2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1991, 1993, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2007-2009 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 #include "opt_ipsec.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/malloc.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
67 #include <sys/refcount.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/ip_var.h>
90 #include <netinet/tcp_var.h>
92 #include <netinet/tcp_hpts.h>
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
98 #include <netinet/in_var.h>
101 #include <netinet/ip6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
107 #include <netipsec/ipsec_support.h>
109 #include <security/mac/mac_framework.h>
111 #define INPCBLBGROUP_SIZMIN 8
112 #define INPCBLBGROUP_SIZMAX 256
114 static struct callout ipport_tick_callout;
117 * These configure the range of local port addresses assigned to
118 * "unspecified" outgoing connections/packets/whatever.
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
128 * Reserved ports accessible only to root. There are significant
129 * security considerations that must be accounted for when changing these,
130 * but the security benefits can be great. Please be careful.
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
145 static void in_pcbremlists(struct inpcb *inp);
147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 struct in_addr faddr, u_int fport_arg,
149 struct in_addr laddr, u_int lport_arg,
150 int lookupflags, struct ifnet *ifp);
152 #define RANGECHK(var, min, max) \
153 if ((var) < (min)) { (var) = (min); } \
154 else if ((var) > (max)) { (var) = (max); }
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
161 error = sysctl_handle_int(oidp, arg1, arg2, req);
163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 &VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 CTLFLAG_VNET | CTLFLAG_RW,
206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 "allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(ipport_randomtime), 0,
211 "Minimum time to keep sequental port "
212 "allocation before switching to a random one");
216 * in_pcb.c: manage the Protocol Control Blocks.
218 * NOTE: It is assumed that most of these functions will be called with
219 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
220 * functions often modify hash chains or addresses in pcbs.
223 static struct inpcblbgroup *
224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
225 uint16_t port, const union in_dependaddr *addr, int size)
227 struct inpcblbgroup *grp;
230 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
231 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
234 grp->il_vflag = vflag;
235 grp->il_lport = port;
236 grp->il_dependladdr = *addr;
237 grp->il_inpsiz = size;
238 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
243 in_pcblbgroup_free_deferred(epoch_context_t ctx)
245 struct inpcblbgroup *grp;
247 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
252 in_pcblbgroup_free(struct inpcblbgroup *grp)
255 CK_LIST_REMOVE(grp, il_list);
256 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
257 in_pcblbgroup_free_deferred);
260 static struct inpcblbgroup *
261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
262 struct inpcblbgroup *old_grp, int size)
264 struct inpcblbgroup *grp;
267 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
268 old_grp->il_lport, &old_grp->il_dependladdr, size);
272 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
273 ("invalid new local group size %d and old local group count %d",
274 grp->il_inpsiz, old_grp->il_inpcnt));
276 for (i = 0; i < old_grp->il_inpcnt; ++i)
277 grp->il_inp[i] = old_grp->il_inp[i];
278 grp->il_inpcnt = old_grp->il_inpcnt;
279 in_pcblbgroup_free(old_grp);
284 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
285 * and shrink group if possible.
288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
291 struct inpcblbgroup *grp, *new_grp;
294 for (; i + 1 < grp->il_inpcnt; ++i)
295 grp->il_inp[i] = grp->il_inp[i + 1];
298 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
299 grp->il_inpcnt <= grp->il_inpsiz / 4) {
300 /* Shrink this group. */
301 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
308 * Add PCB to load balance group for SO_REUSEPORT_LB option.
311 in_pcbinslbgrouphash(struct inpcb *inp)
313 const static struct timeval interval = { 60, 0 };
314 static struct timeval lastprint;
315 struct inpcbinfo *pcbinfo;
316 struct inpcblbgrouphead *hdr;
317 struct inpcblbgroup *grp;
320 pcbinfo = inp->inp_pcbinfo;
322 INP_WLOCK_ASSERT(inp);
323 INP_HASH_WLOCK_ASSERT(pcbinfo);
326 * Don't allow jailed socket to join local group.
328 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
333 * Don't allow IPv4 mapped INET6 wild socket.
335 if ((inp->inp_vflag & INP_IPV4) &&
336 inp->inp_laddr.s_addr == INADDR_ANY &&
337 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
342 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
343 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
344 CK_LIST_FOREACH(grp, hdr, il_list) {
345 if (grp->il_vflag == inp->inp_vflag &&
346 grp->il_lport == inp->inp_lport &&
347 memcmp(&grp->il_dependladdr,
348 &inp->inp_inc.inc_ie.ie_dependladdr,
349 sizeof(grp->il_dependladdr)) == 0)
353 /* Create new load balance group. */
354 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
355 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
356 INPCBLBGROUP_SIZMIN);
359 } else if (grp->il_inpcnt == grp->il_inpsiz) {
360 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
361 if (ratecheck(&lastprint, &interval))
362 printf("lb group port %d, limit reached\n",
363 ntohs(grp->il_lport));
367 /* Expand this local group. */
368 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
373 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
374 ("invalid local group size %d and count %d", grp->il_inpsiz,
377 grp->il_inp[grp->il_inpcnt] = inp;
383 * Remove PCB from load balance group.
386 in_pcbremlbgrouphash(struct inpcb *inp)
388 struct inpcbinfo *pcbinfo;
389 struct inpcblbgrouphead *hdr;
390 struct inpcblbgroup *grp;
393 pcbinfo = inp->inp_pcbinfo;
395 INP_WLOCK_ASSERT(inp);
396 INP_HASH_WLOCK_ASSERT(pcbinfo);
398 hdr = &pcbinfo->ipi_lbgrouphashbase[
399 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
400 CK_LIST_FOREACH(grp, hdr, il_list) {
401 for (i = 0; i < grp->il_inpcnt; ++i) {
402 if (grp->il_inp[i] != inp)
405 if (grp->il_inpcnt == 1) {
406 /* We are the last, free this local group. */
407 in_pcblbgroup_free(grp);
409 /* Pull up inpcbs, shrink group if possible. */
410 in_pcblbgroup_reorder(hdr, &grp, i);
418 * Different protocols initialize their inpcbs differently - giving
419 * different name to the lock. But they all are disposed the same.
422 inpcb_fini(void *mem, int size)
424 struct inpcb *inp = mem;
426 INP_LOCK_DESTROY(inp);
430 * Initialize an inpcbinfo -- we should be able to reduce the number of
434 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
435 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
436 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
439 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
441 INP_INFO_LOCK_INIT(pcbinfo, name);
442 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
443 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
445 pcbinfo->ipi_vnet = curvnet;
447 pcbinfo->ipi_listhead = listhead;
448 CK_LIST_INIT(pcbinfo->ipi_listhead);
449 pcbinfo->ipi_count = 0;
450 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
451 &pcbinfo->ipi_hashmask);
452 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
453 &pcbinfo->ipi_porthashmask);
454 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
455 &pcbinfo->ipi_lbgrouphashmask);
457 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
459 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
460 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
461 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
462 uma_zone_set_warning(pcbinfo->ipi_zone,
463 "kern.ipc.maxsockets limit reached");
467 * Destroy an inpcbinfo.
470 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
473 KASSERT(pcbinfo->ipi_count == 0,
474 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
476 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
477 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
478 pcbinfo->ipi_porthashmask);
479 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
480 pcbinfo->ipi_lbgrouphashmask);
482 in_pcbgroup_destroy(pcbinfo);
484 uma_zdestroy(pcbinfo->ipi_zone);
485 INP_LIST_LOCK_DESTROY(pcbinfo);
486 INP_HASH_LOCK_DESTROY(pcbinfo);
487 INP_INFO_LOCK_DESTROY(pcbinfo);
491 * Allocate a PCB and associate it with the socket.
492 * On success return with the PCB locked.
