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, 0);
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,
872 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
873 priv_check_cred(inp->inp_cred,
874 PRIV_NETINET_REUSEPORT, 0) != 0) {
875 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
876 lport, INPLOOKUP_WILDCARD, cred);
879 * This entire block sorely needs a rewrite.
882 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
883 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
884 (so->so_type != SOCK_STREAM ||
885 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
886 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
887 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
888 (t->inp_flags2 & INP_REUSEPORT) ||
889 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
890 (inp->inp_cred->cr_uid !=
891 t->inp_cred->cr_uid))
895 * If the socket is a BINDMULTI socket, then
896 * the credentials need to match and the
897 * original socket also has to have been bound
900 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
903 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
904 lport, lookupflags, cred);
905 if (t && (t->inp_flags & INP_TIMEWAIT)) {
907 * XXXRW: If an incpb has had its timewait
908 * state recycled, we treat the address as
909 * being in use (for now). This is better
910 * than a panic, but not desirable.
914 ((reuseport & tw->tw_so_options) == 0 &&
916 tw->tw_so_options) == 0)) {
920 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
921 (reuseport & inp_so_options(t)) == 0 &&
922 (reuseport_lb & inp_so_options(t)) == 0) {
924 if (ntohl(sin->sin_addr.s_addr) !=
926 ntohl(t->inp_laddr.s_addr) !=
928 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
929 (t->inp_vflag & INP_IPV6PROTO) == 0)
932 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
940 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
945 *laddrp = laddr.s_addr;
951 * Connect from a socket to a specified address.
952 * Both address and port must be specified in argument sin.
953 * If don't have a local address for this socket yet,
957 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
958 struct ucred *cred, struct mbuf *m)
960 u_short lport, fport;
961 in_addr_t laddr, faddr;
964 INP_WLOCK_ASSERT(inp);
965 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
967 lport = inp->inp_lport;
968 laddr = inp->inp_laddr.s_addr;
969 anonport = (lport == 0);
970 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
975 /* Do the initial binding of the local address if required. */
976 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
977 inp->inp_lport = lport;
978 inp->inp_laddr.s_addr = laddr;
979 if (in_pcbinshash(inp) != 0) {
980 inp->inp_laddr.s_addr = INADDR_ANY;
986 /* Commit the remaining changes. */
987 inp->inp_lport = lport;
988 inp->inp_laddr.s_addr = laddr;
989 inp->inp_faddr.s_addr = faddr;
990 inp->inp_fport = fport;
991 in_pcbrehash_mbuf(inp, m);
994 inp->inp_flags |= INP_ANONPORT;
999 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1002 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1006 * Do proper source address selection on an unbound socket in case
1007 * of connect. Take jails into account as well.
1010 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1014 struct sockaddr *sa;
1015 struct sockaddr_in *sin;
1019 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1021 * Bypass source address selection and use the primary jail IP
1024 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1028 bzero(&sro, sizeof(sro));
1030 sin = (struct sockaddr_in *)&sro.ro_dst;
1031 sin->sin_family = AF_INET;
1032 sin->sin_len = sizeof(struct sockaddr_in);
1033 sin->sin_addr.s_addr = faddr->s_addr;
1036 * If route is known our src addr is taken from the i/f,
1039 * Find out route to destination.
1041 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1042 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1045 * If we found a route, use the address corresponding to
1046 * the outgoing interface.
1048 * Otherwise assume faddr is reachable on a directly connected
1049 * network and try to find a corresponding interface to take
1050 * the source address from.
1053 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1054 struct in_ifaddr *ia;
1057 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1058 inp->inp_socket->so_fibnum));
1060 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1061 inp->inp_socket->so_fibnum));
1065 error = ENETUNREACH;
1069 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1070 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1076 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1079 if (sa->sa_family != AF_INET)
1081 sin = (struct sockaddr_in *)sa;
1082 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1083 ia = (struct in_ifaddr *)ifa;
1088 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1092 /* 3. As a last resort return the 'default' jail address. */
1093 error = prison_get_ip4(cred, laddr);
1098 * If the outgoing interface on the route found is not
1099 * a loopback interface, use the address from that interface.
1100 * In case of jails do those three steps:
1101 * 1. check if the interface address belongs to the jail. If so use it.
1102 * 2. check if we have any address on the outgoing interface
1103 * belonging to this jail. If so use it.
1104 * 3. as a last resort return the 'default' jail address.
1106 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1107 struct in_ifaddr *ia;
1110 /* If not jailed, use the default returned. */
1111 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1112 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1113 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1118 /* 1. Check if the iface address belongs to the jail. */
1119 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1120 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1121 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1122 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1127 * 2. Check if we have any address on the outgoing interface
1128 * belonging to this jail.
1131 ifp = sro.ro_rt->rt_ifp;
1132 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1134 if (sa->sa_family != AF_INET)
1136 sin = (struct sockaddr_in *)sa;
1137 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1138 ia = (struct in_ifaddr *)ifa;
1143 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1147 /* 3. As a last resort return the 'default' jail address. */
1148 error = prison_get_ip4(cred, laddr);
1153 * The outgoing interface is marked with 'loopback net', so a route
1154 * to ourselves is here.
1155 * Try to find the interface of the destination address and then
1156 * take the address from there. That interface is not necessarily
1157 * a loopback interface.
1158 * In case of jails, check that it is an address of the jail
1159 * and if we cannot find, fall back to the 'default' jail address.
1161 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1162 struct sockaddr_in sain;
1163 struct in_ifaddr *ia;
1165 bzero(&sain, sizeof(struct sockaddr_in));
1166 sain.sin_family = AF_INET;
1167 sain.sin_len = sizeof(struct sockaddr_in);
1168 sain.sin_addr.s_addr = faddr->s_addr;
1170 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1171 inp->inp_socket->so_fibnum));
1173 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1174 inp->inp_socket->so_fibnum));
1176 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1178 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1180 error = ENETUNREACH;
1183 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1193 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1195 if (sa->sa_family != AF_INET)
1197 sin = (struct sockaddr_in *)sa;
1198 if (prison_check_ip4(cred,
1199 &sin->sin_addr) == 0) {
1200 ia = (struct in_ifaddr *)ifa;
1205 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1210 /* 3. As a last resort return the 'default' jail address. */
1211 error = prison_get_ip4(cred, laddr);
1217 if (sro.ro_rt != NULL)
1223 * Set up for a connect from a socket to the specified address.
