect(struct inpcb *inp)
1372 INP_WLOCK_ASSERT(inp);
1373 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1375 inp->inp_faddr.s_addr = INADDR_ANY;
1382 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1383 * For most protocols, this will be invoked immediately prior to calling
1384 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1385 * socket, in which case in_pcbfree() is deferred.
1388 in_pcbdetach(struct inpcb *inp)
1391 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1394 if (inp->inp_snd_tag != NULL)
1395 in_pcbdetach_txrtlmt(inp);
1397 inp->inp_socket->so_pcb = NULL;
1398 inp->inp_socket = NULL;
1402 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1403 * stability of an inpcb pointer despite the inpcb lock being released. This
1404 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1405 * but where the inpcb lock may already held, or when acquiring a reference
1408 * in_pcbref() should be used only to provide brief memory stability, and
1409 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1410 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1411 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1412 * lock and rele are the *only* safe operations that may be performed on the
1415 * While the inpcb will not be freed, releasing the inpcb lock means that the
1416 * connection's state may change, so the caller should be careful to
1417 * revalidate any cached state on reacquiring the lock. Drop the reference
1418 * using in_pcbrele().
1421 in_pcbref(struct inpcb *inp)
1424 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1426 refcount_acquire(&inp->inp_refcount);
1430 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1431 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1432 * return a flag indicating whether or not the inpcb remains valid. If it is
1433 * valid, we return with the inpcb lock held.
1435 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1436 * reference on an inpcb. Historically more work was done here (actually, in
1437 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1438 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1439 * about memory stability (and continued use of the write lock).
1442 in_pcbrele_rlocked(struct inpcb *inp)
1444 struct inpcbinfo *pcbinfo;
1446 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1448 INP_RLOCK_ASSERT(inp);
1450 if (refcount_release(&inp->inp_refcount) == 0) {
1452 * If the inpcb has been freed, let the caller know, even if
1453 * this isn't the last reference.
1455 if (inp->inp_flags2 & INP_FREED) {
1462 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1464 if (inp->inp_in_hpts || inp->inp_in_input) {
1465 struct tcp_hpts_entry *hpts;
1467 * We should not be on the hpts at
1468 * this point in any form. we must
1469 * get the lock to be sure.
1471 hpts = tcp_hpts_lock(inp);
1472 if (inp->inp_in_hpts)
1473 panic("Hpts:%p inp:%p at free still on hpts",
1475 mtx_unlock(&hpts->p_mtx);
1476 hpts = tcp_input_lock(inp);
1477 if (inp->inp_in_input)
1478 panic("Hpts:%p inp:%p at free still on input hpts",
1480 mtx_unlock(&hpts->p_mtx);
1484 pcbinfo = inp->inp_pcbinfo;
1485 uma_zfree(pcbinfo->ipi_zone, inp);
1490 in_pcbrele_wlocked(struct inpcb *inp)
1492 struct inpcbinfo *pcbinfo;
1494 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1496 INP_WLOCK_ASSERT(inp);
1498 if (refcount_release(&inp->inp_refcount) == 0) {
1500 * If the inpcb has been freed, let the caller know, even if
1501 * this isn't the last reference.
1503 if (inp->inp_flags2 & INP_FREED) {
1510 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1512 if (inp->inp_in_hpts || inp->inp_in_input) {
1513 struct tcp_hpts_entry *hpts;
1515 * We should not be on the hpts at
1516 * this point in any form. we must
1517 * get the lock to be sure.
1519 hpts = tcp_hpts_lock(inp);
1520 if (inp->inp_in_hpts)
1521 panic("Hpts:%p inp:%p at free still on hpts",
1523 mtx_unlock(&hpts->p_mtx);
1524 hpts = tcp_input_lock(inp);
1525 if (inp->inp_in_input)
1526 panic("Hpts:%p inp:%p at free still on input hpts",
1528 mtx_unlock(&hpts->p_mtx);
1532 pcbinfo = inp->inp_pcbinfo;
1533 uma_zfree(pcbinfo->ipi_zone, inp);
1538 * Temporary wrapper.
1541 in_pcbrele(struct inpcb *inp)
1544 return (in_pcbrele_wlocked(inp));
1548 in_pcblist_rele_rlocked(epoch_context_t ctx)
1550 struct in_pcblist *il;
1552 struct inpcbinfo *pcbinfo;
1555 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1556 pcbinfo = il->il_pcbinfo;
1558 INP_INFO_WLOCK(pcbinfo);
1559 for (i = 0; i < n; i++) {
1560 inp = il->il_inp_list[i];
1562 if (!in_pcbrele_rlocked(inp))
1565 INP_INFO_WUNLOCK(pcbinfo);
1570 inpcbport_free(epoch_context_t ctx)
1572 struct inpcbport *phd;
1574 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1579 in_pcbfree_deferred(epoch_context_t ctx)
1582 int released __unused;
1584 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1587 CURVNET_SET(inp->inp_vnet);
1589 struct ip_moptions *imo = inp->inp_moptions;
1590 inp->inp_moptions = NULL;
1592 /* XXXRW: Do as much as possible here. */
1593 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1594 if (inp->inp_sp != NULL)
1595 ipsec_delete_pcbpolicy(inp);
1598 struct ip6_moptions *im6o = NULL;
1599 if (inp->inp_vflag & INP_IPV6PROTO) {
1600 ip6_freepcbopts(inp->in6p_outputopts);
1601 im6o = inp->in6p_moptions;
1602 inp->in6p_moptions = NULL;
1605 if (inp->inp_options)
1606 (void)m_free(inp->inp_options);
1608 crfree(inp->inp_cred);
1610 mac_inpcb_destroy(inp);
1612 released = in_pcbrele_wlocked(inp);
1615 ip6_freemoptions(im6o);
1618 inp_freemoptions(imo);
1624 * Unconditionally schedule an inpcb to be freed by decrementing its
1625 * reference count, which should occur only after the inpcb has been detached
1626 * from its socket. If another thread holds a temporary reference (acquired
1627 * using in_pcbref()) then the free is deferred until that reference is
1628 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1629 * work, including removal from global lists, is done in this context, where
1630 * the pcbinfo lock is held.
