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 static VNET_DEFINE(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;
1088 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1091 if (sa->sa_family != AF_INET)
1093 sin = (struct sockaddr_in *)sa;
1094 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1095 ia = (struct in_ifaddr *)ifa;
1100 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1101 IF_ADDR_RUNLOCK(ifp);
1104 IF_ADDR_RUNLOCK(ifp);
1106 /* 3. As a last resort return the 'default' jail address. */
1107 error = prison_get_ip4(cred, laddr);
1112 * If the outgoing interface on the route found is not
1113 * a loopback interface, use the address from that interface.
1114 * In case of jails do those three steps:
1115 * 1. check if the interface address belongs to the jail. If so use it.
1116 * 2. check if we have any address on the outgoing interface
1117 * belonging to this jail. If so use it.
1118 * 3. as a last resort return the 'default' jail address.
1120 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1121 struct in_ifaddr *ia;
1124 /* If not jailed, use the default returned. */
1125 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1126 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1127 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1132 /* 1. Check if the iface address belongs to the jail. */
1133 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1134 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1135 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1136 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1141 * 2. Check if we have any address on the outgoing interface
1142 * belonging to this jail.
1145 ifp = sro.ro_rt->rt_ifp;
1147 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1149 if (sa->sa_family != AF_INET)
1151 sin = (struct sockaddr_in *)sa;
1152 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1153 ia = (struct in_ifaddr *)ifa;
1158 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1159 IF_ADDR_RUNLOCK(ifp);
1162 IF_ADDR_RUNLOCK(ifp);
1164 /* 3. As a last resort return the 'default' jail address. */
1165 error = prison_get_ip4(cred, laddr);
1170 * The outgoing interface is marked with 'loopback net', so a route
1171 * to ourselves is here.
1172 * Try to find the interface of the destination address and then
1173 * take the address from there. That interface is not necessarily
1174 * a loopback interface.
1175 * In case of jails, check that it is an address of the jail
1176 * and if we cannot find, fall back to the 'default' jail address.
1178 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1179 struct sockaddr_in sain;
1180 struct in_ifaddr *ia;
1182 bzero(&sain, sizeof(struct sockaddr_in));
1183 sain.sin_family = AF_INET;
1184 sain.sin_len = sizeof(struct sockaddr_in);
1185 sain.sin_addr.s_addr = faddr->s_addr;
1187 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1188 inp->inp_socket->so_fibnum));
1190 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1191 inp->inp_socket->so_fibnum));
1193 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1195 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1197 error = ENETUNREACH;
1200 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1211 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1214 if (sa->sa_family != AF_INET)
1216 sin = (struct sockaddr_in *)sa;
1217 if (prison_check_ip4(cred,
1218 &sin->sin_addr) == 0) {
1219 ia = (struct in_ifaddr *)ifa;
1224 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1225 IF_ADDR_RUNLOCK(ifp);
1228 IF_ADDR_RUNLOCK(ifp);
1231 /* 3. As a last resort return the 'default' jail address. */
1232 error = prison_get_ip4(cred, laddr);
1238 if (sro.ro_rt != NULL)
1244 * Set up for a connect from a socket to the specified address.
1245 * On entry, *laddrp and *lportp should contain the current local
1246 * address and port for the PCB; these are updated to the values
1247 * that should be placed in inp_laddr and inp_lport to complete
1250 * On success, *faddrp and *fportp will be set to the remote address
1251 * and port. These are not updated in the error case.
1253 * If the operation fails because the connection already exists,
1254 * *oinpp will be set to the PCB of that connection so that the
1255 * caller can decide to override it. In all other cases, *oinpp
1259 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1260 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1261 struct inpcb **oinpp, struct ucred *cred)
1263 struct rm_priotracker in_ifa_tracker;
1264 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1265 struct in_ifaddr *ia;
1267 struct in_addr laddr, faddr;
1268 u_short lport, fport;
1272 * Because a global state change doesn't actually occur here, a read
1273 * lock is sufficient.
1275 INP_LOCK_ASSERT(inp);
1276 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1280 if (nam->sa_len != sizeof (*sin))
1282 if (sin->sin_family != AF_INET)
1283 return (EAFNOSUPPORT);
1284 if (sin->sin_port == 0)
1285 return (EADDRNOTAVAIL);
1286 laddr.s_addr = *laddrp;
1288 faddr = sin->sin_addr;
1289 fport = sin->sin_port;
1291 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1293 * If the destination address is INADDR_ANY,
1294 * use the primary local address.
1295 * If the supplied address is INADDR_BROADCAST,
1296 * and the primary interface supports broadcast,
1297 * choose the broadcast address for that interface.
