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 struct ip_moptions *imo = 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 struct ip6_moptions *im6o = NULL;
1591 if (inp->inp_vflag & INP_IPV6PROTO) {
1592 ip6_freepcbopts(inp->in6p_outputopts);
1593 im6o = inp->in6p_moptions;
1594 inp->in6p_moptions = NULL;
1597 if (inp->inp_options)
1598 (void)m_free(inp->inp_options);
1600 crfree(inp->inp_cred);
1602 mac_inpcb_destroy(inp);
1604 released = in_pcbrele_wlocked(inp);
1607 ip6_freemoptions(im6o);
1610 inp_freemoptions(imo);
1615 * Unconditionally schedule an inpcb to be freed by decrementing its
1616 * reference count, which should occur only after the inpcb has been detached
1617 * from its socket. If another thread holds a temporary reference (acquired
1618 * using in_pcbref()) then the free is deferred until that reference is
1619 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1620 * work, including removal from global lists, is done in this context, where
1621 * the pcbinfo lock is held.
1624 in_pcbfree(struct inpcb *inp)
1626 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1628 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1629 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1630 ("%s: called twice for pcb %p", __func__, inp));
1631 if (inp->inp_flags2 & INP_FREED) {
1637 if (pcbinfo == &V_tcbinfo) {
1638 INP_INFO_LOCK_ASSERT(pcbinfo);
1640 INP_INFO_WLOCK_ASSERT(pcbinfo);
1643 INP_WLOCK_ASSERT(inp);
1644 INP_LIST_WLOCK(pcbinfo);
1645 in_pcbremlists(inp);
1646 INP_LIST_WUNLOCK(pcbinfo);
1647 RO_INVALIDATE_CACHE(&inp->inp_route);
1648 /* mark as destruction in progress */
1649 inp->inp_flags2 |= INP_FREED;
1651 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1655 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1656 * port reservation, and preventing it from being returned by inpcb lookups.
1658 * It is used by TCP to mark an inpcb as unused and avoid future packet
1659 * delivery or event notification when a socket remains open but TCP has
1660 * closed. This might occur as a result of a shutdown()-initiated TCP close
1661 * or a RST on the wire, and allows the port binding to be reused while still
1662 * maintaining the invariant that so_pcb always points to a valid inpcb until
1665 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1666 * in_pcbnotifyall() and in_pcbpurgeif0()?
1669 in_pcbdrop(struct inpcb *inp)
1672 INP_WLOCK_ASSERT(inp);
1674 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1675 MPASS(inp->inp_refcount > 1);
1679 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1682 inp->inp_flags |= INP_DROPPED;
1683 if (inp->inp_flags & INP_INHASHLIST) {
1684 struct inpcbport *phd = inp->inp_phd;
1686 INP_HASH_WLOCK(inp->inp_pcbinfo);
1687 in_pcbremlbgrouphash(inp);
1688 CK_LIST_REMOVE(inp, inp_hash);
1689 CK_LIST_REMOVE(inp, inp_portlist);
1690 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1691 CK_LIST_REMOVE(phd, phd_hash);
1692 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1694 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1695 inp->inp_flags &= ~INP_INHASHLIST;
1697 in_pcbgroup_remove(inp);
1704 * Common routines to return the socket addresses associated with inpcbs.
1707 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1709 struct sockaddr_in *sin;
1711 sin = malloc(sizeof *sin, M_SONAME,
1713 sin->sin_family = AF_INET;
1714 sin->sin_len = sizeof(*sin);
1715 sin->sin_addr = *addr_p;
1716 sin->sin_port = port;
1718 return (struct sockaddr *)sin;
1722 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1725 struct in_addr addr;
1728 inp = sotoinpcb(so);
1729 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1732 port = inp->inp_lport;
1733 addr = inp->inp_laddr;
1736 *nam = in_sockaddr(port, &addr);
1741 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1744 struct in_addr addr;
1747 inp = sotoinpcb(so);
1748 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1751 port = inp->inp_fport;
1752 addr = inp->inp_faddr;
1755 *nam = in_sockaddr(port, &addr);
1760 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1761 struct inpcb *(*notify)(struct inpcb *, int))
1763 struct inpcb *inp, *inp_temp;
1765 INP_INFO_WLOCK(pcbinfo);
1766 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1769 if ((inp->inp_vflag & INP_IPV4) == 0) {
1774 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1775 inp->inp_socket == NULL) {
1779 if ((*notify)(inp, errno))
1782 INP_INFO_WUNLOCK(pcbinfo);
1786 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1789 struct ip_moptions *imo;
1792 INP_INFO_WLOCK(pcbinfo);
1793 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1795 imo = inp->inp_moptions;
1796 if ((inp->inp_vflag & INP_IPV4) &&
1799 * Unselect the outgoing interface if it is being
1802 if (imo->imo_multicast_ifp == ifp)
1803 imo->imo_multicast_ifp = NULL;
1806 * Drop multicast group membership if we joined
1807 * through the interface being detached.