1633 in_pcbfree(struct inpcb *inp)
1635 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1637 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1638 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1639 ("%s: called twice for pcb %p", __func__, inp));
1640 if (inp->inp_flags2 & INP_FREED) {
1646 if (pcbinfo == &V_tcbinfo) {
1647 INP_INFO_LOCK_ASSERT(pcbinfo);
1649 INP_INFO_WLOCK_ASSERT(pcbinfo);
1652 INP_WLOCK_ASSERT(inp);
1653 INP_LIST_WLOCK(pcbinfo);
1654 in_pcbremlists(inp);
1655 INP_LIST_WUNLOCK(pcbinfo);
1656 RO_INVALIDATE_CACHE(&inp->inp_route);
1657 /* mark as destruction in progress */
1658 inp->inp_flags2 |= INP_FREED;
1660 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1664 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1665 * port reservation, and preventing it from being returned by inpcb lookups.
1667 * It is used by TCP to mark an inpcb as unused and avoid future packet
1668 * delivery or event notification when a socket remains open but TCP has
1669 * closed. This might occur as a result of a shutdown()-initiated TCP close
1670 * or a RST on the wire, and allows the port binding to be reused while still
1671 * maintaining the invariant that so_pcb always points to a valid inpcb until
1674 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1675 * in_pcbnotifyall() and in_pcbpurgeif0()?
1678 in_pcbdrop(struct inpcb *inp)
1681 INP_WLOCK_ASSERT(inp);
1683 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1684 MPASS(inp->inp_refcount > 1);
1688 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1691 inp->inp_flags |= INP_DROPPED;
1692 if (inp->inp_flags & INP_INHASHLIST) {
1693 struct inpcbport *phd = inp->inp_phd;
1695 INP_HASH_WLOCK(inp->inp_pcbinfo);
1696 in_pcbremlbgrouphash(inp);
1697 CK_LIST_REMOVE(inp, inp_hash);
1698 CK_LIST_REMOVE(inp, inp_portlist);
1699 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1700 CK_LIST_REMOVE(phd, phd_hash);
1701 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1703 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1704 inp->inp_flags &= ~INP_INHASHLIST;
1706 in_pcbgroup_remove(inp);
1713 * Common routines to return the socket addresses associated with inpcbs.
1716 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1718 struct sockaddr_in *sin;
1720 sin = malloc(sizeof *sin, M_SONAME,
1722 sin->sin_family = AF_INET;
1723 sin->sin_len = sizeof(*sin);
1724 sin->sin_addr = *addr_p;
1725 sin->sin_port = port;
1727 return (struct sockaddr *)sin;
1731 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1734 struct in_addr addr;
1737 inp = sotoinpcb(so);
1738 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1741 port = inp->inp_lport;
1742 addr = inp->inp_laddr;
1745 *nam = in_sockaddr(port, &addr);
1750 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1753 struct in_addr addr;
1756 inp = sotoinpcb(so);
1757 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1760 port = inp->inp_fport;
1761 addr = inp->inp_faddr;
1764 *nam = in_sockaddr(port, &addr);
1769 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1770 struct inpcb *(*notify)(struct inpcb *, int))
1772 struct inpcb *inp, *inp_temp;
1774 INP_INFO_WLOCK(pcbinfo);
1775 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1778 if ((inp->inp_vflag & INP_IPV4) == 0) {
1783 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1784 inp->inp_socket == NULL) {
1788 if ((*notify)(inp, errno))
1791 INP_INFO_WUNLOCK(pcbinfo);
1795 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1798 struct in_multi *inm;
1799 struct in_mfilter *imf;
1800 struct ip_moptions *imo;
1802 INP_INFO_WLOCK(pcbinfo);
1803 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1805 imo = inp->inp_moptions;
1806 if ((inp->inp_vflag & INP_IPV4) &&
1809 * Unselect the outgoing interface if it is being
1812 if (imo->imo_multicast_ifp == ifp)
1813 imo->imo_multicast_ifp = NULL;
1816 * Drop multicast group membership if we joined
1817 * through the interface being detached.
1819 * XXX This can all be deferred to an epoch_call
1822 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1823 if ((inm = imf->imf_inm) == NULL)
1825 if (inm->inm_ifp != ifp)
1827 ip_mfilter_remove(&imo->imo_head, imf);
1828 IN_MULTI_LOCK_ASSERT();
1829 in_leavegroup_locked(inm, NULL);
1830 ip_mfilter_free(imf);
1836 INP_INFO_WUNLOCK(pcbinfo);
1840 * Lookup a PCB based on the local address and port. Caller must hold the
1841 * hash lock. No inpcb locks or references are acquired.