1299 if (faddr.s_addr == INADDR_ANY) {
1300 IN_IFADDR_RLOCK(&in_ifa_tracker);
1302 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1303 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1305 (error = prison_get_ip4(cred, &faddr)) != 0)
1307 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1308 IN_IFADDR_RLOCK(&in_ifa_tracker);
1309 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1311 faddr = satosin(&CK_STAILQ_FIRST(
1312 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1313 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1316 if (laddr.s_addr == INADDR_ANY) {
1317 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1319 * If the destination address is multicast and an outgoing
1320 * interface has been set as a multicast option, prefer the
1321 * address of that interface as our source address.
1323 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1324 inp->inp_moptions != NULL) {
1325 struct ip_moptions *imo;
1328 imo = inp->inp_moptions;
1329 if (imo->imo_multicast_ifp != NULL) {
1330 ifp = imo->imo_multicast_ifp;
1331 IN_IFADDR_RLOCK(&in_ifa_tracker);
1332 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1333 if ((ia->ia_ifp == ifp) &&
1335 prison_check_ip4(cred,
1336 &ia->ia_addr.sin_addr) == 0))
1340 error = EADDRNOTAVAIL;
1342 laddr = ia->ia_addr.sin_addr;
1345 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1351 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1352 laddr, lport, 0, NULL);
1356 return (EADDRINUSE);
1359 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1364 *laddrp = laddr.s_addr;
1366 *faddrp = faddr.s_addr;
1372 in_pcbdisconnect(struct inpcb *inp)
1375 INP_WLOCK_ASSERT(inp);
1376 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1378 inp->inp_faddr.s_addr = INADDR_ANY;
1385 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1386 * For most protocols, this will be invoked immediately prior to calling
1387 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1388 * socket, in which case in_pcbfree() is deferred.
1391 in_pcbdetach(struct inpcb *inp)
1394 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1397 if (inp->inp_snd_tag != NULL)
1398 in_pcbdetach_txrtlmt(inp);
1400 inp->inp_socket->so_pcb = NULL;
1401 inp->inp_socket = NULL;
1405 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1406 * stability of an inpcb pointer despite the inpcb lock being released. This
1407 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1408 * but where the inpcb lock may already held, or when acquiring a reference
1411 * in_pcbref() should be used only to provide brief memory stability, and
1412 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1413 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1414 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1415 * lock and rele are the *only* safe operations that may be performed on the
1418 * While the inpcb will not be freed, releasing the inpcb lock means that the
1419 * connection's state may change, so the caller should be careful to
1420 * revalidate any cached state on reacquiring the lock. Drop the reference
1421 * using in_pcbrele().
1424 in_pcbref(struct inpcb *inp)
1427 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1429 refcount_acquire(&inp->inp_refcount);
1433 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1434 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1435 * return a flag indicating whether or not the inpcb remains valid. If it is
1436 * valid, we return with the inpcb lock held.
1438 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1439 * reference on an inpcb. Historically more work was done here (actually, in
1440 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1441 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1442 * about memory stability (and continued use of the write lock).
1445 in_pcbrele_rlocked(struct inpcb *inp)
1447 struct inpcbinfo *pcbinfo;
1449 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1451 INP_RLOCK_ASSERT(inp);
1453 if (refcount_release(&inp->inp_refcount) == 0) {
1455 * If the inpcb has been freed, let the caller know, even if
1456 * this isn't the last reference.
1458 if (inp->inp_flags2 & INP_FREED) {
1465 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1467 if (inp->inp_in_hpts || inp->inp_in_input) {
1468 struct tcp_hpts_entry *hpts;
1470 * We should not be on the hpts at
1471 * this point in any form. we must
1472 * get the lock to be sure.
1474 hpts = tcp_hpts_lock(inp);
1475 if (inp->inp_in_hpts)
1476 panic("Hpts:%p inp:%p at free still on hpts",
1478 mtx_unlock(&hpts->p_mtx);
1479 hpts = tcp_input_lock(inp);
1480 if (inp->inp_in_input)
1481 panic("Hpts:%p inp:%p at free still on input hpts",
1483 mtx_unlock(&hpts->p_mtx);
1487 pcbinfo = inp->inp_pcbinfo;
1488 uma_zfree(pcbinfo->ipi_zone, inp);
1493 in_pcbrele_wlocked(struct inpcb *inp)
1495 struct inpcbinfo *pcbinfo;
1497 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1499 INP_WLOCK_ASSERT(inp);
1501 if (refcount_release(&inp->inp_refcount) == 0) {
1503 * If the inpcb has been freed, let the caller know, even if
1504 * this isn't the last reference.
1506 if (inp->inp_flags2 & INP_FREED) {
1513 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1515 if (inp->inp_in_hpts || inp->inp_in_input) {
1516 struct tcp_hpts_entry *hpts;
1518 * We should not be on the hpts at
1519 * this point in any form. we must
1520 * get the lock to be sure.
