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");
225 #endif /* RATELIMIT */
229 VNET_DEFINE(uint32_t, in_pcbhashseed);
231 in_pcbhashseed_init(void)
234 V_in_pcbhashseed = arc4random();
236 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
237 in_pcbhashseed_init, 0);
239 static void in_pcbremhash(struct inpcb *);
242 * in_pcb.c: manage the Protocol Control Blocks.
244 * NOTE: It is assumed that most of these functions will be called with
245 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
246 * functions often modify hash chains or addresses in pcbs.
249 static struct inpcblbgroup *
250 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred,
251 u_char vflag, uint16_t port, const union in_dependaddr *addr, int size,
254 struct inpcblbgroup *grp;
257 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
258 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
261 grp->il_cred = crhold(cred);
262 grp->il_vflag = vflag;
263 grp->il_lport = port;
264 grp->il_numa_domain = numa_domain;
265 grp->il_dependladdr = *addr;
266 grp->il_inpsiz = size;
267 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
272 in_pcblbgroup_free_deferred(epoch_context_t ctx)
274 struct inpcblbgroup *grp;
276 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
277 crfree(grp->il_cred);
282 in_pcblbgroup_free(struct inpcblbgroup *grp)
285 CK_LIST_REMOVE(grp, il_list);
286 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
289 static struct inpcblbgroup *
290 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
291 struct inpcblbgroup *old_grp, int size)
293 struct inpcblbgroup *grp;
296 grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag,
297 old_grp->il_lport, &old_grp->il_dependladdr, size,
298 old_grp->il_numa_domain);
302 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
303 ("invalid new local group size %d and old local group count %d",
304 grp->il_inpsiz, old_grp->il_inpcnt));
306 for (i = 0; i < old_grp->il_inpcnt; ++i)
307 grp->il_inp[i] = old_grp->il_inp[i];
308 grp->il_inpcnt = old_grp->il_inpcnt;
309 in_pcblbgroup_free(old_grp);
314 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
315 * and shrink group if possible.
318 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
321 struct inpcblbgroup *grp, *new_grp;
324 for (; i + 1 < grp->il_inpcnt; ++i)
325 grp->il_inp[i] = grp->il_inp[i + 1];
328 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
329 grp->il_inpcnt <= grp->il_inpsiz / 4) {
330 /* Shrink this group. */
331 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
338 * Add PCB to load balance group for SO_REUSEPORT_LB option.
341 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain)
343 const static struct timeval interval = { 60, 0 };
344 static struct timeval lastprint;
345 struct inpcbinfo *pcbinfo;
346 struct inpcblbgrouphead *hdr;
347 struct inpcblbgroup *grp;
350 pcbinfo = inp->inp_pcbinfo;
352 INP_WLOCK_ASSERT(inp);
353 INP_HASH_WLOCK_ASSERT(pcbinfo);
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_cred->cr_prison == inp->inp_cred->cr_prison &&
370 grp->il_vflag == inp->inp_vflag &&
371 grp->il_lport == inp->inp_lport &&
372 grp->il_numa_domain == numa_domain &&
373 memcmp(&grp->il_dependladdr,
374 &inp->inp_inc.inc_ie.ie_dependladdr,
375 sizeof(grp->il_dependladdr)) == 0) {
380 /* Create new load balance group. */
381 grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag,
382 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
383 INPCBLBGROUP_SIZMIN, numa_domain);
386 } else if (grp->il_inpcnt == grp->il_inpsiz) {
387 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
388 if (ratecheck(&lastprint, &interval))
389 printf("lb group port %d, limit reached\n",
390 ntohs(grp->il_lport));
394 /* Expand this local group. */
395 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
400 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
401 ("invalid local group size %d and count %d", grp->il_inpsiz,
404 grp->il_inp[grp->il_inpcnt] = inp;
410 * Remove PCB from load balance group.
413 in_pcbremlbgrouphash(struct inpcb *inp)
415 struct inpcbinfo *pcbinfo;
416 struct inpcblbgrouphead *hdr;
417 struct inpcblbgroup *grp;
420 pcbinfo = inp->inp_pcbinfo;
422 INP_WLOCK_ASSERT(inp);
423 INP_HASH_WLOCK_ASSERT(pcbinfo);
425 hdr = &pcbinfo->ipi_lbgrouphashbase[
426 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
427 CK_LIST_FOREACH(grp, hdr, il_list) {
428 for (i = 0; i < grp->il_inpcnt; ++i) {
429 if (grp->il_inp[i] != inp)
432 if (grp->il_inpcnt == 1) {
433 /* We are the last, free this local group. */
434 in_pcblbgroup_free(grp);
436 /* Pull up inpcbs, shrink group if possible. */
437 in_pcblbgroup_reorder(hdr, &grp, i);
445 in_pcblbgroup_numa(struct inpcb *inp, int arg)
447 struct inpcbinfo *pcbinfo;
448 struct inpcblbgrouphead *hdr;
449 struct inpcblbgroup *grp;
454 case TCP_REUSPORT_LB_NUMA_NODOM:
455 numa_domain = M_NODOM;
457 case TCP_REUSPORT_LB_NUMA_CURDOM:
458 numa_domain = PCPU_GET(domain);
461 if (arg < 0 || arg >= vm_ndomains)
467 pcbinfo = inp->inp_pcbinfo;
468 INP_WLOCK_ASSERT(inp);
469 INP_HASH_WLOCK(pcbinfo);
470 hdr = &pcbinfo->ipi_lbgrouphashbase[
471 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
472 CK_LIST_FOREACH(grp, hdr, il_list) {
473 for (i = 0; i < grp->il_inpcnt; ++i) {
474 if (grp->il_inp[i] != inp)
477 if (grp->il_numa_domain == numa_domain) {
478 goto abort_with_hash_wlock;
481 /* Remove it from the old group. */
482 in_pcbremlbgrouphash(inp);
484 /* Add it to the new group based on numa domain. */
485 in_pcbinslbgrouphash(inp, numa_domain);
486 goto abort_with_hash_wlock;
490 abort_with_hash_wlock:
491 INP_HASH_WUNLOCK(pcbinfo);
495 /* Make sure it is safe to use hashinit(9) on CK_LIST. */
496 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb)));
499 * Initialize an inpcbinfo - a per-VNET instance of connections db.
502 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor,
503 u_int hash_nelements, u_int porthash_nelements)
506 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF);
507 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name,
510 pcbinfo->ipi_vnet = curvnet;
512 CK_LIST_INIT(&pcbinfo->ipi_listhead);
513 pcbinfo->ipi_count = 0;
514 pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB,
515 &pcbinfo->ipi_hashmask);
516 pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB,
517 &pcbinfo->ipi_hashmask);
518 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
519 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
520 &pcbinfo->ipi_porthashmask);
521 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
522 &pcbinfo->ipi_lbgrouphashmask);
523 pcbinfo->ipi_zone = pcbstor->ips_zone;
524 pcbinfo->ipi_portzone = pcbstor->ips_portzone;
525 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone);
529 * Destroy an inpcbinfo.
532 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
535 KASSERT(pcbinfo->ipi_count == 0,
536 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
538 hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask);
539 hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask);
540 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
541 pcbinfo->ipi_porthashmask);
542 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
543 pcbinfo->ipi_lbgrouphashmask);
544 mtx_destroy(&pcbinfo->ipi_hash_lock);
545 mtx_destroy(&pcbinfo->ipi_lock);
549 * Initialize a pcbstorage - per protocol zones to allocate inpcbs.
551 static void inpcb_fini(void *, int);
553 in_pcbstorage_init(void *arg)
555 struct inpcbstorage *pcbstor = arg;
557 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name,
558 pcbstor->ips_size, NULL, NULL, pcbstor->ips_pcbinit,
559 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR);
560 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name,
561 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
562 uma_zone_set_smr(pcbstor->ips_portzone,
563 uma_zone_get_smr(pcbstor->ips_zone));
567 * Destroy a pcbstorage - used by unloadable protocols.
570 in_pcbstorage_destroy(void *arg)
572 struct inpcbstorage *pcbstor = arg;
574 uma_zdestroy(pcbstor->ips_zone);
575 uma_zdestroy(pcbstor->ips_portzone);
579 * Allocate a PCB and associate it with the socket.
580 * On success return with the PCB locked.
