2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1991, 1993, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2007-2009 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 #include "opt_ipsec.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/malloc.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
67 #include <sys/refcount.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/ip_var.h>
90 #include <netinet/tcp_var.h>
92 #include <netinet/tcp_hpts.h>
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
98 #include <netinet/in_var.h>
101 #include <netinet/ip6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
107 #include <netipsec/ipsec_support.h>
109 #include <security/mac/mac_framework.h>
111 #define INPCBLBGROUP_SIZMIN 8
112 #define INPCBLBGROUP_SIZMAX 256
114 static struct callout ipport_tick_callout;
117 * These configure the range of local port addresses assigned to
118 * "unspecified" outgoing connections/packets/whatever.
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
128 * Reserved ports accessible only to root. There are significant
129 * security considerations that must be accounted for when changing these,
130 * but the security benefits can be great. Please be careful.
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
145 static void in_pcbremlists(struct inpcb *inp);
147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 struct in_addr faddr, u_int fport_arg,
149 struct in_addr laddr, u_int lport_arg,
150 int lookupflags, struct ifnet *ifp);
152 #define RANGECHK(var, min, max) \
153 if ((var) < (min)) { (var) = (min); } \
154 else if ((var) > (max)) { (var) = (max); }
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
161 error = sysctl_handle_int(oidp, arg1, arg2, req);
163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 &VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 CTLFLAG_VNET | CTLFLAG_RW,
206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 "allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(ipport_randomtime), 0,
211 "Minimum time to keep sequental port "
212 "allocation before switching to a random one");
216 * in_pcb.c: manage the Protocol Control Blocks.
218 * NOTE: It is assumed that most of these functions will be called with
219 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
220 * functions often modify hash chains or addresses in pcbs.
223 static struct inpcblbgroup *
224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
225 uint16_t port, const union in_dependaddr *addr, int size)
227 struct inpcblbgroup *grp;
230 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
231 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
234 grp->il_vflag = vflag;
235 grp->il_lport = port;
236 grp->il_dependladdr = *addr;
237 grp->il_inpsiz = size;
238 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
243 in_pcblbgroup_free_deferred(epoch_context_t ctx)
245 struct inpcblbgroup *grp;
247 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
252 in_pcblbgroup_free(struct inpcblbgroup *grp)
255 CK_LIST_REMOVE(grp, il_list);
256 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
257 in_pcblbgroup_free_deferred);
260 static struct inpcblbgroup *
261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
262 struct inpcblbgroup *old_grp, int size)
264 struct inpcblbgroup *grp;
267 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
268 old_grp->il_lport, &old_grp->il_dependladdr, size);
272 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
273 ("invalid new local group size %d and old local group count %d",
274 grp->il_inpsiz, old_grp->il_inpcnt));
276 for (i = 0; i < old_grp->il_inpcnt; ++i)
277 grp->il_inp[i] = old_grp->il_inp[i];
278 grp->il_inpcnt = old_grp->il_inpcnt;
279 in_pcblbgroup_free(old_grp);
284 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
285 * and shrink group if possible.
288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
291 struct inpcblbgroup *grp, *new_grp;
294 for (; i + 1 < grp->il_inpcnt; ++i)
295 grp->il_inp[i] = grp->il_inp[i + 1];
298 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
299 grp->il_inpcnt <= grp->il_inpsiz / 4) {
300 /* Shrink this group. */
301 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
308 * Add PCB to load balance group for SO_REUSEPORT_LB option.
311 in_pcbinslbgrouphash(struct inpcb *inp)
313 const static struct timeval interval = { 60, 0 };
314 static struct timeval lastprint;
315 struct inpcbinfo *pcbinfo;
316 struct inpcblbgrouphead *hdr;
317 struct inpcblbgroup *grp;
320 pcbinfo = inp->inp_pcbinfo;
322 INP_WLOCK_ASSERT(inp);
323 INP_HASH_WLOCK_ASSERT(pcbinfo);
326 * Don't allow jailed socket to join local group.
328 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
333 * Don't allow IPv4 mapped INET6 wild socket.
335 if ((inp->inp_vflag & INP_IPV4) &&
336 inp->inp_laddr.s_addr == INADDR_ANY &&
337 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
342 idx = INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
343 pcbinfo->ipi_lbgrouphashmask);
344 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
345 CK_LIST_FOREACH(grp, hdr, il_list) {
346 if (grp->il_vflag == inp->inp_vflag &&
347 grp->il_lport == inp->inp_lport &&
348 memcmp(&grp->il_dependladdr,
349 &inp->inp_inc.inc_ie.ie_dependladdr,
350 sizeof(grp->il_dependladdr)) == 0)
354 /* Create new load balance group. */
355 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
356 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
357 INPCBLBGROUP_SIZMIN);
360 } else if (grp->il_inpcnt == grp->il_inpsiz) {
361 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
362 if (ratecheck(&lastprint, &interval))
363 printf("lb group port %d, limit reached\n",
364 ntohs(grp->il_lport));
368 /* Expand this local group. */
369 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
374 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
375 ("invalid local group size %d and count %d", grp->il_inpsiz,
378 grp->il_inp[grp->il_inpcnt] = inp;
384 * Remove PCB from load balance group.
387 in_pcbremlbgrouphash(struct inpcb *inp)
389 struct inpcbinfo *pcbinfo;
390 struct inpcblbgrouphead *hdr;
391 struct inpcblbgroup *grp;
394 pcbinfo = inp->inp_pcbinfo;
396 INP_WLOCK_ASSERT(inp);
397 INP_HASH_WLOCK_ASSERT(pcbinfo);
399 hdr = &pcbinfo->ipi_lbgrouphashbase[
400 INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
401 pcbinfo->ipi_lbgrouphashmask)];
403 CK_LIST_FOREACH(grp, hdr, il_list) {
404 for (i = 0; i < grp->il_inpcnt; ++i) {
405 if (grp->il_inp[i] != inp)
408 if (grp->il_inpcnt == 1) {
409 /* We are the last, free this local group. */
410 in_pcblbgroup_free(grp);
412 /* Pull up inpcbs, shrink group if possible. */
413 in_pcblbgroup_reorder(hdr, &grp, i);
421 * Different protocols initialize their inpcbs differently - giving
422 * different name to the lock. But they all are disposed the same.
425 inpcb_fini(void *mem, int size)
427 struct inpcb *inp = mem;
429 INP_LOCK_DESTROY(inp);
433 * Initialize an inpcbinfo -- we should be able to reduce the number of
437 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
438 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
439 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
442 INP_INFO_LOCK_INIT(pcbinfo, name);
443 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
444 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
446 pcbinfo->ipi_vnet = curvnet;
448 pcbinfo->ipi_listhead = listhead;
449 CK_LIST_INIT(pcbinfo->ipi_listhead);
450 pcbinfo->ipi_count = 0;
451 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
452 &pcbinfo->ipi_hashmask);
453 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
454 &pcbinfo->ipi_porthashmask);
455 pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
456 &pcbinfo->ipi_lbgrouphashmask);
458 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
460 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
461 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
462 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
463 uma_zone_set_warning(pcbinfo->ipi_zone,
464 "kern.ipc.maxsockets limit reached");
468 * Destroy an inpcbinfo.
471 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
474 KASSERT(pcbinfo->ipi_count == 0,
475 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
477 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
478 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
479 pcbinfo->ipi_porthashmask);
480 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
481 pcbinfo->ipi_lbgrouphashmask);
483 in_pcbgroup_destroy(pcbinfo);
485 uma_zdestroy(pcbinfo->ipi_zone);
486 INP_LIST_LOCK_DESTROY(pcbinfo);
487 INP_HASH_LOCK_DESTROY(pcbinfo);
488 INP_INFO_LOCK_DESTROY(pcbinfo);
492 * Allocate a PCB and associate it with the socket.
493 * On success return with the PCB locked.