1522 hpts = tcp_hpts_lock(inp);
1523 if (inp->inp_in_hpts)
1524 panic("Hpts:%p inp:%p at free still on hpts",
1526 mtx_unlock(&hpts->p_mtx);
1527 hpts = tcp_input_lock(inp);
1528 if (inp->inp_in_input)
1529 panic("Hpts:%p inp:%p at free still on input hpts",
1531 mtx_unlock(&hpts->p_mtx);
1535 pcbinfo = inp->inp_pcbinfo;
1536 uma_zfree(pcbinfo->ipi_zone, inp);
1541 * Temporary wrapper.
1544 in_pcbrele(struct inpcb *inp)
1547 return (in_pcbrele_wlocked(inp));
1551 in_pcblist_rele_rlocked(epoch_context_t ctx)
1553 struct in_pcblist *il;
1555 struct inpcbinfo *pcbinfo;
1558 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1559 pcbinfo = il->il_pcbinfo;
1561 INP_INFO_WLOCK(pcbinfo);
1562 for (i = 0; i < n; i++) {
1563 inp = il->il_inp_list[i];
1565 if (!in_pcbrele_rlocked(inp))
1568 INP_INFO_WUNLOCK(pcbinfo);
1573 in_pcbfree_deferred(epoch_context_t ctx)
1576 struct inpcbinfo *pcbinfo;
1577 int released __unused;
1579 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1580 pcbinfo = inp->inp_pcbinfo;
1584 inp_freemoptions(inp->inp_moptions);
1585 inp->inp_moptions = NULL;
1587 /* XXXRW: Do as much as possible here. */
1588 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1589 if (inp->inp_sp != NULL)
1590 ipsec_delete_pcbpolicy(inp);
1593 if (inp->inp_vflag & INP_IPV6PROTO) {
1594 ip6_freepcbopts(inp->in6p_outputopts);
1595 ip6_freemoptions(inp->in6p_moptions);
1596 inp->in6p_moptions = NULL;
1599 if (inp->inp_options)
1600 (void)m_free(inp->inp_options);
1603 crfree(inp->inp_cred);
1605 mac_inpcb_destroy(inp);
1607 released = in_pcbrele_wlocked(inp);
1612 * Unconditionally schedule an inpcb to be freed by decrementing its
1613 * reference count, which should occur only after the inpcb has been detached
1614 * from its socket. If another thread holds a temporary reference (acquired
1615 * using in_pcbref()) then the free is deferred until that reference is
1616 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1617 * work, including removal from global lists, is done in this context, where
1618 * the pcbinfo lock is held.
1621 in_pcbfree(struct inpcb *inp)
1623 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1625 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1626 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1627 ("%s: called twice for pcb %p", __func__, inp));
1628 if (inp->inp_flags2 & INP_FREED) {
1634 if (pcbinfo == &V_tcbinfo) {
1635 INP_INFO_LOCK_ASSERT(pcbinfo);
1637 INP_INFO_WLOCK_ASSERT(pcbinfo);
1640 INP_WLOCK_ASSERT(inp);
1641 INP_LIST_WLOCK(pcbinfo);
1642 in_pcbremlists(inp);
1643 INP_LIST_WUNLOCK(pcbinfo);
1644 RO_INVALIDATE_CACHE(&inp->inp_route);
1645 /* mark as destruction in progress */
1646 inp->inp_flags2 |= INP_FREED;
1648 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1652 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1653 * port reservation, and preventing it from being returned by inpcb lookups.
1655 * It is used by TCP to mark an inpcb as unused and avoid future packet
1656 * delivery or event notification when a socket remains open but TCP has
1657 * closed. This might occur as a result of a shutdown()-initiated TCP close
1658 * or a RST on the wire, and allows the port binding to be reused while still
1659 * maintaining the invariant that so_pcb always points to a valid inpcb until
1662 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1663 * in_pcbnotifyall() and in_pcbpurgeif0()?
1666 in_pcbdrop(struct inpcb *inp)
1669 INP_WLOCK_ASSERT(inp);
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);
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 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 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 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__);
2206 INP_GROUP_UNLOCK(pcbgroup);
2208 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2210 if (in_pcbrele_wlocked(inp))
2214 if (in_pcbrele_rlocked(inp))
2219 if (lookupflags & INPLOOKUP_WLOCKPCB)
2220 INP_WLOCK_ASSERT(inp);
2222 INP_RLOCK_ASSERT(inp);
2226 #endif /* PCBGROUP */
2229 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2230 * that the caller has locked the hash list, and will not perform any further
2231 * locking or reference operations on either the hash list or the connection.
2233 static struct inpcb *
2234 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2235 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2238 struct inpcbhead *head;
2239 struct inpcb *inp, *tmpinp;
2240 u_short fport = fport_arg, lport = lport_arg;
2242 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2243 ("%s: invalid lookup flags %d", __func__, lookupflags));
2245 INP_HASH_LOCK_ASSERT(pcbinfo);
2248 * First look for an exact match.
