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 struct inpcbinfo *pcbinfo;
305 struct inpcblbgrouphead *hdr;
306 struct inpcblbgroup *grp;
307 uint16_t hashmask, lport;
308 uint32_t group_index;
310 static int limit_logged = 0;
312 pcbinfo = inp->inp_pcbinfo;
314 INP_WLOCK_ASSERT(inp);
315 INP_HASH_WLOCK_ASSERT(pcbinfo);
317 if (pcbinfo->ipi_lbgrouphashbase == NULL)
320 hashmask = pcbinfo->ipi_lbgrouphashmask;
321 lport = inp->inp_lport;
322 group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask);
323 hdr = &pcbinfo->ipi_lbgrouphashbase[group_index];
326 * Don't allow jailed socket to join local group.
328 if (inp->inp_socket != NULL)
329 cred = inp->inp_socket->so_cred;
332 if (cred != NULL && jailed(cred))
337 * Don't allow IPv4 mapped INET6 wild socket.
339 if ((inp->inp_vflag & INP_IPV4) &&
340 inp->inp_laddr.s_addr == INADDR_ANY &&
341 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
346 hdr = &pcbinfo->ipi_lbgrouphashbase[
347 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
348 pcbinfo->ipi_lbgrouphashmask)];
349 LIST_FOREACH(grp, hdr, il_list) {
350 if (grp->il_vflag == inp->inp_vflag &&
351 grp->il_lport == inp->inp_lport &&
352 memcmp(&grp->il_dependladdr,
353 &inp->inp_inc.inc_ie.ie_dependladdr,
354 sizeof(grp->il_dependladdr)) == 0) {
359 /* Create new load balance group. */
360 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
361 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
362 INPCBLBGROUP_SIZMIN);
365 } else if (grp->il_inpcnt == grp->il_inpsiz) {
366 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
369 printf("lb group port %d, limit reached\n",
370 ntohs(grp->il_lport));
375 /* Expand this local group. */
376 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
381 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
382 ("invalid local group size %d and count %d",
383 grp->il_inpsiz, grp->il_inpcnt));
385 grp->il_inp[grp->il_inpcnt] = inp;
391 * Remove PCB from load balance group.
394 in_pcbremlbgrouphash(struct inpcb *inp)
396 struct inpcbinfo *pcbinfo;
397 struct inpcblbgrouphead *hdr;
398 struct inpcblbgroup *grp;
401 pcbinfo = inp->inp_pcbinfo;
403 INP_WLOCK_ASSERT(inp);
404 INP_HASH_WLOCK_ASSERT(pcbinfo);
406 if (pcbinfo->ipi_lbgrouphashbase == NULL)
409 hdr = &pcbinfo->ipi_lbgrouphashbase[
410 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
411 pcbinfo->ipi_lbgrouphashmask)];
413 LIST_FOREACH(grp, hdr, il_list) {
414 for (i = 0; i < grp->il_inpcnt; ++i) {
415 if (grp->il_inp[i] != inp)
418 if (grp->il_inpcnt == 1) {
419 /* We are the last, free this local group. */
420 in_pcblbgroup_free(grp);
422 /* Pull up inpcbs, shrink group if possible. */
423 in_pcblbgroup_reorder(hdr, &grp, i);
431 * Different protocols initialize their inpcbs differently - giving
432 * different name to the lock. But they all are disposed the same.
435 inpcb_fini(void *mem, int size)
437 struct inpcb *inp = mem;
439 INP_LOCK_DESTROY(inp);
443 * Initialize an inpcbinfo -- we should be able to reduce the number of
447 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
448 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
449 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
452 INP_INFO_LOCK_INIT(pcbinfo, name);
453 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
454 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
456 pcbinfo->ipi_vnet = curvnet;
458 pcbinfo->ipi_listhead = listhead;
459 CK_LIST_INIT(pcbinfo->ipi_listhead);
460 pcbinfo->ipi_count = 0;
461 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
462 &pcbinfo->ipi_hashmask);
463 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
464 &pcbinfo->ipi_porthashmask);
465 pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
466 &pcbinfo->ipi_lbgrouphashmask);
468 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
470 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
471 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
472 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
473 uma_zone_set_warning(pcbinfo->ipi_zone,
474 "kern.ipc.maxsockets limit reached");
478 * Destroy an inpcbinfo.
481 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
484 KASSERT(pcbinfo->ipi_count == 0,
485 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
487 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
488 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
489 pcbinfo->ipi_porthashmask);
490 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
491 pcbinfo->ipi_lbgrouphashmask);
493 in_pcbgroup_destroy(pcbinfo);
495 uma_zdestroy(pcbinfo->ipi_zone);
496 INP_LIST_LOCK_DESTROY(pcbinfo);
497 INP_HASH_LOCK_DESTROY(pcbinfo);
498 INP_INFO_LOCK_DESTROY(pcbinfo);
502 * Allocate a PCB and associate it with the socket.
503 * On success return with the PCB locked.
506 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
512 if (pcbinfo == &V_tcbinfo) {
513 INP_INFO_RLOCK_ASSERT(pcbinfo);
515 INP_INFO_WLOCK_ASSERT(pcbinfo);
520 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
523 bzero(&inp->inp_start_zero, inp_zero_size);
524 inp->inp_pcbinfo = pcbinfo;
525 inp->inp_socket = so;
526 inp->inp_cred = crhold(so->so_cred);
527 inp->inp_inc.inc_fibnum = so->so_fibnum;
529 error = mac_inpcb_init(inp, M_NOWAIT);
532 mac_inpcb_create(so, inp);
534 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
535 error = ipsec_init_pcbpolicy(inp);
538 mac_inpcb_destroy(inp);
544 if (INP_SOCKAF(so) == AF_INET6) {
545 inp->inp_vflag |= INP_IPV6PROTO;
547 inp->inp_flags |= IN6P_IPV6_V6ONLY;
551 INP_LIST_WLOCK(pcbinfo);
552 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
553 pcbinfo->ipi_count++;
554 so->so_pcb = (caddr_t)inp;
556 if (V_ip6_auto_flowlabel)
557 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
559 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
560 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
563 * Routes in inpcb's can cache L2 as well; they are guaranteed
566 inp->inp_route.ro_flags = RT_LLE_CACHE;
567 INP_LIST_WUNLOCK(pcbinfo);
568 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
571 crfree(inp->inp_cred);
572 uma_zfree(pcbinfo->ipi_zone, inp);
580 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
584 INP_WLOCK_ASSERT(inp);
585 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
587 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
589 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
590 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
591 &inp->inp_lport, cred);
594 if (in_pcbinshash(inp) != 0) {
595 inp->inp_laddr.s_addr = INADDR_ANY;
600 inp->inp_flags |= INP_ANONPORT;
606 * Select a local port (number) to use.
608 #if defined(INET) || defined(INET6)
610 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
611 struct ucred *cred, int lookupflags)
613 struct inpcbinfo *pcbinfo;
614 struct inpcb *tmpinp;
615 unsigned short *lastport;
616 int count, dorandom, error;
617 u_short aux, first, last, lport;
619 struct in_addr laddr;
622 pcbinfo = inp->inp_pcbinfo;
625 * Because no actual state changes occur here, a global write lock on
626 * the pcbinfo isn't required.
