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 LIST_INSERT_HEAD(hdr, grp, il_list);
243 in_pcblbgroup_free(struct inpcblbgroup *grp)
246 LIST_REMOVE(grp, il_list);
250 static struct inpcblbgroup *
251 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
252 struct inpcblbgroup *old_grp, int size)
254 struct inpcblbgroup *grp;
257 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
258 old_grp->il_lport, &old_grp->il_dependladdr, size);
262 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
263 ("invalid new local group size %d and old local group count %d",
264 grp->il_inpsiz, old_grp->il_inpcnt));
266 for (i = 0; i < old_grp->il_inpcnt; ++i)
267 grp->il_inp[i] = old_grp->il_inp[i];
268 grp->il_inpcnt = old_grp->il_inpcnt;
269 in_pcblbgroup_free(old_grp);
274 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
275 * and shrink group if possible.
278 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
281 struct inpcblbgroup *grp = *grpp;
283 for (; i + 1 < grp->il_inpcnt; ++i)
284 grp->il_inp[i] = grp->il_inp[i + 1];
287 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
288 grp->il_inpcnt <= (grp->il_inpsiz / 4)) {
289 /* Shrink this group. */
290 struct inpcblbgroup *new_grp =
291 in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
299 * Add PCB to load balance group for SO_REUSEPORT_LB option.
302 in_pcbinslbgrouphash(struct inpcb *inp)
304 const static struct timeval interval = { 60, 0 };
305 static struct timeval lastprint;
306 struct inpcbinfo *pcbinfo;
307 struct inpcblbgrouphead *hdr;
308 struct inpcblbgroup *grp;
309 uint16_t hashmask, lport;
310 uint32_t group_index;
313 pcbinfo = inp->inp_pcbinfo;
315 INP_WLOCK_ASSERT(inp);
316 INP_HASH_WLOCK_ASSERT(pcbinfo);
318 if (pcbinfo->ipi_lbgrouphashbase == NULL)
321 hashmask = pcbinfo->ipi_lbgrouphashmask;
322 lport = inp->inp_lport;
323 group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask);
324 hdr = &pcbinfo->ipi_lbgrouphashbase[group_index];
327 * Don't allow jailed socket to join local group.
329 if (inp->inp_socket != NULL)
330 cred = inp->inp_socket->so_cred;
333 if (cred != NULL && jailed(cred))
338 * Don't allow IPv4 mapped INET6 wild socket.
340 if ((inp->inp_vflag & INP_IPV4) &&
341 inp->inp_laddr.s_addr == INADDR_ANY &&
342 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
347 hdr = &pcbinfo->ipi_lbgrouphashbase[
348 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
349 pcbinfo->ipi_lbgrouphashmask)];
350 LIST_FOREACH(grp, hdr, il_list) {
351 if (grp->il_vflag == inp->inp_vflag &&
352 grp->il_lport == inp->inp_lport &&
353 memcmp(&grp->il_dependladdr,
354 &inp->inp_inc.inc_ie.ie_dependladdr,
355 sizeof(grp->il_dependladdr)) == 0) {
360 /* Create new load balance group. */
361 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
362 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
363 INPCBLBGROUP_SIZMIN);
366 } else if (grp->il_inpcnt == grp->il_inpsiz) {
367 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
368 if (ratecheck(&lastprint, &interval))
369 printf("lb group port %d, limit reached\n",
370 ntohs(grp->il_lport));
374 /* Expand this local group. */
375 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
380 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
381 ("invalid local group size %d and count %d",
382 grp->il_inpsiz, grp->il_inpcnt));
384 grp->il_inp[grp->il_inpcnt] = inp;
390 * Remove PCB from load balance group.
393 in_pcbremlbgrouphash(struct inpcb *inp)
395 struct inpcbinfo *pcbinfo;
396 struct inpcblbgrouphead *hdr;
397 struct inpcblbgroup *grp;
400 pcbinfo = inp->inp_pcbinfo;
402 INP_WLOCK_ASSERT(inp);
403 INP_HASH_WLOCK_ASSERT(pcbinfo);
405 if (pcbinfo->ipi_lbgrouphashbase == NULL)
408 hdr = &pcbinfo->ipi_lbgrouphashbase[
409 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
410 pcbinfo->ipi_lbgrouphashmask)];
412 LIST_FOREACH(grp, hdr, il_list) {
413 for (i = 0; i < grp->il_inpcnt; ++i) {
414 if (grp->il_inp[i] != inp)
417 if (grp->il_inpcnt == 1) {
418 /* We are the last, free this local group. */
419 in_pcblbgroup_free(grp);
421 /* Pull up inpcbs, shrink group if possible. */
422 in_pcblbgroup_reorder(hdr, &grp, i);
430 * Different protocols initialize their inpcbs differently - giving
431 * different name to the lock. But they all are disposed the same.
434 inpcb_fini(void *mem, int size)
436 struct inpcb *inp = mem;
438 INP_LOCK_DESTROY(inp);
442 * Initialize an inpcbinfo -- we should be able to reduce the number of
446 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
447 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
448 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
451 INP_INFO_LOCK_INIT(pcbinfo, name);
452 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
453 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
455 pcbinfo->ipi_vnet = curvnet;
457 pcbinfo->ipi_listhead = listhead;
458 CK_LIST_INIT(pcbinfo->ipi_listhead);
459 pcbinfo->ipi_count = 0;
460 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
461 &pcbinfo->ipi_hashmask);
462 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
463 &pcbinfo->ipi_porthashmask);
464 pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
465 &pcbinfo->ipi_lbgrouphashmask);
467 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
469 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
470 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
471 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
472 uma_zone_set_warning(pcbinfo->ipi_zone,
473 "kern.ipc.maxsockets limit reached");
477 * Destroy an inpcbinfo.
480 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
483 KASSERT(pcbinfo->ipi_count == 0,
484 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
486 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
487 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
488 pcbinfo->ipi_porthashmask);
489 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
490 pcbinfo->ipi_lbgrouphashmask);
492 in_pcbgroup_destroy(pcbinfo);
494 uma_zdestroy(pcbinfo->ipi_zone);
495 INP_LIST_LOCK_DESTROY(pcbinfo);
496 INP_HASH_LOCK_DESTROY(pcbinfo);
497 INP_INFO_LOCK_DESTROY(pcbinfo);
501 * Allocate a PCB and associate it with the socket.
502 * On success return with the PCB locked.
505 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
511 if (pcbinfo == &V_tcbinfo) {
512 INP_INFO_RLOCK_ASSERT(pcbinfo);
514 INP_INFO_WLOCK_ASSERT(pcbinfo);
519 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
522 bzero(&inp->inp_start_zero, inp_zero_size);
523 inp->inp_pcbinfo = pcbinfo;
524 inp->inp_socket = so;
525 inp->inp_cred = crhold(so->so_cred);
526 inp->inp_inc.inc_fibnum = so->so_fibnum;
528 error = mac_inpcb_init(inp, M_NOWAIT);
531 mac_inpcb_create(so, inp);
533 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
534 error = ipsec_init_pcbpolicy(inp);
537 mac_inpcb_destroy(inp);
543 if (INP_SOCKAF(so) == AF_INET6) {
544 inp->inp_vflag |= INP_IPV6PROTO;
546 inp->inp_flags |= IN6P_IPV6_V6ONLY;
550 INP_LIST_WLOCK(pcbinfo);
551 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
552 pcbinfo->ipi_count++;
553 so->so_pcb = (caddr_t)inp;
555 if (V_ip6_auto_flowlabel)
556 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
558 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
559 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
562 * Routes in inpcb's can cache L2 as well; they are guaranteed
565 inp->inp_route.ro_flags = RT_LLE_CACHE;
566 INP_LIST_WUNLOCK(pcbinfo);
567 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
570 crfree(inp->inp_cred);
571 uma_zfree(pcbinfo->ipi_zone, inp);
579 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
583 INP_WLOCK_ASSERT(inp);
584 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
586 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
588 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
589 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
590 &inp->inp_lport, cred);
593 if (in_pcbinshash(inp) != 0) {
594 inp->inp_laddr.s_addr = INADDR_ANY;
599 inp->inp_flags |= INP_ANONPORT;
605 * Select a local port (number) to use.