2251 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2252 pcbinfo->ipi_hashmask)];
2253 CK_LIST_FOREACH(inp, head, inp_hash) {
2255 /* XXX inp locking */
2256 if ((inp->inp_vflag & INP_IPV4) == 0)
2259 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2260 inp->inp_laddr.s_addr == laddr.s_addr &&
2261 inp->inp_fport == fport &&
2262 inp->inp_lport == lport) {
2264 * XXX We should be able to directly return
2265 * the inp here, without any checks.
2266 * Well unless both bound with SO_REUSEPORT?
2268 if (prison_flag(inp->inp_cred, PR_IP4))
2278 * Then look in lb group (for wildcard match).
2280 if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2281 (lookupflags & INPLOOKUP_WILDCARD)) {
2282 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2283 fport, lookupflags);
2290 * Then look for a wildcard match, if requested.
2292 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2293 struct inpcb *local_wild = NULL, *local_exact = NULL;
2295 struct inpcb *local_wild_mapped = NULL;
2297 struct inpcb *jail_wild = NULL;
2301 * Order of socket selection - we always prefer jails.
2302 * 1. jailed, non-wild.
2304 * 3. non-jailed, non-wild.
2305 * 4. non-jailed, wild.
2308 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2309 0, pcbinfo->ipi_hashmask)];
2310 CK_LIST_FOREACH(inp, head, inp_hash) {
2312 /* XXX inp locking */
2313 if ((inp->inp_vflag & INP_IPV4) == 0)
2316 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2317 inp->inp_lport != lport)
2320 injail = prison_flag(inp->inp_cred, PR_IP4);
2322 if (prison_check_ip4(inp->inp_cred,
2326 if (local_exact != NULL)
2330 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2335 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2337 /* XXX inp locking, NULL check */
2338 if (inp->inp_vflag & INP_IPV6PROTO)
2339 local_wild_mapped = inp;
2347 } /* LIST_FOREACH */
2348 if (jail_wild != NULL)
2350 if (local_exact != NULL)
2351 return (local_exact);
2352 if (local_wild != NULL)
2353 return (local_wild);
2355 if (local_wild_mapped != NULL)
2356 return (local_wild_mapped);
2358 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2364 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2365 * hash list lock, and will return the inpcb locked (i.e., requires
2366 * INPLOOKUP_LOCKPCB).
2368 static struct inpcb *
2369 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2370 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2376 INP_HASH_RLOCK(pcbinfo);
2377 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2378 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2380 if (lookupflags & INPLOOKUP_WLOCKPCB)
2381 locked = INP_TRY_WLOCK(inp);
2382 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2383 locked = INP_TRY_RLOCK(inp);
2385 panic("%s: locking bug", __func__);
2388 INP_HASH_RUNLOCK(pcbinfo);
2390 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2392 if (in_pcbrele_wlocked(inp))
2396 if (in_pcbrele_rlocked(inp))
2401 if (lookupflags & INPLOOKUP_WLOCKPCB)
2402 INP_WLOCK_ASSERT(inp);
2404 INP_RLOCK_ASSERT(inp);
2407 INP_HASH_RUNLOCK(pcbinfo);
2412 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2413 * from which a pre-calculated hash value may be extracted.
2415 * Possibly more of this logic should be in in_pcbgroup.c.
2418 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2419 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2421 #if defined(PCBGROUP) && !defined(RSS)
2422 struct inpcbgroup *pcbgroup;
2425 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2426 ("%s: invalid lookup flags %d", __func__, lookupflags));
2427 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2428 ("%s: LOCKPCB not set", __func__));
2431 * When not using RSS, use connection groups in preference to the
2432 * reservation table when looking up 4-tuples. When using RSS, just
2433 * use the reservation table, due to the cost of the Toeplitz hash
2436 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2437 * we could be doing RSS with a non-Toeplitz hash that is affordable
2440 #if defined(PCBGROUP) && !defined(RSS)
2441 if (in_pcbgroup_enabled(pcbinfo)) {
2442 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2444 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2445 laddr, lport, lookupflags, ifp));
2448 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2453 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2454 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2455 struct ifnet *ifp, struct mbuf *m)
2458 struct inpcbgroup *pcbgroup;
2461 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2462 ("%s: invalid lookup flags %d", __func__, lookupflags));
2463 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2464 ("%s: LOCKPCB not set", __func__));
2468 * If we can use a hardware-generated hash to look up the connection
2469 * group, use that connection group to find the inpcb. Otherwise
2470 * fall back on a software hash -- or the reservation table if we're
2473 * XXXRW: As above, that policy belongs in the pcbgroup code.