495 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
501 if (pcbinfo == &V_tcbinfo) {
502 INP_INFO_RLOCK_ASSERT(pcbinfo);
504 INP_INFO_WLOCK_ASSERT(pcbinfo);
509 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
512 bzero(&inp->inp_start_zero, inp_zero_size);
514 inp->inp_numa_domain = M_NODOM;
516 inp->inp_pcbinfo = pcbinfo;
517 inp->inp_socket = so;
518 inp->inp_cred = crhold(so->so_cred);
519 inp->inp_inc.inc_fibnum = so->so_fibnum;
521 error = mac_inpcb_init(inp, M_NOWAIT);
524 mac_inpcb_create(so, inp);
526 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
527 error = ipsec_init_pcbpolicy(inp);
530 mac_inpcb_destroy(inp);
536 if (INP_SOCKAF(so) == AF_INET6) {
537 inp->inp_vflag |= INP_IPV6PROTO;
539 inp->inp_flags |= IN6P_IPV6_V6ONLY;
543 INP_LIST_WLOCK(pcbinfo);
544 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
545 pcbinfo->ipi_count++;
546 so->so_pcb = (caddr_t)inp;
548 if (V_ip6_auto_flowlabel)
549 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
551 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
552 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
555 * Routes in inpcb's can cache L2 as well; they are guaranteed
558 inp->inp_route.ro_flags = RT_LLE_CACHE;
559 INP_LIST_WUNLOCK(pcbinfo);
560 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
563 crfree(inp->inp_cred);
564 uma_zfree(pcbinfo->ipi_zone, inp);
572 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
576 INP_WLOCK_ASSERT(inp);
577 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
579 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
581 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
582 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
583 &inp->inp_lport, cred);
586 if (in_pcbinshash(inp) != 0) {
587 inp->inp_laddr.s_addr = INADDR_ANY;
592 inp->inp_flags |= INP_ANONPORT;
598 * Select a local port (number) to use.
600 #if defined(INET) || defined(INET6)
602 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
603 struct ucred *cred, int lookupflags)
605 struct inpcbinfo *pcbinfo;
606 struct inpcb *tmpinp;
607 unsigned short *lastport;
608 int count, dorandom, error;
609 u_short aux, first, last, lport;
611 struct in_addr laddr;
614 pcbinfo = inp->inp_pcbinfo;
617 * Because no actual state changes occur here, a global write lock on
618 * the pcbinfo isn't required.
620 INP_LOCK_ASSERT(inp);
621 INP_HASH_LOCK_ASSERT(pcbinfo);
623 if (inp->inp_flags & INP_HIGHPORT) {
624 first = V_ipport_hifirstauto; /* sysctl */
625 last = V_ipport_hilastauto;
626 lastport = &pcbinfo->ipi_lasthi;
627 } else if (inp->inp_flags & INP_LOWPORT) {
628 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
631 first = V_ipport_lowfirstauto; /* 1023 */
632 last = V_ipport_lowlastauto; /* 600 */
633 lastport = &pcbinfo->ipi_lastlow;
635 first = V_ipport_firstauto; /* sysctl */
636 last = V_ipport_lastauto;
637 lastport = &pcbinfo->ipi_lastport;
640 * For UDP(-Lite), use random port allocation as long as the user
641 * allows it. For TCP (and as of yet unknown) connections,
642 * use random port allocation only if the user allows it AND
643 * ipport_tick() allows it.
645 if (V_ipport_randomized &&
646 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
647 pcbinfo == &V_ulitecbinfo))
652 * It makes no sense to do random port allocation if
653 * we have the only port available.
657 /* Make sure to not include UDP(-Lite) packets in the count. */
658 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
659 V_ipport_tcpallocs++;
661 * Instead of having two loops further down counting up or down
662 * make sure that first is always <= last and go with only one
663 * code path implementing all logic.
672 /* Make the compiler happy. */
674 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
675 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
680 tmpinp = NULL; /* Make compiler happy. */
684 *lastport = first + (arc4random() % (last - first));
686 count = last - first;
689 if (count-- < 0) /* completely used? */
690 return (EADDRNOTAVAIL);
692 if (*lastport < first || *lastport > last)
694 lport = htons(*lastport);
697 if ((inp->inp_vflag & INP_IPV6) != 0)
698 tmpinp = in6_pcblookup_local(pcbinfo,
699 &inp->in6p_laddr, lport, lookupflags, cred);
701 #if defined(INET) && defined(INET6)
705 tmpinp = in_pcblookup_local(pcbinfo, laddr,
706 lport, lookupflags, cred);
708 } while (tmpinp != NULL);
711 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
712 laddrp->s_addr = laddr.s_addr;
720 * Return cached socket options.
723 inp_so_options(const struct inpcb *inp)
729 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
730 so_options |= SO_REUSEPORT_LB;
731 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
732 so_options |= SO_REUSEPORT;
733 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
734 so_options |= SO_REUSEADDR;
737 #endif /* INET || INET6 */
740 * Check if a new BINDMULTI socket is allowed to be created.
742 * ni points to the new inp.
743 * oi points to the exisitng inp.
745 * This checks whether the existing inp also has BINDMULTI and
746 * whether the credentials match.
749 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
751 /* Check permissions match */
752 if ((ni->inp_flags2 & INP_BINDMULTI) &&
753 (ni->inp_cred->cr_uid !=
754 oi->inp_cred->cr_uid))
757 /* Check the existing inp has BINDMULTI set */
758 if ((ni->inp_flags2 & INP_BINDMULTI) &&
759 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
763 * We're okay - either INP_BINDMULTI isn't set on ni, or
764 * it is and it matches the checks.
771 * Set up a bind operation on a PCB, performing port allocation
772 * as required, but do not actually modify the PCB. Callers can
773 * either complete the bind by setting inp_laddr/inp_lport and
774 * calling in_pcbinshash(), or they can just use the resulting
775 * port and address to authorise the sending of a once-off packet.
777 * On error, the values of *laddrp and *lportp are not changed.
780 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
781 u_short *lportp, struct ucred *cred)
783 struct socket *so = inp->inp_socket;
784 struct sockaddr_in *sin;
785 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
786 struct in_addr laddr;
788 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
792 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
793 * so that we don't have to add to the (already messy) code below.
795 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
798 * No state changes, so read locks are sufficient here.
800 INP_LOCK_ASSERT(inp);
801 INP_HASH_LOCK_ASSERT(pcbinfo);
803 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
804 return (EADDRNOTAVAIL);
805 laddr.s_addr = *laddrp;
806 if (nam != NULL && laddr.s_addr != INADDR_ANY)
808 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
809 lookupflags = INPLOOKUP_WILDCARD;
811 if ((error = prison_local_ip4(cred, &laddr)) != 0)
814 sin = (struct sockaddr_in *)nam;
815 if (nam->sa_len != sizeof (*sin))
819 * We should check the family, but old programs
820 * incorrectly fail to initialize it.
822 if (sin->sin_family != AF_INET)
823 return (EAFNOSUPPORT);
825 error = prison_local_ip4(cred, &sin->sin_addr);
828 if (sin->sin_port != *lportp) {
829 /* Don't allow the port to change. */
832 lport = sin->sin_port;
834 /* NB: lport is left as 0 if the port isn't being changed. */
835 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
837 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
838 * allow complete duplication of binding if
839 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
840 * and a multicast address is bound on both
841 * new and duplicated sockets.
843 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
844 reuseport = SO_REUSEADDR|SO_REUSEPORT;
846 * XXX: How to deal with SO_REUSEPORT_LB here?
847 * Treat same as SO_REUSEPORT for now.
849 if ((so->so_options &
850 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
851 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
852 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
853 sin->sin_port = 0; /* yech... */
854 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
856 * Is the address a local IP address?
857 * If INP_BINDANY is set, then the socket may be bound
858 * to any endpoint address, local or not.
860 if ((inp->inp_flags & INP_BINDANY) == 0 &&
861 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
862 return (EADDRNOTAVAIL);
864 laddr = sin->sin_addr;
870 if (ntohs(lport) <= V_ipport_reservedhigh &&
871 ntohs(lport) >= V_ipport_reservedlow &&
872 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
874 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
875 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
876 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
877 lport, INPLOOKUP_WILDCARD, cred);
880 * This entire block sorely needs a rewrite.
883 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
884 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
885 (so->so_type != SOCK_STREAM ||
886 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
887 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
888 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
889 (t->inp_flags2 & INP_REUSEPORT) ||
890 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
891 (inp->inp_cred->cr_uid !=
892 t->inp_cred->cr_uid))
896 * If the socket is a BINDMULTI socket, then
897 * the credentials need to match and the
898 * original socket also has to have been bound
901 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
904 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
905 lport, lookupflags, cred);
906 if (t && (t->inp_flags & INP_TIMEWAIT)) {
908 * XXXRW: If an incpb has had its timewait
909 * state recycled, we treat the address as
910 * being in use (for now). This is better
911 * than a panic, but not desirable.
