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>
90 #include <netinet/in_var.h>
91 #include <netinet/in_fib.h>
93 #include <netinet/ip_var.h>
94 #include <netinet/tcp_var.h>
96 #include <netinet/tcp_hpts.h>
98 #include <netinet/udp.h>
99 #include <netinet/udp_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>
106 #include <net/route/nhop.h>
109 #include <netipsec/ipsec_support.h>
111 #include <security/mac/mac_framework.h>
113 #define INPCBLBGROUP_SIZMIN 8
114 #define INPCBLBGROUP_SIZMAX 256
116 static struct callout ipport_tick_callout;
119 * These configure the range of local port addresses assigned to
120 * "unspecified" outgoing connections/packets/whatever.
122 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
123 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
124 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
125 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
126 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
127 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
130 * Reserved ports accessible only to root. There are significant
131 * security considerations that must be accounted for when changing these,
132 * but the security benefits can be great. Please be careful.
134 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
135 VNET_DEFINE(int, ipport_reservedlow);
137 /* Variables dealing with random ephemeral port allocation. */
138 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
140 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
141 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
142 VNET_DEFINE(int, ipport_tcpallocs);
143 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
145 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
147 static void in_pcbremlists(struct inpcb *inp);
149 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
150 struct in_addr faddr, u_int fport_arg,
151 struct in_addr laddr, u_int lport_arg,
152 int lookupflags, struct ifnet *ifp);
154 #define RANGECHK(var, min, max) \
155 if ((var) < (min)) { (var) = (min); } \
156 else if ((var) > (max)) { (var) = (max); }
159 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
163 error = sysctl_handle_int(oidp, arg1, arg2, req);
165 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
166 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
167 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
169 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
170 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
177 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
178 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
183 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
186 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
187 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
190 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
191 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
195 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
197 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
198 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
199 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
201 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
202 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
203 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
205 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
206 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
207 &VNET_NAME(ipport_reservedhigh), 0, "");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
209 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
210 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
211 CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
213 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
214 CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
216 "allocations before switching to a sequental one");
217 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
218 CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(ipport_randomtime), 0,
220 "Minimum time to keep sequental port "
221 "allocation before switching to a random one");
224 counter_u64_t rate_limit_active;
225 counter_u64_t rate_limit_alloc_fail;
226 counter_u64_t rate_limit_set_ok;
228 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
230 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
231 &rate_limit_active, "Active rate limited connections");
232 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
233 &rate_limit_alloc_fail, "Rate limited connection failures");
234 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
235 &rate_limit_set_ok, "Rate limited setting succeeded");
236 #endif /* RATELIMIT */
241 * in_pcb.c: manage the Protocol Control Blocks.
243 * NOTE: It is assumed that most of these functions will be called with
244 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
245 * functions often modify hash chains or addresses in pcbs.
248 static struct inpcblbgroup *
249 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
250 uint16_t port, const union in_dependaddr *addr, int size)
252 struct inpcblbgroup *grp;
255 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
256 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
259 grp->il_vflag = vflag;
260 grp->il_lport = port;
261 grp->il_dependladdr = *addr;
262 grp->il_inpsiz = size;
263 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
268 in_pcblbgroup_free_deferred(epoch_context_t ctx)
270 struct inpcblbgroup *grp;
272 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
277 in_pcblbgroup_free(struct inpcblbgroup *grp)
280 CK_LIST_REMOVE(grp, il_list);
281 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
284 static struct inpcblbgroup *
285 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
286 struct inpcblbgroup *old_grp, int size)
288 struct inpcblbgroup *grp;
291 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
292 old_grp->il_lport, &old_grp->il_dependladdr, size);
296 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
297 ("invalid new local group size %d and old local group count %d",
298 grp->il_inpsiz, old_grp->il_inpcnt));
300 for (i = 0; i < old_grp->il_inpcnt; ++i)
301 grp->il_inp[i] = old_grp->il_inp[i];
302 grp->il_inpcnt = old_grp->il_inpcnt;
303 in_pcblbgroup_free(old_grp);
308 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
309 * and shrink group if possible.
312 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
315 struct inpcblbgroup *grp, *new_grp;
318 for (; i + 1 < grp->il_inpcnt; ++i)
319 grp->il_inp[i] = grp->il_inp[i + 1];
322 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
323 grp->il_inpcnt <= grp->il_inpsiz / 4) {
324 /* Shrink this group. */
325 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
332 * Add PCB to load balance group for SO_REUSEPORT_LB option.
335 in_pcbinslbgrouphash(struct inpcb *inp)
337 const static struct timeval interval = { 60, 0 };
338 static struct timeval lastprint;
339 struct inpcbinfo *pcbinfo;
340 struct inpcblbgrouphead *hdr;
341 struct inpcblbgroup *grp;
344 pcbinfo = inp->inp_pcbinfo;
346 INP_WLOCK_ASSERT(inp);
347 INP_HASH_WLOCK_ASSERT(pcbinfo);
350 * Don't allow jailed socket to join local group.
352 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
357 * Don't allow IPv4 mapped INET6 wild socket.
359 if ((inp->inp_vflag & INP_IPV4) &&
360 inp->inp_laddr.s_addr == INADDR_ANY &&
361 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
366 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
367 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
368 CK_LIST_FOREACH(grp, hdr, il_list) {
369 if (grp->il_vflag == inp->inp_vflag &&
370 grp->il_lport == inp->inp_lport &&
371 memcmp(&grp->il_dependladdr,
372 &inp->inp_inc.inc_ie.ie_dependladdr,
373 sizeof(grp->il_dependladdr)) == 0)
377 /* Create new load balance group. */
378 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
379 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
380 INPCBLBGROUP_SIZMIN);
383 } else if (grp->il_inpcnt == grp->il_inpsiz) {
384 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
385 if (ratecheck(&lastprint, &interval))
386 printf("lb group port %d, limit reached\n",
387 ntohs(grp->il_lport));
391 /* Expand this local group. */
392 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
397 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
398 ("invalid local group size %d and count %d", grp->il_inpsiz,
401 grp->il_inp[grp->il_inpcnt] = inp;
407 * Remove PCB from load balance group.
410 in_pcbremlbgrouphash(struct inpcb *inp)
412 struct inpcbinfo *pcbinfo;
413 struct inpcblbgrouphead *hdr;
414 struct inpcblbgroup *grp;
417 pcbinfo = inp->inp_pcbinfo;
419 INP_WLOCK_ASSERT(inp);
420 INP_HASH_WLOCK_ASSERT(pcbinfo);
422 hdr = &pcbinfo->ipi_lbgrouphashbase[
423 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
424 CK_LIST_FOREACH(grp, hdr, il_list) {
425 for (i = 0; i < grp->il_inpcnt; ++i) {
426 if (grp->il_inp[i] != inp)
429 if (grp->il_inpcnt == 1) {
430 /* We are the last, free this local group. */
431 in_pcblbgroup_free(grp);
433 /* Pull up inpcbs, shrink group if possible. */
434 in_pcblbgroup_reorder(hdr, &grp, i);
442 * Different protocols initialize their inpcbs differently - giving
443 * different name to the lock. But they all are disposed the same.
446 inpcb_fini(void *mem, int size)
448 struct inpcb *inp = mem;
450 INP_LOCK_DESTROY(inp);
454 * Initialize an inpcbinfo -- we should be able to reduce the number of
458 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
459 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
460 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
463 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
465 INP_INFO_LOCK_INIT(pcbinfo, name);
466 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
467 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
469 pcbinfo->ipi_vnet = curvnet;
471 pcbinfo->ipi_listhead = listhead;
472 CK_LIST_INIT(pcbinfo->ipi_listhead);
473 pcbinfo->ipi_count = 0;
474 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
475 &pcbinfo->ipi_hashmask);
476 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
477 &pcbinfo->ipi_porthashmask);
478 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
479 &pcbinfo->ipi_lbgrouphashmask);
481 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
483 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
484 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
485 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
486 uma_zone_set_warning(pcbinfo->ipi_zone,
487 "kern.ipc.maxsockets limit reached");
491 * Destroy an inpcbinfo.
494 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
497 KASSERT(pcbinfo->ipi_count == 0,
498 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
500 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
501 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
502 pcbinfo->ipi_porthashmask);
503 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
504 pcbinfo->ipi_lbgrouphashmask);
506 in_pcbgroup_destroy(pcbinfo);
508 uma_zdestroy(pcbinfo->ipi_zone);
509 INP_LIST_LOCK_DESTROY(pcbinfo);
510 INP_HASH_LOCK_DESTROY(pcbinfo);
511 INP_INFO_LOCK_DESTROY(pcbinfo);
515 * Allocate a PCB and associate it with the socket.