2475 if (in_pcbgroup_enabled(pcbinfo) &&
2476 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2477 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2478 m->m_pkthdr.flowid);
2479 if (pcbgroup != NULL)
2480 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2481 fport, laddr, lport, lookupflags, ifp));
2483 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2485 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2486 laddr, lport, lookupflags, ifp));
2490 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2496 * Insert PCB onto various hash lists.
2499 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2501 struct inpcbhead *pcbhash;
2502 struct inpcbporthead *pcbporthash;
2503 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2504 struct inpcbport *phd;
2505 u_int32_t hashkey_faddr;
2508 INP_WLOCK_ASSERT(inp);
2509 INP_HASH_WLOCK_ASSERT(pcbinfo);
2511 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2512 ("in_pcbinshash: INP_INHASHLIST"));
2515 if (inp->inp_vflag & INP_IPV6)
2516 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2519 hashkey_faddr = inp->inp_faddr.s_addr;
2521 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2522 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2524 pcbporthash = &pcbinfo->ipi_porthashbase[
2525 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2528 * Add entry to load balance group.
2529 * Only do this if SO_REUSEPORT_LB is set.
2531 so_options = inp_so_options(inp);
2532 if (so_options & SO_REUSEPORT_LB) {
2533 int ret = in_pcbinslbgrouphash(inp);
2535 /* pcb lb group malloc fail (ret=ENOBUFS). */
2541 * Go through port list and look for a head for this lport.
2543 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2544 if (phd->phd_port == inp->inp_lport)
2548 * If none exists, malloc one and tack it on.
2551 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2553 return (ENOBUFS); /* XXX */
2555 phd->phd_port = inp->inp_lport;
2556 CK_LIST_INIT(&phd->phd_pcblist);
2557 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2560 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2561 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2562 inp->inp_flags |= INP_INHASHLIST;
2564 if (do_pcbgroup_update)
2565 in_pcbgroup_update(inp);
2571 * For now, there are two public interfaces to insert an inpcb into the hash
2572 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2573 * is used only in the TCP syncache, where in_pcbinshash is called before the
2574 * full 4-tuple is set for the inpcb, and we don't want to install in the
2575 * pcbgroup until later.
2577 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2578 * connection groups, and partially initialised inpcbs should not be exposed
2579 * to either reservation hash tables or pcbgroups.
2582 in_pcbinshash(struct inpcb *inp)
2585 return (in_pcbinshash_internal(inp, 1));
2589 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2592 return (in_pcbinshash_internal(inp, 0));
2596 * Move PCB to the proper hash bucket when { faddr, fport } have been
2597 * changed. NOTE: This does not handle the case of the lport changing (the
2598 * hashed port list would have to be updated as well), so the lport must
2599 * not change after in_pcbinshash() has been called.
2602 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2604 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2605 struct inpcbhead *head;
2606 u_int32_t hashkey_faddr;
2608 INP_WLOCK_ASSERT(inp);
2609 INP_HASH_WLOCK_ASSERT(pcbinfo);
2611 KASSERT(inp->inp_flags & INP_INHASHLIST,
2612 ("in_pcbrehash: !INP_INHASHLIST"));
2615 if (inp->inp_vflag & INP_IPV6)
2616 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2619 hashkey_faddr = inp->inp_faddr.s_addr;
2621 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2622 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2624 CK_LIST_REMOVE(inp, inp_hash);
2625 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2629 in_pcbgroup_update_mbuf(inp, m);
2631 in_pcbgroup_update(inp);
2636 in_pcbrehash(struct inpcb *inp)
2639 in_pcbrehash_mbuf(inp, NULL);
2643 * Remove PCB from various lists.
2646 in_pcbremlists(struct inpcb *inp)
2648 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2651 if (pcbinfo == &V_tcbinfo) {
2652 INP_INFO_RLOCK_ASSERT(pcbinfo);
2654 INP_INFO_WLOCK_ASSERT(pcbinfo);
2658 INP_WLOCK_ASSERT(inp);
2659 INP_LIST_WLOCK_ASSERT(pcbinfo);
2661 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2662 if (inp->inp_flags & INP_INHASHLIST) {
2663 struct inpcbport *phd = inp->inp_phd;
2665 INP_HASH_WLOCK(pcbinfo);
2667 /* XXX: Only do if SO_REUSEPORT_LB set? */
2668 in_pcbremlbgrouphash(inp);
2670 CK_LIST_REMOVE(inp, inp_hash);
2671 CK_LIST_REMOVE(inp, inp_portlist);
2672 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2673 CK_LIST_REMOVE(phd, phd_hash);
2676 INP_HASH_WUNLOCK(pcbinfo);
2677 inp->inp_flags &= ~INP_INHASHLIST;
2679 CK_LIST_REMOVE(inp, inp_list);
2680 pcbinfo->ipi_count--;
2682 in_pcbgroup_remove(inp);
2687 * Check for alternatives when higher level complains
2688 * about service problems. For now, invalidate cached
2689 * routing information. If the route was created dynamically
2690 * (by a redirect), time to try a default gateway again.