496 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
502 if (pcbinfo == &V_tcbinfo) {
503 INP_INFO_RLOCK_ASSERT(pcbinfo);
505 INP_INFO_WLOCK_ASSERT(pcbinfo);
510 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
513 bzero(&inp->inp_start_zero, inp_zero_size);
514 inp->inp_pcbinfo = pcbinfo;
515 inp->inp_socket = so;
516 inp->inp_cred = crhold(so->so_cred);
517 inp->inp_inc.inc_fibnum = so->so_fibnum;
519 error = mac_inpcb_init(inp, M_NOWAIT);
522 mac_inpcb_create(so, inp);
524 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
525 error = ipsec_init_pcbpolicy(inp);
528 mac_inpcb_destroy(inp);
534 if (INP_SOCKAF(so) == AF_INET6) {
535 inp->inp_vflag |= INP_IPV6PROTO;
537 inp->inp_flags |= IN6P_IPV6_V6ONLY;
541 INP_LIST_WLOCK(pcbinfo);
542 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
543 pcbinfo->ipi_count++;
544 so->so_pcb = (caddr_t)inp;
546 if (V_ip6_auto_flowlabel)
547 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
549 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
550 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
553 * Routes in inpcb's can cache L2 as well; they are guaranteed
556 inp->inp_route.ro_flags = RT_LLE_CACHE;
557 INP_LIST_WUNLOCK(pcbinfo);
558 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
561 crfree(inp->inp_cred);
562 uma_zfree(pcbinfo->ipi_zone, inp);
570 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
574 INP_WLOCK_ASSERT(inp);
575 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
577 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
579 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
580 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
581 &inp->inp_lport, cred);
584 if (in_pcbinshash(inp) != 0) {
585 inp->inp_laddr.s_addr = INADDR_ANY;
590 inp->inp_flags |= INP_ANONPORT;
596 * Select a local port (number) to use.
598 #if defined(INET) || defined(INET6)
600 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
601 struct ucred *cred, int lookupflags)
603 struct inpcbinfo *pcbinfo;
604 struct inpcb *tmpinp;
605 unsigned short *lastport;
606 int count, dorandom, error;
607 u_short aux, first, last, lport;
609 struct in_addr laddr;
612 pcbinfo = inp->inp_pcbinfo;
615 * Because no actual state changes occur here, a global write lock on
616 * the pcbinfo isn't required.
618 INP_LOCK_ASSERT(inp);
619 INP_HASH_LOCK_ASSERT(pcbinfo);
621 if (inp->inp_flags & INP_HIGHPORT) {
622 first = V_ipport_hifirstauto; /* sysctl */
623 last = V_ipport_hilastauto;
624 lastport = &pcbinfo->ipi_lasthi;
625 } else if (inp->inp_flags & INP_LOWPORT) {
626 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
629 first = V_ipport_lowfirstauto; /* 1023 */
630 last = V_ipport_lowlastauto; /* 600 */
631 lastport = &pcbinfo->ipi_lastlow;
633 first = V_ipport_firstauto; /* sysctl */
634 last = V_ipport_lastauto;
635 lastport = &pcbinfo->ipi_lastport;
638 * For UDP(-Lite), use random port allocation as long as the user
639 * allows it. For TCP (and as of yet unknown) connections,
640 * use random port allocation only if the user allows it AND
641 * ipport_tick() allows it.
643 if (V_ipport_randomized &&
644 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
645 pcbinfo == &V_ulitecbinfo))
650 * It makes no sense to do random port allocation if
651 * we have the only port available.
655 /* Make sure to not include UDP(-Lite) packets in the count. */
656 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
657 V_ipport_tcpallocs++;
659 * Instead of having two loops further down counting up or down
660 * make sure that first is always <= last and go with only one
661 * code path implementing all logic.
670 /* Make the compiler happy. */
672 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
673 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
678 tmpinp = NULL; /* Make compiler happy. */
682 *lastport = first + (arc4random() % (last - first));
684 count = last - first;
687 if (count-- < 0) /* completely used? */
688 return (EADDRNOTAVAIL);
690 if (*lastport < first || *lastport > last)
692 lport = htons(*lastport);
695 if ((inp->inp_vflag & INP_IPV6) != 0)
696 tmpinp = in6_pcblookup_local(pcbinfo,
697 &inp->in6p_laddr, lport, lookupflags, cred);
699 #if defined(INET) && defined(INET6)
703 tmpinp = in_pcblookup_local(pcbinfo, laddr,
704 lport, lookupflags, cred);
706 } while (tmpinp != NULL);
709 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
710 laddrp->s_addr = laddr.s_addr;
718 * Return cached socket options.
721 inp_so_options(const struct inpcb *inp)
727 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
728 so_options |= SO_REUSEPORT_LB;
729 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
730 so_options |= SO_REUSEPORT;
731 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
732 so_options |= SO_REUSEADDR;
735 #endif /* INET || INET6 */
738 * Check if a new BINDMULTI socket is allowed to be created.
740 * ni points to the new inp.
741 * oi points to the exisitng inp.
743 * This checks whether the existing inp also has BINDMULTI and
744 * whether the credentials match.
747 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
749 /* Check permissions match */
750 if ((ni->inp_flags2 & INP_BINDMULTI) &&
751 (ni->inp_cred->cr_uid !=
752 oi->inp_cred->cr_uid))
755 /* Check the existing inp has BINDMULTI set */
756 if ((ni->inp_flags2 & INP_BINDMULTI) &&
757 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
761 * We're okay - either INP_BINDMULTI isn't set on ni, or
762 * it is and it matches the checks.
769 * Set up a bind operation on a PCB, performing port allocation
770 * as required, but do not actually modify the PCB. Callers can
771 * either complete the bind by setting inp_laddr/inp_lport and
772 * calling in_pcbinshash(), or they can just use the resulting
773 * port and address to authorise the sending of a once-off packet.
775 * On error, the values of *laddrp and *lportp are not changed.
778 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
779 u_short *lportp, struct ucred *cred)
781 struct socket *so = inp->inp_socket;
782 struct sockaddr_in *sin;
783 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
784 struct in_addr laddr;
786 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
790 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
791 * so that we don't have to add to the (already messy) code below.
793 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
796 * No state changes, so read locks are sufficient here.
798 INP_LOCK_ASSERT(inp);
799 INP_HASH_LOCK_ASSERT(pcbinfo);
801 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
802 return (EADDRNOTAVAIL);
803 laddr.s_addr = *laddrp;
804 if (nam != NULL && laddr.s_addr != INADDR_ANY)
806 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
807 lookupflags = INPLOOKUP_WILDCARD;
809 if ((error = prison_local_ip4(cred, &laddr)) != 0)
812 sin = (struct sockaddr_in *)nam;
813 if (nam->sa_len != sizeof (*sin))
817 * We should check the family, but old programs
818 * incorrectly fail to initialize it.
820 if (sin->sin_family != AF_INET)
821 return (EAFNOSUPPORT);
823 error = prison_local_ip4(cred, &sin->sin_addr);
826 if (sin->sin_port != *lportp) {
827 /* Don't allow the port to change. */
830 lport = sin->sin_port;
832 /* NB: lport is left as 0 if the port isn't being changed. */
833 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
835 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
836 * allow complete duplication of binding if
837 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
838 * and a multicast address is bound on both
839 * new and duplicated sockets.
841 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
842 reuseport = SO_REUSEADDR|SO_REUSEPORT;
844 * XXX: How to deal with SO_REUSEPORT_LB here?
845 * Treat same as SO_REUSEPORT for now.
847 if ((so->so_options &
848 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
849 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
850 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
851 sin->sin_port = 0; /* yech... */
852 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
854 * Is the address a local IP address?
855 * If INP_BINDANY is set, then the socket may be bound
856 * to any endpoint address, local or not.
858 if ((inp->inp_flags & INP_BINDANY) == 0 &&
859 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
860 return (EADDRNOTAVAIL);
862 laddr = sin->sin_addr;
868 if (ntohs(lport) <= V_ipport_reservedhigh &&
869 ntohs(lport) >= V_ipport_reservedlow &&
870 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
873 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
874 priv_check_cred(inp->inp_cred,
875 PRIV_NETINET_REUSEPORT, 0) != 0) {
876 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
877 lport, INPLOOKUP_WILDCARD, cred);
880 * This entire block sorely needs a rewrite.
883 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
884 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
885 (so->so_type != SOCK_STREAM ||
886 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
887 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
888 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
889 (t->inp_flags2 & INP_REUSEPORT) ||
890 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
891 (inp->inp_cred->cr_uid !=
892 t->inp_cred->cr_uid))
896 * If the socket is a BINDMULTI socket, then
897 * the credentials need to match and the
898 * original socket also has to have been bound
901 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
904 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
905 lport, lookupflags, cred);
906 if (t && (t->inp_flags & INP_TIMEWAIT)) {
908 * XXXRW: If an incpb has had its timewait
909 * state recycled, we treat the address as
910 * being in use (for now). This is better
911 * than a panic, but not desirable.