1224 * On entry, *laddrp and *lportp should contain the current local
1225 * address and port for the PCB; these are updated to the values
1226 * that should be placed in inp_laddr and inp_lport to complete
1229 * On success, *faddrp and *fportp will be set to the remote address
1230 * and port. These are not updated in the error case.
1232 * If the operation fails because the connection already exists,
1233 * *oinpp will be set to the PCB of that connection so that the
1234 * caller can decide to override it. In all other cases, *oinpp
1238 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1239 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1240 struct inpcb **oinpp, struct ucred *cred)
1242 struct rm_priotracker in_ifa_tracker;
1243 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1244 struct in_ifaddr *ia;
1246 struct in_addr laddr, faddr;
1247 u_short lport, fport;
1251 * Because a global state change doesn't actually occur here, a read
1252 * lock is sufficient.
1254 INP_LOCK_ASSERT(inp);
1255 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1259 if (nam->sa_len != sizeof (*sin))
1261 if (sin->sin_family != AF_INET)
1262 return (EAFNOSUPPORT);
1263 if (sin->sin_port == 0)
1264 return (EADDRNOTAVAIL);
1265 laddr.s_addr = *laddrp;
1267 faddr = sin->sin_addr;
1268 fport = sin->sin_port;
1270 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1272 * If the destination address is INADDR_ANY,
1273 * use the primary local address.
1274 * If the supplied address is INADDR_BROADCAST,
1275 * and the primary interface supports broadcast,
1276 * choose the broadcast address for that interface.
1278 if (faddr.s_addr == INADDR_ANY) {
1279 IN_IFADDR_RLOCK(&in_ifa_tracker);
1281 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1282 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1284 (error = prison_get_ip4(cred, &faddr)) != 0)
1286 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1287 IN_IFADDR_RLOCK(&in_ifa_tracker);
1288 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1290 faddr = satosin(&CK_STAILQ_FIRST(
1291 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1292 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1295 if (laddr.s_addr == INADDR_ANY) {
1296 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1298 * If the destination address is multicast and an outgoing
1299 * interface has been set as a multicast option, prefer the
1300 * address of that interface as our source address.
1302 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1303 inp->inp_moptions != NULL) {
1304 struct ip_moptions *imo;
1307 imo = inp->inp_moptions;
1308 if (imo->imo_multicast_ifp != NULL) {
1309 ifp = imo->imo_multicast_ifp;
1310 IN_IFADDR_RLOCK(&in_ifa_tracker);
1311 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1312 if ((ia->ia_ifp == ifp) &&
1314 prison_check_ip4(cred,
1315 &ia->ia_addr.sin_addr) == 0))
1319 error = EADDRNOTAVAIL;
1321 laddr = ia->ia_addr.sin_addr;
1324 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1330 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1331 laddr, lport, 0, NULL);
1335 return (EADDRINUSE);
1338 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1343 *laddrp = laddr.s_addr;
1345 *faddrp = faddr.s_addr;
1351 in_pcbdisconnect(struct inpcb *inp)
1354 INP_WLOCK_ASSERT(inp);
1355 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1357 inp->inp_faddr.s_addr = INADDR_ANY;
1364 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1365 * For most protocols, this will be invoked immediately prior to calling
1366 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1367 * socket, in which case in_pcbfree() is deferred.
1370 in_pcbdetach(struct inpcb *inp)
1373 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1376 if (inp->inp_snd_tag != NULL)
1377 in_pcbdetach_txrtlmt(inp);
1379 inp->inp_socket->so_pcb = NULL;
1380 inp->inp_socket = NULL;
1384 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1385 * stability of an inpcb pointer despite the inpcb lock being released. This
1386 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1387 * but where the inpcb lock may already held, or when acquiring a reference
1390 * in_pcbref() should be used only to provide brief memory stability, and
1391 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1392 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1393 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1394 * lock and rele are the *only* safe operations that may be performed on the
1397 * While the inpcb will not be freed, releasing the inpcb lock means that the
1398 * connection's state may change, so the caller should be careful to
1399 * revalidate any cached state on reacquiring the lock. Drop the reference
1400 * using in_pcbrele().
1403 in_pcbref(struct inpcb *inp)
1406 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1408 refcount_acquire(&inp->inp_refcount);
1412 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1413 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1414 * return a flag indicating whether or not the inpcb remains valid. If it is
1415 * valid, we return with the inpcb lock held.
1417 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1418 * reference on an inpcb. Historically more work was done here (actually, in
1419 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1420 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1421 * about memory stability (and continued use of the write lock).
1424 in_pcbrele_rlocked(struct inpcb *inp)
1426 struct inpcbinfo *pcbinfo;
1428 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1430 INP_RLOCK_ASSERT(inp);
1432 if (refcount_release(&inp->inp_refcount) == 0) {
1434 * If the inpcb has been freed, let the caller know, even if
1435 * this isn't the last reference.
1437 if (inp->inp_flags2 & INP_FREED) {
1444 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1446 if (inp->inp_in_hpts || inp->inp_in_input) {
1447 struct tcp_hpts_entry *hpts;
1449 * We should not be on the hpts at
1450 * this point in any form. we must
1451 * get the lock to be sure.
1453 hpts = tcp_hpts_lock(inp);
1454 if (inp->inp_in_hpts)
1455 panic("Hpts:%p inp:%p at free still on hpts",
1457 mtx_unlock(&hpts->p_mtx);
1458 hpts = tcp_input_lock(inp);
1459 if (inp->inp_in_input)
1460 panic("Hpts:%p inp:%p at free still on input hpts",
1462 mtx_unlock(&hpts->p_mtx);
1466 pcbinfo = inp->inp_pcbinfo;
1467 uma_zfree(pcbinfo->ipi_zone, inp);
1472 in_pcbrele_wlocked(struct inpcb *inp)
1474 struct inpcbinfo *pcbinfo;
1476 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1478 INP_WLOCK_ASSERT(inp);
1480 if (refcount_release(&inp->inp_refcount) == 0) {
1482 * If the inpcb has been freed, let the caller know, even if
1483 * this isn't the last reference.
1485 if (inp->inp_flags2 & INP_FREED) {
1492 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1494 if (inp->inp_in_hpts || inp->inp_in_input) {
1495 struct tcp_hpts_entry *hpts;
1497 * We should not be on the hpts at
1498 * this point in any form. we must
1499 * get the lock to be sure.
