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.
8 * Copyright (c) 2021-2022 Gleb Smirnoff <glebius@FreeBSD.org>
11 * Portions of this software were developed by Robert N. M. Watson under
12 * contract to Juniper Networks, Inc.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
45 #include "opt_ipsec.h"
47 #include "opt_inet6.h"
48 #include "opt_ratelimit.h"
49 #include "opt_route.h"
52 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/libkern.h>
57 #include <sys/malloc.h>
59 #include <sys/eventhandler.h>
60 #include <sys/domain.h>
62 #include <sys/protosw.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
67 #include <sys/sockio.h>
70 #include <sys/refcount.h>
72 #include <sys/kernel.h>
73 #include <sys/sysctl.h>
83 #include <net/if_var.h>
84 #include <net/if_private.h>
85 #include <net/if_types.h>
86 #include <net/if_llatbl.h>
87 #include <net/route.h>
88 #include <net/rss_config.h>
91 #if defined(INET) || defined(INET6)
92 #include <netinet/in.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_pcb_var.h>
95 #include <netinet/tcp.h>
97 #include <netinet/in_var.h>
98 #include <netinet/in_fib.h>
100 #include <netinet/ip_var.h>
102 #include <netinet/ip6.h>
103 #include <netinet6/in6_pcb.h>
104 #include <netinet6/in6_var.h>
105 #include <netinet6/ip6_var.h>
107 #include <net/route/nhop.h>
110 #include <netipsec/ipsec_support.h>
112 #include <security/mac/mac_framework.h>
114 #define INPCBLBGROUP_SIZMIN 8
115 #define INPCBLBGROUP_SIZMAX 256
116 #define INP_FREED 0x00000200 /* See in_pcb.h. */
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 /* Enable random ephemeral port allocation by default. */
138 VNET_DEFINE(int, ipport_randomized) = 1;
141 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
142 struct in_addr faddr, u_int fport_arg,
143 struct in_addr laddr, u_int lport_arg,
144 int lookupflags, uint8_t numa_domain);
146 #define RANGECHK(var, min, max) \
147 if ((var) < (min)) { (var) = (min); } \
148 else if ((var) > (max)) { (var) = (max); }
151 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
155 error = sysctl_handle_int(oidp, arg1, arg2, req);
157 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
158 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
159 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
160 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
161 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
162 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
169 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
175 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
177 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
178 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
179 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
183 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
186 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
187 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
190 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
191 &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 | CTLFLAG_NEEDGIANT,
195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
197 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
198 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
199 &VNET_NAME(ipport_reservedhigh), 0, "");
200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
201 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
202 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
203 CTLFLAG_VNET | CTLFLAG_RW,
204 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
207 counter_u64_t rate_limit_new;
208 counter_u64_t rate_limit_chg;
209 counter_u64_t rate_limit_active;
210 counter_u64_t rate_limit_alloc_fail;
211 counter_u64_t rate_limit_set_ok;
213 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
215 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
216 &rate_limit_active, "Active rate limited connections");
217 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
218 &rate_limit_alloc_fail, "Rate limited connection failures");
219 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
220 &rate_limit_set_ok, "Rate limited setting succeeded");
221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD,
222 &rate_limit_new, "Total Rate limit new attempts");
223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD,
224 &rate_limit_chg, "Total Rate limited change attempts");
226 #endif /* RATELIMIT */
230 VNET_DEFINE(uint32_t, in_pcbhashseed);
232 in_pcbhashseed_init(void)
235 V_in_pcbhashseed = arc4random();
237 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
238 in_pcbhashseed_init, 0);
240 static void in_pcbremhash(struct inpcb *);
243 * in_pcb.c: manage the Protocol Control Blocks.
245 * NOTE: It is assumed that most of these functions will be called with
246 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
247 * functions often modify hash chains or addresses in pcbs.
250 static struct inpcblbgroup *
251 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred,
252 u_char vflag, uint16_t port, const union in_dependaddr *addr, int size,
255 struct inpcblbgroup *grp;
258 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
259 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
262 grp->il_cred = crhold(cred);
263 grp->il_vflag = vflag;
264 grp->il_lport = port;
265 grp->il_numa_domain = numa_domain;
266 grp->il_dependladdr = *addr;
267 grp->il_inpsiz = size;
268 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
273 in_pcblbgroup_free_deferred(epoch_context_t ctx)
275 struct inpcblbgroup *grp;
277 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
278 crfree(grp->il_cred);
283 in_pcblbgroup_free(struct inpcblbgroup *grp)
286 CK_LIST_REMOVE(grp, il_list);
287 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
290 static struct inpcblbgroup *
291 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
292 struct inpcblbgroup *old_grp, int size)
294 struct inpcblbgroup *grp;
297 grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag,
298 old_grp->il_lport, &old_grp->il_dependladdr, size,
299 old_grp->il_numa_domain);
303 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
304 ("invalid new local group size %d and old local group count %d",
305 grp->il_inpsiz, old_grp->il_inpcnt));
307 for (i = 0; i < old_grp->il_inpcnt; ++i)
308 grp->il_inp[i] = old_grp->il_inp[i];
309 grp->il_inpcnt = old_grp->il_inpcnt;
310 in_pcblbgroup_free(old_grp);
315 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
316 * and shrink group if possible.
319 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
322 struct inpcblbgroup *grp, *new_grp;
325 for (; i + 1 < grp->il_inpcnt; ++i)
326 grp->il_inp[i] = grp->il_inp[i + 1];
329 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
330 grp->il_inpcnt <= grp->il_inpsiz / 4) {
331 /* Shrink this group. */
332 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
339 * Add PCB to load balance group for SO_REUSEPORT_LB option.
342 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain)
344 const static struct timeval interval = { 60, 0 };
345 static struct timeval lastprint;
346 struct inpcbinfo *pcbinfo;
347 struct inpcblbgrouphead *hdr;
348 struct inpcblbgroup *grp;
351 pcbinfo = inp->inp_pcbinfo;
353 INP_WLOCK_ASSERT(inp);
354 INP_HASH_WLOCK_ASSERT(pcbinfo);
358 * Don't allow IPv4 mapped INET6 wild socket.
360 if ((inp->inp_vflag & INP_IPV4) &&
361 inp->inp_laddr.s_addr == INADDR_ANY &&
362 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
367 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
368 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
369 CK_LIST_FOREACH(grp, hdr, il_list) {
370 if (grp->il_cred->cr_prison == inp->inp_cred->cr_prison &&
371 grp->il_vflag == inp->inp_vflag &&
372 grp->il_lport == inp->inp_lport &&
373 grp->il_numa_domain == numa_domain &&
374 memcmp(&grp->il_dependladdr,
375 &inp->inp_inc.inc_ie.ie_dependladdr,
376 sizeof(grp->il_dependladdr)) == 0) {
381 /* Create new load balance group. */
382 grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag,
383 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
384 INPCBLBGROUP_SIZMIN, numa_domain);
387 } else if (grp->il_inpcnt == grp->il_inpsiz) {
388 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
389 if (ratecheck(&lastprint, &interval))
390 printf("lb group port %d, limit reached\n",
391 ntohs(grp->il_lport));
395 /* Expand this local group. */
396 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
401 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
402 ("invalid local group size %d and count %d", grp->il_inpsiz,
405 grp->il_inp[grp->il_inpcnt] = inp;
411 * Remove PCB from load balance group.
414 in_pcbremlbgrouphash(struct inpcb *inp)
416 struct inpcbinfo *pcbinfo;
417 struct inpcblbgrouphead *hdr;
418 struct inpcblbgroup *grp;
421 pcbinfo = inp->inp_pcbinfo;
423 INP_WLOCK_ASSERT(inp);
424 INP_HASH_WLOCK_ASSERT(pcbinfo);
426 hdr = &pcbinfo->ipi_lbgrouphashbase[
427 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
428 CK_LIST_FOREACH(grp, hdr, il_list) {
429 for (i = 0; i < grp->il_inpcnt; ++i) {
430 if (grp->il_inp[i] != inp)
433 if (grp->il_inpcnt == 1) {
434 /* We are the last, free this local group. */
435 in_pcblbgroup_free(grp);
437 /* Pull up inpcbs, shrink group if possible. */
438 in_pcblbgroup_reorder(hdr, &grp, i);
446 in_pcblbgroup_numa(struct inpcb *inp, int arg)
448 struct inpcbinfo *pcbinfo;
449 struct inpcblbgrouphead *hdr;
450 struct inpcblbgroup *grp;
455 case TCP_REUSPORT_LB_NUMA_NODOM:
456 numa_domain = M_NODOM;
458 case TCP_REUSPORT_LB_NUMA_CURDOM:
459 numa_domain = PCPU_GET(domain);
462 if (arg < 0 || arg >= vm_ndomains)
468 pcbinfo = inp->inp_pcbinfo;
469 INP_WLOCK_ASSERT(inp);
470 INP_HASH_WLOCK(pcbinfo);
471 hdr = &pcbinfo->ipi_lbgrouphashbase[
472 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
473 CK_LIST_FOREACH(grp, hdr, il_list) {
474 for (i = 0; i < grp->il_inpcnt; ++i) {
475 if (grp->il_inp[i] != inp)
478 if (grp->il_numa_domain == numa_domain) {
479 goto abort_with_hash_wlock;
482 /* Remove it from the old group. */
483 in_pcbremlbgrouphash(inp);
485 /* Add it to the new group based on numa domain. */
486 in_pcbinslbgrouphash(inp, numa_domain);
487 goto abort_with_hash_wlock;
491 abort_with_hash_wlock:
492 INP_HASH_WUNLOCK(pcbinfo);
496 /* Make sure it is safe to use hashinit(9) on CK_LIST. */
497 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb)));
500 * Initialize an inpcbinfo - a per-VNET instance of connections db.
