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_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
343 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
344 CK_LIST_FOREACH(grp, hdr, il_list) {
345 if (grp->il_vflag == inp->inp_vflag &&
346 grp->il_lport == inp->inp_lport &&
347 memcmp(&grp->il_dependladdr,
348 &inp->inp_inc.inc_ie.ie_dependladdr,
349 sizeof(grp->il_dependladdr)) == 0)
353 /* Create new load balance group. */
354 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
355 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
356 INPCBLBGROUP_SIZMIN);
359 } else if (grp->il_inpcnt == grp->il_inpsiz) {
360 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
361 if (ratecheck(&lastprint, &interval))
362 printf("lb group port %d, limit reached\n",
363 ntohs(grp->il_lport));
367 /* Expand this local group. */
368 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
373 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
374 ("invalid local group size %d and count %d", grp->il_inpsiz,
377 grp->il_inp[grp->il_inpcnt] = inp;
383 * Remove PCB from load balance group.
386 in_pcbremlbgrouphash(struct inpcb *inp)
388 struct inpcbinfo *pcbinfo;
389 struct inpcblbgrouphead *hdr;
390 struct inpcblbgroup *grp;
393 pcbinfo = inp->inp_pcbinfo;
395 INP_WLOCK_ASSERT(inp);
396 INP_HASH_WLOCK_ASSERT(pcbinfo);
398 hdr = &pcbinfo->ipi_lbgrouphashbase[
399 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
400 CK_LIST_FOREACH(grp, hdr, il_list) {
401 for (i = 0; i < grp->il_inpcnt; ++i) {
402 if (grp->il_inp[i] != inp)
405 if (grp->il_inpcnt == 1) {
406 /* We are the last, free this local group. */
407 in_pcblbgroup_free(grp);
409 /* Pull up inpcbs, shrink group if possible. */
410 in_pcblbgroup_reorder(hdr, &grp, i);
418 * Different protocols initialize their inpcbs differently - giving
419 * different name to the lock. But they all are disposed the same.
422 inpcb_fini(void *mem, int size)
424 struct inpcb *inp = mem;
426 INP_LOCK_DESTROY(inp);
430 * Initialize an inpcbinfo -- we should be able to reduce the number of
434 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
435 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
436 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
439 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
441 INP_INFO_LOCK_INIT(pcbinfo, name);
442 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
443 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
445 pcbinfo->ipi_vnet = curvnet;
447 pcbinfo->ipi_listhead = listhead;
448 CK_LIST_INIT(pcbinfo->ipi_listhead);
449 pcbinfo->ipi_count = 0;
450 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
451 &pcbinfo->ipi_hashmask);
452 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
453 &pcbinfo->ipi_porthashmask);
454 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
455 &pcbinfo->ipi_lbgrouphashmask);
457 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
459 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
460 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
461 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
462 uma_zone_set_warning(pcbinfo->ipi_zone,
463 "kern.ipc.maxsockets limit reached");
467 * Destroy an inpcbinfo.
470 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
473 KASSERT(pcbinfo->ipi_count == 0,
474 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
476 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
477 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
478 pcbinfo->ipi_porthashmask);
479 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
480 pcbinfo->ipi_lbgrouphashmask);
482 in_pcbgroup_destroy(pcbinfo);
484 uma_zdestroy(pcbinfo->ipi_zone);
485 INP_LIST_LOCK_DESTROY(pcbinfo);
486 INP_HASH_LOCK_DESTROY(pcbinfo);
487 INP_INFO_LOCK_DESTROY(pcbinfo);
491 * Allocate a PCB and associate it with the socket.
492 * On success return with the PCB locked.
495 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
501 if (pcbinfo == &V_tcbinfo) {
502 INP_INFO_RLOCK_ASSERT(pcbinfo);
504 INP_INFO_WLOCK_ASSERT(pcbinfo);
509 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
512 bzero(&inp->inp_start_zero, inp_zero_size);
513 inp->inp_pcbinfo = pcbinfo;
514 inp->inp_socket = so;
515 inp->inp_cred = crhold(so->so_cred);
516 inp->inp_inc.inc_fibnum = so->so_fibnum;
518 error = mac_inpcb_init(inp, M_NOWAIT);
521 mac_inpcb_create(so, inp);
523 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
524 error = ipsec_init_pcbpolicy(inp);
527 mac_inpcb_destroy(inp);
533 if (INP_SOCKAF(so) == AF_INET6) {
534 inp->inp_vflag |= INP_IPV6PROTO;
536 inp->inp_flags |= IN6P_IPV6_V6ONLY;
540 INP_LIST_WLOCK(pcbinfo);
541 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
542 pcbinfo->ipi_count++;
543 so->so_pcb = (caddr_t)inp;
545 if (V_ip6_auto_flowlabel)
546 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
548 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
549 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
552 * Routes in inpcb's can cache L2 as well; they are guaranteed
555 inp->inp_route.ro_flags = RT_LLE_CACHE;
556 INP_LIST_WUNLOCK(pcbinfo);
557 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
560 crfree(inp->inp_cred);
561 uma_zfree(pcbinfo->ipi_zone, inp);
569 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
573 INP_WLOCK_ASSERT(inp);
574 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
576 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
578 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
579 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
580 &inp->inp_lport, cred);
583 if (in_pcbinshash(inp) != 0) {
584 inp->inp_laddr.s_addr = INADDR_ANY;
589 inp->inp_flags |= INP_ANONPORT;
595 * Select a local port (number) to use.
597 #if defined(INET) || defined(INET6)
599 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
600 struct ucred *cred, int lookupflags)
602 struct inpcbinfo *pcbinfo;
603 struct inpcb *tmpinp;
604 unsigned short *lastport;
605 int count, dorandom, error;
606 u_short aux, first, last, lport;
608 struct in_addr laddr;
611 pcbinfo = inp->inp_pcbinfo;
614 * Because no actual state changes occur here, a global write lock on
615 * the pcbinfo isn't required.
617 INP_LOCK_ASSERT(inp);
618 INP_HASH_LOCK_ASSERT(pcbinfo);
620 if (inp->inp_flags & INP_HIGHPORT) {
621 first = V_ipport_hifirstauto; /* sysctl */
622 last = V_ipport_hilastauto;
623 lastport = &pcbinfo->ipi_lasthi;
624 } else if (inp->inp_flags & INP_LOWPORT) {
625 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
628 first = V_ipport_lowfirstauto; /* 1023 */
629 last = V_ipport_lowlastauto; /* 600 */
630 lastport = &pcbinfo->ipi_lastlow;
632 first = V_ipport_firstauto; /* sysctl */
633 last = V_ipport_lastauto;
634 lastport = &pcbinfo->ipi_lastport;
637 * For UDP(-Lite), use random port allocation as long as the user
638 * allows it. For TCP (and as of yet unknown) connections,
639 * use random port allocation only if the user allows it AND
640 * ipport_tick() allows it.
642 if (V_ipport_randomized &&
643 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
644 pcbinfo == &V_ulitecbinfo))
649 * It makes no sense to do random port allocation if
650 * we have the only port available.
654 /* Make sure to not include UDP(-Lite) packets in the count. */
655 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
656 V_ipport_tcpallocs++;
658 * Instead of having two loops further down counting up or down
659 * make sure that first is always <= last and go with only one
660 * code path implementing all logic.
669 /* Make the compiler happy. */
671 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
672 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
677 tmpinp = NULL; /* Make compiler happy. */
681 *lastport = first + (arc4random() % (last - first));
683 count = last - first;
686 if (count-- < 0) /* completely used? */
687 return (EADDRNOTAVAIL);
689 if (*lastport < first || *lastport > last)
691 lport = htons(*lastport);
694 if ((inp->inp_vflag & INP_IPV6) != 0)
695 tmpinp = in6_pcblookup_local(pcbinfo,
696 &inp->in6p_laddr, lport, lookupflags, cred);
698 #if defined(INET) && defined(INET6)
702 tmpinp = in_pcblookup_local(pcbinfo, laddr,
703 lport, lookupflags, cred);
705 } while (tmpinp != NULL);
708 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
709 laddrp->s_addr = laddr.s_addr;
717 * Return cached socket options.