1501 hpts = tcp_hpts_lock(inp);
1502 if (inp->inp_in_hpts)
1503 panic("Hpts:%p inp:%p at free still on hpts",
1505 mtx_unlock(&hpts->p_mtx);
1506 hpts = tcp_input_lock(inp);
1507 if (inp->inp_in_input)
1508 panic("Hpts:%p inp:%p at free still on input hpts",
1510 mtx_unlock(&hpts->p_mtx);
1514 pcbinfo = inp->inp_pcbinfo;
1515 uma_zfree(pcbinfo->ipi_zone, inp);
1520 * Temporary wrapper.
1523 in_pcbrele(struct inpcb *inp)
1526 return (in_pcbrele_wlocked(inp));
1530 in_pcblist_rele_rlocked(epoch_context_t ctx)
1532 struct in_pcblist *il;
1534 struct inpcbinfo *pcbinfo;
1537 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1538 pcbinfo = il->il_pcbinfo;
1540 INP_INFO_WLOCK(pcbinfo);
1541 for (i = 0; i < n; i++) {
1542 inp = il->il_inp_list[i];
1544 if (!in_pcbrele_rlocked(inp))
1547 INP_INFO_WUNLOCK(pcbinfo);
1552 inpcbport_free(epoch_context_t ctx)
1554 struct inpcbport *phd;
1556 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1561 in_pcbfree_deferred(epoch_context_t ctx)
1564 int released __unused;
1566 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1570 struct ip_moptions *imo = inp->inp_moptions;
1571 inp->inp_moptions = NULL;
1573 /* XXXRW: Do as much as possible here. */
1574 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1575 if (inp->inp_sp != NULL)
1576 ipsec_delete_pcbpolicy(inp);
1579 struct ip6_moptions *im6o = NULL;
1580 if (inp->inp_vflag & INP_IPV6PROTO) {
1581 ip6_freepcbopts(inp->in6p_outputopts);
1582 im6o = inp->in6p_moptions;
1583 inp->in6p_moptions = NULL;
1586 if (inp->inp_options)
1587 (void)m_free(inp->inp_options);
1589 crfree(inp->inp_cred);
1591 mac_inpcb_destroy(inp);
1593 released = in_pcbrele_wlocked(inp);
1596 ip6_freemoptions(im6o);
1599 inp_freemoptions(imo);
1604 * Unconditionally schedule an inpcb to be freed by decrementing its
1605 * reference count, which should occur only after the inpcb has been detached
1606 * from its socket. If another thread holds a temporary reference (acquired
1607 * using in_pcbref()) then the free is deferred until that reference is
1608 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1609 * work, including removal from global lists, is done in this context, where
1610 * the pcbinfo lock is held.
1613 in_pcbfree(struct inpcb *inp)
1615 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1617 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1618 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1619 ("%s: called twice for pcb %p", __func__, inp));
1620 if (inp->inp_flags2 & INP_FREED) {
1626 if (pcbinfo == &V_tcbinfo) {
1627 INP_INFO_LOCK_ASSERT(pcbinfo);
1629 INP_INFO_WLOCK_ASSERT(pcbinfo);
1632 INP_WLOCK_ASSERT(inp);
1633 INP_LIST_WLOCK(pcbinfo);
1634 in_pcbremlists(inp);
1635 INP_LIST_WUNLOCK(pcbinfo);
1636 RO_INVALIDATE_CACHE(&inp->inp_route);
1637 /* mark as destruction in progress */
1638 inp->inp_flags2 |= INP_FREED;
1640 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1644 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1645 * port reservation, and preventing it from being returned by inpcb lookups.
1647 * It is used by TCP to mark an inpcb as unused and avoid future packet
1648 * delivery or event notification when a socket remains open but TCP has
1649 * closed. This might occur as a result of a shutdown()-initiated TCP close
1650 * or a RST on the wire, and allows the port binding to be reused while still
1651 * maintaining the invariant that so_pcb always points to a valid inpcb until
1654 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1655 * in_pcbnotifyall() and in_pcbpurgeif0()?
1658 in_pcbdrop(struct inpcb *inp)
1661 INP_WLOCK_ASSERT(inp);
1663 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1664 MPASS(inp->inp_refcount > 1);
1668 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1671 inp->inp_flags |= INP_DROPPED;
1672 if (inp->inp_flags & INP_INHASHLIST) {
1673 struct inpcbport *phd = inp->inp_phd;
1675 INP_HASH_WLOCK(inp->inp_pcbinfo);
1676 in_pcbremlbgrouphash(inp);
1677 CK_LIST_REMOVE(inp, inp_hash);
1678 CK_LIST_REMOVE(inp, inp_portlist);
1679 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1680 CK_LIST_REMOVE(phd, phd_hash);
1681 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1683 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1684 inp->inp_flags &= ~INP_INHASHLIST;
1686 in_pcbgroup_remove(inp);
1693 * Common routines to return the socket addresses associated with inpcbs.
1696 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1698 struct sockaddr_in *sin;
1700 sin = malloc(sizeof *sin, M_SONAME,
1702 sin->sin_family = AF_INET;
1703 sin->sin_len = sizeof(*sin);
1704 sin->sin_addr = *addr_p;
1705 sin->sin_port = port;
1707 return (struct sockaddr *)sin;
1711 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1714 struct in_addr addr;
1717 inp = sotoinpcb(so);
1718 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1721 port = inp->inp_lport;
1722 addr = inp->inp_laddr;
1725 *nam = in_sockaddr(port, &addr);
1730 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1733 struct in_addr addr;
1736 inp = sotoinpcb(so);
1737 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1740 port = inp->inp_fport;
1741 addr = inp->inp_faddr;
1744 *nam = in_sockaddr(port, &addr);
1749 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1750 struct inpcb *(*notify)(struct inpcb *, int))
1752 struct inpcb *inp, *inp_temp;
1754 INP_INFO_WLOCK(pcbinfo);
1755 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1758 if ((inp->inp_vflag & INP_IPV4) == 0) {
1763 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1764 inp->inp_socket == NULL) {
1768 if ((*notify)(inp, errno))
1771 INP_INFO_WUNLOCK(pcbinfo);
1775 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1778 struct ip_moptions *imo;
1781 INP_INFO_WLOCK(pcbinfo);
1782 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1784 imo = inp->inp_moptions;
1785 if ((inp->inp_vflag & INP_IPV4) &&
1788 * Unselect the outgoing interface if it is being
1791 if (imo->imo_multicast_ifp == ifp)
1792 imo->imo_multicast_ifp = NULL;
1795 * Drop multicast group membership if we joined
1796 * through the interface being detached.