516 * On success return with the PCB locked.
519 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
525 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
528 bzero(&inp->inp_start_zero, inp_zero_size);
530 inp->inp_numa_domain = M_NODOM;
532 inp->inp_pcbinfo = pcbinfo;
533 inp->inp_socket = so;
534 inp->inp_cred = crhold(so->so_cred);
535 inp->inp_inc.inc_fibnum = so->so_fibnum;
537 error = mac_inpcb_init(inp, M_NOWAIT);
540 mac_inpcb_create(so, inp);
542 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
543 error = ipsec_init_pcbpolicy(inp);
546 mac_inpcb_destroy(inp);
552 if (INP_SOCKAF(so) == AF_INET6) {
553 inp->inp_vflag |= INP_IPV6PROTO;
555 inp->inp_flags |= IN6P_IPV6_V6ONLY;
559 INP_LIST_WLOCK(pcbinfo);
560 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
561 pcbinfo->ipi_count++;
562 so->so_pcb = (caddr_t)inp;
564 if (V_ip6_auto_flowlabel)
565 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
567 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
568 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
571 * Routes in inpcb's can cache L2 as well; they are guaranteed
574 inp->inp_route.ro_flags = RT_LLE_CACHE;
575 INP_LIST_WUNLOCK(pcbinfo);
576 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
579 crfree(inp->inp_cred);
580 uma_zfree(pcbinfo->ipi_zone, inp);
588 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
592 INP_WLOCK_ASSERT(inp);
593 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
595 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
597 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
598 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
599 &inp->inp_lport, cred);
602 if (in_pcbinshash(inp) != 0) {
603 inp->inp_laddr.s_addr = INADDR_ANY;
608 inp->inp_flags |= INP_ANONPORT;
614 * Select a local port (number) to use.
616 #if defined(INET) || defined(INET6)
618 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
619 struct ucred *cred, int lookupflags)
621 struct inpcbinfo *pcbinfo;
622 struct inpcb *tmpinp;
623 unsigned short *lastport;
624 int count, dorandom, error;
625 u_short aux, first, last, lport;
627 struct in_addr laddr;
630 pcbinfo = inp->inp_pcbinfo;
633 * Because no actual state changes occur here, a global write lock on
634 * the pcbinfo isn't required.
636 INP_LOCK_ASSERT(inp);
637 INP_HASH_LOCK_ASSERT(pcbinfo);
639 if (inp->inp_flags & INP_HIGHPORT) {
640 first = V_ipport_hifirstauto; /* sysctl */
641 last = V_ipport_hilastauto;
642 lastport = &pcbinfo->ipi_lasthi;
643 } else if (inp->inp_flags & INP_LOWPORT) {
644 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
647 first = V_ipport_lowfirstauto; /* 1023 */
648 last = V_ipport_lowlastauto; /* 600 */
649 lastport = &pcbinfo->ipi_lastlow;
651 first = V_ipport_firstauto; /* sysctl */
652 last = V_ipport_lastauto;
653 lastport = &pcbinfo->ipi_lastport;
656 * For UDP(-Lite), use random port allocation as long as the user
657 * allows it. For TCP (and as of yet unknown) connections,
658 * use random port allocation only if the user allows it AND
659 * ipport_tick() allows it.
661 if (V_ipport_randomized &&
662 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
663 pcbinfo == &V_ulitecbinfo))
668 * It makes no sense to do random port allocation if
669 * we have the only port available.
673 /* Make sure to not include UDP(-Lite) packets in the count. */
674 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
675 V_ipport_tcpallocs++;
677 * Instead of having two loops further down counting up or down
678 * make sure that first is always <= last and go with only one
679 * code path implementing all logic.
688 /* Make the compiler happy. */
690 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
691 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
696 tmpinp = NULL; /* Make compiler happy. */
700 *lastport = first + (arc4random() % (last - first));
702 count = last - first;
705 if (count-- < 0) /* completely used? */
706 return (EADDRNOTAVAIL);
708 if (*lastport < first || *lastport > last)
710 lport = htons(*lastport);
713 if ((inp->inp_vflag & INP_IPV6) != 0)
714 tmpinp = in6_pcblookup_local(pcbinfo,
715 &inp->in6p_laddr, lport, lookupflags, cred);
717 #if defined(INET) && defined(INET6)
721 tmpinp = in_pcblookup_local(pcbinfo, laddr,
722 lport, lookupflags, cred);
724 } while (tmpinp != NULL);
727 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
728 laddrp->s_addr = laddr.s_addr;
736 * Return cached socket options.
739 inp_so_options(const struct inpcb *inp)
745 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
746 so_options |= SO_REUSEPORT_LB;
747 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
748 so_options |= SO_REUSEPORT;
749 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
750 so_options |= SO_REUSEADDR;
753 #endif /* INET || INET6 */
756 * Check if a new BINDMULTI socket is allowed to be created.
758 * ni points to the new inp.
759 * oi points to the exisitng inp.
761 * This checks whether the existing inp also has BINDMULTI and
762 * whether the credentials match.
765 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
767 /* Check permissions match */
768 if ((ni->inp_flags2 & INP_BINDMULTI) &&
769 (ni->inp_cred->cr_uid !=
770 oi->inp_cred->cr_uid))
773 /* Check the existing inp has BINDMULTI set */
774 if ((ni->inp_flags2 & INP_BINDMULTI) &&
775 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
779 * We're okay - either INP_BINDMULTI isn't set on ni, or
780 * it is and it matches the checks.
787 * Set up a bind operation on a PCB, performing port allocation
788 * as required, but do not actually modify the PCB. Callers can
789 * either complete the bind by setting inp_laddr/inp_lport and
790 * calling in_pcbinshash(), or they can just use the resulting
791 * port and address to authorise the sending of a once-off packet.
793 * On error, the values of *laddrp and *lportp are not changed.
796 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
797 u_short *lportp, struct ucred *cred)
799 struct socket *so = inp->inp_socket;
800 struct sockaddr_in *sin;
801 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
802 struct in_addr laddr;
804 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
808 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
809 * so that we don't have to add to the (already messy) code below.
811 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
814 * No state changes, so read locks are sufficient here.
816 INP_LOCK_ASSERT(inp);
817 INP_HASH_LOCK_ASSERT(pcbinfo);
819 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
820 return (EADDRNOTAVAIL);
821 laddr.s_addr = *laddrp;
822 if (nam != NULL && laddr.s_addr != INADDR_ANY)
824 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
825 lookupflags = INPLOOKUP_WILDCARD;
827 if ((error = prison_local_ip4(cred, &laddr)) != 0)
830 sin = (struct sockaddr_in *)nam;
831 if (nam->sa_len != sizeof (*sin))
835 * We should check the family, but old programs
836 * incorrectly fail to initialize it.
838 if (sin->sin_family != AF_INET)
839 return (EAFNOSUPPORT);
841 error = prison_local_ip4(cred, &sin->sin_addr);
844 if (sin->sin_port != *lportp) {
845 /* Don't allow the port to change. */
848 lport = sin->sin_port;
850 /* NB: lport is left as 0 if the port isn't being changed. */
851 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
853 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
854 * allow complete duplication of binding if
855 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
856 * and a multicast address is bound on both
857 * new and duplicated sockets.
859 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
860 reuseport = SO_REUSEADDR|SO_REUSEPORT;
862 * XXX: How to deal with SO_REUSEPORT_LB here?
863 * Treat same as SO_REUSEPORT for now.
865 if ((so->so_options &
866 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
867 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
868 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
869 sin->sin_port = 0; /* yech... */
870 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
872 * Is the address a local IP address?
873 * If INP_BINDANY is set, then the socket may be bound
874 * to any endpoint address, local or not.
876 if ((inp->inp_flags & INP_BINDANY) == 0 &&
877 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
878 return (EADDRNOTAVAIL);
880 laddr = sin->sin_addr;
886 if (ntohs(lport) <= V_ipport_reservedhigh &&
887 ntohs(lport) >= V_ipport_reservedlow &&
888 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
890 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
891 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
892 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
893 lport, INPLOOKUP_WILDCARD, cred);
896 * This entire block sorely needs a rewrite.