583 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
586 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
590 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT);
593 bzero(&inp->inp_start_zero, inp_zero_size);
595 inp->inp_numa_domain = M_NODOM;
597 inp->inp_pcbinfo = pcbinfo;
598 inp->inp_socket = so;
599 inp->inp_cred = crhold(so->so_cred);
600 inp->inp_inc.inc_fibnum = so->so_fibnum;
602 error = mac_inpcb_init(inp, M_NOWAIT);
605 mac_inpcb_create(so, inp);
607 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
608 error = ipsec_init_pcbpolicy(inp);
611 mac_inpcb_destroy(inp);
617 if (INP_SOCKAF(so) == AF_INET6) {
618 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6;
620 inp->inp_flags |= IN6P_IPV6_V6ONLY;
623 inp->inp_vflag |= INP_IPV4;
625 if (V_ip6_auto_flowlabel)
626 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
627 inp->in6p_hops = -1; /* use kernel default */
630 #if defined(INET) && defined(INET6)
634 inp->inp_vflag |= INP_IPV4;
636 inp->inp_smr = SMR_SEQ_INVALID;
639 * Routes in inpcb's can cache L2 as well; they are guaranteed
642 inp->inp_route.ro_flags = RT_LLE_CACHE;
643 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */
645 INP_INFO_WLOCK(pcbinfo);
646 pcbinfo->ipi_count++;
647 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
648 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list);
649 INP_INFO_WUNLOCK(pcbinfo);
654 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
656 uma_zfree_smr(pcbinfo->ipi_zone, inp);
663 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred)
667 KASSERT(sin == NULL || sin->sin_family == AF_INET,
668 ("%s: invalid address family for %p", __func__, sin));
669 KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in),
670 ("%s: invalid address length for %p", __func__, sin));
671 INP_WLOCK_ASSERT(inp);
672 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
674 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
676 anonport = sin == NULL || sin->sin_port == 0;
677 error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr,
678 &inp->inp_lport, cred);
681 if (in_pcbinshash(inp) != 0) {
682 inp->inp_laddr.s_addr = INADDR_ANY;
687 inp->inp_flags |= INP_ANONPORT;
692 #if defined(INET) || defined(INET6)
694 * Assign a local port like in_pcb_lport(), but also used with connect()
695 * and a foreign address and port. If fsa is non-NULL, choose a local port
696 * that is unused with those, otherwise one that is completely unused.
697 * lsa can be NULL for IPv6.
700 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
701 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
703 struct inpcbinfo *pcbinfo;
704 struct inpcb *tmpinp;
705 unsigned short *lastport;
707 u_short aux, first, last, lport;
709 struct in_addr laddr, faddr;
712 struct in6_addr *laddr6, *faddr6;
715 pcbinfo = inp->inp_pcbinfo;
718 * Because no actual state changes occur here, a global write lock on
719 * the pcbinfo isn't required.
721 INP_LOCK_ASSERT(inp);
722 INP_HASH_LOCK_ASSERT(pcbinfo);
724 if (inp->inp_flags & INP_HIGHPORT) {
725 first = V_ipport_hifirstauto; /* sysctl */
726 last = V_ipport_hilastauto;
727 lastport = &pcbinfo->ipi_lasthi;
728 } else if (inp->inp_flags & INP_LOWPORT) {
729 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
732 first = V_ipport_lowfirstauto; /* 1023 */
733 last = V_ipport_lowlastauto; /* 600 */
734 lastport = &pcbinfo->ipi_lastlow;
736 first = V_ipport_firstauto; /* sysctl */
737 last = V_ipport_lastauto;
738 lastport = &pcbinfo->ipi_lastport;
742 * Instead of having two loops further down counting up or down
743 * make sure that first is always <= last and go with only one
744 * code path implementing all logic.
753 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */
754 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
756 laddr = ((struct sockaddr_in *)lsa)->sin_addr;
758 faddr = ((struct sockaddr_in *)fsa)->sin_addr;
763 if ((inp->inp_vflag & INP_IPV6) != 0) {
765 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
767 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
774 if (V_ipport_randomized)
775 *lastport = first + (arc4random() % (last - first));
777 count = last - first;
780 if (count-- < 0) /* completely used? */
781 return (EADDRNOTAVAIL);
783 if (*lastport < first || *lastport > last)
785 lport = htons(*lastport);
789 if (lsa->sa_family == AF_INET) {
790 tmpinp = in_pcblookup_hash_locked(pcbinfo,
791 faddr, fport, laddr, lport, lookupflags,
796 if (lsa->sa_family == AF_INET6) {
797 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
798 faddr6, fport, laddr6, lport, lookupflags,
804 if ((inp->inp_vflag & INP_IPV6) != 0) {
805 tmpinp = in6_pcblookup_local(pcbinfo,
806 &inp->in6p_laddr, lport, lookupflags, cred);
808 if (tmpinp == NULL &&
809 (inp->inp_vflag & INP_IPV4))
810 tmpinp = in_pcblookup_local(pcbinfo,
811 laddr, lport, lookupflags, cred);
815 #if defined(INET) && defined(INET6)
819 tmpinp = in_pcblookup_local(pcbinfo, laddr,
820 lport, lookupflags, cred);
823 } while (tmpinp != NULL);
831 * Select a local port (number) to use.
834 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
835 struct ucred *cred, int lookupflags)
837 struct sockaddr_in laddr;
840 bzero(&laddr, sizeof(laddr));
841 laddr.sin_family = AF_INET;
842 laddr.sin_addr = *laddrp;
844 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
845 NULL, lportp, NULL, 0, cred, lookupflags));
849 * Return cached socket options.
852 inp_so_options(const struct inpcb *inp)
858 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
859 so_options |= SO_REUSEPORT_LB;
860 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
861 so_options |= SO_REUSEPORT;
862 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
863 so_options |= SO_REUSEADDR;
866 #endif /* INET || INET6 */
870 * Set up a bind operation on a PCB, performing port allocation
871 * as required, but do not actually modify the PCB. Callers can
872 * either complete the bind by setting inp_laddr/inp_lport and
873 * calling in_pcbinshash(), or they can just use the resulting
874 * port and address to authorise the sending of a once-off packet.
876 * On error, the values of *laddrp and *lportp are not changed.
879 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp,
880 u_short *lportp, struct ucred *cred)
882 struct socket *so = inp->inp_socket;
883 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
884 struct in_addr laddr;
886 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
890 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
891 * so that we don't have to add to the (already messy) code below.
893 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
896 * No state changes, so read locks are sufficient here.
898 INP_LOCK_ASSERT(inp);
899 INP_HASH_LOCK_ASSERT(pcbinfo);
901 laddr.s_addr = *laddrp;
902 if (sin != NULL && laddr.s_addr != INADDR_ANY)
904 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
905 lookupflags = INPLOOKUP_WILDCARD;
907 if ((error = prison_local_ip4(cred, &laddr)) != 0)
910 KASSERT(sin->sin_family == AF_INET,
911 ("%s: invalid family for address %p", __func__, sin));
912 KASSERT(sin->sin_len == sizeof(*sin),
913 ("%s: invalid length for address %p", __func__, sin));
915 error = prison_local_ip4(cred, &sin->sin_addr);
918 if (sin->sin_port != *lportp) {
919 /* Don't allow the port to change. */
922 lport = sin->sin_port;
924 /* NB: lport is left as 0 if the port isn't being changed. */
925 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
927 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
928 * allow complete duplication of binding if
929 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
930 * and a multicast address is bound on both
931 * new and duplicated sockets.
933 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
934 reuseport = SO_REUSEADDR|SO_REUSEPORT;
936 * XXX: How to deal with SO_REUSEPORT_LB here?
937 * Treat same as SO_REUSEPORT for now.
939 if ((so->so_options &
940 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
941 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
942 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
943 sin->sin_port = 0; /* yech... */
944 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
946 * Is the address a local IP address?
947 * If INP_BINDANY is set, then the socket may be bound
948 * to any endpoint address, local or not.
950 if ((inp->inp_flags & INP_BINDANY) == 0 &&
951 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
952 return (EADDRNOTAVAIL);
954 laddr = sin->sin_addr;
959 if (ntohs(lport) <= V_ipport_reservedhigh &&
960 ntohs(lport) >= V_ipport_reservedlow &&
961 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
963 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
964 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
965 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
966 lport, INPLOOKUP_WILDCARD, cred);
969 * This entire block sorely needs a rewrite.
972 (so->so_type != SOCK_STREAM ||
973 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
974 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
975 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
976 (t->inp_flags2 & INP_REUSEPORT) ||
977 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
978 (inp->inp_cred->cr_uid !=
979 t->inp_cred->cr_uid))
982 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
983 lport, lookupflags, cred);
984 if (t != NULL && (reuseport & inp_so_options(t)) == 0 &&
985 (reuseport_lb & inp_so_options(t)) == 0) {
987 if (ntohl(sin->sin_addr.s_addr) !=
989 ntohl(t->inp_laddr.s_addr) !=
991 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
992 (t->inp_vflag & INP_IPV6PROTO) == 0)
1001 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1005 *laddrp = laddr.s_addr;
1011 * Connect from a socket to a specified address.
1012 * Both address and port must be specified in argument sin.
1013 * If don't have a local address for this socket yet,
1017 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred,
1018 bool rehash __unused)
1020 u_short lport, fport;
1021 in_addr_t laddr, faddr;
1022 int anonport, error;
1024 INP_WLOCK_ASSERT(inp);
1025 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1026 KASSERT(in_nullhost(inp->inp_faddr),
1027 ("%s: inp is already connected", __func__));
1029 lport = inp->inp_lport;
1030 laddr = inp->inp_laddr.s_addr;
1031 anonport = (lport == 0);
1032 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport,
1037 inp->inp_faddr.s_addr = faddr;
1038 inp->inp_fport = fport;
1040 /* Do the initial binding of the local address if required. */
1041 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1042 inp->inp_lport = lport;
1043 inp->inp_laddr.s_addr = laddr;
1044 if (in_pcbinshash(inp) != 0) {
1045 inp->inp_laddr.s_addr = inp->inp_faddr.s_addr =
1047 inp->inp_lport = inp->inp_fport = 0;
1051 inp->inp_lport = lport;
1052 inp->inp_laddr.s_addr = laddr;
1053 if ((inp->inp_flags & INP_INHASHLIST) != 0)
1060 inp->inp_flags |= INP_ANONPORT;
1065 * Do proper source address selection on an unbound socket in case
1066 * of connect. Take jails into account as well.