720 inp_so_options(const struct inpcb *inp)
726 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
727 so_options |= SO_REUSEPORT_LB;
728 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
729 so_options |= SO_REUSEPORT;
730 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
731 so_options |= SO_REUSEADDR;
734 #endif /* INET || INET6 */
737 * Check if a new BINDMULTI socket is allowed to be created.
739 * ni points to the new inp.
740 * oi points to the exisitng inp.
742 * This checks whether the existing inp also has BINDMULTI and
743 * whether the credentials match.
746 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
748 /* Check permissions match */
749 if ((ni->inp_flags2 & INP_BINDMULTI) &&
750 (ni->inp_cred->cr_uid !=
751 oi->inp_cred->cr_uid))
754 /* Check the existing inp has BINDMULTI set */
755 if ((ni->inp_flags2 & INP_BINDMULTI) &&
756 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
760 * We're okay - either INP_BINDMULTI isn't set on ni, or
761 * it is and it matches the checks.
768 * Set up a bind operation on a PCB, performing port allocation
769 * as required, but do not actually modify the PCB. Callers can
770 * either complete the bind by setting inp_laddr/inp_lport and
771 * calling in_pcbinshash(), or they can just use the resulting
772 * port and address to authorise the sending of a once-off packet.
774 * On error, the values of *laddrp and *lportp are not changed.
777 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
778 u_short *lportp, struct ucred *cred)
780 struct socket *so = inp->inp_socket;
781 struct sockaddr_in *sin;
782 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
783 struct in_addr laddr;
785 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
789 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
790 * so that we don't have to add to the (already messy) code below.
792 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
795 * No state changes, so read locks are sufficient here.
797 INP_LOCK_ASSERT(inp);
798 INP_HASH_LOCK_ASSERT(pcbinfo);
800 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
801 return (EADDRNOTAVAIL);
802 laddr.s_addr = *laddrp;
803 if (nam != NULL && laddr.s_addr != INADDR_ANY)
805 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
806 lookupflags = INPLOOKUP_WILDCARD;
808 if ((error = prison_local_ip4(cred, &laddr)) != 0)
811 sin = (struct sockaddr_in *)nam;
812 if (nam->sa_len != sizeof (*sin))
816 * We should check the family, but old programs
817 * incorrectly fail to initialize it.
819 if (sin->sin_family != AF_INET)
820 return (EAFNOSUPPORT);
822 error = prison_local_ip4(cred, &sin->sin_addr);
825 if (sin->sin_port != *lportp) {
826 /* Don't allow the port to change. */
829 lport = sin->sin_port;
831 /* NB: lport is left as 0 if the port isn't being changed. */
832 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
834 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
835 * allow complete duplication of binding if
836 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
837 * and a multicast address is bound on both
838 * new and duplicated sockets.
840 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
841 reuseport = SO_REUSEADDR|SO_REUSEPORT;
843 * XXX: How to deal with SO_REUSEPORT_LB here?
844 * Treat same as SO_REUSEPORT for now.
846 if ((so->so_options &
847 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
848 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
849 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
850 sin->sin_port = 0; /* yech... */
851 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
853 * Is the address a local IP address?
854 * If INP_BINDANY is set, then the socket may be bound
855 * to any endpoint address, local or not.
857 if ((inp->inp_flags & INP_BINDANY) == 0 &&
858 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
859 return (EADDRNOTAVAIL);
861 laddr = sin->sin_addr;
867 if (ntohs(lport) <= V_ipport_reservedhigh &&
868 ntohs(lport) >= V_ipport_reservedlow &&
869 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
871 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
872 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
873 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
874 lport, INPLOOKUP_WILDCARD, cred);
877 * This entire block sorely needs a rewrite.
880 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
881 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
882 (so->so_type != SOCK_STREAM ||
883 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
884 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
885 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
886 (t->inp_flags2 & INP_REUSEPORT) ||
887 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
888 (inp->inp_cred->cr_uid !=
889 t->inp_cred->cr_uid))
893 * If the socket is a BINDMULTI socket, then
894 * the credentials need to match and the
895 * original socket also has to have been bound
898 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
901 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
902 lport, lookupflags, cred);
903 if (t && (t->inp_flags & INP_TIMEWAIT)) {
905 * XXXRW: If an incpb has had its timewait
906 * state recycled, we treat the address as
907 * being in use (for now). This is better
908 * than a panic, but not desirable.
912 ((reuseport & tw->tw_so_options) == 0 &&
914 tw->tw_so_options) == 0)) {
918 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
919 (reuseport & inp_so_options(t)) == 0 &&
920 (reuseport_lb & inp_so_options(t)) == 0) {
922 if (ntohl(sin->sin_addr.s_addr) !=
924 ntohl(t->inp_laddr.s_addr) !=
926 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
927 (t->inp_vflag & INP_IPV6PROTO) == 0)
930 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
938 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
943 *laddrp = laddr.s_addr;
949 * Connect from a socket to a specified address.
950 * Both address and port must be specified in argument sin.
951 * If don't have a local address for this socket yet,
955 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
956 struct ucred *cred, struct mbuf *m)
958 u_short lport, fport;
959 in_addr_t laddr, faddr;
962 INP_WLOCK_ASSERT(inp);
963 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
965 lport = inp->inp_lport;
966 laddr = inp->inp_laddr.s_addr;
967 anonport = (lport == 0);
968 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
973 /* Do the initial binding of the local address if required. */
974 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
975 inp->inp_lport = lport;
976 inp->inp_laddr.s_addr = laddr;
977 if (in_pcbinshash(inp) != 0) {
978 inp->inp_laddr.s_addr = INADDR_ANY;
984 /* Commit the remaining changes. */
985 inp->inp_lport = lport;
986 inp->inp_laddr.s_addr = laddr;
987 inp->inp_faddr.s_addr = faddr;
988 inp->inp_fport = fport;
989 in_pcbrehash_mbuf(inp, m);
992 inp->inp_flags |= INP_ANONPORT;
997 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1000 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1004 * Do proper source address selection on an unbound socket in case
1005 * of connect. Take jails into account as well.
1008 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1012 struct sockaddr *sa;
1013 struct sockaddr_in *sin;
1015 struct epoch_tracker et;
1018 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1020 * Bypass source address selection and use the primary jail IP
1023 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1027 bzero(&sro, sizeof(sro));
1029 sin = (struct sockaddr_in *)&sro.ro_dst;
1030 sin->sin_family = AF_INET;
1031 sin->sin_len = sizeof(struct sockaddr_in);
1032 sin->sin_addr.s_addr = faddr->s_addr;
1035 * If route is known our src addr is taken from the i/f,
1038 * Find out route to destination.
1040 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1041 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1044 * If we found a route, use the address corresponding to
1045 * the outgoing interface.
1047 * Otherwise assume faddr is reachable on a directly connected
1048 * network and try to find a corresponding interface to take
1049 * the source address from.
1051 NET_EPOCH_ENTER(et);
1052 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1053 struct in_ifaddr *ia;
1056 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1057 inp->inp_socket->so_fibnum));
1059 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1060 inp->inp_socket->so_fibnum));
1064 error = ENETUNREACH;
1068 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1069 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1075 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1078 if (sa->sa_family != AF_INET)
1080 sin = (struct sockaddr_in *)sa;
1081 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1082 ia = (struct in_ifaddr *)ifa;
1087 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1091 /* 3. As a last resort return the 'default' jail address. */
1092 error = prison_get_ip4(cred, laddr);
1097 * If the outgoing interface on the route found is not
1098 * a loopback interface, use the address from that interface.