915 ((reuseport & tw->tw_so_options) == 0 &&
917 tw->tw_so_options) == 0)) {
921 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
922 (reuseport & inp_so_options(t)) == 0 &&
923 (reuseport_lb & inp_so_options(t)) == 0) {
925 if (ntohl(sin->sin_addr.s_addr) !=
927 ntohl(t->inp_laddr.s_addr) !=
929 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
930 (t->inp_vflag & INP_IPV6PROTO) == 0)
933 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
941 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
946 *laddrp = laddr.s_addr;
952 * Connect from a socket to a specified address.
953 * Both address and port must be specified in argument sin.
954 * If don't have a local address for this socket yet,
958 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
959 struct ucred *cred, struct mbuf *m)
961 u_short lport, fport;
962 in_addr_t laddr, faddr;
965 INP_WLOCK_ASSERT(inp);
966 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
968 lport = inp->inp_lport;
969 laddr = inp->inp_laddr.s_addr;
970 anonport = (lport == 0);
971 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
976 /* Do the initial binding of the local address if required. */
977 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
978 inp->inp_lport = lport;
979 inp->inp_laddr.s_addr = laddr;
980 if (in_pcbinshash(inp) != 0) {
981 inp->inp_laddr.s_addr = INADDR_ANY;
987 /* Commit the remaining changes. */
988 inp->inp_lport = lport;
989 inp->inp_laddr.s_addr = laddr;
990 inp->inp_faddr.s_addr = faddr;
991 inp->inp_fport = fport;
992 in_pcbrehash_mbuf(inp, m);
995 inp->inp_flags |= INP_ANONPORT;
1000 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1003 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1007 * Do proper source address selection on an unbound socket in case
1008 * of connect. Take jails into account as well.
1011 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1015 struct sockaddr *sa;
1016 struct sockaddr_in *sin;
1018 struct epoch_tracker et;
1021 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1023 * Bypass source address selection and use the primary jail IP
1026 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1030 bzero(&sro, sizeof(sro));
1032 sin = (struct sockaddr_in *)&sro.ro_dst;
1033 sin->sin_family = AF_INET;
1034 sin->sin_len = sizeof(struct sockaddr_in);
1035 sin->sin_addr.s_addr = faddr->s_addr;
1038 * If route is known our src addr is taken from the i/f,
1041 * Find out route to destination.
1043 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1044 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1047 * If we found a route, use the address corresponding to
1048 * the outgoing interface.
1050 * Otherwise assume faddr is reachable on a directly connected
1051 * network and try to find a corresponding interface to take
1052 * the source address from.
1054 NET_EPOCH_ENTER(et);
1055 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1056 struct in_ifaddr *ia;
1059 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1060 inp->inp_socket->so_fibnum));
1062 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1063 inp->inp_socket->so_fibnum));
1067 error = ENETUNREACH;
1071 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1072 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1078 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1081 if (sa->sa_family != AF_INET)
1083 sin = (struct sockaddr_in *)sa;
1084 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1085 ia = (struct in_ifaddr *)ifa;
1090 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1094 /* 3. As a last resort return the 'default' jail address. */
1095 error = prison_get_ip4(cred, laddr);
1100 * If the outgoing interface on the route found is not
1101 * a loopback interface, use the address from that interface.
1102 * In case of jails do those three steps:
1103 * 1. check if the interface address belongs to the jail. If so use it.
1104 * 2. check if we have any address on the outgoing interface
1105 * belonging to this jail. If so use it.
1106 * 3. as a last resort return the 'default' jail address.
1108 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1109 struct in_ifaddr *ia;
1112 /* If not jailed, use the default returned. */
1113 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1114 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1115 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1120 /* 1. Check if the iface address belongs to the jail. */
1121 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1122 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1123 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1124 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1129 * 2. Check if we have any address on the outgoing interface
1130 * belonging to this jail.
1133 ifp = sro.ro_rt->rt_ifp;
1134 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1136 if (sa->sa_family != AF_INET)
1138 sin = (struct sockaddr_in *)sa;
1139 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1140 ia = (struct in_ifaddr *)ifa;
1145 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1149 /* 3. As a last resort return the 'default' jail address. */
1150 error = prison_get_ip4(cred, laddr);
1155 * The outgoing interface is marked with 'loopback net', so a route
1156 * to ourselves is here.
1157 * Try to find the interface of the destination address and then
1158 * take the address from there. That interface is not necessarily
1159 * a loopback interface.
1160 * In case of jails, check that it is an address of the jail
1161 * and if we cannot find, fall back to the 'default' jail address.
1163 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1164 struct sockaddr_in sain;
1165 struct in_ifaddr *ia;
1167 bzero(&sain, sizeof(struct sockaddr_in));
1168 sain.sin_family = AF_INET;
1169 sain.sin_len = sizeof(struct sockaddr_in);
1170 sain.sin_addr.s_addr = faddr->s_addr;
1172 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1173 inp->inp_socket->so_fibnum));
1175 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1176 inp->inp_socket->so_fibnum));
1178 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1180 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1182 error = ENETUNREACH;
1185 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1195 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1197 if (sa->sa_family != AF_INET)
1199 sin = (struct sockaddr_in *)sa;
1200 if (prison_check_ip4(cred,
1201 &sin->sin_addr) == 0) {
1202 ia = (struct in_ifaddr *)ifa;
1207 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1212 /* 3. As a last resort return the 'default' jail address. */
1213 error = prison_get_ip4(cred, laddr);
1219 if (sro.ro_rt != NULL)
1225 * Set up for a connect from a socket to the specified address.
1226 * On entry, *laddrp and *lportp should contain the current local
1227 * address and port for the PCB; these are updated to the values
1228 * that should be placed in inp_laddr and inp_lport to complete
1231 * On success, *faddrp and *fportp will be set to the remote address
1232 * and port. These are not updated in the error case.
1234 * If the operation fails because the connection already exists,
1235 * *oinpp will be set to the PCB of that connection so that the
1236 * caller can decide to override it. In all other cases, *oinpp
1240 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1241 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1242 struct inpcb **oinpp, struct ucred *cred)
1244 struct rm_priotracker in_ifa_tracker;
1245 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1246 struct in_ifaddr *ia;
1248 struct in_addr laddr, faddr;
1249 u_short lport, fport;
1253 * Because a global state change doesn't actually occur here, a read
1254 * lock is sufficient.
1256 INP_LOCK_ASSERT(inp);
1257 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1261 if (nam->sa_len != sizeof (*sin))
1263 if (sin->sin_family != AF_INET)
1264 return (EAFNOSUPPORT);
1265 if (sin->sin_port == 0)
1266 return (EADDRNOTAVAIL);
1267 laddr.s_addr = *laddrp;
1269 faddr = sin->sin_addr;
1270 fport = sin->sin_port;
1272 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1274 * If the destination address is INADDR_ANY,
1275 * use the primary local address.
1276 * If the supplied address is INADDR_BROADCAST,
1277 * and the primary interface supports broadcast,
1278 * choose the broadcast address for that interface.
1280 if (faddr.s_addr == INADDR_ANY) {
1281 IN_IFADDR_RLOCK(&in_ifa_tracker);
1283 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1284 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1286 (error = prison_get_ip4(cred, &faddr)) != 0)
1288 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1289 IN_IFADDR_RLOCK(&in_ifa_tracker);
1290 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1292 faddr = satosin(&CK_STAILQ_FIRST(
1293 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1294 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1297 if (laddr.s_addr == INADDR_ANY) {
1298 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1300 * If the destination address is multicast and an outgoing
1301 * interface has been set as a multicast option, prefer the
1302 * address of that interface as our source address.
1304 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1305 inp->inp_moptions != NULL) {
1306 struct ip_moptions *imo;
1309 imo = inp->inp_moptions;
1310 if (imo->imo_multicast_ifp != NULL) {
1311 ifp = imo->imo_multicast_ifp;
1312 IN_IFADDR_RLOCK(&in_ifa_tracker);
1313 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1314 if ((ia->ia_ifp == ifp) &&
1316 prison_check_ip4(cred,
1317 &ia->ia_addr.sin_addr) == 0))
1321 error = EADDRNOTAVAIL;
1323 laddr = ia->ia_addr.sin_addr;
1326 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1332 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1333 laddr, lport, 0, NULL);
1337 return (EADDRINUSE);
1340 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1345 *laddrp = laddr.s_addr;
1347 *faddrp = faddr.s_addr;
1353 in_pcbdisconnect(struct inpcb *inp)
1356 INP_WLOCK_ASSERT(inp);
1357 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1359 inp->inp_faddr.s_addr = INADDR_ANY;
1366 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1367 * For most protocols, this will be invoked immediately prior to calling
1368 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1369 * socket, in which case in_pcbfree() is deferred.