915 ((reuseport & tw->tw_so_options) == 0 &&
917 tw->tw_so_options) == 0)) {
921 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
922 (reuseport & inp_so_options(t)) == 0 &&
923 (reuseport_lb & inp_so_options(t)) == 0) {
925 if (ntohl(sin->sin_addr.s_addr) !=
927 ntohl(t->inp_laddr.s_addr) !=
929 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
930 (t->inp_vflag & INP_IPV6PROTO) == 0)
933 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
941 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
946 *laddrp = laddr.s_addr;
952 * Connect from a socket to a specified address.
953 * Both address and port must be specified in argument sin.
954 * If don't have a local address for this socket yet,
958 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
959 struct ucred *cred, struct mbuf *m)
961 u_short lport, fport;
962 in_addr_t laddr, faddr;
965 INP_WLOCK_ASSERT(inp);
966 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
968 lport = inp->inp_lport;
969 laddr = inp->inp_laddr.s_addr;
970 anonport = (lport == 0);
971 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
976 /* Do the initial binding of the local address if required. */
977 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
978 inp->inp_lport = lport;
979 inp->inp_laddr.s_addr = laddr;
980 if (in_pcbinshash(inp) != 0) {
981 inp->inp_laddr.s_addr = INADDR_ANY;
987 /* Commit the remaining changes. */
988 inp->inp_lport = lport;
989 inp->inp_laddr.s_addr = laddr;
990 inp->inp_faddr.s_addr = faddr;
991 inp->inp_fport = fport;
992 in_pcbrehash_mbuf(inp, m);
995 inp->inp_flags |= INP_ANONPORT;
1000 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1003 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1007 * Do proper source address selection on an unbound socket in case
1008 * of connect. Take jails into account as well.
1011 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1015 struct sockaddr *sa;
1016 struct sockaddr_in *sin;
1020 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1022 * Bypass source address selection and use the primary jail IP
1025 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1029 bzero(&sro, sizeof(sro));
1031 sin = (struct sockaddr_in *)&sro.ro_dst;
1032 sin->sin_family = AF_INET;
1033 sin->sin_len = sizeof(struct sockaddr_in);
1034 sin->sin_addr.s_addr = faddr->s_addr;
1037 * If route is known our src addr is taken from the i/f,
1040 * Find out route to destination.
1042 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1043 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1046 * If we found a route, use the address corresponding to
1047 * the outgoing interface.
1049 * Otherwise assume faddr is reachable on a directly connected
1050 * network and try to find a corresponding interface to take
1051 * the source address from.
1054 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1055 struct in_ifaddr *ia;
1058 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1059 inp->inp_socket->so_fibnum));
1061 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1062 inp->inp_socket->so_fibnum));
1066 error = ENETUNREACH;
1070 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1071 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1077 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1080 if (sa->sa_family != AF_INET)
1082 sin = (struct sockaddr_in *)sa;
1083 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1084 ia = (struct in_ifaddr *)ifa;
1089 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1093 /* 3. As a last resort return the 'default' jail address. */
1094 error = prison_get_ip4(cred, laddr);
1099 * If the outgoing interface on the route found is not
1100 * a loopback interface, use the address from that interface.
1101 * In case of jails do those three steps:
1102 * 1. check if the interface address belongs to the jail. If so use it.
1103 * 2. check if we have any address on the outgoing interface
1104 * belonging to this jail. If so use it.
1105 * 3. as a last resort return the 'default' jail address.
1107 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1108 struct in_ifaddr *ia;
1111 /* If not jailed, use the default returned. */
1112 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1113 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1114 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1119 /* 1. Check if the iface address belongs to the jail. */
1120 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1121 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1122 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1123 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1128 * 2. Check if we have any address on the outgoing interface
1129 * belonging to this jail.
1132 ifp = sro.ro_rt->rt_ifp;
1133 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1135 if (sa->sa_family != AF_INET)
1137 sin = (struct sockaddr_in *)sa;
1138 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1139 ia = (struct in_ifaddr *)ifa;
1144 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1148 /* 3. As a last resort return the 'default' jail address. */
1149 error = prison_get_ip4(cred, laddr);
1154 * The outgoing interface is marked with 'loopback net', so a route
1155 * to ourselves is here.
1156 * Try to find the interface of the destination address and then
1157 * take the address from there. That interface is not necessarily
1158 * a loopback interface.
1159 * In case of jails, check that it is an address of the jail
1160 * and if we cannot find, fall back to the 'default' jail address.
1162 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1163 struct sockaddr_in sain;
1164 struct in_ifaddr *ia;
1166 bzero(&sain, sizeof(struct sockaddr_in));
1167 sain.sin_family = AF_INET;
1168 sain.sin_len = sizeof(struct sockaddr_in);
1169 sain.sin_addr.s_addr = faddr->s_addr;
1171 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1172 inp->inp_socket->so_fibnum));
1174 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1175 inp->inp_socket->so_fibnum));
1177 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1179 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1181 error = ENETUNREACH;
1184 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1194 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1196 if (sa->sa_family != AF_INET)
1198 sin = (struct sockaddr_in *)sa;
1199 if (prison_check_ip4(cred,
1200 &sin->sin_addr) == 0) {
1201 ia = (struct in_ifaddr *)ifa;
1206 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1211 /* 3. As a last resort return the 'default' jail address. */
1212 error = prison_get_ip4(cred, laddr);
1218 if (sro.ro_rt != NULL)
1224 * Set up for a connect from a socket to the specified address.
1225 * On entry, *laddrp and *lportp should contain the current local
1226 * address and port for the PCB; these are updated to the values
1227 * that should be placed in inp_laddr and inp_lport to complete
1230 * On success, *faddrp and *fportp will be set to the remote address
1231 * and port. These are not updated in the error case.
1233 * If the operation fails because the connection already exists,
1234 * *oinpp will be set to the PCB of that connection so that the
1235 * caller can decide to override it. In all other cases, *oinpp
1239 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1240 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1241 struct inpcb **oinpp, struct ucred *cred)
1243 struct rm_priotracker in_ifa_tracker;
1244 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1245 struct in_ifaddr *ia;
1247 struct in_addr laddr, faddr;
1248 u_short lport, fport;
1252 * Because a global state change doesn't actually occur here, a read
1253 * lock is sufficient.
1255 INP_LOCK_ASSERT(inp);
1256 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1260 if (nam->sa_len != sizeof (*sin))
1262 if (sin->sin_family != AF_INET)
1263 return (EAFNOSUPPORT);
1264 if (sin->sin_port == 0)
1265 return (EADDRNOTAVAIL);
1266 laddr.s_addr = *laddrp;
1268 faddr = sin->sin_addr;
1269 fport = sin->sin_port;
1271 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1273 * If the destination address is INADDR_ANY,
1274 * use the primary local address.
1275 * If the supplied address is INADDR_BROADCAST,
1276 * and the primary interface supports broadcast,
1277 * choose the broadcast address for that interface.
1279 if (faddr.s_addr == INADDR_ANY) {
1280 IN_IFADDR_RLOCK(&in_ifa_tracker);
1282 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1283 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1285 (error = prison_get_ip4(cred, &faddr)) != 0)
1287 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1288 IN_IFADDR_RLOCK(&in_ifa_tracker);
1289 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1291 faddr = satosin(&CK_STAILQ_FIRST(
1292 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1293 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1296 if (laddr.s_addr == INADDR_ANY) {
1297 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1299 * If the destination address is multicast and an outgoing
1300 * interface has been set as a multicast option, prefer the
1301 * address of that interface as our source address.
1303 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1304 inp->inp_moptions != NULL) {
1305 struct ip_moptions *imo;
1308 imo = inp->inp_moptions;
1309 if (imo->imo_multicast_ifp != NULL) {
1310 ifp = imo->imo_multicast_ifp;
1311 IN_IFADDR_RLOCK(&in_ifa_tracker);
1312 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1313 if ((ia->ia_ifp == ifp) &&
1315 prison_check_ip4(cred,
1316 &ia->ia_addr.sin_addr) == 0))
1320 error = EADDRNOTAVAIL;
1322 laddr = ia->ia_addr.sin_addr;
1325 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1331 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1332 laddr, lport, 0, NULL);
1336 return (EADDRINUSE);
1339 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1344 *laddrp = laddr.s_addr;
1346 *faddrp = faddr.s_addr;
1352 in_pcbdisconnect(struct inpcb *inp)
1355 INP_WLOCK_ASSERT(inp);
1356 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1358 inp->inp_faddr.s_addr = INADDR_ANY;
1365 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1366 * For most protocols, this will be invoked immediately prior to calling
1367 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1368 * socket, in which case in_pcbfree() is deferred.