1809 * XXX This can all be deferred to an epoch_call
1811 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1813 if (imo->imo_membership[i]->inm_ifp == ifp) {
1814 IN_MULTI_LOCK_ASSERT();
1815 in_leavegroup_locked(imo->imo_membership[i], NULL);
1817 } else if (gap != 0)
1818 imo->imo_membership[i - gap] =
1819 imo->imo_membership[i];
1821 imo->imo_num_memberships -= gap;
1825 INP_INFO_WUNLOCK(pcbinfo);
1829 * Lookup a PCB based on the local address and port. Caller must hold the
1830 * hash lock. No inpcb locks or references are acquired.
1832 #define INP_LOOKUP_MAPPED_PCB_COST 3
1834 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1835 u_short lport, int lookupflags, struct ucred *cred)
1839 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1845 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1846 ("%s: invalid lookup flags %d", __func__, lookupflags));
1848 INP_HASH_LOCK_ASSERT(pcbinfo);
1850 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1851 struct inpcbhead *head;
1853 * Look for an unconnected (wildcard foreign addr) PCB that
1854 * matches the local address and port we're looking for.
1856 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1857 0, pcbinfo->ipi_hashmask)];
1858 CK_LIST_FOREACH(inp, head, inp_hash) {
1860 /* XXX inp locking */
1861 if ((inp->inp_vflag & INP_IPV4) == 0)
1864 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1865 inp->inp_laddr.s_addr == laddr.s_addr &&
1866 inp->inp_lport == lport) {
1871 prison_equal_ip4(cred->cr_prison,
1872 inp->inp_cred->cr_prison))
1881 struct inpcbporthead *porthash;
1882 struct inpcbport *phd;
1883 struct inpcb *match = NULL;
1885 * Best fit PCB lookup.
1887 * First see if this local port is in use by looking on the
1890 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1891 pcbinfo->ipi_porthashmask)];
1892 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1893 if (phd->phd_port == lport)
1898 * Port is in use by one or more PCBs. Look for best
1901 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1904 !prison_equal_ip4(inp->inp_cred->cr_prison,
1908 /* XXX inp locking */
1909 if ((inp->inp_vflag & INP_IPV4) == 0)
1912 * We never select the PCB that has
1913 * INP_IPV6 flag and is bound to :: if
1914 * we have another PCB which is bound
1915 * to 0.0.0.0. If a PCB has the
1916 * INP_IPV6 flag, then we set its cost
1917 * higher than IPv4 only PCBs.
1919 * Note that the case only happens
1920 * when a socket is bound to ::, under
1921 * the condition that the use of the
1922 * mapped address is allowed.
1924 if ((inp->inp_vflag & INP_IPV6) != 0)
1925 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1927 if (inp->inp_faddr.s_addr != INADDR_ANY)
1929 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1930 if (laddr.s_addr == INADDR_ANY)
1932 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1935 if (laddr.s_addr != INADDR_ANY)
1938 if (wildcard < matchwild) {
1940 matchwild = wildcard;
1949 #undef INP_LOOKUP_MAPPED_PCB_COST
1951 static struct inpcb *
1952 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1953 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1954 uint16_t fport, int lookupflags)
1956 struct inpcb *local_wild;
1957 const struct inpcblbgrouphead *hdr;
1958 struct inpcblbgroup *grp;
1961 INP_HASH_LOCK_ASSERT(pcbinfo);
1963 hdr = &pcbinfo->ipi_lbgrouphashbase[INP_PCBLBGROUP_PORTHASH(lport,
1964 pcbinfo->ipi_lbgrouphashmask)];
1967 * Order of socket selection:
1969 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1972 * - Load balanced group does not contain jailed sockets
1973 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1976 LIST_FOREACH(grp, hdr, il_list) {
1978 if (!(grp->il_vflag & INP_IPV4))
1981 if (grp->il_lport != lport)
1984 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1986 if (grp->il_laddr.s_addr == laddr->s_addr)
1987 return (grp->il_inp[idx]);
1988 if (grp->il_laddr.s_addr == INADDR_ANY &&
1989 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1990 local_wild = grp->il_inp[idx];
1992 return (local_wild);
1997 * Lookup PCB in hash list, using pcbgroup tables.
1999 static struct inpcb *
2000 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2001 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2002 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2004 struct inpcbhead *head;
2005 struct inpcb *inp, *tmpinp;
2006 u_short fport = fport_arg, lport = lport_arg;
2010 * First look for an exact match.