1372 in_pcbdetach(struct inpcb *inp)
1375 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1378 if (inp->inp_snd_tag != NULL)
1379 in_pcbdetach_txrtlmt(inp);
1381 inp->inp_socket->so_pcb = NULL;
1382 inp->inp_socket = NULL;
1386 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1387 * stability of an inpcb pointer despite the inpcb lock being released. This
1388 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1389 * but where the inpcb lock may already held, or when acquiring a reference
1392 * in_pcbref() should be used only to provide brief memory stability, and
1393 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1394 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1395 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1396 * lock and rele are the *only* safe operations that may be performed on the
1399 * While the inpcb will not be freed, releasing the inpcb lock means that the
1400 * connection's state may change, so the caller should be careful to
1401 * revalidate any cached state on reacquiring the lock. Drop the reference
1402 * using in_pcbrele().
1405 in_pcbref(struct inpcb *inp)
1408 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1410 refcount_acquire(&inp->inp_refcount);
1414 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1415 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1416 * return a flag indicating whether or not the inpcb remains valid. If it is
1417 * valid, we return with the inpcb lock held.
1419 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1420 * reference on an inpcb. Historically more work was done here (actually, in
1421 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1422 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1423 * about memory stability (and continued use of the write lock).
1426 in_pcbrele_rlocked(struct inpcb *inp)
1428 struct inpcbinfo *pcbinfo;
1430 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1432 INP_RLOCK_ASSERT(inp);
1434 if (refcount_release(&inp->inp_refcount) == 0) {
1436 * If the inpcb has been freed, let the caller know, even if
1437 * this isn't the last reference.
1439 if (inp->inp_flags2 & INP_FREED) {
1446 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1448 if (inp->inp_in_hpts || inp->inp_in_input) {
1449 struct tcp_hpts_entry *hpts;
1451 * We should not be on the hpts at
1452 * this point in any form. we must
1453 * get the lock to be sure.
1455 hpts = tcp_hpts_lock(inp);
1456 if (inp->inp_in_hpts)
1457 panic("Hpts:%p inp:%p at free still on hpts",
1459 mtx_unlock(&hpts->p_mtx);
1460 hpts = tcp_input_lock(inp);
1461 if (inp->inp_in_input)
1462 panic("Hpts:%p inp:%p at free still on input hpts",
1464 mtx_unlock(&hpts->p_mtx);
1468 pcbinfo = inp->inp_pcbinfo;
1469 uma_zfree(pcbinfo->ipi_zone, inp);
1474 in_pcbrele_wlocked(struct inpcb *inp)
1476 struct inpcbinfo *pcbinfo;
1478 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1480 INP_WLOCK_ASSERT(inp);
1482 if (refcount_release(&inp->inp_refcount) == 0) {
1484 * If the inpcb has been freed, let the caller know, even if
1485 * this isn't the last reference.
1487 if (inp->inp_flags2 & INP_FREED) {
1494 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1496 if (inp->inp_in_hpts || inp->inp_in_input) {
1497 struct tcp_hpts_entry *hpts;
1499 * We should not be on the hpts at
1500 * this point in any form. we must
1501 * get the lock to be sure.
1503 hpts = tcp_hpts_lock(inp);
1504 if (inp->inp_in_hpts)
1505 panic("Hpts:%p inp:%p at free still on hpts",
1507 mtx_unlock(&hpts->p_mtx);
1508 hpts = tcp_input_lock(inp);
1509 if (inp->inp_in_input)
1510 panic("Hpts:%p inp:%p at free still on input hpts",
1512 mtx_unlock(&hpts->p_mtx);
1516 pcbinfo = inp->inp_pcbinfo;
1517 uma_zfree(pcbinfo->ipi_zone, inp);
1522 * Temporary wrapper.
1525 in_pcbrele(struct inpcb *inp)
1528 return (in_pcbrele_wlocked(inp));
1532 in_pcblist_rele_rlocked(epoch_context_t ctx)
1534 struct in_pcblist *il;
1536 struct inpcbinfo *pcbinfo;
1539 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1540 pcbinfo = il->il_pcbinfo;
1542 INP_INFO_WLOCK(pcbinfo);
1543 for (i = 0; i < n; i++) {
1544 inp = il->il_inp_list[i];
1546 if (!in_pcbrele_rlocked(inp))
1549 INP_INFO_WUNLOCK(pcbinfo);
1554 inpcbport_free(epoch_context_t ctx)
1556 struct inpcbport *phd;
1558 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1563 in_pcbfree_deferred(epoch_context_t ctx)
1566 int released __unused;
1568 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1571 CURVNET_SET(inp->inp_vnet);
1573 struct ip_moptions *imo = inp->inp_moptions;
1574 inp->inp_moptions = NULL;
1576 /* XXXRW: Do as much as possible here. */
1577 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1578 if (inp->inp_sp != NULL)
1579 ipsec_delete_pcbpolicy(inp);
1582 struct ip6_moptions *im6o = NULL;
1583 if (inp->inp_vflag & INP_IPV6PROTO) {
1584 ip6_freepcbopts(inp->in6p_outputopts);
1585 im6o = inp->in6p_moptions;
1586 inp->in6p_moptions = NULL;
1589 if (inp->inp_options)
1590 (void)m_free(inp->inp_options);
1592 crfree(inp->inp_cred);
1594 mac_inpcb_destroy(inp);
1596 released = in_pcbrele_wlocked(inp);
1599 ip6_freemoptions(im6o);
1602 inp_freemoptions(imo);
1608 * Unconditionally schedule an inpcb to be freed by decrementing its
1609 * reference count, which should occur only after the inpcb has been detached
1610 * from its socket. If another thread holds a temporary reference (acquired
1611 * using in_pcbref()) then the free is deferred until that reference is
1612 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1613 * work, including removal from global lists, is done in this context, where
1614 * the pcbinfo lock is held.
1617 in_pcbfree(struct inpcb *inp)
1619 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1621 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1622 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1623 ("%s: called twice for pcb %p", __func__, inp));
1624 if (inp->inp_flags2 & INP_FREED) {
1630 if (pcbinfo == &V_tcbinfo) {
1631 INP_INFO_LOCK_ASSERT(pcbinfo);
1633 INP_INFO_WLOCK_ASSERT(pcbinfo);
1636 INP_WLOCK_ASSERT(inp);
1637 INP_LIST_WLOCK(pcbinfo);
1638 in_pcbremlists(inp);
1639 INP_LIST_WUNLOCK(pcbinfo);
1640 RO_INVALIDATE_CACHE(&inp->inp_route);
1641 /* mark as destruction in progress */
1642 inp->inp_flags2 |= INP_FREED;
1644 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1648 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1649 * port reservation, and preventing it from being returned by inpcb lookups.
1651 * It is used by TCP to mark an inpcb as unused and avoid future packet
1652 * delivery or event notification when a socket remains open but TCP has
1653 * closed. This might occur as a result of a shutdown()-initiated TCP close
1654 * or a RST on the wire, and allows the port binding to be reused while still
1655 * maintaining the invariant that so_pcb always points to a valid inpcb until
1658 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1659 * in_pcbnotifyall() and in_pcbpurgeif0()?
1662 in_pcbdrop(struct inpcb *inp)
1665 INP_WLOCK_ASSERT(inp);
1667 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1668 MPASS(inp->inp_refcount > 1);
1672 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1675 inp->inp_flags |= INP_DROPPED;
1676 if (inp->inp_flags & INP_INHASHLIST) {
1677 struct inpcbport *phd = inp->inp_phd;
1679 INP_HASH_WLOCK(inp->inp_pcbinfo);
1680 in_pcbremlbgrouphash(inp);
1681 CK_LIST_REMOVE(inp, inp_hash);
1682 CK_LIST_REMOVE(inp, inp_portlist);
1683 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1684 CK_LIST_REMOVE(phd, phd_hash);
1685 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1687 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1688 inp->inp_flags &= ~INP_INHASHLIST;
1690 in_pcbgroup_remove(inp);
1697 * Common routines to return the socket addresses associated with inpcbs.