899 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
900 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
901 (so->so_type != SOCK_STREAM ||
902 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
903 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
904 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
905 (t->inp_flags2 & INP_REUSEPORT) ||
906 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
907 (inp->inp_cred->cr_uid !=
908 t->inp_cred->cr_uid))
912 * If the socket is a BINDMULTI socket, then
913 * the credentials need to match and the
914 * original socket also has to have been bound
917 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
920 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
921 lport, lookupflags, cred);
922 if (t && (t->inp_flags & INP_TIMEWAIT)) {
924 * XXXRW: If an incpb has had its timewait
925 * state recycled, we treat the address as
926 * being in use (for now). This is better
927 * than a panic, but not desirable.
931 ((reuseport & tw->tw_so_options) == 0 &&
933 tw->tw_so_options) == 0)) {
937 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
938 (reuseport & inp_so_options(t)) == 0 &&
939 (reuseport_lb & inp_so_options(t)) == 0) {
941 if (ntohl(sin->sin_addr.s_addr) !=
943 ntohl(t->inp_laddr.s_addr) !=
945 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
946 (t->inp_vflag & INP_IPV6PROTO) == 0)
949 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
957 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
962 *laddrp = laddr.s_addr;
968 * Connect from a socket to a specified address.
969 * Both address and port must be specified in argument sin.
970 * If don't have a local address for this socket yet,
974 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
975 struct ucred *cred, struct mbuf *m, bool rehash)
977 u_short lport, fport;
978 in_addr_t laddr, faddr;
981 INP_WLOCK_ASSERT(inp);
982 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
984 lport = inp->inp_lport;
985 laddr = inp->inp_laddr.s_addr;
986 anonport = (lport == 0);
987 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
992 /* Do the initial binding of the local address if required. */
993 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
994 KASSERT(rehash == true,
995 ("Rehashing required for unbound inps"));
996 inp->inp_lport = lport;
997 inp->inp_laddr.s_addr = laddr;
998 if (in_pcbinshash(inp) != 0) {
999 inp->inp_laddr.s_addr = INADDR_ANY;
1005 /* Commit the remaining changes. */
1006 inp->inp_lport = lport;
1007 inp->inp_laddr.s_addr = laddr;
1008 inp->inp_faddr.s_addr = faddr;
1009 inp->inp_fport = fport;
1011 in_pcbrehash_mbuf(inp, m);
1013 in_pcbinshash_mbuf(inp, m);
1017 inp->inp_flags |= INP_ANONPORT;
1022 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1025 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1029 * Do proper source address selection on an unbound socket in case
1030 * of connect. Take jails into account as well.
1033 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1037 struct sockaddr *sa;
1038 struct sockaddr_in *sin, dst;
1039 struct nhop_object *nh;
1043 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1045 * Bypass source address selection and use the primary jail IP
1048 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1054 bzero(&dst, sizeof(dst));
1056 sin->sin_family = AF_INET;
1057 sin->sin_len = sizeof(struct sockaddr_in);
1058 sin->sin_addr.s_addr = faddr->s_addr;
1061 * If route is known our src addr is taken from the i/f,
1064 * Find out route to destination.
1066 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1067 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1071 * If we found a route, use the address corresponding to
1072 * the outgoing interface.
1074 * Otherwise assume faddr is reachable on a directly connected
1075 * network and try to find a corresponding interface to take
1076 * the source address from.
1078 if (nh == NULL || nh->nh_ifp == NULL) {
1079 struct in_ifaddr *ia;
1082 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1083 inp->inp_socket->so_fibnum));
1085 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1086 inp->inp_socket->so_fibnum));
1090 error = ENETUNREACH;
1094 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1095 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1101 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1104 if (sa->sa_family != AF_INET)
1106 sin = (struct sockaddr_in *)sa;
1107 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1108 ia = (struct in_ifaddr *)ifa;
1113 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1117 /* 3. As a last resort return the 'default' jail address. */
1118 error = prison_get_ip4(cred, laddr);
1123 * If the outgoing interface on the route found is not
1124 * a loopback interface, use the address from that interface.
1125 * In case of jails do those three steps:
1126 * 1. check if the interface address belongs to the jail. If so use it.
1127 * 2. check if we have any address on the outgoing interface
1128 * belonging to this jail. If so use it.
1129 * 3. as a last resort return the 'default' jail address.
1131 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1132 struct in_ifaddr *ia;
1135 /* If not jailed, use the default returned. */
1136 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1137 ia = (struct in_ifaddr *)nh->nh_ifa;
1138 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1143 /* 1. Check if the iface address belongs to the jail. */
1144 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1145 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1146 ia = (struct in_ifaddr *)nh->nh_ifa;
1147 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1152 * 2. Check if we have any address on the outgoing interface
1153 * belonging to this jail.
1157 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1159 if (sa->sa_family != AF_INET)
1161 sin = (struct sockaddr_in *)sa;
1162 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1163 ia = (struct in_ifaddr *)ifa;
1168 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1172 /* 3. As a last resort return the 'default' jail address. */
1173 error = prison_get_ip4(cred, laddr);
1178 * The outgoing interface is marked with 'loopback net', so a route
1179 * to ourselves is here.
1180 * Try to find the interface of the destination address and then
1181 * take the address from there. That interface is not necessarily
1182 * a loopback interface.
1183 * In case of jails, check that it is an address of the jail
1184 * and if we cannot find, fall back to the 'default' jail address.
1186 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1187 struct in_ifaddr *ia;
1189 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1190 inp->inp_socket->so_fibnum));
1192 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1193 inp->inp_socket->so_fibnum));
1195 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1197 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1199 error = ENETUNREACH;
1202 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1212 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1214 if (sa->sa_family != AF_INET)
1216 sin = (struct sockaddr_in *)sa;
1217 if (prison_check_ip4(cred,
1218 &sin->sin_addr) == 0) {
1219 ia = (struct in_ifaddr *)ifa;
1224 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1229 /* 3. As a last resort return the 'default' jail address. */
1230 error = prison_get_ip4(cred, laddr);
1239 * Set up for a connect from a socket to the specified address.
1240 * On entry, *laddrp and *lportp should contain the current local
1241 * address and port for the PCB; these are updated to the values
1242 * that should be placed in inp_laddr and inp_lport to complete
1245 * On success, *faddrp and *fportp will be set to the remote address
1246 * and port. These are not updated in the error case.
1248 * If the operation fails because the connection already exists,
1249 * *oinpp will be set to the PCB of that connection so that the
1250 * caller can decide to override it. In all other cases, *oinpp
1254 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1255 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1256 struct inpcb **oinpp, struct ucred *cred)
1258 struct rm_priotracker in_ifa_tracker;
1259 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1260 struct in_ifaddr *ia;
1262 struct in_addr laddr, faddr;
1263 u_short lport, fport;
1267 * Because a global state change doesn't actually occur here, a read
1268 * lock is sufficient.
1271 INP_LOCK_ASSERT(inp);
1272 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1276 if (nam->sa_len != sizeof (*sin))
1278 if (sin->sin_family != AF_INET)
1279 return (EAFNOSUPPORT);
1280 if (sin->sin_port == 0)
1281 return (EADDRNOTAVAIL);
1282 laddr.s_addr = *laddrp;
1284 faddr = sin->sin_addr;
1285 fport = sin->sin_port;
1287 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1289 * If the destination address is INADDR_ANY,
1290 * use the primary local address.
1291 * If the supplied address is INADDR_BROADCAST,
1292 * and the primary interface supports broadcast,
1293 * choose the broadcast address for that interface.
1295 if (faddr.s_addr == INADDR_ANY) {
1296 IN_IFADDR_RLOCK(&in_ifa_tracker);
1298 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1299 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1301 (error = prison_get_ip4(cred, &faddr)) != 0)
1303 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1304 IN_IFADDR_RLOCK(&in_ifa_tracker);
1305 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1307 faddr = satosin(&CK_STAILQ_FIRST(
1308 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1309 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1312 if (laddr.s_addr == INADDR_ANY) {
1313 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1315 * If the destination address is multicast and an outgoing
1316 * interface has been set as a multicast option, prefer the
1317 * address of that interface as our source address.