628 INP_LOCK_ASSERT(inp);
629 INP_HASH_LOCK_ASSERT(pcbinfo);
631 if (inp->inp_flags & INP_HIGHPORT) {
632 first = V_ipport_hifirstauto; /* sysctl */
633 last = V_ipport_hilastauto;
634 lastport = &pcbinfo->ipi_lasthi;
635 } else if (inp->inp_flags & INP_LOWPORT) {
636 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
639 first = V_ipport_lowfirstauto; /* 1023 */
640 last = V_ipport_lowlastauto; /* 600 */
641 lastport = &pcbinfo->ipi_lastlow;
643 first = V_ipport_firstauto; /* sysctl */
644 last = V_ipport_lastauto;
645 lastport = &pcbinfo->ipi_lastport;
648 * For UDP(-Lite), use random port allocation as long as the user
649 * allows it. For TCP (and as of yet unknown) connections,
650 * use random port allocation only if the user allows it AND
651 * ipport_tick() allows it.
653 if (V_ipport_randomized &&
654 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
655 pcbinfo == &V_ulitecbinfo))
660 * It makes no sense to do random port allocation if
661 * we have the only port available.
665 /* Make sure to not include UDP(-Lite) packets in the count. */
666 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
667 V_ipport_tcpallocs++;
669 * Instead of having two loops further down counting up or down
670 * make sure that first is always <= last and go with only one
671 * code path implementing all logic.
680 /* Make the compiler happy. */
682 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
683 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
688 tmpinp = NULL; /* Make compiler happy. */
692 *lastport = first + (arc4random() % (last - first));
694 count = last - first;
697 if (count-- < 0) /* completely used? */
698 return (EADDRNOTAVAIL);
700 if (*lastport < first || *lastport > last)
702 lport = htons(*lastport);
705 if ((inp->inp_vflag & INP_IPV6) != 0)
706 tmpinp = in6_pcblookup_local(pcbinfo,
707 &inp->in6p_laddr, lport, lookupflags, cred);
709 #if defined(INET) && defined(INET6)
713 tmpinp = in_pcblookup_local(pcbinfo, laddr,
714 lport, lookupflags, cred);
716 } while (tmpinp != NULL);
719 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
720 laddrp->s_addr = laddr.s_addr;
728 * Return cached socket options.
731 inp_so_options(const struct inpcb *inp)
737 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
738 so_options |= SO_REUSEPORT_LB;
739 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
740 so_options |= SO_REUSEPORT;
741 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
742 so_options |= SO_REUSEADDR;
745 #endif /* INET || INET6 */
748 * Check if a new BINDMULTI socket is allowed to be created.
750 * ni points to the new inp.
751 * oi points to the exisitng inp.
753 * This checks whether the existing inp also has BINDMULTI and
754 * whether the credentials match.
757 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
759 /* Check permissions match */
760 if ((ni->inp_flags2 & INP_BINDMULTI) &&
761 (ni->inp_cred->cr_uid !=
762 oi->inp_cred->cr_uid))
765 /* Check the existing inp has BINDMULTI set */
766 if ((ni->inp_flags2 & INP_BINDMULTI) &&
767 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
771 * We're okay - either INP_BINDMULTI isn't set on ni, or
772 * it is and it matches the checks.
779 * Set up a bind operation on a PCB, performing port allocation
780 * as required, but do not actually modify the PCB. Callers can
781 * either complete the bind by setting inp_laddr/inp_lport and
782 * calling in_pcbinshash(), or they can just use the resulting
783 * port and address to authorise the sending of a once-off packet.
785 * On error, the values of *laddrp and *lportp are not changed.
788 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
789 u_short *lportp, struct ucred *cred)
791 struct socket *so = inp->inp_socket;
792 struct sockaddr_in *sin;
793 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
794 struct in_addr laddr;
796 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
800 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
801 * so that we don't have to add to the (already messy) code below.
803 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
806 * No state changes, so read locks are sufficient here.
808 INP_LOCK_ASSERT(inp);
809 INP_HASH_LOCK_ASSERT(pcbinfo);
811 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
812 return (EADDRNOTAVAIL);
813 laddr.s_addr = *laddrp;
814 if (nam != NULL && laddr.s_addr != INADDR_ANY)
816 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
817 lookupflags = INPLOOKUP_WILDCARD;
819 if ((error = prison_local_ip4(cred, &laddr)) != 0)
822 sin = (struct sockaddr_in *)nam;
823 if (nam->sa_len != sizeof (*sin))
827 * We should check the family, but old programs
828 * incorrectly fail to initialize it.
830 if (sin->sin_family != AF_INET)
831 return (EAFNOSUPPORT);
833 error = prison_local_ip4(cred, &sin->sin_addr);
836 if (sin->sin_port != *lportp) {
837 /* Don't allow the port to change. */
840 lport = sin->sin_port;
842 /* NB: lport is left as 0 if the port isn't being changed. */
843 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
845 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
846 * allow complete duplication of binding if
847 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
848 * and a multicast address is bound on both
849 * new and duplicated sockets.
851 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
852 reuseport = SO_REUSEADDR|SO_REUSEPORT;
854 * XXX: How to deal with SO_REUSEPORT_LB here?
855 * Treat same as SO_REUSEPORT for now.
857 if ((so->so_options &
858 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
859 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
860 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
861 sin->sin_port = 0; /* yech... */
862 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
864 * Is the address a local IP address?
865 * If INP_BINDANY is set, then the socket may be bound
866 * to any endpoint address, local or not.
868 if ((inp->inp_flags & INP_BINDANY) == 0 &&
869 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
870 return (EADDRNOTAVAIL);
872 laddr = sin->sin_addr;
878 if (ntohs(lport) <= V_ipport_reservedhigh &&
879 ntohs(lport) >= V_ipport_reservedlow &&
880 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
883 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
884 priv_check_cred(inp->inp_cred,
885 PRIV_NETINET_REUSEPORT, 0) != 0) {
886 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
887 lport, INPLOOKUP_WILDCARD, cred);
890 * This entire block sorely needs a rewrite.
893 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
894 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
895 (so->so_type != SOCK_STREAM ||
896 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
897 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
898 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
899 (t->inp_flags2 & INP_REUSEPORT) ||
900 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
901 (inp->inp_cred->cr_uid !=
902 t->inp_cred->cr_uid))
906 * If the socket is a BINDMULTI socket, then
907 * the credentials need to match and the
908 * original socket also has to have been bound
911 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
914 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
915 lport, lookupflags, cred);
916 if (t && (t->inp_flags & INP_TIMEWAIT)) {
918 * XXXRW: If an incpb has had its timewait
919 * state recycled, we treat the address as
920 * being in use (for now). This is better
921 * than a panic, but not desirable.