1843 #define INP_LOOKUP_MAPPED_PCB_COST 3
1845 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1846 u_short lport, int lookupflags, struct ucred *cred)
1850 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1856 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1857 ("%s: invalid lookup flags %d", __func__, lookupflags));
1859 INP_HASH_LOCK_ASSERT(pcbinfo);
1861 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1862 struct inpcbhead *head;
1864 * Look for an unconnected (wildcard foreign addr) PCB that
1865 * matches the local address and port we're looking for.
1867 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1868 0, pcbinfo->ipi_hashmask)];
1869 CK_LIST_FOREACH(inp, head, inp_hash) {
1871 /* XXX inp locking */
1872 if ((inp->inp_vflag & INP_IPV4) == 0)
1875 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1876 inp->inp_laddr.s_addr == laddr.s_addr &&
1877 inp->inp_lport == lport) {
1882 prison_equal_ip4(cred->cr_prison,
1883 inp->inp_cred->cr_prison))
1892 struct inpcbporthead *porthash;
1893 struct inpcbport *phd;
1894 struct inpcb *match = NULL;
1896 * Best fit PCB lookup.
1898 * First see if this local port is in use by looking on the
1901 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1902 pcbinfo->ipi_porthashmask)];
1903 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1904 if (phd->phd_port == lport)
1909 * Port is in use by one or more PCBs. Look for best
1912 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1915 !prison_equal_ip4(inp->inp_cred->cr_prison,
1919 /* XXX inp locking */
1920 if ((inp->inp_vflag & INP_IPV4) == 0)
1923 * We never select the PCB that has
1924 * INP_IPV6 flag and is bound to :: if
1925 * we have another PCB which is bound
1926 * to 0.0.0.0. If a PCB has the
1927 * INP_IPV6 flag, then we set its cost
1928 * higher than IPv4 only PCBs.
1930 * Note that the case only happens
1931 * when a socket is bound to ::, under
1932 * the condition that the use of the
1933 * mapped address is allowed.
1935 if ((inp->inp_vflag & INP_IPV6) != 0)
1936 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1938 if (inp->inp_faddr.s_addr != INADDR_ANY)
1940 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1941 if (laddr.s_addr == INADDR_ANY)
1943 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1946 if (laddr.s_addr != INADDR_ANY)
1949 if (wildcard < matchwild) {
1951 matchwild = wildcard;
1960 #undef INP_LOOKUP_MAPPED_PCB_COST
1962 static struct inpcb *
1963 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1964 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1965 uint16_t fport, int lookupflags)
1967 struct inpcb *local_wild;
1968 const struct inpcblbgrouphead *hdr;
1969 struct inpcblbgroup *grp;
1972 INP_HASH_LOCK_ASSERT(pcbinfo);
1974 hdr = &pcbinfo->ipi_lbgrouphashbase[
1975 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1978 * Order of socket selection:
1980 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1983 * - Load balanced group does not contain jailed sockets
1984 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1987 CK_LIST_FOREACH(grp, hdr, il_list) {
1989 if (!(grp->il_vflag & INP_IPV4))
1992 if (grp->il_lport != lport)
1995 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1997 if (grp->il_laddr.s_addr == laddr->s_addr)
1998 return (grp->il_inp[idx]);
1999 if (grp->il_laddr.s_addr == INADDR_ANY &&
2000 (lookupflags & INPLOOKUP_WILDCARD) != 0)
2001 local_wild = grp->il_inp[idx];
2003 return (local_wild);
2008 * Lookup PCB in hash list, using pcbgroup tables.
2010 static struct inpcb *
2011 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2012 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2013 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2015 struct inpcbhead *head;
2016 struct inpcb *inp, *tmpinp;
2017 u_short fport = fport_arg, lport = lport_arg;
2021 * First look for an exact match.
2024 INP_GROUP_LOCK(pcbgroup);
2025 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2026 pcbgroup->ipg_hashmask)];
2027 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2029 /* XXX inp locking */
2030 if ((inp->inp_vflag & INP_IPV4) == 0)
2033 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2034 inp->inp_laddr.s_addr == laddr.s_addr &&
2035 inp->inp_fport == fport &&
2036 inp->inp_lport == lport) {
2038 * XXX We should be able to directly return
2039 * the inp here, without any checks.
2040 * Well unless both bound with SO_REUSEPORT?
2042 if (prison_flag(inp->inp_cred, PR_IP4))
2048 if (tmpinp != NULL) {
2055 * For incoming connections, we may wish to do a wildcard
2056 * match for an RSS-local socket.
2058 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2059 struct inpcb *local_wild = NULL, *local_exact = NULL;
2061 struct inpcb *local_wild_mapped = NULL;
2063 struct inpcb *jail_wild = NULL;
2064 struct inpcbhead *head;
2068 * Order of socket selection - we always prefer jails.
2069 * 1. jailed, non-wild.
2071 * 3. non-jailed, non-wild.
2072 * 4. non-jailed, wild.