1069 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1073 struct sockaddr *sa;
1074 struct sockaddr_in *sin, dst;
1075 struct nhop_object *nh;
1079 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1082 * Bypass source address selection and use the primary jail IP
1085 if (!prison_saddrsel_ip4(cred, laddr))
1091 bzero(&dst, sizeof(dst));
1093 sin->sin_family = AF_INET;
1094 sin->sin_len = sizeof(struct sockaddr_in);
1095 sin->sin_addr.s_addr = faddr->s_addr;
1098 * If route is known our src addr is taken from the i/f,
1101 * Find out route to destination.
1103 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1104 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1108 * If we found a route, use the address corresponding to
1109 * the outgoing interface.
1111 * Otherwise assume faddr is reachable on a directly connected
1112 * network and try to find a corresponding interface to take
1113 * the source address from.
1115 if (nh == NULL || nh->nh_ifp == NULL) {
1116 struct in_ifaddr *ia;
1119 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1120 inp->inp_socket->so_fibnum));
1122 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1123 inp->inp_socket->so_fibnum));
1126 error = ENETUNREACH;
1130 if (!prison_flag(cred, PR_IP4)) {
1131 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1137 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1139 if (sa->sa_family != AF_INET)
1141 sin = (struct sockaddr_in *)sa;
1142 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1143 ia = (struct in_ifaddr *)ifa;
1148 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1152 /* 3. As a last resort return the 'default' jail address. */
1153 error = prison_get_ip4(cred, laddr);
1158 * If the outgoing interface on the route found is not
1159 * a loopback interface, use the address from that interface.
1160 * In case of jails do those three steps:
1161 * 1. check if the interface address belongs to the jail. If so use it.
1162 * 2. check if we have any address on the outgoing interface
1163 * belonging to this jail. If so use it.
1164 * 3. as a last resort return the 'default' jail address.
1166 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1167 struct in_ifaddr *ia;
1170 /* If not jailed, use the default returned. */
1171 if (!prison_flag(cred, PR_IP4)) {
1172 ia = (struct in_ifaddr *)nh->nh_ifa;
1173 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1178 /* 1. Check if the iface address belongs to the jail. */
1179 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1180 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1181 ia = (struct in_ifaddr *)nh->nh_ifa;
1182 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1187 * 2. Check if we have any address on the outgoing interface
1188 * belonging to this jail.
1192 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1194 if (sa->sa_family != AF_INET)
1196 sin = (struct sockaddr_in *)sa;
1197 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1198 ia = (struct in_ifaddr *)ifa;
1203 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1207 /* 3. As a last resort return the 'default' jail address. */
1208 error = prison_get_ip4(cred, laddr);
1213 * The outgoing interface is marked with 'loopback net', so a route
1214 * to ourselves is here.
1215 * Try to find the interface of the destination address and then
1216 * take the address from there. That interface is not necessarily
1217 * a loopback interface.
1218 * In case of jails, check that it is an address of the jail
1219 * and if we cannot find, fall back to the 'default' jail address.
1221 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1222 struct in_ifaddr *ia;
1224 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1225 inp->inp_socket->so_fibnum));
1227 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1228 inp->inp_socket->so_fibnum));
1230 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1232 if (!prison_flag(cred, PR_IP4)) {
1234 error = ENETUNREACH;
1237 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1247 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1249 if (sa->sa_family != AF_INET)
1251 sin = (struct sockaddr_in *)sa;
1252 if (prison_check_ip4(cred,
1253 &sin->sin_addr) == 0) {
1254 ia = (struct in_ifaddr *)ifa;
1259 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1264 /* 3. As a last resort return the 'default' jail address. */
1265 error = prison_get_ip4(cred, laddr);
1270 if (error == 0 && laddr->s_addr == INADDR_ANY)
1271 return (EHOSTUNREACH);
1276 * Set up for a connect from a socket to the specified address.
1277 * On entry, *laddrp and *lportp should contain the current local
1278 * address and port for the PCB; these are updated to the values
1279 * that should be placed in inp_laddr and inp_lport to complete
1282 * On success, *faddrp and *fportp will be set to the remote address
1283 * and port. These are not updated in the error case.
1286 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin,
1287 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1290 struct in_ifaddr *ia;
1291 struct in_addr laddr, faddr;
1292 u_short lport, fport;
1295 KASSERT(sin->sin_family == AF_INET,
1296 ("%s: invalid address family for %p", __func__, sin));
1297 KASSERT(sin->sin_len == sizeof(*sin),
1298 ("%s: invalid address length for %p", __func__, sin));
1301 * Because a global state change doesn't actually occur here, a read
1302 * lock is sufficient.
1305 INP_LOCK_ASSERT(inp);
1306 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1308 if (sin->sin_port == 0)
1309 return (EADDRNOTAVAIL);
1310 laddr.s_addr = *laddrp;
1312 faddr = sin->sin_addr;
1313 fport = sin->sin_port;
1315 if (CALC_FLOWID_OUTBOUND) {
1316 uint32_t hash_val, hash_type;
1318 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1319 inp->inp_socket->so_proto->pr_protocol, &hash_type);
1321 inp->inp_flowid = hash_val;
1322 inp->inp_flowtype = hash_type;
1325 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1327 * If the destination address is INADDR_ANY,
1328 * use the primary local address.
1329 * If the supplied address is INADDR_BROADCAST,
1330 * and the primary interface supports broadcast,
1331 * choose the broadcast address for that interface.
1333 if (faddr.s_addr == INADDR_ANY) {
1335 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1336 if ((error = prison_get_ip4(cred, &faddr)) != 0)
1338 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1339 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1341 faddr = satosin(&CK_STAILQ_FIRST(
1342 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1345 if (laddr.s_addr == INADDR_ANY) {
1346 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1348 * If the destination address is multicast and an outgoing
1349 * interface has been set as a multicast option, prefer the
1350 * address of that interface as our source address.
1352 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1353 inp->inp_moptions != NULL) {
1354 struct ip_moptions *imo;
1357 imo = inp->inp_moptions;
1358 if (imo->imo_multicast_ifp != NULL) {
1359 ifp = imo->imo_multicast_ifp;
1360 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1361 if (ia->ia_ifp == ifp &&
1362 prison_check_ip4(cred,
1363 &ia->ia_addr.sin_addr) == 0)
1367 error = EADDRNOTAVAIL;
1369 laddr = ia->ia_addr.sin_addr;
1379 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1380 fport, laddr, lport, 0, M_NODOM) != NULL)
1381 return (EADDRINUSE);
1383 struct sockaddr_in lsin, fsin;
1385 bzero(&lsin, sizeof(lsin));
1386 bzero(&fsin, sizeof(fsin));
1387 lsin.sin_family = AF_INET;
1388 lsin.sin_addr = laddr;
1389 fsin.sin_family = AF_INET;
1390 fsin.sin_addr = faddr;
1391 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1392 &lport, (struct sockaddr *)& fsin, fport, cred,
1393 INPLOOKUP_WILDCARD);
1397 *laddrp = laddr.s_addr;
1399 *faddrp = faddr.s_addr;
1405 in_pcbdisconnect(struct inpcb *inp)
1408 INP_WLOCK_ASSERT(inp);
1409 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1410 KASSERT(inp->inp_smr == SMR_SEQ_INVALID,
1411 ("%s: inp %p was already disconnected", __func__, inp));
1413 in_pcbremhash_locked(inp);
1415 /* See the comment in in_pcbinshash(). */
1416 inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr);
1417 inp->inp_laddr.s_addr = INADDR_ANY;
1418 inp->inp_faddr.s_addr = INADDR_ANY;
1424 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1425 * For most protocols, this will be invoked immediately prior to calling
1426 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1427 * socket, in which case in_pcbfree() is deferred.
1430 in_pcbdetach(struct inpcb *inp)
1433 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1436 if (inp->inp_snd_tag != NULL)
1437 in_pcbdetach_txrtlmt(inp);
1439 inp->inp_socket->so_pcb = NULL;
1440 inp->inp_socket = NULL;
1444 * inpcb hash lookups are protected by SMR section.
1446 * Once desired pcb has been found, switching from SMR section to a pcb
1447 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1448 * here because SMR is a critical section.
1449 * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1452 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1455 lock == INPLOOKUP_RLOCKPCB ?
1456 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1460 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1463 lock == INPLOOKUP_RLOCKPCB ?
1464 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1468 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1471 return (lock == INPLOOKUP_RLOCKPCB ?
1472 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1476 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags)
1479 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1480 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1482 if (__predict_true(inp_trylock(inp, lock))) {
1483 if (__predict_false(inp->inp_flags & ignflags)) {
1484 smr_exit(inp->inp_pcbinfo->ipi_smr);
1485 inp_unlock(inp, lock);
1488 smr_exit(inp->inp_pcbinfo->ipi_smr);
1492 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1493 smr_exit(inp->inp_pcbinfo->ipi_smr);
1494 inp_lock(inp, lock);
1495 if (__predict_false(in_pcbrele(inp, lock)))
1498 * inp acquired through refcount & lock for sure didn't went
1499 * through uma_zfree(). However, it may have already went
1500 * through in_pcbfree() and has another reference, that
1501 * prevented its release by our in_pcbrele().