607 #if defined(INET) || defined(INET6)
609 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
610 struct ucred *cred, int lookupflags)
612 struct inpcbinfo *pcbinfo;
613 struct inpcb *tmpinp;
614 unsigned short *lastport;
615 int count, dorandom, error;
616 u_short aux, first, last, lport;
618 struct in_addr laddr;
621 pcbinfo = inp->inp_pcbinfo;
624 * Because no actual state changes occur here, a global write lock on
625 * the pcbinfo isn't required.
627 INP_LOCK_ASSERT(inp);
628 INP_HASH_LOCK_ASSERT(pcbinfo);
630 if (inp->inp_flags & INP_HIGHPORT) {
631 first = V_ipport_hifirstauto; /* sysctl */
632 last = V_ipport_hilastauto;
633 lastport = &pcbinfo->ipi_lasthi;
634 } else if (inp->inp_flags & INP_LOWPORT) {
635 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
638 first = V_ipport_lowfirstauto; /* 1023 */
639 last = V_ipport_lowlastauto; /* 600 */
640 lastport = &pcbinfo->ipi_lastlow;
642 first = V_ipport_firstauto; /* sysctl */
643 last = V_ipport_lastauto;
644 lastport = &pcbinfo->ipi_lastport;
647 * For UDP(-Lite), use random port allocation as long as the user
648 * allows it. For TCP (and as of yet unknown) connections,
649 * use random port allocation only if the user allows it AND
650 * ipport_tick() allows it.
652 if (V_ipport_randomized &&
653 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
654 pcbinfo == &V_ulitecbinfo))
659 * It makes no sense to do random port allocation if
660 * we have the only port available.
664 /* Make sure to not include UDP(-Lite) packets in the count. */
665 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
666 V_ipport_tcpallocs++;
668 * Instead of having two loops further down counting up or down
669 * make sure that first is always <= last and go with only one
670 * code path implementing all logic.
679 /* Make the compiler happy. */
681 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
682 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
687 tmpinp = NULL; /* Make compiler happy. */
691 *lastport = first + (arc4random() % (last - first));
693 count = last - first;
696 if (count-- < 0) /* completely used? */
697 return (EADDRNOTAVAIL);
699 if (*lastport < first || *lastport > last)
701 lport = htons(*lastport);
704 if ((inp->inp_vflag & INP_IPV6) != 0)
705 tmpinp = in6_pcblookup_local(pcbinfo,
706 &inp->in6p_laddr, lport, lookupflags, cred);
708 #if defined(INET) && defined(INET6)
712 tmpinp = in_pcblookup_local(pcbinfo, laddr,
713 lport, lookupflags, cred);
715 } while (tmpinp != NULL);
718 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
719 laddrp->s_addr = laddr.s_addr;
727 * Return cached socket options.
730 inp_so_options(const struct inpcb *inp)
736 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
737 so_options |= SO_REUSEPORT_LB;
738 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
739 so_options |= SO_REUSEPORT;
740 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
741 so_options |= SO_REUSEADDR;
744 #endif /* INET || INET6 */
747 * Check if a new BINDMULTI socket is allowed to be created.
749 * ni points to the new inp.
750 * oi points to the exisitng inp.
752 * This checks whether the existing inp also has BINDMULTI and
753 * whether the credentials match.
756 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
758 /* Check permissions match */
759 if ((ni->inp_flags2 & INP_BINDMULTI) &&
760 (ni->inp_cred->cr_uid !=
761 oi->inp_cred->cr_uid))
764 /* Check the existing inp has BINDMULTI set */
765 if ((ni->inp_flags2 & INP_BINDMULTI) &&
766 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
770 * We're okay - either INP_BINDMULTI isn't set on ni, or
771 * it is and it matches the checks.
778 * Set up a bind operation on a PCB, performing port allocation
779 * as required, but do not actually modify the PCB. Callers can
780 * either complete the bind by setting inp_laddr/inp_lport and
781 * calling in_pcbinshash(), or they can just use the resulting
782 * port and address to authorise the sending of a once-off packet.
784 * On error, the values of *laddrp and *lportp are not changed.
787 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
788 u_short *lportp, struct ucred *cred)
790 struct socket *so = inp->inp_socket;
791 struct sockaddr_in *sin;
792 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
793 struct in_addr laddr;
795 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
799 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
800 * so that we don't have to add to the (already messy) code below.
802 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
805 * No state changes, so read locks are sufficient here.
807 INP_LOCK_ASSERT(inp);
808 INP_HASH_LOCK_ASSERT(pcbinfo);
810 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
811 return (EADDRNOTAVAIL);
812 laddr.s_addr = *laddrp;
813 if (nam != NULL && laddr.s_addr != INADDR_ANY)
815 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
816 lookupflags = INPLOOKUP_WILDCARD;
818 if ((error = prison_local_ip4(cred, &laddr)) != 0)
821 sin = (struct sockaddr_in *)nam;
822 if (nam->sa_len != sizeof (*sin))
826 * We should check the family, but old programs
827 * incorrectly fail to initialize it.
829 if (sin->sin_family != AF_INET)
830 return (EAFNOSUPPORT);
832 error = prison_local_ip4(cred, &sin->sin_addr);
835 if (sin->sin_port != *lportp) {
836 /* Don't allow the port to change. */
839 lport = sin->sin_port;
841 /* NB: lport is left as 0 if the port isn't being changed. */
842 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
844 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
845 * allow complete duplication of binding if
846 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
847 * and a multicast address is bound on both
848 * new and duplicated sockets.
850 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
851 reuseport = SO_REUSEADDR|SO_REUSEPORT;
853 * XXX: How to deal with SO_REUSEPORT_LB here?
854 * Treat same as SO_REUSEPORT for now.
856 if ((so->so_options &
857 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
858 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
859 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
860 sin->sin_port = 0; /* yech... */
861 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
863 * Is the address a local IP address?
864 * If INP_BINDANY is set, then the socket may be bound
865 * to any endpoint address, local or not.
867 if ((inp->inp_flags & INP_BINDANY) == 0 &&
868 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
869 return (EADDRNOTAVAIL);
871 laddr = sin->sin_addr;
877 if (ntohs(lport) <= V_ipport_reservedhigh &&
878 ntohs(lport) >= V_ipport_reservedlow &&
879 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
882 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
883 priv_check_cred(inp->inp_cred,
884 PRIV_NETINET_REUSEPORT, 0) != 0) {
885 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
886 lport, INPLOOKUP_WILDCARD, cred);
889 * This entire block sorely needs a rewrite.
892 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
893 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
894 (so->so_type != SOCK_STREAM ||
895 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
896 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
897 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
898 (t->inp_flags2 & INP_REUSEPORT) ||
899 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
900 (inp->inp_cred->cr_uid !=
901 t->inp_cred->cr_uid))
905 * If the socket is a BINDMULTI socket, then
906 * the credentials need to match and the
907 * original socket also has to have been bound
910 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
913 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
914 lport, lookupflags, cred);
915 if (t && (t->inp_flags & INP_TIMEWAIT)) {
917 * XXXRW: If an incpb has had its timewait
918 * state recycled, we treat the address as
919 * being in use (for now). This is better
920 * than a panic, but not desirable.
924 ((reuseport & tw->tw_so_options) == 0 &&
926 tw->tw_so_options) == 0)) {
930 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
931 (reuseport & inp_so_options(t)) == 0 &&
932 (reuseport_lb & inp_so_options(t)) == 0) {
934 if (ntohl(sin->sin_addr.s_addr) !=
936 ntohl(t->inp_laddr.s_addr) !=
938 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
939 (t->inp_vflag & INP_IPV6PROTO) == 0)
942 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
950 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
955 *laddrp = laddr.s_addr;
961 * Connect from a socket to a specified address.
962 * Both address and port must be specified in argument sin.