2075 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2076 lport, 0, pcbgroup->ipg_hashmask)];
2077 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2079 /* XXX inp locking */
2080 if ((inp->inp_vflag & INP_IPV4) == 0)
2083 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2084 inp->inp_lport != lport)
2087 injail = prison_flag(inp->inp_cred, PR_IP4);
2089 if (prison_check_ip4(inp->inp_cred,
2093 if (local_exact != NULL)
2097 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2102 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2104 /* XXX inp locking, NULL check */
2105 if (inp->inp_vflag & INP_IPV6PROTO)
2106 local_wild_mapped = inp;
2114 } /* LIST_FOREACH */
2123 inp = local_wild_mapped;
2131 * Then look for a wildcard match, if requested.
2133 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2134 struct inpcb *local_wild = NULL, *local_exact = NULL;
2136 struct inpcb *local_wild_mapped = NULL;
2138 struct inpcb *jail_wild = NULL;
2139 struct inpcbhead *head;
2143 * Order of socket selection - we always prefer jails.
2144 * 1. jailed, non-wild.
2146 * 3. non-jailed, non-wild.
2147 * 4. non-jailed, wild.
2149 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2150 0, pcbinfo->ipi_wildmask)];
2151 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2153 /* XXX inp locking */
2154 if ((inp->inp_vflag & INP_IPV4) == 0)
2157 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2158 inp->inp_lport != lport)
2161 injail = prison_flag(inp->inp_cred, PR_IP4);
2163 if (prison_check_ip4(inp->inp_cred,
2167 if (local_exact != NULL)
2171 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2176 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2178 /* XXX inp locking, NULL check */
2179 if (inp->inp_vflag & INP_IPV6PROTO)
2180 local_wild_mapped = inp;
2188 } /* LIST_FOREACH */
2196 inp = local_wild_mapped;
2200 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2201 INP_GROUP_UNLOCK(pcbgroup);
2205 if (lookupflags & INPLOOKUP_WLOCKPCB)
2206 locked = INP_TRY_WLOCK(inp);
2207 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2208 locked = INP_TRY_RLOCK(inp);
2210 panic("%s: locking bug", __func__);
2211 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2212 if (lookupflags & INPLOOKUP_WLOCKPCB)
2219 INP_GROUP_UNLOCK(pcbgroup);
2221 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2223 if (in_pcbrele_wlocked(inp))
2227 if (in_pcbrele_rlocked(inp))
2232 if (lookupflags & INPLOOKUP_WLOCKPCB)
2233 INP_WLOCK_ASSERT(inp);
2235 INP_RLOCK_ASSERT(inp);
2239 #endif /* PCBGROUP */
2242 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2243 * that the caller has locked the hash list, and will not perform any further
2244 * locking or reference operations on either the hash list or the connection.
2246 static struct inpcb *
2247 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2248 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2251 struct inpcbhead *head;
2252 struct inpcb *inp, *tmpinp;
2253 u_short fport = fport_arg, lport = lport_arg;
2256 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2257 ("%s: invalid lookup flags %d", __func__, lookupflags));
2258 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2259 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2262 * First look for an exact match.
2265 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2266 pcbinfo->ipi_hashmask)];
2267 CK_LIST_FOREACH(inp, head, inp_hash) {
2269 /* XXX inp locking */
2270 if ((inp->inp_vflag & INP_IPV4) == 0)
2273 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2274 inp->inp_laddr.s_addr == laddr.s_addr &&
2275 inp->inp_fport == fport &&
2276 inp->inp_lport == lport) {
2278 * XXX We should be able to directly return
2279 * the inp here, without any checks.
2280 * Well unless both bound with SO_REUSEPORT?
2282 if (prison_flag(inp->inp_cred, PR_IP4))
2292 * Then look in lb group (for wildcard match).
2294 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2295 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2296 fport, lookupflags);
2302 * Then look for a wildcard match, if requested.
2304 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2305 struct inpcb *local_wild = NULL, *local_exact = NULL;
2307 struct inpcb *local_wild_mapped = NULL;
2309 struct inpcb *jail_wild = NULL;
2313 * Order of socket selection - we always prefer jails.
2314 * 1. jailed, non-wild.
2316 * 3. non-jailed, non-wild.
2317 * 4. non-jailed, wild.
2320 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2321 0, pcbinfo->ipi_hashmask)];
2322 CK_LIST_FOREACH(inp, head, inp_hash) {
2324 /* XXX inp locking */
2325 if ((inp->inp_vflag & INP_IPV4) == 0)
2328 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2329 inp->inp_lport != lport)
2332 injail = prison_flag(inp->inp_cred, PR_IP4);
2334 if (prison_check_ip4(inp->inp_cred,
2338 if (local_exact != NULL)
2342 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2347 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2349 /* XXX inp locking, NULL check */
2350 if (inp->inp_vflag & INP_IPV6PROTO)
2351 local_wild_mapped = inp;
2359 } /* LIST_FOREACH */
2360 if (jail_wild != NULL)
2362 if (local_exact != NULL)
2363 return (local_exact);
2364 if (local_wild != NULL)
2365 return (local_wild);
2367 if (local_wild_mapped != NULL)
2368 return (local_wild_mapped);
2370 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2376 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2377 * hash list lock, and will return the inpcb locked (i.e., requires
2378 * INPLOOKUP_LOCKPCB).
2380 static struct inpcb *
2381 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2382 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2387 INP_HASH_RLOCK(pcbinfo);
2388 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2389 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2391 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2393 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2397 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2399 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2404 panic("%s: locking bug", __func__);
2407 if (lookupflags & INPLOOKUP_WLOCKPCB)
2408 INP_WLOCK_ASSERT(inp);
2410 INP_RLOCK_ASSERT(inp);
2414 INP_HASH_RUNLOCK(pcbinfo);
2419 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2420 * from which a pre-calculated hash value may be extracted.