1700 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1702 struct sockaddr_in *sin;
1704 sin = malloc(sizeof *sin, M_SONAME,
1706 sin->sin_family = AF_INET;
1707 sin->sin_len = sizeof(*sin);
1708 sin->sin_addr = *addr_p;
1709 sin->sin_port = port;
1711 return (struct sockaddr *)sin;
1715 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1718 struct in_addr addr;
1721 inp = sotoinpcb(so);
1722 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1725 port = inp->inp_lport;
1726 addr = inp->inp_laddr;
1729 *nam = in_sockaddr(port, &addr);
1734 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1737 struct in_addr addr;
1740 inp = sotoinpcb(so);
1741 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1744 port = inp->inp_fport;
1745 addr = inp->inp_faddr;
1748 *nam = in_sockaddr(port, &addr);
1753 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1754 struct inpcb *(*notify)(struct inpcb *, int))
1756 struct inpcb *inp, *inp_temp;
1758 INP_INFO_WLOCK(pcbinfo);
1759 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1762 if ((inp->inp_vflag & INP_IPV4) == 0) {
1767 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1768 inp->inp_socket == NULL) {
1772 if ((*notify)(inp, errno))
1775 INP_INFO_WUNLOCK(pcbinfo);
1779 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1782 struct ip_moptions *imo;
1785 INP_INFO_WLOCK(pcbinfo);
1786 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1788 imo = inp->inp_moptions;
1789 if ((inp->inp_vflag & INP_IPV4) &&
1792 * Unselect the outgoing interface if it is being
1795 if (imo->imo_multicast_ifp == ifp)
1796 imo->imo_multicast_ifp = NULL;
1799 * Drop multicast group membership if we joined
1800 * through the interface being detached.
1802 * XXX This can all be deferred to an epoch_call
1804 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1806 if (imo->imo_membership[i]->inm_ifp == ifp) {
1807 IN_MULTI_LOCK_ASSERT();
1808 in_leavegroup_locked(imo->imo_membership[i], NULL);
1810 } else if (gap != 0)
1811 imo->imo_membership[i - gap] =
1812 imo->imo_membership[i];
1814 imo->imo_num_memberships -= gap;
1818 INP_INFO_WUNLOCK(pcbinfo);
1822 * Lookup a PCB based on the local address and port. Caller must hold the
1823 * hash lock. No inpcb locks or references are acquired.
1825 #define INP_LOOKUP_MAPPED_PCB_COST 3
1827 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1828 u_short lport, int lookupflags, struct ucred *cred)
1832 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1838 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1839 ("%s: invalid lookup flags %d", __func__, lookupflags));
1841 INP_HASH_LOCK_ASSERT(pcbinfo);
1843 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1844 struct inpcbhead *head;
1846 * Look for an unconnected (wildcard foreign addr) PCB that
1847 * matches the local address and port we're looking for.
1849 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1850 0, pcbinfo->ipi_hashmask)];
1851 CK_LIST_FOREACH(inp, head, inp_hash) {
1853 /* XXX inp locking */
1854 if ((inp->inp_vflag & INP_IPV4) == 0)
1857 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1858 inp->inp_laddr.s_addr == laddr.s_addr &&
1859 inp->inp_lport == lport) {
1864 prison_equal_ip4(cred->cr_prison,
1865 inp->inp_cred->cr_prison))
1874 struct inpcbporthead *porthash;
1875 struct inpcbport *phd;
1876 struct inpcb *match = NULL;
1878 * Best fit PCB lookup.
1880 * First see if this local port is in use by looking on the
1883 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1884 pcbinfo->ipi_porthashmask)];
1885 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1886 if (phd->phd_port == lport)
1891 * Port is in use by one or more PCBs. Look for best
1894 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1897 !prison_equal_ip4(inp->inp_cred->cr_prison,
1901 /* XXX inp locking */
1902 if ((inp->inp_vflag & INP_IPV4) == 0)
1905 * We never select the PCB that has
1906 * INP_IPV6 flag and is bound to :: if
1907 * we have another PCB which is bound
1908 * to 0.0.0.0. If a PCB has the
1909 * INP_IPV6 flag, then we set its cost
1910 * higher than IPv4 only PCBs.
1912 * Note that the case only happens
1913 * when a socket is bound to ::, under
1914 * the condition that the use of the
1915 * mapped address is allowed.
1917 if ((inp->inp_vflag & INP_IPV6) != 0)
1918 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1920 if (inp->inp_faddr.s_addr != INADDR_ANY)
1922 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1923 if (laddr.s_addr == INADDR_ANY)
1925 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1928 if (laddr.s_addr != INADDR_ANY)
1931 if (wildcard < matchwild) {
1933 matchwild = wildcard;
1942 #undef INP_LOOKUP_MAPPED_PCB_COST
1944 static struct inpcb *
1945 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1946 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1947 uint16_t fport, int lookupflags)
1949 struct inpcb *local_wild;
1950 const struct inpcblbgrouphead *hdr;
1951 struct inpcblbgroup *grp;
1954 INP_HASH_LOCK_ASSERT(pcbinfo);
1956 hdr = &pcbinfo->ipi_lbgrouphashbase[
1957 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1960 * Order of socket selection:
1962 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1965 * - Load balanced group does not contain jailed sockets
1966 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1969 CK_LIST_FOREACH(grp, hdr, il_list) {
1971 if (!(grp->il_vflag & INP_IPV4))
1974 if (grp->il_lport != lport)
1977 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1979 if (grp->il_laddr.s_addr == laddr->s_addr)
1980 return (grp->il_inp[idx]);
1981 if (grp->il_laddr.s_addr == INADDR_ANY &&
1982 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1983 local_wild = grp->il_inp[idx];
1985 return (local_wild);
1990 * Lookup PCB in hash list, using pcbgroup tables.
1992 static struct inpcb *
1993 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1994 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1995 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1997 struct inpcbhead *head;
1998 struct inpcb *inp, *tmpinp;
1999 u_short fport = fport_arg, lport = lport_arg;
2003 * First look for an exact match.
2006 INP_GROUP_LOCK(pcbgroup);
2007 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2008 pcbgroup->ipg_hashmask)];
2009 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2011 /* XXX inp locking */
2012 if ((inp->inp_vflag & INP_IPV4) == 0)
2015 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2016 inp->inp_laddr.s_addr == laddr.s_addr &&
2017 inp->inp_fport == fport &&
2018 inp->inp_lport == lport) {
2020 * XXX We should be able to directly return
2021 * the inp here, without any checks.
2022 * Well unless both bound with SO_REUSEPORT?
2024 if (prison_flag(inp->inp_cred, PR_IP4))
2030 if (tmpinp != NULL) {
2037 * For incoming connections, we may wish to do a wildcard
2038 * match for an RSS-local socket.
2040 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2041 struct inpcb *local_wild = NULL, *local_exact = NULL;
2043 struct inpcb *local_wild_mapped = NULL;
2045 struct inpcb *jail_wild = NULL;
2046 struct inpcbhead *head;
2050 * Order of socket selection - we always prefer jails.
2051 * 1. jailed, non-wild.
2053 * 3. non-jailed, non-wild.
2054 * 4. non-jailed, wild.
2057 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2058 lport, 0, pcbgroup->ipg_hashmask)];
2059 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2061 /* XXX inp locking */
2062 if ((inp->inp_vflag & INP_IPV4) == 0)
2065 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2066 inp->inp_lport != lport)
2069 injail = prison_flag(inp->inp_cred, PR_IP4);
2071 if (prison_check_ip4(inp->inp_cred,
2075 if (local_exact != NULL)
2079 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2084 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2086 /* XXX inp locking, NULL check */
2087 if (inp->inp_vflag & INP_IPV6PROTO)
2088 local_wild_mapped = inp;
2096 } /* LIST_FOREACH */
2105 inp = local_wild_mapped;
2113 * Then look for a wildcard match, if requested.
2115 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2116 struct inpcb *local_wild = NULL, *local_exact = NULL;
2118 struct inpcb *local_wild_mapped = NULL;
2120 struct inpcb *jail_wild = NULL;
2121 struct inpcbhead *head;
2125 * Order of socket selection - we always prefer jails.
2126 * 1. jailed, non-wild.
2128 * 3. non-jailed, non-wild.
2129 * 4. non-jailed, wild.