925 ((reuseport & tw->tw_so_options) == 0 &&
927 tw->tw_so_options) == 0)) {
931 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
932 (reuseport & inp_so_options(t)) == 0 &&
933 (reuseport_lb & inp_so_options(t)) == 0) {
935 if (ntohl(sin->sin_addr.s_addr) !=
937 ntohl(t->inp_laddr.s_addr) !=
939 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
940 (t->inp_vflag & INP_IPV6PROTO) == 0)
943 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
951 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
956 *laddrp = laddr.s_addr;
962 * Connect from a socket to a specified address.
963 * Both address and port must be specified in argument sin.
964 * If don't have a local address for this socket yet,
968 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
969 struct ucred *cred, struct mbuf *m)
971 u_short lport, fport;
972 in_addr_t laddr, faddr;
975 INP_WLOCK_ASSERT(inp);
976 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
978 lport = inp->inp_lport;
979 laddr = inp->inp_laddr.s_addr;
980 anonport = (lport == 0);
981 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
986 /* Do the initial binding of the local address if required. */
987 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
988 inp->inp_lport = lport;
989 inp->inp_laddr.s_addr = laddr;
990 if (in_pcbinshash(inp) != 0) {
991 inp->inp_laddr.s_addr = INADDR_ANY;
997 /* Commit the remaining changes. */
998 inp->inp_lport = lport;
999 inp->inp_laddr.s_addr = laddr;
1000 inp->inp_faddr.s_addr = faddr;
1001 inp->inp_fport = fport;
1002 in_pcbrehash_mbuf(inp, m);
1005 inp->inp_flags |= INP_ANONPORT;
1010 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1013 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1017 * Do proper source address selection on an unbound socket in case
1018 * of connect. Take jails into account as well.
1021 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1025 struct sockaddr *sa;
1026 struct sockaddr_in *sin;
1030 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1032 * Bypass source address selection and use the primary jail IP
1035 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1039 bzero(&sro, sizeof(sro));
1041 sin = (struct sockaddr_in *)&sro.ro_dst;
1042 sin->sin_family = AF_INET;
1043 sin->sin_len = sizeof(struct sockaddr_in);
1044 sin->sin_addr.s_addr = faddr->s_addr;
1047 * If route is known our src addr is taken from the i/f,
1050 * Find out route to destination.
1052 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1053 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1056 * If we found a route, use the address corresponding to
1057 * the outgoing interface.
1059 * Otherwise assume faddr is reachable on a directly connected
1060 * network and try to find a corresponding interface to take
1061 * the source address from.
1064 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1065 struct in_ifaddr *ia;
1068 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1069 inp->inp_socket->so_fibnum));
1071 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1072 inp->inp_socket->so_fibnum));
1076 error = ENETUNREACH;
1080 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1081 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1087 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1090 if (sa->sa_family != AF_INET)
1092 sin = (struct sockaddr_in *)sa;
1093 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1094 ia = (struct in_ifaddr *)ifa;
1099 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1103 /* 3. As a last resort return the 'default' jail address. */
1104 error = prison_get_ip4(cred, laddr);
1109 * If the outgoing interface on the route found is not
1110 * a loopback interface, use the address from that interface.
1111 * In case of jails do those three steps:
1112 * 1. check if the interface address belongs to the jail. If so use it.
1113 * 2. check if we have any address on the outgoing interface
1114 * belonging to this jail. If so use it.
1115 * 3. as a last resort return the 'default' jail address.
1117 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1118 struct in_ifaddr *ia;
1121 /* If not jailed, use the default returned. */
1122 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1123 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1124 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1129 /* 1. Check if the iface address belongs to the jail. */
1130 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1131 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1132 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1133 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1138 * 2. Check if we have any address on the outgoing interface
1139 * belonging to this jail.
1142 ifp = sro.ro_rt->rt_ifp;
1143 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1145 if (sa->sa_family != AF_INET)
1147 sin = (struct sockaddr_in *)sa;
1148 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1149 ia = (struct in_ifaddr *)ifa;
1154 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1158 /* 3. As a last resort return the 'default' jail address. */
1159 error = prison_get_ip4(cred, laddr);
1164 * The outgoing interface is marked with 'loopback net', so a route
1165 * to ourselves is here.
1166 * Try to find the interface of the destination address and then
1167 * take the address from there. That interface is not necessarily
1168 * a loopback interface.
1169 * In case of jails, check that it is an address of the jail
1170 * and if we cannot find, fall back to the 'default' jail address.
1172 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1173 struct sockaddr_in sain;
1174 struct in_ifaddr *ia;
1176 bzero(&sain, sizeof(struct sockaddr_in));
1177 sain.sin_family = AF_INET;
1178 sain.sin_len = sizeof(struct sockaddr_in);
1179 sain.sin_addr.s_addr = faddr->s_addr;
1181 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1182 inp->inp_socket->so_fibnum));
1184 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1185 inp->inp_socket->so_fibnum));
1187 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1189 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1191 error = ENETUNREACH;
1194 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1204 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1206 if (sa->sa_family != AF_INET)
1208 sin = (struct sockaddr_in *)sa;
1209 if (prison_check_ip4(cred,
1210 &sin->sin_addr) == 0) {
1211 ia = (struct in_ifaddr *)ifa;
1216 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1221 /* 3. As a last resort return the 'default' jail address. */
1222 error = prison_get_ip4(cred, laddr);
1228 if (sro.ro_rt != NULL)
1234 * Set up for a connect from a socket to the specified address.
1235 * On entry, *laddrp and *lportp should contain the current local
1236 * address and port for the PCB; these are updated to the values
1237 * that should be placed in inp_laddr and inp_lport to complete
1240 * On success, *faddrp and *fportp will be set to the remote address
1241 * and port. These are not updated in the error case.
1243 * If the operation fails because the connection already exists,
1244 * *oinpp will be set to the PCB of that connection so that the
1245 * caller can decide to override it. In all other cases, *oinpp
1249 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1250 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1251 struct inpcb **oinpp, struct ucred *cred)
1253 struct rm_priotracker in_ifa_tracker;
1254 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1255 struct in_ifaddr *ia;
1257 struct in_addr laddr, faddr;
1258 u_short lport, fport;
1262 * Because a global state change doesn't actually occur here, a read
1263 * lock is sufficient.
1265 INP_LOCK_ASSERT(inp);
1266 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1270 if (nam->sa_len != sizeof (*sin))
1272 if (sin->sin_family != AF_INET)
1273 return (EAFNOSUPPORT);
1274 if (sin->sin_port == 0)
1275 return (EADDRNOTAVAIL);
1276 laddr.s_addr = *laddrp;
1278 faddr = sin->sin_addr;
1279 fport = sin->sin_port;
1281 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1283 * If the destination address is INADDR_ANY,
1284 * use the primary local address.
1285 * If the supplied address is INADDR_BROADCAST,
1286 * and the primary interface supports broadcast,
1287 * choose the broadcast address for that interface.
1289 if (faddr.s_addr == INADDR_ANY) {
1290 IN_IFADDR_RLOCK(&in_ifa_tracker);
1292 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1293 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1295 (error = prison_get_ip4(cred, &faddr)) != 0)
1297 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1298 IN_IFADDR_RLOCK(&in_ifa_tracker);
1299 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1301 faddr = satosin(&CK_STAILQ_FIRST(
1302 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1303 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1306 if (laddr.s_addr == INADDR_ANY) {
1307 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1309 * If the destination address is multicast and an outgoing
1310 * interface has been set as a multicast option, prefer the
1311 * address of that interface as our source address.