1099 * In case of jails do those three steps:
1100 * 1. check if the interface address belongs to the jail. If so use it.
1101 * 2. check if we have any address on the outgoing interface
1102 * belonging to this jail. If so use it.
1103 * 3. as a last resort return the 'default' jail address.
1105 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1106 struct in_ifaddr *ia;
1109 /* If not jailed, use the default returned. */
1110 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1111 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1112 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1117 /* 1. Check if the iface address belongs to the jail. */
1118 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1119 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1120 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1121 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1126 * 2. Check if we have any address on the outgoing interface
1127 * belonging to this jail.
1130 ifp = sro.ro_rt->rt_ifp;
1131 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1133 if (sa->sa_family != AF_INET)
1135 sin = (struct sockaddr_in *)sa;
1136 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1137 ia = (struct in_ifaddr *)ifa;
1142 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1146 /* 3. As a last resort return the 'default' jail address. */
1147 error = prison_get_ip4(cred, laddr);
1152 * The outgoing interface is marked with 'loopback net', so a route
1153 * to ourselves is here.
1154 * Try to find the interface of the destination address and then
1155 * take the address from there. That interface is not necessarily
1156 * a loopback interface.
1157 * In case of jails, check that it is an address of the jail
1158 * and if we cannot find, fall back to the 'default' jail address.
1160 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1161 struct sockaddr_in sain;
1162 struct in_ifaddr *ia;
1164 bzero(&sain, sizeof(struct sockaddr_in));
1165 sain.sin_family = AF_INET;
1166 sain.sin_len = sizeof(struct sockaddr_in);
1167 sain.sin_addr.s_addr = faddr->s_addr;
1169 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1170 inp->inp_socket->so_fibnum));
1172 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1173 inp->inp_socket->so_fibnum));
1175 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1177 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1179 error = ENETUNREACH;
1182 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1192 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1194 if (sa->sa_family != AF_INET)
1196 sin = (struct sockaddr_in *)sa;
1197 if (prison_check_ip4(cred,
1198 &sin->sin_addr) == 0) {
1199 ia = (struct in_ifaddr *)ifa;
1204 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1209 /* 3. As a last resort return the 'default' jail address. */
1210 error = prison_get_ip4(cred, laddr);
1216 if (sro.ro_rt != NULL)
1222 * Set up for a connect from a socket to the specified address.
1223 * On entry, *laddrp and *lportp should contain the current local
1224 * address and port for the PCB; these are updated to the values
1225 * that should be placed in inp_laddr and inp_lport to complete
1228 * On success, *faddrp and *fportp will be set to the remote address
1229 * and port. These are not updated in the error case.
1231 * If the operation fails because the connection already exists,
1232 * *oinpp will be set to the PCB of that connection so that the
1233 * caller can decide to override it. In all other cases, *oinpp
1237 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1238 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1239 struct inpcb **oinpp, struct ucred *cred)
1241 struct rm_priotracker in_ifa_tracker;
1242 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1243 struct in_ifaddr *ia;
1245 struct in_addr laddr, faddr;
1246 u_short lport, fport;
1250 * Because a global state change doesn't actually occur here, a read
1251 * lock is sufficient.
1253 INP_LOCK_ASSERT(inp);
1254 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1258 if (nam->sa_len != sizeof (*sin))
1260 if (sin->sin_family != AF_INET)
1261 return (EAFNOSUPPORT);
1262 if (sin->sin_port == 0)
1263 return (EADDRNOTAVAIL);
1264 laddr.s_addr = *laddrp;
1266 faddr = sin->sin_addr;
1267 fport = sin->sin_port;
1269 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1271 * If the destination address is INADDR_ANY,
1272 * use the primary local address.
1273 * If the supplied address is INADDR_BROADCAST,
1274 * and the primary interface supports broadcast,
1275 * choose the broadcast address for that interface.
1277 if (faddr.s_addr == INADDR_ANY) {
1278 IN_IFADDR_RLOCK(&in_ifa_tracker);
1280 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1281 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1283 (error = prison_get_ip4(cred, &faddr)) != 0)
1285 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1286 IN_IFADDR_RLOCK(&in_ifa_tracker);
1287 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1289 faddr = satosin(&CK_STAILQ_FIRST(
1290 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1291 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1294 if (laddr.s_addr == INADDR_ANY) {
1295 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1297 * If the destination address is multicast and an outgoing
1298 * interface has been set as a multicast option, prefer the
1299 * address of that interface as our source address.
1301 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1302 inp->inp_moptions != NULL) {
1303 struct ip_moptions *imo;
1306 imo = inp->inp_moptions;
1307 if (imo->imo_multicast_ifp != NULL) {
1308 ifp = imo->imo_multicast_ifp;
1309 IN_IFADDR_RLOCK(&in_ifa_tracker);
1310 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1311 if ((ia->ia_ifp == ifp) &&
1313 prison_check_ip4(cred,
1314 &ia->ia_addr.sin_addr) == 0))
1318 error = EADDRNOTAVAIL;
1320 laddr = ia->ia_addr.sin_addr;
1323 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1329 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1330 laddr, lport, 0, NULL);
1334 return (EADDRINUSE);
1337 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1342 *laddrp = laddr.s_addr;
1344 *faddrp = faddr.s_addr;
1350 in_pcbdisconnect(struct inpcb *inp)
1353 INP_WLOCK_ASSERT(inp);
1354 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1356 inp->inp_faddr.s_addr = INADDR_ANY;
1363 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1364 * For most protocols, this will be invoked immediately prior to calling
1365 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1366 * socket, in which case in_pcbfree() is deferred.
1369 in_pcbdetach(struct inpcb *inp)
1372 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1375 if (inp->inp_snd_tag != NULL)
1376 in_pcbdetach_txrtlmt(inp);
1378 inp->inp_socket->so_pcb = NULL;
1379 inp->inp_socket = NULL;
1383 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1384 * stability of an inpcb pointer despite the inpcb lock being released. This
1385 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1386 * but where the inpcb lock may already held, or when acquiring a reference
1389 * in_pcbref() should be used only to provide brief memory stability, and
1390 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1391 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1392 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1393 * lock and rele are the *only* safe operations that may be performed on the
1396 * While the inpcb will not be freed, releasing the inpcb lock means that the
1397 * connection's state may change, so the caller should be careful to
1398 * revalidate any cached state on reacquiring the lock. Drop the reference
1399 * using in_pcbrele().
1402 in_pcbref(struct inpcb *inp)
1405 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1407 refcount_acquire(&inp->inp_refcount);
1411 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1412 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1413 * return a flag indicating whether or not the inpcb remains valid. If it is
1414 * valid, we return with the inpcb lock held.
1416 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1417 * reference on an inpcb. Historically more work was done here (actually, in
1418 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1419 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1420 * about memory stability (and continued use of the write lock).
1423 in_pcbrele_rlocked(struct inpcb *inp)
1425 struct inpcbinfo *pcbinfo;
1427 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1429 INP_RLOCK_ASSERT(inp);
1431 if (refcount_release(&inp->inp_refcount) == 0) {
1433 * If the inpcb has been freed, let the caller know, even if
1434 * this isn't the last reference.
1436 if (inp->inp_flags2 & INP_FREED) {
1443 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1445 if (inp->inp_in_hpts || inp->inp_in_input) {
1446 struct tcp_hpts_entry *hpts;
1448 * We should not be on the hpts at
1449 * this point in any form. we must
1450 * get the lock to be sure.