1798 * XXX This can all be deferred to an epoch_call
1800 for (i = 0, gap = 0; i < imo->imo_num_memberships;
1802 if (imo->imo_membership[i]->inm_ifp == ifp) {
1803 IN_MULTI_LOCK_ASSERT();
1804 in_leavegroup_locked(imo->imo_membership[i], NULL);
1806 } else if (gap != 0)
1807 imo->imo_membership[i - gap] =
1808 imo->imo_membership[i];
1810 imo->imo_num_memberships -= gap;
1814 INP_INFO_WUNLOCK(pcbinfo);
1818 * Lookup a PCB based on the local address and port. Caller must hold the
1819 * hash lock. No inpcb locks or references are acquired.
1821 #define INP_LOOKUP_MAPPED_PCB_COST 3
1823 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1824 u_short lport, int lookupflags, struct ucred *cred)
1828 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1834 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1835 ("%s: invalid lookup flags %d", __func__, lookupflags));
1837 INP_HASH_LOCK_ASSERT(pcbinfo);
1839 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1840 struct inpcbhead *head;
1842 * Look for an unconnected (wildcard foreign addr) PCB that
1843 * matches the local address and port we're looking for.
1845 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1846 0, pcbinfo->ipi_hashmask)];
1847 CK_LIST_FOREACH(inp, head, inp_hash) {
1849 /* XXX inp locking */
1850 if ((inp->inp_vflag & INP_IPV4) == 0)
1853 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1854 inp->inp_laddr.s_addr == laddr.s_addr &&
1855 inp->inp_lport == lport) {
1860 prison_equal_ip4(cred->cr_prison,
1861 inp->inp_cred->cr_prison))
1870 struct inpcbporthead *porthash;
1871 struct inpcbport *phd;
1872 struct inpcb *match = NULL;
1874 * Best fit PCB lookup.
1876 * First see if this local port is in use by looking on the
1879 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1880 pcbinfo->ipi_porthashmask)];
1881 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1882 if (phd->phd_port == lport)
1887 * Port is in use by one or more PCBs. Look for best
1890 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1893 !prison_equal_ip4(inp->inp_cred->cr_prison,
1897 /* XXX inp locking */
1898 if ((inp->inp_vflag & INP_IPV4) == 0)
1901 * We never select the PCB that has
1902 * INP_IPV6 flag and is bound to :: if
1903 * we have another PCB which is bound
1904 * to 0.0.0.0. If a PCB has the
1905 * INP_IPV6 flag, then we set its cost
1906 * higher than IPv4 only PCBs.
1908 * Note that the case only happens
1909 * when a socket is bound to ::, under
1910 * the condition that the use of the
1911 * mapped address is allowed.
1913 if ((inp->inp_vflag & INP_IPV6) != 0)
1914 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1916 if (inp->inp_faddr.s_addr != INADDR_ANY)
1918 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1919 if (laddr.s_addr == INADDR_ANY)
1921 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1924 if (laddr.s_addr != INADDR_ANY)
1927 if (wildcard < matchwild) {
1929 matchwild = wildcard;
1938 #undef INP_LOOKUP_MAPPED_PCB_COST
1940 static struct inpcb *
1941 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1942 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1943 uint16_t fport, int lookupflags)
1945 struct inpcb *local_wild;
1946 const struct inpcblbgrouphead *hdr;
1947 struct inpcblbgroup *grp;
1950 INP_HASH_LOCK_ASSERT(pcbinfo);
1952 hdr = &pcbinfo->ipi_lbgrouphashbase[
1953 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1956 * Order of socket selection:
1958 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1961 * - Load balanced group does not contain jailed sockets
1962 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1965 CK_LIST_FOREACH(grp, hdr, il_list) {
1967 if (!(grp->il_vflag & INP_IPV4))
1970 if (grp->il_lport != lport)
1973 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1975 if (grp->il_laddr.s_addr == laddr->s_addr)
1976 return (grp->il_inp[idx]);
1977 if (grp->il_laddr.s_addr == INADDR_ANY &&
1978 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1979 local_wild = grp->il_inp[idx];
1981 return (local_wild);
1986 * Lookup PCB in hash list, using pcbgroup tables.
1988 static struct inpcb *
1989 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1990 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1991 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1993 struct inpcbhead *head;
1994 struct inpcb *inp, *tmpinp;
1995 u_short fport = fport_arg, lport = lport_arg;
1999 * First look for an exact match.
2002 INP_GROUP_LOCK(pcbgroup);
2003 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2004 pcbgroup->ipg_hashmask)];
2005 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2007 /* XXX inp locking */
2008 if ((inp->inp_vflag & INP_IPV4) == 0)
2011 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2012 inp->inp_laddr.s_addr == laddr.s_addr &&
2013 inp->inp_fport == fport &&
2014 inp->inp_lport == lport) {
2016 * XXX We should be able to directly return
2017 * the inp here, without any checks.
2018 * Well unless both bound with SO_REUSEPORT?
2020 if (prison_flag(inp->inp_cred, PR_IP4))
2026 if (tmpinp != NULL) {
2033 * For incoming connections, we may wish to do a wildcard
2034 * match for an RSS-local socket.
2036 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2037 struct inpcb *local_wild = NULL, *local_exact = NULL;
2039 struct inpcb *local_wild_mapped = NULL;
2041 struct inpcb *jail_wild = NULL;
2042 struct inpcbhead *head;
2046 * Order of socket selection - we always prefer jails.
2047 * 1. jailed, non-wild.
2049 * 3. non-jailed, non-wild.
2050 * 4. non-jailed, wild.
2053 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2054 lport, 0, pcbgroup->ipg_hashmask)];
2055 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2057 /* XXX inp locking */
2058 if ((inp->inp_vflag & INP_IPV4) == 0)
2061 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2062 inp->inp_lport != lport)
2065 injail = prison_flag(inp->inp_cred, PR_IP4);
2067 if (prison_check_ip4(inp->inp_cred,
2071 if (local_exact != NULL)
2075 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2080 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2082 /* XXX inp locking, NULL check */
2083 if (inp->inp_vflag & INP_IPV6PROTO)
2084 local_wild_mapped = inp;
2092 } /* LIST_FOREACH */
2101 inp = local_wild_mapped;
2109 * Then look for a wildcard match, if requested.