963 * If don't have a local address for this socket yet,
967 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
968 struct ucred *cred, struct mbuf *m)
970 u_short lport, fport;
971 in_addr_t laddr, faddr;
974 INP_WLOCK_ASSERT(inp);
975 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
977 lport = inp->inp_lport;
978 laddr = inp->inp_laddr.s_addr;
979 anonport = (lport == 0);
980 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
985 /* Do the initial binding of the local address if required. */
986 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
987 inp->inp_lport = lport;
988 inp->inp_laddr.s_addr = laddr;
989 if (in_pcbinshash(inp) != 0) {
990 inp->inp_laddr.s_addr = INADDR_ANY;
996 /* Commit the remaining changes. */
997 inp->inp_lport = lport;
998 inp->inp_laddr.s_addr = laddr;
999 inp->inp_faddr.s_addr = faddr;
1000 inp->inp_fport = fport;
1001 in_pcbrehash_mbuf(inp, m);
1004 inp->inp_flags |= INP_ANONPORT;
1009 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1012 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1016 * Do proper source address selection on an unbound socket in case
1017 * of connect. Take jails into account as well.
1020 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1024 struct sockaddr *sa;
1025 struct sockaddr_in *sin;
1029 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1031 * Bypass source address selection and use the primary jail IP
1034 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1038 bzero(&sro, sizeof(sro));
1040 sin = (struct sockaddr_in *)&sro.ro_dst;
1041 sin->sin_family = AF_INET;
1042 sin->sin_len = sizeof(struct sockaddr_in);
1043 sin->sin_addr.s_addr = faddr->s_addr;
1046 * If route is known our src addr is taken from the i/f,
1049 * Find out route to destination.
1051 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1052 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1055 * If we found a route, use the address corresponding to
1056 * the outgoing interface.
1058 * Otherwise assume faddr is reachable on a directly connected
1059 * network and try to find a corresponding interface to take
1060 * the source address from.
1063 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1064 struct in_ifaddr *ia;
1067 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1068 inp->inp_socket->so_fibnum));
1070 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1071 inp->inp_socket->so_fibnum));
1075 error = ENETUNREACH;
1079 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1080 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1086 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1089 if (sa->sa_family != AF_INET)
1091 sin = (struct sockaddr_in *)sa;
1092 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1093 ia = (struct in_ifaddr *)ifa;
1098 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1102 /* 3. As a last resort return the 'default' jail address. */
1103 error = prison_get_ip4(cred, laddr);
1108 * If the outgoing interface on the route found is not
1109 * a loopback interface, use the address from that interface.
1110 * In case of jails do those three steps:
1111 * 1. check if the interface address belongs to the jail. If so use it.
1112 * 2. check if we have any address on the outgoing interface
1113 * belonging to this jail. If so use it.
1114 * 3. as a last resort return the 'default' jail address.
1116 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1117 struct in_ifaddr *ia;
1120 /* If not jailed, use the default returned. */
1121 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1122 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1123 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1128 /* 1. Check if the iface address belongs to the jail. */
1129 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1130 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1131 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1132 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1137 * 2. Check if we have any address on the outgoing interface
1138 * belonging to this jail.
1141 ifp = sro.ro_rt->rt_ifp;
1142 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1144 if (sa->sa_family != AF_INET)
1146 sin = (struct sockaddr_in *)sa;
1147 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1148 ia = (struct in_ifaddr *)ifa;
1153 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1157 /* 3. As a last resort return the 'default' jail address. */
1158 error = prison_get_ip4(cred, laddr);
1163 * The outgoing interface is marked with 'loopback net', so a route
1164 * to ourselves is here.
1165 * Try to find the interface of the destination address and then
1166 * take the address from there. That interface is not necessarily
1167 * a loopback interface.
1168 * In case of jails, check that it is an address of the jail
1169 * and if we cannot find, fall back to the 'default' jail address.
1171 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1172 struct sockaddr_in sain;
1173 struct in_ifaddr *ia;
1175 bzero(&sain, sizeof(struct sockaddr_in));
1176 sain.sin_family = AF_INET;
1177 sain.sin_len = sizeof(struct sockaddr_in);
1178 sain.sin_addr.s_addr = faddr->s_addr;
1180 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1181 inp->inp_socket->so_fibnum));
1183 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1184 inp->inp_socket->so_fibnum));
1186 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1188 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1190 error = ENETUNREACH;
1193 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1203 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1205 if (sa->sa_family != AF_INET)
1207 sin = (struct sockaddr_in *)sa;
1208 if (prison_check_ip4(cred,
1209 &sin->sin_addr) == 0) {
1210 ia = (struct in_ifaddr *)ifa;
1215 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1220 /* 3. As a last resort return the 'default' jail address. */
1221 error = prison_get_ip4(cred, laddr);
1227 if (sro.ro_rt != NULL)
1233 * Set up for a connect from a socket to the specified address.
1234 * On entry, *laddrp and *lportp should contain the current local
1235 * address and port for the PCB; these are updated to the values
1236 * that should be placed in inp_laddr and inp_lport to complete
1239 * On success, *faddrp and *fportp will be set to the remote address
1240 * and port. These are not updated in the error case.
1242 * If the operation fails because the connection already exists,
1243 * *oinpp will be set to the PCB of that connection so that the
1244 * caller can decide to override it. In all other cases, *oinpp
1248 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1249 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1250 struct inpcb **oinpp, struct ucred *cred)
1252 struct rm_priotracker in_ifa_tracker;
1253 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1254 struct in_ifaddr *ia;
1256 struct in_addr laddr, faddr;
1257 u_short lport, fport;
1261 * Because a global state change doesn't actually occur here, a read
1262 * lock is sufficient.
1264 INP_LOCK_ASSERT(inp);
1265 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1269 if (nam->sa_len != sizeof (*sin))
1271 if (sin->sin_family != AF_INET)
1272 return (EAFNOSUPPORT);
1273 if (sin->sin_port == 0)
1274 return (EADDRNOTAVAIL);
1275 laddr.s_addr = *laddrp;
1277 faddr = sin->sin_addr;
1278 fport = sin->sin_port;
1280 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1282 * If the destination address is INADDR_ANY,
1283 * use the primary local address.
1284 * If the supplied address is INADDR_BROADCAST,
1285 * and the primary interface supports broadcast,
1286 * choose the broadcast address for that interface.
1288 if (faddr.s_addr == INADDR_ANY) {
1289 IN_IFADDR_RLOCK(&in_ifa_tracker);
1291 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1292 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1294 (error = prison_get_ip4(cred, &faddr)) != 0)
1296 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1297 IN_IFADDR_RLOCK(&in_ifa_tracker);
1298 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1300 faddr = satosin(&CK_STAILQ_FIRST(
1301 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1302 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1305 if (laddr.s_addr == INADDR_ANY) {
1306 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1308 * If the destination address is multicast and an outgoing
1309 * interface has been set as a multicast option, prefer the
1310 * address of that interface as our source address.
1312 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1313 inp->inp_moptions != NULL) {
1314 struct ip_moptions *imo;
1317 imo = inp->inp_moptions;
1318 if (imo->imo_multicast_ifp != NULL) {
1319 ifp = imo->imo_multicast_ifp;
1320 IN_IFADDR_RLOCK(&in_ifa_tracker);
1321 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1322 if ((ia->ia_ifp == ifp) &&
1324 prison_check_ip4(cred,
1325 &ia->ia_addr.sin_addr) == 0))
1329 error = EADDRNOTAVAIL;
1331 laddr = ia->ia_addr.sin_addr;
1334 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1340 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1341 laddr, lport, 0, NULL);
1345 return (EADDRINUSE);
1348 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1353 *laddrp = laddr.s_addr;
1355 *faddrp = faddr.s_addr;
1361 in_pcbdisconnect(struct inpcb *inp)
1364 INP_WLOCK_ASSERT(inp);
1365 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1367 inp->inp_faddr.s_addr = INADDR_ANY;
1374 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1375 * For most protocols, this will be invoked immediately prior to calling
1376 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1377 * socket, in which case in_pcbfree() is deferred.
1380 in_pcbdetach(struct inpcb *inp)
1383 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1386 if (inp->inp_snd_tag != NULL)
1387 in_pcbdetach_txrtlmt(inp);
1389 inp->inp_socket->so_pcb = NULL;
1390 inp->inp_socket = NULL;
1394 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1395 * stability of an inpcb pointer despite the inpcb lock being released. This
1396 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1397 * but where the inpcb lock may already held, or when acquiring a reference
1400 * in_pcbref() should be used only to provide brief memory stability, and
1401 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1402 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1403 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1404 * lock and rele are the *only* safe operations that may be performed on the
1407 * While the inpcb will not be freed, releasing the inpcb lock means that the
1408 * connection's state may change, so the caller should be careful to
1409 * revalidate any cached state on reacquiring the lock. Drop the reference
1410 * using in_pcbrele().