1452 hpts = tcp_hpts_lock(inp);
1453 if (inp->inp_in_hpts)
1454 panic("Hpts:%p inp:%p at free still on hpts",
1456 mtx_unlock(&hpts->p_mtx);
1457 hpts = tcp_input_lock(inp);
1458 if (inp->inp_in_input)
1459 panic("Hpts:%p inp:%p at free still on input hpts",
1461 mtx_unlock(&hpts->p_mtx);
1465 pcbinfo = inp->inp_pcbinfo;
1466 uma_zfree(pcbinfo->ipi_zone, inp);
1471 in_pcbrele_wlocked(struct inpcb *inp)
1473 struct inpcbinfo *pcbinfo;
1475 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1477 INP_WLOCK_ASSERT(inp);
1479 if (refcount_release(&inp->inp_refcount) == 0) {
1481 * If the inpcb has been freed, let the caller know, even if
1482 * this isn't the last reference.
1484 if (inp->inp_flags2 & INP_FREED) {
1491 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1493 if (inp->inp_in_hpts || inp->inp_in_input) {
1494 struct tcp_hpts_entry *hpts;
1496 * We should not be on the hpts at
1497 * this point in any form. we must
1498 * get the lock to be sure.
1500 hpts = tcp_hpts_lock(inp);
1501 if (inp->inp_in_hpts)
1502 panic("Hpts:%p inp:%p at free still on hpts",
1504 mtx_unlock(&hpts->p_mtx);
1505 hpts = tcp_input_lock(inp);
1506 if (inp->inp_in_input)
1507 panic("Hpts:%p inp:%p at free still on input hpts",
1509 mtx_unlock(&hpts->p_mtx);
1513 pcbinfo = inp->inp_pcbinfo;
1514 uma_zfree(pcbinfo->ipi_zone, inp);
1519 * Temporary wrapper.
1522 in_pcbrele(struct inpcb *inp)
1525 return (in_pcbrele_wlocked(inp));
1529 in_pcblist_rele_rlocked(epoch_context_t ctx)
1531 struct in_pcblist *il;
1533 struct inpcbinfo *pcbinfo;
1536 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1537 pcbinfo = il->il_pcbinfo;
1539 INP_INFO_WLOCK(pcbinfo);
1540 for (i = 0; i < n; i++) {
1541 inp = il->il_inp_list[i];
1543 if (!in_pcbrele_rlocked(inp))
1546 INP_INFO_WUNLOCK(pcbinfo);
1551 inpcbport_free(epoch_context_t ctx)
1553 struct inpcbport *phd;
1555 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1560 in_pcbfree_deferred(epoch_context_t ctx)
1563 int released __unused;
1565 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1569 struct ip_moptions *imo = inp->inp_moptions;
1570 inp->inp_moptions = NULL;
1572 /* XXXRW: Do as much as possible here. */
1573 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1574 if (inp->inp_sp != NULL)
1575 ipsec_delete_pcbpolicy(inp);
1578 struct ip6_moptions *im6o = NULL;
1579 if (inp->inp_vflag & INP_IPV6PROTO) {
1580 ip6_freepcbopts(inp->in6p_outputopts);
1581 im6o = inp->in6p_moptions;
1582 inp->in6p_moptions = NULL;
1585 if (inp->inp_options)
1586 (void)m_free(inp->inp_options);
1588 crfree(inp->inp_cred);
1590 mac_inpcb_destroy(inp);
1592 released = in_pcbrele_wlocked(inp);
1595 ip6_freemoptions(im6o);
1598 inp_freemoptions(imo);
1603 * Unconditionally schedule an inpcb to be freed by decrementing its
1604 * reference count, which should occur only after the inpcb has been detached
1605 * from its socket. If another thread holds a temporary reference (acquired
1606 * using in_pcbref()) then the free is deferred until that reference is
1607 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1608 * work, including removal from global lists, is done in this context, where
1609 * the pcbinfo lock is held.
1612 in_pcbfree(struct inpcb *inp)
1614 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1616 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1617 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1618 ("%s: called twice for pcb %p", __func__, inp));
1619 if (inp->inp_flags2 & INP_FREED) {
1625 if (pcbinfo == &V_tcbinfo) {
1626 INP_INFO_LOCK_ASSERT(pcbinfo);
1628 INP_INFO_WLOCK_ASSERT(pcbinfo);
1631 INP_WLOCK_ASSERT(inp);
1632 INP_LIST_WLOCK(pcbinfo);
1633 in_pcbremlists(inp);
1634 INP_LIST_WUNLOCK(pcbinfo);
1635 RO_INVALIDATE_CACHE(&inp->inp_route);
1636 /* mark as destruction in progress */
1637 inp->inp_flags2 |= INP_FREED;
1639 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1643 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1644 * port reservation, and preventing it from being returned by inpcb lookups.
1646 * It is used by TCP to mark an inpcb as unused and avoid future packet
1647 * delivery or event notification when a socket remains open but TCP has
1648 * closed. This might occur as a result of a shutdown()-initiated TCP close
1649 * or a RST on the wire, and allows the port binding to be reused while still
1650 * maintaining the invariant that so_pcb always points to a valid inpcb until
1653 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1654 * in_pcbnotifyall() and in_pcbpurgeif0()?
1657 in_pcbdrop(struct inpcb *inp)
1660 INP_WLOCK_ASSERT(inp);
1662 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1663 MPASS(inp->inp_refcount > 1);
1667 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1670 inp->inp_flags |= INP_DROPPED;
1671 if (inp->inp_flags & INP_INHASHLIST) {
1672 struct inpcbport *phd = inp->inp_phd;
1674 INP_HASH_WLOCK(inp->inp_pcbinfo);
1675 in_pcbremlbgrouphash(inp);
1676 CK_LIST_REMOVE(inp, inp_hash);
1677 CK_LIST_REMOVE(inp, inp_portlist);
1678 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1679 CK_LIST_REMOVE(phd, phd_hash);
1680 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1682 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1683 inp->inp_flags &= ~INP_INHASHLIST;
1685 in_pcbgroup_remove(inp);
1692 * Common routines to return the socket addresses associated with inpcbs.
1695 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1697 struct sockaddr_in *sin;
1699 sin = malloc(sizeof *sin, M_SONAME,
1701 sin->sin_family = AF_INET;
1702 sin->sin_len = sizeof(*sin);
1703 sin->sin_addr = *addr_p;
1704 sin->sin_port = port;
1706 return (struct sockaddr *)sin;
1710 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1713 struct in_addr addr;
1716 inp = sotoinpcb(so);
1717 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1720 port = inp->inp_lport;
1721 addr = inp->inp_laddr;
1724 *nam = in_sockaddr(port, &addr);
1729 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1732 struct in_addr addr;
1735 inp = sotoinpcb(so);
1736 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1739 port = inp->inp_fport;
1740 addr = inp->inp_faddr;
1743 *nam = in_sockaddr(port, &addr);
1748 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1749 struct inpcb *(*notify)(struct inpcb *, int))
1751 struct inpcb *inp, *inp_temp;
1753 INP_INFO_WLOCK(pcbinfo);
1754 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1757 if ((inp->inp_vflag & INP_IPV4) == 0) {
1762 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1763 inp->inp_socket == NULL) {
1767 if ((*notify)(inp, errno))
1770 INP_INFO_WUNLOCK(pcbinfo);
1774 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1777 struct ip_moptions *imo;
1780 INP_INFO_WLOCK(pcbinfo);
1781 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1783 imo = inp->inp_moptions;
1784 if ((inp->inp_vflag & INP_IPV4) &&
1787 * Unselect the outgoing interface if it is being
1790 if (imo->imo_multicast_ifp == ifp)
1791 imo->imo_multicast_ifp = NULL;
1794 * Drop multicast group membership if we joined
1795 * through the interface being detached.
1797 * XXX This can all be deferred to an epoch_call
1799 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1801 if (imo->imo_membership[i]->inm_ifp == ifp) {
1802 IN_MULTI_LOCK_ASSERT();
1803 in_leavegroup_locked(imo->imo_membership[i], NULL);
1805 } else if (gap != 0)
1806 imo->imo_membership[i - gap] =
1807 imo->imo_membership[i];
1809 imo->imo_num_memberships -= gap;
1813 INP_INFO_WUNLOCK(pcbinfo);
1817 * Lookup a PCB based on the local address and port. Caller must hold the
1818 * hash lock. No inpcb locks or references are acquired.