2111 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2112 struct inpcb *local_wild = NULL, *local_exact = NULL;
2114 struct inpcb *local_wild_mapped = NULL;
2116 struct inpcb *jail_wild = NULL;
2117 struct inpcbhead *head;
2121 * Order of socket selection - we always prefer jails.
2122 * 1. jailed, non-wild.
2124 * 3. non-jailed, non-wild.
2125 * 4. non-jailed, wild.
2127 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2128 0, pcbinfo->ipi_wildmask)];
2129 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2131 /* XXX inp locking */
2132 if ((inp->inp_vflag & INP_IPV4) == 0)
2135 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2136 inp->inp_lport != lport)
2139 injail = prison_flag(inp->inp_cred, PR_IP4);
2141 if (prison_check_ip4(inp->inp_cred,
2145 if (local_exact != NULL)
2149 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2154 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2156 /* XXX inp locking, NULL check */
2157 if (inp->inp_vflag & INP_IPV6PROTO)
2158 local_wild_mapped = inp;
2166 } /* LIST_FOREACH */
2174 inp = local_wild_mapped;
2178 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2179 INP_GROUP_UNLOCK(pcbgroup);
2183 if (lookupflags & INPLOOKUP_WLOCKPCB)
2184 locked = INP_TRY_WLOCK(inp);
2185 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2186 locked = INP_TRY_RLOCK(inp);
2188 panic("%s: locking bug", __func__);
2189 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2190 if (lookupflags & INPLOOKUP_WLOCKPCB)
2197 INP_GROUP_UNLOCK(pcbgroup);
2199 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2201 if (in_pcbrele_wlocked(inp))
2205 if (in_pcbrele_rlocked(inp))
2210 if (lookupflags & INPLOOKUP_WLOCKPCB)
2211 INP_WLOCK_ASSERT(inp);
2213 INP_RLOCK_ASSERT(inp);
2217 #endif /* PCBGROUP */
2220 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2221 * that the caller has locked the hash list, and will not perform any further
2222 * locking or reference operations on either the hash list or the connection.
2224 static struct inpcb *
2225 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2226 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2229 struct inpcbhead *head;
2230 struct inpcb *inp, *tmpinp;
2231 u_short fport = fport_arg, lport = lport_arg;
2234 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2235 ("%s: invalid lookup flags %d", __func__, lookupflags));
2236 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2237 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2240 * First look for an exact match.
2243 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2244 pcbinfo->ipi_hashmask)];
2245 CK_LIST_FOREACH(inp, head, inp_hash) {
2247 /* XXX inp locking */
2248 if ((inp->inp_vflag & INP_IPV4) == 0)
2251 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2252 inp->inp_laddr.s_addr == laddr.s_addr &&
2253 inp->inp_fport == fport &&
2254 inp->inp_lport == lport) {
2256 * XXX We should be able to directly return
2257 * the inp here, without any checks.
2258 * Well unless both bound with SO_REUSEPORT?
2260 if (prison_flag(inp->inp_cred, PR_IP4))
2270 * Then look in lb group (for wildcard match).
2272 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2273 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2274 fport, lookupflags);
2280 * Then look for a wildcard match, if requested.
2282 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2283 struct inpcb *local_wild = NULL, *local_exact = NULL;
2285 struct inpcb *local_wild_mapped = NULL;
2287 struct inpcb *jail_wild = NULL;
2291 * Order of socket selection - we always prefer jails.
2292 * 1. jailed, non-wild.
2294 * 3. non-jailed, non-wild.
2295 * 4. non-jailed, wild.
2298 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2299 0, pcbinfo->ipi_hashmask)];
2300 CK_LIST_FOREACH(inp, head, inp_hash) {
2302 /* XXX inp locking */
2303 if ((inp->inp_vflag & INP_IPV4) == 0)
2306 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2307 inp->inp_lport != lport)
2310 injail = prison_flag(inp->inp_cred, PR_IP4);
2312 if (prison_check_ip4(inp->inp_cred,
2316 if (local_exact != NULL)
2320 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2325 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2327 /* XXX inp locking, NULL check */
2328 if (inp->inp_vflag & INP_IPV6PROTO)
2329 local_wild_mapped = inp;
2337 } /* LIST_FOREACH */
2338 if (jail_wild != NULL)
2340 if (local_exact != NULL)
2341 return (local_exact);
2342 if (local_wild != NULL)
2343 return (local_wild);
2345 if (local_wild_mapped != NULL)
2346 return (local_wild_mapped);
2348 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2354 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2355 * hash list lock, and will return the inpcb locked (i.e., requires
2356 * INPLOOKUP_LOCKPCB).
2358 static struct inpcb *
2359 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2360 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2365 INP_HASH_RLOCK(pcbinfo);
2366 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2367 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2369 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2371 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2375 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2377 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2382 panic("%s: locking bug", __func__);
2385 if (lookupflags & INPLOOKUP_WLOCKPCB)
2386 INP_WLOCK_ASSERT(inp);
2388 INP_RLOCK_ASSERT(inp);
2392 INP_HASH_RUNLOCK(pcbinfo);
2397 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2398 * from which a pre-calculated hash value may be extracted.
2400 * Possibly more of this logic should be in in_pcbgroup.c.
2403 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2404 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2406 #if defined(PCBGROUP) && !defined(RSS)
2407 struct inpcbgroup *pcbgroup;
2410 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2411 ("%s: invalid lookup flags %d", __func__, lookupflags));
2412 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2413 ("%s: LOCKPCB not set", __func__));
2416 * When not using RSS, use connection groups in preference to the
2417 * reservation table when looking up 4-tuples. When using RSS, just
2418 * use the reservation table, due to the cost of the Toeplitz hash
2421 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2422 * we could be doing RSS with a non-Toeplitz hash that is affordable
2425 #if defined(PCBGROUP) && !defined(RSS)
2426 if (in_pcbgroup_enabled(pcbinfo)) {
2427 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2429 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2430 laddr, lport, lookupflags, ifp));
2433 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2438 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2439 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2440 struct ifnet *ifp, struct mbuf *m)
2443 struct inpcbgroup *pcbgroup;
2446 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2447 ("%s: invalid lookup flags %d", __func__, lookupflags));
2448 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2449 ("%s: LOCKPCB not set", __func__));
2453 * If we can use a hardware-generated hash to look up the connection
2454 * group, use that connection group to find the inpcb. Otherwise
2455 * fall back on a software hash -- or the reservation table if we're
2458 * XXXRW: As above, that policy belongs in the pcbgroup code.