503 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor,
504 u_int hash_nelements, u_int porthash_nelements)
507 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF);
508 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name,
511 pcbinfo->ipi_vnet = curvnet;
513 CK_LIST_INIT(&pcbinfo->ipi_listhead);
514 pcbinfo->ipi_count = 0;
515 pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB,
516 &pcbinfo->ipi_hashmask);
517 pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB,
518 &pcbinfo->ipi_hashmask);
519 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
520 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
521 &pcbinfo->ipi_porthashmask);
522 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
523 &pcbinfo->ipi_lbgrouphashmask);
524 pcbinfo->ipi_zone = pcbstor->ips_zone;
525 pcbinfo->ipi_portzone = pcbstor->ips_portzone;
526 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone);
530 * Destroy an inpcbinfo.
533 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
536 KASSERT(pcbinfo->ipi_count == 0,
537 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
539 hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask);
540 hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask);
541 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
542 pcbinfo->ipi_porthashmask);
543 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
544 pcbinfo->ipi_lbgrouphashmask);
545 mtx_destroy(&pcbinfo->ipi_hash_lock);
546 mtx_destroy(&pcbinfo->ipi_lock);
550 * Initialize a pcbstorage - per protocol zones to allocate inpcbs.
552 static void inpcb_dtor(void *, int, void *);
553 static void inpcb_fini(void *, int);
555 in_pcbstorage_init(void *arg)
557 struct inpcbstorage *pcbstor = arg;
559 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name,
560 pcbstor->ips_size, NULL, inpcb_dtor, pcbstor->ips_pcbinit,
561 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR);
562 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name,
563 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
564 uma_zone_set_smr(pcbstor->ips_portzone,
565 uma_zone_get_smr(pcbstor->ips_zone));
569 * Destroy a pcbstorage - used by unloadable protocols.
572 in_pcbstorage_destroy(void *arg)
574 struct inpcbstorage *pcbstor = arg;
576 uma_zdestroy(pcbstor->ips_zone);
577 uma_zdestroy(pcbstor->ips_portzone);
581 * Allocate a PCB and associate it with the socket.
582 * On success return with the PCB locked.
585 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
588 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
592 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT);
595 bzero(&inp->inp_start_zero, inp_zero_size);
597 inp->inp_numa_domain = M_NODOM;
599 inp->inp_pcbinfo = pcbinfo;
600 inp->inp_socket = so;
601 inp->inp_cred = crhold(so->so_cred);
602 inp->inp_inc.inc_fibnum = so->so_fibnum;
604 error = mac_inpcb_init(inp, M_NOWAIT);
607 mac_inpcb_create(so, inp);
609 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
610 error = ipsec_init_pcbpolicy(inp);
613 mac_inpcb_destroy(inp);
619 if (INP_SOCKAF(so) == AF_INET6) {
620 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6;
622 inp->inp_flags |= IN6P_IPV6_V6ONLY;
625 inp->inp_vflag |= INP_IPV4;
627 if (V_ip6_auto_flowlabel)
628 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
629 inp->in6p_hops = -1; /* use kernel default */
632 #if defined(INET) && defined(INET6)
636 inp->inp_vflag |= INP_IPV4;
638 inp->inp_smr = SMR_SEQ_INVALID;
641 * Routes in inpcb's can cache L2 as well; they are guaranteed
644 inp->inp_route.ro_flags = RT_LLE_CACHE;
645 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */
647 INP_INFO_WLOCK(pcbinfo);
648 pcbinfo->ipi_count++;
649 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
650 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list);
651 INP_INFO_WUNLOCK(pcbinfo);
656 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
658 uma_zfree_smr(pcbinfo->ipi_zone, inp);
665 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred)
669 KASSERT(sin == NULL || sin->sin_family == AF_INET,
670 ("%s: invalid address family for %p", __func__, sin));
671 KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in),
672 ("%s: invalid address length for %p", __func__, sin));
673 INP_WLOCK_ASSERT(inp);
674 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
676 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
678 anonport = sin == NULL || sin->sin_port == 0;
679 error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr,
680 &inp->inp_lport, cred);
683 if (in_pcbinshash(inp) != 0) {
684 inp->inp_laddr.s_addr = INADDR_ANY;
689 inp->inp_flags |= INP_ANONPORT;
694 #if defined(INET) || defined(INET6)
696 * Assign a local port like in_pcb_lport(), but also used with connect()
697 * and a foreign address and port. If fsa is non-NULL, choose a local port
698 * that is unused with those, otherwise one that is completely unused.
699 * lsa can be NULL for IPv6.
702 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
703 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
705 struct inpcbinfo *pcbinfo;
706 struct inpcb *tmpinp;
707 unsigned short *lastport;
709 u_short aux, first, last, lport;
711 struct in_addr laddr, faddr;
714 struct in6_addr *laddr6, *faddr6;
717 pcbinfo = inp->inp_pcbinfo;
720 * Because no actual state changes occur here, a global write lock on
721 * the pcbinfo isn't required.
723 INP_LOCK_ASSERT(inp);
724 INP_HASH_LOCK_ASSERT(pcbinfo);
726 if (inp->inp_flags & INP_HIGHPORT) {
727 first = V_ipport_hifirstauto; /* sysctl */
728 last = V_ipport_hilastauto;
729 lastport = &pcbinfo->ipi_lasthi;
730 } else if (inp->inp_flags & INP_LOWPORT) {
731 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
734 first = V_ipport_lowfirstauto; /* 1023 */
735 last = V_ipport_lowlastauto; /* 600 */
736 lastport = &pcbinfo->ipi_lastlow;
738 first = V_ipport_firstauto; /* sysctl */
739 last = V_ipport_lastauto;
740 lastport = &pcbinfo->ipi_lastport;
744 * Instead of having two loops further down counting up or down
745 * make sure that first is always <= last and go with only one
746 * code path implementing all logic.
755 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */
756 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
758 laddr = ((struct sockaddr_in *)lsa)->sin_addr;
760 faddr = ((struct sockaddr_in *)fsa)->sin_addr;
765 if ((inp->inp_vflag & INP_IPV6) != 0) {
767 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
769 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
776 if (V_ipport_randomized)
777 *lastport = first + (arc4random() % (last - first));
779 count = last - first;
782 if (count-- < 0) /* completely used? */
783 return (EADDRNOTAVAIL);
785 if (*lastport < first || *lastport > last)
787 lport = htons(*lastport);
791 if (lsa->sa_family == AF_INET) {
792 tmpinp = in_pcblookup_hash_locked(pcbinfo,
793 faddr, fport, laddr, lport, lookupflags,
798 if (lsa->sa_family == AF_INET6) {
799 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
800 faddr6, fport, laddr6, lport, lookupflags,
806 if ((inp->inp_vflag & INP_IPV6) != 0) {
807 tmpinp = in6_pcblookup_local(pcbinfo,
808 &inp->in6p_laddr, lport, lookupflags, cred);
810 if (tmpinp == NULL &&
811 (inp->inp_vflag & INP_IPV4))
812 tmpinp = in_pcblookup_local(pcbinfo,
813 laddr, lport, lookupflags, cred);
817 #if defined(INET) && defined(INET6)
821 tmpinp = in_pcblookup_local(pcbinfo, laddr,
822 lport, lookupflags, cred);
825 } while (tmpinp != NULL);
833 * Select a local port (number) to use.
836 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
837 struct ucred *cred, int lookupflags)
839 struct sockaddr_in laddr;
842 bzero(&laddr, sizeof(laddr));
843 laddr.sin_family = AF_INET;
844 laddr.sin_addr = *laddrp;
846 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
847 NULL, lportp, NULL, 0, cred, lookupflags));
851 * Return cached socket options.
854 inp_so_options(const struct inpcb *inp)
860 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
861 so_options |= SO_REUSEPORT_LB;
862 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
863 so_options |= SO_REUSEPORT;
864 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
865 so_options |= SO_REUSEADDR;
868 #endif /* INET || INET6 */
872 * Set up a bind operation on a PCB, performing port allocation
873 * as required, but do not actually modify the PCB. Callers can
874 * either complete the bind by setting inp_laddr/inp_lport and
875 * calling in_pcbinshash(), or they can just use the resulting
876 * port and address to authorise the sending of a once-off packet.