2422 * Possibly more of this logic should be in in_pcbgroup.c.
2425 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2426 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2428 #if defined(PCBGROUP) && !defined(RSS)
2429 struct inpcbgroup *pcbgroup;
2432 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2433 ("%s: invalid lookup flags %d", __func__, lookupflags));
2434 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2435 ("%s: LOCKPCB not set", __func__));
2438 * When not using RSS, use connection groups in preference to the
2439 * reservation table when looking up 4-tuples. When using RSS, just
2440 * use the reservation table, due to the cost of the Toeplitz hash
2443 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2444 * we could be doing RSS with a non-Toeplitz hash that is affordable
2447 #if defined(PCBGROUP) && !defined(RSS)
2448 if (in_pcbgroup_enabled(pcbinfo)) {
2449 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2451 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2452 laddr, lport, lookupflags, ifp));
2455 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2460 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2461 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2462 struct ifnet *ifp, struct mbuf *m)
2465 struct inpcbgroup *pcbgroup;
2468 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2469 ("%s: invalid lookup flags %d", __func__, lookupflags));
2470 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2471 ("%s: LOCKPCB not set", __func__));
2475 * If we can use a hardware-generated hash to look up the connection
2476 * group, use that connection group to find the inpcb. Otherwise
2477 * fall back on a software hash -- or the reservation table if we're
2480 * XXXRW: As above, that policy belongs in the pcbgroup code.
2482 if (in_pcbgroup_enabled(pcbinfo) &&
2483 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2484 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2485 m->m_pkthdr.flowid);
2486 if (pcbgroup != NULL)
2487 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2488 fport, laddr, lport, lookupflags, ifp));
2490 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2492 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2493 laddr, lport, lookupflags, ifp));
2497 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2503 * Insert PCB onto various hash lists.
2506 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2508 struct inpcbhead *pcbhash;
2509 struct inpcbporthead *pcbporthash;
2510 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2511 struct inpcbport *phd;
2512 u_int32_t hashkey_faddr;
2515 INP_WLOCK_ASSERT(inp);
2516 INP_HASH_WLOCK_ASSERT(pcbinfo);
2518 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2519 ("in_pcbinshash: INP_INHASHLIST"));
2522 if (inp->inp_vflag & INP_IPV6)
2523 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2526 hashkey_faddr = inp->inp_faddr.s_addr;
2528 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2529 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2531 pcbporthash = &pcbinfo->ipi_porthashbase[
2532 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2535 * Add entry to load balance group.
2536 * Only do this if SO_REUSEPORT_LB is set.
2538 so_options = inp_so_options(inp);
2539 if (so_options & SO_REUSEPORT_LB) {
2540 int ret = in_pcbinslbgrouphash(inp);
2542 /* pcb lb group malloc fail (ret=ENOBUFS). */
2548 * Go through port list and look for a head for this lport.
2550 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2551 if (phd->phd_port == inp->inp_lport)
2555 * If none exists, malloc one and tack it on.
2558 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2560 return (ENOBUFS); /* XXX */
2562 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2563 phd->phd_port = inp->inp_lport;
2564 CK_LIST_INIT(&phd->phd_pcblist);
2565 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2568 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2569 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2570 inp->inp_flags |= INP_INHASHLIST;
2572 if (do_pcbgroup_update)
2573 in_pcbgroup_update(inp);
2579 * For now, there are two public interfaces to insert an inpcb into the hash
2580 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2581 * is used only in the TCP syncache, where in_pcbinshash is called before the
2582 * full 4-tuple is set for the inpcb, and we don't want to install in the
2583 * pcbgroup until later.
2585 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2586 * connection groups, and partially initialised inpcbs should not be exposed
2587 * to either reservation hash tables or pcbgroups.
2590 in_pcbinshash(struct inpcb *inp)
2593 return (in_pcbinshash_internal(inp, 1));
2597 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2600 return (in_pcbinshash_internal(inp, 0));
2604 * Move PCB to the proper hash bucket when { faddr, fport } have been
2605 * changed. NOTE: This does not handle the case of the lport changing (the
2606 * hashed port list would have to be updated as well), so the lport must
2607 * not change after in_pcbinshash() has been called.
2610 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2612 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2613 struct inpcbhead *head;
2614 u_int32_t hashkey_faddr;
2616 INP_WLOCK_ASSERT(inp);
2617 INP_HASH_WLOCK_ASSERT(pcbinfo);
2619 KASSERT(inp->inp_flags & INP_INHASHLIST,
2620 ("in_pcbrehash: !INP_INHASHLIST"));
2623 if (inp->inp_vflag & INP_IPV6)
2624 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2627 hashkey_faddr = inp->inp_faddr.s_addr;
2629 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2630 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2632 CK_LIST_REMOVE(inp, inp_hash);
2633 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2637 in_pcbgroup_update_mbuf(inp, m);
2639 in_pcbgroup_update(inp);
2644 in_pcbrehash(struct inpcb *inp)
2647 in_pcbrehash_mbuf(inp, NULL);
2651 * Remove PCB from various lists.