2460 if (in_pcbgroup_enabled(pcbinfo) &&
2461 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2462 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2463 m->m_pkthdr.flowid);
2464 if (pcbgroup != NULL)
2465 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2466 fport, laddr, lport, lookupflags, ifp));
2468 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2470 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2471 laddr, lport, lookupflags, ifp));
2475 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2481 * Insert PCB onto various hash lists.
2484 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2486 struct inpcbhead *pcbhash;
2487 struct inpcbporthead *pcbporthash;
2488 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2489 struct inpcbport *phd;
2490 u_int32_t hashkey_faddr;
2493 INP_WLOCK_ASSERT(inp);
2494 INP_HASH_WLOCK_ASSERT(pcbinfo);
2496 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2497 ("in_pcbinshash: INP_INHASHLIST"));
2500 if (inp->inp_vflag & INP_IPV6)
2501 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2504 hashkey_faddr = inp->inp_faddr.s_addr;
2506 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2507 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2509 pcbporthash = &pcbinfo->ipi_porthashbase[
2510 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2513 * Add entry to load balance group.
2514 * Only do this if SO_REUSEPORT_LB is set.
2516 so_options = inp_so_options(inp);
2517 if (so_options & SO_REUSEPORT_LB) {
2518 int ret = in_pcbinslbgrouphash(inp);
2520 /* pcb lb group malloc fail (ret=ENOBUFS). */
2526 * Go through port list and look for a head for this lport.
2528 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2529 if (phd->phd_port == inp->inp_lport)
2533 * If none exists, malloc one and tack it on.
2536 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2538 return (ENOBUFS); /* XXX */
2540 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2541 phd->phd_port = inp->inp_lport;
2542 CK_LIST_INIT(&phd->phd_pcblist);
2543 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2546 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2547 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2548 inp->inp_flags |= INP_INHASHLIST;
2550 if (do_pcbgroup_update)
2551 in_pcbgroup_update(inp);
2557 * For now, there are two public interfaces to insert an inpcb into the hash
2558 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2559 * is used only in the TCP syncache, where in_pcbinshash is called before the
2560 * full 4-tuple is set for the inpcb, and we don't want to install in the
2561 * pcbgroup until later.
2563 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2564 * connection groups, and partially initialised inpcbs should not be exposed
2565 * to either reservation hash tables or pcbgroups.
2568 in_pcbinshash(struct inpcb *inp)
2571 return (in_pcbinshash_internal(inp, 1));
2575 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2578 return (in_pcbinshash_internal(inp, 0));
2582 * Move PCB to the proper hash bucket when { faddr, fport } have been
2583 * changed. NOTE: This does not handle the case of the lport changing (the
2584 * hashed port list would have to be updated as well), so the lport must
2585 * not change after in_pcbinshash() has been called.
2588 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2590 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2591 struct inpcbhead *head;
2592 u_int32_t hashkey_faddr;
2594 INP_WLOCK_ASSERT(inp);
2595 INP_HASH_WLOCK_ASSERT(pcbinfo);
2597 KASSERT(inp->inp_flags & INP_INHASHLIST,
2598 ("in_pcbrehash: !INP_INHASHLIST"));
2601 if (inp->inp_vflag & INP_IPV6)
2602 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2605 hashkey_faddr = inp->inp_faddr.s_addr;
2607 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2608 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2610 CK_LIST_REMOVE(inp, inp_hash);
2611 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2615 in_pcbgroup_update_mbuf(inp, m);
2617 in_pcbgroup_update(inp);
2622 in_pcbrehash(struct inpcb *inp)
2625 in_pcbrehash_mbuf(inp, NULL);
2629 * Remove PCB from various lists.
2632 in_pcbremlists(struct inpcb *inp)
2634 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2637 if (pcbinfo == &V_tcbinfo) {
2638 INP_INFO_RLOCK_ASSERT(pcbinfo);
2640 INP_INFO_WLOCK_ASSERT(pcbinfo);
2644 INP_WLOCK_ASSERT(inp);
2645 INP_LIST_WLOCK_ASSERT(pcbinfo);
2647 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2648 if (inp->inp_flags & INP_INHASHLIST) {
2649 struct inpcbport *phd = inp->inp_phd;
2651 INP_HASH_WLOCK(pcbinfo);
2653 /* XXX: Only do if SO_REUSEPORT_LB set? */
2654 in_pcbremlbgrouphash(inp);
2656 CK_LIST_REMOVE(inp, inp_hash);
2657 CK_LIST_REMOVE(inp, inp_portlist);
2658 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2659 CK_LIST_REMOVE(phd, phd_hash);
2660 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2662 INP_HASH_WUNLOCK(pcbinfo);
2663 inp->inp_flags &= ~INP_INHASHLIST;
2665 CK_LIST_REMOVE(inp, inp_list);
2666 pcbinfo->ipi_count--;
2668 in_pcbgroup_remove(inp);
2673 * Check for alternatives when higher level complains
2674 * about service problems. For now, invalidate cached
2675 * routing information. If the route was created dynamically
2676 * (by a redirect), time to try a default gateway again.
2679 in_losing(struct inpcb *inp)
2682 RO_INVALIDATE_CACHE(&inp->inp_route);
2687 * A set label operation has occurred at the socket layer, propagate the
2688 * label change into the in_pcb for the socket.
2691 in_pcbsosetlabel(struct socket *so)
2696 inp = sotoinpcb(so);
2697 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2701 mac_inpcb_sosetlabel(so, inp);
2708 * ipport_tick runs once per second, determining if random port allocation
2709 * should be continued. If more than ipport_randomcps ports have been
2710 * allocated in the last second, then we return to sequential port
2711 * allocation. We return to random allocation only once we drop below
2712 * ipport_randomcps for at least ipport_randomtime seconds.
2715 ipport_tick(void *xtp)
2717 VNET_ITERATOR_DECL(vnet_iter);
2719 VNET_LIST_RLOCK_NOSLEEP();
2720 VNET_FOREACH(vnet_iter) {
2721 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2722 if (V_ipport_tcpallocs <=
2723 V_ipport_tcplastcount + V_ipport_randomcps) {
2724 if (V_ipport_stoprandom > 0)
2725 V_ipport_stoprandom--;
2727 V_ipport_stoprandom = V_ipport_randomtime;
2728 V_ipport_tcplastcount = V_ipport_tcpallocs;
2731 VNET_LIST_RUNLOCK_NOSLEEP();
2732 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2739 callout_stop(&ipport_tick_callout);
2743 * The ipport_callout should start running at about the time we attach the
2744 * inet or inet6 domains.