1313 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1314 inp->inp_moptions != NULL) {
1315 struct ip_moptions *imo;
1318 imo = inp->inp_moptions;
1319 if (imo->imo_multicast_ifp != NULL) {
1320 ifp = imo->imo_multicast_ifp;
1321 IN_IFADDR_RLOCK(&in_ifa_tracker);
1322 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1323 if ((ia->ia_ifp == ifp) &&
1325 prison_check_ip4(cred,
1326 &ia->ia_addr.sin_addr) == 0))
1330 error = EADDRNOTAVAIL;
1332 laddr = ia->ia_addr.sin_addr;
1335 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1341 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1342 laddr, lport, 0, NULL);
1346 return (EADDRINUSE);
1349 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1354 *laddrp = laddr.s_addr;
1356 *faddrp = faddr.s_addr;
1362 in_pcbdisconnect(struct inpcb *inp)
1365 INP_WLOCK_ASSERT(inp);
1366 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1368 inp->inp_faddr.s_addr = INADDR_ANY;
1375 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1376 * For most protocols, this will be invoked immediately prior to calling
1377 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1378 * socket, in which case in_pcbfree() is deferred.
1381 in_pcbdetach(struct inpcb *inp)
1384 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1387 if (inp->inp_snd_tag != NULL)
1388 in_pcbdetach_txrtlmt(inp);
1390 inp->inp_socket->so_pcb = NULL;
1391 inp->inp_socket = NULL;
1395 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1396 * stability of an inpcb pointer despite the inpcb lock being released. This
1397 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1398 * but where the inpcb lock may already held, or when acquiring a reference
1401 * in_pcbref() should be used only to provide brief memory stability, and
1402 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1403 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1404 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1405 * lock and rele are the *only* safe operations that may be performed on the
1408 * While the inpcb will not be freed, releasing the inpcb lock means that the
1409 * connection's state may change, so the caller should be careful to
1410 * revalidate any cached state on reacquiring the lock. Drop the reference
1411 * using in_pcbrele().
1414 in_pcbref(struct inpcb *inp)
1417 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1419 refcount_acquire(&inp->inp_refcount);
1423 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1424 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1425 * return a flag indicating whether or not the inpcb remains valid. If it is
1426 * valid, we return with the inpcb lock held.
1428 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1429 * reference on an inpcb. Historically more work was done here (actually, in
1430 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1431 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1432 * about memory stability (and continued use of the write lock).
1435 in_pcbrele_rlocked(struct inpcb *inp)
1437 struct inpcbinfo *pcbinfo;
1439 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1441 INP_RLOCK_ASSERT(inp);
1443 if (refcount_release(&inp->inp_refcount) == 0) {
1445 * If the inpcb has been freed, let the caller know, even if
1446 * this isn't the last reference.
1448 if (inp->inp_flags2 & INP_FREED) {
1455 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1457 if (inp->inp_in_hpts || inp->inp_in_input) {
1458 struct tcp_hpts_entry *hpts;
1460 * We should not be on the hpts at
1461 * this point in any form. we must
1462 * get the lock to be sure.
1464 hpts = tcp_hpts_lock(inp);
1465 if (inp->inp_in_hpts)
1466 panic("Hpts:%p inp:%p at free still on hpts",
1468 mtx_unlock(&hpts->p_mtx);
1469 hpts = tcp_input_lock(inp);
1470 if (inp->inp_in_input)
1471 panic("Hpts:%p inp:%p at free still on input hpts",
1473 mtx_unlock(&hpts->p_mtx);
1477 pcbinfo = inp->inp_pcbinfo;
1478 uma_zfree(pcbinfo->ipi_zone, inp);
1483 in_pcbrele_wlocked(struct inpcb *inp)
1485 struct inpcbinfo *pcbinfo;
1487 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1489 INP_WLOCK_ASSERT(inp);
1491 if (refcount_release(&inp->inp_refcount) == 0) {
1493 * If the inpcb has been freed, let the caller know, even if
1494 * this isn't the last reference.
1496 if (inp->inp_flags2 & INP_FREED) {
1503 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1505 if (inp->inp_in_hpts || inp->inp_in_input) {
1506 struct tcp_hpts_entry *hpts;
1508 * We should not be on the hpts at
1509 * this point in any form. we must
1510 * get the lock to be sure.
1512 hpts = tcp_hpts_lock(inp);
1513 if (inp->inp_in_hpts)
1514 panic("Hpts:%p inp:%p at free still on hpts",
1516 mtx_unlock(&hpts->p_mtx);
1517 hpts = tcp_input_lock(inp);
1518 if (inp->inp_in_input)
1519 panic("Hpts:%p inp:%p at free still on input hpts",
1521 mtx_unlock(&hpts->p_mtx);
1525 pcbinfo = inp->inp_pcbinfo;
1526 uma_zfree(pcbinfo->ipi_zone, inp);
1531 * Temporary wrapper.
1534 in_pcbrele(struct inpcb *inp)
1537 return (in_pcbrele_wlocked(inp));
1541 in_pcblist_rele_rlocked(epoch_context_t ctx)
1543 struct in_pcblist *il;
1545 struct inpcbinfo *pcbinfo;
1548 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1549 pcbinfo = il->il_pcbinfo;
1551 INP_INFO_WLOCK(pcbinfo);
1552 for (i = 0; i < n; i++) {
1553 inp = il->il_inp_list[i];
1555 if (!in_pcbrele_rlocked(inp))
1558 INP_INFO_WUNLOCK(pcbinfo);
1563 inpcbport_free(epoch_context_t ctx)
1565 struct inpcbport *phd;
1567 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1572 in_pcbfree_deferred(epoch_context_t ctx)
1575 int released __unused;
1577 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1581 inp_freemoptions(inp->inp_moptions);
1582 inp->inp_moptions = NULL;
1584 /* XXXRW: Do as much as possible here. */
1585 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1586 if (inp->inp_sp != NULL)
1587 ipsec_delete_pcbpolicy(inp);
1590 if (inp->inp_vflag & INP_IPV6PROTO) {
1591 ip6_freepcbopts(inp->in6p_outputopts);
1592 ip6_freemoptions(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);
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 = NULL;
1950 const struct inpcblbgrouphead *hdr;
1951 struct inpcblbgroup *grp;
1952 struct inpcblbgroup *grp_local_wild;
1954 INP_HASH_LOCK_ASSERT(pcbinfo);
1956 hdr = &pcbinfo->ipi_lbgrouphashbase[
1957 INP_PCBLBGROUP_PORTHASH(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
1968 LIST_FOREACH(grp, hdr, il_list) {
1970 if (!(grp->il_vflag & INP_IPV4))
1974 if (grp->il_lport == lport) {
1977 int pkt_hash = INP_PCBLBGROUP_PKTHASH(faddr->s_addr,
1980 idx = pkt_hash % grp->il_inpcnt;
1982 if (grp->il_laddr.s_addr == laddr->s_addr) {
1983 return (grp->il_inp[idx]);
1985 if (grp->il_laddr.s_addr == INADDR_ANY &&
1986 (lookupflags & INPLOOKUP_WILDCARD)) {
1987 local_wild = grp->il_inp[idx];
1988 grp_local_wild = grp;
1993 if (local_wild != NULL) {
1994 return (local_wild);
2001 * Lookup PCB in hash list, using pcbgroup tables.