1371 in_pcbdetach(struct inpcb *inp)
1374 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1377 if (inp->inp_snd_tag != NULL)
1378 in_pcbdetach_txrtlmt(inp);
1380 inp->inp_socket->so_pcb = NULL;
1381 inp->inp_socket = NULL;
1385 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1386 * stability of an inpcb pointer despite the inpcb lock being released. This
1387 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1388 * but where the inpcb lock may already held, or when acquiring a reference
1391 * in_pcbref() should be used only to provide brief memory stability, and
1392 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1393 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1394 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1395 * lock and rele are the *only* safe operations that may be performed on the
1398 * While the inpcb will not be freed, releasing the inpcb lock means that the
1399 * connection's state may change, so the caller should be careful to
1400 * revalidate any cached state on reacquiring the lock. Drop the reference
1401 * using in_pcbrele().
1404 in_pcbref(struct inpcb *inp)
1407 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1409 refcount_acquire(&inp->inp_refcount);
1413 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1414 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1415 * return a flag indicating whether or not the inpcb remains valid. If it is
1416 * valid, we return with the inpcb lock held.
1418 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1419 * reference on an inpcb. Historically more work was done here (actually, in
1420 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1421 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1422 * about memory stability (and continued use of the write lock).
1425 in_pcbrele_rlocked(struct inpcb *inp)
1427 struct inpcbinfo *pcbinfo;
1429 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1431 INP_RLOCK_ASSERT(inp);
1433 if (refcount_release(&inp->inp_refcount) == 0) {
1435 * If the inpcb has been freed, let the caller know, even if
1436 * this isn't the last reference.
1438 if (inp->inp_flags2 & INP_FREED) {
1445 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1447 if (inp->inp_in_hpts || inp->inp_in_input) {
1448 struct tcp_hpts_entry *hpts;
1450 * We should not be on the hpts at
1451 * this point in any form. we must
1452 * get the lock to be sure.
1454 hpts = tcp_hpts_lock(inp);
1455 if (inp->inp_in_hpts)
1456 panic("Hpts:%p inp:%p at free still on hpts",
1458 mtx_unlock(&hpts->p_mtx);
1459 hpts = tcp_input_lock(inp);
1460 if (inp->inp_in_input)
1461 panic("Hpts:%p inp:%p at free still on input hpts",
1463 mtx_unlock(&hpts->p_mtx);
1467 pcbinfo = inp->inp_pcbinfo;
1468 uma_zfree(pcbinfo->ipi_zone, inp);
1473 in_pcbrele_wlocked(struct inpcb *inp)
1475 struct inpcbinfo *pcbinfo;
1477 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1479 INP_WLOCK_ASSERT(inp);
1481 if (refcount_release(&inp->inp_refcount) == 0) {
1483 * If the inpcb has been freed, let the caller know, even if
1484 * this isn't the last reference.
1486 if (inp->inp_flags2 & INP_FREED) {
1493 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1495 if (inp->inp_in_hpts || inp->inp_in_input) {
1496 struct tcp_hpts_entry *hpts;
1498 * We should not be on the hpts at
1499 * this point in any form. we must
1500 * get the lock to be sure.
1502 hpts = tcp_hpts_lock(inp);
1503 if (inp->inp_in_hpts)
1504 panic("Hpts:%p inp:%p at free still on hpts",
1506 mtx_unlock(&hpts->p_mtx);
1507 hpts = tcp_input_lock(inp);
1508 if (inp->inp_in_input)
1509 panic("Hpts:%p inp:%p at free still on input hpts",
1511 mtx_unlock(&hpts->p_mtx);
1515 pcbinfo = inp->inp_pcbinfo;
1516 uma_zfree(pcbinfo->ipi_zone, inp);
1521 * Temporary wrapper.
1524 in_pcbrele(struct inpcb *inp)
1527 return (in_pcbrele_wlocked(inp));
1531 in_pcblist_rele_rlocked(epoch_context_t ctx)
1533 struct in_pcblist *il;
1535 struct inpcbinfo *pcbinfo;
1538 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1539 pcbinfo = il->il_pcbinfo;
1541 INP_INFO_WLOCK(pcbinfo);
1542 for (i = 0; i < n; i++) {
1543 inp = il->il_inp_list[i];
1545 if (!in_pcbrele_rlocked(inp))
1548 INP_INFO_WUNLOCK(pcbinfo);
1553 inpcbport_free(epoch_context_t ctx)
1555 struct inpcbport *phd;
1557 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1562 in_pcbfree_deferred(epoch_context_t ctx)
1565 int released __unused;
1567 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1571 struct ip_moptions *imo = inp->inp_moptions;
1572 inp->inp_moptions = NULL;
1574 /* XXXRW: Do as much as possible here. */
1575 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1576 if (inp->inp_sp != NULL)
1577 ipsec_delete_pcbpolicy(inp);
1580 struct ip6_moptions *im6o = NULL;
1581 if (inp->inp_vflag & INP_IPV6PROTO) {
1582 ip6_freepcbopts(inp->in6p_outputopts);
1583 im6o = inp->in6p_moptions;
1584 inp->in6p_moptions = NULL;
1587 if (inp->inp_options)
1588 (void)m_free(inp->inp_options);
1590 crfree(inp->inp_cred);
1592 mac_inpcb_destroy(inp);
1594 released = in_pcbrele_wlocked(inp);
1597 ip6_freemoptions(im6o);
1600 inp_freemoptions(imo);
1605 * Unconditionally schedule an inpcb to be freed by decrementing its
1606 * reference count, which should occur only after the inpcb has been detached
1607 * from its socket. If another thread holds a temporary reference (acquired
1608 * using in_pcbref()) then the free is deferred until that reference is
1609 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1610 * work, including removal from global lists, is done in this context, where
1611 * the pcbinfo lock is held.
1614 in_pcbfree(struct inpcb *inp)
1616 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1618 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1619 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1620 ("%s: called twice for pcb %p", __func__, inp));
1621 if (inp->inp_flags2 & INP_FREED) {
1627 if (pcbinfo == &V_tcbinfo) {
1628 INP_INFO_LOCK_ASSERT(pcbinfo);
1630 INP_INFO_WLOCK_ASSERT(pcbinfo);
1633 INP_WLOCK_ASSERT(inp);
1634 INP_LIST_WLOCK(pcbinfo);
1635 in_pcbremlists(inp);
1636 INP_LIST_WUNLOCK(pcbinfo);
1637 RO_INVALIDATE_CACHE(&inp->inp_route);
1638 /* mark as destruction in progress */
1639 inp->inp_flags2 |= INP_FREED;
1641 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1645 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1646 * port reservation, and preventing it from being returned by inpcb lookups.
1648 * It is used by TCP to mark an inpcb as unused and avoid future packet
1649 * delivery or event notification when a socket remains open but TCP has
1650 * closed. This might occur as a result of a shutdown()-initiated TCP close
1651 * or a RST on the wire, and allows the port binding to be reused while still
1652 * maintaining the invariant that so_pcb always points to a valid inpcb until
1655 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1656 * in_pcbnotifyall() and in_pcbpurgeif0()?
1659 in_pcbdrop(struct inpcb *inp)
1662 INP_WLOCK_ASSERT(inp);
1664 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1665 MPASS(inp->inp_refcount > 1);
1669 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1672 inp->inp_flags |= INP_DROPPED;
1673 if (inp->inp_flags & INP_INHASHLIST) {
1674 struct inpcbport *phd = inp->inp_phd;
1676 INP_HASH_WLOCK(inp->inp_pcbinfo);
1677 in_pcbremlbgrouphash(inp);
1678 CK_LIST_REMOVE(inp, inp_hash);
1679 CK_LIST_REMOVE(inp, inp_portlist);
1680 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1681 CK_LIST_REMOVE(phd, phd_hash);
1682 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1684 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1685 inp->inp_flags &= ~INP_INHASHLIST;
1687 in_pcbgroup_remove(inp);
1694 * Common routines to return the socket addresses associated with inpcbs.