1503 if (__predict_false(inp->inp_flags & ignflags)) {
1504 inp_unlock(inp, lock);
1509 smr_exit(inp->inp_pcbinfo->ipi_smr);
1515 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1519 * in_pcblookup() family of functions ignore not only freed entries,
1520 * that may be found due to lockless access to the hash, but dropped
1523 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED));
1527 * inp_next() - inpcb hash/list traversal iterator
1529 * Requires initialized struct inpcb_iterator for context.
1530 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1532 * - Iterator can have either write-lock or read-lock semantics, that can not
1534 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1535 * a single hash slot. Note: only rip_input() does the latter.
1536 * - Iterator may have optional bool matching function. The matching function
1537 * will be executed for each inpcb in the SMR context, so it can not acquire
1538 * locks and can safely access only immutable fields of inpcb.
1540 * A fresh initialized iterator has NULL inpcb in its context and that
1541 * means that inp_next() call would return the very first inpcb on the list
1542 * locked with desired semantic. In all following calls the context pointer
1543 * shall hold the current inpcb pointer. The KPI user is not supposed to
1544 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL
1545 * and write NULL to its context. After end of traversal an iterator can be
1548 * List traversals have the following features/constraints:
1549 * - New entries won't be seen, as they are always added to the head of a list.
1550 * - Removed entries won't stop traversal as long as they are not added to
1551 * a different list. This is violated by in_pcbrehash().
1553 #define II_LIST_FIRST(ipi, hash) \
1554 (((hash) == INP_ALL_LIST) ? \
1555 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \
1556 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)]))
1557 #define II_LIST_NEXT(inp, hash) \
1558 (((hash) == INP_ALL_LIST) ? \
1559 CK_LIST_NEXT((inp), inp_list) : \
1560 CK_LIST_NEXT((inp), inp_hash_exact))
1561 #define II_LOCK_ASSERT(inp, lock) \
1562 rw_assert(&(inp)->inp_lock, \
1563 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED )
1565 inp_next(struct inpcb_iterator *ii)
1567 const struct inpcbinfo *ipi = ii->ipi;
1568 inp_match_t *match = ii->match;
1569 void *ctx = ii->ctx;
1570 inp_lookup_t lock = ii->lock;
1571 int hash = ii->hash;
1574 if (ii->inp == NULL) { /* First call. */
1575 smr_enter(ipi->ipi_smr);
1576 /* This is unrolled CK_LIST_FOREACH(). */
1577 for (inp = II_LIST_FIRST(ipi, hash);
1579 inp = II_LIST_NEXT(inp, hash)) {
1580 if (match != NULL && (match)(inp, ctx) == false)
1582 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED)))
1585 smr_enter(ipi->ipi_smr);
1586 MPASS(inp != II_LIST_FIRST(ipi, hash));
1587 inp = II_LIST_FIRST(ipi, hash);
1594 smr_exit(ipi->ipi_smr);
1601 /* Not a first call. */
1602 smr_enter(ipi->ipi_smr);
1605 II_LOCK_ASSERT(inp, lock);
1607 inp = II_LIST_NEXT(inp, hash);
1609 smr_exit(ipi->ipi_smr);
1613 if (match != NULL && (match)(inp, ctx) == false)
1616 if (__predict_true(inp_trylock(inp, lock))) {
1617 if (__predict_false(inp->inp_flags & INP_FREED)) {
1619 * Entries are never inserted in middle of a list, thus
1620 * as long as we are in SMR, we can continue traversal.
1621 * Jump to 'restart' should yield in the same result,
1622 * but could produce unnecessary looping. Could this
1623 * looping be unbound?
1625 inp_unlock(inp, lock);
1628 smr_exit(ipi->ipi_smr);
1634 * Can't obtain lock immediately, thus going hard. Once we exit the
1635 * SMR section we can no longer jump to 'next', and our only stable
1636 * anchoring point is ii->inp, which we keep locked for this case, so
1637 * we jump to 'restart'.
1639 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1640 smr_exit(ipi->ipi_smr);
1641 inp_lock(inp, lock);
1642 if (__predict_false(in_pcbrele(inp, lock))) {
1643 smr_enter(ipi->ipi_smr);
1647 * See comment in inp_smr_lock().
1649 if (__predict_false(inp->inp_flags & INP_FREED)) {
1650 inp_unlock(inp, lock);
1651 smr_enter(ipi->ipi_smr);
1658 inp_unlock(ii->inp, lock);
1665 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1666 * stability of an inpcb pointer despite the inpcb lock being released or
1667 * SMR section exited.
1669 * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1672 in_pcbref(struct inpcb *inp)
1674 u_int old __diagused;
1676 old = refcount_acquire(&inp->inp_refcount);
1677 KASSERT(old > 0, ("%s: refcount 0", __func__));
1681 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1682 * freeing the pcb, if the reference was very last.
1685 in_pcbrele_rlocked(struct inpcb *inp)
1688 INP_RLOCK_ASSERT(inp);
1690 if (!refcount_release(&inp->inp_refcount))
1693 MPASS(inp->inp_flags & INP_FREED);
1694 MPASS(inp->inp_socket == NULL);
1695 crfree(inp->inp_cred);
1697 inp->inp_cred = NULL;
1700 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1705 in_pcbrele_wlocked(struct inpcb *inp)
1708 INP_WLOCK_ASSERT(inp);
1710 if (!refcount_release(&inp->inp_refcount))
1713 MPASS(inp->inp_flags & INP_FREED);
1714 MPASS(inp->inp_socket == NULL);
1715 crfree(inp->inp_cred);
1717 inp->inp_cred = NULL;
1720 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1725 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1728 return (lock == INPLOOKUP_RLOCKPCB ?
1729 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1733 * Unconditionally schedule an inpcb to be freed by decrementing its
1734 * reference count, which should occur only after the inpcb has been detached
1735 * from its socket. If another thread holds a temporary reference (acquired
1736 * using in_pcbref()) then the free is deferred until that reference is
1737 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1738 * Almost all work, including removal from global lists, is done in this
1739 * context, where the pcbinfo lock is held.
1742 in_pcbfree(struct inpcb *inp)
1744 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1746 struct ip_moptions *imo;
1749 struct ip6_moptions *im6o;
1752 INP_WLOCK_ASSERT(inp);
1753 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1754 KASSERT((inp->inp_flags & INP_FREED) == 0,
1755 ("%s: called twice for pcb %p", __func__, inp));
1757 inp->inp_flags |= INP_FREED;
1758 INP_INFO_WLOCK(pcbinfo);
1759 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1760 pcbinfo->ipi_count--;
1761 CK_LIST_REMOVE(inp, inp_list);
1762 INP_INFO_WUNLOCK(pcbinfo);
1764 if (inp->inp_flags & INP_INHASHLIST)
1767 RO_INVALIDATE_CACHE(&inp->inp_route);
1769 mac_inpcb_destroy(inp);
1771 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1772 if (inp->inp_sp != NULL)
1773 ipsec_delete_pcbpolicy(inp);
1776 if (inp->inp_options)
1777 (void)m_free(inp->inp_options);
1778 imo = inp->inp_moptions;
1781 if (inp->inp_vflag & INP_IPV6PROTO) {
1782 ip6_freepcbopts(inp->in6p_outputopts);
1783 im6o = inp->in6p_moptions;
1788 if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1792 ip6_freemoptions(im6o);
1795 inp_freemoptions(imo);
1800 * Different protocols initialize their inpcbs differently - giving
1801 * different name to the lock. But they all are disposed the same.
1804 inpcb_fini(void *mem, int size)
1806 struct inpcb *inp = mem;
1808 INP_LOCK_DESTROY(inp);
1812 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1813 * port reservation, and preventing it from being returned by inpcb lookups.
1815 * It is used by TCP to mark an inpcb as unused and avoid future packet
1816 * delivery or event notification when a socket remains open but TCP has
1817 * closed. This might occur as a result of a shutdown()-initiated TCP close
1818 * or a RST on the wire, and allows the port binding to be reused while still
1819 * maintaining the invariant that so_pcb always points to a valid inpcb until
1822 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1823 * in_pcbnotifyall() and in_pcbpurgeif0()?
1826 in_pcbdrop(struct inpcb *inp)
1829 INP_WLOCK_ASSERT(inp);
1831 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1832 MPASS(inp->inp_refcount > 1);
1835 inp->inp_flags |= INP_DROPPED;
1836 if (inp->inp_flags & INP_INHASHLIST)
1842 * Common routines to return the socket addresses associated with inpcbs.