2693 in_losing(struct inpcb *inp)
2696 RO_INVALIDATE_CACHE(&inp->inp_route);
2701 * A set label operation has occurred at the socket layer, propagate the
2702 * label change into the in_pcb for the socket.
2705 in_pcbsosetlabel(struct socket *so)
2710 inp = sotoinpcb(so);
2711 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2715 mac_inpcb_sosetlabel(so, inp);
2722 * ipport_tick runs once per second, determining if random port allocation
2723 * should be continued. If more than ipport_randomcps ports have been
2724 * allocated in the last second, then we return to sequential port
2725 * allocation. We return to random allocation only once we drop below
2726 * ipport_randomcps for at least ipport_randomtime seconds.
2729 ipport_tick(void *xtp)
2731 VNET_ITERATOR_DECL(vnet_iter);
2733 VNET_LIST_RLOCK_NOSLEEP();
2734 VNET_FOREACH(vnet_iter) {
2735 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2736 if (V_ipport_tcpallocs <=
2737 V_ipport_tcplastcount + V_ipport_randomcps) {
2738 if (V_ipport_stoprandom > 0)
2739 V_ipport_stoprandom--;
2741 V_ipport_stoprandom = V_ipport_randomtime;
2742 V_ipport_tcplastcount = V_ipport_tcpallocs;
2745 VNET_LIST_RUNLOCK_NOSLEEP();
2746 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2753 callout_stop(&ipport_tick_callout);
2757 * The ipport_callout should start running at about the time we attach the
2758 * inet or inet6 domains.
2761 ipport_tick_init(const void *unused __unused)
2764 /* Start ipport_tick. */
2765 callout_init(&ipport_tick_callout, 1);
2766 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2767 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2768 SHUTDOWN_PRI_DEFAULT);
2770 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2771 ipport_tick_init, NULL);
2774 inp_wlock(struct inpcb *inp)
2781 inp_wunlock(struct inpcb *inp)
2788 inp_rlock(struct inpcb *inp)
2795 inp_runlock(struct inpcb *inp)
2801 #ifdef INVARIANT_SUPPORT
2803 inp_lock_assert(struct inpcb *inp)
2806 INP_WLOCK_ASSERT(inp);
2810 inp_unlock_assert(struct inpcb *inp)
2813 INP_UNLOCK_ASSERT(inp);
2818 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2822 INP_INFO_WLOCK(&V_tcbinfo);
2823 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2828 INP_INFO_WUNLOCK(&V_tcbinfo);
2832 inp_inpcbtosocket(struct inpcb *inp)
2835 INP_WLOCK_ASSERT(inp);
2836 return (inp->inp_socket);
2840 inp_inpcbtotcpcb(struct inpcb *inp)
2843 INP_WLOCK_ASSERT(inp);
2844 return ((struct tcpcb *)inp->inp_ppcb);
2848 inp_ip_tos_get(const struct inpcb *inp)
2851 return (inp->inp_ip_tos);
2855 inp_ip_tos_set(struct inpcb *inp, int val)
2858 inp->inp_ip_tos = val;
2862 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2863 uint32_t *faddr, uint16_t *fp)
2866 INP_LOCK_ASSERT(inp);
2867 *laddr = inp->inp_laddr.s_addr;
2868 *faddr = inp->inp_faddr.s_addr;
2869 *lp = inp->inp_lport;
2870 *fp = inp->inp_fport;
2874 so_sotoinpcb(struct socket *so)
2877 return (sotoinpcb(so));
2881 so_sototcpcb(struct socket *so)
2884 return (sototcpcb(so));
2888 * Create an external-format (``xinpcb'') structure using the information in
2889 * the kernel-format in_pcb structure pointed to by inp. This is done to
2890 * reduce the spew of irrelevant information over this interface, to isolate
2891 * user code from changes in the kernel structure, and potentially to provide
2892 * information-hiding if we decide that some of this information should be
2893 * hidden from users.