2013 INP_GROUP_LOCK(pcbgroup);
2014 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2015 pcbgroup->ipg_hashmask)];
2016 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2018 /* XXX inp locking */
2019 if ((inp->inp_vflag & INP_IPV4) == 0)
2022 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2023 inp->inp_laddr.s_addr == laddr.s_addr &&
2024 inp->inp_fport == fport &&
2025 inp->inp_lport == lport) {
2027 * XXX We should be able to directly return
2028 * the inp here, without any checks.
2029 * Well unless both bound with SO_REUSEPORT?
2031 if (prison_flag(inp->inp_cred, PR_IP4))
2037 if (tmpinp != NULL) {
2044 * For incoming connections, we may wish to do a wildcard
2045 * match for an RSS-local socket.
2047 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2048 struct inpcb *local_wild = NULL, *local_exact = NULL;
2050 struct inpcb *local_wild_mapped = NULL;
2052 struct inpcb *jail_wild = NULL;
2053 struct inpcbhead *head;
2057 * Order of socket selection - we always prefer jails.
2058 * 1. jailed, non-wild.
2060 * 3. non-jailed, non-wild.
2061 * 4. non-jailed, wild.
2064 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2065 lport, 0, pcbgroup->ipg_hashmask)];
2066 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2068 /* XXX inp locking */
2069 if ((inp->inp_vflag & INP_IPV4) == 0)
2072 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2073 inp->inp_lport != lport)
2076 injail = prison_flag(inp->inp_cred, PR_IP4);
2078 if (prison_check_ip4(inp->inp_cred,
2082 if (local_exact != NULL)
2086 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2091 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2093 /* XXX inp locking, NULL check */
2094 if (inp->inp_vflag & INP_IPV6PROTO)
2095 local_wild_mapped = inp;
2103 } /* LIST_FOREACH */
2112 inp = local_wild_mapped;
2120 * Then look for a wildcard match, if requested.
2122 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2123 struct inpcb *local_wild = NULL, *local_exact = NULL;
2125 struct inpcb *local_wild_mapped = NULL;
2127 struct inpcb *jail_wild = NULL;
2128 struct inpcbhead *head;
2132 * Order of socket selection - we always prefer jails.
2133 * 1. jailed, non-wild.
2135 * 3. non-jailed, non-wild.
2136 * 4. non-jailed, wild.
2138 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2139 0, pcbinfo->ipi_wildmask)];
2140 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2142 /* XXX inp locking */
2143 if ((inp->inp_vflag & INP_IPV4) == 0)
2146 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2147 inp->inp_lport != lport)
2150 injail = prison_flag(inp->inp_cred, PR_IP4);
2152 if (prison_check_ip4(inp->inp_cred,
2156 if (local_exact != NULL)
2160 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2165 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2167 /* XXX inp locking, NULL check */
2168 if (inp->inp_vflag & INP_IPV6PROTO)
2169 local_wild_mapped = inp;
2177 } /* LIST_FOREACH */
2185 inp = local_wild_mapped;
2189 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2190 INP_GROUP_UNLOCK(pcbgroup);
2194 if (lookupflags & INPLOOKUP_WLOCKPCB)
2195 locked = INP_TRY_WLOCK(inp);
2196 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2197 locked = INP_TRY_RLOCK(inp);
2199 panic("%s: locking bug", __func__);
2200 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2201 if (lookupflags & INPLOOKUP_WLOCKPCB)
2208 INP_GROUP_UNLOCK(pcbgroup);
2210 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2212 if (in_pcbrele_wlocked(inp))
2216 if (in_pcbrele_rlocked(inp))
2221 if (lookupflags & INPLOOKUP_WLOCKPCB)
2222 INP_WLOCK_ASSERT(inp);
2224 INP_RLOCK_ASSERT(inp);
2228 #endif /* PCBGROUP */
2231 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2232 * that the caller has locked the hash list, and will not perform any further
2233 * locking or reference operations on either the hash list or the connection.
2235 static struct inpcb *
2236 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2237 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2240 struct inpcbhead *head;
2241 struct inpcb *inp, *tmpinp;
2242 u_short fport = fport_arg, lport = lport_arg;
2245 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2246 ("%s: invalid lookup flags %d", __func__, lookupflags));
2247 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2248 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2251 * First look for an exact match.
2254 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2255 pcbinfo->ipi_hashmask)];
2256 CK_LIST_FOREACH(inp, head, inp_hash) {
2258 /* XXX inp locking */
2259 if ((inp->inp_vflag & INP_IPV4) == 0)
2262 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2263 inp->inp_laddr.s_addr == laddr.s_addr &&
2264 inp->inp_fport == fport &&
2265 inp->inp_lport == lport) {
2267 * XXX We should be able to directly return
2268 * the inp here, without any checks.