1413 in_pcbref(struct inpcb *inp)
1416 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1418 refcount_acquire(&inp->inp_refcount);
1422 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1423 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1424 * return a flag indicating whether or not the inpcb remains valid. If it is
1425 * valid, we return with the inpcb lock held.
1427 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1428 * reference on an inpcb. Historically more work was done here (actually, in
1429 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1430 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1431 * about memory stability (and continued use of the write lock).
1434 in_pcbrele_rlocked(struct inpcb *inp)
1436 struct inpcbinfo *pcbinfo;
1438 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1440 INP_RLOCK_ASSERT(inp);
1442 if (refcount_release(&inp->inp_refcount) == 0) {
1444 * If the inpcb has been freed, let the caller know, even if
1445 * this isn't the last reference.
1447 if (inp->inp_flags2 & INP_FREED) {
1454 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1456 if (inp->inp_in_hpts || inp->inp_in_input) {
1457 struct tcp_hpts_entry *hpts;
1459 * We should not be on the hpts at
1460 * this point in any form. we must
1461 * get the lock to be sure.
1463 hpts = tcp_hpts_lock(inp);
1464 if (inp->inp_in_hpts)
1465 panic("Hpts:%p inp:%p at free still on hpts",
1467 mtx_unlock(&hpts->p_mtx);
1468 hpts = tcp_input_lock(inp);
1469 if (inp->inp_in_input)
1470 panic("Hpts:%p inp:%p at free still on input hpts",
1472 mtx_unlock(&hpts->p_mtx);
1476 pcbinfo = inp->inp_pcbinfo;
1477 uma_zfree(pcbinfo->ipi_zone, inp);
1482 in_pcbrele_wlocked(struct inpcb *inp)
1484 struct inpcbinfo *pcbinfo;
1486 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1488 INP_WLOCK_ASSERT(inp);
1490 if (refcount_release(&inp->inp_refcount) == 0) {
1492 * If the inpcb has been freed, let the caller know, even if
1493 * this isn't the last reference.
1495 if (inp->inp_flags2 & INP_FREED) {
1502 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1504 if (inp->inp_in_hpts || inp->inp_in_input) {
1505 struct tcp_hpts_entry *hpts;
1507 * We should not be on the hpts at
1508 * this point in any form. we must
1509 * get the lock to be sure.
1511 hpts = tcp_hpts_lock(inp);
1512 if (inp->inp_in_hpts)
1513 panic("Hpts:%p inp:%p at free still on hpts",
1515 mtx_unlock(&hpts->p_mtx);
1516 hpts = tcp_input_lock(inp);
1517 if (inp->inp_in_input)
1518 panic("Hpts:%p inp:%p at free still on input hpts",
1520 mtx_unlock(&hpts->p_mtx);
1524 pcbinfo = inp->inp_pcbinfo;
1525 uma_zfree(pcbinfo->ipi_zone, inp);
1530 * Temporary wrapper.
1533 in_pcbrele(struct inpcb *inp)
1536 return (in_pcbrele_wlocked(inp));
1540 in_pcblist_rele_rlocked(epoch_context_t ctx)
1542 struct in_pcblist *il;
1544 struct inpcbinfo *pcbinfo;
1547 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1548 pcbinfo = il->il_pcbinfo;
1550 INP_INFO_WLOCK(pcbinfo);
1551 for (i = 0; i < n; i++) {
1552 inp = il->il_inp_list[i];
1554 if (!in_pcbrele_rlocked(inp))
1557 INP_INFO_WUNLOCK(pcbinfo);
1562 inpcbport_free(epoch_context_t ctx)
1564 struct inpcbport *phd;
1566 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1571 in_pcbfree_deferred(epoch_context_t ctx)
1574 int released __unused;
1576 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1580 struct ip_moptions *imo = inp->inp_moptions;
1581 inp->inp_moptions = NULL;
1583 /* XXXRW: Do as much as possible here. */
1584 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1585 if (inp->inp_sp != NULL)
1586 ipsec_delete_pcbpolicy(inp);
1589 struct ip6_moptions *im6o = NULL;
1590 if (inp->inp_vflag & INP_IPV6PROTO) {
1591 ip6_freepcbopts(inp->in6p_outputopts);
1592 im6o = inp->in6p_moptions;
1593 inp->in6p_moptions = NULL;
1596 if (inp->inp_options)
1597 (void)m_free(inp->inp_options);
1599 crfree(inp->inp_cred);
1601 mac_inpcb_destroy(inp);
1603 released = in_pcbrele_wlocked(inp);
1606 ip6_freemoptions(im6o);
1609 inp_freemoptions(imo);
1614 * Unconditionally schedule an inpcb to be freed by decrementing its
1615 * reference count, which should occur only after the inpcb has been detached
1616 * from its socket. If another thread holds a temporary reference (acquired
1617 * using in_pcbref()) then the free is deferred until that reference is
1618 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1619 * work, including removal from global lists, is done in this context, where
1620 * the pcbinfo lock is held.
1623 in_pcbfree(struct inpcb *inp)
1625 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1627 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1628 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1629 ("%s: called twice for pcb %p", __func__, inp));
1630 if (inp->inp_flags2 & INP_FREED) {
1636 if (pcbinfo == &V_tcbinfo) {
1637 INP_INFO_LOCK_ASSERT(pcbinfo);
1639 INP_INFO_WLOCK_ASSERT(pcbinfo);
1642 INP_WLOCK_ASSERT(inp);
1643 INP_LIST_WLOCK(pcbinfo);
1644 in_pcbremlists(inp);
1645 INP_LIST_WUNLOCK(pcbinfo);
1646 RO_INVALIDATE_CACHE(&inp->inp_route);
1647 /* mark as destruction in progress */
1648 inp->inp_flags2 |= INP_FREED;
1650 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1654 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1655 * port reservation, and preventing it from being returned by inpcb lookups.
1657 * It is used by TCP to mark an inpcb as unused and avoid future packet
1658 * delivery or event notification when a socket remains open but TCP has
1659 * closed. This might occur as a result of a shutdown()-initiated TCP close
1660 * or a RST on the wire, and allows the port binding to be reused while still
1661 * maintaining the invariant that so_pcb always points to a valid inpcb until
1664 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1665 * in_pcbnotifyall() and in_pcbpurgeif0()?
1668 in_pcbdrop(struct inpcb *inp)
1671 INP_WLOCK_ASSERT(inp);
1673 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1674 MPASS(inp->inp_refcount > 1);
1678 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1681 inp->inp_flags |= INP_DROPPED;
1682 if (inp->inp_flags & INP_INHASHLIST) {
1683 struct inpcbport *phd = inp->inp_phd;
1685 INP_HASH_WLOCK(inp->inp_pcbinfo);
1686 in_pcbremlbgrouphash(inp);
1687 CK_LIST_REMOVE(inp, inp_hash);
1688 CK_LIST_REMOVE(inp, inp_portlist);
1689 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1690 CK_LIST_REMOVE(phd, phd_hash);
1691 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1693 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1694 inp->inp_flags &= ~INP_INHASHLIST;
1696 in_pcbgroup_remove(inp);
1703 * Common routines to return the socket addresses associated with inpcbs.