1697 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1699 struct sockaddr_in *sin;
1701 sin = malloc(sizeof *sin, M_SONAME,
1703 sin->sin_family = AF_INET;
1704 sin->sin_len = sizeof(*sin);
1705 sin->sin_addr = *addr_p;
1706 sin->sin_port = port;
1708 return (struct sockaddr *)sin;
1712 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1715 struct in_addr addr;
1718 inp = sotoinpcb(so);
1719 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1722 port = inp->inp_lport;
1723 addr = inp->inp_laddr;
1726 *nam = in_sockaddr(port, &addr);
1731 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1734 struct in_addr addr;
1737 inp = sotoinpcb(so);
1738 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1741 port = inp->inp_fport;
1742 addr = inp->inp_faddr;
1745 *nam = in_sockaddr(port, &addr);
1750 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1751 struct inpcb *(*notify)(struct inpcb *, int))
1753 struct inpcb *inp, *inp_temp;
1755 INP_INFO_WLOCK(pcbinfo);
1756 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1759 if ((inp->inp_vflag & INP_IPV4) == 0) {
1764 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1765 inp->inp_socket == NULL) {
1769 if ((*notify)(inp, errno))
1772 INP_INFO_WUNLOCK(pcbinfo);
1776 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1779 struct ip_moptions *imo;
1782 INP_INFO_WLOCK(pcbinfo);
1783 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1785 imo = inp->inp_moptions;
1786 if ((inp->inp_vflag & INP_IPV4) &&
1789 * Unselect the outgoing interface if it is being
1792 if (imo->imo_multicast_ifp == ifp)
1793 imo->imo_multicast_ifp = NULL;
1796 * Drop multicast group membership if we joined
1797 * through the interface being detached.
1799 * XXX This can all be deferred to an epoch_call
1801 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1803 if (imo->imo_membership[i]->inm_ifp == ifp) {
1804 IN_MULTI_LOCK_ASSERT();
1805 in_leavegroup_locked(imo->imo_membership[i], NULL);
1807 } else if (gap != 0)
1808 imo->imo_membership[i - gap] =
1809 imo->imo_membership[i];
1811 imo->imo_num_memberships -= gap;
1815 INP_INFO_WUNLOCK(pcbinfo);
1819 * Lookup a PCB based on the local address and port. Caller must hold the
1820 * hash lock. No inpcb locks or references are acquired.
1822 #define INP_LOOKUP_MAPPED_PCB_COST 3
1824 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1825 u_short lport, int lookupflags, struct ucred *cred)
1829 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1835 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1836 ("%s: invalid lookup flags %d", __func__, lookupflags));
1838 INP_HASH_LOCK_ASSERT(pcbinfo);
1840 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1841 struct inpcbhead *head;
1843 * Look for an unconnected (wildcard foreign addr) PCB that
1844 * matches the local address and port we're looking for.
1846 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1847 0, pcbinfo->ipi_hashmask)];
1848 CK_LIST_FOREACH(inp, head, inp_hash) {
1850 /* XXX inp locking */
1851 if ((inp->inp_vflag & INP_IPV4) == 0)
1854 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1855 inp->inp_laddr.s_addr == laddr.s_addr &&
1856 inp->inp_lport == lport) {
1861 prison_equal_ip4(cred->cr_prison,
1862 inp->inp_cred->cr_prison))
1871 struct inpcbporthead *porthash;
1872 struct inpcbport *phd;
1873 struct inpcb *match = NULL;
1875 * Best fit PCB lookup.
1877 * First see if this local port is in use by looking on the
1880 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1881 pcbinfo->ipi_porthashmask)];
1882 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1883 if (phd->phd_port == lport)
1888 * Port is in use by one or more PCBs. Look for best
1891 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1894 !prison_equal_ip4(inp->inp_cred->cr_prison,
1898 /* XXX inp locking */
1899 if ((inp->inp_vflag & INP_IPV4) == 0)
1902 * We never select the PCB that has
1903 * INP_IPV6 flag and is bound to :: if
1904 * we have another PCB which is bound
1905 * to 0.0.0.0. If a PCB has the
1906 * INP_IPV6 flag, then we set its cost
1907 * higher than IPv4 only PCBs.
1909 * Note that the case only happens
1910 * when a socket is bound to ::, under
1911 * the condition that the use of the
1912 * mapped address is allowed.
1914 if ((inp->inp_vflag & INP_IPV6) != 0)
1915 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1917 if (inp->inp_faddr.s_addr != INADDR_ANY)
1919 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1920 if (laddr.s_addr == INADDR_ANY)
1922 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1925 if (laddr.s_addr != INADDR_ANY)
1928 if (wildcard < matchwild) {
1930 matchwild = wildcard;
1939 #undef INP_LOOKUP_MAPPED_PCB_COST
1941 static struct inpcb *
1942 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1943 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1944 uint16_t fport, int lookupflags)
1946 struct inpcb *local_wild;
1947 const struct inpcblbgrouphead *hdr;
1948 struct inpcblbgroup *grp;
1951 INP_HASH_LOCK_ASSERT(pcbinfo);
1953 hdr = &pcbinfo->ipi_lbgrouphashbase[INP_PCBLBGROUP_PORTHASH(lport,
1954 pcbinfo->ipi_lbgrouphashmask)];
1957 * Order of socket selection:
1959 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1962 * - Load balanced group does not contain jailed sockets
1963 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1966 CK_LIST_FOREACH(grp, hdr, il_list) {
1968 if (!(grp->il_vflag & INP_IPV4))
1971 if (grp->il_lport != lport)
1974 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1976 if (grp->il_laddr.s_addr == laddr->s_addr)
1977 return (grp->il_inp[idx]);
1978 if (grp->il_laddr.s_addr == INADDR_ANY &&
1979 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1980 local_wild = grp->il_inp[idx];
1982 return (local_wild);
1987 * Lookup PCB in hash list, using pcbgroup tables.
1989 static struct inpcb *
1990 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1991 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1992 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1994 struct inpcbhead *head;
1995 struct inpcb *inp, *tmpinp;
1996 u_short fport = fport_arg, lport = lport_arg;
2000 * First look for an exact match.
2003 INP_GROUP_LOCK(pcbgroup);
2004 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2005 pcbgroup->ipg_hashmask)];
2006 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2008 /* XXX inp locking */
2009 if ((inp->inp_vflag & INP_IPV4) == 0)
2012 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2013 inp->inp_laddr.s_addr == laddr.s_addr &&
2014 inp->inp_fport == fport &&
2015 inp->inp_lport == lport) {
2017 * XXX We should be able to directly return
2018 * the inp here, without any checks.
2019 * Well unless both bound with SO_REUSEPORT?
2021 if (prison_flag(inp->inp_cred, PR_IP4))
2027 if (tmpinp != NULL) {
2034 * For incoming connections, we may wish to do a wildcard
2035 * match for an RSS-local socket.
2037 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2038 struct inpcb *local_wild = NULL, *local_exact = NULL;
2040 struct inpcb *local_wild_mapped = NULL;
2042 struct inpcb *jail_wild = NULL;
2043 struct inpcbhead *head;
2047 * Order of socket selection - we always prefer jails.
2048 * 1. jailed, non-wild.
2050 * 3. non-jailed, non-wild.
2051 * 4. non-jailed, wild.
2054 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2055 lport, 0, pcbgroup->ipg_hashmask)];
2056 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2058 /* XXX inp locking */
2059 if ((inp->inp_vflag & INP_IPV4) == 0)
2062 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2063 inp->inp_lport != lport)
2066 injail = prison_flag(inp->inp_cred, PR_IP4);
2068 if (prison_check_ip4(inp->inp_cred,
2072 if (local_exact != NULL)
2076 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2081 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2083 /* XXX inp locking, NULL check */
2084 if (inp->inp_vflag & INP_IPV6PROTO)
2085 local_wild_mapped = inp;
2093 } /* LIST_FOREACH */
2102 inp = local_wild_mapped;
2110 * Then look for a wildcard match, if requested.
2112 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2113 struct inpcb *local_wild = NULL, *local_exact = NULL;
2115 struct inpcb *local_wild_mapped = NULL;
2117 struct inpcb *jail_wild = NULL;
2118 struct inpcbhead *head;
2122 * Order of socket selection - we always prefer jails.
2123 * 1. jailed, non-wild.
2125 * 3. non-jailed, non-wild.
2126 * 4. non-jailed, wild.