2896 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2899 xi->xi_len = sizeof(struct xinpcb);
2900 if (inp->inp_socket)
2901 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2903 bzero(&xi->xi_socket, sizeof(struct xsocket));
2904 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2905 xi->inp_gencnt = inp->inp_gencnt;
2906 xi->inp_ppcb = inp->inp_ppcb;
2907 xi->inp_flow = inp->inp_flow;
2908 xi->inp_flowid = inp->inp_flowid;
2909 xi->inp_flowtype = inp->inp_flowtype;
2910 xi->inp_flags = inp->inp_flags;
2911 xi->inp_flags2 = inp->inp_flags2;
2912 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2913 xi->in6p_cksum = inp->in6p_cksum;
2914 xi->in6p_hops = inp->in6p_hops;
2915 xi->inp_ip_tos = inp->inp_ip_tos;
2916 xi->inp_vflag = inp->inp_vflag;
2917 xi->inp_ip_ttl = inp->inp_ip_ttl;
2918 xi->inp_ip_p = inp->inp_ip_p;
2919 xi->inp_ip_minttl = inp->inp_ip_minttl;
2924 db_print_indent(int indent)
2928 for (i = 0; i < indent; i++)
2933 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2935 char faddr_str[48], laddr_str[48];
2937 db_print_indent(indent);
2938 db_printf("%s at %p\n", name, inc);
2943 if (inc->inc_flags & INC_ISIPV6) {
2945 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2946 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2951 inet_ntoa_r(inc->inc_laddr, laddr_str);
2952 inet_ntoa_r(inc->inc_faddr, faddr_str);
2954 db_print_indent(indent);
2955 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2956 ntohs(inc->inc_lport));
2957 db_print_indent(indent);
2958 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2959 ntohs(inc->inc_fport));
2963 db_print_inpflags(int inp_flags)
2968 if (inp_flags & INP_RECVOPTS) {
2969 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2972 if (inp_flags & INP_RECVRETOPTS) {
2973 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2976 if (inp_flags & INP_RECVDSTADDR) {
2977 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2980 if (inp_flags & INP_ORIGDSTADDR) {
2981 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2984 if (inp_flags & INP_HDRINCL) {
2985 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2988 if (inp_flags & INP_HIGHPORT) {
2989 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2992 if (inp_flags & INP_LOWPORT) {
2993 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2996 if (inp_flags & INP_ANONPORT) {
2997 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3000 if (inp_flags & INP_RECVIF) {
3001 db_printf("%sINP_RECVIF", comma ? ", " : "");
3004 if (inp_flags & INP_MTUDISC) {
3005 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3008 if (inp_flags & INP_RECVTTL) {
3009 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3012 if (inp_flags & INP_DONTFRAG) {
3013 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3016 if (inp_flags & INP_RECVTOS) {
3017 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3020 if (inp_flags & IN6P_IPV6_V6ONLY) {
3021 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3024 if (inp_flags & IN6P_PKTINFO) {
3025 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3028 if (inp_flags & IN6P_HOPLIMIT) {
3029 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3032 if (inp_flags & IN6P_HOPOPTS) {
3033 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3036 if (inp_flags & IN6P_DSTOPTS) {
3037 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3040 if (inp_flags & IN6P_RTHDR) {
3041 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3044 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3045 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3048 if (inp_flags & IN6P_TCLASS) {
3049 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3052 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3053 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3056 if (inp_flags & INP_TIMEWAIT) {
3057 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3060 if (inp_flags & INP_ONESBCAST) {
3061 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3064 if (inp_flags & INP_DROPPED) {
3065 db_printf("%sINP_DROPPED", comma ? ", " : "");
3068 if (inp_flags & INP_SOCKREF) {
3069 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3072 if (inp_flags & IN6P_RFC2292) {
3073 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3076 if (inp_flags & IN6P_MTU) {
3077 db_printf("IN6P_MTU%s", comma ? ", " : "");
3083 db_print_inpvflag(u_char inp_vflag)
3088 if (inp_vflag & INP_IPV4) {
3089 db_printf("%sINP_IPV4", comma ? ", " : "");
3092 if (inp_vflag & INP_IPV6) {
3093 db_printf("%sINP_IPV6", comma ? ", " : "");
3096 if (inp_vflag & INP_IPV6PROTO) {
3097 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3103 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3106 db_print_indent(indent);
3107 db_printf("%s at %p\n", name, inp);
3111 db_print_indent(indent);
3112 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3114 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3116 db_print_indent(indent);
3117 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3118 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3120 db_print_indent(indent);
3121 db_printf("inp_label: %p inp_flags: 0x%x (",
3122 inp->inp_label, inp->inp_flags);
3123 db_print_inpflags(inp->inp_flags);
3126 db_print_indent(indent);
3127 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3129 db_print_inpvflag(inp->inp_vflag);
3132 db_print_indent(indent);
3133 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3134 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3136 db_print_indent(indent);
3138 if (inp->inp_vflag & INP_IPV6) {
3139 db_printf("in6p_options: %p in6p_outputopts: %p "
3140 "in6p_moptions: %p\n", inp->in6p_options,
3141 inp->in6p_outputopts, inp->in6p_moptions);
3142 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3143 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3148 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3149 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3150 inp->inp_options, inp->inp_moptions);
3153 db_print_indent(indent);
3154 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3155 (uintmax_t)inp->inp_gencnt);
3158 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3163 db_printf("usage: show inpcb <addr>\n");
3166 inp = (struct inpcb *)addr;
3168 db_print_inpcb(inp, "inpcb", 0);
3174 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3178 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3180 union if_snd_tag_modify_params params = {
3181 .rate_limit.max_rate = max_pacing_rate,
3183 struct m_snd_tag *mst;
3187 mst = inp->inp_snd_tag;
3195 if (ifp->if_snd_tag_modify == NULL) {
3198 error = ifp->if_snd_tag_modify(mst, ¶ms);
3204 * Query existing TX rate limit based on the existing
3205 * "inp->inp_snd_tag", if any.