1319 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1320 inp->inp_moptions != NULL) {
1321 struct ip_moptions *imo;
1324 imo = inp->inp_moptions;
1325 if (imo->imo_multicast_ifp != NULL) {
1326 ifp = imo->imo_multicast_ifp;
1327 IN_IFADDR_RLOCK(&in_ifa_tracker);
1328 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1329 if ((ia->ia_ifp == ifp) &&
1331 prison_check_ip4(cred,
1332 &ia->ia_addr.sin_addr) == 0))
1336 error = EADDRNOTAVAIL;
1338 laddr = ia->ia_addr.sin_addr;
1341 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1347 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1348 laddr, lport, 0, NULL);
1352 return (EADDRINUSE);
1355 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1360 *laddrp = laddr.s_addr;
1362 *faddrp = faddr.s_addr;
1368 in_pcbdisconnect(struct inpcb *inp)
1371 INP_WLOCK_ASSERT(inp);
1372 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1374 inp->inp_faddr.s_addr = INADDR_ANY;
1381 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1382 * For most protocols, this will be invoked immediately prior to calling
1383 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1384 * socket, in which case in_pcbfree() is deferred.
1387 in_pcbdetach(struct inpcb *inp)
1390 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1393 if (inp->inp_snd_tag != NULL)
1394 in_pcbdetach_txrtlmt(inp);
1396 inp->inp_socket->so_pcb = NULL;
1397 inp->inp_socket = NULL;
1401 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1402 * stability of an inpcb pointer despite the inpcb lock being released. This
1403 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1404 * but where the inpcb lock may already held, or when acquiring a reference
1407 * in_pcbref() should be used only to provide brief memory stability, and
1408 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1409 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1410 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1411 * lock and rele are the *only* safe operations that may be performed on the
1414 * While the inpcb will not be freed, releasing the inpcb lock means that the
1415 * connection's state may change, so the caller should be careful to
1416 * revalidate any cached state on reacquiring the lock. Drop the reference
1417 * using in_pcbrele().
1420 in_pcbref(struct inpcb *inp)
1423 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1425 refcount_acquire(&inp->inp_refcount);
1429 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1430 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1431 * return a flag indicating whether or not the inpcb remains valid. If it is
1432 * valid, we return with the inpcb lock held.
1434 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1435 * reference on an inpcb. Historically more work was done here (actually, in
1436 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1437 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1438 * about memory stability (and continued use of the write lock).
1441 in_pcbrele_rlocked(struct inpcb *inp)
1443 struct inpcbinfo *pcbinfo;
1445 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1447 INP_RLOCK_ASSERT(inp);
1449 if (refcount_release(&inp->inp_refcount) == 0) {
1451 * If the inpcb has been freed, let the caller know, even if
1452 * this isn't the last reference.
1454 if (inp->inp_flags2 & INP_FREED) {
1461 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1463 if (inp->inp_in_hpts || inp->inp_in_input) {
1464 struct tcp_hpts_entry *hpts;
1466 * We should not be on the hpts at
1467 * this point in any form. we must
1468 * get the lock to be sure.
1470 hpts = tcp_hpts_lock(inp);
1471 if (inp->inp_in_hpts)
1472 panic("Hpts:%p inp:%p at free still on hpts",
1474 mtx_unlock(&hpts->p_mtx);
1475 hpts = tcp_input_lock(inp);
1476 if (inp->inp_in_input)
1477 panic("Hpts:%p inp:%p at free still on input hpts",
1479 mtx_unlock(&hpts->p_mtx);
1483 pcbinfo = inp->inp_pcbinfo;
1484 uma_zfree(pcbinfo->ipi_zone, inp);
1489 in_pcbrele_wlocked(struct inpcb *inp)
1491 struct inpcbinfo *pcbinfo;
1493 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1495 INP_WLOCK_ASSERT(inp);
1497 if (refcount_release(&inp->inp_refcount) == 0) {
1499 * If the inpcb has been freed, let the caller know, even if
1500 * this isn't the last reference.
1502 if (inp->inp_flags2 & INP_FREED) {
1509 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1511 if (inp->inp_in_hpts || inp->inp_in_input) {
1512 struct tcp_hpts_entry *hpts;
1514 * We should not be on the hpts at
1515 * this point in any form. we must
1516 * get the lock to be sure.
1518 hpts = tcp_hpts_lock(inp);
1519 if (inp->inp_in_hpts)
1520 panic("Hpts:%p inp:%p at free still on hpts",
1522 mtx_unlock(&hpts->p_mtx);
1523 hpts = tcp_input_lock(inp);
1524 if (inp->inp_in_input)
1525 panic("Hpts:%p inp:%p at free still on input hpts",
1527 mtx_unlock(&hpts->p_mtx);
1531 pcbinfo = inp->inp_pcbinfo;
1532 uma_zfree(pcbinfo->ipi_zone, inp);
1537 * Temporary wrapper.
1540 in_pcbrele(struct inpcb *inp)
1543 return (in_pcbrele_wlocked(inp));
1547 in_pcblist_rele_rlocked(epoch_context_t ctx)
1549 struct in_pcblist *il;
1551 struct inpcbinfo *pcbinfo;
1554 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1555 pcbinfo = il->il_pcbinfo;
1557 INP_INFO_WLOCK(pcbinfo);
1558 for (i = 0; i < n; i++) {
1559 inp = il->il_inp_list[i];
1561 if (!in_pcbrele_rlocked(inp))
1564 INP_INFO_WUNLOCK(pcbinfo);
1569 inpcbport_free(epoch_context_t ctx)
1571 struct inpcbport *phd;
1573 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1578 in_pcbfree_deferred(epoch_context_t ctx)
1581 int released __unused;
1583 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1586 CURVNET_SET(inp->inp_vnet);
1588 struct ip_moptions *imo = inp->inp_moptions;
1589 inp->inp_moptions = NULL;
1591 /* XXXRW: Do as much as possible here. */
1592 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1593 if (inp->inp_sp != NULL)
1594 ipsec_delete_pcbpolicy(inp);
1597 struct ip6_moptions *im6o = NULL;
1598 if (inp->inp_vflag & INP_IPV6PROTO) {
1599 ip6_freepcbopts(inp->in6p_outputopts);
1600 im6o = inp->in6p_moptions;
1601 inp->in6p_moptions = NULL;
1604 if (inp->inp_options)
1605 (void)m_free(inp->inp_options);
1607 crfree(inp->inp_cred);
1609 mac_inpcb_destroy(inp);
1611 released = in_pcbrele_wlocked(inp);
1614 ip6_freemoptions(im6o);
1617 inp_freemoptions(imo);
1623 * Unconditionally schedule an inpcb to be freed by decrementing its
1624 * reference count, which should occur only after the inpcb has been detached
1625 * from its socket. If another thread holds a temporary reference (acquired
1626 * using in_pcbref()) then the free is deferred until that reference is
1627 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1628 * work, including removal from global lists, is done in this context, where
1629 * the pcbinfo lock is held.
1632 in_pcbfree(struct inpcb *inp)
1634 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1636 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1637 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1638 ("%s: called twice for pcb %p", __func__, inp));
1639 if (inp->inp_flags2 & INP_FREED) {
1644 INP_WLOCK_ASSERT(inp);
1645 INP_LIST_WLOCK(pcbinfo);
1646 in_pcbremlists(inp);
1647 INP_LIST_WUNLOCK(pcbinfo);
1648 RO_INVALIDATE_CACHE(&inp->inp_route);
1649 /* mark as destruction in progress */
1650 inp->inp_flags2 |= INP_FREED;
1652 NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx);
1656 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1657 * port reservation, and preventing it from being returned by inpcb lookups.
1659 * It is used by TCP to mark an inpcb as unused and avoid future packet
1660 * delivery or event notification when a socket remains open but TCP has
1661 * closed. This might occur as a result of a shutdown()-initiated TCP close
1662 * or a RST on the wire, and allows the port binding to be reused while still
1663 * maintaining the invariant that so_pcb always points to a valid inpcb until
1666 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1667 * in_pcbnotifyall() and in_pcbpurgeif0()?
1670 in_pcbdrop(struct inpcb *inp)
1673 INP_WLOCK_ASSERT(inp);
1675 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1676 MPASS(inp->inp_refcount > 1);
1680 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1683 inp->inp_flags |= INP_DROPPED;
1684 if (inp->inp_flags & INP_INHASHLIST) {
1685 struct inpcbport *phd = inp->inp_phd;
1687 INP_HASH_WLOCK(inp->inp_pcbinfo);
1688 in_pcbremlbgrouphash(inp);
1689 CK_LIST_REMOVE(inp, inp_hash);
1690 CK_LIST_REMOVE(inp, inp_portlist);
1691 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1692 CK_LIST_REMOVE(phd, phd_hash);
1693 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
1695 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1696 inp->inp_flags &= ~INP_INHASHLIST;
1698 in_pcbgroup_remove(inp);
1705 * Common routines to return the socket addresses associated with inpcbs.