2654 in_pcbremlists(struct inpcb *inp)
2656 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2659 if (pcbinfo == &V_tcbinfo) {
2660 INP_INFO_RLOCK_ASSERT(pcbinfo);
2662 INP_INFO_WLOCK_ASSERT(pcbinfo);
2666 INP_WLOCK_ASSERT(inp);
2667 INP_LIST_WLOCK_ASSERT(pcbinfo);
2669 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2670 if (inp->inp_flags & INP_INHASHLIST) {
2671 struct inpcbport *phd = inp->inp_phd;
2673 INP_HASH_WLOCK(pcbinfo);
2675 /* XXX: Only do if SO_REUSEPORT_LB set? */
2676 in_pcbremlbgrouphash(inp);
2678 CK_LIST_REMOVE(inp, inp_hash);
2679 CK_LIST_REMOVE(inp, inp_portlist);
2680 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2681 CK_LIST_REMOVE(phd, phd_hash);
2682 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2684 INP_HASH_WUNLOCK(pcbinfo);
2685 inp->inp_flags &= ~INP_INHASHLIST;
2687 CK_LIST_REMOVE(inp, inp_list);
2688 pcbinfo->ipi_count--;
2690 in_pcbgroup_remove(inp);
2695 * Check for alternatives when higher level complains
2696 * about service problems. For now, invalidate cached
2697 * routing information. If the route was created dynamically
2698 * (by a redirect), time to try a default gateway again.
2701 in_losing(struct inpcb *inp)
2704 RO_INVALIDATE_CACHE(&inp->inp_route);
2709 * A set label operation has occurred at the socket layer, propagate the
2710 * label change into the in_pcb for the socket.
2713 in_pcbsosetlabel(struct socket *so)
2718 inp = sotoinpcb(so);
2719 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2723 mac_inpcb_sosetlabel(so, inp);
2730 * ipport_tick runs once per second, determining if random port allocation
2731 * should be continued. If more than ipport_randomcps ports have been
2732 * allocated in the last second, then we return to sequential port
2733 * allocation. We return to random allocation only once we drop below
2734 * ipport_randomcps for at least ipport_randomtime seconds.
2737 ipport_tick(void *xtp)
2739 VNET_ITERATOR_DECL(vnet_iter);
2741 VNET_LIST_RLOCK_NOSLEEP();
2742 VNET_FOREACH(vnet_iter) {
2743 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2744 if (V_ipport_tcpallocs <=
2745 V_ipport_tcplastcount + V_ipport_randomcps) {
2746 if (V_ipport_stoprandom > 0)
2747 V_ipport_stoprandom--;
2749 V_ipport_stoprandom = V_ipport_randomtime;
2750 V_ipport_tcplastcount = V_ipport_tcpallocs;
2753 VNET_LIST_RUNLOCK_NOSLEEP();
2754 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2761 callout_stop(&ipport_tick_callout);
2765 * The ipport_callout should start running at about the time we attach the
2766 * inet or inet6 domains.
2769 ipport_tick_init(const void *unused __unused)
2772 /* Start ipport_tick. */
2773 callout_init(&ipport_tick_callout, 1);
2774 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2775 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2776 SHUTDOWN_PRI_DEFAULT);
2778 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2779 ipport_tick_init, NULL);
2782 inp_wlock(struct inpcb *inp)
2789 inp_wunlock(struct inpcb *inp)
2796 inp_rlock(struct inpcb *inp)
2803 inp_runlock(struct inpcb *inp)
2809 #ifdef INVARIANT_SUPPORT
2811 inp_lock_assert(struct inpcb *inp)
2814 INP_WLOCK_ASSERT(inp);
2818 inp_unlock_assert(struct inpcb *inp)
2821 INP_UNLOCK_ASSERT(inp);
2826 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2830 INP_INFO_WLOCK(&V_tcbinfo);
2831 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2836 INP_INFO_WUNLOCK(&V_tcbinfo);
2840 inp_inpcbtosocket(struct inpcb *inp)
2843 INP_WLOCK_ASSERT(inp);
2844 return (inp->inp_socket);
2848 inp_inpcbtotcpcb(struct inpcb *inp)
2851 INP_WLOCK_ASSERT(inp);
2852 return ((struct tcpcb *)inp->inp_ppcb);
2856 inp_ip_tos_get(const struct inpcb *inp)
2859 return (inp->inp_ip_tos);
2863 inp_ip_tos_set(struct inpcb *inp, int val)
2866 inp->inp_ip_tos = val;
2870 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2871 uint32_t *faddr, uint16_t *fp)
2874 INP_LOCK_ASSERT(inp);
2875 *laddr = inp->inp_laddr.s_addr;
2876 *faddr = inp->inp_faddr.s_addr;
2877 *lp = inp->inp_lport;
2878 *fp = inp->inp_fport;
2882 so_sotoinpcb(struct socket *so)
2885 return (sotoinpcb(so));
2889 so_sototcpcb(struct socket *so)
2892 return (sototcpcb(so));
2896 * Create an external-format (``xinpcb'') structure using the information in
2897 * the kernel-format in_pcb structure pointed to by inp. This is done to
2898 * reduce the spew of irrelevant information over this interface, to isolate
2899 * user code from changes in the kernel structure, and potentially to provide
2900 * information-hiding if we decide that some of this information should be
2901 * hidden from users.