1820 #define INP_LOOKUP_MAPPED_PCB_COST 3
1822 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1823 u_short lport, int lookupflags, struct ucred *cred)
1827 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1833 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1834 ("%s: invalid lookup flags %d", __func__, lookupflags));
1836 INP_HASH_LOCK_ASSERT(pcbinfo);
1838 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1839 struct inpcbhead *head;
1841 * Look for an unconnected (wildcard foreign addr) PCB that
1842 * matches the local address and port we're looking for.
1844 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1845 0, pcbinfo->ipi_hashmask)];
1846 CK_LIST_FOREACH(inp, head, inp_hash) {
1848 /* XXX inp locking */
1849 if ((inp->inp_vflag & INP_IPV4) == 0)
1852 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1853 inp->inp_laddr.s_addr == laddr.s_addr &&
1854 inp->inp_lport == lport) {
1859 prison_equal_ip4(cred->cr_prison,
1860 inp->inp_cred->cr_prison))
1869 struct inpcbporthead *porthash;
1870 struct inpcbport *phd;
1871 struct inpcb *match = NULL;
1873 * Best fit PCB lookup.
1875 * First see if this local port is in use by looking on the
1878 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1879 pcbinfo->ipi_porthashmask)];
1880 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1881 if (phd->phd_port == lport)
1886 * Port is in use by one or more PCBs. Look for best
1889 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1892 !prison_equal_ip4(inp->inp_cred->cr_prison,
1896 /* XXX inp locking */
1897 if ((inp->inp_vflag & INP_IPV4) == 0)
1900 * We never select the PCB that has
1901 * INP_IPV6 flag and is bound to :: if
1902 * we have another PCB which is bound
1903 * to 0.0.0.0. If a PCB has the
1904 * INP_IPV6 flag, then we set its cost
1905 * higher than IPv4 only PCBs.
1907 * Note that the case only happens
1908 * when a socket is bound to ::, under
1909 * the condition that the use of the
1910 * mapped address is allowed.
1912 if ((inp->inp_vflag & INP_IPV6) != 0)
1913 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1915 if (inp->inp_faddr.s_addr != INADDR_ANY)
1917 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1918 if (laddr.s_addr == INADDR_ANY)
1920 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1923 if (laddr.s_addr != INADDR_ANY)
1926 if (wildcard < matchwild) {
1928 matchwild = wildcard;
1937 #undef INP_LOOKUP_MAPPED_PCB_COST
1939 static struct inpcb *
1940 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1941 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1942 uint16_t fport, int lookupflags)
1944 struct inpcb *local_wild;
1945 const struct inpcblbgrouphead *hdr;
1946 struct inpcblbgroup *grp;
1949 INP_HASH_LOCK_ASSERT(pcbinfo);
1951 hdr = &pcbinfo->ipi_lbgrouphashbase[
1952 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1955 * Order of socket selection:
1957 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1960 * - Load balanced group does not contain jailed sockets
1961 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1964 CK_LIST_FOREACH(grp, hdr, il_list) {
1966 if (!(grp->il_vflag & INP_IPV4))
1969 if (grp->il_lport != lport)
1972 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1974 if (grp->il_laddr.s_addr == laddr->s_addr)
1975 return (grp->il_inp[idx]);
1976 if (grp->il_laddr.s_addr == INADDR_ANY &&
1977 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1978 local_wild = grp->il_inp[idx];
1980 return (local_wild);
1985 * Lookup PCB in hash list, using pcbgroup tables.
1987 static struct inpcb *
1988 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1989 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1990 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1992 struct inpcbhead *head;
1993 struct inpcb *inp, *tmpinp;
1994 u_short fport = fport_arg, lport = lport_arg;
1998 * First look for an exact match.
2001 INP_GROUP_LOCK(pcbgroup);
2002 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2003 pcbgroup->ipg_hashmask)];
2004 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2006 /* XXX inp locking */
2007 if ((inp->inp_vflag & INP_IPV4) == 0)
2010 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2011 inp->inp_laddr.s_addr == laddr.s_addr &&
2012 inp->inp_fport == fport &&
2013 inp->inp_lport == lport) {
2015 * XXX We should be able to directly return
2016 * the inp here, without any checks.
2017 * Well unless both bound with SO_REUSEPORT?
2019 if (prison_flag(inp->inp_cred, PR_IP4))
2025 if (tmpinp != NULL) {
2032 * For incoming connections, we may wish to do a wildcard
2033 * match for an RSS-local socket.
2035 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2036 struct inpcb *local_wild = NULL, *local_exact = NULL;
2038 struct inpcb *local_wild_mapped = NULL;
2040 struct inpcb *jail_wild = NULL;
2041 struct inpcbhead *head;
2045 * Order of socket selection - we always prefer jails.
2046 * 1. jailed, non-wild.
2048 * 3. non-jailed, non-wild.
2049 * 4. non-jailed, wild.
2052 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2053 lport, 0, pcbgroup->ipg_hashmask)];
2054 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2056 /* XXX inp locking */
2057 if ((inp->inp_vflag & INP_IPV4) == 0)
2060 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2061 inp->inp_lport != lport)
2064 injail = prison_flag(inp->inp_cred, PR_IP4);
2066 if (prison_check_ip4(inp->inp_cred,
2070 if (local_exact != NULL)
2074 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2079 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2081 /* XXX inp locking, NULL check */
2082 if (inp->inp_vflag & INP_IPV6PROTO)
2083 local_wild_mapped = inp;
2091 } /* LIST_FOREACH */
2100 inp = local_wild_mapped;
2108 * Then look for a wildcard match, if requested.
2110 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2111 struct inpcb *local_wild = NULL, *local_exact = NULL;
2113 struct inpcb *local_wild_mapped = NULL;
2115 struct inpcb *jail_wild = NULL;
2116 struct inpcbhead *head;
2120 * Order of socket selection - we always prefer jails.
2121 * 1. jailed, non-wild.
2123 * 3. non-jailed, non-wild.
2124 * 4. non-jailed, wild.
2126 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2127 0, pcbinfo->ipi_wildmask)];
2128 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2130 /* XXX inp locking */
2131 if ((inp->inp_vflag & INP_IPV4) == 0)
2134 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2135 inp->inp_lport != lport)
2138 injail = prison_flag(inp->inp_cred, PR_IP4);
2140 if (prison_check_ip4(inp->inp_cred,
2144 if (local_exact != NULL)
2148 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2153 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2155 /* XXX inp locking, NULL check */
2156 if (inp->inp_vflag & INP_IPV6PROTO)
2157 local_wild_mapped = inp;
2165 } /* LIST_FOREACH */
2173 inp = local_wild_mapped;
2177 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2178 INP_GROUP_UNLOCK(pcbgroup);
2182 if (lookupflags & INPLOOKUP_WLOCKPCB)
2183 locked = INP_TRY_WLOCK(inp);
2184 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2185 locked = INP_TRY_RLOCK(inp);
2187 panic("%s: locking bug", __func__);
2188 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2189 if (lookupflags & INPLOOKUP_WLOCKPCB)
2196 INP_GROUP_UNLOCK(pcbgroup);
2198 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2200 if (in_pcbrele_wlocked(inp))
2204 if (in_pcbrele_rlocked(inp))
2209 if (lookupflags & INPLOOKUP_WLOCKPCB)
2210 INP_WLOCK_ASSERT(inp);
2212 INP_RLOCK_ASSERT(inp);
2216 #endif /* PCBGROUP */
2219 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2220 * that the caller has locked the hash list, and will not perform any further
2221 * locking or reference operations on either the hash list or the connection.