2131 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2132 0, pcbinfo->ipi_wildmask)];
2133 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2135 /* XXX inp locking */
2136 if ((inp->inp_vflag & INP_IPV4) == 0)
2139 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2140 inp->inp_lport != lport)
2143 injail = prison_flag(inp->inp_cred, PR_IP4);
2145 if (prison_check_ip4(inp->inp_cred,
2149 if (local_exact != NULL)
2153 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2158 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2160 /* XXX inp locking, NULL check */
2161 if (inp->inp_vflag & INP_IPV6PROTO)
2162 local_wild_mapped = inp;
2170 } /* LIST_FOREACH */
2178 inp = local_wild_mapped;
2182 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2183 INP_GROUP_UNLOCK(pcbgroup);
2187 if (lookupflags & INPLOOKUP_WLOCKPCB)
2188 locked = INP_TRY_WLOCK(inp);
2189 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2190 locked = INP_TRY_RLOCK(inp);
2192 panic("%s: locking bug", __func__);
2193 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2194 if (lookupflags & INPLOOKUP_WLOCKPCB)
2201 INP_GROUP_UNLOCK(pcbgroup);
2203 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2205 if (in_pcbrele_wlocked(inp))
2209 if (in_pcbrele_rlocked(inp))
2214 if (lookupflags & INPLOOKUP_WLOCKPCB)
2215 INP_WLOCK_ASSERT(inp);
2217 INP_RLOCK_ASSERT(inp);
2221 #endif /* PCBGROUP */
2224 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2225 * that the caller has locked the hash list, and will not perform any further
2226 * locking or reference operations on either the hash list or the connection.
2228 static struct inpcb *
2229 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2230 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2233 struct inpcbhead *head;
2234 struct inpcb *inp, *tmpinp;
2235 u_short fport = fport_arg, lport = lport_arg;
2238 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2239 ("%s: invalid lookup flags %d", __func__, lookupflags));
2240 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2241 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2244 * First look for an exact match.
2247 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2248 pcbinfo->ipi_hashmask)];
2249 CK_LIST_FOREACH(inp, head, inp_hash) {
2251 /* XXX inp locking */
2252 if ((inp->inp_vflag & INP_IPV4) == 0)
2255 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2256 inp->inp_laddr.s_addr == laddr.s_addr &&
2257 inp->inp_fport == fport &&
2258 inp->inp_lport == lport) {
2260 * XXX We should be able to directly return
2261 * the inp here, without any checks.
2262 * Well unless both bound with SO_REUSEPORT?
2264 if (prison_flag(inp->inp_cred, PR_IP4))
2274 * Then look in lb group (for wildcard match).
2276 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2277 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2278 fport, lookupflags);
2284 * Then look for a wildcard match, if requested.
2286 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2287 struct inpcb *local_wild = NULL, *local_exact = NULL;
2289 struct inpcb *local_wild_mapped = NULL;
2291 struct inpcb *jail_wild = NULL;
2295 * Order of socket selection - we always prefer jails.
2296 * 1. jailed, non-wild.
2298 * 3. non-jailed, non-wild.
2299 * 4. non-jailed, wild.
2302 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2303 0, pcbinfo->ipi_hashmask)];
2304 CK_LIST_FOREACH(inp, head, inp_hash) {
2306 /* XXX inp locking */
2307 if ((inp->inp_vflag & INP_IPV4) == 0)
2310 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2311 inp->inp_lport != lport)
2314 injail = prison_flag(inp->inp_cred, PR_IP4);
2316 if (prison_check_ip4(inp->inp_cred,
2320 if (local_exact != NULL)
2324 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2329 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2331 /* XXX inp locking, NULL check */
2332 if (inp->inp_vflag & INP_IPV6PROTO)
2333 local_wild_mapped = inp;
2341 } /* LIST_FOREACH */
2342 if (jail_wild != NULL)
2344 if (local_exact != NULL)
2345 return (local_exact);
2346 if (local_wild != NULL)
2347 return (local_wild);
2349 if (local_wild_mapped != NULL)
2350 return (local_wild_mapped);
2352 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2358 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2359 * hash list lock, and will return the inpcb locked (i.e., requires
2360 * INPLOOKUP_LOCKPCB).
2362 static struct inpcb *
2363 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2364 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2369 INP_HASH_RLOCK(pcbinfo);
2370 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2371 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2373 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2375 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2379 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2381 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2386 panic("%s: locking bug", __func__);
2389 if (lookupflags & INPLOOKUP_WLOCKPCB)
2390 INP_WLOCK_ASSERT(inp);
2392 INP_RLOCK_ASSERT(inp);
2396 INP_HASH_RUNLOCK(pcbinfo);
2401 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2402 * from which a pre-calculated hash value may be extracted.
2404 * Possibly more of this logic should be in in_pcbgroup.c.
2407 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2408 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2410 #if defined(PCBGROUP) && !defined(RSS)
2411 struct inpcbgroup *pcbgroup;
2414 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2415 ("%s: invalid lookup flags %d", __func__, lookupflags));
2416 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2417 ("%s: LOCKPCB not set", __func__));
2420 * When not using RSS, use connection groups in preference to the
2421 * reservation table when looking up 4-tuples. When using RSS, just
2422 * use the reservation table, due to the cost of the Toeplitz hash
2425 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2426 * we could be doing RSS with a non-Toeplitz hash that is affordable
2429 #if defined(PCBGROUP) && !defined(RSS)
2430 if (in_pcbgroup_enabled(pcbinfo)) {
2431 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2433 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2434 laddr, lport, lookupflags, ifp));
2437 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2442 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2443 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2444 struct ifnet *ifp, struct mbuf *m)
2447 struct inpcbgroup *pcbgroup;
2450 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2451 ("%s: invalid lookup flags %d", __func__, lookupflags));
2452 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2453 ("%s: LOCKPCB not set", __func__));
2457 * If we can use a hardware-generated hash to look up the connection
2458 * group, use that connection group to find the inpcb. Otherwise
2459 * fall back on a software hash -- or the reservation table if we're
2462 * XXXRW: As above, that policy belongs in the pcbgroup code.
2464 if (in_pcbgroup_enabled(pcbinfo) &&
2465 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2466 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2467 m->m_pkthdr.flowid);
2468 if (pcbgroup != NULL)
2469 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2470 fport, laddr, lport, lookupflags, ifp));
2472 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2474 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2475 laddr, lport, lookupflags, ifp));
2479 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2485 * Insert PCB onto various hash lists.
2488 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2490 struct inpcbhead *pcbhash;
2491 struct inpcbporthead *pcbporthash;
2492 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2493 struct inpcbport *phd;
2494 u_int32_t hashkey_faddr;
2497 INP_WLOCK_ASSERT(inp);
2498 INP_HASH_WLOCK_ASSERT(pcbinfo);
2500 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2501 ("in_pcbinshash: INP_INHASHLIST"));
2504 if (inp->inp_vflag & INP_IPV6)
2505 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2508 hashkey_faddr = inp->inp_faddr.s_addr;
2510 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2511 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2513 pcbporthash = &pcbinfo->ipi_porthashbase[
2514 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2517 * Add entry to load balance group.
2518 * Only do this if SO_REUSEPORT_LB is set.
2520 so_options = inp_so_options(inp);
2521 if (so_options & SO_REUSEPORT_LB) {
2522 int ret = in_pcbinslbgrouphash(inp);
2524 /* pcb lb group malloc fail (ret=ENOBUFS). */
2530 * Go through port list and look for a head for this lport.
2532 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2533 if (phd->phd_port == inp->inp_lport)
2537 * If none exists, malloc one and tack it on.
2540 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2542 return (ENOBUFS); /* XXX */
2544 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2545 phd->phd_port = inp->inp_lport;
2546 CK_LIST_INIT(&phd->phd_pcblist);
2547 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2550 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2551 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2552 inp->inp_flags |= INP_INHASHLIST;
2554 if (do_pcbgroup_update)
2555 in_pcbgroup_update(inp);
2561 * For now, there are two public interfaces to insert an inpcb into the hash
2562 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2563 * is used only in the TCP syncache, where in_pcbinshash is called before the
2564 * full 4-tuple is set for the inpcb, and we don't want to install in the
2565 * pcbgroup until later.
2567 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2568 * connection groups, and partially initialised inpcbs should not be exposed
2569 * to either reservation hash tables or pcbgroups.