2904 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2907 bzero(xi, sizeof(*xi));
2908 xi->xi_len = sizeof(struct xinpcb);
2909 if (inp->inp_socket)
2910 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2911 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2912 xi->inp_gencnt = inp->inp_gencnt;
2913 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2914 xi->inp_flow = inp->inp_flow;
2915 xi->inp_flowid = inp->inp_flowid;
2916 xi->inp_flowtype = inp->inp_flowtype;
2917 xi->inp_flags = inp->inp_flags;
2918 xi->inp_flags2 = inp->inp_flags2;
2919 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2920 xi->in6p_cksum = inp->in6p_cksum;
2921 xi->in6p_hops = inp->in6p_hops;
2922 xi->inp_ip_tos = inp->inp_ip_tos;
2923 xi->inp_vflag = inp->inp_vflag;
2924 xi->inp_ip_ttl = inp->inp_ip_ttl;
2925 xi->inp_ip_p = inp->inp_ip_p;
2926 xi->inp_ip_minttl = inp->inp_ip_minttl;
2931 db_print_indent(int indent)
2935 for (i = 0; i < indent; i++)
2940 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2942 char faddr_str[48], laddr_str[48];
2944 db_print_indent(indent);
2945 db_printf("%s at %p\n", name, inc);
2950 if (inc->inc_flags & INC_ISIPV6) {
2952 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2953 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2958 inet_ntoa_r(inc->inc_laddr, laddr_str);
2959 inet_ntoa_r(inc->inc_faddr, faddr_str);
2961 db_print_indent(indent);
2962 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2963 ntohs(inc->inc_lport));
2964 db_print_indent(indent);
2965 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2966 ntohs(inc->inc_fport));
2970 db_print_inpflags(int inp_flags)
2975 if (inp_flags & INP_RECVOPTS) {
2976 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2979 if (inp_flags & INP_RECVRETOPTS) {
2980 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2983 if (inp_flags & INP_RECVDSTADDR) {
2984 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2987 if (inp_flags & INP_ORIGDSTADDR) {
2988 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2991 if (inp_flags & INP_HDRINCL) {
2992 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2995 if (inp_flags & INP_HIGHPORT) {
2996 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2999 if (inp_flags & INP_LOWPORT) {
3000 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3003 if (inp_flags & INP_ANONPORT) {
3004 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3007 if (inp_flags & INP_RECVIF) {
3008 db_printf("%sINP_RECVIF", comma ? ", " : "");
3011 if (inp_flags & INP_MTUDISC) {
3012 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3015 if (inp_flags & INP_RECVTTL) {
3016 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3019 if (inp_flags & INP_DONTFRAG) {
3020 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3023 if (inp_flags & INP_RECVTOS) {
3024 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3027 if (inp_flags & IN6P_IPV6_V6ONLY) {
3028 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3031 if (inp_flags & IN6P_PKTINFO) {
3032 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3035 if (inp_flags & IN6P_HOPLIMIT) {
3036 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3039 if (inp_flags & IN6P_HOPOPTS) {
3040 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3043 if (inp_flags & IN6P_DSTOPTS) {
3044 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3047 if (inp_flags & IN6P_RTHDR) {
3048 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3051 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3052 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3055 if (inp_flags & IN6P_TCLASS) {
3056 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3059 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3060 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3063 if (inp_flags & INP_TIMEWAIT) {
3064 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3067 if (inp_flags & INP_ONESBCAST) {
3068 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3071 if (inp_flags & INP_DROPPED) {
3072 db_printf("%sINP_DROPPED", comma ? ", " : "");
3075 if (inp_flags & INP_SOCKREF) {
3076 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3079 if (inp_flags & IN6P_RFC2292) {
3080 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3083 if (inp_flags & IN6P_MTU) {
3084 db_printf("IN6P_MTU%s", comma ? ", " : "");
3090 db_print_inpvflag(u_char inp_vflag)
3095 if (inp_vflag & INP_IPV4) {
3096 db_printf("%sINP_IPV4", comma ? ", " : "");
3099 if (inp_vflag & INP_IPV6) {
3100 db_printf("%sINP_IPV6", comma ? ", " : "");
3103 if (inp_vflag & INP_IPV6PROTO) {
3104 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3110 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3113 db_print_indent(indent);
3114 db_printf("%s at %p\n", name, inp);
3118 db_print_indent(indent);
3119 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3121 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3123 db_print_indent(indent);
3124 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3125 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3127 db_print_indent(indent);
3128 db_printf("inp_label: %p inp_flags: 0x%x (",
3129 inp->inp_label, inp->inp_flags);
3130 db_print_inpflags(inp->inp_flags);
3133 db_print_indent(indent);
3134 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3136 db_print_inpvflag(inp->inp_vflag);
3139 db_print_indent(indent);
3140 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3141 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3143 db_print_indent(indent);
3145 if (inp->inp_vflag & INP_IPV6) {
3146 db_printf("in6p_options: %p in6p_outputopts: %p "
3147 "in6p_moptions: %p\n", inp->in6p_options,
3148 inp->in6p_outputopts, inp->in6p_moptions);
3149 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3150 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3155 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3156 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3157 inp->inp_options, inp->inp_moptions);
3160 db_print_indent(indent);
3161 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3162 (uintmax_t)inp->inp_gencnt);
3165 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3170 db_printf("usage: show inpcb <addr>\n");
3173 inp = (struct inpcb *)addr;
3175 db_print_inpcb(inp, "inpcb", 0);
3181 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3185 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3187 union if_snd_tag_modify_params params = {
3188 .rate_limit.max_rate = max_pacing_rate,
3189 .rate_limit.flags = M_NOWAIT,
3191 struct m_snd_tag *mst;
3195 mst = inp->inp_snd_tag;
3203 if (ifp->if_snd_tag_modify == NULL) {
3206 error = ifp->if_snd_tag_modify(mst, ¶ms);
3212 * Query existing TX rate limit based on the existing
3213 * "inp->inp_snd_tag", if any.