2269 * Well unless both bound with SO_REUSEPORT?
2271 if (prison_flag(inp->inp_cred, PR_IP4))
2281 * Then look in lb group (for wildcard match).
2283 if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2284 (lookupflags & INPLOOKUP_WILDCARD)) {
2285 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2286 fport, lookupflags);
2293 * Then look for a wildcard match, if requested.
2295 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2296 struct inpcb *local_wild = NULL, *local_exact = NULL;
2298 struct inpcb *local_wild_mapped = NULL;
2300 struct inpcb *jail_wild = NULL;
2304 * Order of socket selection - we always prefer jails.
2305 * 1. jailed, non-wild.
2307 * 3. non-jailed, non-wild.
2308 * 4. non-jailed, wild.
2311 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2312 0, pcbinfo->ipi_hashmask)];
2313 CK_LIST_FOREACH(inp, head, inp_hash) {
2315 /* XXX inp locking */
2316 if ((inp->inp_vflag & INP_IPV4) == 0)
2319 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2320 inp->inp_lport != lport)
2323 injail = prison_flag(inp->inp_cred, PR_IP4);
2325 if (prison_check_ip4(inp->inp_cred,
2329 if (local_exact != NULL)
2333 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2338 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2340 /* XXX inp locking, NULL check */
2341 if (inp->inp_vflag & INP_IPV6PROTO)
2342 local_wild_mapped = inp;
2350 } /* LIST_FOREACH */
2351 if (jail_wild != NULL)
2353 if (local_exact != NULL)
2354 return (local_exact);
2355 if (local_wild != NULL)
2356 return (local_wild);
2358 if (local_wild_mapped != NULL)
2359 return (local_wild_mapped);
2361 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2367 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2368 * hash list lock, and will return the inpcb locked (i.e., requires
2369 * INPLOOKUP_LOCKPCB).
2371 static struct inpcb *
2372 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2373 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2378 INP_HASH_RLOCK(pcbinfo);
2379 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2380 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2382 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2384 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2388 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2390 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2395 panic("%s: locking bug", __func__);
2398 if (lookupflags & INPLOOKUP_WLOCKPCB)
2399 INP_WLOCK_ASSERT(inp);
2401 INP_RLOCK_ASSERT(inp);
2405 INP_HASH_RUNLOCK(pcbinfo);
2410 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2411 * from which a pre-calculated hash value may be extracted.
2413 * Possibly more of this logic should be in in_pcbgroup.c.
2416 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2417 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2419 #if defined(PCBGROUP) && !defined(RSS)
2420 struct inpcbgroup *pcbgroup;
2423 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2424 ("%s: invalid lookup flags %d", __func__, lookupflags));
2425 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2426 ("%s: LOCKPCB not set", __func__));
2429 * When not using RSS, use connection groups in preference to the
2430 * reservation table when looking up 4-tuples. When using RSS, just
2431 * use the reservation table, due to the cost of the Toeplitz hash
2434 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2435 * we could be doing RSS with a non-Toeplitz hash that is affordable
2438 #if defined(PCBGROUP) && !defined(RSS)
2439 if (in_pcbgroup_enabled(pcbinfo)) {
2440 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2442 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2443 laddr, lport, lookupflags, ifp));
2446 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2451 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2452 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2453 struct ifnet *ifp, struct mbuf *m)
2456 struct inpcbgroup *pcbgroup;
2459 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2460 ("%s: invalid lookup flags %d", __func__, lookupflags));
2461 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2462 ("%s: LOCKPCB not set", __func__));
2466 * If we can use a hardware-generated hash to look up the connection
2467 * group, use that connection group to find the inpcb. Otherwise
2468 * fall back on a software hash -- or the reservation table if we're
2471 * XXXRW: As above, that policy belongs in the pcbgroup code.
2473 if (in_pcbgroup_enabled(pcbinfo) &&
2474 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2475 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2476 m->m_pkthdr.flowid);
2477 if (pcbgroup != NULL)
2478 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2479 fport, laddr, lport, lookupflags, ifp));
2481 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2483 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2484 laddr, lport, lookupflags, ifp));
2488 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2494 * Insert PCB onto various hash lists.