2572 in_pcbinshash(struct inpcb *inp)
2575 return (in_pcbinshash_internal(inp, 1));
2579 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2582 return (in_pcbinshash_internal(inp, 0));
2586 * Move PCB to the proper hash bucket when { faddr, fport } have been
2587 * changed. NOTE: This does not handle the case of the lport changing (the
2588 * hashed port list would have to be updated as well), so the lport must
2589 * not change after in_pcbinshash() has been called.
2592 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2594 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2595 struct inpcbhead *head;
2596 u_int32_t hashkey_faddr;
2598 INP_WLOCK_ASSERT(inp);
2599 INP_HASH_WLOCK_ASSERT(pcbinfo);
2601 KASSERT(inp->inp_flags & INP_INHASHLIST,
2602 ("in_pcbrehash: !INP_INHASHLIST"));
2605 if (inp->inp_vflag & INP_IPV6)
2606 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2609 hashkey_faddr = inp->inp_faddr.s_addr;
2611 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2612 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2614 CK_LIST_REMOVE(inp, inp_hash);
2615 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2619 in_pcbgroup_update_mbuf(inp, m);
2621 in_pcbgroup_update(inp);
2626 in_pcbrehash(struct inpcb *inp)
2629 in_pcbrehash_mbuf(inp, NULL);
2633 * Remove PCB from various lists.
2636 in_pcbremlists(struct inpcb *inp)
2638 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2641 if (pcbinfo == &V_tcbinfo) {
2642 INP_INFO_RLOCK_ASSERT(pcbinfo);
2644 INP_INFO_WLOCK_ASSERT(pcbinfo);
2648 INP_WLOCK_ASSERT(inp);
2649 INP_LIST_WLOCK_ASSERT(pcbinfo);
2651 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2652 if (inp->inp_flags & INP_INHASHLIST) {
2653 struct inpcbport *phd = inp->inp_phd;
2655 INP_HASH_WLOCK(pcbinfo);
2657 /* XXX: Only do if SO_REUSEPORT_LB set? */
2658 in_pcbremlbgrouphash(inp);
2660 CK_LIST_REMOVE(inp, inp_hash);
2661 CK_LIST_REMOVE(inp, inp_portlist);
2662 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2663 CK_LIST_REMOVE(phd, phd_hash);
2664 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2666 INP_HASH_WUNLOCK(pcbinfo);
2667 inp->inp_flags &= ~INP_INHASHLIST;
2669 CK_LIST_REMOVE(inp, inp_list);
2670 pcbinfo->ipi_count--;
2672 in_pcbgroup_remove(inp);
2677 * Check for alternatives when higher level complains
2678 * about service problems. For now, invalidate cached
2679 * routing information. If the route was created dynamically
2680 * (by a redirect), time to try a default gateway again.
2683 in_losing(struct inpcb *inp)
2686 RO_INVALIDATE_CACHE(&inp->inp_route);
2691 * A set label operation has occurred at the socket layer, propagate the
2692 * label change into the in_pcb for the socket.
2695 in_pcbsosetlabel(struct socket *so)
2700 inp = sotoinpcb(so);
2701 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2705 mac_inpcb_sosetlabel(so, inp);
2712 * ipport_tick runs once per second, determining if random port allocation
2713 * should be continued. If more than ipport_randomcps ports have been
2714 * allocated in the last second, then we return to sequential port
2715 * allocation. We return to random allocation only once we drop below
2716 * ipport_randomcps for at least ipport_randomtime seconds.
2719 ipport_tick(void *xtp)
2721 VNET_ITERATOR_DECL(vnet_iter);
2723 VNET_LIST_RLOCK_NOSLEEP();
2724 VNET_FOREACH(vnet_iter) {
2725 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2726 if (V_ipport_tcpallocs <=
2727 V_ipport_tcplastcount + V_ipport_randomcps) {
2728 if (V_ipport_stoprandom > 0)
2729 V_ipport_stoprandom--;
2731 V_ipport_stoprandom = V_ipport_randomtime;
2732 V_ipport_tcplastcount = V_ipport_tcpallocs;
2735 VNET_LIST_RUNLOCK_NOSLEEP();
2736 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2743 callout_stop(&ipport_tick_callout);
2747 * The ipport_callout should start running at about the time we attach the
2748 * inet or inet6 domains.
2751 ipport_tick_init(const void *unused __unused)
2754 /* Start ipport_tick. */
2755 callout_init(&ipport_tick_callout, 1);
2756 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2757 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2758 SHUTDOWN_PRI_DEFAULT);
2760 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2761 ipport_tick_init, NULL);
2764 inp_wlock(struct inpcb *inp)
2771 inp_wunlock(struct inpcb *inp)
2778 inp_rlock(struct inpcb *inp)
2785 inp_runlock(struct inpcb *inp)
2791 #ifdef INVARIANT_SUPPORT
2793 inp_lock_assert(struct inpcb *inp)
2796 INP_WLOCK_ASSERT(inp);
2800 inp_unlock_assert(struct inpcb *inp)
2803 INP_UNLOCK_ASSERT(inp);
2808 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2812 INP_INFO_WLOCK(&V_tcbinfo);
2813 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2818 INP_INFO_WUNLOCK(&V_tcbinfo);
2822 inp_inpcbtosocket(struct inpcb *inp)
2825 INP_WLOCK_ASSERT(inp);
2826 return (inp->inp_socket);
2830 inp_inpcbtotcpcb(struct inpcb *inp)
2833 INP_WLOCK_ASSERT(inp);
2834 return ((struct tcpcb *)inp->inp_ppcb);
2838 inp_ip_tos_get(const struct inpcb *inp)
2841 return (inp->inp_ip_tos);
2845 inp_ip_tos_set(struct inpcb *inp, int val)
2848 inp->inp_ip_tos = val;
2852 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2853 uint32_t *faddr, uint16_t *fp)
2856 INP_LOCK_ASSERT(inp);
2857 *laddr = inp->inp_laddr.s_addr;
2858 *faddr = inp->inp_faddr.s_addr;
2859 *lp = inp->inp_lport;
2860 *fp = inp->inp_fport;
2864 so_sotoinpcb(struct socket *so)
2867 return (sotoinpcb(so));
2871 so_sototcpcb(struct socket *so)
2874 return (sototcpcb(so));
2878 * Create an external-format (``xinpcb'') structure using the information in
2879 * the kernel-format in_pcb structure pointed to by inp. This is done to
2880 * reduce the spew of irrelevant information over this interface, to isolate
2881 * user code from changes in the kernel structure, and potentially to provide
2882 * information-hiding if we decide that some of this information should be
2883 * hidden from users.