878 * On error, the values of *laddrp and *lportp are not changed.
881 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp,
882 u_short *lportp, struct ucred *cred)
884 struct socket *so = inp->inp_socket;
885 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
886 struct in_addr laddr;
888 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
892 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
893 * so that we don't have to add to the (already messy) code below.
895 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
898 * No state changes, so read locks are sufficient here.
900 INP_LOCK_ASSERT(inp);
901 INP_HASH_LOCK_ASSERT(pcbinfo);
903 laddr.s_addr = *laddrp;
904 if (sin != NULL && laddr.s_addr != INADDR_ANY)
906 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
907 lookupflags = INPLOOKUP_WILDCARD;
909 if ((error = prison_local_ip4(cred, &laddr)) != 0)
912 KASSERT(sin->sin_family == AF_INET,
913 ("%s: invalid family for address %p", __func__, sin));
914 KASSERT(sin->sin_len == sizeof(*sin),
915 ("%s: invalid length for address %p", __func__, sin));
917 error = prison_local_ip4(cred, &sin->sin_addr);
920 if (sin->sin_port != *lportp) {
921 /* Don't allow the port to change. */
924 lport = sin->sin_port;
926 /* NB: lport is left as 0 if the port isn't being changed. */
927 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
929 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
930 * allow complete duplication of binding if
931 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
932 * and a multicast address is bound on both
933 * new and duplicated sockets.
935 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
936 reuseport = SO_REUSEADDR|SO_REUSEPORT;
938 * XXX: How to deal with SO_REUSEPORT_LB here?
939 * Treat same as SO_REUSEPORT for now.
941 if ((so->so_options &
942 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
943 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
944 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
945 sin->sin_port = 0; /* yech... */
946 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
948 * Is the address a local IP address?
949 * If INP_BINDANY is set, then the socket may be bound
950 * to any endpoint address, local or not.
952 if ((inp->inp_flags & INP_BINDANY) == 0 &&
953 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
954 return (EADDRNOTAVAIL);
956 laddr = sin->sin_addr;
961 if (ntohs(lport) <= V_ipport_reservedhigh &&
962 ntohs(lport) >= V_ipport_reservedlow &&
963 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
965 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
966 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
967 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
968 lport, INPLOOKUP_WILDCARD, cred);
971 * This entire block sorely needs a rewrite.
974 (so->so_type != SOCK_STREAM ||
975 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
976 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
977 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
978 (t->inp_flags2 & INP_REUSEPORT) ||
979 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
980 (inp->inp_cred->cr_uid !=
981 t->inp_cred->cr_uid))
984 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
985 lport, lookupflags, cred);
986 if (t != NULL && (reuseport & inp_so_options(t)) == 0 &&
987 (reuseport_lb & inp_so_options(t)) == 0) {
989 if (ntohl(sin->sin_addr.s_addr) !=
991 ntohl(t->inp_laddr.s_addr) !=
993 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
994 (t->inp_vflag & INP_IPV6PROTO) == 0)
1003 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1007 *laddrp = laddr.s_addr;
1013 * Connect from a socket to a specified address.
1014 * Both address and port must be specified in argument sin.
1015 * If don't have a local address for this socket yet,
1019 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred,
1020 bool rehash __unused)
1022 u_short lport, fport;
1023 in_addr_t laddr, faddr;
1024 int anonport, error;
1026 INP_WLOCK_ASSERT(inp);
1027 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1028 KASSERT(in_nullhost(inp->inp_faddr),
1029 ("%s: inp is already connected", __func__));
1031 lport = inp->inp_lport;
1032 laddr = inp->inp_laddr.s_addr;
1033 anonport = (lport == 0);
1034 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport,
1039 inp->inp_faddr.s_addr = faddr;
1040 inp->inp_fport = fport;
1042 /* Do the initial binding of the local address if required. */
1043 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1044 inp->inp_lport = lport;
1045 inp->inp_laddr.s_addr = laddr;
1046 if (in_pcbinshash(inp) != 0) {
1047 inp->inp_laddr.s_addr = inp->inp_faddr.s_addr =
1049 inp->inp_lport = inp->inp_fport = 0;
1053 inp->inp_lport = lport;
1054 inp->inp_laddr.s_addr = laddr;
1055 if ((inp->inp_flags & INP_INHASHLIST) != 0)
1062 inp->inp_flags |= INP_ANONPORT;
1067 * Do proper source address selection on an unbound socket in case
1068 * of connect. Take jails into account as well.
1071 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1075 struct sockaddr *sa;
1076 struct sockaddr_in *sin, dst;
1077 struct nhop_object *nh;
1081 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1084 * Bypass source address selection and use the primary jail IP
1087 if (!prison_saddrsel_ip4(cred, laddr))
1093 bzero(&dst, sizeof(dst));
1095 sin->sin_family = AF_INET;
1096 sin->sin_len = sizeof(struct sockaddr_in);
1097 sin->sin_addr.s_addr = faddr->s_addr;
1100 * If route is known our src addr is taken from the i/f,
1103 * Find out route to destination.
1105 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1106 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1110 * If we found a route, use the address corresponding to
1111 * the outgoing interface.
1113 * Otherwise assume faddr is reachable on a directly connected
1114 * network and try to find a corresponding interface to take
1115 * the source address from.
1117 if (nh == NULL || nh->nh_ifp == NULL) {
1118 struct in_ifaddr *ia;
1121 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1122 inp->inp_socket->so_fibnum));
1124 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1125 inp->inp_socket->so_fibnum));
1128 error = ENETUNREACH;
1132 if (!prison_flag(cred, PR_IP4)) {
1133 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1139 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1141 if (sa->sa_family != AF_INET)
1143 sin = (struct sockaddr_in *)sa;
1144 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1145 ia = (struct in_ifaddr *)ifa;
1150 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1154 /* 3. As a last resort return the 'default' jail address. */
1155 error = prison_get_ip4(cred, laddr);
1160 * If the outgoing interface on the route found is not
1161 * a loopback interface, use the address from that interface.
1162 * In case of jails do those three steps:
1163 * 1. check if the interface address belongs to the jail. If so use it.
1164 * 2. check if we have any address on the outgoing interface
1165 * belonging to this jail. If so use it.
1166 * 3. as a last resort return the 'default' jail address.
1168 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1169 struct in_ifaddr *ia;
1172 /* If not jailed, use the default returned. */
1173 if (!prison_flag(cred, PR_IP4)) {
1174 ia = (struct in_ifaddr *)nh->nh_ifa;
1175 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1180 /* 1. Check if the iface address belongs to the jail. */
1181 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1182 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1183 ia = (struct in_ifaddr *)nh->nh_ifa;
1184 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1189 * 2. Check if we have any address on the outgoing interface
1190 * belonging to this jail.
1194 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1196 if (sa->sa_family != AF_INET)
1198 sin = (struct sockaddr_in *)sa;
1199 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1200 ia = (struct in_ifaddr *)ifa;
1205 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1209 /* 3. As a last resort return the 'default' jail address. */
1210 error = prison_get_ip4(cred, laddr);
1215 * The outgoing interface is marked with 'loopback net', so a route
1216 * to ourselves is here.
1217 * Try to find the interface of the destination address and then
1218 * take the address from there. That interface is not necessarily
1219 * a loopback interface.
1220 * In case of jails, check that it is an address of the jail
1221 * and if we cannot find, fall back to the 'default' jail address.
1223 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1224 struct in_ifaddr *ia;
1226 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1227 inp->inp_socket->so_fibnum));
1229 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1230 inp->inp_socket->so_fibnum));
1232 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1234 if (!prison_flag(cred, PR_IP4)) {
1236 error = ENETUNREACH;
1239 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1249 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1251 if (sa->sa_family != AF_INET)
1253 sin = (struct sockaddr_in *)sa;
1254 if (prison_check_ip4(cred,
1255 &sin->sin_addr) == 0) {
1256 ia = (struct in_ifaddr *)ifa;
1261 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1266 /* 3. As a last resort return the 'default' jail address. */
1267 error = prison_get_ip4(cred, laddr);
1272 if (error == 0 && laddr->s_addr == INADDR_ANY)
1273 return (EHOSTUNREACH);
1278 * Set up for a connect from a socket to the specified address.
1279 * On entry, *laddrp and *lportp should contain the current local
1280 * address and port for the PCB; these are updated to the values
1281 * that should be placed in inp_laddr and inp_lport to complete
1284 * On success, *faddrp and *fportp will be set to the remote address
1285 * and port. These are not updated in the error case.
1288 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin,
1289 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1292 struct in_ifaddr *ia;
1293 struct in_addr laddr, faddr;
1294 u_short lport, fport;
1297 KASSERT(sin->sin_family == AF_INET,
1298 ("%s: invalid address family for %p", __func__, sin));
1299 KASSERT(sin->sin_len == sizeof(*sin),
1300 ("%s: invalid address length for %p", __func__, sin));
1303 * Because a global state change doesn't actually occur here, a read
1304 * lock is sufficient.