1708 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1710 struct sockaddr_in *sin;
1712 sin = malloc(sizeof *sin, M_SONAME,
1714 sin->sin_family = AF_INET;
1715 sin->sin_len = sizeof(*sin);
1716 sin->sin_addr = *addr_p;
1717 sin->sin_port = port;
1719 return (struct sockaddr *)sin;
1723 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1726 struct in_addr addr;
1729 inp = sotoinpcb(so);
1730 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1733 port = inp->inp_lport;
1734 addr = inp->inp_laddr;
1737 *nam = in_sockaddr(port, &addr);
1742 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1745 struct in_addr addr;
1748 inp = sotoinpcb(so);
1749 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1752 port = inp->inp_fport;
1753 addr = inp->inp_faddr;
1756 *nam = in_sockaddr(port, &addr);
1761 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1762 struct inpcb *(*notify)(struct inpcb *, int))
1764 struct inpcb *inp, *inp_temp;
1766 INP_INFO_WLOCK(pcbinfo);
1767 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1770 if ((inp->inp_vflag & INP_IPV4) == 0) {
1775 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1776 inp->inp_socket == NULL) {
1780 if ((*notify)(inp, errno))
1783 INP_INFO_WUNLOCK(pcbinfo);
1787 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1790 struct in_multi *inm;
1791 struct in_mfilter *imf;
1792 struct ip_moptions *imo;
1794 INP_INFO_WLOCK(pcbinfo);
1795 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1797 imo = inp->inp_moptions;
1798 if ((inp->inp_vflag & INP_IPV4) &&
1801 * Unselect the outgoing interface if it is being
1804 if (imo->imo_multicast_ifp == ifp)
1805 imo->imo_multicast_ifp = NULL;
1808 * Drop multicast group membership if we joined
1809 * through the interface being detached.
1811 * XXX This can all be deferred to an epoch_call
1814 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1815 if ((inm = imf->imf_inm) == NULL)
1817 if (inm->inm_ifp != ifp)
1819 ip_mfilter_remove(&imo->imo_head, imf);
1820 IN_MULTI_LOCK_ASSERT();
1821 in_leavegroup_locked(inm, NULL);
1822 ip_mfilter_free(imf);
1828 INP_INFO_WUNLOCK(pcbinfo);
1832 * Lookup a PCB based on the local address and port. Caller must hold the
1833 * hash lock. No inpcb locks or references are acquired.
1835 #define INP_LOOKUP_MAPPED_PCB_COST 3
1837 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1838 u_short lport, int lookupflags, struct ucred *cred)
1842 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1848 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1849 ("%s: invalid lookup flags %d", __func__, lookupflags));
1851 INP_HASH_LOCK_ASSERT(pcbinfo);
1853 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1854 struct inpcbhead *head;
1856 * Look for an unconnected (wildcard foreign addr) PCB that
1857 * matches the local address and port we're looking for.
1859 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1860 0, pcbinfo->ipi_hashmask)];
1861 CK_LIST_FOREACH(inp, head, inp_hash) {
1863 /* XXX inp locking */
1864 if ((inp->inp_vflag & INP_IPV4) == 0)
1867 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1868 inp->inp_laddr.s_addr == laddr.s_addr &&
1869 inp->inp_lport == lport) {
1874 prison_equal_ip4(cred->cr_prison,
1875 inp->inp_cred->cr_prison))
1884 struct inpcbporthead *porthash;
1885 struct inpcbport *phd;
1886 struct inpcb *match = NULL;
1888 * Best fit PCB lookup.
1890 * First see if this local port is in use by looking on the
1893 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1894 pcbinfo->ipi_porthashmask)];
1895 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1896 if (phd->phd_port == lport)
1901 * Port is in use by one or more PCBs. Look for best
1904 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1907 !prison_equal_ip4(inp->inp_cred->cr_prison,
1911 /* XXX inp locking */
1912 if ((inp->inp_vflag & INP_IPV4) == 0)
1915 * We never select the PCB that has
1916 * INP_IPV6 flag and is bound to :: if
1917 * we have another PCB which is bound
1918 * to 0.0.0.0. If a PCB has the
1919 * INP_IPV6 flag, then we set its cost
1920 * higher than IPv4 only PCBs.
1922 * Note that the case only happens
1923 * when a socket is bound to ::, under
1924 * the condition that the use of the
1925 * mapped address is allowed.
1927 if ((inp->inp_vflag & INP_IPV6) != 0)
1928 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1930 if (inp->inp_faddr.s_addr != INADDR_ANY)
1932 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1933 if (laddr.s_addr == INADDR_ANY)
1935 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1938 if (laddr.s_addr != INADDR_ANY)
1941 if (wildcard < matchwild) {
1943 matchwild = wildcard;
1952 #undef INP_LOOKUP_MAPPED_PCB_COST
1954 static struct inpcb *
1955 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1956 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1957 uint16_t fport, int lookupflags)
1959 struct inpcb *local_wild;
1960 const struct inpcblbgrouphead *hdr;
1961 struct inpcblbgroup *grp;
1964 INP_HASH_LOCK_ASSERT(pcbinfo);
1966 hdr = &pcbinfo->ipi_lbgrouphashbase[
1967 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1970 * Order of socket selection:
1972 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1975 * - Load balanced group does not contain jailed sockets
1976 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1979 CK_LIST_FOREACH(grp, hdr, il_list) {
1981 if (!(grp->il_vflag & INP_IPV4))
1984 if (grp->il_lport != lport)
1987 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1989 if (grp->il_laddr.s_addr == laddr->s_addr)
1990 return (grp->il_inp[idx]);
1991 if (grp->il_laddr.s_addr == INADDR_ANY &&
1992 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1993 local_wild = grp->il_inp[idx];
1995 return (local_wild);
2000 * Lookup PCB in hash list, using pcbgroup tables.
2002 static struct inpcb *
2003 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2004 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2005 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2007 struct inpcbhead *head;
2008 struct inpcb *inp, *tmpinp;
2009 u_short fport = fport_arg, lport = lport_arg;
2013 * First look for an exact match.
2016 INP_GROUP_LOCK(pcbgroup);
2017 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2018 pcbgroup->ipg_hashmask)];
2019 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2021 /* XXX inp locking */
2022 if ((inp->inp_vflag & INP_IPV4) == 0)
2025 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2026 inp->inp_laddr.s_addr == laddr.s_addr &&
2027 inp->inp_fport == fport &&
2028 inp->inp_lport == lport) {
2030 * XXX We should be able to directly return
2031 * the inp here, without any checks.
2032 * Well unless both bound with SO_REUSEPORT?
2034 if (prison_flag(inp->inp_cred, PR_IP4))
2040 if (tmpinp != NULL) {
2047 * For incoming connections, we may wish to do a wildcard
2048 * match for an RSS-local socket.
2050 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2051 struct inpcb *local_wild = NULL, *local_exact = NULL;
2053 struct inpcb *local_wild_mapped = NULL;
2055 struct inpcb *jail_wild = NULL;
2056 struct inpcbhead *head;
2060 * Order of socket selection - we always prefer jails.
2061 * 1. jailed, non-wild.
2063 * 3. non-jailed, non-wild.
2064 * 4. non-jailed, wild.
2067 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2068 lport, 0, pcbgroup->ipg_hashmask)];
2069 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2071 /* XXX inp locking */
2072 if ((inp->inp_vflag & INP_IPV4) == 0)
2075 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2076 inp->inp_lport != lport)
2079 injail = prison_flag(inp->inp_cred, PR_IP4);
2081 if (prison_check_ip4(inp->inp_cred,
2085 if (local_exact != NULL)
2089 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2094 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2096 /* XXX inp locking, NULL check */
2097 if (inp->inp_vflag & INP_IPV6PROTO)
2098 local_wild_mapped = inp;
2106 } /* LIST_FOREACH */
2115 inp = local_wild_mapped;
2123 * Then look for a wildcard match, if requested.
2125 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2126 struct inpcb *local_wild = NULL, *local_exact = NULL;
2128 struct inpcb *local_wild_mapped = NULL;
2130 struct inpcb *jail_wild = NULL;
2131 struct inpcbhead *head;
2135 * Order of socket selection - we always prefer jails.