1845 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1847 struct sockaddr_in *sin;
1849 sin = malloc(sizeof *sin, M_SONAME,
1851 sin->sin_family = AF_INET;
1852 sin->sin_len = sizeof(*sin);
1853 sin->sin_addr = *addr_p;
1854 sin->sin_port = port;
1856 return (struct sockaddr *)sin;
1860 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1863 struct in_addr addr;
1866 inp = sotoinpcb(so);
1867 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1870 port = inp->inp_lport;
1871 addr = inp->inp_laddr;
1874 *nam = in_sockaddr(port, &addr);
1879 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1882 struct in_addr addr;
1885 inp = sotoinpcb(so);
1886 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1889 port = inp->inp_fport;
1890 addr = inp->inp_faddr;
1893 *nam = in_sockaddr(port, &addr);
1898 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1899 struct inpcb *(*notify)(struct inpcb *, int))
1901 struct inpcb *inp, *inp_temp;
1903 INP_INFO_WLOCK(pcbinfo);
1904 CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
1907 if ((inp->inp_vflag & INP_IPV4) == 0) {
1912 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1913 inp->inp_socket == NULL) {
1917 if ((*notify)(inp, errno))
1920 INP_INFO_WUNLOCK(pcbinfo);
1924 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
1927 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
1934 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1936 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
1937 inp_v4_multi_match, NULL);
1939 struct in_multi *inm;
1940 struct in_mfilter *imf;
1941 struct ip_moptions *imo;
1943 IN_MULTI_LOCK_ASSERT();
1945 while ((inp = inp_next(&inpi)) != NULL) {
1946 INP_WLOCK_ASSERT(inp);
1948 imo = inp->inp_moptions;
1950 * Unselect the outgoing interface if it is being
1953 if (imo->imo_multicast_ifp == ifp)
1954 imo->imo_multicast_ifp = NULL;
1957 * Drop multicast group membership if we joined
1958 * through the interface being detached.
1960 * XXX This can all be deferred to an epoch_call
1963 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1964 if ((inm = imf->imf_inm) == NULL)
1966 if (inm->inm_ifp != ifp)
1968 ip_mfilter_remove(&imo->imo_head, imf);
1969 in_leavegroup_locked(inm, NULL);
1970 ip_mfilter_free(imf);
1977 * Lookup a PCB based on the local address and port. Caller must hold the
1978 * hash lock. No inpcb locks or references are acquired.
1980 #define INP_LOOKUP_MAPPED_PCB_COST 3
1982 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1983 u_short lport, int lookupflags, struct ucred *cred)
1987 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1993 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1994 ("%s: invalid lookup flags %d", __func__, lookupflags));
1995 INP_HASH_LOCK_ASSERT(pcbinfo);
1997 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1998 struct inpcbhead *head;
2000 * Look for an unconnected (wildcard foreign addr) PCB that
2001 * matches the local address and port we're looking for.
2003 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2004 pcbinfo->ipi_hashmask)];
2005 CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2007 /* XXX inp locking */
2008 if ((inp->inp_vflag & INP_IPV4) == 0)
2011 if (inp->inp_faddr.s_addr == INADDR_ANY &&
2012 inp->inp_laddr.s_addr == laddr.s_addr &&
2013 inp->inp_lport == lport) {
2017 if (prison_equal_ip4(cred->cr_prison,
2018 inp->inp_cred->cr_prison))
2027 struct inpcbporthead *porthash;
2028 struct inpcbport *phd;
2029 struct inpcb *match = NULL;
2031 * Best fit PCB lookup.
2033 * First see if this local port is in use by looking on the
2036 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2037 pcbinfo->ipi_porthashmask)];
2038 CK_LIST_FOREACH(phd, porthash, phd_hash) {
2039 if (phd->phd_port == lport)
2044 * Port is in use by one or more PCBs. Look for best
2047 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2049 if (!prison_equal_ip4(inp->inp_cred->cr_prison,
2053 /* XXX inp locking */
2054 if ((inp->inp_vflag & INP_IPV4) == 0)
2057 * We never select the PCB that has
2058 * INP_IPV6 flag and is bound to :: if
2059 * we have another PCB which is bound
2060 * to 0.0.0.0. If a PCB has the
2061 * INP_IPV6 flag, then we set its cost
2062 * higher than IPv4 only PCBs.
2064 * Note that the case only happens
2065 * when a socket is bound to ::, under
2066 * the condition that the use of the
2067 * mapped address is allowed.
2069 if ((inp->inp_vflag & INP_IPV6) != 0)
2070 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2072 if (inp->inp_faddr.s_addr != INADDR_ANY)
2074 if (inp->inp_laddr.s_addr != INADDR_ANY) {
2075 if (laddr.s_addr == INADDR_ANY)
2077 else if (inp->inp_laddr.s_addr != laddr.s_addr)
2080 if (laddr.s_addr != INADDR_ANY)
2083 if (wildcard < matchwild) {
2085 matchwild = wildcard;
2094 #undef INP_LOOKUP_MAPPED_PCB_COST
2097 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain)
2099 return (domain == M_NODOM || domain == grp->il_numa_domain);
2102 static struct inpcb *
2103 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2104 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr,
2105 uint16_t lport, int domain)
2107 const struct inpcblbgrouphead *hdr;
2108 struct inpcblbgroup *grp;
2109 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild;
2111 INP_HASH_LOCK_ASSERT(pcbinfo);
2113 hdr = &pcbinfo->ipi_lbgrouphashbase[
2114 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2117 * Search for an LB group match based on the following criteria:
2118 * - prefer jailed groups to non-jailed groups
2119 * - prefer exact source address matches to wildcard matches
2120 * - prefer groups bound to the specified NUMA domain
2122 jail_exact = jail_wild = local_exact = local_wild = NULL;
2123 CK_LIST_FOREACH(grp, hdr, il_list) {
2127 if (!(grp->il_vflag & INP_IPV4))
2130 if (grp->il_lport != lport)
2133 injail = prison_flag(grp->il_cred, PR_IP4) != 0;
2134 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison,
2138 if (grp->il_laddr.s_addr == laddr->s_addr) {
2141 if (in_pcblookup_lb_numa_match(grp, domain))
2142 /* This is a perfect match. */
2144 } else if (local_exact == NULL ||
2145 in_pcblookup_lb_numa_match(grp, domain)) {
2148 } else if (grp->il_laddr.s_addr == INADDR_ANY) {
2150 if (jail_wild == NULL ||
2151 in_pcblookup_lb_numa_match(grp, domain))
2153 } else if (local_wild == NULL ||
2154 in_pcblookup_lb_numa_match(grp, domain)) {
2160 if (jail_exact != NULL)
2162 else if (jail_wild != NULL)
2164 else if (local_exact != NULL)
2171 return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2176 in_pcblookup_exact_match(const struct inpcb *inp, struct in_addr faddr,
2177 u_short fport, struct in_addr laddr, u_short lport)
2180 /* XXX inp locking */
2181 if ((inp->inp_vflag & INP_IPV4) == 0)
2184 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2185 inp->inp_laddr.s_addr == laddr.s_addr &&
2186 inp->inp_fport == fport &&
2187 inp->inp_lport == lport)
2192 static struct inpcb *
2193 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2194 u_short fport, struct in_addr laddr, u_short lport)
2196 struct inpcbhead *head;
2199 INP_HASH_LOCK_ASSERT(pcbinfo);
2201 head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport,
2202 pcbinfo->ipi_hashmask)];
2203 CK_LIST_FOREACH(inp, head, inp_hash_exact) {
2204 if (in_pcblookup_exact_match(inp, faddr, fport, laddr, lport))
2211 INPLOOKUP_MATCH_NONE = 0,
2212 INPLOOKUP_MATCH_WILD = 1,
2213 INPLOOKUP_MATCH_LADDR = 2,
2214 } inp_lookup_match_t;
2216 static inp_lookup_match_t
2217 in_pcblookup_wild_match(const struct inpcb *inp, struct in_addr laddr,
2221 /* XXX inp locking */
2222 if ((inp->inp_vflag & INP_IPV4) == 0)
2223 return (INPLOOKUP_MATCH_NONE);
2225 if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport)
2226 return (INPLOOKUP_MATCH_NONE);
2227 if (inp->inp_laddr.s_addr == INADDR_ANY)
2228 return (INPLOOKUP_MATCH_WILD);
2229 if (inp->inp_laddr.s_addr == laddr.s_addr)
2230 return (INPLOOKUP_MATCH_LADDR);
2231 return (INPLOOKUP_MATCH_NONE);
2234 #define INP_LOOKUP_AGAIN ((struct inpcb *)(uintptr_t)-1)
2236 static struct inpcb *
2237 in_pcblookup_hash_wild_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2238 u_short fport, struct in_addr laddr, u_short lport,
2239 const inp_lookup_t lockflags)
2241 struct inpcbhead *head;
2244 KASSERT(SMR_ENTERED(pcbinfo->ipi_smr),
2245 ("%s: not in SMR read section", __func__));
2247 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2248 pcbinfo->ipi_hashmask)];
2249 CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2250 inp_lookup_match_t match;
2252 match = in_pcblookup_wild_match(inp, laddr, lport);
2253 if (match == INPLOOKUP_MATCH_NONE)
2256 if (__predict_true(inp_smr_lock(inp, lockflags))) {
2257 match = in_pcblookup_wild_match(inp, laddr, lport);
2258 if (match != INPLOOKUP_MATCH_NONE &&
2259 prison_check_ip4_locked(inp->inp_cred->cr_prison,
2262 inp_unlock(inp, lockflags);
2266 * The matching socket disappeared out from under us. Fall back
2267 * to a serialized lookup.