1307 INP_LOCK_ASSERT(inp);
1308 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1310 if (sin->sin_port == 0)
1311 return (EADDRNOTAVAIL);
1312 laddr.s_addr = *laddrp;
1314 faddr = sin->sin_addr;
1315 fport = sin->sin_port;
1317 if (CALC_FLOWID_OUTBOUND) {
1318 uint32_t hash_val, hash_type;
1320 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1321 inp->inp_socket->so_proto->pr_protocol, &hash_type);
1323 inp->inp_flowid = hash_val;
1324 inp->inp_flowtype = hash_type;
1327 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1329 * If the destination address is INADDR_ANY,
1330 * use the primary local address.
1331 * If the supplied address is INADDR_BROADCAST,
1332 * and the primary interface supports broadcast,
1333 * choose the broadcast address for that interface.
1335 if (faddr.s_addr == INADDR_ANY) {
1337 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1338 if ((error = prison_get_ip4(cred, &faddr)) != 0)
1340 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1341 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1343 faddr = satosin(&CK_STAILQ_FIRST(
1344 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1347 if (laddr.s_addr == INADDR_ANY) {
1348 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1350 * If the destination address is multicast and an outgoing
1351 * interface has been set as a multicast option, prefer the
1352 * address of that interface as our source address.
1354 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1355 inp->inp_moptions != NULL) {
1356 struct ip_moptions *imo;
1359 imo = inp->inp_moptions;
1360 if (imo->imo_multicast_ifp != NULL) {
1361 ifp = imo->imo_multicast_ifp;
1362 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1363 if (ia->ia_ifp == ifp &&
1364 prison_check_ip4(cred,
1365 &ia->ia_addr.sin_addr) == 0)
1369 error = EADDRNOTAVAIL;
1371 laddr = ia->ia_addr.sin_addr;
1381 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1382 fport, laddr, lport, 0, M_NODOM) != NULL)
1383 return (EADDRINUSE);
1385 struct sockaddr_in lsin, fsin;
1387 bzero(&lsin, sizeof(lsin));
1388 bzero(&fsin, sizeof(fsin));
1389 lsin.sin_family = AF_INET;
1390 lsin.sin_addr = laddr;
1391 fsin.sin_family = AF_INET;
1392 fsin.sin_addr = faddr;
1393 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1394 &lport, (struct sockaddr *)& fsin, fport, cred,
1395 INPLOOKUP_WILDCARD);
1399 *laddrp = laddr.s_addr;
1401 *faddrp = faddr.s_addr;
1407 in_pcbdisconnect(struct inpcb *inp)
1410 INP_WLOCK_ASSERT(inp);
1411 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1412 KASSERT(inp->inp_smr == SMR_SEQ_INVALID,
1413 ("%s: inp %p was already disconnected", __func__, inp));
1415 in_pcbremhash_locked(inp);
1417 /* See the comment in in_pcbinshash(). */
1418 inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr);
1419 inp->inp_laddr.s_addr = INADDR_ANY;
1420 inp->inp_faddr.s_addr = INADDR_ANY;
1426 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1427 * For most protocols, this will be invoked immediately prior to calling
1428 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1429 * socket, in which case in_pcbfree() is deferred.
1432 in_pcbdetach(struct inpcb *inp)
1435 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1438 if (inp->inp_snd_tag != NULL)
1439 in_pcbdetach_txrtlmt(inp);
1441 inp->inp_socket->so_pcb = NULL;
1442 inp->inp_socket = NULL;
1446 * inpcb hash lookups are protected by SMR section.
1448 * Once desired pcb has been found, switching from SMR section to a pcb
1449 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1450 * here because SMR is a critical section.
1451 * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1454 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1457 lock == INPLOOKUP_RLOCKPCB ?
1458 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1462 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1465 lock == INPLOOKUP_RLOCKPCB ?
1466 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1470 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1473 return (lock == INPLOOKUP_RLOCKPCB ?
1474 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1478 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1481 return (lock == INPLOOKUP_RLOCKPCB ?
1482 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1486 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags)
1489 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1490 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1492 if (__predict_true(inp_trylock(inp, lock))) {
1493 if (__predict_false(inp->inp_flags & ignflags)) {
1494 smr_exit(inp->inp_pcbinfo->ipi_smr);
1495 inp_unlock(inp, lock);
1498 smr_exit(inp->inp_pcbinfo->ipi_smr);
1502 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1503 smr_exit(inp->inp_pcbinfo->ipi_smr);
1504 inp_lock(inp, lock);
1505 if (__predict_false(in_pcbrele(inp, lock)))
1508 * inp acquired through refcount & lock for sure didn't went
1509 * through uma_zfree(). However, it may have already went
1510 * through in_pcbfree() and has another reference, that
1511 * prevented its release by our in_pcbrele().
1513 if (__predict_false(inp->inp_flags & ignflags)) {
1514 inp_unlock(inp, lock);
1519 smr_exit(inp->inp_pcbinfo->ipi_smr);
1525 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1529 * in_pcblookup() family of functions ignore not only freed entries,
1530 * that may be found due to lockless access to the hash, but dropped
1533 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED));
1537 * inp_next() - inpcb hash/list traversal iterator
1539 * Requires initialized struct inpcb_iterator for context.
1540 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1542 * - Iterator can have either write-lock or read-lock semantics, that can not
1544 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1545 * a single hash slot. Note: only rip_input() does the latter.
1546 * - Iterator may have optional bool matching function. The matching function
1547 * will be executed for each inpcb in the SMR context, so it can not acquire
1548 * locks and can safely access only immutable fields of inpcb.
1550 * A fresh initialized iterator has NULL inpcb in its context and that
1551 * means that inp_next() call would return the very first inpcb on the list
1552 * locked with desired semantic. In all following calls the context pointer
1553 * shall hold the current inpcb pointer. The KPI user is not supposed to
1554 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL
1555 * and write NULL to its context. After end of traversal an iterator can be
1558 * List traversals have the following features/constraints:
1559 * - New entries won't be seen, as they are always added to the head of a list.
1560 * - Removed entries won't stop traversal as long as they are not added to
1561 * a different list. This is violated by in_pcbrehash().
1563 #define II_LIST_FIRST(ipi, hash) \
1564 (((hash) == INP_ALL_LIST) ? \
1565 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \
1566 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)]))
1567 #define II_LIST_NEXT(inp, hash) \
1568 (((hash) == INP_ALL_LIST) ? \
1569 CK_LIST_NEXT((inp), inp_list) : \
1570 CK_LIST_NEXT((inp), inp_hash_exact))
1571 #define II_LOCK_ASSERT(inp, lock) \
1572 rw_assert(&(inp)->inp_lock, \
1573 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED )
1575 inp_next(struct inpcb_iterator *ii)
1577 const struct inpcbinfo *ipi = ii->ipi;
1578 inp_match_t *match = ii->match;
1579 void *ctx = ii->ctx;
1580 inp_lookup_t lock = ii->lock;
1581 int hash = ii->hash;
1584 if (ii->inp == NULL) { /* First call. */
1585 smr_enter(ipi->ipi_smr);
1586 /* This is unrolled CK_LIST_FOREACH(). */
1587 for (inp = II_LIST_FIRST(ipi, hash);
1589 inp = II_LIST_NEXT(inp, hash)) {
1590 if (match != NULL && (match)(inp, ctx) == false)
1592 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED)))
1595 smr_enter(ipi->ipi_smr);
1596 MPASS(inp != II_LIST_FIRST(ipi, hash));
1597 inp = II_LIST_FIRST(ipi, hash);
1604 smr_exit(ipi->ipi_smr);
1611 /* Not a first call. */
1612 smr_enter(ipi->ipi_smr);
1615 II_LOCK_ASSERT(inp, lock);
1617 inp = II_LIST_NEXT(inp, hash);
1619 smr_exit(ipi->ipi_smr);
1623 if (match != NULL && (match)(inp, ctx) == false)
1626 if (__predict_true(inp_trylock(inp, lock))) {
1627 if (__predict_false(inp->inp_flags & INP_FREED)) {
1629 * Entries are never inserted in middle of a list, thus
1630 * as long as we are in SMR, we can continue traversal.
1631 * Jump to 'restart' should yield in the same result,
1632 * but could produce unnecessary looping. Could this
1633 * looping be unbound?
1635 inp_unlock(inp, lock);
1638 smr_exit(ipi->ipi_smr);
1644 * Can't obtain lock immediately, thus going hard. Once we exit the
1645 * SMR section we can no longer jump to 'next', and our only stable
1646 * anchoring point is ii->inp, which we keep locked for this case, so
1647 * we jump to 'restart'.
1649 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1650 smr_exit(ipi->ipi_smr);
1651 inp_lock(inp, lock);
1652 if (__predict_false(in_pcbrele(inp, lock))) {
1653 smr_enter(ipi->ipi_smr);
1657 * See comment in inp_smr_lock().