2497 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2499 struct inpcbhead *pcbhash;
2500 struct inpcbporthead *pcbporthash;
2501 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2502 struct inpcbport *phd;
2503 u_int32_t hashkey_faddr;
2506 INP_WLOCK_ASSERT(inp);
2507 INP_HASH_WLOCK_ASSERT(pcbinfo);
2509 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2510 ("in_pcbinshash: INP_INHASHLIST"));
2513 if (inp->inp_vflag & INP_IPV6)
2514 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2517 hashkey_faddr = inp->inp_faddr.s_addr;
2519 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2520 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2522 pcbporthash = &pcbinfo->ipi_porthashbase[
2523 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2526 * Add entry to load balance group.
2527 * Only do this if SO_REUSEPORT_LB is set.
2529 so_options = inp_so_options(inp);
2530 if (so_options & SO_REUSEPORT_LB) {
2531 int ret = in_pcbinslbgrouphash(inp);
2533 /* pcb lb group malloc fail (ret=ENOBUFS). */
2539 * Go through port list and look for a head for this lport.
2541 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2542 if (phd->phd_port == inp->inp_lport)
2546 * If none exists, malloc one and tack it on.
2549 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2551 return (ENOBUFS); /* XXX */
2553 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2554 phd->phd_port = inp->inp_lport;
2555 CK_LIST_INIT(&phd->phd_pcblist);
2556 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2559 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2560 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2561 inp->inp_flags |= INP_INHASHLIST;
2563 if (do_pcbgroup_update)
2564 in_pcbgroup_update(inp);
2570 * For now, there are two public interfaces to insert an inpcb into the hash
2571 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2572 * is used only in the TCP syncache, where in_pcbinshash is called before the
2573 * full 4-tuple is set for the inpcb, and we don't want to install in the
2574 * pcbgroup until later.
2576 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2577 * connection groups, and partially initialised inpcbs should not be exposed
2578 * to either reservation hash tables or pcbgroups.
2581 in_pcbinshash(struct inpcb *inp)
2584 return (in_pcbinshash_internal(inp, 1));
2588 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2591 return (in_pcbinshash_internal(inp, 0));
2595 * Move PCB to the proper hash bucket when { faddr, fport } have been
2596 * changed. NOTE: This does not handle the case of the lport changing (the
2597 * hashed port list would have to be updated as well), so the lport must
2598 * not change after in_pcbinshash() has been called.
2601 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2603 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2604 struct inpcbhead *head;
2605 u_int32_t hashkey_faddr;
2607 INP_WLOCK_ASSERT(inp);
2608 INP_HASH_WLOCK_ASSERT(pcbinfo);
2610 KASSERT(inp->inp_flags & INP_INHASHLIST,
2611 ("in_pcbrehash: !INP_INHASHLIST"));
2614 if (inp->inp_vflag & INP_IPV6)
2615 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2618 hashkey_faddr = inp->inp_faddr.s_addr;
2620 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2621 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2623 CK_LIST_REMOVE(inp, inp_hash);
2624 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2628 in_pcbgroup_update_mbuf(inp, m);
2630 in_pcbgroup_update(inp);
2635 in_pcbrehash(struct inpcb *inp)
2638 in_pcbrehash_mbuf(inp, NULL);
2642 * Remove PCB from various lists.
2645 in_pcbremlists(struct inpcb *inp)
2647 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2650 if (pcbinfo == &V_tcbinfo) {
2651 INP_INFO_RLOCK_ASSERT(pcbinfo);
2653 INP_INFO_WLOCK_ASSERT(pcbinfo);
2657 INP_WLOCK_ASSERT(inp);
2658 INP_LIST_WLOCK_ASSERT(pcbinfo);
2660 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2661 if (inp->inp_flags & INP_INHASHLIST) {
2662 struct inpcbport *phd = inp->inp_phd;
2664 INP_HASH_WLOCK(pcbinfo);
2666 /* XXX: Only do if SO_REUSEPORT_LB set? */
2667 in_pcbremlbgrouphash(inp);
2669 CK_LIST_REMOVE(inp, inp_hash);
2670 CK_LIST_REMOVE(inp, inp_portlist);
2671 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2672 CK_LIST_REMOVE(phd, phd_hash);
2673 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2675 INP_HASH_WUNLOCK(pcbinfo);
2676 inp->inp_flags &= ~INP_INHASHLIST;
2678 CK_LIST_REMOVE(inp, inp_list);
2679 pcbinfo->ipi_count--;
2681 in_pcbgroup_remove(inp);
2686 * Check for alternatives when higher level complains
2687 * about service problems. For now, invalidate cached
2688 * routing information. If the route was created dynamically
2689 * (by a redirect), time to try a default gateway again.
2692 in_losing(struct inpcb *inp)
2695 RO_INVALIDATE_CACHE(&inp->inp_route);
2700 * A set label operation has occurred at the socket layer, propagate the
2701 * label change into the in_pcb for the socket.