2136 * 1. jailed, non-wild.
2138 * 3. non-jailed, non-wild.
2139 * 4. non-jailed, wild.
2141 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2142 0, pcbinfo->ipi_wildmask)];
2143 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2145 /* XXX inp locking */
2146 if ((inp->inp_vflag & INP_IPV4) == 0)
2149 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2150 inp->inp_lport != lport)
2153 injail = prison_flag(inp->inp_cred, PR_IP4);
2155 if (prison_check_ip4(inp->inp_cred,
2159 if (local_exact != NULL)
2163 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2168 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2170 /* XXX inp locking, NULL check */
2171 if (inp->inp_vflag & INP_IPV6PROTO)
2172 local_wild_mapped = inp;
2180 } /* LIST_FOREACH */
2188 inp = local_wild_mapped;
2192 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2193 INP_GROUP_UNLOCK(pcbgroup);
2197 if (lookupflags & INPLOOKUP_WLOCKPCB)
2198 locked = INP_TRY_WLOCK(inp);
2199 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2200 locked = INP_TRY_RLOCK(inp);
2202 panic("%s: locking bug", __func__);
2203 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2204 if (lookupflags & INPLOOKUP_WLOCKPCB)
2211 INP_GROUP_UNLOCK(pcbgroup);
2213 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2215 if (in_pcbrele_wlocked(inp))
2219 if (in_pcbrele_rlocked(inp))
2224 if (lookupflags & INPLOOKUP_WLOCKPCB)
2225 INP_WLOCK_ASSERT(inp);
2227 INP_RLOCK_ASSERT(inp);
2231 #endif /* PCBGROUP */
2234 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2235 * that the caller has locked the hash list, and will not perform any further
2236 * locking or reference operations on either the hash list or the connection.
2238 static struct inpcb *
2239 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2240 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2243 struct inpcbhead *head;
2244 struct inpcb *inp, *tmpinp;
2245 u_short fport = fport_arg, lport = lport_arg;
2247 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2248 ("%s: invalid lookup flags %d", __func__, lookupflags));
2249 INP_HASH_LOCK_ASSERT(pcbinfo);
2252 * First look for an exact match.
2255 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2256 pcbinfo->ipi_hashmask)];
2257 CK_LIST_FOREACH(inp, head, inp_hash) {
2259 /* XXX inp locking */
2260 if ((inp->inp_vflag & INP_IPV4) == 0)
2263 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2264 inp->inp_laddr.s_addr == laddr.s_addr &&
2265 inp->inp_fport == fport &&
2266 inp->inp_lport == lport) {
2268 * XXX We should be able to directly return
2269 * the inp here, without any checks.
2270 * Well unless both bound with SO_REUSEPORT?
2272 if (prison_flag(inp->inp_cred, PR_IP4))
2282 * Then look in lb group (for wildcard match).
2284 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2285 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2286 fport, lookupflags);
2292 * Then look for a wildcard match, if requested.
2294 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2295 struct inpcb *local_wild = NULL, *local_exact = NULL;
2297 struct inpcb *local_wild_mapped = NULL;
2299 struct inpcb *jail_wild = NULL;
2303 * Order of socket selection - we always prefer jails.
2304 * 1. jailed, non-wild.
2306 * 3. non-jailed, non-wild.
2307 * 4. non-jailed, wild.
2310 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2311 0, pcbinfo->ipi_hashmask)];
2312 CK_LIST_FOREACH(inp, head, inp_hash) {
2314 /* XXX inp locking */
2315 if ((inp->inp_vflag & INP_IPV4) == 0)
2318 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2319 inp->inp_lport != lport)
2322 injail = prison_flag(inp->inp_cred, PR_IP4);
2324 if (prison_check_ip4(inp->inp_cred,
2328 if (local_exact != NULL)
2332 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2337 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2339 /* XXX inp locking, NULL check */
2340 if (inp->inp_vflag & INP_IPV6PROTO)
2341 local_wild_mapped = inp;
2349 } /* LIST_FOREACH */
2350 if (jail_wild != NULL)
2352 if (local_exact != NULL)
2353 return (local_exact);
2354 if (local_wild != NULL)
2355 return (local_wild);
2357 if (local_wild_mapped != NULL)
2358 return (local_wild_mapped);
2360 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2366 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2367 * hash list lock, and will return the inpcb locked (i.e., requires
2368 * INPLOOKUP_LOCKPCB).
2370 static struct inpcb *
2371 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2372 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2377 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2378 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2380 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2382 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2386 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2388 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2393 panic("%s: locking bug", __func__);
2396 if (lookupflags & INPLOOKUP_WLOCKPCB)
2397 INP_WLOCK_ASSERT(inp);
2399 INP_RLOCK_ASSERT(inp);
2408 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2409 * from which a pre-calculated hash value may be extracted.
2411 * Possibly more of this logic should be in in_pcbgroup.c.
2414 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2415 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2417 #if defined(PCBGROUP) && !defined(RSS)
2418 struct inpcbgroup *pcbgroup;
2421 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2422 ("%s: invalid lookup flags %d", __func__, lookupflags));
2423 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2424 ("%s: LOCKPCB not set", __func__));
2427 * When not using RSS, use connection groups in preference to the
2428 * reservation table when looking up 4-tuples. When using RSS, just
2429 * use the reservation table, due to the cost of the Toeplitz hash
2432 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2433 * we could be doing RSS with a non-Toeplitz hash that is affordable
2436 #if defined(PCBGROUP) && !defined(RSS)
2437 if (in_pcbgroup_enabled(pcbinfo)) {
2438 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2440 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2441 laddr, lport, lookupflags, ifp));
2444 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2449 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2450 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2451 struct ifnet *ifp, struct mbuf *m)
2454 struct inpcbgroup *pcbgroup;
2457 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2458 ("%s: invalid lookup flags %d", __func__, lookupflags));
2459 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2460 ("%s: LOCKPCB not set", __func__));
2464 * If we can use a hardware-generated hash to look up the connection
2465 * group, use that connection group to find the inpcb. Otherwise
2466 * fall back on a software hash -- or the reservation table if we're
2469 * XXXRW: As above, that policy belongs in the pcbgroup code.
2471 if (in_pcbgroup_enabled(pcbinfo) &&
2472 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2473 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2474 m->m_pkthdr.flowid);
2475 if (pcbgroup != NULL)
2476 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2477 fport, laddr, lport, lookupflags, ifp));
2479 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2481 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2482 laddr, lport, lookupflags, ifp));
2486 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2492 * Insert PCB onto various hash lists.
2495 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2497 struct inpcbhead *pcbhash;
2498 struct inpcbporthead *pcbporthash;
2499 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2500 struct inpcbport *phd;
2501 u_int32_t hashkey_faddr;
2504 INP_WLOCK_ASSERT(inp);
2505 INP_HASH_WLOCK_ASSERT(pcbinfo);
2507 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2508 ("in_pcbinshash: INP_INHASHLIST"));
2511 if (inp->inp_vflag & INP_IPV6)
2512 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2515 hashkey_faddr = inp->inp_faddr.s_addr;
2517 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2518 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2520 pcbporthash = &pcbinfo->ipi_porthashbase[
2521 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2524 * Add entry to load balance group.
2525 * Only do this if SO_REUSEPORT_LB is set.
2527 so_options = inp_so_options(inp);
2528 if (so_options & SO_REUSEPORT_LB) {
2529 int ret = in_pcbinslbgrouphash(inp);
2531 /* pcb lb group malloc fail (ret=ENOBUFS). */
2537 * Go through port list and look for a head for this lport.
2539 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2540 if (phd->phd_port == inp->inp_lport)
2544 * If none exists, malloc one and tack it on.
2547 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2549 return (ENOBUFS); /* XXX */
2551 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2552 phd->phd_port = inp->inp_lport;
2553 CK_LIST_INIT(&phd->phd_pcblist);
2554 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2557 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2558 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2559 inp->inp_flags |= INP_INHASHLIST;
2562 in_pcbgroup_update_mbuf(inp, m);
2564 in_pcbgroup_update(inp);
2571 in_pcbinshash(struct inpcb *inp)
2574 return (in_pcbinshash_internal(inp, NULL));
2578 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2581 return (in_pcbinshash_internal(inp, m));
2585 * Move PCB to the proper hash bucket when { faddr, fport } have been
2586 * changed. NOTE: This does not handle the case of the lport changing (the
2587 * hashed port list would have to be updated as well), so the lport must
2588 * not change after in_pcbinshash() has been called.