2886 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2889 bzero(xi, sizeof(*xi));
2890 xi->xi_len = sizeof(struct xinpcb);
2891 if (inp->inp_socket)
2892 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2893 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2894 xi->inp_gencnt = inp->inp_gencnt;
2895 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2896 xi->inp_flow = inp->inp_flow;
2897 xi->inp_flowid = inp->inp_flowid;
2898 xi->inp_flowtype = inp->inp_flowtype;
2899 xi->inp_flags = inp->inp_flags;
2900 xi->inp_flags2 = inp->inp_flags2;
2901 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2902 xi->in6p_cksum = inp->in6p_cksum;
2903 xi->in6p_hops = inp->in6p_hops;
2904 xi->inp_ip_tos = inp->inp_ip_tos;
2905 xi->inp_vflag = inp->inp_vflag;
2906 xi->inp_ip_ttl = inp->inp_ip_ttl;
2907 xi->inp_ip_p = inp->inp_ip_p;
2908 xi->inp_ip_minttl = inp->inp_ip_minttl;
2913 db_print_indent(int indent)
2917 for (i = 0; i < indent; i++)
2922 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2924 char faddr_str[48], laddr_str[48];
2926 db_print_indent(indent);
2927 db_printf("%s at %p\n", name, inc);
2932 if (inc->inc_flags & INC_ISIPV6) {
2934 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2935 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2940 inet_ntoa_r(inc->inc_laddr, laddr_str);
2941 inet_ntoa_r(inc->inc_faddr, faddr_str);
2943 db_print_indent(indent);
2944 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2945 ntohs(inc->inc_lport));
2946 db_print_indent(indent);
2947 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2948 ntohs(inc->inc_fport));
2952 db_print_inpflags(int inp_flags)
2957 if (inp_flags & INP_RECVOPTS) {
2958 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2961 if (inp_flags & INP_RECVRETOPTS) {
2962 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2965 if (inp_flags & INP_RECVDSTADDR) {
2966 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2969 if (inp_flags & INP_ORIGDSTADDR) {
2970 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2973 if (inp_flags & INP_HDRINCL) {
2974 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2977 if (inp_flags & INP_HIGHPORT) {
2978 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2981 if (inp_flags & INP_LOWPORT) {
2982 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2985 if (inp_flags & INP_ANONPORT) {
2986 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2989 if (inp_flags & INP_RECVIF) {
2990 db_printf("%sINP_RECVIF", comma ? ", " : "");
2993 if (inp_flags & INP_MTUDISC) {
2994 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2997 if (inp_flags & INP_RECVTTL) {
2998 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3001 if (inp_flags & INP_DONTFRAG) {
3002 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3005 if (inp_flags & INP_RECVTOS) {
3006 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3009 if (inp_flags & IN6P_IPV6_V6ONLY) {
3010 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3013 if (inp_flags & IN6P_PKTINFO) {
3014 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3017 if (inp_flags & IN6P_HOPLIMIT) {
3018 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3021 if (inp_flags & IN6P_HOPOPTS) {
3022 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3025 if (inp_flags & IN6P_DSTOPTS) {
3026 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3029 if (inp_flags & IN6P_RTHDR) {
3030 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3033 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3034 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3037 if (inp_flags & IN6P_TCLASS) {
3038 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3041 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3042 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3045 if (inp_flags & INP_TIMEWAIT) {
3046 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3049 if (inp_flags & INP_ONESBCAST) {
3050 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3053 if (inp_flags & INP_DROPPED) {
3054 db_printf("%sINP_DROPPED", comma ? ", " : "");
3057 if (inp_flags & INP_SOCKREF) {
3058 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3061 if (inp_flags & IN6P_RFC2292) {
3062 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3065 if (inp_flags & IN6P_MTU) {
3066 db_printf("IN6P_MTU%s", comma ? ", " : "");
3072 db_print_inpvflag(u_char inp_vflag)
3077 if (inp_vflag & INP_IPV4) {
3078 db_printf("%sINP_IPV4", comma ? ", " : "");
3081 if (inp_vflag & INP_IPV6) {
3082 db_printf("%sINP_IPV6", comma ? ", " : "");
3085 if (inp_vflag & INP_IPV6PROTO) {
3086 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3092 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3095 db_print_indent(indent);
3096 db_printf("%s at %p\n", name, inp);
3100 db_print_indent(indent);
3101 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3103 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3105 db_print_indent(indent);
3106 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3107 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3109 db_print_indent(indent);
3110 db_printf("inp_label: %p inp_flags: 0x%x (",
3111 inp->inp_label, inp->inp_flags);
3112 db_print_inpflags(inp->inp_flags);
3115 db_print_indent(indent);
3116 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3118 db_print_inpvflag(inp->inp_vflag);
3121 db_print_indent(indent);
3122 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3123 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3125 db_print_indent(indent);
3127 if (inp->inp_vflag & INP_IPV6) {
3128 db_printf("in6p_options: %p in6p_outputopts: %p "
3129 "in6p_moptions: %p\n", inp->in6p_options,
3130 inp->in6p_outputopts, inp->in6p_moptions);
3131 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3132 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3137 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3138 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3139 inp->inp_options, inp->inp_moptions);
3142 db_print_indent(indent);
3143 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3144 (uintmax_t)inp->inp_gencnt);
3147 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3152 db_printf("usage: show inpcb <addr>\n");
3155 inp = (struct inpcb *)addr;
3157 db_print_inpcb(inp, "inpcb", 0);
3163 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3167 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3169 union if_snd_tag_modify_params params = {
3170 .rate_limit.max_rate = max_pacing_rate,
3172 struct m_snd_tag *mst;
3176 mst = inp->inp_snd_tag;
3184 if (ifp->if_snd_tag_modify == NULL) {
3187 error = ifp->if_snd_tag_modify(mst, ¶ms);
3193 * Query existing TX rate limit based on the existing
3194 * "inp->inp_snd_tag", if any.
3197 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3199 union if_snd_tag_query_params params = { };
3200 struct m_snd_tag *mst;
3204 mst = inp->inp_snd_tag;
3212 if (ifp->if_snd_tag_query == NULL) {
3215 error = ifp->if_snd_tag_query(mst, ¶ms);
3216 if (error == 0 && p_max_pacing_rate != NULL)
3217 *p_max_pacing_rate = params.rate_limit.max_rate;
3223 * Query existing TX queue level based on the existing
3224 * "inp->inp_snd_tag", if any.
3227 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3229 union if_snd_tag_query_params params = { };
3230 struct m_snd_tag *mst;
3234 mst = inp->inp_snd_tag;
3242 if (ifp->if_snd_tag_query == NULL)
3243 return (EOPNOTSUPP);
3245 error = ifp->if_snd_tag_query(mst, ¶ms);
3246 if (error == 0 && p_txqueue_level != NULL)
3247 *p_txqueue_level = params.rate_limit.queue_level;
3252 * Allocate a new TX rate limit send tag from the network interface
3253 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3256 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3257 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3259 union if_snd_tag_alloc_params params = {
3260 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3261 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3262 .rate_limit.hdr.flowid = flowid,
3263 .rate_limit.hdr.flowtype = flowtype,
3264 .rate_limit.max_rate = max_pacing_rate,
3268 INP_WLOCK_ASSERT(inp);
3270 if (inp->inp_snd_tag != NULL)
3273 if (ifp->if_snd_tag_alloc == NULL) {
3276 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3279 * At success increment the refcount on
3280 * the send tag's network interface:
3283 if_ref(inp->inp_snd_tag->ifp);
3289 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3293 in_pcbdetach_txrtlmt(struct inpcb *inp)
3295 struct m_snd_tag *mst;
3298 INP_WLOCK_ASSERT(inp);
3300 mst = inp->inp_snd_tag;
3301 inp->inp_snd_tag = NULL;
3311 * If the device was detached while we still had reference(s)
3312 * on the ifp, we assume if_snd_tag_free() was replaced with
3315 ifp->if_snd_tag_free(mst);
3317 /* release reference count on network interface */
3322 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3323 * is set in the fast path and will attach/detach/modify the TX rate
3324 * limit send tag based on the socket's so_max_pacing_rate value.
3327 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3329 struct socket *socket;
3330 uint32_t max_pacing_rate;
3337 socket = inp->inp_socket;
3341 if (!INP_WLOCKED(inp)) {
3343 * NOTE: If the write locking fails, we need to bail
3344 * out and use the non-ratelimited ring for the
3345 * transmit until there is a new chance to get the
3348 if (!INP_TRY_UPGRADE(inp))
3356 * NOTE: The so_max_pacing_rate value is read unlocked,
3357 * because atomic updates are not required since the variable
3358 * is checked at every mbuf we send. It is assumed that the
3359 * variable read itself will be atomic.
3361 max_pacing_rate = socket->so_max_pacing_rate;
3364 * NOTE: When attaching to a network interface a reference is
3365 * made to ensure the network interface doesn't go away until
3366 * all ratelimit connections are gone. The network interface
3367 * pointers compared below represent valid network interfaces,
3368 * except when comparing towards NULL.
3370 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3372 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3373 if (inp->inp_snd_tag != NULL)
3374 in_pcbdetach_txrtlmt(inp);
3376 } else if (inp->inp_snd_tag == NULL) {
3378 * In order to utilize packet pacing with RSS, we need
3379 * to wait until there is a valid RSS hash before we
3382 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3385 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3386 mb->m_pkthdr.flowid, max_pacing_rate);
3389 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3391 if (error == 0 || error == EOPNOTSUPP)
3392 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3398 * Track route changes for TX rate limiting.
3401 in_pcboutput_eagain(struct inpcb *inp)
3408 if (inp->inp_snd_tag == NULL)
3411 if (!INP_WLOCKED(inp)) {
3413 * NOTE: If the write locking fails, we need to bail
3414 * out and use the non-ratelimited ring for the
3415 * transmit until there is a new chance to get the
3418 if (!INP_TRY_UPGRADE(inp))
3425 /* detach rate limiting */
3426 in_pcbdetach_txrtlmt(inp);
3428 /* make sure new mbuf send tag allocation is made */
3429 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3434 #endif /* RATELIMIT */