2704 in_pcbsosetlabel(struct socket *so)
2709 inp = sotoinpcb(so);
2710 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2714 mac_inpcb_sosetlabel(so, inp);
2721 * ipport_tick runs once per second, determining if random port allocation
2722 * should be continued. If more than ipport_randomcps ports have been
2723 * allocated in the last second, then we return to sequential port
2724 * allocation. We return to random allocation only once we drop below
2725 * ipport_randomcps for at least ipport_randomtime seconds.
2728 ipport_tick(void *xtp)
2730 VNET_ITERATOR_DECL(vnet_iter);
2732 VNET_LIST_RLOCK_NOSLEEP();
2733 VNET_FOREACH(vnet_iter) {
2734 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2735 if (V_ipport_tcpallocs <=
2736 V_ipport_tcplastcount + V_ipport_randomcps) {
2737 if (V_ipport_stoprandom > 0)
2738 V_ipport_stoprandom--;
2740 V_ipport_stoprandom = V_ipport_randomtime;
2741 V_ipport_tcplastcount = V_ipport_tcpallocs;
2744 VNET_LIST_RUNLOCK_NOSLEEP();
2745 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2752 callout_stop(&ipport_tick_callout);
2756 * The ipport_callout should start running at about the time we attach the
2757 * inet or inet6 domains.
2760 ipport_tick_init(const void *unused __unused)
2763 /* Start ipport_tick. */
2764 callout_init(&ipport_tick_callout, 1);
2765 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2766 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2767 SHUTDOWN_PRI_DEFAULT);
2769 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2770 ipport_tick_init, NULL);
2773 inp_wlock(struct inpcb *inp)
2780 inp_wunlock(struct inpcb *inp)
2787 inp_rlock(struct inpcb *inp)
2794 inp_runlock(struct inpcb *inp)
2800 #ifdef INVARIANT_SUPPORT
2802 inp_lock_assert(struct inpcb *inp)
2805 INP_WLOCK_ASSERT(inp);
2809 inp_unlock_assert(struct inpcb *inp)
2812 INP_UNLOCK_ASSERT(inp);
2817 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2821 INP_INFO_WLOCK(&V_tcbinfo);
2822 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2827 INP_INFO_WUNLOCK(&V_tcbinfo);
2831 inp_inpcbtosocket(struct inpcb *inp)
2834 INP_WLOCK_ASSERT(inp);
2835 return (inp->inp_socket);
2839 inp_inpcbtotcpcb(struct inpcb *inp)
2842 INP_WLOCK_ASSERT(inp);
2843 return ((struct tcpcb *)inp->inp_ppcb);
2847 inp_ip_tos_get(const struct inpcb *inp)
2850 return (inp->inp_ip_tos);
2854 inp_ip_tos_set(struct inpcb *inp, int val)
2857 inp->inp_ip_tos = val;
2861 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2862 uint32_t *faddr, uint16_t *fp)
2865 INP_LOCK_ASSERT(inp);
2866 *laddr = inp->inp_laddr.s_addr;
2867 *faddr = inp->inp_faddr.s_addr;
2868 *lp = inp->inp_lport;
2869 *fp = inp->inp_fport;
2873 so_sotoinpcb(struct socket *so)
2876 return (sotoinpcb(so));
2880 so_sototcpcb(struct socket *so)
2883 return (sototcpcb(so));
2887 * Create an external-format (``xinpcb'') structure using the information in
2888 * the kernel-format in_pcb structure pointed to by inp. This is done to
2889 * reduce the spew of irrelevant information over this interface, to isolate
2890 * user code from changes in the kernel structure, and potentially to provide
2891 * information-hiding if we decide that some of this information should be
2892 * hidden from users.