2747 ipport_tick_init(const void *unused __unused)
2750 /* Start ipport_tick. */
2751 callout_init(&ipport_tick_callout, 1);
2752 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2753 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2754 SHUTDOWN_PRI_DEFAULT);
2756 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2757 ipport_tick_init, NULL);
2760 inp_wlock(struct inpcb *inp)
2767 inp_wunlock(struct inpcb *inp)
2774 inp_rlock(struct inpcb *inp)
2781 inp_runlock(struct inpcb *inp)
2787 #ifdef INVARIANT_SUPPORT
2789 inp_lock_assert(struct inpcb *inp)
2792 INP_WLOCK_ASSERT(inp);
2796 inp_unlock_assert(struct inpcb *inp)
2799 INP_UNLOCK_ASSERT(inp);
2804 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2808 INP_INFO_WLOCK(&V_tcbinfo);
2809 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2814 INP_INFO_WUNLOCK(&V_tcbinfo);
2818 inp_inpcbtosocket(struct inpcb *inp)
2821 INP_WLOCK_ASSERT(inp);
2822 return (inp->inp_socket);
2826 inp_inpcbtotcpcb(struct inpcb *inp)
2829 INP_WLOCK_ASSERT(inp);
2830 return ((struct tcpcb *)inp->inp_ppcb);
2834 inp_ip_tos_get(const struct inpcb *inp)
2837 return (inp->inp_ip_tos);
2841 inp_ip_tos_set(struct inpcb *inp, int val)
2844 inp->inp_ip_tos = val;
2848 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2849 uint32_t *faddr, uint16_t *fp)
2852 INP_LOCK_ASSERT(inp);
2853 *laddr = inp->inp_laddr.s_addr;
2854 *faddr = inp->inp_faddr.s_addr;
2855 *lp = inp->inp_lport;
2856 *fp = inp->inp_fport;
2860 so_sotoinpcb(struct socket *so)
2863 return (sotoinpcb(so));
2867 so_sototcpcb(struct socket *so)
2870 return (sototcpcb(so));
2874 * Create an external-format (``xinpcb'') structure using the information in
2875 * the kernel-format in_pcb structure pointed to by inp. This is done to
2876 * reduce the spew of irrelevant information over this interface, to isolate
2877 * user code from changes in the kernel structure, and potentially to provide
2878 * information-hiding if we decide that some of this information should be
2879 * hidden from users.
2882 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2885 bzero(xi, sizeof(*xi));
2886 xi->xi_len = sizeof(struct xinpcb);
2887 if (inp->inp_socket)
2888 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2889 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2890 xi->inp_gencnt = inp->inp_gencnt;
2891 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2892 xi->inp_flow = inp->inp_flow;
2893 xi->inp_flowid = inp->inp_flowid;
2894 xi->inp_flowtype = inp->inp_flowtype;
2895 xi->inp_flags = inp->inp_flags;
2896 xi->inp_flags2 = inp->inp_flags2;
2897 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2898 xi->in6p_cksum = inp->in6p_cksum;
2899 xi->in6p_hops = inp->in6p_hops;
2900 xi->inp_ip_tos = inp->inp_ip_tos;
2901 xi->inp_vflag = inp->inp_vflag;
2902 xi->inp_ip_ttl = inp->inp_ip_ttl;
2903 xi->inp_ip_p = inp->inp_ip_p;
2904 xi->inp_ip_minttl = inp->inp_ip_minttl;
2909 db_print_indent(int indent)
2913 for (i = 0; i < indent; i++)
2918 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2920 char faddr_str[48], laddr_str[48];
2922 db_print_indent(indent);
2923 db_printf("%s at %p\n", name, inc);
2928 if (inc->inc_flags & INC_ISIPV6) {
2930 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2931 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2936 inet_ntoa_r(inc->inc_laddr, laddr_str);
2937 inet_ntoa_r(inc->inc_faddr, faddr_str);
2939 db_print_indent(indent);
2940 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2941 ntohs(inc->inc_lport));
2942 db_print_indent(indent);
2943 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2944 ntohs(inc->inc_fport));
2948 db_print_inpflags(int inp_flags)
2953 if (inp_flags & INP_RECVOPTS) {
2954 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2957 if (inp_flags & INP_RECVRETOPTS) {
2958 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2961 if (inp_flags & INP_RECVDSTADDR) {
2962 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2965 if (inp_flags & INP_ORIGDSTADDR) {
2966 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2969 if (inp_flags & INP_HDRINCL) {
2970 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2973 if (inp_flags & INP_HIGHPORT) {
2974 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2977 if (inp_flags & INP_LOWPORT) {
2978 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2981 if (inp_flags & INP_ANONPORT) {
2982 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2985 if (inp_flags & INP_RECVIF) {
2986 db_printf("%sINP_RECVIF", comma ? ", " : "");
2989 if (inp_flags & INP_MTUDISC) {
2990 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2993 if (inp_flags & INP_RECVTTL) {
2994 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2997 if (inp_flags & INP_DONTFRAG) {
2998 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3001 if (inp_flags & INP_RECVTOS) {
3002 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3005 if (inp_flags & IN6P_IPV6_V6ONLY) {
3006 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3009 if (inp_flags & IN6P_PKTINFO) {
3010 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3013 if (inp_flags & IN6P_HOPLIMIT) {
3014 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3017 if (inp_flags & IN6P_HOPOPTS) {
3018 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3021 if (inp_flags & IN6P_DSTOPTS) {
3022 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3025 if (inp_flags & IN6P_RTHDR) {
3026 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3029 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3030 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3033 if (inp_flags & IN6P_TCLASS) {
3034 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3037 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3038 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3041 if (inp_flags & INP_TIMEWAIT) {
3042 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3045 if (inp_flags & INP_ONESBCAST) {
3046 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3049 if (inp_flags & INP_DROPPED) {
3050 db_printf("%sINP_DROPPED", comma ? ", " : "");
3053 if (inp_flags & INP_SOCKREF) {
3054 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3057 if (inp_flags & IN6P_RFC2292) {
3058 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3061 if (inp_flags & IN6P_MTU) {
3062 db_printf("IN6P_MTU%s", comma ? ", " : "");
3068 db_print_inpvflag(u_char inp_vflag)
3073 if (inp_vflag & INP_IPV4) {
3074 db_printf("%sINP_IPV4", comma ? ", " : "");
3077 if (inp_vflag & INP_IPV6) {
3078 db_printf("%sINP_IPV6", comma ? ", " : "");
3081 if (inp_vflag & INP_IPV6PROTO) {