2003 static struct inpcb *
2004 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2005 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2006 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2008 struct inpcbhead *head;
2009 struct inpcb *inp, *tmpinp;
2010 u_short fport = fport_arg, lport = lport_arg;
2014 * First look for an exact match.
2017 INP_GROUP_LOCK(pcbgroup);
2018 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2019 pcbgroup->ipg_hashmask)];
2020 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2022 /* XXX inp locking */
2023 if ((inp->inp_vflag & INP_IPV4) == 0)
2026 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2027 inp->inp_laddr.s_addr == laddr.s_addr &&
2028 inp->inp_fport == fport &&
2029 inp->inp_lport == lport) {
2031 * XXX We should be able to directly return
2032 * the inp here, without any checks.
2033 * Well unless both bound with SO_REUSEPORT?
2035 if (prison_flag(inp->inp_cred, PR_IP4))
2041 if (tmpinp != NULL) {
2048 * For incoming connections, we may wish to do a wildcard
2049 * match for an RSS-local socket.
2051 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2052 struct inpcb *local_wild = NULL, *local_exact = NULL;
2054 struct inpcb *local_wild_mapped = NULL;
2056 struct inpcb *jail_wild = NULL;
2057 struct inpcbhead *head;
2061 * Order of socket selection - we always prefer jails.
2062 * 1. jailed, non-wild.
2064 * 3. non-jailed, non-wild.
2065 * 4. non-jailed, wild.
2068 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2069 lport, 0, pcbgroup->ipg_hashmask)];
2070 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2072 /* XXX inp locking */
2073 if ((inp->inp_vflag & INP_IPV4) == 0)
2076 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2077 inp->inp_lport != lport)
2080 injail = prison_flag(inp->inp_cred, PR_IP4);
2082 if (prison_check_ip4(inp->inp_cred,
2086 if (local_exact != NULL)
2090 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2095 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2097 /* XXX inp locking, NULL check */
2098 if (inp->inp_vflag & INP_IPV6PROTO)
2099 local_wild_mapped = inp;
2107 } /* LIST_FOREACH */
2116 inp = local_wild_mapped;
2124 * Then look for a wildcard match, if requested.
2126 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2127 struct inpcb *local_wild = NULL, *local_exact = NULL;
2129 struct inpcb *local_wild_mapped = NULL;
2131 struct inpcb *jail_wild = NULL;
2132 struct inpcbhead *head;
2136 * Order of socket selection - we always prefer jails.
2137 * 1. jailed, non-wild.
2139 * 3. non-jailed, non-wild.
2140 * 4. non-jailed, wild.
2142 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2143 0, pcbinfo->ipi_wildmask)];
2144 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2146 /* XXX inp locking */
2147 if ((inp->inp_vflag & INP_IPV4) == 0)
2150 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2151 inp->inp_lport != lport)
2154 injail = prison_flag(inp->inp_cred, PR_IP4);
2156 if (prison_check_ip4(inp->inp_cred,
2160 if (local_exact != NULL)
2164 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2169 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2171 /* XXX inp locking, NULL check */
2172 if (inp->inp_vflag & INP_IPV6PROTO)
2173 local_wild_mapped = inp;
2181 } /* LIST_FOREACH */
2189 inp = local_wild_mapped;
2193 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2194 INP_GROUP_UNLOCK(pcbgroup);
2198 if (lookupflags & INPLOOKUP_WLOCKPCB)
2199 locked = INP_TRY_WLOCK(inp);
2200 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2201 locked = INP_TRY_RLOCK(inp);
2203 panic("%s: locking bug", __func__);
2204 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2205 if (lookupflags & INPLOOKUP_WLOCKPCB)
2212 INP_GROUP_UNLOCK(pcbgroup);
2214 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2216 if (in_pcbrele_wlocked(inp))
2220 if (in_pcbrele_rlocked(inp))
2225 if (lookupflags & INPLOOKUP_WLOCKPCB)
2226 INP_WLOCK_ASSERT(inp);
2228 INP_RLOCK_ASSERT(inp);
2232 #endif /* PCBGROUP */
2235 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2236 * that the caller has locked the hash list, and will not perform any further
2237 * locking or reference operations on either the hash list or the connection.
2239 static struct inpcb *
2240 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2241 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2244 struct inpcbhead *head;
2245 struct inpcb *inp, *tmpinp;
2246 u_short fport = fport_arg, lport = lport_arg;
2249 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2250 ("%s: invalid lookup flags %d", __func__, lookupflags));
2251 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2252 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2255 * First look for an exact match.
2258 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2259 pcbinfo->ipi_hashmask)];
2260 CK_LIST_FOREACH(inp, head, inp_hash) {
2262 /* XXX inp locking */
2263 if ((inp->inp_vflag & INP_IPV4) == 0)
2266 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2267 inp->inp_laddr.s_addr == laddr.s_addr &&
2268 inp->inp_fport == fport &&
2269 inp->inp_lport == lport) {
2271 * XXX We should be able to directly return
2272 * the inp here, without any checks.
2273 * Well unless both bound with SO_REUSEPORT?
2275 if (prison_flag(inp->inp_cred, PR_IP4))
2285 * Then look in lb group (for wildcard match).
2287 if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2288 (lookupflags & INPLOOKUP_WILDCARD)) {
2289 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2290 fport, lookupflags);
2297 * Then look for a wildcard match, if requested.
2299 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2300 struct inpcb *local_wild = NULL, *local_exact = NULL;
2302 struct inpcb *local_wild_mapped = NULL;
2304 struct inpcb *jail_wild = NULL;
2308 * Order of socket selection - we always prefer jails.
2309 * 1. jailed, non-wild.
2311 * 3. non-jailed, non-wild.
2312 * 4. non-jailed, wild.
2315 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2316 0, pcbinfo->ipi_hashmask)];
2317 CK_LIST_FOREACH(inp, head, inp_hash) {
2319 /* XXX inp locking */
2320 if ((inp->inp_vflag & INP_IPV4) == 0)
2323 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2324 inp->inp_lport != lport)
2327 injail = prison_flag(inp->inp_cred, PR_IP4);
2329 if (prison_check_ip4(inp->inp_cred,
2333 if (local_exact != NULL)
2337 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2342 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2344 /* XXX inp locking, NULL check */
2345 if (inp->inp_vflag & INP_IPV6PROTO)
2346 local_wild_mapped = inp;
2354 } /* LIST_FOREACH */
2355 if (jail_wild != NULL)
2357 if (local_exact != NULL)
2358 return (local_exact);
2359 if (local_wild != NULL)
2360 return (local_wild);
2362 if (local_wild_mapped != NULL)
2363 return (local_wild_mapped);
2365 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2371 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2372 * hash list lock, and will return the inpcb locked (i.e., requires
2373 * INPLOOKUP_LOCKPCB).