1706 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1708 struct sockaddr_in *sin;
1710 sin = malloc(sizeof *sin, M_SONAME,
1712 sin->sin_family = AF_INET;
1713 sin->sin_len = sizeof(*sin);
1714 sin->sin_addr = *addr_p;
1715 sin->sin_port = port;
1717 return (struct sockaddr *)sin;
1721 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1724 struct in_addr addr;
1727 inp = sotoinpcb(so);
1728 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1731 port = inp->inp_lport;
1732 addr = inp->inp_laddr;
1735 *nam = in_sockaddr(port, &addr);
1740 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1743 struct in_addr addr;
1746 inp = sotoinpcb(so);
1747 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1750 port = inp->inp_fport;
1751 addr = inp->inp_faddr;
1754 *nam = in_sockaddr(port, &addr);
1759 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1760 struct inpcb *(*notify)(struct inpcb *, int))
1762 struct inpcb *inp, *inp_temp;
1764 INP_INFO_WLOCK(pcbinfo);
1765 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1768 if ((inp->inp_vflag & INP_IPV4) == 0) {
1773 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1774 inp->inp_socket == NULL) {
1778 if ((*notify)(inp, errno))
1781 INP_INFO_WUNLOCK(pcbinfo);
1785 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1788 struct ip_moptions *imo;
1791 INP_INFO_WLOCK(pcbinfo);
1792 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1794 imo = inp->inp_moptions;
1795 if ((inp->inp_vflag & INP_IPV4) &&
1798 * Unselect the outgoing interface if it is being
1801 if (imo->imo_multicast_ifp == ifp)
1802 imo->imo_multicast_ifp = NULL;
1805 * Drop multicast group membership if we joined
1806 * through the interface being detached.
1808 * XXX This can all be deferred to an epoch_call
1810 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1812 if (imo->imo_membership[i]->inm_ifp == ifp) {
1813 IN_MULTI_LOCK_ASSERT();
1814 in_leavegroup_locked(imo->imo_membership[i], NULL);
1816 } else if (gap != 0)
1817 imo->imo_membership[i - gap] =
1818 imo->imo_membership[i];
1820 imo->imo_num_memberships -= gap;
1824 INP_INFO_WUNLOCK(pcbinfo);
1828 * Lookup a PCB based on the local address and port. Caller must hold the
1829 * hash lock. No inpcb locks or references are acquired.
1831 #define INP_LOOKUP_MAPPED_PCB_COST 3
1833 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1834 u_short lport, int lookupflags, struct ucred *cred)
1838 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1844 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1845 ("%s: invalid lookup flags %d", __func__, lookupflags));
1847 INP_HASH_LOCK_ASSERT(pcbinfo);
1849 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1850 struct inpcbhead *head;
1852 * Look for an unconnected (wildcard foreign addr) PCB that
1853 * matches the local address and port we're looking for.
1855 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1856 0, pcbinfo->ipi_hashmask)];
1857 CK_LIST_FOREACH(inp, head, inp_hash) {
1859 /* XXX inp locking */
1860 if ((inp->inp_vflag & INP_IPV4) == 0)
1863 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1864 inp->inp_laddr.s_addr == laddr.s_addr &&
1865 inp->inp_lport == lport) {
1870 prison_equal_ip4(cred->cr_prison,
1871 inp->inp_cred->cr_prison))
1880 struct inpcbporthead *porthash;
1881 struct inpcbport *phd;
1882 struct inpcb *match = NULL;
1884 * Best fit PCB lookup.
1886 * First see if this local port is in use by looking on the
1889 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1890 pcbinfo->ipi_porthashmask)];
1891 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1892 if (phd->phd_port == lport)
1897 * Port is in use by one or more PCBs. Look for best
1900 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1903 !prison_equal_ip4(inp->inp_cred->cr_prison,
1907 /* XXX inp locking */
1908 if ((inp->inp_vflag & INP_IPV4) == 0)
1911 * We never select the PCB that has
1912 * INP_IPV6 flag and is bound to :: if
1913 * we have another PCB which is bound
1914 * to 0.0.0.0. If a PCB has the
1915 * INP_IPV6 flag, then we set its cost
1916 * higher than IPv4 only PCBs.
1918 * Note that the case only happens
1919 * when a socket is bound to ::, under
1920 * the condition that the use of the
1921 * mapped address is allowed.
1923 if ((inp->inp_vflag & INP_IPV6) != 0)
1924 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1926 if (inp->inp_faddr.s_addr != INADDR_ANY)
1928 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1929 if (laddr.s_addr == INADDR_ANY)
1931 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1934 if (laddr.s_addr != INADDR_ANY)
1937 if (wildcard < matchwild) {
1939 matchwild = wildcard;
1948 #undef INP_LOOKUP_MAPPED_PCB_COST
1950 static struct inpcb *
1951 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1952 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1953 uint16_t fport, int lookupflags)
1955 struct inpcb *local_wild;
1956 const struct inpcblbgrouphead *hdr;
1957 struct inpcblbgroup *grp;
1960 INP_HASH_LOCK_ASSERT(pcbinfo);
1962 hdr = &pcbinfo->ipi_lbgrouphashbase[INP_PCBLBGROUP_PORTHASH(lport,
1963 pcbinfo->ipi_lbgrouphashmask)];
1966 * Order of socket selection:
1968 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1971 * - Load balanced group does not contain jailed sockets
1972 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1975 LIST_FOREACH(grp, hdr, il_list) {
1977 if (!(grp->il_vflag & INP_IPV4))
1980 if (grp->il_lport != lport)
1983 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1985 if (grp->il_laddr.s_addr == laddr->s_addr)
1986 return (grp->il_inp[idx]);
1987 if (grp->il_laddr.s_addr == INADDR_ANY &&
1988 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1989 local_wild = grp->il_inp[idx];
1991 return (local_wild);
1996 * Lookup PCB in hash list, using pcbgroup tables.
1998 static struct inpcb *
1999 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2000 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2001 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2003 struct inpcbhead *head;
2004 struct inpcb *inp, *tmpinp;
2005 u_short fport = fport_arg, lport = lport_arg;
2009 * First look for an exact match.
2012 INP_GROUP_LOCK(pcbgroup);
2013 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2014 pcbgroup->ipg_hashmask)];
2015 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2017 /* XXX inp locking */
2018 if ((inp->inp_vflag & INP_IPV4) == 0)
2021 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2022 inp->inp_laddr.s_addr == laddr.s_addr &&
2023 inp->inp_fport == fport &&
2024 inp->inp_lport == lport) {
2026 * XXX We should be able to directly return
2027 * the inp here, without any checks.
2028 * Well unless both bound with SO_REUSEPORT?
2030 if (prison_flag(inp->inp_cred, PR_IP4))
2036 if (tmpinp != NULL) {
2043 * For incoming connections, we may wish to do a wildcard
2044 * match for an RSS-local socket.
2046 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2047 struct inpcb *local_wild = NULL, *local_exact = NULL;
2049 struct inpcb *local_wild_mapped = NULL;
2051 struct inpcb *jail_wild = NULL;
2052 struct inpcbhead *head;
2056 * Order of socket selection - we always prefer jails.
2057 * 1. jailed, non-wild.
2059 * 3. non-jailed, non-wild.
2060 * 4. non-jailed, wild.
2063 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2064 lport, 0, pcbgroup->ipg_hashmask)];
2065 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2067 /* XXX inp locking */
2068 if ((inp->inp_vflag & INP_IPV4) == 0)
2071 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2072 inp->inp_lport != lport)
2075 injail = prison_flag(inp->inp_cred, PR_IP4);
2077 if (prison_check_ip4(inp->inp_cred,
2081 if (local_exact != NULL)
2085 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2090 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2092 /* XXX inp locking, NULL check */
2093 if (inp->inp_vflag & INP_IPV6PROTO)
2094 local_wild_mapped = inp;
2102 } /* LIST_FOREACH */
2111 inp = local_wild_mapped;
2119 * Then look for a wildcard match, if requested.
2121 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2122 struct inpcb *local_wild = NULL, *local_exact = NULL;
2124 struct inpcb *local_wild_mapped = NULL;
2126 struct inpcb *jail_wild = NULL;
2127 struct inpcbhead *head;
2131 * Order of socket selection - we always prefer jails.
2132 * 1. jailed, non-wild.
2134 * 3. non-jailed, non-wild.
2135 * 4. non-jailed, wild.