2223 static struct inpcb *
2224 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2225 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2228 struct inpcbhead *head;
2229 struct inpcb *inp, *tmpinp;
2230 u_short fport = fport_arg, lport = lport_arg;
2233 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2234 ("%s: invalid lookup flags %d", __func__, lookupflags));
2235 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2236 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2239 * First look for an exact match.
2242 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2243 pcbinfo->ipi_hashmask)];
2244 CK_LIST_FOREACH(inp, head, inp_hash) {
2246 /* XXX inp locking */
2247 if ((inp->inp_vflag & INP_IPV4) == 0)
2250 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2251 inp->inp_laddr.s_addr == laddr.s_addr &&
2252 inp->inp_fport == fport &&
2253 inp->inp_lport == lport) {
2255 * XXX We should be able to directly return
2256 * the inp here, without any checks.
2257 * Well unless both bound with SO_REUSEPORT?
2259 if (prison_flag(inp->inp_cred, PR_IP4))
2269 * Then look in lb group (for wildcard match).
2271 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2272 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2273 fport, lookupflags);
2279 * Then look for a wildcard match, if requested.
2281 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2282 struct inpcb *local_wild = NULL, *local_exact = NULL;
2284 struct inpcb *local_wild_mapped = NULL;
2286 struct inpcb *jail_wild = NULL;
2290 * Order of socket selection - we always prefer jails.
2291 * 1. jailed, non-wild.
2293 * 3. non-jailed, non-wild.
2294 * 4. non-jailed, wild.
2297 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2298 0, pcbinfo->ipi_hashmask)];
2299 CK_LIST_FOREACH(inp, head, inp_hash) {
2301 /* XXX inp locking */
2302 if ((inp->inp_vflag & INP_IPV4) == 0)
2305 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2306 inp->inp_lport != lport)
2309 injail = prison_flag(inp->inp_cred, PR_IP4);
2311 if (prison_check_ip4(inp->inp_cred,
2315 if (local_exact != NULL)
2319 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2324 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2326 /* XXX inp locking, NULL check */
2327 if (inp->inp_vflag & INP_IPV6PROTO)
2328 local_wild_mapped = inp;
2336 } /* LIST_FOREACH */
2337 if (jail_wild != NULL)
2339 if (local_exact != NULL)
2340 return (local_exact);
2341 if (local_wild != NULL)
2342 return (local_wild);
2344 if (local_wild_mapped != NULL)
2345 return (local_wild_mapped);
2347 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2353 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2354 * hash list lock, and will return the inpcb locked (i.e., requires
2355 * INPLOOKUP_LOCKPCB).
2357 static struct inpcb *
2358 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2359 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2364 INP_HASH_RLOCK(pcbinfo);
2365 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2366 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2368 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2370 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2374 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2376 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2381 panic("%s: locking bug", __func__);
2384 if (lookupflags & INPLOOKUP_WLOCKPCB)
2385 INP_WLOCK_ASSERT(inp);
2387 INP_RLOCK_ASSERT(inp);
2391 INP_HASH_RUNLOCK(pcbinfo);
2396 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2397 * from which a pre-calculated hash value may be extracted.
2399 * Possibly more of this logic should be in in_pcbgroup.c.
2402 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2403 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2405 #if defined(PCBGROUP) && !defined(RSS)
2406 struct inpcbgroup *pcbgroup;
2409 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2410 ("%s: invalid lookup flags %d", __func__, lookupflags));
2411 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2412 ("%s: LOCKPCB not set", __func__));
2415 * When not using RSS, use connection groups in preference to the
2416 * reservation table when looking up 4-tuples. When using RSS, just
2417 * use the reservation table, due to the cost of the Toeplitz hash
2420 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2421 * we could be doing RSS with a non-Toeplitz hash that is affordable
2424 #if defined(PCBGROUP) && !defined(RSS)
2425 if (in_pcbgroup_enabled(pcbinfo)) {
2426 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2428 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2429 laddr, lport, lookupflags, ifp));
2432 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2437 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2438 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2439 struct ifnet *ifp, struct mbuf *m)
2442 struct inpcbgroup *pcbgroup;
2445 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2446 ("%s: invalid lookup flags %d", __func__, lookupflags));
2447 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2448 ("%s: LOCKPCB not set", __func__));
2452 * If we can use a hardware-generated hash to look up the connection
2453 * group, use that connection group to find the inpcb. Otherwise
2454 * fall back on a software hash -- or the reservation table if we're
2457 * XXXRW: As above, that policy belongs in the pcbgroup code.
2459 if (in_pcbgroup_enabled(pcbinfo) &&
2460 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2461 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2462 m->m_pkthdr.flowid);
2463 if (pcbgroup != NULL)
2464 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2465 fport, laddr, lport, lookupflags, ifp));
2467 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2469 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2470 laddr, lport, lookupflags, ifp));
2474 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2480 * Insert PCB onto various hash lists.
2483 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2485 struct inpcbhead *pcbhash;
2486 struct inpcbporthead *pcbporthash;
2487 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2488 struct inpcbport *phd;
2489 u_int32_t hashkey_faddr;
2492 INP_WLOCK_ASSERT(inp);
2493 INP_HASH_WLOCK_ASSERT(pcbinfo);
2495 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2496 ("in_pcbinshash: INP_INHASHLIST"));
2499 if (inp->inp_vflag & INP_IPV6)
2500 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2503 hashkey_faddr = inp->inp_faddr.s_addr;
2505 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2506 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2508 pcbporthash = &pcbinfo->ipi_porthashbase[
2509 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2512 * Add entry to load balance group.
2513 * Only do this if SO_REUSEPORT_LB is set.
2515 so_options = inp_so_options(inp);
2516 if (so_options & SO_REUSEPORT_LB) {
2517 int ret = in_pcbinslbgrouphash(inp);
2519 /* pcb lb group malloc fail (ret=ENOBUFS). */
2525 * Go through port list and look for a head for this lport.
2527 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2528 if (phd->phd_port == inp->inp_lport)
2532 * If none exists, malloc one and tack it on.
2535 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2537 return (ENOBUFS); /* XXX */
2539 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2540 phd->phd_port = inp->inp_lport;
2541 CK_LIST_INIT(&phd->phd_pcblist);
2542 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2545 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2546 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2547 inp->inp_flags |= INP_INHASHLIST;
2549 if (do_pcbgroup_update)
2550 in_pcbgroup_update(inp);
2556 * For now, there are two public interfaces to insert an inpcb into the hash
2557 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2558 * is used only in the TCP syncache, where in_pcbinshash is called before the
2559 * full 4-tuple is set for the inpcb, and we don't want to install in the
2560 * pcbgroup until later.
2562 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2563 * connection groups, and partially initialised inpcbs should not be exposed
2564 * to either reservation hash tables or pcbgroups.
2567 in_pcbinshash(struct inpcb *inp)
2570 return (in_pcbinshash_internal(inp, 1));
2574 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2577 return (in_pcbinshash_internal(inp, 0));
2581 * Move PCB to the proper hash bucket when { faddr, fport } have been
2582 * changed. NOTE: This does not handle the case of the lport changing (the
2583 * hashed port list would have to be updated as well), so the lport must
2584 * not change after in_pcbinshash() has been called.
2587 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2589 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2590 struct inpcbhead *head;
2591 u_int32_t hashkey_faddr;
2593 INP_WLOCK_ASSERT(inp);
2594 INP_HASH_WLOCK_ASSERT(pcbinfo);
2596 KASSERT(inp->inp_flags & INP_INHASHLIST,
2597 ("in_pcbrehash: !INP_INHASHLIST"));
2600 if (inp->inp_vflag & INP_IPV6)
2601 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2604 hashkey_faddr = inp->inp_faddr.s_addr;
2606 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2607 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2609 CK_LIST_REMOVE(inp, inp_hash);
2610 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2614 in_pcbgroup_update_mbuf(inp, m);
2616 in_pcbgroup_update(inp);
2621 in_pcbrehash(struct inpcb *inp)
2624 in_pcbrehash_mbuf(inp, NULL);
2628 * Remove PCB from various lists.