2895 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2898 xi->xi_len = sizeof(struct xinpcb);
2899 if (inp->inp_socket)
2900 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2902 bzero(&xi->xi_socket, sizeof(struct xsocket));
2903 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2904 xi->inp_gencnt = inp->inp_gencnt;
2905 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2906 xi->inp_flow = inp->inp_flow;
2907 xi->inp_flowid = inp->inp_flowid;
2908 xi->inp_flowtype = inp->inp_flowtype;
2909 xi->inp_flags = inp->inp_flags;
2910 xi->inp_flags2 = inp->inp_flags2;
2911 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2912 xi->in6p_cksum = inp->in6p_cksum;
2913 xi->in6p_hops = inp->in6p_hops;
2914 xi->inp_ip_tos = inp->inp_ip_tos;
2915 xi->inp_vflag = inp->inp_vflag;
2916 xi->inp_ip_ttl = inp->inp_ip_ttl;
2917 xi->inp_ip_p = inp->inp_ip_p;
2918 xi->inp_ip_minttl = inp->inp_ip_minttl;
2923 db_print_indent(int indent)
2927 for (i = 0; i < indent; i++)
2932 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2934 char faddr_str[48], laddr_str[48];
2936 db_print_indent(indent);
2937 db_printf("%s at %p\n", name, inc);
2942 if (inc->inc_flags & INC_ISIPV6) {
2944 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2945 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2950 inet_ntoa_r(inc->inc_laddr, laddr_str);
2951 inet_ntoa_r(inc->inc_faddr, faddr_str);
2953 db_print_indent(indent);
2954 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2955 ntohs(inc->inc_lport));
2956 db_print_indent(indent);
2957 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2958 ntohs(inc->inc_fport));
2962 db_print_inpflags(int inp_flags)
2967 if (inp_flags & INP_RECVOPTS) {
2968 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2971 if (inp_flags & INP_RECVRETOPTS) {
2972 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2975 if (inp_flags & INP_RECVDSTADDR) {
2976 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2979 if (inp_flags & INP_ORIGDSTADDR) {
2980 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2983 if (inp_flags & INP_HDRINCL) {
2984 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2987 if (inp_flags & INP_HIGHPORT) {
2988 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2991 if (inp_flags & INP_LOWPORT) {
2992 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2995 if (inp_flags & INP_ANONPORT) {
2996 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2999 if (inp_flags & INP_RECVIF) {
3000 db_printf("%sINP_RECVIF", comma ? ", " : "");
3003 if (inp_flags & INP_MTUDISC) {
3004 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3007 if (inp_flags & INP_RECVTTL) {
3008 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3011 if (inp_flags & INP_DONTFRAG) {
3012 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3015 if (inp_flags & INP_RECVTOS) {
3016 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3019 if (inp_flags & IN6P_IPV6_V6ONLY) {
3020 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3023 if (inp_flags & IN6P_PKTINFO) {
3024 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3027 if (inp_flags & IN6P_HOPLIMIT) {
3028 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3031 if (inp_flags & IN6P_HOPOPTS) {
3032 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3035 if (inp_flags & IN6P_DSTOPTS) {
3036 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3039 if (inp_flags & IN6P_RTHDR) {
3040 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3043 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3044 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3047 if (inp_flags & IN6P_TCLASS) {
3048 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3051 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3052 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3055 if (inp_flags & INP_TIMEWAIT) {
3056 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3059 if (inp_flags & INP_ONESBCAST) {
3060 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3063 if (inp_flags & INP_DROPPED) {
3064 db_printf("%sINP_DROPPED", comma ? ", " : "");
3067 if (inp_flags & INP_SOCKREF) {
3068 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3071 if (inp_flags & IN6P_RFC2292) {
3072 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3075 if (inp_flags & IN6P_MTU) {
3076 db_printf("IN6P_MTU%s", comma ? ", " : "");
3082 db_print_inpvflag(u_char inp_vflag)
3087 if (inp_vflag & INP_IPV4) {
3088 db_printf("%sINP_IPV4", comma ? ", " : "");
3091 if (inp_vflag & INP_IPV6) {
3092 db_printf("%sINP_IPV6", comma ? ", " : "");
3095 if (inp_vflag & INP_IPV6PROTO) {
3096 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3102 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3105 db_print_indent(indent);
3106 db_printf("%s at %p\n", name, inp);
3110 db_print_indent(indent);
3111 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3113 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3115 db_print_indent(indent);
3116 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3117 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3119 db_print_indent(indent);
3120 db_printf("inp_label: %p inp_flags: 0x%x (",
3121 inp->inp_label, inp->inp_flags);
3122 db_print_inpflags(inp->inp_flags);
3125 db_print_indent(indent);
3126 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3128 db_print_inpvflag(inp->inp_vflag);
3131 db_print_indent(indent);
3132 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3133 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3135 db_print_indent(indent);
3137 if (inp->inp_vflag & INP_IPV6) {
3138 db_printf("in6p_options: %p in6p_outputopts: %p "
3139 "in6p_moptions: %p\n", inp->in6p_options,
3140 inp->in6p_outputopts, inp->in6p_moptions);
3141 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3142 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3147 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3148 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3149 inp->inp_options, inp->inp_moptions);
3152 db_print_indent(indent);
3153 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3154 (uintmax_t)inp->inp_gencnt);
3157 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3162 db_printf("usage: show inpcb <addr>\n");
3165 inp = (struct inpcb *)addr;
3167 db_print_inpcb(inp, "inpcb", 0);
3173 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3177 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3179 union if_snd_tag_modify_params params = {
3180 .rate_limit.max_rate = max_pacing_rate,
3182 struct m_snd_tag *mst;
3186 mst = inp->inp_snd_tag;
3194 if (ifp->if_snd_tag_modify == NULL) {
3197 error = ifp->if_snd_tag_modify(mst, ¶ms);
3203 * Query existing TX rate limit based on the existing
3204 * "inp->inp_snd_tag", if any.