1659 if (__predict_false(inp->inp_flags & INP_FREED)) {
1660 inp_unlock(inp, lock);
1661 smr_enter(ipi->ipi_smr);
1668 inp_unlock(ii->inp, lock);
1675 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1676 * stability of an inpcb pointer despite the inpcb lock being released or
1677 * SMR section exited.
1679 * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1682 in_pcbref(struct inpcb *inp)
1684 u_int old __diagused;
1686 old = refcount_acquire(&inp->inp_refcount);
1687 KASSERT(old > 0, ("%s: refcount 0", __func__));
1691 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1692 * freeing the pcb, if the reference was very last.
1695 in_pcbrele_rlocked(struct inpcb *inp)
1698 INP_RLOCK_ASSERT(inp);
1700 if (!refcount_release(&inp->inp_refcount))
1703 MPASS(inp->inp_flags & INP_FREED);
1704 MPASS(inp->inp_socket == NULL);
1705 MPASS(inp->inp_in_hpts == 0);
1707 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1712 in_pcbrele_wlocked(struct inpcb *inp)
1715 INP_WLOCK_ASSERT(inp);
1717 if (!refcount_release(&inp->inp_refcount))
1720 MPASS(inp->inp_flags & INP_FREED);
1721 MPASS(inp->inp_socket == NULL);
1722 MPASS(inp->inp_in_hpts == 0);
1724 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1729 * Unconditionally schedule an inpcb to be freed by decrementing its
1730 * reference count, which should occur only after the inpcb has been detached
1731 * from its socket. If another thread holds a temporary reference (acquired
1732 * using in_pcbref()) then the free is deferred until that reference is
1733 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1734 * Almost all work, including removal from global lists, is done in this
1735 * context, where the pcbinfo lock is held.
1738 in_pcbfree(struct inpcb *inp)
1740 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1742 struct ip_moptions *imo;
1745 struct ip6_moptions *im6o;
1748 INP_WLOCK_ASSERT(inp);
1749 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1750 KASSERT((inp->inp_flags & INP_FREED) == 0,
1751 ("%s: called twice for pcb %p", __func__, inp));
1753 inp->inp_flags |= INP_FREED;
1754 INP_INFO_WLOCK(pcbinfo);
1755 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1756 pcbinfo->ipi_count--;
1757 CK_LIST_REMOVE(inp, inp_list);
1758 INP_INFO_WUNLOCK(pcbinfo);
1760 if (inp->inp_flags & INP_INHASHLIST)
1763 RO_INVALIDATE_CACHE(&inp->inp_route);
1765 mac_inpcb_destroy(inp);
1767 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1768 if (inp->inp_sp != NULL)
1769 ipsec_delete_pcbpolicy(inp);
1772 if (inp->inp_options)
1773 (void)m_free(inp->inp_options);
1774 imo = inp->inp_moptions;
1777 if (inp->inp_vflag & INP_IPV6PROTO) {
1778 ip6_freepcbopts(inp->in6p_outputopts);
1779 im6o = inp->in6p_moptions;
1784 if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1788 ip6_freemoptions(im6o);
1791 inp_freemoptions(imo);
1793 /* Destruction is finalized in inpcb_dtor(). */
1797 inpcb_dtor(void *mem, int size, void *arg)
1799 struct inpcb *inp = mem;
1801 crfree(inp->inp_cred);
1803 inp->inp_cred = NULL;
1808 * Different protocols initialize their inpcbs differently - giving
1809 * different name to the lock. But they all are disposed the same.
1812 inpcb_fini(void *mem, int size)
1814 struct inpcb *inp = mem;
1816 INP_LOCK_DESTROY(inp);
1820 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1821 * port reservation, and preventing it from being returned by inpcb lookups.
1823 * It is used by TCP to mark an inpcb as unused and avoid future packet
1824 * delivery or event notification when a socket remains open but TCP has
1825 * closed. This might occur as a result of a shutdown()-initiated TCP close
1826 * or a RST on the wire, and allows the port binding to be reused while still
1827 * maintaining the invariant that so_pcb always points to a valid inpcb until
1830 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1831 * in_pcbnotifyall() and in_pcbpurgeif0()?
1834 in_pcbdrop(struct inpcb *inp)
1837 INP_WLOCK_ASSERT(inp);
1839 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1840 MPASS(inp->inp_refcount > 1);
1843 inp->inp_flags |= INP_DROPPED;
1844 if (inp->inp_flags & INP_INHASHLIST)
1850 * Common routines to return the socket addresses associated with inpcbs.
1853 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1855 struct sockaddr_in *sin;
1857 sin = malloc(sizeof *sin, M_SONAME,
1859 sin->sin_family = AF_INET;
1860 sin->sin_len = sizeof(*sin);
1861 sin->sin_addr = *addr_p;
1862 sin->sin_port = port;
1864 return (struct sockaddr *)sin;
1868 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1871 struct in_addr addr;
1874 inp = sotoinpcb(so);
1875 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1878 port = inp->inp_lport;
1879 addr = inp->inp_laddr;
1882 *nam = in_sockaddr(port, &addr);
1887 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1890 struct in_addr addr;
1893 inp = sotoinpcb(so);
1894 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1897 port = inp->inp_fport;
1898 addr = inp->inp_faddr;
1901 *nam = in_sockaddr(port, &addr);
1906 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1907 struct inpcb *(*notify)(struct inpcb *, int))
1909 struct inpcb *inp, *inp_temp;
1911 INP_INFO_WLOCK(pcbinfo);
1912 CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
1915 if ((inp->inp_vflag & INP_IPV4) == 0) {
1920 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1921 inp->inp_socket == NULL) {
1925 if ((*notify)(inp, errno))
1928 INP_INFO_WUNLOCK(pcbinfo);
1932 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
1935 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
1942 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1944 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
1945 inp_v4_multi_match, NULL);
1947 struct in_multi *inm;
1948 struct in_mfilter *imf;
1949 struct ip_moptions *imo;
1951 IN_MULTI_LOCK_ASSERT();
1953 while ((inp = inp_next(&inpi)) != NULL) {
1954 INP_WLOCK_ASSERT(inp);
1956 imo = inp->inp_moptions;
1958 * Unselect the outgoing interface if it is being
1961 if (imo->imo_multicast_ifp == ifp)
1962 imo->imo_multicast_ifp = NULL;
1965 * Drop multicast group membership if we joined
1966 * through the interface being detached.
1968 * XXX This can all be deferred to an epoch_call
1971 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1972 if ((inm = imf->imf_inm) == NULL)
1974 if (inm->inm_ifp != ifp)
1976 ip_mfilter_remove(&imo->imo_head, imf);
1977 in_leavegroup_locked(inm, NULL);
1978 ip_mfilter_free(imf);
1985 * Lookup a PCB based on the local address and port. Caller must hold the
1986 * hash lock. No inpcb locks or references are acquired.
1988 #define INP_LOOKUP_MAPPED_PCB_COST 3
1990 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1991 u_short lport, int lookupflags, struct ucred *cred)
1995 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
2001 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2002 ("%s: invalid lookup flags %d", __func__, lookupflags));
2003 INP_HASH_LOCK_ASSERT(pcbinfo);
2005 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
2006 struct inpcbhead *head;
2008 * Look for an unconnected (wildcard foreign addr) PCB that
2009 * matches the local address and port we're looking for.
2011 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2012 pcbinfo->ipi_hashmask)];
2013 CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2015 /* XXX inp locking */
2016 if ((inp->inp_vflag & INP_IPV4) == 0)
2019 if (inp->inp_faddr.s_addr == INADDR_ANY &&
2020 inp->inp_laddr.s_addr == laddr.s_addr &&
2021 inp->inp_lport == lport) {
2025 if (prison_equal_ip4(cred->cr_prison,
2026 inp->inp_cred->cr_prison))
2035 struct inpcbporthead *porthash;
2036 struct inpcbport *phd;
2037 struct inpcb *match = NULL;
2039 * Best fit PCB lookup.
2041 * First see if this local port is in use by looking on the
2044 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2045 pcbinfo->ipi_porthashmask)];
2046 CK_LIST_FOREACH(phd, porthash, phd_hash) {
2047 if (phd->phd_port == lport)
2052 * Port is in use by one or more PCBs. Look for best
2055 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2057 if (!prison_equal_ip4(inp->inp_cred->cr_prison,
2061 /* XXX inp locking */
2062 if ((inp->inp_vflag & INP_IPV4) == 0)
2065 * We never select the PCB that has
2066 * INP_IPV6 flag and is bound to :: if
2067 * we have another PCB which is bound
2068 * to 0.0.0.0. If a PCB has the
2069 * INP_IPV6 flag, then we set its cost
2070 * higher than IPv4 only PCBs.
2072 * Note that the case only happens
2073 * when a socket is bound to ::, under
2074 * the condition that the use of the
2075 * mapped address is allowed.