2137 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2138 0, pcbinfo->ipi_wildmask)];
2139 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2141 /* XXX inp locking */
2142 if ((inp->inp_vflag & INP_IPV4) == 0)
2145 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2146 inp->inp_lport != lport)
2149 injail = prison_flag(inp->inp_cred, PR_IP4);
2151 if (prison_check_ip4(inp->inp_cred,
2155 if (local_exact != NULL)
2159 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2164 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2166 /* XXX inp locking, NULL check */
2167 if (inp->inp_vflag & INP_IPV6PROTO)
2168 local_wild_mapped = inp;
2176 } /* LIST_FOREACH */
2184 inp = local_wild_mapped;
2188 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2189 INP_GROUP_UNLOCK(pcbgroup);
2193 if (lookupflags & INPLOOKUP_WLOCKPCB)
2194 locked = INP_TRY_WLOCK(inp);
2195 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2196 locked = INP_TRY_RLOCK(inp);
2198 panic("%s: locking bug", __func__);
2199 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2200 if (lookupflags & INPLOOKUP_WLOCKPCB)
2207 INP_GROUP_UNLOCK(pcbgroup);
2209 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2211 if (in_pcbrele_wlocked(inp))
2215 if (in_pcbrele_rlocked(inp))
2220 if (lookupflags & INPLOOKUP_WLOCKPCB)
2221 INP_WLOCK_ASSERT(inp);
2223 INP_RLOCK_ASSERT(inp);
2227 #endif /* PCBGROUP */
2230 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2231 * that the caller has locked the hash list, and will not perform any further
2232 * locking or reference operations on either the hash list or the connection.
2234 static struct inpcb *
2235 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2236 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2239 struct inpcbhead *head;
2240 struct inpcb *inp, *tmpinp;
2241 u_short fport = fport_arg, lport = lport_arg;
2244 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2245 ("%s: invalid lookup flags %d", __func__, lookupflags));
2246 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2247 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2250 * First look for an exact match.
2253 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2254 pcbinfo->ipi_hashmask)];
2255 CK_LIST_FOREACH(inp, head, inp_hash) {
2257 /* XXX inp locking */
2258 if ((inp->inp_vflag & INP_IPV4) == 0)
2261 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2262 inp->inp_laddr.s_addr == laddr.s_addr &&
2263 inp->inp_fport == fport &&
2264 inp->inp_lport == lport) {
2266 * XXX We should be able to directly return
2267 * the inp here, without any checks.
2268 * Well unless both bound with SO_REUSEPORT?
2270 if (prison_flag(inp->inp_cred, PR_IP4))
2280 * Then look in lb group (for wildcard match).
2282 if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2283 (lookupflags & INPLOOKUP_WILDCARD)) {
2284 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2285 fport, lookupflags);
2292 * Then look for a wildcard match, if requested.
2294 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2295 struct inpcb *local_wild = NULL, *local_exact = NULL;
2297 struct inpcb *local_wild_mapped = NULL;
2299 struct inpcb *jail_wild = NULL;
2303 * Order of socket selection - we always prefer jails.
2304 * 1. jailed, non-wild.
2306 * 3. non-jailed, non-wild.
2307 * 4. non-jailed, wild.
2310 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2311 0, pcbinfo->ipi_hashmask)];
2312 CK_LIST_FOREACH(inp, head, inp_hash) {
2314 /* XXX inp locking */
2315 if ((inp->inp_vflag & INP_IPV4) == 0)
2318 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2319 inp->inp_lport != lport)
2322 injail = prison_flag(inp->inp_cred, PR_IP4);
2324 if (prison_check_ip4(inp->inp_cred,
2328 if (local_exact != NULL)
2332 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2337 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2339 /* XXX inp locking, NULL check */
2340 if (inp->inp_vflag & INP_IPV6PROTO)
2341 local_wild_mapped = inp;
2349 } /* LIST_FOREACH */
2350 if (jail_wild != NULL)
2352 if (local_exact != NULL)
2353 return (local_exact);
2354 if (local_wild != NULL)
2355 return (local_wild);
2357 if (local_wild_mapped != NULL)
2358 return (local_wild_mapped);
2360 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2366 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2367 * hash list lock, and will return the inpcb locked (i.e., requires
2368 * INPLOOKUP_LOCKPCB).
2370 static struct inpcb *
2371 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2372 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2377 INP_HASH_RLOCK(pcbinfo);
2378 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2379 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2381 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2383 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2387 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2389 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2394 panic("%s: locking bug", __func__);
2397 if (lookupflags & INPLOOKUP_WLOCKPCB)
2398 INP_WLOCK_ASSERT(inp);
2400 INP_RLOCK_ASSERT(inp);
2404 INP_HASH_RUNLOCK(pcbinfo);
2409 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2410 * from which a pre-calculated hash value may be extracted.
2412 * Possibly more of this logic should be in in_pcbgroup.c.
2415 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2416 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2418 #if defined(PCBGROUP) && !defined(RSS)
2419 struct inpcbgroup *pcbgroup;
2422 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2423 ("%s: invalid lookup flags %d", __func__, lookupflags));
2424 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2425 ("%s: LOCKPCB not set", __func__));
2428 * When not using RSS, use connection groups in preference to the
2429 * reservation table when looking up 4-tuples. When using RSS, just
2430 * use the reservation table, due to the cost of the Toeplitz hash
2433 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2434 * we could be doing RSS with a non-Toeplitz hash that is affordable
2437 #if defined(PCBGROUP) && !defined(RSS)
2438 if (in_pcbgroup_enabled(pcbinfo)) {
2439 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2441 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2442 laddr, lport, lookupflags, ifp));
2445 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2450 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2451 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2452 struct ifnet *ifp, struct mbuf *m)
2455 struct inpcbgroup *pcbgroup;
2458 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2459 ("%s: invalid lookup flags %d", __func__, lookupflags));
2460 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2461 ("%s: LOCKPCB not set", __func__));
2465 * If we can use a hardware-generated hash to look up the connection
2466 * group, use that connection group to find the inpcb. Otherwise
2467 * fall back on a software hash -- or the reservation table if we're
2470 * XXXRW: As above, that policy belongs in the pcbgroup code.
2472 if (in_pcbgroup_enabled(pcbinfo) &&
2473 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2474 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2475 m->m_pkthdr.flowid);
2476 if (pcbgroup != NULL)
2477 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2478 fport, laddr, lport, lookupflags, ifp));
2480 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2482 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2483 laddr, lport, lookupflags, ifp));
2487 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2493 * Insert PCB onto various hash lists.
2496 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2498 struct inpcbhead *pcbhash;
2499 struct inpcbporthead *pcbporthash;
2500 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2501 struct inpcbport *phd;
2502 u_int32_t hashkey_faddr;
2505 INP_WLOCK_ASSERT(inp);
2506 INP_HASH_WLOCK_ASSERT(pcbinfo);
2508 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2509 ("in_pcbinshash: INP_INHASHLIST"));
2512 if (inp->inp_vflag & INP_IPV6)
2513 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2516 hashkey_faddr = inp->inp_faddr.s_addr;
2518 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2519 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2521 pcbporthash = &pcbinfo->ipi_porthashbase[
2522 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2525 * Add entry to load balance group.
2526 * Only do this if SO_REUSEPORT_LB is set.
2528 so_options = inp_so_options(inp);
2529 if (so_options & SO_REUSEPORT_LB) {
2530 int ret = in_pcbinslbgrouphash(inp);
2532 /* pcb lb group malloc fail (ret=ENOBUFS). */
2538 * Go through port list and look for a head for this lport.
2540 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2541 if (phd->phd_port == inp->inp_lport)
2545 * If none exists, malloc one and tack it on.
2548 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2550 return (ENOBUFS); /* XXX */
2552 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2553 phd->phd_port = inp->inp_lport;
2554 CK_LIST_INIT(&phd->phd_pcblist);
2555 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2558 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2559 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2560 inp->inp_flags |= INP_INHASHLIST;
2562 if (do_pcbgroup_update)
2563 in_pcbgroup_update(inp);
2569 * For now, there are two public interfaces to insert an inpcb into the hash
2570 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2571 * is used only in the TCP syncache, where in_pcbinshash is called before the
2572 * full 4-tuple is set for the inpcb, and we don't want to install in the
2573 * pcbgroup until later.
2575 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2576 * connection groups, and partially initialised inpcbs should not be exposed
2577 * to either reservation hash tables or pcbgroups.
2580 in_pcbinshash(struct inpcb *inp)
2583 return (in_pcbinshash_internal(inp, 1));
2587 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2590 return (in_pcbinshash_internal(inp, 0));
2594 * Move PCB to the proper hash bucket when { faddr, fport } have been
2595 * changed. NOTE: This does not handle the case of the lport changing (the
2596 * hashed port list would have to be updated as well), so the lport must
2597 * not change after in_pcbinshash() has been called.