2128 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2129 0, pcbinfo->ipi_wildmask)];
2130 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2132 /* XXX inp locking */
2133 if ((inp->inp_vflag & INP_IPV4) == 0)
2136 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2137 inp->inp_lport != lport)
2140 injail = prison_flag(inp->inp_cred, PR_IP4);
2142 if (prison_check_ip4(inp->inp_cred,
2146 if (local_exact != NULL)
2150 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2155 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2157 /* XXX inp locking, NULL check */
2158 if (inp->inp_vflag & INP_IPV6PROTO)
2159 local_wild_mapped = inp;
2167 } /* LIST_FOREACH */
2175 inp = local_wild_mapped;
2179 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2180 INP_GROUP_UNLOCK(pcbgroup);
2184 if (lookupflags & INPLOOKUP_WLOCKPCB)
2185 locked = INP_TRY_WLOCK(inp);
2186 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2187 locked = INP_TRY_RLOCK(inp);
2189 panic("%s: locking bug", __func__);
2190 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2191 if (lookupflags & INPLOOKUP_WLOCKPCB)
2198 INP_GROUP_UNLOCK(pcbgroup);
2200 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2202 if (in_pcbrele_wlocked(inp))
2206 if (in_pcbrele_rlocked(inp))
2211 if (lookupflags & INPLOOKUP_WLOCKPCB)
2212 INP_WLOCK_ASSERT(inp);
2214 INP_RLOCK_ASSERT(inp);
2218 #endif /* PCBGROUP */
2221 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2222 * that the caller has locked the hash list, and will not perform any further
2223 * locking or reference operations on either the hash list or the connection.
2225 static struct inpcb *
2226 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2227 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2230 struct inpcbhead *head;
2231 struct inpcb *inp, *tmpinp;
2232 u_short fport = fport_arg, lport = lport_arg;
2235 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2236 ("%s: invalid lookup flags %d", __func__, lookupflags));
2237 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2238 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2241 * First look for an exact match.
2244 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2245 pcbinfo->ipi_hashmask)];
2246 CK_LIST_FOREACH(inp, head, inp_hash) {
2248 /* XXX inp locking */
2249 if ((inp->inp_vflag & INP_IPV4) == 0)
2252 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2253 inp->inp_laddr.s_addr == laddr.s_addr &&
2254 inp->inp_fport == fport &&
2255 inp->inp_lport == lport) {
2257 * XXX We should be able to directly return
2258 * the inp here, without any checks.
2259 * Well unless both bound with SO_REUSEPORT?
2261 if (prison_flag(inp->inp_cred, PR_IP4))
2271 * Then look in lb group (for wildcard match).
2273 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2274 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2275 fport, lookupflags);
2281 * Then look for a wildcard match, if requested.
2283 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2284 struct inpcb *local_wild = NULL, *local_exact = NULL;
2286 struct inpcb *local_wild_mapped = NULL;
2288 struct inpcb *jail_wild = NULL;
2292 * Order of socket selection - we always prefer jails.
2293 * 1. jailed, non-wild.
2295 * 3. non-jailed, non-wild.
2296 * 4. non-jailed, wild.
2299 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2300 0, pcbinfo->ipi_hashmask)];
2301 CK_LIST_FOREACH(inp, head, inp_hash) {
2303 /* XXX inp locking */
2304 if ((inp->inp_vflag & INP_IPV4) == 0)
2307 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2308 inp->inp_lport != lport)
2311 injail = prison_flag(inp->inp_cred, PR_IP4);
2313 if (prison_check_ip4(inp->inp_cred,
2317 if (local_exact != NULL)
2321 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2326 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2328 /* XXX inp locking, NULL check */
2329 if (inp->inp_vflag & INP_IPV6PROTO)
2330 local_wild_mapped = inp;
2338 } /* LIST_FOREACH */
2339 if (jail_wild != NULL)
2341 if (local_exact != NULL)
2342 return (local_exact);
2343 if (local_wild != NULL)
2344 return (local_wild);
2346 if (local_wild_mapped != NULL)
2347 return (local_wild_mapped);
2349 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2355 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2356 * hash list lock, and will return the inpcb locked (i.e., requires
2357 * INPLOOKUP_LOCKPCB).
2359 static struct inpcb *
2360 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2361 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2366 INP_HASH_RLOCK(pcbinfo);
2367 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2368 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2370 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2372 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2376 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2378 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2383 panic("%s: locking bug", __func__);
2386 if (lookupflags & INPLOOKUP_WLOCKPCB)
2387 INP_WLOCK_ASSERT(inp);
2389 INP_RLOCK_ASSERT(inp);
2393 INP_HASH_RUNLOCK(pcbinfo);
2398 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2399 * from which a pre-calculated hash value may be extracted.
2401 * Possibly more of this logic should be in in_pcbgroup.c.
2404 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2405 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2407 #if defined(PCBGROUP) && !defined(RSS)
2408 struct inpcbgroup *pcbgroup;
2411 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2412 ("%s: invalid lookup flags %d", __func__, lookupflags));
2413 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2414 ("%s: LOCKPCB not set", __func__));
2417 * When not using RSS, use connection groups in preference to the
2418 * reservation table when looking up 4-tuples. When using RSS, just
2419 * use the reservation table, due to the cost of the Toeplitz hash
2422 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2423 * we could be doing RSS with a non-Toeplitz hash that is affordable
2426 #if defined(PCBGROUP) && !defined(RSS)
2427 if (in_pcbgroup_enabled(pcbinfo)) {
2428 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2430 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2431 laddr, lport, lookupflags, ifp));
2434 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2439 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2440 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2441 struct ifnet *ifp, struct mbuf *m)
2444 struct inpcbgroup *pcbgroup;
2447 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2448 ("%s: invalid lookup flags %d", __func__, lookupflags));
2449 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2450 ("%s: LOCKPCB not set", __func__));
2454 * If we can use a hardware-generated hash to look up the connection
2455 * group, use that connection group to find the inpcb. Otherwise
2456 * fall back on a software hash -- or the reservation table if we're
2459 * XXXRW: As above, that policy belongs in the pcbgroup code.
2461 if (in_pcbgroup_enabled(pcbinfo) &&
2462 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2463 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2464 m->m_pkthdr.flowid);
2465 if (pcbgroup != NULL)
2466 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2467 fport, laddr, lport, lookupflags, ifp));
2469 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2471 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2472 laddr, lport, lookupflags, ifp));
2476 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2482 * Insert PCB onto various hash lists.
2485 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2487 struct inpcbhead *pcbhash;
2488 struct inpcbporthead *pcbporthash;
2489 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2490 struct inpcbport *phd;
2491 u_int32_t hashkey_faddr;
2494 INP_WLOCK_ASSERT(inp);
2495 INP_HASH_WLOCK_ASSERT(pcbinfo);
2497 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2498 ("in_pcbinshash: INP_INHASHLIST"));
2501 if (inp->inp_vflag & INP_IPV6)
2502 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2505 hashkey_faddr = inp->inp_faddr.s_addr;
2507 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2508 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2510 pcbporthash = &pcbinfo->ipi_porthashbase[
2511 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2514 * Add entry to load balance group.
2515 * Only do this if SO_REUSEPORT_LB is set.
2517 so_options = inp_so_options(inp);
2518 if (so_options & SO_REUSEPORT_LB) {
2519 int ret = in_pcbinslbgrouphash(inp);
2521 /* pcb lb group malloc fail (ret=ENOBUFS). */
2527 * Go through port list and look for a head for this lport.
2529 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2530 if (phd->phd_port == inp->inp_lport)
2534 * If none exists, malloc one and tack it on.
2537 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2539 return (ENOBUFS); /* XXX */
2541 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2542 phd->phd_port = inp->inp_lport;
2543 CK_LIST_INIT(&phd->phd_pcblist);
2544 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2547 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2548 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2549 inp->inp_flags |= INP_INHASHLIST;
2551 if (do_pcbgroup_update)
2552 in_pcbgroup_update(inp);
2558 * For now, there are two public interfaces to insert an inpcb into the hash
2559 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2560 * is used only in the TCP syncache, where in_pcbinshash is called before the
2561 * full 4-tuple is set for the inpcb, and we don't want to install in the
2562 * pcbgroup until later.
2564 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2565 * connection groups, and partially initialised inpcbs should not be exposed
2566 * to either reservation hash tables or pcbgroups.