2077 if ((inp->inp_vflag & INP_IPV6) != 0)
2078 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2080 if (inp->inp_faddr.s_addr != INADDR_ANY)
2082 if (inp->inp_laddr.s_addr != INADDR_ANY) {
2083 if (laddr.s_addr == INADDR_ANY)
2085 else if (inp->inp_laddr.s_addr != laddr.s_addr)
2088 if (laddr.s_addr != INADDR_ANY)
2091 if (wildcard < matchwild) {
2093 matchwild = wildcard;
2102 #undef INP_LOOKUP_MAPPED_PCB_COST
2105 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain)
2107 return (domain == M_NODOM || domain == grp->il_numa_domain);
2110 static struct inpcb *
2111 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2112 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr,
2113 uint16_t lport, int domain)
2115 const struct inpcblbgrouphead *hdr;
2116 struct inpcblbgroup *grp;
2117 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild;
2119 INP_HASH_LOCK_ASSERT(pcbinfo);
2121 hdr = &pcbinfo->ipi_lbgrouphashbase[
2122 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2125 * Search for an LB group match based on the following criteria:
2126 * - prefer jailed groups to non-jailed groups
2127 * - prefer exact source address matches to wildcard matches
2128 * - prefer groups bound to the specified NUMA domain
2130 jail_exact = jail_wild = local_exact = local_wild = NULL;
2131 CK_LIST_FOREACH(grp, hdr, il_list) {
2135 if (!(grp->il_vflag & INP_IPV4))
2138 if (grp->il_lport != lport)
2141 injail = prison_flag(grp->il_cred, PR_IP4) != 0;
2142 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison,
2146 if (grp->il_laddr.s_addr == laddr->s_addr) {
2149 if (in_pcblookup_lb_numa_match(grp, domain))
2150 /* This is a perfect match. */
2152 } else if (local_exact == NULL ||
2153 in_pcblookup_lb_numa_match(grp, domain)) {
2156 } else if (grp->il_laddr.s_addr == INADDR_ANY) {
2158 if (jail_wild == NULL ||
2159 in_pcblookup_lb_numa_match(grp, domain))
2161 } else if (local_wild == NULL ||
2162 in_pcblookup_lb_numa_match(grp, domain)) {
2168 if (jail_exact != NULL)
2170 else if (jail_wild != NULL)
2172 else if (local_exact != NULL)
2179 return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2183 static struct inpcb *
2184 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2185 u_short fport, struct in_addr laddr, u_short lport)
2187 struct inpcbhead *head;
2188 struct inpcb *inp, *match;
2190 INP_HASH_LOCK_ASSERT(pcbinfo);
2193 head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport,
2194 pcbinfo->ipi_hashmask)];
2195 CK_LIST_FOREACH(inp, head, inp_hash_exact) {
2197 /* XXX inp locking */
2198 if ((inp->inp_vflag & INP_IPV4) == 0)
2201 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2202 inp->inp_laddr.s_addr == laddr.s_addr &&
2203 inp->inp_fport == fport &&
2204 inp->inp_lport == lport)
2210 static struct inpcb *
2211 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2212 u_short fport, struct in_addr laddr, u_short lport)
2214 struct inpcbhead *head;
2215 struct inpcb *inp, *local_wild, *local_exact, *jail_wild;
2217 struct inpcb *local_wild_mapped;
2220 INP_HASH_LOCK_ASSERT(pcbinfo);
2223 * Order of socket selection - we always prefer jails.
2224 * 1. jailed, non-wild.
2226 * 3. non-jailed, non-wild.
2227 * 4. non-jailed, wild.
2229 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2230 pcbinfo->ipi_hashmask)];
2231 local_wild = local_exact = jail_wild = NULL;
2233 local_wild_mapped = NULL;
2235 CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2239 /* XXX inp locking */
2240 if ((inp->inp_vflag & INP_IPV4) == 0)
2243 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2244 inp->inp_lport != lport)
2247 injail = prison_flag(inp->inp_cred, PR_IP4) != 0;
2249 if (prison_check_ip4_locked(inp->inp_cred->cr_prison,
2253 if (local_exact != NULL)
2257 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2261 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2263 /* XXX inp locking, NULL check */
2264 if (inp->inp_vflag & INP_IPV6PROTO)
2265 local_wild_mapped = inp;
2274 if (jail_wild != NULL)
2276 if (local_exact != NULL)
2277 return (local_exact);
2278 if (local_wild != NULL)
2279 return (local_wild);
2281 if (local_wild_mapped != NULL)
2282 return (local_wild_mapped);
2288 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2289 * that the caller has either locked the hash list, which usually happens
2290 * for bind(2) operations, or is in SMR section, which happens when sorting
2291 * out incoming packets.
2293 static struct inpcb *
2294 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2295 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2296 uint8_t numa_domain)
2299 const u_short fport = fport_arg, lport = lport_arg;
2301 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0,
2302 ("%s: invalid lookup flags %d", __func__, lookupflags));
2303 KASSERT(faddr.s_addr != INADDR_ANY,
2304 ("%s: invalid foreign address", __func__));
2305 KASSERT(laddr.s_addr != INADDR_ANY,
2306 ("%s: invalid local address", __func__));
2307 INP_HASH_LOCK_ASSERT(pcbinfo);
2309 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2313 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2314 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, &laddr,
2315 lport, numa_domain);
2317 inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr,
2318 fport, laddr, lport);
2325 static struct inpcb *
2326 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2327 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2328 uint8_t numa_domain)
2332 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2333 ("%s: invalid lookup flags %d", __func__, lookupflags));
2334 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2335 ("%s: LOCKPCB not set", __func__));
2337 smr_enter(pcbinfo->ipi_smr);
2338 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2339 lookupflags & INPLOOKUP_WILDCARD, numa_domain);
2341 if (__predict_false(inp_smr_lock(inp,
2342 (lookupflags & INPLOOKUP_LOCKMASK)) == false))
2345 smr_exit(pcbinfo->ipi_smr);
2351 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2352 * from which a pre-calculated hash value may be extracted.
2355 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2356 struct in_addr laddr, u_int lport, int lookupflags,
2357 struct ifnet *ifp __unused)
2359 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2360 lookupflags, M_NODOM));
2364 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2365 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2366 struct ifnet *ifp __unused, struct mbuf *m)
2368 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2369 lookupflags, m->m_pkthdr.numa_domain));
2374 * Insert PCB onto various hash lists.
2377 in_pcbinshash(struct inpcb *inp)
2379 struct inpcbhead *pcbhash;
2380 struct inpcbporthead *pcbporthash;
2381 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2382 struct inpcbport *phd;
2386 INP_WLOCK_ASSERT(inp);
2387 INP_HASH_WLOCK_ASSERT(pcbinfo);
2388 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2389 ("in_pcbinshash: INP_INHASHLIST"));
2392 if (inp->inp_vflag & INP_IPV6) {
2393 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport,
2394 inp->inp_fport, pcbinfo->ipi_hashmask);
2395 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr);
2399 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport,
2400 inp->inp_fport, pcbinfo->ipi_hashmask);
2401 connected = !in_nullhost(inp->inp_faddr);
2405 pcbhash = &pcbinfo->ipi_hash_exact[hash];
2407 pcbhash = &pcbinfo->ipi_hash_wild[hash];
2409 pcbporthash = &pcbinfo->ipi_porthashbase[
2410 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2413 * Add entry to load balance group.
2414 * Only do this if SO_REUSEPORT_LB is set.
2416 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) {
2417 int error = in_pcbinslbgrouphash(inp, M_NODOM);
2423 * Go through port list and look for a head for this lport.
2425 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2426 if (phd->phd_port == inp->inp_lport)
2431 * If none exists, malloc one and tack it on.
2434 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2436 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2437 in_pcbremlbgrouphash(inp);
2440 phd->phd_port = inp->inp_lport;
2441 CK_LIST_INIT(&phd->phd_pcblist);
2442 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2445 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2448 * The PCB may have been disconnected in the past. Before we can safely
2449 * make it visible in the hash table, we must wait for all readers which
2450 * may be traversing this PCB to finish.
2452 if (inp->inp_smr != SMR_SEQ_INVALID) {
2453 smr_wait(pcbinfo->ipi_smr, inp->inp_smr);
2454 inp->inp_smr = SMR_SEQ_INVALID;
2458 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact);
2460 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2461 inp->inp_flags |= INP_INHASHLIST;
2467 in_pcbremhash_locked(struct inpcb *inp)
2469 struct inpcbport *phd = inp->inp_phd;
2471 INP_WLOCK_ASSERT(inp);
2472 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2473 MPASS(inp->inp_flags & INP_INHASHLIST);
2475 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2476 in_pcbremlbgrouphash(inp);
2478 if (inp->inp_vflag & INP_IPV6) {
2479 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
2480 CK_LIST_REMOVE(inp, inp_hash_wild);
2482 CK_LIST_REMOVE(inp, inp_hash_exact);
2486 if (in_nullhost(inp->inp_faddr))
2487 CK_LIST_REMOVE(inp, inp_hash_wild);
2489 CK_LIST_REMOVE(inp, inp_hash_exact);
2491 CK_LIST_REMOVE(inp, inp_portlist);
2492 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2493 CK_LIST_REMOVE(phd, phd_hash);
2494 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
2496 inp->inp_flags &= ~INP_INHASHLIST;
2500 in_pcbremhash(struct inpcb *inp)
2502 INP_HASH_WLOCK(inp->inp_pcbinfo);
2503 in_pcbremhash_locked(inp);
2504 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
2508 * Move PCB to the proper hash bucket when { faddr, fport } have been
2509 * changed. NOTE: This does not handle the case of the lport changing (the
2510 * hashed port list would have to be updated as well), so the lport must
2511 * not change after in_pcbinshash() has been called.