2375 static struct inpcb *
2376 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2377 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2382 INP_HASH_RLOCK(pcbinfo);
2383 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2384 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2386 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2388 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2392 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2394 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2399 panic("%s: locking bug", __func__);
2402 if (lookupflags & INPLOOKUP_WLOCKPCB)
2403 INP_WLOCK_ASSERT(inp);
2405 INP_RLOCK_ASSERT(inp);
2409 INP_HASH_RUNLOCK(pcbinfo);
2414 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2415 * from which a pre-calculated hash value may be extracted.
2417 * Possibly more of this logic should be in in_pcbgroup.c.
2420 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2421 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2423 #if defined(PCBGROUP) && !defined(RSS)
2424 struct inpcbgroup *pcbgroup;
2427 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2428 ("%s: invalid lookup flags %d", __func__, lookupflags));
2429 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2430 ("%s: LOCKPCB not set", __func__));
2433 * When not using RSS, use connection groups in preference to the
2434 * reservation table when looking up 4-tuples. When using RSS, just
2435 * use the reservation table, due to the cost of the Toeplitz hash
2438 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2439 * we could be doing RSS with a non-Toeplitz hash that is affordable
2442 #if defined(PCBGROUP) && !defined(RSS)
2443 if (in_pcbgroup_enabled(pcbinfo)) {
2444 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2446 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2447 laddr, lport, lookupflags, ifp));
2450 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2455 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2456 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2457 struct ifnet *ifp, struct mbuf *m)
2460 struct inpcbgroup *pcbgroup;
2463 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2464 ("%s: invalid lookup flags %d", __func__, lookupflags));
2465 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2466 ("%s: LOCKPCB not set", __func__));
2470 * If we can use a hardware-generated hash to look up the connection
2471 * group, use that connection group to find the inpcb. Otherwise
2472 * fall back on a software hash -- or the reservation table if we're
2475 * XXXRW: As above, that policy belongs in the pcbgroup code.
2477 if (in_pcbgroup_enabled(pcbinfo) &&
2478 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2479 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2480 m->m_pkthdr.flowid);
2481 if (pcbgroup != NULL)
2482 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2483 fport, laddr, lport, lookupflags, ifp));
2485 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2487 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2488 laddr, lport, lookupflags, ifp));
2492 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2498 * Insert PCB onto various hash lists.
2501 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2503 struct inpcbhead *pcbhash;
2504 struct inpcbporthead *pcbporthash;
2505 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2506 struct inpcbport *phd;
2507 u_int32_t hashkey_faddr;
2510 INP_WLOCK_ASSERT(inp);
2511 INP_HASH_WLOCK_ASSERT(pcbinfo);
2513 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2514 ("in_pcbinshash: INP_INHASHLIST"));
2517 if (inp->inp_vflag & INP_IPV6)
2518 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2521 hashkey_faddr = inp->inp_faddr.s_addr;
2523 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2524 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2526 pcbporthash = &pcbinfo->ipi_porthashbase[
2527 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2530 * Add entry to load balance group.
2531 * Only do this if SO_REUSEPORT_LB is set.
2533 so_options = inp_so_options(inp);
2534 if (so_options & SO_REUSEPORT_LB) {
2535 int ret = in_pcbinslbgrouphash(inp);
2537 /* pcb lb group malloc fail (ret=ENOBUFS). */
2543 * Go through port list and look for a head for this lport.
2545 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2546 if (phd->phd_port == inp->inp_lport)
2550 * If none exists, malloc one and tack it on.
2553 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2555 return (ENOBUFS); /* XXX */
2557 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2558 phd->phd_port = inp->inp_lport;
2559 CK_LIST_INIT(&phd->phd_pcblist);
2560 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2563 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2564 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2565 inp->inp_flags |= INP_INHASHLIST;
2567 if (do_pcbgroup_update)
2568 in_pcbgroup_update(inp);
2574 * For now, there are two public interfaces to insert an inpcb into the hash
2575 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2576 * is used only in the TCP syncache, where in_pcbinshash is called before the
2577 * full 4-tuple is set for the inpcb, and we don't want to install in the
2578 * pcbgroup until later.
2580 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2581 * connection groups, and partially initialised inpcbs should not be exposed
2582 * to either reservation hash tables or pcbgroups.
2585 in_pcbinshash(struct inpcb *inp)
2588 return (in_pcbinshash_internal(inp, 1));
2592 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2595 return (in_pcbinshash_internal(inp, 0));
2599 * Move PCB to the proper hash bucket when { faddr, fport } have been
2600 * changed. NOTE: This does not handle the case of the lport changing (the
2601 * hashed port list would have to be updated as well), so the lport must
2602 * not change after in_pcbinshash() has been called.
2605 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2607 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2608 struct inpcbhead *head;
2609 u_int32_t hashkey_faddr;
2611 INP_WLOCK_ASSERT(inp);
2612 INP_HASH_WLOCK_ASSERT(pcbinfo);
2614 KASSERT(inp->inp_flags & INP_INHASHLIST,
2615 ("in_pcbrehash: !INP_INHASHLIST"));
2618 if (inp->inp_vflag & INP_IPV6)
2619 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2622 hashkey_faddr = inp->inp_faddr.s_addr;
2624 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2625 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2627 CK_LIST_REMOVE(inp, inp_hash);
2628 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2632 in_pcbgroup_update_mbuf(inp, m);
2634 in_pcbgroup_update(inp);
2639 in_pcbrehash(struct inpcb *inp)
2642 in_pcbrehash_mbuf(inp, NULL);
2646 * Remove PCB from various lists.
2649 in_pcbremlists(struct inpcb *inp)
2651 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2654 if (pcbinfo == &V_tcbinfo) {
2655 INP_INFO_RLOCK_ASSERT(pcbinfo);
2657 INP_INFO_WLOCK_ASSERT(pcbinfo);
2661 INP_WLOCK_ASSERT(inp);
2662 INP_LIST_WLOCK_ASSERT(pcbinfo);
2664 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2665 if (inp->inp_flags & INP_INHASHLIST) {
2666 struct inpcbport *phd = inp->inp_phd;
2668 INP_HASH_WLOCK(pcbinfo);
2670 /* XXX: Only do if SO_REUSEPORT_LB set? */
2671 in_pcbremlbgrouphash(inp);
2673 CK_LIST_REMOVE(inp, inp_hash);
2674 CK_LIST_REMOVE(inp, inp_portlist);
2675 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2676 CK_LIST_REMOVE(phd, phd_hash);
2677 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2679 INP_HASH_WUNLOCK(pcbinfo);
2680 inp->inp_flags &= ~INP_INHASHLIST;
2682 CK_LIST_REMOVE(inp, inp_list);
2683 pcbinfo->ipi_count--;
2685 in_pcbgroup_remove(inp);
2690 * Check for alternatives when higher level complains
2691 * about service problems. For now, invalidate cached
2692 * routing information. If the route was created dynamically
2693 * (by a redirect), time to try a default gateway again.