3207 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3209 union if_snd_tag_query_params params = { };
3210 struct m_snd_tag *mst;
3214 mst = inp->inp_snd_tag;
3222 if (ifp->if_snd_tag_query == NULL) {
3225 error = ifp->if_snd_tag_query(mst, ¶ms);
3226 if (error == 0 && p_max_pacing_rate != NULL)
3227 *p_max_pacing_rate = params.rate_limit.max_rate;
3233 * Query existing TX queue level based on the existing
3234 * "inp->inp_snd_tag", if any.
3237 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3239 union if_snd_tag_query_params params = { };
3240 struct m_snd_tag *mst;
3244 mst = inp->inp_snd_tag;
3252 if (ifp->if_snd_tag_query == NULL)
3253 return (EOPNOTSUPP);
3255 error = ifp->if_snd_tag_query(mst, ¶ms);
3256 if (error == 0 && p_txqueue_level != NULL)
3257 *p_txqueue_level = params.rate_limit.queue_level;
3262 * Allocate a new TX rate limit send tag from the network interface
3263 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3266 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3267 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3269 union if_snd_tag_alloc_params params = {
3270 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3271 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3272 .rate_limit.hdr.flowid = flowid,
3273 .rate_limit.hdr.flowtype = flowtype,
3274 .rate_limit.max_rate = max_pacing_rate,
3278 INP_WLOCK_ASSERT(inp);
3280 if (inp->inp_snd_tag != NULL)
3283 if (ifp->if_snd_tag_alloc == NULL) {
3286 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3289 * At success increment the refcount on
3290 * the send tag's network interface:
3293 if_ref(inp->inp_snd_tag->ifp);
3299 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3303 in_pcbdetach_txrtlmt(struct inpcb *inp)
3305 struct m_snd_tag *mst;
3308 INP_WLOCK_ASSERT(inp);
3310 mst = inp->inp_snd_tag;
3311 inp->inp_snd_tag = NULL;
3321 * If the device was detached while we still had reference(s)
3322 * on the ifp, we assume if_snd_tag_free() was replaced with
3325 ifp->if_snd_tag_free(mst);
3327 /* release reference count on network interface */
3332 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3333 * is set in the fast path and will attach/detach/modify the TX rate
3334 * limit send tag based on the socket's so_max_pacing_rate value.
3337 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3339 struct socket *socket;
3340 uint32_t max_pacing_rate;
3347 socket = inp->inp_socket;
3351 if (!INP_WLOCKED(inp)) {
3353 * NOTE: If the write locking fails, we need to bail
3354 * out and use the non-ratelimited ring for the
3355 * transmit until there is a new chance to get the
3358 if (!INP_TRY_UPGRADE(inp))
3366 * NOTE: The so_max_pacing_rate value is read unlocked,
3367 * because atomic updates are not required since the variable
3368 * is checked at every mbuf we send. It is assumed that the
3369 * variable read itself will be atomic.
3371 max_pacing_rate = socket->so_max_pacing_rate;
3374 * NOTE: When attaching to a network interface a reference is
3375 * made to ensure the network interface doesn't go away until
3376 * all ratelimit connections are gone. The network interface
3377 * pointers compared below represent valid network interfaces,
3378 * except when comparing towards NULL.
3380 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3382 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3383 if (inp->inp_snd_tag != NULL)
3384 in_pcbdetach_txrtlmt(inp);
3386 } else if (inp->inp_snd_tag == NULL) {
3388 * In order to utilize packet pacing with RSS, we need
3389 * to wait until there is a valid RSS hash before we
3392 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3395 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3396 mb->m_pkthdr.flowid, max_pacing_rate);
3399 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3401 if (error == 0 || error == EOPNOTSUPP)
3402 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3408 * Track route changes for TX rate limiting.
3411 in_pcboutput_eagain(struct inpcb *inp)
3413 struct socket *socket;
3419 socket = inp->inp_socket;
3423 if (inp->inp_snd_tag == NULL)
3426 if (!INP_WLOCKED(inp)) {
3428 * NOTE: If the write locking fails, we need to bail
3429 * out and use the non-ratelimited ring for the
3430 * transmit until there is a new chance to get the
3433 if (!INP_TRY_UPGRADE(inp))
3440 /* detach rate limiting */
3441 in_pcbdetach_txrtlmt(inp);
3443 /* make sure new mbuf send tag allocation is made */
3444 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3449 #endif /* RATELIMIT */