3208 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3210 union if_snd_tag_query_params params = { };
3211 struct m_snd_tag *mst;
3215 mst = inp->inp_snd_tag;
3223 if (ifp->if_snd_tag_query == NULL) {
3226 error = ifp->if_snd_tag_query(mst, ¶ms);
3227 if (error == 0 && p_max_pacing_rate != NULL)
3228 *p_max_pacing_rate = params.rate_limit.max_rate;
3234 * Query existing TX queue level based on the existing
3235 * "inp->inp_snd_tag", if any.
3238 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3240 union if_snd_tag_query_params params = { };
3241 struct m_snd_tag *mst;
3245 mst = inp->inp_snd_tag;
3253 if (ifp->if_snd_tag_query == NULL)
3254 return (EOPNOTSUPP);
3256 error = ifp->if_snd_tag_query(mst, ¶ms);
3257 if (error == 0 && p_txqueue_level != NULL)
3258 *p_txqueue_level = params.rate_limit.queue_level;
3263 * Allocate a new TX rate limit send tag from the network interface
3264 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3267 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3268 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3270 union if_snd_tag_alloc_params params = {
3271 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3272 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3273 .rate_limit.hdr.flowid = flowid,
3274 .rate_limit.hdr.flowtype = flowtype,
3275 .rate_limit.max_rate = max_pacing_rate,
3279 INP_WLOCK_ASSERT(inp);
3281 if (inp->inp_snd_tag != NULL)
3284 if (ifp->if_snd_tag_alloc == NULL) {
3287 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3290 * At success increment the refcount on
3291 * the send tag's network interface:
3294 if_ref(inp->inp_snd_tag->ifp);
3300 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3304 in_pcbdetach_txrtlmt(struct inpcb *inp)
3306 struct m_snd_tag *mst;
3309 INP_WLOCK_ASSERT(inp);
3311 mst = inp->inp_snd_tag;
3312 inp->inp_snd_tag = NULL;
3322 * If the device was detached while we still had reference(s)
3323 * on the ifp, we assume if_snd_tag_free() was replaced with
3326 ifp->if_snd_tag_free(mst);
3328 /* release reference count on network interface */
3333 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3334 * is set in the fast path and will attach/detach/modify the TX rate
3335 * limit send tag based on the socket's so_max_pacing_rate value.
3338 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3340 struct socket *socket;
3341 uint32_t max_pacing_rate;
3348 socket = inp->inp_socket;
3352 if (!INP_WLOCKED(inp)) {
3354 * NOTE: If the write locking fails, we need to bail
3355 * out and use the non-ratelimited ring for the
3356 * transmit until there is a new chance to get the
3359 if (!INP_TRY_UPGRADE(inp))
3367 * NOTE: The so_max_pacing_rate value is read unlocked,
3368 * because atomic updates are not required since the variable
3369 * is checked at every mbuf we send. It is assumed that the
3370 * variable read itself will be atomic.
3372 max_pacing_rate = socket->so_max_pacing_rate;
3375 * NOTE: When attaching to a network interface a reference is
3376 * made to ensure the network interface doesn't go away until
3377 * all ratelimit connections are gone. The network interface
3378 * pointers compared below represent valid network interfaces,
3379 * except when comparing towards NULL.
3381 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3383 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3384 if (inp->inp_snd_tag != NULL)
3385 in_pcbdetach_txrtlmt(inp);
3387 } else if (inp->inp_snd_tag == NULL) {
3389 * In order to utilize packet pacing with RSS, we need
3390 * to wait until there is a valid RSS hash before we
3393 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3396 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3397 mb->m_pkthdr.flowid, max_pacing_rate);
3400 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3402 if (error == 0 || error == EOPNOTSUPP)
3403 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3409 * Track route changes for TX rate limiting.
3412 in_pcboutput_eagain(struct inpcb *inp)
3414 struct socket *socket;
3420 socket = inp->inp_socket;
3424 if (inp->inp_snd_tag == NULL)
3427 if (!INP_WLOCKED(inp)) {
3429 * NOTE: If the write locking fails, we need to bail
3430 * out and use the non-ratelimited ring for the
3431 * transmit until there is a new chance to get the
3434 if (!INP_TRY_UPGRADE(inp))
3441 /* detach rate limiting */
3442 in_pcbdetach_txrtlmt(inp);
3444 /* make sure new mbuf send tag allocation is made */
3445 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3450 #endif /* RATELIMIT */