2269 return (INP_LOOKUP_AGAIN);
2274 static struct inpcb *
2275 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2276 u_short fport, struct in_addr laddr, u_short lport)
2278 struct inpcbhead *head;
2279 struct inpcb *inp, *local_wild, *local_exact, *jail_wild;
2281 struct inpcb *local_wild_mapped;
2284 INP_HASH_LOCK_ASSERT(pcbinfo);
2287 * Order of socket selection - we always prefer jails.
2288 * 1. jailed, non-wild.
2290 * 3. non-jailed, non-wild.
2291 * 4. non-jailed, wild.
2293 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2294 pcbinfo->ipi_hashmask)];
2295 local_wild = local_exact = jail_wild = NULL;
2297 local_wild_mapped = NULL;
2299 CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2300 inp_lookup_match_t match;
2303 match = in_pcblookup_wild_match(inp, laddr, lport);
2304 if (match == INPLOOKUP_MATCH_NONE)
2307 injail = prison_flag(inp->inp_cred, PR_IP4) != 0;
2309 if (prison_check_ip4_locked(inp->inp_cred->cr_prison,
2313 if (local_exact != NULL)
2317 if (match == INPLOOKUP_MATCH_LADDR) {
2323 /* XXX inp locking, NULL check */
2324 if (inp->inp_vflag & INP_IPV6PROTO)
2325 local_wild_mapped = inp;
2334 if (jail_wild != NULL)
2336 if (local_exact != NULL)
2337 return (local_exact);
2338 if (local_wild != NULL)
2339 return (local_wild);
2341 if (local_wild_mapped != NULL)
2342 return (local_wild_mapped);
2348 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2349 * that the caller has either locked the hash list, which usually happens
2350 * for bind(2) operations, or is in SMR section, which happens when sorting
2351 * out incoming packets.
2353 static struct inpcb *
2354 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2355 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2356 uint8_t numa_domain)
2359 const u_short fport = fport_arg, lport = lport_arg;
2361 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0,
2362 ("%s: invalid lookup flags %d", __func__, lookupflags));
2363 KASSERT(faddr.s_addr != INADDR_ANY,
2364 ("%s: invalid foreign address", __func__));
2365 KASSERT(laddr.s_addr != INADDR_ANY,
2366 ("%s: invalid local address", __func__));
2367 INP_HASH_WLOCK_ASSERT(pcbinfo);
2369 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2373 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2374 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport,
2375 &laddr, lport, numa_domain);
2377 inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr,
2378 fport, laddr, lport);
2385 static struct inpcb *
2386 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2387 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2388 uint8_t numa_domain)
2391 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK;
2393 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2394 ("%s: LOCKPCB not set", __func__));
2396 INP_HASH_WLOCK(pcbinfo);
2397 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2398 lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain);
2399 if (inp != NULL && !inp_trylock(inp, lockflags)) {
2401 INP_HASH_WUNLOCK(pcbinfo);
2402 inp_lock(inp, lockflags);
2403 if (in_pcbrele(inp, lockflags))
2404 /* XXX-MJ or retry until we get a negative match? */
2407 INP_HASH_WUNLOCK(pcbinfo);
2412 static struct inpcb *
2413 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2414 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2415 uint8_t numa_domain)
2418 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK;
2419 const u_short fport = fport_arg, lport = lport_arg;
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__));
2426 smr_enter(pcbinfo->ipi_smr);
2427 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2429 if (__predict_true(inp_smr_lock(inp, lockflags))) {
2431 * Revalidate the 4-tuple, the socket could have been
2434 if (__predict_true(in_pcblookup_exact_match(inp,
2435 faddr, fport, laddr, lport)))
2437 inp_unlock(inp, lockflags);
2441 * We failed to lock the inpcb, or its connection state changed
2442 * out from under us. Fall back to a precise search.
2444 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2445 lookupflags, numa_domain));
2448 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2449 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport,
2450 &laddr, lport, numa_domain);
2452 if (__predict_true(inp_smr_lock(inp, lockflags))) {
2453 if (__predict_true(in_pcblookup_wild_match(inp,
2454 laddr, lport) != INPLOOKUP_MATCH_NONE))
2456 inp_unlock(inp, lockflags);
2458 inp = INP_LOOKUP_AGAIN;
2460 inp = in_pcblookup_hash_wild_smr(pcbinfo, faddr, fport,
2461 laddr, lport, lockflags);
2463 if (inp == INP_LOOKUP_AGAIN) {
2464 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr,
2465 lport, lookupflags, numa_domain));
2470 smr_exit(pcbinfo->ipi_smr);
2476 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2477 * from which a pre-calculated hash value may be extracted.
2480 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2481 struct in_addr laddr, u_int lport, int lookupflags,
2482 struct ifnet *ifp __unused)
2484 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2485 lookupflags, M_NODOM));
2489 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2490 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2491 struct ifnet *ifp __unused, struct mbuf *m)
2493 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2494 lookupflags, m->m_pkthdr.numa_domain));
2499 in_pcbjailed(const struct inpcb *inp, unsigned int flag)
2501 return (prison_flag(inp->inp_cred, flag) != 0);
2505 * Insert the PCB into a hash chain using ordering rules which ensure that
2506 * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first.
2508 * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs
2509 * with exact local addresses ahead of wildcard PCBs. Unbound v4-mapped v6 PCBs
2510 * always appear last no matter whether they are jailed.
2513 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp)
2518 INP_LOCK_ASSERT(inp);
2519 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2522 bound = inp->inp_laddr.s_addr != INADDR_ANY;
2523 if (!bound && (inp->inp_vflag & INP_IPV6PROTO) != 0) {
2524 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) {
2525 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2526 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2530 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2534 injail = in_pcbjailed(inp, PR_IP4);
2536 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) {
2537 if (!in_pcbjailed(last, PR_IP4))
2539 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2540 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2544 } else if (!CK_LIST_EMPTY(pcbhash) &&
2545 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) {
2546 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2550 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) {
2551 if (last->inp_laddr.s_addr == INADDR_ANY)
2553 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2554 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2560 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2562 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild);
2567 * See the comment above _in_pcbinshash_wild().
2570 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp)
2575 INP_LOCK_ASSERT(inp);
2576 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2579 bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr);
2580 injail = in_pcbjailed(inp, PR_IP6);
2582 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) {
2583 if (!in_pcbjailed(last, PR_IP6))
2585 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2586 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2590 } else if (!CK_LIST_EMPTY(pcbhash) &&
2591 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) {
2592 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2596 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) {
2597 if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr))
2599 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2600 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2606 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2608 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild);
2613 * Insert PCB onto various hash lists.
2616 in_pcbinshash(struct inpcb *inp)
2618 struct inpcbhead *pcbhash;
2619 struct inpcbporthead *pcbporthash;
2620 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2621 struct inpcbport *phd;
2625 INP_WLOCK_ASSERT(inp);
2626 INP_HASH_WLOCK_ASSERT(pcbinfo);
2627 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2628 ("in_pcbinshash: INP_INHASHLIST"));
2631 if (inp->inp_vflag & INP_IPV6) {
2632 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport,
2633 inp->inp_fport, pcbinfo->ipi_hashmask);
2634 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr);
2638 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport,
2639 inp->inp_fport, pcbinfo->ipi_hashmask);
2640 connected = !in_nullhost(inp->inp_faddr);
2644 pcbhash = &pcbinfo->ipi_hash_exact[hash];
2646 pcbhash = &pcbinfo->ipi_hash_wild[hash];
2648 pcbporthash = &pcbinfo->ipi_porthashbase[
2649 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2652 * Add entry to load balance group.
2653 * Only do this if SO_REUSEPORT_LB is set.
2655 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) {
2656 int error = in_pcbinslbgrouphash(inp, M_NODOM);
2662 * Go through port list and look for a head for this lport.
2664 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2665 if (phd->phd_port == inp->inp_lport)
2670 * If none exists, malloc one and tack it on.
2673 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2675 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2676 in_pcbremlbgrouphash(inp);
2679 phd->phd_port = inp->inp_lport;
2680 CK_LIST_INIT(&phd->phd_pcblist);
2681 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2684 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2687 * The PCB may have been disconnected in the past. Before we can safely
2688 * make it visible in the hash table, we must wait for all readers which
2689 * may be traversing this PCB to finish.