2591 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2593 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2594 struct inpcbhead *head;
2595 u_int32_t hashkey_faddr;
2597 INP_WLOCK_ASSERT(inp);
2598 INP_HASH_WLOCK_ASSERT(pcbinfo);
2600 KASSERT(inp->inp_flags & INP_INHASHLIST,
2601 ("in_pcbrehash: !INP_INHASHLIST"));
2604 if (inp->inp_vflag & INP_IPV6)
2605 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2608 hashkey_faddr = inp->inp_faddr.s_addr;
2610 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2611 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2613 CK_LIST_REMOVE(inp, inp_hash);
2614 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2618 in_pcbgroup_update_mbuf(inp, m);
2620 in_pcbgroup_update(inp);
2625 in_pcbrehash(struct inpcb *inp)
2628 in_pcbrehash_mbuf(inp, NULL);
2632 * Remove PCB from various lists.
2635 in_pcbremlists(struct inpcb *inp)
2637 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2639 INP_WLOCK_ASSERT(inp);
2640 INP_LIST_WLOCK_ASSERT(pcbinfo);
2642 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2643 if (inp->inp_flags & INP_INHASHLIST) {
2644 struct inpcbport *phd = inp->inp_phd;
2646 INP_HASH_WLOCK(pcbinfo);
2648 /* XXX: Only do if SO_REUSEPORT_LB set? */
2649 in_pcbremlbgrouphash(inp);
2651 CK_LIST_REMOVE(inp, inp_hash);
2652 CK_LIST_REMOVE(inp, inp_portlist);
2653 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2654 CK_LIST_REMOVE(phd, phd_hash);
2655 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
2657 INP_HASH_WUNLOCK(pcbinfo);
2658 inp->inp_flags &= ~INP_INHASHLIST;
2660 CK_LIST_REMOVE(inp, inp_list);
2661 pcbinfo->ipi_count--;
2663 in_pcbgroup_remove(inp);
2668 * Check for alternatives when higher level complains
2669 * about service problems. For now, invalidate cached
2670 * routing information. If the route was created dynamically
2671 * (by a redirect), time to try a default gateway again.
2674 in_losing(struct inpcb *inp)
2677 RO_INVALIDATE_CACHE(&inp->inp_route);
2682 * A set label operation has occurred at the socket layer, propagate the
2683 * label change into the in_pcb for the socket.
2686 in_pcbsosetlabel(struct socket *so)
2691 inp = sotoinpcb(so);
2692 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2696 mac_inpcb_sosetlabel(so, inp);
2703 * ipport_tick runs once per second, determining if random port allocation
2704 * should be continued. If more than ipport_randomcps ports have been
2705 * allocated in the last second, then we return to sequential port
2706 * allocation. We return to random allocation only once we drop below
2707 * ipport_randomcps for at least ipport_randomtime seconds.
2710 ipport_tick(void *xtp)
2712 VNET_ITERATOR_DECL(vnet_iter);
2714 VNET_LIST_RLOCK_NOSLEEP();
2715 VNET_FOREACH(vnet_iter) {
2716 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2717 if (V_ipport_tcpallocs <=
2718 V_ipport_tcplastcount + V_ipport_randomcps) {
2719 if (V_ipport_stoprandom > 0)
2720 V_ipport_stoprandom--;
2722 V_ipport_stoprandom = V_ipport_randomtime;
2723 V_ipport_tcplastcount = V_ipport_tcpallocs;
2726 VNET_LIST_RUNLOCK_NOSLEEP();
2727 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2734 callout_stop(&ipport_tick_callout);
2738 * The ipport_callout should start running at about the time we attach the
2739 * inet or inet6 domains.
2742 ipport_tick_init(const void *unused __unused)
2745 /* Start ipport_tick. */
2746 callout_init(&ipport_tick_callout, 1);
2747 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2748 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2749 SHUTDOWN_PRI_DEFAULT);
2751 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2752 ipport_tick_init, NULL);
2755 inp_wlock(struct inpcb *inp)
2762 inp_wunlock(struct inpcb *inp)
2769 inp_rlock(struct inpcb *inp)
2776 inp_runlock(struct inpcb *inp)
2782 #ifdef INVARIANT_SUPPORT
2784 inp_lock_assert(struct inpcb *inp)
2787 INP_WLOCK_ASSERT(inp);
2791 inp_unlock_assert(struct inpcb *inp)
2794 INP_UNLOCK_ASSERT(inp);
2799 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2803 INP_INFO_WLOCK(&V_tcbinfo);
2804 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2809 INP_INFO_WUNLOCK(&V_tcbinfo);
2813 inp_inpcbtosocket(struct inpcb *inp)
2816 INP_WLOCK_ASSERT(inp);
2817 return (inp->inp_socket);
2821 inp_inpcbtotcpcb(struct inpcb *inp)
2824 INP_WLOCK_ASSERT(inp);
2825 return ((struct tcpcb *)inp->inp_ppcb);
2829 inp_ip_tos_get(const struct inpcb *inp)
2832 return (inp->inp_ip_tos);
2836 inp_ip_tos_set(struct inpcb *inp, int val)
2839 inp->inp_ip_tos = val;
2843 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2844 uint32_t *faddr, uint16_t *fp)
2847 INP_LOCK_ASSERT(inp);
2848 *laddr = inp->inp_laddr.s_addr;
2849 *faddr = inp->inp_faddr.s_addr;
2850 *lp = inp->inp_lport;
2851 *fp = inp->inp_fport;
2855 so_sotoinpcb(struct socket *so)
2858 return (sotoinpcb(so));
2862 so_sototcpcb(struct socket *so)
2865 return (sototcpcb(so));
2869 * Create an external-format (``xinpcb'') structure using the information in
2870 * the kernel-format in_pcb structure pointed to by inp. This is done to
2871 * reduce the spew of irrelevant information over this interface, to isolate
2872 * user code from changes in the kernel structure, and potentially to provide
2873 * information-hiding if we decide that some of this information should be
2874 * hidden from users.