2600 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2602 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2603 struct inpcbhead *head;
2604 u_int32_t hashkey_faddr;
2606 INP_WLOCK_ASSERT(inp);
2607 INP_HASH_WLOCK_ASSERT(pcbinfo);
2609 KASSERT(inp->inp_flags & INP_INHASHLIST,
2610 ("in_pcbrehash: !INP_INHASHLIST"));
2613 if (inp->inp_vflag & INP_IPV6)
2614 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2617 hashkey_faddr = inp->inp_faddr.s_addr;
2619 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2620 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2622 CK_LIST_REMOVE(inp, inp_hash);
2623 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2627 in_pcbgroup_update_mbuf(inp, m);
2629 in_pcbgroup_update(inp);
2634 in_pcbrehash(struct inpcb *inp)
2637 in_pcbrehash_mbuf(inp, NULL);
2641 * Remove PCB from various lists.
2644 in_pcbremlists(struct inpcb *inp)
2646 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2649 if (pcbinfo == &V_tcbinfo) {
2650 INP_INFO_RLOCK_ASSERT(pcbinfo);
2652 INP_INFO_WLOCK_ASSERT(pcbinfo);
2656 INP_WLOCK_ASSERT(inp);
2657 INP_LIST_WLOCK_ASSERT(pcbinfo);
2659 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2660 if (inp->inp_flags & INP_INHASHLIST) {
2661 struct inpcbport *phd = inp->inp_phd;
2663 INP_HASH_WLOCK(pcbinfo);
2665 /* XXX: Only do if SO_REUSEPORT_LB set? */
2666 in_pcbremlbgrouphash(inp);
2668 CK_LIST_REMOVE(inp, inp_hash);
2669 CK_LIST_REMOVE(inp, inp_portlist);
2670 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2671 CK_LIST_REMOVE(phd, phd_hash);
2672 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2674 INP_HASH_WUNLOCK(pcbinfo);
2675 inp->inp_flags &= ~INP_INHASHLIST;
2677 CK_LIST_REMOVE(inp, inp_list);
2678 pcbinfo->ipi_count--;
2680 in_pcbgroup_remove(inp);
2685 * Check for alternatives when higher level complains
2686 * about service problems. For now, invalidate cached
2687 * routing information. If the route was created dynamically
2688 * (by a redirect), time to try a default gateway again.
2691 in_losing(struct inpcb *inp)
2694 RO_INVALIDATE_CACHE(&inp->inp_route);
2699 * A set label operation has occurred at the socket layer, propagate the
2700 * label change into the in_pcb for the socket.
2703 in_pcbsosetlabel(struct socket *so)
2708 inp = sotoinpcb(so);
2709 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2713 mac_inpcb_sosetlabel(so, inp);
2720 * ipport_tick runs once per second, determining if random port allocation
2721 * should be continued. If more than ipport_randomcps ports have been
2722 * allocated in the last second, then we return to sequential port
2723 * allocation. We return to random allocation only once we drop below
2724 * ipport_randomcps for at least ipport_randomtime seconds.
2727 ipport_tick(void *xtp)
2729 VNET_ITERATOR_DECL(vnet_iter);
2731 VNET_LIST_RLOCK_NOSLEEP();
2732 VNET_FOREACH(vnet_iter) {
2733 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2734 if (V_ipport_tcpallocs <=
2735 V_ipport_tcplastcount + V_ipport_randomcps) {
2736 if (V_ipport_stoprandom > 0)
2737 V_ipport_stoprandom--;
2739 V_ipport_stoprandom = V_ipport_randomtime;
2740 V_ipport_tcplastcount = V_ipport_tcpallocs;
2743 VNET_LIST_RUNLOCK_NOSLEEP();
2744 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2751 callout_stop(&ipport_tick_callout);
2755 * The ipport_callout should start running at about the time we attach the
2756 * inet or inet6 domains.
2759 ipport_tick_init(const void *unused __unused)
2762 /* Start ipport_tick. */
2763 callout_init(&ipport_tick_callout, 1);
2764 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2765 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2766 SHUTDOWN_PRI_DEFAULT);
2768 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2769 ipport_tick_init, NULL);
2772 inp_wlock(struct inpcb *inp)
2779 inp_wunlock(struct inpcb *inp)
2786 inp_rlock(struct inpcb *inp)
2793 inp_runlock(struct inpcb *inp)
2799 #ifdef INVARIANT_SUPPORT
2801 inp_lock_assert(struct inpcb *inp)
2804 INP_WLOCK_ASSERT(inp);
2808 inp_unlock_assert(struct inpcb *inp)
2811 INP_UNLOCK_ASSERT(inp);
2816 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2820 INP_INFO_WLOCK(&V_tcbinfo);
2821 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2826 INP_INFO_WUNLOCK(&V_tcbinfo);
2830 inp_inpcbtosocket(struct inpcb *inp)
2833 INP_WLOCK_ASSERT(inp);
2834 return (inp->inp_socket);
2838 inp_inpcbtotcpcb(struct inpcb *inp)
2841 INP_WLOCK_ASSERT(inp);
2842 return ((struct tcpcb *)inp->inp_ppcb);
2846 inp_ip_tos_get(const struct inpcb *inp)
2849 return (inp->inp_ip_tos);
2853 inp_ip_tos_set(struct inpcb *inp, int val)
2856 inp->inp_ip_tos = val;
2860 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2861 uint32_t *faddr, uint16_t *fp)
2864 INP_LOCK_ASSERT(inp);
2865 *laddr = inp->inp_laddr.s_addr;
2866 *faddr = inp->inp_faddr.s_addr;
2867 *lp = inp->inp_lport;
2868 *fp = inp->inp_fport;
2872 so_sotoinpcb(struct socket *so)
2875 return (sotoinpcb(so));
2879 so_sototcpcb(struct socket *so)
2882 return (sototcpcb(so));
2886 * Create an external-format (``xinpcb'') structure using the information in
2887 * the kernel-format in_pcb structure pointed to by inp. This is done to
2888 * reduce the spew of irrelevant information over this interface, to isolate
2889 * user code from changes in the kernel structure, and potentially to provide
2890 * information-hiding if we decide that some of this information should be
2891 * hidden from users.