2691 if (inp->inp_smr != SMR_SEQ_INVALID) {
2692 smr_wait(pcbinfo->ipi_smr, inp->inp_smr);
2693 inp->inp_smr = SMR_SEQ_INVALID;
2697 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact);
2700 if ((inp->inp_vflag & INP_IPV6) != 0)
2701 _in6_pcbinshash_wild(pcbhash, inp);
2704 _in_pcbinshash_wild(pcbhash, inp);
2706 inp->inp_flags |= INP_INHASHLIST;
2712 in_pcbremhash_locked(struct inpcb *inp)
2714 struct inpcbport *phd = inp->inp_phd;
2716 INP_WLOCK_ASSERT(inp);
2717 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2718 MPASS(inp->inp_flags & INP_INHASHLIST);
2720 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2721 in_pcbremlbgrouphash(inp);
2723 if (inp->inp_vflag & INP_IPV6) {
2724 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
2725 CK_LIST_REMOVE(inp, inp_hash_wild);
2727 CK_LIST_REMOVE(inp, inp_hash_exact);
2731 if (in_nullhost(inp->inp_faddr))
2732 CK_LIST_REMOVE(inp, inp_hash_wild);
2734 CK_LIST_REMOVE(inp, inp_hash_exact);
2736 CK_LIST_REMOVE(inp, inp_portlist);
2737 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2738 CK_LIST_REMOVE(phd, phd_hash);
2739 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
2741 inp->inp_flags &= ~INP_INHASHLIST;
2745 in_pcbremhash(struct inpcb *inp)
2747 INP_HASH_WLOCK(inp->inp_pcbinfo);
2748 in_pcbremhash_locked(inp);
2749 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
2753 * Move PCB to the proper hash bucket when { faddr, fport } have been
2754 * changed. NOTE: This does not handle the case of the lport changing (the
2755 * hashed port list would have to be updated as well), so the lport must
2756 * not change after in_pcbinshash() has been called.
2759 in_pcbrehash(struct inpcb *inp)
2761 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2762 struct inpcbhead *head;
2766 INP_WLOCK_ASSERT(inp);
2767 INP_HASH_WLOCK_ASSERT(pcbinfo);
2768 KASSERT(inp->inp_flags & INP_INHASHLIST,
2769 ("%s: !INP_INHASHLIST", __func__));
2770 KASSERT(inp->inp_smr == SMR_SEQ_INVALID,
2771 ("%s: inp was disconnected", __func__));
2774 if (inp->inp_vflag & INP_IPV6) {
2775 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport,
2776 inp->inp_fport, pcbinfo->ipi_hashmask);
2777 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr);
2781 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport,
2782 inp->inp_fport, pcbinfo->ipi_hashmask);
2783 connected = !in_nullhost(inp->inp_faddr);
2787 * When rehashing, the caller must ensure that either the new or the old
2788 * foreign address was unspecified.
2791 CK_LIST_REMOVE(inp, inp_hash_wild);
2793 CK_LIST_REMOVE(inp, inp_hash_exact);
2796 head = &pcbinfo->ipi_hash_exact[hash];
2797 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact);
2799 head = &pcbinfo->ipi_hash_wild[hash];
2800 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild);
2805 * Check for alternatives when higher level complains
2806 * about service problems. For now, invalidate cached
2807 * routing information. If the route was created dynamically
2808 * (by a redirect), time to try a default gateway again.
2811 in_losing(struct inpcb *inp)
2814 RO_INVALIDATE_CACHE(&inp->inp_route);
2819 * A set label operation has occurred at the socket layer, propagate the
2820 * label change into the in_pcb for the socket.
2823 in_pcbsosetlabel(struct socket *so)
2828 inp = sotoinpcb(so);
2829 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2833 mac_inpcb_sosetlabel(so, inp);
2840 inp_wlock(struct inpcb *inp)
2847 inp_wunlock(struct inpcb *inp)
2854 inp_rlock(struct inpcb *inp)
2861 inp_runlock(struct inpcb *inp)
2867 #ifdef INVARIANT_SUPPORT
2869 inp_lock_assert(struct inpcb *inp)
2872 INP_WLOCK_ASSERT(inp);
2876 inp_unlock_assert(struct inpcb *inp)
2879 INP_UNLOCK_ASSERT(inp);
2884 inp_apply_all(struct inpcbinfo *pcbinfo,
2885 void (*func)(struct inpcb *, void *), void *arg)
2887 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2888 INPLOOKUP_WLOCKPCB);
2891 while ((inp = inp_next(&inpi)) != NULL)
2896 inp_inpcbtosocket(struct inpcb *inp)
2899 INP_WLOCK_ASSERT(inp);
2900 return (inp->inp_socket);
2904 inp_inpcbtotcpcb(struct inpcb *inp)
2907 INP_WLOCK_ASSERT(inp);
2908 return ((struct tcpcb *)inp->inp_ppcb);
2912 inp_ip_tos_get(const struct inpcb *inp)
2915 return (inp->inp_ip_tos);
2919 inp_ip_tos_set(struct inpcb *inp, int val)
2922 inp->inp_ip_tos = val;
2926 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2927 uint32_t *faddr, uint16_t *fp)
2930 INP_LOCK_ASSERT(inp);
2931 *laddr = inp->inp_laddr.s_addr;
2932 *faddr = inp->inp_faddr.s_addr;
2933 *lp = inp->inp_lport;
2934 *fp = inp->inp_fport;
2938 so_sotoinpcb(struct socket *so)
2941 return (sotoinpcb(so));
2945 * Create an external-format (``xinpcb'') structure using the information in
2946 * the kernel-format in_pcb structure pointed to by inp. This is done to
2947 * reduce the spew of irrelevant information over this interface, to isolate
2948 * user code from changes in the kernel structure, and potentially to provide
2949 * information-hiding if we decide that some of this information should be
2950 * hidden from users.
2953 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2956 bzero(xi, sizeof(*xi));
2957 xi->xi_len = sizeof(struct xinpcb);
2958 if (inp->inp_socket)
2959 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2960 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2961 xi->inp_gencnt = inp->inp_gencnt;
2962 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2963 xi->inp_flow = inp->inp_flow;
2964 xi->inp_flowid = inp->inp_flowid;
2965 xi->inp_flowtype = inp->inp_flowtype;
2966 xi->inp_flags = inp->inp_flags;
2967 xi->inp_flags2 = inp->inp_flags2;
2968 xi->in6p_cksum = inp->in6p_cksum;
2969 xi->in6p_hops = inp->in6p_hops;
2970 xi->inp_ip_tos = inp->inp_ip_tos;
2971 xi->inp_vflag = inp->inp_vflag;
2972 xi->inp_ip_ttl = inp->inp_ip_ttl;
2973 xi->inp_ip_p = inp->inp_ip_p;
2974 xi->inp_ip_minttl = inp->inp_ip_minttl;
2978 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2979 int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2981 struct sockopt sopt;
2982 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2983 INPLOOKUP_WLOCKPCB);
2985 struct sockopt_parameters *params;
2990 if (req->oldptr != NULL || req->oldlen != 0)
2992 if (req->newptr == NULL)
2994 if (req->newlen > sizeof(buf))
2996 error = SYSCTL_IN(req, buf, req->newlen);
2999 if (req->newlen < sizeof(struct sockopt_parameters))
3001 params = (struct sockopt_parameters *)buf;
3002 sopt.sopt_level = params->sop_level;
3003 sopt.sopt_name = params->sop_optname;
3004 sopt.sopt_dir = SOPT_SET;
3005 sopt.sopt_val = params->sop_optval;
3006 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
3007 sopt.sopt_td = NULL;
3009 if (params->sop_inc.inc_flags & INC_ISIPV6) {
3010 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr))
3011 params->sop_inc.inc6_laddr.s6_addr16[1] =
3012 htons(params->sop_inc.inc6_zoneid & 0xffff);
3013 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr))
3014 params->sop_inc.inc6_faddr.s6_addr16[1] =
3015 htons(params->sop_inc.inc6_zoneid & 0xffff);
3018 if (params->sop_inc.inc_lport != htons(0)) {
3019 if (params->sop_inc.inc_fport == htons(0))
3020 inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
3021 pcbinfo->ipi_hashmask);
3024 if (params->sop_inc.inc_flags & INC_ISIPV6)
3025 inpi.hash = INP6_PCBHASH(
3026 ¶ms->sop_inc.inc6_faddr,
3027 params->sop_inc.inc_lport,
3028 params->sop_inc.inc_fport,
3029 pcbinfo->ipi_hashmask);
3032 inpi.hash = INP_PCBHASH(
3033 ¶ms->sop_inc.inc_faddr,
3034 params->sop_inc.inc_lport,
3035 params->sop_inc.inc_fport,
3036 pcbinfo->ipi_hashmask);
3038 while ((inp = inp_next(&inpi)) != NULL)
3039 if (inp->inp_gencnt == params->sop_id) {
3040 if (inp->inp_flags & INP_DROPPED) {
3042 return (ECONNRESET);
3044 so = inp->inp_socket;
3045 KASSERT(so != NULL, ("inp_socket == NULL"));
3047 error = (*ctloutput_set)(inp, &sopt);
3058 db_print_indent(int indent)
3062 for (i = 0; i < indent; i++)
3067 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
3069 char faddr_str[48], laddr_str[48];
3071 db_print_indent(indent);
3072 db_printf("%s at %p\n", name, inc);
3077 if (inc->inc_flags & INC_ISIPV6) {
3079 ip6_sprintf(laddr_str, &inc->inc6_laddr);
3080 ip6_sprintf(faddr_str, &inc->inc6_faddr);
3085 inet_ntoa_r(inc->inc_laddr, laddr_str);
3086 inet_ntoa_r(inc->inc_faddr, faddr_str);
3088 db_print_indent(indent);
3089 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
3090 ntohs(inc->inc_lport));
3091 db_print_indent(indent);
3092 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
3093 ntohs(inc->inc_fport));
3097 db_print_inpflags(int inp_flags)
3102 if (inp_flags & INP_RECVOPTS) {
3103 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
3106 if (inp_flags & INP_RECVRETOPTS) {
3107 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
3110 if (inp_flags & INP_RECVDSTADDR) {
3111 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
3114 if (inp_flags & INP_ORIGDSTADDR) {
3115 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
3118 if (inp_flags & INP_HDRINCL) {
3119 db_printf("%sINP_HDRINCL", comma ? ", " : "");
3122 if (inp_flags & INP_HIGHPORT) {
3123 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3126 if (inp_flags & INP_LOWPORT) {
3127 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3130 if (inp_flags & INP_ANONPORT) {
3131 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3134 if (inp_flags & INP_RECVIF) {
3135 db_printf("%sINP_RECVIF", comma ? ", " : "");
3138 if (inp_flags & INP_MTUDISC) {
3139 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3142 if (inp_flags & INP_RECVTTL) {