2514 in_pcbrehash(struct inpcb *inp)
2516 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2517 struct inpcbhead *head;
2521 INP_WLOCK_ASSERT(inp);
2522 INP_HASH_WLOCK_ASSERT(pcbinfo);
2523 KASSERT(inp->inp_flags & INP_INHASHLIST,
2524 ("%s: !INP_INHASHLIST", __func__));
2525 KASSERT(inp->inp_smr == SMR_SEQ_INVALID,
2526 ("%s: inp was disconnected", __func__));
2529 if (inp->inp_vflag & INP_IPV6) {
2530 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport,
2531 inp->inp_fport, pcbinfo->ipi_hashmask);
2532 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr);
2536 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport,
2537 inp->inp_fport, pcbinfo->ipi_hashmask);
2538 connected = !in_nullhost(inp->inp_faddr);
2542 * When rehashing, the caller must ensure that either the new or the old
2543 * foreign address was unspecified.
2546 CK_LIST_REMOVE(inp, inp_hash_wild);
2548 CK_LIST_REMOVE(inp, inp_hash_exact);
2551 head = &pcbinfo->ipi_hash_exact[hash];
2552 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact);
2554 head = &pcbinfo->ipi_hash_wild[hash];
2555 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild);
2560 * Check for alternatives when higher level complains
2561 * about service problems. For now, invalidate cached
2562 * routing information. If the route was created dynamically
2563 * (by a redirect), time to try a default gateway again.
2566 in_losing(struct inpcb *inp)
2569 RO_INVALIDATE_CACHE(&inp->inp_route);
2574 * A set label operation has occurred at the socket layer, propagate the
2575 * label change into the in_pcb for the socket.
2578 in_pcbsosetlabel(struct socket *so)
2583 inp = sotoinpcb(so);
2584 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2588 mac_inpcb_sosetlabel(so, inp);
2595 inp_wlock(struct inpcb *inp)
2602 inp_wunlock(struct inpcb *inp)
2609 inp_rlock(struct inpcb *inp)
2616 inp_runlock(struct inpcb *inp)
2622 #ifdef INVARIANT_SUPPORT
2624 inp_lock_assert(struct inpcb *inp)
2627 INP_WLOCK_ASSERT(inp);
2631 inp_unlock_assert(struct inpcb *inp)
2634 INP_UNLOCK_ASSERT(inp);
2639 inp_apply_all(struct inpcbinfo *pcbinfo,
2640 void (*func)(struct inpcb *, void *), void *arg)
2642 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2643 INPLOOKUP_WLOCKPCB);
2646 while ((inp = inp_next(&inpi)) != NULL)
2651 inp_inpcbtosocket(struct inpcb *inp)
2654 INP_WLOCK_ASSERT(inp);
2655 return (inp->inp_socket);
2659 inp_inpcbtotcpcb(struct inpcb *inp)
2662 INP_WLOCK_ASSERT(inp);
2663 return ((struct tcpcb *)inp->inp_ppcb);
2667 inp_ip_tos_get(const struct inpcb *inp)
2670 return (inp->inp_ip_tos);
2674 inp_ip_tos_set(struct inpcb *inp, int val)
2677 inp->inp_ip_tos = val;
2681 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2682 uint32_t *faddr, uint16_t *fp)
2685 INP_LOCK_ASSERT(inp);
2686 *laddr = inp->inp_laddr.s_addr;
2687 *faddr = inp->inp_faddr.s_addr;
2688 *lp = inp->inp_lport;
2689 *fp = inp->inp_fport;
2693 so_sotoinpcb(struct socket *so)
2696 return (sotoinpcb(so));
2700 * Create an external-format (``xinpcb'') structure using the information in
2701 * the kernel-format in_pcb structure pointed to by inp. This is done to
2702 * reduce the spew of irrelevant information over this interface, to isolate
2703 * user code from changes in the kernel structure, and potentially to provide
2704 * information-hiding if we decide that some of this information should be
2705 * hidden from users.
2708 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2711 bzero(xi, sizeof(*xi));
2712 xi->xi_len = sizeof(struct xinpcb);
2713 if (inp->inp_socket)
2714 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2715 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2716 xi->inp_gencnt = inp->inp_gencnt;
2717 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2718 xi->inp_flow = inp->inp_flow;
2719 xi->inp_flowid = inp->inp_flowid;
2720 xi->inp_flowtype = inp->inp_flowtype;
2721 xi->inp_flags = inp->inp_flags;
2722 xi->inp_flags2 = inp->inp_flags2;
2723 xi->in6p_cksum = inp->in6p_cksum;
2724 xi->in6p_hops = inp->in6p_hops;
2725 xi->inp_ip_tos = inp->inp_ip_tos;
2726 xi->inp_vflag = inp->inp_vflag;
2727 xi->inp_ip_ttl = inp->inp_ip_ttl;
2728 xi->inp_ip_p = inp->inp_ip_p;
2729 xi->inp_ip_minttl = inp->inp_ip_minttl;
2733 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2734 int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2736 struct sockopt sopt;
2737 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2738 INPLOOKUP_WLOCKPCB);
2740 struct sockopt_parameters *params;
2745 if (req->oldptr != NULL || req->oldlen != 0)
2747 if (req->newptr == NULL)
2749 if (req->newlen > sizeof(buf))
2751 error = SYSCTL_IN(req, buf, req->newlen);
2754 if (req->newlen < sizeof(struct sockopt_parameters))
2756 params = (struct sockopt_parameters *)buf;
2757 sopt.sopt_level = params->sop_level;
2758 sopt.sopt_name = params->sop_optname;
2759 sopt.sopt_dir = SOPT_SET;
2760 sopt.sopt_val = params->sop_optval;
2761 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
2762 sopt.sopt_td = NULL;
2764 if (params->sop_inc.inc_flags & INC_ISIPV6) {
2765 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr))
2766 params->sop_inc.inc6_laddr.s6_addr16[1] =
2767 htons(params->sop_inc.inc6_zoneid & 0xffff);
2768 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr))
2769 params->sop_inc.inc6_faddr.s6_addr16[1] =
2770 htons(params->sop_inc.inc6_zoneid & 0xffff);
2773 if (params->sop_inc.inc_lport != htons(0)) {
2774 if (params->sop_inc.inc_fport == htons(0))
2775 inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
2776 pcbinfo->ipi_hashmask);
2779 if (params->sop_inc.inc_flags & INC_ISIPV6)
2780 inpi.hash = INP6_PCBHASH(
2781 ¶ms->sop_inc.inc6_faddr,
2782 params->sop_inc.inc_lport,
2783 params->sop_inc.inc_fport,
2784 pcbinfo->ipi_hashmask);
2787 inpi.hash = INP_PCBHASH(
2788 ¶ms->sop_inc.inc_faddr,
2789 params->sop_inc.inc_lport,
2790 params->sop_inc.inc_fport,
2791 pcbinfo->ipi_hashmask);
2793 while ((inp = inp_next(&inpi)) != NULL)
2794 if (inp->inp_gencnt == params->sop_id) {
2795 if (inp->inp_flags & INP_DROPPED) {
2797 return (ECONNRESET);
2799 so = inp->inp_socket;
2800 KASSERT(so != NULL, ("inp_socket == NULL"));
2802 error = (*ctloutput_set)(inp, &sopt);
2813 db_print_indent(int indent)
2817 for (i = 0; i < indent; i++)
2822 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2824 char faddr_str[48], laddr_str[48];
2826 db_print_indent(indent);
2827 db_printf("%s at %p\n", name, inc);
2832 if (inc->inc_flags & INC_ISIPV6) {
2834 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2835 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2840 inet_ntoa_r(inc->inc_laddr, laddr_str);
2841 inet_ntoa_r(inc->inc_faddr, faddr_str);
2843 db_print_indent(indent);
2844 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2845 ntohs(inc->inc_lport));
2846 db_print_indent(indent);
2847 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2848 ntohs(inc->inc_fport));
2852 db_print_inpflags(int inp_flags)
2857 if (inp_flags & INP_RECVOPTS) {
2858 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2861 if (inp_flags & INP_RECVRETOPTS) {
2862 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2865 if (inp_flags & INP_RECVDSTADDR) {
2866 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2869 if (inp_flags & INP_ORIGDSTADDR) {
2870 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2873 if (inp_flags & INP_HDRINCL) {
2874 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2877 if (inp_flags & INP_HIGHPORT) {
2878 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2881 if (inp_flags & INP_LOWPORT) {
2882 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2885 if (inp_flags & INP_ANONPORT) {
2886 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2889 if (inp_flags & INP_RECVIF) {
2890 db_printf("%sINP_RECVIF", comma ? ", " : "");
2893 if (inp_flags & INP_MTUDISC) {
2894 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2897 if (inp_flags & INP_RECVTTL) {
2898 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2901 if (inp_flags & INP_DONTFRAG) {
2902 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2905 if (inp_flags & INP_RECVTOS) {
2906 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2909 if (inp_flags & IN6P_IPV6_V6ONLY) {
2910 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2913 if (inp_flags & IN6P_PKTINFO) {
2914 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2917 if (inp_flags & IN6P_HOPLIMIT) {
2918 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2921 if (inp_flags & IN6P_HOPOPTS) {
2922 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2925 if (inp_flags & IN6P_DSTOPTS) {
2926 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2929 if (inp_flags & IN6P_RTHDR) {