2894 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2897 xi->xi_len = sizeof(struct xinpcb);
2898 if (inp->inp_socket)
2899 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2901 bzero(&xi->xi_socket, sizeof(struct xsocket));
2902 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2903 xi->inp_gencnt = inp->inp_gencnt;
2904 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2905 xi->inp_flow = inp->inp_flow;
2906 xi->inp_flowid = inp->inp_flowid;
2907 xi->inp_flowtype = inp->inp_flowtype;
2908 xi->inp_flags = inp->inp_flags;
2909 xi->inp_flags2 = inp->inp_flags2;
2910 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2911 xi->in6p_cksum = inp->in6p_cksum;
2912 xi->in6p_hops = inp->in6p_hops;
2913 xi->inp_ip_tos = inp->inp_ip_tos;
2914 xi->inp_vflag = inp->inp_vflag;
2915 xi->inp_ip_ttl = inp->inp_ip_ttl;
2916 xi->inp_ip_p = inp->inp_ip_p;
2917 xi->inp_ip_minttl = inp->inp_ip_minttl;
2922 db_print_indent(int indent)
2926 for (i = 0; i < indent; i++)
2931 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2933 char faddr_str[48], laddr_str[48];
2935 db_print_indent(indent);
2936 db_printf("%s at %p\n", name, inc);
2941 if (inc->inc_flags & INC_ISIPV6) {
2943 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2944 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2949 inet_ntoa_r(inc->inc_laddr, laddr_str);
2950 inet_ntoa_r(inc->inc_faddr, faddr_str);
2952 db_print_indent(indent);
2953 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2954 ntohs(inc->inc_lport));
2955 db_print_indent(indent);
2956 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2957 ntohs(inc->inc_fport));
2961 db_print_inpflags(int inp_flags)
2966 if (inp_flags & INP_RECVOPTS) {
2967 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2970 if (inp_flags & INP_RECVRETOPTS) {
2971 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2974 if (inp_flags & INP_RECVDSTADDR) {
2975 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2978 if (inp_flags & INP_ORIGDSTADDR) {
2979 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2982 if (inp_flags & INP_HDRINCL) {
2983 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2986 if (inp_flags & INP_HIGHPORT) {
2987 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2990 if (inp_flags & INP_LOWPORT) {
2991 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2994 if (inp_flags & INP_ANONPORT) {
2995 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2998 if (inp_flags & INP_RECVIF) {
2999 db_printf("%sINP_RECVIF", comma ? ", " : "");
3002 if (inp_flags & INP_MTUDISC) {
3003 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3006 if (inp_flags & INP_RECVTTL) {
3007 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3010 if (inp_flags & INP_DONTFRAG) {
3011 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3014 if (inp_flags & INP_RECVTOS) {
3015 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3018 if (inp_flags & IN6P_IPV6_V6ONLY) {
3019 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3022 if (inp_flags & IN6P_PKTINFO) {
3023 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3026 if (inp_flags & IN6P_HOPLIMIT) {
3027 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3030 if (inp_flags & IN6P_HOPOPTS) {
3031 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3034 if (inp_flags & IN6P_DSTOPTS) {
3035 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3038 if (inp_flags & IN6P_RTHDR) {
3039 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3042 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3043 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3046 if (inp_flags & IN6P_TCLASS) {
3047 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3050 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3051 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3054 if (inp_flags & INP_TIMEWAIT) {
3055 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3058 if (inp_flags & INP_ONESBCAST) {
3059 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3062 if (inp_flags & INP_DROPPED) {
3063 db_printf("%sINP_DROPPED", comma ? ", " : "");
3066 if (inp_flags & INP_SOCKREF) {
3067 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3070 if (inp_flags & IN6P_RFC2292) {
3071 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3074 if (inp_flags & IN6P_MTU) {
3075 db_printf("IN6P_MTU%s", comma ? ", " : "");
3081 db_print_inpvflag(u_char inp_vflag)
3086 if (inp_vflag & INP_IPV4) {
3087 db_printf("%sINP_IPV4", comma ? ", " : "");
3090 if (inp_vflag & INP_IPV6) {
3091 db_printf("%sINP_IPV6", comma ? ", " : "");
3094 if (inp_vflag & INP_IPV6PROTO) {
3095 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3101 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3104 db_print_indent(indent);
3105 db_printf("%s at %p\n", name, inp);
3109 db_print_indent(indent);
3110 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3112 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3114 db_print_indent(indent);
3115 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3116 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3118 db_print_indent(indent);
3119 db_printf("inp_label: %p inp_flags: 0x%x (",
3120 inp->inp_label, inp->inp_flags);
3121 db_print_inpflags(inp->inp_flags);
3124 db_print_indent(indent);
3125 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3127 db_print_inpvflag(inp->inp_vflag);
3130 db_print_indent(indent);
3131 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3132 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3134 db_print_indent(indent);
3136 if (inp->inp_vflag & INP_IPV6) {
3137 db_printf("in6p_options: %p in6p_outputopts: %p "
3138 "in6p_moptions: %p\n", inp->in6p_options,
3139 inp->in6p_outputopts, inp->in6p_moptions);
3140 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3141 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3146 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3147 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3148 inp->inp_options, inp->inp_moptions);
3151 db_print_indent(indent);
3152 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3153 (uintmax_t)inp->inp_gencnt);
3156 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3161 db_printf("usage: show inpcb <addr>\n");
3164 inp = (struct inpcb *)addr;
3166 db_print_inpcb(inp, "inpcb", 0);
3172 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3176 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3178 union if_snd_tag_modify_params params = {
3179 .rate_limit.max_rate = max_pacing_rate,
3181 struct m_snd_tag *mst;
3185 mst = inp->inp_snd_tag;
3193 if (ifp->if_snd_tag_modify == NULL) {
3196 error = ifp->if_snd_tag_modify(mst, ¶ms);
3202 * Query existing TX rate limit based on the existing
3203 * "inp->inp_snd_tag", if any.
3206 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3208 union if_snd_tag_query_params params = { };
3209 struct m_snd_tag *mst;
3213 mst = inp->inp_snd_tag;
3221 if (ifp->if_snd_tag_query == NULL) {
3224 error = ifp->if_snd_tag_query(mst, ¶ms);
3225 if (error == 0 && p_max_pacing_rate != NULL)
3226 *p_max_pacing_rate = params.rate_limit.max_rate;
3232 * Query existing TX queue level based on the existing
3233 * "inp->inp_snd_tag", if any.
3236 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3238 union if_snd_tag_query_params params = { };
3239 struct m_snd_tag *mst;
3243 mst = inp->inp_snd_tag;
3251 if (ifp->if_snd_tag_query == NULL)
3252 return (EOPNOTSUPP);
3254 error = ifp->if_snd_tag_query(mst, ¶ms);
3255 if (error == 0 && p_txqueue_level != NULL)
3256 *p_txqueue_level = params.rate_limit.queue_level;
3261 * Allocate a new TX rate limit send tag from the network interface
3262 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3265 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3266 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3268 union if_snd_tag_alloc_params params = {
3269 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3270 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3271 .rate_limit.hdr.flowid = flowid,
3272 .rate_limit.hdr.flowtype = flowtype,
3273 .rate_limit.max_rate = max_pacing_rate,
3277 INP_WLOCK_ASSERT(inp);
3279 if (inp->inp_snd_tag != NULL)
3282 if (ifp->if_snd_tag_alloc == NULL) {
3285 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3288 * At success increment the refcount on
3289 * the send tag's network interface:
3292 if_ref(inp->inp_snd_tag->ifp);
3298 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3302 in_pcbdetach_txrtlmt(struct inpcb *inp)
3304 struct m_snd_tag *mst;
3307 INP_WLOCK_ASSERT(inp);
3309 mst = inp->inp_snd_tag;
3310 inp->inp_snd_tag = NULL;
3320 * If the device was detached while we still had reference(s)
3321 * on the ifp, we assume if_snd_tag_free() was replaced with
3324 ifp->if_snd_tag_free(mst);
3326 /* release reference count on network interface */
3331 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3332 * is set in the fast path and will attach/detach/modify the TX rate
3333 * limit send tag based on the socket's so_max_pacing_rate value.
3336 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3338 struct socket *socket;
3339 uint32_t max_pacing_rate;
3346 socket = inp->inp_socket;
3350 if (!INP_WLOCKED(inp)) {
3352 * NOTE: If the write locking fails, we need to bail
3353 * out and use the non-ratelimited ring for the
3354 * transmit until there is a new chance to get the
3357 if (!INP_TRY_UPGRADE(inp))
3365 * NOTE: The so_max_pacing_rate value is read unlocked,
3366 * because atomic updates are not required since the variable
3367 * is checked at every mbuf we send. It is assumed that the
3368 * variable read itself will be atomic.
3370 max_pacing_rate = socket->so_max_pacing_rate;
3373 * NOTE: When attaching to a network interface a reference is
3374 * made to ensure the network interface doesn't go away until
3375 * all ratelimit connections are gone. The network interface
3376 * pointers compared below represent valid network interfaces,
3377 * except when comparing towards NULL.
3379 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3381 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3382 if (inp->inp_snd_tag != NULL)
3383 in_pcbdetach_txrtlmt(inp);
3385 } else if (inp->inp_snd_tag == NULL) {
3387 * In order to utilize packet pacing with RSS, we need
3388 * to wait until there is a valid RSS hash before we
3391 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3394 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3395 mb->m_pkthdr.flowid, max_pacing_rate);
3398 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3400 if (error == 0 || error == EOPNOTSUPP)
3401 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3407 * Track route changes for TX rate limiting.
3410 in_pcboutput_eagain(struct inpcb *inp)
3412 struct socket *socket;
3418 socket = inp->inp_socket;
3422 if (inp->inp_snd_tag == NULL)
3425 if (!INP_WLOCKED(inp)) {
3427 * NOTE: If the write locking fails, we need to bail
3428 * out and use the non-ratelimited ring for the
3429 * transmit until there is a new chance to get the
3432 if (!INP_TRY_UPGRADE(inp))
3439 /* detach rate limiting */
3440 in_pcbdetach_txrtlmt(inp);
3442 /* make sure new mbuf send tag allocation is made */
3443 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3448 #endif /* RATELIMIT */