2696 in_losing(struct inpcb *inp)
2699 RO_INVALIDATE_CACHE(&inp->inp_route);
2704 * A set label operation has occurred at the socket layer, propagate the
2705 * label change into the in_pcb for the socket.
2708 in_pcbsosetlabel(struct socket *so)
2713 inp = sotoinpcb(so);
2714 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2718 mac_inpcb_sosetlabel(so, inp);
2725 * ipport_tick runs once per second, determining if random port allocation
2726 * should be continued. If more than ipport_randomcps ports have been
2727 * allocated in the last second, then we return to sequential port
2728 * allocation. We return to random allocation only once we drop below
2729 * ipport_randomcps for at least ipport_randomtime seconds.
2732 ipport_tick(void *xtp)
2734 VNET_ITERATOR_DECL(vnet_iter);
2736 VNET_LIST_RLOCK_NOSLEEP();
2737 VNET_FOREACH(vnet_iter) {
2738 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2739 if (V_ipport_tcpallocs <=
2740 V_ipport_tcplastcount + V_ipport_randomcps) {
2741 if (V_ipport_stoprandom > 0)
2742 V_ipport_stoprandom--;
2744 V_ipport_stoprandom = V_ipport_randomtime;
2745 V_ipport_tcplastcount = V_ipport_tcpallocs;
2748 VNET_LIST_RUNLOCK_NOSLEEP();
2749 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2756 callout_stop(&ipport_tick_callout);
2760 * The ipport_callout should start running at about the time we attach the
2761 * inet or inet6 domains.
2764 ipport_tick_init(const void *unused __unused)
2767 /* Start ipport_tick. */
2768 callout_init(&ipport_tick_callout, 1);
2769 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2770 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2771 SHUTDOWN_PRI_DEFAULT);
2773 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2774 ipport_tick_init, NULL);
2777 inp_wlock(struct inpcb *inp)
2784 inp_wunlock(struct inpcb *inp)
2791 inp_rlock(struct inpcb *inp)
2798 inp_runlock(struct inpcb *inp)
2804 #ifdef INVARIANT_SUPPORT
2806 inp_lock_assert(struct inpcb *inp)
2809 INP_WLOCK_ASSERT(inp);
2813 inp_unlock_assert(struct inpcb *inp)
2816 INP_UNLOCK_ASSERT(inp);
2821 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2825 INP_INFO_WLOCK(&V_tcbinfo);
2826 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2831 INP_INFO_WUNLOCK(&V_tcbinfo);
2835 inp_inpcbtosocket(struct inpcb *inp)
2838 INP_WLOCK_ASSERT(inp);
2839 return (inp->inp_socket);
2843 inp_inpcbtotcpcb(struct inpcb *inp)
2846 INP_WLOCK_ASSERT(inp);
2847 return ((struct tcpcb *)inp->inp_ppcb);
2851 inp_ip_tos_get(const struct inpcb *inp)
2854 return (inp->inp_ip_tos);
2858 inp_ip_tos_set(struct inpcb *inp, int val)
2861 inp->inp_ip_tos = val;
2865 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2866 uint32_t *faddr, uint16_t *fp)
2869 INP_LOCK_ASSERT(inp);
2870 *laddr = inp->inp_laddr.s_addr;
2871 *faddr = inp->inp_faddr.s_addr;
2872 *lp = inp->inp_lport;
2873 *fp = inp->inp_fport;
2877 so_sotoinpcb(struct socket *so)
2880 return (sotoinpcb(so));
2884 so_sototcpcb(struct socket *so)
2887 return (sototcpcb(so));
2891 * Create an external-format (``xinpcb'') structure using the information in
2892 * the kernel-format in_pcb structure pointed to by inp. This is done to
2893 * reduce the spew of irrelevant information over this interface, to isolate
2894 * user code from changes in the kernel structure, and potentially to provide
2895 * information-hiding if we decide that some of this information should be
2896 * hidden from users.
2899 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2902 xi->xi_len = sizeof(struct xinpcb);
2903 if (inp->inp_socket)
2904 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2906 bzero(&xi->xi_socket, sizeof(struct xsocket));
2907 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2908 xi->inp_gencnt = inp->inp_gencnt;
2909 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2910 xi->inp_flow = inp->inp_flow;
2911 xi->inp_flowid = inp->inp_flowid;
2912 xi->inp_flowtype = inp->inp_flowtype;
2913 xi->inp_flags = inp->inp_flags;
2914 xi->inp_flags2 = inp->inp_flags2;
2915 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2916 xi->in6p_cksum = inp->in6p_cksum;
2917 xi->in6p_hops = inp->in6p_hops;
2918 xi->inp_ip_tos = inp->inp_ip_tos;
2919 xi->inp_vflag = inp->inp_vflag;
2920 xi->inp_ip_ttl = inp->inp_ip_ttl;
2921 xi->inp_ip_p = inp->inp_ip_p;
2922 xi->inp_ip_minttl = inp->inp_ip_minttl;
2927 db_print_indent(int indent)
2931 for (i = 0; i < indent; i++)
2936 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2938 char faddr_str[48], laddr_str[48];
2940 db_print_indent(indent);
2941 db_printf("%s at %p\n", name, inc);
2946 if (inc->inc_flags & INC_ISIPV6) {
2948 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2949 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2954 inet_ntoa_r(inc->inc_laddr, laddr_str);
2955 inet_ntoa_r(inc->inc_faddr, faddr_str);
2957 db_print_indent(indent);
2958 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2959 ntohs(inc->inc_lport));
2960 db_print_indent(indent);
2961 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2962 ntohs(inc->inc_fport));
2966 db_print_inpflags(int inp_flags)
2971 if (inp_flags & INP_RECVOPTS) {
2972 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2975 if (inp_flags & INP_RECVRETOPTS) {
2976 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2979 if (inp_flags & INP_RECVDSTADDR) {
2980 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2983 if (inp_flags & INP_ORIGDSTADDR) {
2984 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2987 if (inp_flags & INP_HDRINCL) {
2988 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2991 if (inp_flags & INP_HIGHPORT) {
2992 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2995 if (inp_flags & INP_LOWPORT) {
2996 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2999 if (inp_flags & INP_ANONPORT) {
3000 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3003 if (inp_flags & INP_RECVIF) {
3004 db_printf("%sINP_RECVIF", comma ? ", " : "");
3007 if (inp_flags & INP_MTUDISC) {
3008 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3011 if (inp_flags & INP_RECVTTL) {
3012 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3015 if (inp_flags & INP_DONTFRAG) {
3016 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3019 if (inp_flags & INP_RECVTOS) {
3020 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3023 if (inp_flags & IN6P_IPV6_V6ONLY) {
3024 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3027 if (inp_flags & IN6P_PKTINFO) {
3028 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3031 if (inp_flags & IN6P_HOPLIMIT) {
3032 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3035 if (inp_flags & IN6P_HOPOPTS) {
3036 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3039 if (inp_flags & IN6P_DSTOPTS) {
3040 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3043 if (inp_flags & IN6P_RTHDR) {
3044 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3047 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3048 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3051 if (inp_flags & IN6P_TCLASS) {
3052 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3055 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3056 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3059 if (inp_flags & INP_TIMEWAIT) {
3060 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3063 if (inp_flags & INP_ONESBCAST) {