2569 in_pcbinshash(struct inpcb *inp)
2572 return (in_pcbinshash_internal(inp, 1));
2576 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2579 return (in_pcbinshash_internal(inp, 0));
2583 * Move PCB to the proper hash bucket when { faddr, fport } have been
2584 * changed. NOTE: This does not handle the case of the lport changing (the
2585 * hashed port list would have to be updated as well), so the lport must
2586 * not change after in_pcbinshash() has been called.
2589 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2591 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2592 struct inpcbhead *head;
2593 u_int32_t hashkey_faddr;
2595 INP_WLOCK_ASSERT(inp);
2596 INP_HASH_WLOCK_ASSERT(pcbinfo);
2598 KASSERT(inp->inp_flags & INP_INHASHLIST,
2599 ("in_pcbrehash: !INP_INHASHLIST"));
2602 if (inp->inp_vflag & INP_IPV6)
2603 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2606 hashkey_faddr = inp->inp_faddr.s_addr;
2608 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2609 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2611 CK_LIST_REMOVE(inp, inp_hash);
2612 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2616 in_pcbgroup_update_mbuf(inp, m);
2618 in_pcbgroup_update(inp);
2623 in_pcbrehash(struct inpcb *inp)
2626 in_pcbrehash_mbuf(inp, NULL);
2630 * Remove PCB from various lists.
2633 in_pcbremlists(struct inpcb *inp)
2635 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2638 if (pcbinfo == &V_tcbinfo) {
2639 INP_INFO_RLOCK_ASSERT(pcbinfo);
2641 INP_INFO_WLOCK_ASSERT(pcbinfo);
2645 INP_WLOCK_ASSERT(inp);
2646 INP_LIST_WLOCK_ASSERT(pcbinfo);
2648 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2649 if (inp->inp_flags & INP_INHASHLIST) {
2650 struct inpcbport *phd = inp->inp_phd;
2652 INP_HASH_WLOCK(pcbinfo);
2654 /* XXX: Only do if SO_REUSEPORT_LB set? */
2655 in_pcbremlbgrouphash(inp);
2657 CK_LIST_REMOVE(inp, inp_hash);
2658 CK_LIST_REMOVE(inp, inp_portlist);
2659 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2660 CK_LIST_REMOVE(phd, phd_hash);
2661 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2663 INP_HASH_WUNLOCK(pcbinfo);
2664 inp->inp_flags &= ~INP_INHASHLIST;
2666 CK_LIST_REMOVE(inp, inp_list);
2667 pcbinfo->ipi_count--;
2669 in_pcbgroup_remove(inp);
2674 * Check for alternatives when higher level complains
2675 * about service problems. For now, invalidate cached
2676 * routing information. If the route was created dynamically
2677 * (by a redirect), time to try a default gateway again.
2680 in_losing(struct inpcb *inp)
2683 RO_INVALIDATE_CACHE(&inp->inp_route);
2688 * A set label operation has occurred at the socket layer, propagate the
2689 * label change into the in_pcb for the socket.
2692 in_pcbsosetlabel(struct socket *so)
2697 inp = sotoinpcb(so);
2698 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2702 mac_inpcb_sosetlabel(so, inp);
2709 * ipport_tick runs once per second, determining if random port allocation
2710 * should be continued. If more than ipport_randomcps ports have been
2711 * allocated in the last second, then we return to sequential port
2712 * allocation. We return to random allocation only once we drop below
2713 * ipport_randomcps for at least ipport_randomtime seconds.
2716 ipport_tick(void *xtp)
2718 VNET_ITERATOR_DECL(vnet_iter);
2720 VNET_LIST_RLOCK_NOSLEEP();
2721 VNET_FOREACH(vnet_iter) {
2722 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2723 if (V_ipport_tcpallocs <=
2724 V_ipport_tcplastcount + V_ipport_randomcps) {
2725 if (V_ipport_stoprandom > 0)
2726 V_ipport_stoprandom--;
2728 V_ipport_stoprandom = V_ipport_randomtime;
2729 V_ipport_tcplastcount = V_ipport_tcpallocs;
2732 VNET_LIST_RUNLOCK_NOSLEEP();
2733 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2740 callout_stop(&ipport_tick_callout);
2744 * The ipport_callout should start running at about the time we attach the
2745 * inet or inet6 domains.
2748 ipport_tick_init(const void *unused __unused)
2751 /* Start ipport_tick. */
2752 callout_init(&ipport_tick_callout, 1);
2753 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2754 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2755 SHUTDOWN_PRI_DEFAULT);
2757 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2758 ipport_tick_init, NULL);
2761 inp_wlock(struct inpcb *inp)
2768 inp_wunlock(struct inpcb *inp)
2775 inp_rlock(struct inpcb *inp)
2782 inp_runlock(struct inpcb *inp)
2788 #ifdef INVARIANT_SUPPORT
2790 inp_lock_assert(struct inpcb *inp)
2793 INP_WLOCK_ASSERT(inp);
2797 inp_unlock_assert(struct inpcb *inp)
2800 INP_UNLOCK_ASSERT(inp);
2805 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2809 INP_INFO_WLOCK(&V_tcbinfo);
2810 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2815 INP_INFO_WUNLOCK(&V_tcbinfo);
2819 inp_inpcbtosocket(struct inpcb *inp)
2822 INP_WLOCK_ASSERT(inp);
2823 return (inp->inp_socket);
2827 inp_inpcbtotcpcb(struct inpcb *inp)
2830 INP_WLOCK_ASSERT(inp);
2831 return ((struct tcpcb *)inp->inp_ppcb);
2835 inp_ip_tos_get(const struct inpcb *inp)
2838 return (inp->inp_ip_tos);
2842 inp_ip_tos_set(struct inpcb *inp, int val)
2845 inp->inp_ip_tos = val;
2849 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2850 uint32_t *faddr, uint16_t *fp)
2853 INP_LOCK_ASSERT(inp);
2854 *laddr = inp->inp_laddr.s_addr;
2855 *faddr = inp->inp_faddr.s_addr;
2856 *lp = inp->inp_lport;
2857 *fp = inp->inp_fport;
2861 so_sotoinpcb(struct socket *so)
2864 return (sotoinpcb(so));
2868 so_sototcpcb(struct socket *so)
2871 return (sototcpcb(so));
2875 * Create an external-format (``xinpcb'') structure using the information in
2876 * the kernel-format in_pcb structure pointed to by inp. This is done to
2877 * reduce the spew of irrelevant information over this interface, to isolate
2878 * user code from changes in the kernel structure, and potentially to provide
2879 * information-hiding if we decide that some of this information should be
2880 * hidden from users.