2631 in_pcbremlists(struct inpcb *inp)
2633 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2636 if (pcbinfo == &V_tcbinfo) {
2637 INP_INFO_RLOCK_ASSERT(pcbinfo);
2639 INP_INFO_WLOCK_ASSERT(pcbinfo);
2643 INP_WLOCK_ASSERT(inp);
2644 INP_LIST_WLOCK_ASSERT(pcbinfo);
2646 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2647 if (inp->inp_flags & INP_INHASHLIST) {
2648 struct inpcbport *phd = inp->inp_phd;
2650 INP_HASH_WLOCK(pcbinfo);
2652 /* XXX: Only do if SO_REUSEPORT_LB set? */
2653 in_pcbremlbgrouphash(inp);
2655 CK_LIST_REMOVE(inp, inp_hash);
2656 CK_LIST_REMOVE(inp, inp_portlist);
2657 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2658 CK_LIST_REMOVE(phd, phd_hash);
2659 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2661 INP_HASH_WUNLOCK(pcbinfo);
2662 inp->inp_flags &= ~INP_INHASHLIST;
2664 CK_LIST_REMOVE(inp, inp_list);
2665 pcbinfo->ipi_count--;
2667 in_pcbgroup_remove(inp);
2672 * Check for alternatives when higher level complains
2673 * about service problems. For now, invalidate cached
2674 * routing information. If the route was created dynamically
2675 * (by a redirect), time to try a default gateway again.
2678 in_losing(struct inpcb *inp)
2681 RO_INVALIDATE_CACHE(&inp->inp_route);
2686 * A set label operation has occurred at the socket layer, propagate the
2687 * label change into the in_pcb for the socket.
2690 in_pcbsosetlabel(struct socket *so)
2695 inp = sotoinpcb(so);
2696 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2700 mac_inpcb_sosetlabel(so, inp);
2707 * ipport_tick runs once per second, determining if random port allocation
2708 * should be continued. If more than ipport_randomcps ports have been
2709 * allocated in the last second, then we return to sequential port
2710 * allocation. We return to random allocation only once we drop below
2711 * ipport_randomcps for at least ipport_randomtime seconds.
2714 ipport_tick(void *xtp)
2716 VNET_ITERATOR_DECL(vnet_iter);
2718 VNET_LIST_RLOCK_NOSLEEP();
2719 VNET_FOREACH(vnet_iter) {
2720 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2721 if (V_ipport_tcpallocs <=
2722 V_ipport_tcplastcount + V_ipport_randomcps) {
2723 if (V_ipport_stoprandom > 0)
2724 V_ipport_stoprandom--;
2726 V_ipport_stoprandom = V_ipport_randomtime;
2727 V_ipport_tcplastcount = V_ipport_tcpallocs;
2730 VNET_LIST_RUNLOCK_NOSLEEP();
2731 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2738 callout_stop(&ipport_tick_callout);
2742 * The ipport_callout should start running at about the time we attach the
2743 * inet or inet6 domains.
2746 ipport_tick_init(const void *unused __unused)
2749 /* Start ipport_tick. */
2750 callout_init(&ipport_tick_callout, 1);
2751 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2752 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2753 SHUTDOWN_PRI_DEFAULT);
2755 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2756 ipport_tick_init, NULL);
2759 inp_wlock(struct inpcb *inp)
2766 inp_wunlock(struct inpcb *inp)
2773 inp_rlock(struct inpcb *inp)
2780 inp_runlock(struct inpcb *inp)
2786 #ifdef INVARIANT_SUPPORT
2788 inp_lock_assert(struct inpcb *inp)
2791 INP_WLOCK_ASSERT(inp);
2795 inp_unlock_assert(struct inpcb *inp)
2798 INP_UNLOCK_ASSERT(inp);
2803 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2807 INP_INFO_WLOCK(&V_tcbinfo);
2808 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2813 INP_INFO_WUNLOCK(&V_tcbinfo);
2817 inp_inpcbtosocket(struct inpcb *inp)
2820 INP_WLOCK_ASSERT(inp);
2821 return (inp->inp_socket);
2825 inp_inpcbtotcpcb(struct inpcb *inp)
2828 INP_WLOCK_ASSERT(inp);
2829 return ((struct tcpcb *)inp->inp_ppcb);
2833 inp_ip_tos_get(const struct inpcb *inp)
2836 return (inp->inp_ip_tos);
2840 inp_ip_tos_set(struct inpcb *inp, int val)
2843 inp->inp_ip_tos = val;
2847 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2848 uint32_t *faddr, uint16_t *fp)
2851 INP_LOCK_ASSERT(inp);
2852 *laddr = inp->inp_laddr.s_addr;
2853 *faddr = inp->inp_faddr.s_addr;
2854 *lp = inp->inp_lport;
2855 *fp = inp->inp_fport;
2859 so_sotoinpcb(struct socket *so)
2862 return (sotoinpcb(so));
2866 so_sototcpcb(struct socket *so)
2869 return (sototcpcb(so));
2873 * Create an external-format (``xinpcb'') structure using the information in
2874 * the kernel-format in_pcb structure pointed to by inp. This is done to
2875 * reduce the spew of irrelevant information over this interface, to isolate
2876 * user code from changes in the kernel structure, and potentially to provide
2877 * information-hiding if we decide that some of this information should be
2878 * hidden from users.
2881 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2884 bzero(xi, sizeof(*xi));
2885 xi->xi_len = sizeof(struct xinpcb);
2886 if (inp->inp_socket)
2887 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2888 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2889 xi->inp_gencnt = inp->inp_gencnt;
2890 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2891 xi->inp_flow = inp->inp_flow;
2892 xi->inp_flowid = inp->inp_flowid;
2893 xi->inp_flowtype = inp->inp_flowtype;
2894 xi->inp_flags = inp->inp_flags;
2895 xi->inp_flags2 = inp->inp_flags2;
2896 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2897 xi->in6p_cksum = inp->in6p_cksum;
2898 xi->in6p_hops = inp->in6p_hops;
2899 xi->inp_ip_tos = inp->inp_ip_tos;
2900 xi->inp_vflag = inp->inp_vflag;
2901 xi->inp_ip_ttl = inp->inp_ip_ttl;
2902 xi->inp_ip_p = inp->inp_ip_p;
2903 xi->inp_ip_minttl = inp->inp_ip_minttl;
2908 db_print_indent(int indent)
2912 for (i = 0; i < indent; i++)
2917 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2919 char faddr_str[48], laddr_str[48];
2921 db_print_indent(indent);
2922 db_printf("%s at %p\n", name, inc);
2927 if (inc->inc_flags & INC_ISIPV6) {
2929 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2930 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2935 inet_ntoa_r(inc->inc_laddr, laddr_str);
2936 inet_ntoa_r(inc->inc_faddr, faddr_str);
2938 db_print_indent(indent);
2939 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2940 ntohs(inc->inc_lport));
2941 db_print_indent(indent);
2942 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2943 ntohs(inc->inc_fport));
2947 db_print_inpflags(int inp_flags)
2952 if (inp_flags & INP_RECVOPTS) {
2953 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2956 if (inp_flags & INP_RECVRETOPTS) {
2957 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2960 if (inp_flags & INP_RECVDSTADDR) {
2961 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2964 if (inp_flags & INP_ORIGDSTADDR) {
2965 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2968 if (inp_flags & INP_HDRINCL) {
2969 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2972 if (inp_flags & INP_HIGHPORT) {
2973 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2976 if (inp_flags & INP_LOWPORT) {
2977 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2980 if (inp_flags & INP_ANONPORT) {
2981 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2984 if (inp_flags & INP_RECVIF) {
2985 db_printf("%sINP_RECVIF", comma ? ", " : "");
2988 if (inp_flags & INP_MTUDISC) {
2989 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2992 if (inp_flags & INP_RECVTTL) {
2993 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2996 if (inp_flags & INP_DONTFRAG) {
2997 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3000 if (inp_flags & INP_RECVTOS) {
3001 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3004 if (inp_flags & IN6P_IPV6_V6ONLY) {
3005 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3008 if (inp_flags & IN6P_PKTINFO) {
3009 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3012 if (inp_flags & IN6P_HOPLIMIT) {
3013 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3016 if (inp_flags & IN6P_HOPOPTS) {
3017 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3020 if (inp_flags & IN6P_DSTOPTS) {