3082 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3088 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3091 db_print_indent(indent);
3092 db_printf("%s at %p\n", name, inp);
3096 db_print_indent(indent);
3097 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3099 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3101 db_print_indent(indent);
3102 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3103 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3105 db_print_indent(indent);
3106 db_printf("inp_label: %p inp_flags: 0x%x (",
3107 inp->inp_label, inp->inp_flags);
3108 db_print_inpflags(inp->inp_flags);
3111 db_print_indent(indent);
3112 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3114 db_print_inpvflag(inp->inp_vflag);
3117 db_print_indent(indent);
3118 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3119 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3121 db_print_indent(indent);
3123 if (inp->inp_vflag & INP_IPV6) {
3124 db_printf("in6p_options: %p in6p_outputopts: %p "
3125 "in6p_moptions: %p\n", inp->in6p_options,
3126 inp->in6p_outputopts, inp->in6p_moptions);
3127 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3128 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3133 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3134 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3135 inp->inp_options, inp->inp_moptions);
3138 db_print_indent(indent);
3139 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3140 (uintmax_t)inp->inp_gencnt);
3143 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3148 db_printf("usage: show inpcb <addr>\n");
3151 inp = (struct inpcb *)addr;
3153 db_print_inpcb(inp, "inpcb", 0);
3159 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3163 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3165 union if_snd_tag_modify_params params = {
3166 .rate_limit.max_rate = max_pacing_rate,
3168 struct m_snd_tag *mst;
3172 mst = inp->inp_snd_tag;
3180 if (ifp->if_snd_tag_modify == NULL) {
3183 error = ifp->if_snd_tag_modify(mst, ¶ms);
3189 * Query existing TX rate limit based on the existing
3190 * "inp->inp_snd_tag", if any.
3193 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3195 union if_snd_tag_query_params params = { };
3196 struct m_snd_tag *mst;
3200 mst = inp->inp_snd_tag;
3208 if (ifp->if_snd_tag_query == NULL) {
3211 error = ifp->if_snd_tag_query(mst, ¶ms);
3212 if (error == 0 && p_max_pacing_rate != NULL)
3213 *p_max_pacing_rate = params.rate_limit.max_rate;
3219 * Query existing TX queue level based on the existing
3220 * "inp->inp_snd_tag", if any.
3223 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3225 union if_snd_tag_query_params params = { };
3226 struct m_snd_tag *mst;
3230 mst = inp->inp_snd_tag;
3238 if (ifp->if_snd_tag_query == NULL)
3239 return (EOPNOTSUPP);
3241 error = ifp->if_snd_tag_query(mst, ¶ms);
3242 if (error == 0 && p_txqueue_level != NULL)
3243 *p_txqueue_level = params.rate_limit.queue_level;
3248 * Allocate a new TX rate limit send tag from the network interface
3249 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3252 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3253 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3255 union if_snd_tag_alloc_params params = {
3256 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3257 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3258 .rate_limit.hdr.flowid = flowid,
3259 .rate_limit.hdr.flowtype = flowtype,
3260 .rate_limit.max_rate = max_pacing_rate,
3264 INP_WLOCK_ASSERT(inp);
3266 if (inp->inp_snd_tag != NULL)
3269 if (ifp->if_snd_tag_alloc == NULL) {
3272 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3275 * At success increment the refcount on
3276 * the send tag's network interface:
3279 if_ref(inp->inp_snd_tag->ifp);
3285 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3289 in_pcbdetach_txrtlmt(struct inpcb *inp)
3291 struct m_snd_tag *mst;
3294 INP_WLOCK_ASSERT(inp);
3296 mst = inp->inp_snd_tag;
3297 inp->inp_snd_tag = NULL;
3307 * If the device was detached while we still had reference(s)
3308 * on the ifp, we assume if_snd_tag_free() was replaced with
3311 ifp->if_snd_tag_free(mst);
3313 /* release reference count on network interface */
3318 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3319 * is set in the fast path and will attach/detach/modify the TX rate
3320 * limit send tag based on the socket's so_max_pacing_rate value.
3323 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3325 struct socket *socket;
3326 uint32_t max_pacing_rate;
3333 socket = inp->inp_socket;
3337 if (!INP_WLOCKED(inp)) {
3339 * NOTE: If the write locking fails, we need to bail
3340 * out and use the non-ratelimited ring for the
3341 * transmit until there is a new chance to get the
3344 if (!INP_TRY_UPGRADE(inp))
3352 * NOTE: The so_max_pacing_rate value is read unlocked,
3353 * because atomic updates are not required since the variable
3354 * is checked at every mbuf we send. It is assumed that the
3355 * variable read itself will be atomic.
3357 max_pacing_rate = socket->so_max_pacing_rate;
3360 * NOTE: When attaching to a network interface a reference is
3361 * made to ensure the network interface doesn't go away until
3362 * all ratelimit connections are gone. The network interface
3363 * pointers compared below represent valid network interfaces,
3364 * except when comparing towards NULL.
3366 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3368 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3369 if (inp->inp_snd_tag != NULL)
3370 in_pcbdetach_txrtlmt(inp);
3372 } else if (inp->inp_snd_tag == NULL) {
3374 * In order to utilize packet pacing with RSS, we need
3375 * to wait until there is a valid RSS hash before we
3378 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3381 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3382 mb->m_pkthdr.flowid, max_pacing_rate);
3385 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3387 if (error == 0 || error == EOPNOTSUPP)
3388 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3394 * Track route changes for TX rate limiting.
3397 in_pcboutput_eagain(struct inpcb *inp)
3399 struct socket *socket;
3405 socket = inp->inp_socket;
3409 if (inp->inp_snd_tag == NULL)
3412 if (!INP_WLOCKED(inp)) {
3414 * NOTE: If the write locking fails, we need to bail
3415 * out and use the non-ratelimited ring for the
3416 * transmit until there is a new chance to get the
3419 if (!INP_TRY_UPGRADE(inp))
3426 /* detach rate limiting */
3427 in_pcbdetach_txrtlmt(inp);
3429 /* make sure new mbuf send tag allocation is made */
3430 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3435 #endif /* RATELIMIT */