2877 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2880 bzero(xi, sizeof(*xi));
2881 xi->xi_len = sizeof(struct xinpcb);
2882 if (inp->inp_socket)
2883 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2884 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2885 xi->inp_gencnt = inp->inp_gencnt;
2886 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2887 xi->inp_flow = inp->inp_flow;
2888 xi->inp_flowid = inp->inp_flowid;
2889 xi->inp_flowtype = inp->inp_flowtype;
2890 xi->inp_flags = inp->inp_flags;
2891 xi->inp_flags2 = inp->inp_flags2;
2892 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2893 xi->in6p_cksum = inp->in6p_cksum;
2894 xi->in6p_hops = inp->in6p_hops;
2895 xi->inp_ip_tos = inp->inp_ip_tos;
2896 xi->inp_vflag = inp->inp_vflag;
2897 xi->inp_ip_ttl = inp->inp_ip_ttl;
2898 xi->inp_ip_p = inp->inp_ip_p;
2899 xi->inp_ip_minttl = inp->inp_ip_minttl;
2904 db_print_indent(int indent)
2908 for (i = 0; i < indent; i++)
2913 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2915 char faddr_str[48], laddr_str[48];
2917 db_print_indent(indent);
2918 db_printf("%s at %p\n", name, inc);
2923 if (inc->inc_flags & INC_ISIPV6) {
2925 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2926 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2931 inet_ntoa_r(inc->inc_laddr, laddr_str);
2932 inet_ntoa_r(inc->inc_faddr, faddr_str);
2934 db_print_indent(indent);
2935 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2936 ntohs(inc->inc_lport));
2937 db_print_indent(indent);
2938 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2939 ntohs(inc->inc_fport));
2943 db_print_inpflags(int inp_flags)
2948 if (inp_flags & INP_RECVOPTS) {
2949 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2952 if (inp_flags & INP_RECVRETOPTS) {
2953 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2956 if (inp_flags & INP_RECVDSTADDR) {
2957 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2960 if (inp_flags & INP_ORIGDSTADDR) {
2961 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2964 if (inp_flags & INP_HDRINCL) {
2965 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2968 if (inp_flags & INP_HIGHPORT) {
2969 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2972 if (inp_flags & INP_LOWPORT) {
2973 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2976 if (inp_flags & INP_ANONPORT) {
2977 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2980 if (inp_flags & INP_RECVIF) {
2981 db_printf("%sINP_RECVIF", comma ? ", " : "");
2984 if (inp_flags & INP_MTUDISC) {
2985 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2988 if (inp_flags & INP_RECVTTL) {
2989 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2992 if (inp_flags & INP_DONTFRAG) {
2993 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2996 if (inp_flags & INP_RECVTOS) {
2997 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3000 if (inp_flags & IN6P_IPV6_V6ONLY) {
3001 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3004 if (inp_flags & IN6P_PKTINFO) {
3005 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3008 if (inp_flags & IN6P_HOPLIMIT) {
3009 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3012 if (inp_flags & IN6P_HOPOPTS) {
3013 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3016 if (inp_flags & IN6P_DSTOPTS) {
3017 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3020 if (inp_flags & IN6P_RTHDR) {
3021 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3024 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3025 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3028 if (inp_flags & IN6P_TCLASS) {
3029 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3032 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3033 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3036 if (inp_flags & INP_TIMEWAIT) {
3037 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3040 if (inp_flags & INP_ONESBCAST) {
3041 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3044 if (inp_flags & INP_DROPPED) {
3045 db_printf("%sINP_DROPPED", comma ? ", " : "");
3048 if (inp_flags & INP_SOCKREF) {
3049 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3052 if (inp_flags & IN6P_RFC2292) {
3053 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3056 if (inp_flags & IN6P_MTU) {
3057 db_printf("IN6P_MTU%s", comma ? ", " : "");
3063 db_print_inpvflag(u_char inp_vflag)
3068 if (inp_vflag & INP_IPV4) {
3069 db_printf("%sINP_IPV4", comma ? ", " : "");
3072 if (inp_vflag & INP_IPV6) {
3073 db_printf("%sINP_IPV6", comma ? ", " : "");
3076 if (inp_vflag & INP_IPV6PROTO) {
3077 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3083 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3086 db_print_indent(indent);
3087 db_printf("%s at %p\n", name, inp);
3091 db_print_indent(indent);
3092 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3094 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3096 db_print_indent(indent);
3097 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3098 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3100 db_print_indent(indent);
3101 db_printf("inp_label: %p inp_flags: 0x%x (",
3102 inp->inp_label, inp->inp_flags);
3103 db_print_inpflags(inp->inp_flags);
3106 db_print_indent(indent);
3107 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3109 db_print_inpvflag(inp->inp_vflag);
3112 db_print_indent(indent);
3113 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3114 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3116 db_print_indent(indent);
3118 if (inp->inp_vflag & INP_IPV6) {
3119 db_printf("in6p_options: %p in6p_outputopts: %p "
3120 "in6p_moptions: %p\n", inp->in6p_options,
3121 inp->in6p_outputopts, inp->in6p_moptions);
3122 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3123 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3128 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3129 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3130 inp->inp_options, inp->inp_moptions);
3133 db_print_indent(indent);
3134 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3135 (uintmax_t)inp->inp_gencnt);
3138 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3143 db_printf("usage: show inpcb <addr>\n");
3146 inp = (struct inpcb *)addr;
3148 db_print_inpcb(inp, "inpcb", 0);
3154 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3158 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3160 union if_snd_tag_modify_params params = {
3161 .rate_limit.max_rate = max_pacing_rate,
3162 .rate_limit.flags = M_NOWAIT,
3164 struct m_snd_tag *mst;
3168 mst = inp->inp_snd_tag;
3176 if (ifp->if_snd_tag_modify == NULL) {
3179 error = ifp->if_snd_tag_modify(mst, ¶ms);
3185 * Query existing TX rate limit based on the existing
3186 * "inp->inp_snd_tag", if any.
3189 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3191 union if_snd_tag_query_params params = { };
3192 struct m_snd_tag *mst;
3196 mst = inp->inp_snd_tag;
3204 if (ifp->if_snd_tag_query == NULL) {
3207 error = ifp->if_snd_tag_query(mst, ¶ms);
3208 if (error == 0 && p_max_pacing_rate != NULL)
3209 *p_max_pacing_rate = params.rate_limit.max_rate;
3215 * Query existing TX queue level based on the existing
3216 * "inp->inp_snd_tag", if any.
3219 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3221 union if_snd_tag_query_params params = { };
3222 struct m_snd_tag *mst;
3226 mst = inp->inp_snd_tag;
3234 if (ifp->if_snd_tag_query == NULL)
3235 return (EOPNOTSUPP);
3237 error = ifp->if_snd_tag_query(mst, ¶ms);
3238 if (error == 0 && p_txqueue_level != NULL)
3239 *p_txqueue_level = params.rate_limit.queue_level;
3244 * Allocate a new TX rate limit send tag from the network interface
3245 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3248 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3249 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3252 union if_snd_tag_alloc_params params = {
3253 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3254 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3255 .rate_limit.hdr.flowid = flowid,
3256 .rate_limit.hdr.flowtype = flowtype,
3257 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3258 .rate_limit.max_rate = max_pacing_rate,
3259 .rate_limit.flags = M_NOWAIT,
3263 INP_WLOCK_ASSERT(inp);
3268 if (ifp->if_snd_tag_alloc == NULL) {
3271 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3275 counter_u64_add(rate_limit_set_ok, 1);
3276 counter_u64_add(rate_limit_active, 1);
3278 counter_u64_add(rate_limit_alloc_fail, 1);
3285 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3291 * If the device was detached while we still had reference(s)
3292 * on the ifp, we assume if_snd_tag_free() was replaced with
3295 ifp->if_snd_tag_free(mst);
3297 /* release reference count on network interface */
3300 counter_u64_add(rate_limit_active, -1);
3305 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3309 in_pcbdetach_txrtlmt(struct inpcb *inp)
3311 struct m_snd_tag *mst;
3313 INP_WLOCK_ASSERT(inp);
3315 mst = inp->inp_snd_tag;
3316 inp->inp_snd_tag = NULL;
3321 m_snd_tag_rele(mst);
3325 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3330 * If the existing send tag is for the wrong interface due to
3331 * a route change, first drop the existing tag. Set the
3332 * CHANGED flag so that we will keep trying to allocate a new
3333 * tag if we fail to allocate one this time.
3335 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3336 in_pcbdetach_txrtlmt(inp);
3337 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3341 * NOTE: When attaching to a network interface a reference is
3342 * made to ensure the network interface doesn't go away until
3343 * all ratelimit connections are gone. The network interface
3344 * pointers compared below represent valid network interfaces,
3345 * except when comparing towards NULL.
3347 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3349 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3350 if (inp->inp_snd_tag != NULL)
3351 in_pcbdetach_txrtlmt(inp);
3353 } else if (inp->inp_snd_tag == NULL) {
3355 * In order to utilize packet pacing with RSS, we need
3356 * to wait until there is a valid RSS hash before we
3359 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3362 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3363 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3366 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3368 if (error == 0 || error == EOPNOTSUPP)
3369 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3375 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3376 * is set in the fast path and will attach/detach/modify the TX rate
3377 * limit send tag based on the socket's so_max_pacing_rate value.
3380 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3382 struct socket *socket;
3383 uint32_t max_pacing_rate;
3390 socket = inp->inp_socket;
3394 if (!INP_WLOCKED(inp)) {
3396 * NOTE: If the write locking fails, we need to bail
3397 * out and use the non-ratelimited ring for the
3398 * transmit until there is a new chance to get the
3401 if (!INP_TRY_UPGRADE(inp))
3409 * NOTE: The so_max_pacing_rate value is read unlocked,
3410 * because atomic updates are not required since the variable
3411 * is checked at every mbuf we send. It is assumed that the
3412 * variable read itself will be atomic.
3414 max_pacing_rate = socket->so_max_pacing_rate;
3416 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3423 * Track route changes for TX rate limiting.
3426 in_pcboutput_eagain(struct inpcb *inp)
3433 if (inp->inp_snd_tag == NULL)
3436 if (!INP_WLOCKED(inp)) {
3438 * NOTE: If the write locking fails, we need to bail
3439 * out and use the non-ratelimited ring for the
3440 * transmit until there is a new chance to get the
3443 if (!INP_TRY_UPGRADE(inp))
3450 /* detach rate limiting */
3451 in_pcbdetach_txrtlmt(inp);
3453 /* make sure new mbuf send tag allocation is made */
3454 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3464 rate_limit_active = counter_u64_alloc(M_WAITOK);
3465 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3466 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3469 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3471 #endif /* RATELIMIT */