2883 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2886 xi->xi_len = sizeof(struct xinpcb);
2887 if (inp->inp_socket)
2888 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2890 bzero(&xi->xi_socket, sizeof(struct xsocket));
2891 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2892 xi->inp_gencnt = inp->inp_gencnt;
2893 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2894 xi->inp_flow = inp->inp_flow;
2895 xi->inp_flowid = inp->inp_flowid;
2896 xi->inp_flowtype = inp->inp_flowtype;
2897 xi->inp_flags = inp->inp_flags;
2898 xi->inp_flags2 = inp->inp_flags2;
2899 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2900 xi->in6p_cksum = inp->in6p_cksum;
2901 xi->in6p_hops = inp->in6p_hops;
2902 xi->inp_ip_tos = inp->inp_ip_tos;
2903 xi->inp_vflag = inp->inp_vflag;
2904 xi->inp_ip_ttl = inp->inp_ip_ttl;
2905 xi->inp_ip_p = inp->inp_ip_p;
2906 xi->inp_ip_minttl = inp->inp_ip_minttl;
2911 db_print_indent(int indent)
2915 for (i = 0; i < indent; i++)
2920 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2922 char faddr_str[48], laddr_str[48];
2924 db_print_indent(indent);
2925 db_printf("%s at %p\n", name, inc);
2930 if (inc->inc_flags & INC_ISIPV6) {
2932 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2933 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2938 inet_ntoa_r(inc->inc_laddr, laddr_str);
2939 inet_ntoa_r(inc->inc_faddr, faddr_str);
2941 db_print_indent(indent);
2942 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2943 ntohs(inc->inc_lport));
2944 db_print_indent(indent);
2945 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2946 ntohs(inc->inc_fport));
2950 db_print_inpflags(int inp_flags)
2955 if (inp_flags & INP_RECVOPTS) {
2956 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2959 if (inp_flags & INP_RECVRETOPTS) {
2960 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2963 if (inp_flags & INP_RECVDSTADDR) {
2964 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2967 if (inp_flags & INP_ORIGDSTADDR) {
2968 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2971 if (inp_flags & INP_HDRINCL) {
2972 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2975 if (inp_flags & INP_HIGHPORT) {
2976 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2979 if (inp_flags & INP_LOWPORT) {
2980 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2983 if (inp_flags & INP_ANONPORT) {
2984 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2987 if (inp_flags & INP_RECVIF) {
2988 db_printf("%sINP_RECVIF", comma ? ", " : "");
2991 if (inp_flags & INP_MTUDISC) {
2992 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2995 if (inp_flags & INP_RECVTTL) {
2996 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2999 if (inp_flags & INP_DONTFRAG) {
3000 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3003 if (inp_flags & INP_RECVTOS) {
3004 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3007 if (inp_flags & IN6P_IPV6_V6ONLY) {
3008 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3011 if (inp_flags & IN6P_PKTINFO) {
3012 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3015 if (inp_flags & IN6P_HOPLIMIT) {
3016 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3019 if (inp_flags & IN6P_HOPOPTS) {
3020 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3023 if (inp_flags & IN6P_DSTOPTS) {
3024 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3027 if (inp_flags & IN6P_RTHDR) {
3028 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3031 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3032 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3035 if (inp_flags & IN6P_TCLASS) {
3036 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3039 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3040 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3043 if (inp_flags & INP_TIMEWAIT) {
3044 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3047 if (inp_flags & INP_ONESBCAST) {
3048 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3051 if (inp_flags & INP_DROPPED) {
3052 db_printf("%sINP_DROPPED", comma ? ", " : "");
3055 if (inp_flags & INP_SOCKREF) {
3056 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3059 if (inp_flags & IN6P_RFC2292) {
3060 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3063 if (inp_flags & IN6P_MTU) {
3064 db_printf("IN6P_MTU%s", comma ? ", " : "");
3070 db_print_inpvflag(u_char inp_vflag)
3075 if (inp_vflag & INP_IPV4) {
3076 db_printf("%sINP_IPV4", comma ? ", " : "");
3079 if (inp_vflag & INP_IPV6) {
3080 db_printf("%sINP_IPV6", comma ? ", " : "");
3083 if (inp_vflag & INP_IPV6PROTO) {
3084 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3090 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3093 db_print_indent(indent);
3094 db_printf("%s at %p\n", name, inp);
3098 db_print_indent(indent);
3099 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3101 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3103 db_print_indent(indent);
3104 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3105 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3107 db_print_indent(indent);
3108 db_printf("inp_label: %p inp_flags: 0x%x (",
3109 inp->inp_label, inp->inp_flags);
3110 db_print_inpflags(inp->inp_flags);
3113 db_print_indent(indent);
3114 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3116 db_print_inpvflag(inp->inp_vflag);
3119 db_print_indent(indent);
3120 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3121 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3123 db_print_indent(indent);
3125 if (inp->inp_vflag & INP_IPV6) {
3126 db_printf("in6p_options: %p in6p_outputopts: %p "
3127 "in6p_moptions: %p\n", inp->in6p_options,
3128 inp->in6p_outputopts, inp->in6p_moptions);
3129 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3130 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3135 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3136 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3137 inp->inp_options, inp->inp_moptions);
3140 db_print_indent(indent);
3141 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3142 (uintmax_t)inp->inp_gencnt);
3145 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3150 db_printf("usage: show inpcb <addr>\n");
3153 inp = (struct inpcb *)addr;
3155 db_print_inpcb(inp, "inpcb", 0);
3161 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3165 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3167 union if_snd_tag_modify_params params = {
3168 .rate_limit.max_rate = max_pacing_rate,
3170 struct m_snd_tag *mst;
3174 mst = inp->inp_snd_tag;
3182 if (ifp->if_snd_tag_modify == NULL) {
3185 error = ifp->if_snd_tag_modify(mst, ¶ms);
3191 * Query existing TX rate limit based on the existing
3192 * "inp->inp_snd_tag", if any.
3195 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3197 union if_snd_tag_query_params params = { };
3198 struct m_snd_tag *mst;
3202 mst = inp->inp_snd_tag;
3210 if (ifp->if_snd_tag_query == NULL) {
3213 error = ifp->if_snd_tag_query(mst, ¶ms);
3214 if (error == 0 && p_max_pacing_rate != NULL)
3215 *p_max_pacing_rate = params.rate_limit.max_rate;
3221 * Query existing TX queue level based on the existing
3222 * "inp->inp_snd_tag", if any.
3225 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3227 union if_snd_tag_query_params params = { };
3228 struct m_snd_tag *mst;
3232 mst = inp->inp_snd_tag;
3240 if (ifp->if_snd_tag_query == NULL)
3241 return (EOPNOTSUPP);
3243 error = ifp->if_snd_tag_query(mst, ¶ms);
3244 if (error == 0 && p_txqueue_level != NULL)
3245 *p_txqueue_level = params.rate_limit.queue_level;
3250 * Allocate a new TX rate limit send tag from the network interface
3251 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3254 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3255 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3257 union if_snd_tag_alloc_params params = {
3258 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3259 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3260 .rate_limit.hdr.flowid = flowid,
3261 .rate_limit.hdr.flowtype = flowtype,
3262 .rate_limit.max_rate = max_pacing_rate,
3266 INP_WLOCK_ASSERT(inp);
3268 if (inp->inp_snd_tag != NULL)
3271 if (ifp->if_snd_tag_alloc == NULL) {
3274 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3277 * At success increment the refcount on
3278 * the send tag's network interface:
3281 if_ref(inp->inp_snd_tag->ifp);
3287 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3291 in_pcbdetach_txrtlmt(struct inpcb *inp)
3293 struct m_snd_tag *mst;
3296 INP_WLOCK_ASSERT(inp);
3298 mst = inp->inp_snd_tag;
3299 inp->inp_snd_tag = NULL;
3309 * If the device was detached while we still had reference(s)
3310 * on the ifp, we assume if_snd_tag_free() was replaced with
3313 ifp->if_snd_tag_free(mst);
3315 /* release reference count on network interface */
3320 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3321 * is set in the fast path and will attach/detach/modify the TX rate
3322 * limit send tag based on the socket's so_max_pacing_rate value.
3325 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3327 struct socket *socket;
3328 uint32_t max_pacing_rate;
3335 socket = inp->inp_socket;
3339 if (!INP_WLOCKED(inp)) {
3341 * NOTE: If the write locking fails, we need to bail
3342 * out and use the non-ratelimited ring for the
3343 * transmit until there is a new chance to get the
3346 if (!INP_TRY_UPGRADE(inp))
3354 * NOTE: The so_max_pacing_rate value is read unlocked,
3355 * because atomic updates are not required since the variable
3356 * is checked at every mbuf we send. It is assumed that the
3357 * variable read itself will be atomic.
3359 max_pacing_rate = socket->so_max_pacing_rate;
3362 * NOTE: When attaching to a network interface a reference is
3363 * made to ensure the network interface doesn't go away until
3364 * all ratelimit connections are gone. The network interface
3365 * pointers compared below represent valid network interfaces,
3366 * except when comparing towards NULL.
3368 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3370 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3371 if (inp->inp_snd_tag != NULL)
3372 in_pcbdetach_txrtlmt(inp);
3374 } else if (inp->inp_snd_tag == NULL) {
3376 * In order to utilize packet pacing with RSS, we need
3377 * to wait until there is a valid RSS hash before we
3380 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3383 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3384 mb->m_pkthdr.flowid, max_pacing_rate);
3387 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3389 if (error == 0 || error == EOPNOTSUPP)
3390 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3396 * Track route changes for TX rate limiting.
3399 in_pcboutput_eagain(struct inpcb *inp)
3401 struct socket *socket;
3407 socket = inp->inp_socket;
3411 if (inp->inp_snd_tag == NULL)
3414 if (!INP_WLOCKED(inp)) {
3416 * NOTE: If the write locking fails, we need to bail
3417 * out and use the non-ratelimited ring for the
3418 * transmit until there is a new chance to get the
3421 if (!INP_TRY_UPGRADE(inp))
3428 /* detach rate limiting */
3429 in_pcbdetach_txrtlmt(inp);
3431 /* make sure new mbuf send tag allocation is made */
3432 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3437 #endif /* RATELIMIT */