3143 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3146 if (inp_flags & INP_DONTFRAG) {
3147 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3150 if (inp_flags & INP_RECVTOS) {
3151 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3154 if (inp_flags & IN6P_IPV6_V6ONLY) {
3155 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3158 if (inp_flags & IN6P_PKTINFO) {
3159 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3162 if (inp_flags & IN6P_HOPLIMIT) {
3163 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3166 if (inp_flags & IN6P_HOPOPTS) {
3167 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3170 if (inp_flags & IN6P_DSTOPTS) {
3171 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3174 if (inp_flags & IN6P_RTHDR) {
3175 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3178 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3179 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3182 if (inp_flags & IN6P_TCLASS) {
3183 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3186 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3187 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3190 if (inp_flags & INP_ONESBCAST) {
3191 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3194 if (inp_flags & INP_DROPPED) {
3195 db_printf("%sINP_DROPPED", comma ? ", " : "");
3198 if (inp_flags & INP_SOCKREF) {
3199 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3202 if (inp_flags & IN6P_RFC2292) {
3203 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3206 if (inp_flags & IN6P_MTU) {
3207 db_printf("IN6P_MTU%s", comma ? ", " : "");
3213 db_print_inpvflag(u_char inp_vflag)
3218 if (inp_vflag & INP_IPV4) {
3219 db_printf("%sINP_IPV4", comma ? ", " : "");
3222 if (inp_vflag & INP_IPV6) {
3223 db_printf("%sINP_IPV6", comma ? ", " : "");
3226 if (inp_vflag & INP_IPV6PROTO) {
3227 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3233 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3236 db_print_indent(indent);
3237 db_printf("%s at %p\n", name, inp);
3241 db_print_indent(indent);
3242 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3244 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3246 db_print_indent(indent);
3247 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3248 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3250 db_print_indent(indent);
3251 db_printf("inp_label: %p inp_flags: 0x%x (",
3252 inp->inp_label, inp->inp_flags);
3253 db_print_inpflags(inp->inp_flags);
3256 db_print_indent(indent);
3257 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3259 db_print_inpvflag(inp->inp_vflag);
3262 db_print_indent(indent);
3263 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3264 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3266 db_print_indent(indent);
3268 if (inp->inp_vflag & INP_IPV6) {
3269 db_printf("in6p_options: %p in6p_outputopts: %p "
3270 "in6p_moptions: %p\n", inp->in6p_options,
3271 inp->in6p_outputopts, inp->in6p_moptions);
3272 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3273 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3278 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3279 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3280 inp->inp_options, inp->inp_moptions);
3283 db_print_indent(indent);
3284 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3285 (uintmax_t)inp->inp_gencnt);
3288 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3293 db_printf("usage: show inpcb <addr>\n");
3296 inp = (struct inpcb *)addr;
3298 db_print_inpcb(inp, "inpcb", 0);
3304 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3308 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3310 union if_snd_tag_modify_params params = {
3311 .rate_limit.max_rate = max_pacing_rate,
3312 .rate_limit.flags = M_NOWAIT,
3314 struct m_snd_tag *mst;
3317 mst = inp->inp_snd_tag;
3321 if (mst->sw->snd_tag_modify == NULL) {
3324 error = mst->sw->snd_tag_modify(mst, ¶ms);
3330 * Query existing TX rate limit based on the existing
3331 * "inp->inp_snd_tag", if any.
3334 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3336 union if_snd_tag_query_params params = { };
3337 struct m_snd_tag *mst;
3340 mst = inp->inp_snd_tag;
3344 if (mst->sw->snd_tag_query == NULL) {
3347 error = mst->sw->snd_tag_query(mst, ¶ms);
3348 if (error == 0 && p_max_pacing_rate != NULL)
3349 *p_max_pacing_rate = params.rate_limit.max_rate;
3355 * Query existing TX queue level based on the existing
3356 * "inp->inp_snd_tag", if any.
3359 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3361 union if_snd_tag_query_params params = { };
3362 struct m_snd_tag *mst;
3365 mst = inp->inp_snd_tag;
3369 if (mst->sw->snd_tag_query == NULL)
3370 return (EOPNOTSUPP);
3372 error = mst->sw->snd_tag_query(mst, ¶ms);
3373 if (error == 0 && p_txqueue_level != NULL)
3374 *p_txqueue_level = params.rate_limit.queue_level;
3379 * Allocate a new TX rate limit send tag from the network interface
3380 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3383 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3384 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3387 union if_snd_tag_alloc_params params = {
3388 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3389 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3390 .rate_limit.hdr.flowid = flowid,
3391 .rate_limit.hdr.flowtype = flowtype,
3392 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3393 .rate_limit.max_rate = max_pacing_rate,
3394 .rate_limit.flags = M_NOWAIT,
3398 INP_WLOCK_ASSERT(inp);
3401 * If there is already a send tag, or the INP is being torn
3402 * down, allocating a new send tag is not allowed. Else send
3405 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0)
3408 error = m_snd_tag_alloc(ifp, ¶ms, st);
3411 counter_u64_add(rate_limit_set_ok, 1);
3412 counter_u64_add(rate_limit_active, 1);
3413 } else if (error != EOPNOTSUPP)
3414 counter_u64_add(rate_limit_alloc_fail, 1);
3420 in_pcbdetach_tag(struct m_snd_tag *mst)
3423 m_snd_tag_rele(mst);
3425 counter_u64_add(rate_limit_active, -1);
3430 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3434 in_pcbdetach_txrtlmt(struct inpcb *inp)
3436 struct m_snd_tag *mst;
3438 INP_WLOCK_ASSERT(inp);
3440 mst = inp->inp_snd_tag;
3441 inp->inp_snd_tag = NULL;
3446 m_snd_tag_rele(mst);
3448 counter_u64_add(rate_limit_active, -1);
3453 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3458 * If the existing send tag is for the wrong interface due to
3459 * a route change, first drop the existing tag. Set the
3460 * CHANGED flag so that we will keep trying to allocate a new
3461 * tag if we fail to allocate one this time.
3463 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3464 in_pcbdetach_txrtlmt(inp);
3465 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3469 * NOTE: When attaching to a network interface a reference is
3470 * made to ensure the network interface doesn't go away until
3471 * all ratelimit connections are gone. The network interface
3472 * pointers compared below represent valid network interfaces,
3473 * except when comparing towards NULL.
3475 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3477 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3478 if (inp->inp_snd_tag != NULL)
3479 in_pcbdetach_txrtlmt(inp);
3481 } else if (inp->inp_snd_tag == NULL) {
3483 * In order to utilize packet pacing with RSS, we need
3484 * to wait until there is a valid RSS hash before we
3487 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3490 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3491 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3494 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3496 if (error == 0 || error == EOPNOTSUPP)
3497 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3503 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3504 * is set in the fast path and will attach/detach/modify the TX rate
3505 * limit send tag based on the socket's so_max_pacing_rate value.
3508 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3510 struct socket *socket;
3511 uint32_t max_pacing_rate;
3517 socket = inp->inp_socket;
3521 if (!INP_WLOCKED(inp)) {
3523 * NOTE: If the write locking fails, we need to bail
3524 * out and use the non-ratelimited ring for the
3525 * transmit until there is a new chance to get the
3528 if (!INP_TRY_UPGRADE(inp))
3536 * NOTE: The so_max_pacing_rate value is read unlocked,
3537 * because atomic updates are not required since the variable
3538 * is checked at every mbuf we send. It is assumed that the
3539 * variable read itself will be atomic.
3541 max_pacing_rate = socket->so_max_pacing_rate;
3543 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3550 * Track route changes for TX rate limiting.
3553 in_pcboutput_eagain(struct inpcb *inp)
3560 if (inp->inp_snd_tag == NULL)
3563 if (!INP_WLOCKED(inp)) {
3565 * NOTE: If the write locking fails, we need to bail
3566 * out and use the non-ratelimited ring for the
3567 * transmit until there is a new chance to get the
3570 if (!INP_TRY_UPGRADE(inp))
3577 /* detach rate limiting */
3578 in_pcbdetach_txrtlmt(inp);
3580 /* make sure new mbuf send tag allocation is made */
3581 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3591 rate_limit_new = counter_u64_alloc(M_WAITOK);
3592 rate_limit_chg = counter_u64_alloc(M_WAITOK);
3593 rate_limit_active = counter_u64_alloc(M_WAITOK);
3594 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3595 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3598 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3600 #endif /* RATELIMIT */