3216 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3218 union if_snd_tag_query_params params = { };
3219 struct m_snd_tag *mst;
3223 mst = inp->inp_snd_tag;
3231 if (ifp->if_snd_tag_query == NULL) {
3234 error = ifp->if_snd_tag_query(mst, ¶ms);
3235 if (error == 0 && p_max_pacing_rate != NULL)
3236 *p_max_pacing_rate = params.rate_limit.max_rate;
3242 * Query existing TX queue level based on the existing
3243 * "inp->inp_snd_tag", if any.
3246 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3248 union if_snd_tag_query_params params = { };
3249 struct m_snd_tag *mst;
3253 mst = inp->inp_snd_tag;
3261 if (ifp->if_snd_tag_query == NULL)
3262 return (EOPNOTSUPP);
3264 error = ifp->if_snd_tag_query(mst, ¶ms);
3265 if (error == 0 && p_txqueue_level != NULL)
3266 *p_txqueue_level = params.rate_limit.queue_level;
3271 * Allocate a new TX rate limit send tag from the network interface
3272 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3275 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3276 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3279 union if_snd_tag_alloc_params params = {
3280 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3281 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3282 .rate_limit.hdr.flowid = flowid,
3283 .rate_limit.hdr.flowtype = flowtype,
3284 .rate_limit.max_rate = max_pacing_rate,
3285 .rate_limit.flags = M_NOWAIT,
3289 INP_WLOCK_ASSERT(inp);
3294 if (ifp->if_snd_tag_alloc == NULL) {
3297 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3301 counter_u64_add(rate_limit_set_ok, 1);
3302 counter_u64_add(rate_limit_active, 1);
3304 counter_u64_add(rate_limit_alloc_fail, 1);
3311 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3317 * If the device was detached while we still had reference(s)
3318 * on the ifp, we assume if_snd_tag_free() was replaced with
3321 ifp->if_snd_tag_free(mst);
3323 /* release reference count on network interface */
3326 counter_u64_add(rate_limit_active, -1);
3331 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3335 in_pcbdetach_txrtlmt(struct inpcb *inp)
3337 struct m_snd_tag *mst;
3339 INP_WLOCK_ASSERT(inp);
3341 mst = inp->inp_snd_tag;
3342 inp->inp_snd_tag = NULL;
3347 m_snd_tag_rele(mst);
3351 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3356 * If the existing send tag is for the wrong interface due to
3357 * a route change, first drop the existing tag. Set the
3358 * CHANGED flag so that we will keep trying to allocate a new
3359 * tag if we fail to allocate one this time.
3361 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3362 in_pcbdetach_txrtlmt(inp);
3363 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3367 * NOTE: When attaching to a network interface a reference is
3368 * made to ensure the network interface doesn't go away until
3369 * all ratelimit connections are gone. The network interface
3370 * pointers compared below represent valid network interfaces,
3371 * except when comparing towards NULL.
3373 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3375 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3376 if (inp->inp_snd_tag != NULL)
3377 in_pcbdetach_txrtlmt(inp);
3379 } else if (inp->inp_snd_tag == NULL) {
3381 * In order to utilize packet pacing with RSS, we need
3382 * to wait until there is a valid RSS hash before we
3385 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3388 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3389 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3392 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3394 if (error == 0 || error == EOPNOTSUPP)
3395 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3401 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3402 * is set in the fast path and will attach/detach/modify the TX rate
3403 * limit send tag based on the socket's so_max_pacing_rate value.
3406 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3408 struct socket *socket;
3409 uint32_t max_pacing_rate;
3416 socket = inp->inp_socket;
3420 if (!INP_WLOCKED(inp)) {
3422 * NOTE: If the write locking fails, we need to bail
3423 * out and use the non-ratelimited ring for the
3424 * transmit until there is a new chance to get the
3427 if (!INP_TRY_UPGRADE(inp))
3435 * NOTE: The so_max_pacing_rate value is read unlocked,
3436 * because atomic updates are not required since the variable
3437 * is checked at every mbuf we send. It is assumed that the
3438 * variable read itself will be atomic.
3440 max_pacing_rate = socket->so_max_pacing_rate;
3442 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3449 * Track route changes for TX rate limiting.
3452 in_pcboutput_eagain(struct inpcb *inp)
3459 if (inp->inp_snd_tag == NULL)
3462 if (!INP_WLOCKED(inp)) {
3464 * NOTE: If the write locking fails, we need to bail
3465 * out and use the non-ratelimited ring for the
3466 * transmit until there is a new chance to get the
3469 if (!INP_TRY_UPGRADE(inp))
3476 /* detach rate limiting */
3477 in_pcbdetach_txrtlmt(inp);
3479 /* make sure new mbuf send tag allocation is made */
3480 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3490 rate_limit_active = counter_u64_alloc(M_WAITOK);
3491 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3492 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3495 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3497 #endif /* RATELIMIT */