3064 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3067 if (inp_flags & INP_DROPPED) {
3068 db_printf("%sINP_DROPPED", comma ? ", " : "");
3071 if (inp_flags & INP_SOCKREF) {
3072 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3075 if (inp_flags & IN6P_RFC2292) {
3076 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3079 if (inp_flags & IN6P_MTU) {
3080 db_printf("IN6P_MTU%s", comma ? ", " : "");
3086 db_print_inpvflag(u_char inp_vflag)
3091 if (inp_vflag & INP_IPV4) {
3092 db_printf("%sINP_IPV4", comma ? ", " : "");
3095 if (inp_vflag & INP_IPV6) {
3096 db_printf("%sINP_IPV6", comma ? ", " : "");
3099 if (inp_vflag & INP_IPV6PROTO) {
3100 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3106 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3109 db_print_indent(indent);
3110 db_printf("%s at %p\n", name, inp);
3114 db_print_indent(indent);
3115 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3117 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3119 db_print_indent(indent);
3120 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3121 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3123 db_print_indent(indent);
3124 db_printf("inp_label: %p inp_flags: 0x%x (",
3125 inp->inp_label, inp->inp_flags);
3126 db_print_inpflags(inp->inp_flags);
3129 db_print_indent(indent);
3130 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3132 db_print_inpvflag(inp->inp_vflag);
3135 db_print_indent(indent);
3136 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3137 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3139 db_print_indent(indent);
3141 if (inp->inp_vflag & INP_IPV6) {
3142 db_printf("in6p_options: %p in6p_outputopts: %p "
3143 "in6p_moptions: %p\n", inp->in6p_options,
3144 inp->in6p_outputopts, inp->in6p_moptions);
3145 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3146 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3151 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3152 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3153 inp->inp_options, inp->inp_moptions);
3156 db_print_indent(indent);
3157 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3158 (uintmax_t)inp->inp_gencnt);
3161 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3166 db_printf("usage: show inpcb <addr>\n");
3169 inp = (struct inpcb *)addr;
3171 db_print_inpcb(inp, "inpcb", 0);
3177 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3181 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3183 union if_snd_tag_modify_params params = {
3184 .rate_limit.max_rate = max_pacing_rate,
3186 struct m_snd_tag *mst;
3190 mst = inp->inp_snd_tag;
3198 if (ifp->if_snd_tag_modify == NULL) {
3201 error = ifp->if_snd_tag_modify(mst, ¶ms);
3207 * Query existing TX rate limit based on the existing
3208 * "inp->inp_snd_tag", if any.
3211 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3213 union if_snd_tag_query_params params = { };
3214 struct m_snd_tag *mst;
3218 mst = inp->inp_snd_tag;
3226 if (ifp->if_snd_tag_query == NULL) {
3229 error = ifp->if_snd_tag_query(mst, ¶ms);
3230 if (error == 0 && p_max_pacing_rate != NULL)
3231 *p_max_pacing_rate = params.rate_limit.max_rate;
3237 * Query existing TX queue level based on the existing
3238 * "inp->inp_snd_tag", if any.
3241 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3243 union if_snd_tag_query_params params = { };
3244 struct m_snd_tag *mst;
3248 mst = inp->inp_snd_tag;
3256 if (ifp->if_snd_tag_query == NULL)
3257 return (EOPNOTSUPP);
3259 error = ifp->if_snd_tag_query(mst, ¶ms);
3260 if (error == 0 && p_txqueue_level != NULL)
3261 *p_txqueue_level = params.rate_limit.queue_level;
3266 * Allocate a new TX rate limit send tag from the network interface
3267 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3270 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3271 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3273 union if_snd_tag_alloc_params params = {
3274 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3275 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3276 .rate_limit.hdr.flowid = flowid,
3277 .rate_limit.hdr.flowtype = flowtype,
3278 .rate_limit.max_rate = max_pacing_rate,
3282 INP_WLOCK_ASSERT(inp);
3284 if (inp->inp_snd_tag != NULL)
3287 if (ifp->if_snd_tag_alloc == NULL) {
3290 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3293 * At success increment the refcount on
3294 * the send tag's network interface:
3297 if_ref(inp->inp_snd_tag->ifp);
3303 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3307 in_pcbdetach_txrtlmt(struct inpcb *inp)
3309 struct m_snd_tag *mst;
3312 INP_WLOCK_ASSERT(inp);
3314 mst = inp->inp_snd_tag;
3315 inp->inp_snd_tag = NULL;
3325 * If the device was detached while we still had reference(s)
3326 * on the ifp, we assume if_snd_tag_free() was replaced with
3329 ifp->if_snd_tag_free(mst);
3331 /* release reference count on network interface */
3336 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3337 * is set in the fast path and will attach/detach/modify the TX rate
3338 * limit send tag based on the socket's so_max_pacing_rate value.
3341 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3343 struct socket *socket;
3344 uint32_t max_pacing_rate;
3351 socket = inp->inp_socket;
3355 if (!INP_WLOCKED(inp)) {
3357 * NOTE: If the write locking fails, we need to bail
3358 * out and use the non-ratelimited ring for the
3359 * transmit until there is a new chance to get the
3362 if (!INP_TRY_UPGRADE(inp))
3370 * NOTE: The so_max_pacing_rate value is read unlocked,
3371 * because atomic updates are not required since the variable
3372 * is checked at every mbuf we send. It is assumed that the
3373 * variable read itself will be atomic.
3375 max_pacing_rate = socket->so_max_pacing_rate;
3378 * NOTE: When attaching to a network interface a reference is
3379 * made to ensure the network interface doesn't go away until
3380 * all ratelimit connections are gone. The network interface
3381 * pointers compared below represent valid network interfaces,
3382 * except when comparing towards NULL.
3384 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3386 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3387 if (inp->inp_snd_tag != NULL)
3388 in_pcbdetach_txrtlmt(inp);
3390 } else if (inp->inp_snd_tag == NULL) {
3392 * In order to utilize packet pacing with RSS, we need
3393 * to wait until there is a valid RSS hash before we
3396 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3399 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3400 mb->m_pkthdr.flowid, max_pacing_rate);
3403 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3405 if (error == 0 || error == EOPNOTSUPP)
3406 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3412 * Track route changes for TX rate limiting.
3415 in_pcboutput_eagain(struct inpcb *inp)
3417 struct socket *socket;
3423 socket = inp->inp_socket;
3427 if (inp->inp_snd_tag == NULL)
3430 if (!INP_WLOCKED(inp)) {
3432 * NOTE: If the write locking fails, we need to bail
3433 * out and use the non-ratelimited ring for the
3434 * transmit until there is a new chance to get the
3437 if (!INP_TRY_UPGRADE(inp))
3444 /* detach rate limiting */
3445 in_pcbdetach_txrtlmt(inp);
3447 /* make sure new mbuf send tag allocation is made */
3448 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3453 #endif /* RATELIMIT */