3021 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3024 if (inp_flags & IN6P_RTHDR) {
3025 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3028 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3029 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3032 if (inp_flags & IN6P_TCLASS) {
3033 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3036 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3037 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3040 if (inp_flags & INP_TIMEWAIT) {
3041 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3044 if (inp_flags & INP_ONESBCAST) {
3045 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3048 if (inp_flags & INP_DROPPED) {
3049 db_printf("%sINP_DROPPED", comma ? ", " : "");
3052 if (inp_flags & INP_SOCKREF) {
3053 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3056 if (inp_flags & IN6P_RFC2292) {
3057 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3060 if (inp_flags & IN6P_MTU) {
3061 db_printf("IN6P_MTU%s", comma ? ", " : "");
3067 db_print_inpvflag(u_char inp_vflag)
3072 if (inp_vflag & INP_IPV4) {
3073 db_printf("%sINP_IPV4", comma ? ", " : "");
3076 if (inp_vflag & INP_IPV6) {
3077 db_printf("%sINP_IPV6", comma ? ", " : "");
3080 if (inp_vflag & INP_IPV6PROTO) {
3081 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3087 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3090 db_print_indent(indent);
3091 db_printf("%s at %p\n", name, inp);
3095 db_print_indent(indent);
3096 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3098 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3100 db_print_indent(indent);
3101 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3102 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3104 db_print_indent(indent);
3105 db_printf("inp_label: %p inp_flags: 0x%x (",
3106 inp->inp_label, inp->inp_flags);
3107 db_print_inpflags(inp->inp_flags);
3110 db_print_indent(indent);
3111 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3113 db_print_inpvflag(inp->inp_vflag);
3116 db_print_indent(indent);
3117 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3118 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3120 db_print_indent(indent);
3122 if (inp->inp_vflag & INP_IPV6) {
3123 db_printf("in6p_options: %p in6p_outputopts: %p "
3124 "in6p_moptions: %p\n", inp->in6p_options,
3125 inp->in6p_outputopts, inp->in6p_moptions);
3126 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3127 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3132 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3133 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3134 inp->inp_options, inp->inp_moptions);
3137 db_print_indent(indent);
3138 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3139 (uintmax_t)inp->inp_gencnt);
3142 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3147 db_printf("usage: show inpcb <addr>\n");
3150 inp = (struct inpcb *)addr;
3152 db_print_inpcb(inp, "inpcb", 0);
3158 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3162 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3164 union if_snd_tag_modify_params params = {
3165 .rate_limit.max_rate = max_pacing_rate,
3167 struct m_snd_tag *mst;
3171 mst = inp->inp_snd_tag;
3179 if (ifp->if_snd_tag_modify == NULL) {
3182 error = ifp->if_snd_tag_modify(mst, ¶ms);
3188 * Query existing TX rate limit based on the existing
3189 * "inp->inp_snd_tag", if any.
3192 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3194 union if_snd_tag_query_params params = { };
3195 struct m_snd_tag *mst;
3199 mst = inp->inp_snd_tag;
3207 if (ifp->if_snd_tag_query == NULL) {
3210 error = ifp->if_snd_tag_query(mst, ¶ms);
3211 if (error == 0 && p_max_pacing_rate != NULL)
3212 *p_max_pacing_rate = params.rate_limit.max_rate;
3218 * Query existing TX queue level based on the existing
3219 * "inp->inp_snd_tag", if any.
3222 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3224 union if_snd_tag_query_params params = { };
3225 struct m_snd_tag *mst;
3229 mst = inp->inp_snd_tag;
3237 if (ifp->if_snd_tag_query == NULL)
3238 return (EOPNOTSUPP);
3240 error = ifp->if_snd_tag_query(mst, ¶ms);
3241 if (error == 0 && p_txqueue_level != NULL)
3242 *p_txqueue_level = params.rate_limit.queue_level;
3247 * Allocate a new TX rate limit send tag from the network interface
3248 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3251 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3252 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3254 union if_snd_tag_alloc_params params = {
3255 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3256 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3257 .rate_limit.hdr.flowid = flowid,
3258 .rate_limit.hdr.flowtype = flowtype,
3259 .rate_limit.max_rate = max_pacing_rate,
3263 INP_WLOCK_ASSERT(inp);
3265 if (inp->inp_snd_tag != NULL)
3268 if (ifp->if_snd_tag_alloc == NULL) {
3271 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3274 * At success increment the refcount on
3275 * the send tag's network interface:
3278 if_ref(inp->inp_snd_tag->ifp);
3284 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3288 in_pcbdetach_txrtlmt(struct inpcb *inp)
3290 struct m_snd_tag *mst;
3293 INP_WLOCK_ASSERT(inp);
3295 mst = inp->inp_snd_tag;
3296 inp->inp_snd_tag = NULL;
3306 * If the device was detached while we still had reference(s)
3307 * on the ifp, we assume if_snd_tag_free() was replaced with
3310 ifp->if_snd_tag_free(mst);
3312 /* release reference count on network interface */
3317 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3318 * is set in the fast path and will attach/detach/modify the TX rate
3319 * limit send tag based on the socket's so_max_pacing_rate value.
3322 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3324 struct socket *socket;
3325 uint32_t max_pacing_rate;
3332 socket = inp->inp_socket;
3336 if (!INP_WLOCKED(inp)) {
3338 * NOTE: If the write locking fails, we need to bail
3339 * out and use the non-ratelimited ring for the
3340 * transmit until there is a new chance to get the
3343 if (!INP_TRY_UPGRADE(inp))
3351 * NOTE: The so_max_pacing_rate value is read unlocked,
3352 * because atomic updates are not required since the variable
3353 * is checked at every mbuf we send. It is assumed that the
3354 * variable read itself will be atomic.
3356 max_pacing_rate = socket->so_max_pacing_rate;
3359 * NOTE: When attaching to a network interface a reference is
3360 * made to ensure the network interface doesn't go away until
3361 * all ratelimit connections are gone. The network interface
3362 * pointers compared below represent valid network interfaces,
3363 * except when comparing towards NULL.
3365 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3367 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3368 if (inp->inp_snd_tag != NULL)
3369 in_pcbdetach_txrtlmt(inp);
3371 } else if (inp->inp_snd_tag == NULL) {
3373 * In order to utilize packet pacing with RSS, we need
3374 * to wait until there is a valid RSS hash before we
3377 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3380 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3381 mb->m_pkthdr.flowid, max_pacing_rate);
3384 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3386 if (error == 0 || error == EOPNOTSUPP)
3387 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3393 * Track route changes for TX rate limiting.
3396 in_pcboutput_eagain(struct inpcb *inp)
3398 struct socket *socket;
3404 socket = inp->inp_socket;
3408 if (inp->inp_snd_tag == NULL)
3411 if (!INP_WLOCKED(inp)) {
3413 * NOTE: If the write locking fails, we need to bail
3414 * out and use the non-ratelimited ring for the
3415 * transmit until there is a new chance to get the
3418 if (!INP_TRY_UPGRADE(inp))
3425 /* detach rate limiting */
3426 in_pcbdetach_txrtlmt(inp);
3428 /* make sure new mbuf send tag allocation is made */
3429 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3434 #endif /* RATELIMIT */