2930 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2933 if (inp_flags & IN6P_RTHDRDSTOPTS) {
2934 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2937 if (inp_flags & IN6P_TCLASS) {
2938 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2941 if (inp_flags & IN6P_AUTOFLOWLABEL) {
2942 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2945 if (inp_flags & INP_ONESBCAST) {
2946 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2949 if (inp_flags & INP_DROPPED) {
2950 db_printf("%sINP_DROPPED", comma ? ", " : "");
2953 if (inp_flags & INP_SOCKREF) {
2954 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2957 if (inp_flags & IN6P_RFC2292) {
2958 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2961 if (inp_flags & IN6P_MTU) {
2962 db_printf("IN6P_MTU%s", comma ? ", " : "");
2968 db_print_inpvflag(u_char inp_vflag)
2973 if (inp_vflag & INP_IPV4) {
2974 db_printf("%sINP_IPV4", comma ? ", " : "");
2977 if (inp_vflag & INP_IPV6) {
2978 db_printf("%sINP_IPV6", comma ? ", " : "");
2981 if (inp_vflag & INP_IPV6PROTO) {
2982 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2988 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2991 db_print_indent(indent);
2992 db_printf("%s at %p\n", name, inp);
2996 db_print_indent(indent);
2997 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2999 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3001 db_print_indent(indent);
3002 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3003 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3005 db_print_indent(indent);
3006 db_printf("inp_label: %p inp_flags: 0x%x (",
3007 inp->inp_label, inp->inp_flags);
3008 db_print_inpflags(inp->inp_flags);
3011 db_print_indent(indent);
3012 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3014 db_print_inpvflag(inp->inp_vflag);
3017 db_print_indent(indent);
3018 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3019 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3021 db_print_indent(indent);
3023 if (inp->inp_vflag & INP_IPV6) {
3024 db_printf("in6p_options: %p in6p_outputopts: %p "
3025 "in6p_moptions: %p\n", inp->in6p_options,
3026 inp->in6p_outputopts, inp->in6p_moptions);
3027 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3028 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3033 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3034 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3035 inp->inp_options, inp->inp_moptions);
3038 db_print_indent(indent);
3039 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3040 (uintmax_t)inp->inp_gencnt);
3043 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3048 db_printf("usage: show inpcb <addr>\n");
3051 inp = (struct inpcb *)addr;
3053 db_print_inpcb(inp, "inpcb", 0);
3059 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3063 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3065 union if_snd_tag_modify_params params = {
3066 .rate_limit.max_rate = max_pacing_rate,
3067 .rate_limit.flags = M_NOWAIT,
3069 struct m_snd_tag *mst;
3072 mst = inp->inp_snd_tag;
3076 if (mst->sw->snd_tag_modify == NULL) {
3079 error = mst->sw->snd_tag_modify(mst, ¶ms);
3085 * Query existing TX rate limit based on the existing
3086 * "inp->inp_snd_tag", if any.
3089 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3091 union if_snd_tag_query_params params = { };
3092 struct m_snd_tag *mst;
3095 mst = inp->inp_snd_tag;
3099 if (mst->sw->snd_tag_query == NULL) {
3102 error = mst->sw->snd_tag_query(mst, ¶ms);
3103 if (error == 0 && p_max_pacing_rate != NULL)
3104 *p_max_pacing_rate = params.rate_limit.max_rate;
3110 * Query existing TX queue level based on the existing
3111 * "inp->inp_snd_tag", if any.
3114 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3116 union if_snd_tag_query_params params = { };
3117 struct m_snd_tag *mst;
3120 mst = inp->inp_snd_tag;
3124 if (mst->sw->snd_tag_query == NULL)
3125 return (EOPNOTSUPP);
3127 error = mst->sw->snd_tag_query(mst, ¶ms);
3128 if (error == 0 && p_txqueue_level != NULL)
3129 *p_txqueue_level = params.rate_limit.queue_level;
3134 * Allocate a new TX rate limit send tag from the network interface
3135 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3138 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3139 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3142 union if_snd_tag_alloc_params params = {
3143 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3144 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3145 .rate_limit.hdr.flowid = flowid,
3146 .rate_limit.hdr.flowtype = flowtype,
3147 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3148 .rate_limit.max_rate = max_pacing_rate,
3149 .rate_limit.flags = M_NOWAIT,
3153 INP_WLOCK_ASSERT(inp);
3156 * If there is already a send tag, or the INP is being torn
3157 * down, allocating a new send tag is not allowed. Else send
3160 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0)
3163 error = m_snd_tag_alloc(ifp, ¶ms, st);
3166 counter_u64_add(rate_limit_set_ok, 1);
3167 counter_u64_add(rate_limit_active, 1);
3168 } else if (error != EOPNOTSUPP)
3169 counter_u64_add(rate_limit_alloc_fail, 1);
3175 in_pcbdetach_tag(struct m_snd_tag *mst)
3178 m_snd_tag_rele(mst);
3180 counter_u64_add(rate_limit_active, -1);
3185 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3189 in_pcbdetach_txrtlmt(struct inpcb *inp)
3191 struct m_snd_tag *mst;
3193 INP_WLOCK_ASSERT(inp);
3195 mst = inp->inp_snd_tag;
3196 inp->inp_snd_tag = NULL;
3201 m_snd_tag_rele(mst);
3203 counter_u64_add(rate_limit_active, -1);
3208 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3213 * If the existing send tag is for the wrong interface due to
3214 * a route change, first drop the existing tag. Set the
3215 * CHANGED flag so that we will keep trying to allocate a new
3216 * tag if we fail to allocate one this time.
3218 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3219 in_pcbdetach_txrtlmt(inp);
3220 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3224 * NOTE: When attaching to a network interface a reference is
3225 * made to ensure the network interface doesn't go away until
3226 * all ratelimit connections are gone. The network interface
3227 * pointers compared below represent valid network interfaces,
3228 * except when comparing towards NULL.
3230 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3232 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3233 if (inp->inp_snd_tag != NULL)
3234 in_pcbdetach_txrtlmt(inp);
3236 } else if (inp->inp_snd_tag == NULL) {
3238 * In order to utilize packet pacing with RSS, we need
3239 * to wait until there is a valid RSS hash before we
3242 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3245 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3246 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3249 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3251 if (error == 0 || error == EOPNOTSUPP)
3252 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3258 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3259 * is set in the fast path and will attach/detach/modify the TX rate
3260 * limit send tag based on the socket's so_max_pacing_rate value.
3263 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3265 struct socket *socket;
3266 uint32_t max_pacing_rate;
3272 socket = inp->inp_socket;
3276 if (!INP_WLOCKED(inp)) {
3278 * NOTE: If the write locking fails, we need to bail
3279 * out and use the non-ratelimited ring for the
3280 * transmit until there is a new chance to get the
3283 if (!INP_TRY_UPGRADE(inp))
3291 * NOTE: The so_max_pacing_rate value is read unlocked,
3292 * because atomic updates are not required since the variable
3293 * is checked at every mbuf we send. It is assumed that the
3294 * variable read itself will be atomic.
3296 max_pacing_rate = socket->so_max_pacing_rate;
3298 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3305 * Track route changes for TX rate limiting.
3308 in_pcboutput_eagain(struct inpcb *inp)
3315 if (inp->inp_snd_tag == NULL)
3318 if (!INP_WLOCKED(inp)) {
3320 * NOTE: If the write locking fails, we need to bail
3321 * out and use the non-ratelimited ring for the
3322 * transmit until there is a new chance to get the
3325 if (!INP_TRY_UPGRADE(inp))
3332 /* detach rate limiting */
3333 in_pcbdetach_txrtlmt(inp);
3335 /* make sure new mbuf send tag allocation is made */
3336 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3346 rate_limit_new = counter_u64_alloc(M_WAITOK);
3347 rate_limit_chg = counter_u64_alloc(M_WAITOK);
3348 rate_limit_active = counter_u64_alloc(M_WAITOK);
3349 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3350 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3353 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3355 #endif /* RATELIMIT */