2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1991, 1993, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2007-2009 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 #include "opt_ipsec.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/malloc.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
67 #include <sys/refcount.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
90 #include <netinet/in_var.h>
91 #include <netinet/in_fib.h>
93 #include <netinet/ip_var.h>
94 #include <netinet/tcp_var.h>
96 #include <netinet/tcp_hpts.h>
98 #include <netinet/udp.h>
99 #include <netinet/udp_var.h>
101 #include <netinet/ip6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
106 #include <net/route/nhop.h>
109 #include <netipsec/ipsec_support.h>
111 #include <security/mac/mac_framework.h>
113 #define INPCBLBGROUP_SIZMIN 8
114 #define INPCBLBGROUP_SIZMAX 256
116 static struct callout ipport_tick_callout;
119 * These configure the range of local port addresses assigned to
120 * "unspecified" outgoing connections/packets/whatever.
122 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
123 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
124 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
125 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
126 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
127 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
130 * Reserved ports accessible only to root. There are significant
131 * security considerations that must be accounted for when changing these,
132 * but the security benefits can be great. Please be careful.
134 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
135 VNET_DEFINE(int, ipport_reservedlow);
137 /* Variables dealing with random ephemeral port allocation. */
138 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
140 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
141 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
142 VNET_DEFINE(int, ipport_tcpallocs);
143 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
145 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
147 static void in_pcbremlists(struct inpcb *inp);
149 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
150 struct in_addr faddr, u_int fport_arg,
151 struct in_addr laddr, u_int lport_arg,
152 int lookupflags, struct ifnet *ifp);
154 #define RANGECHK(var, min, max) \
155 if ((var) < (min)) { (var) = (min); } \
156 else if ((var) > (max)) { (var) = (max); }
159 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
163 error = sysctl_handle_int(oidp, arg1, arg2, req);
165 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
166 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
167 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
169 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
170 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
177 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
178 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
183 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
186 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
187 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
190 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
191 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
195 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
197 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
198 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
199 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
201 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
202 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
203 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
205 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
206 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
207 &VNET_NAME(ipport_reservedhigh), 0, "");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
209 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
210 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
211 CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
213 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
214 CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
216 "allocations before switching to a sequental one");
217 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
218 CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(ipport_randomtime), 0,
220 "Minimum time to keep sequental port "
221 "allocation before switching to a random one");
224 counter_u64_t rate_limit_active;
225 counter_u64_t rate_limit_alloc_fail;
226 counter_u64_t rate_limit_set_ok;
228 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
230 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
231 &rate_limit_active, "Active rate limited connections");
232 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
233 &rate_limit_alloc_fail, "Rate limited connection failures");
234 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
235 &rate_limit_set_ok, "Rate limited setting succeeded");
236 #endif /* RATELIMIT */
241 * in_pcb.c: manage the Protocol Control Blocks.
243 * NOTE: It is assumed that most of these functions will be called with
244 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
245 * functions often modify hash chains or addresses in pcbs.
248 static struct inpcblbgroup *
249 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
250 uint16_t port, const union in_dependaddr *addr, int size)
252 struct inpcblbgroup *grp;
255 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
256 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
259 grp->il_vflag = vflag;
260 grp->il_lport = port;
261 grp->il_dependladdr = *addr;
262 grp->il_inpsiz = size;
263 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
268 in_pcblbgroup_free_deferred(epoch_context_t ctx)
270 struct inpcblbgroup *grp;
272 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
277 in_pcblbgroup_free(struct inpcblbgroup *grp)
280 CK_LIST_REMOVE(grp, il_list);
281 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
284 static struct inpcblbgroup *
285 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
286 struct inpcblbgroup *old_grp, int size)
288 struct inpcblbgroup *grp;
291 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
292 old_grp->il_lport, &old_grp->il_dependladdr, size);
296 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
297 ("invalid new local group size %d and old local group count %d",
298 grp->il_inpsiz, old_grp->il_inpcnt));
300 for (i = 0; i < old_grp->il_inpcnt; ++i)
301 grp->il_inp[i] = old_grp->il_inp[i];
302 grp->il_inpcnt = old_grp->il_inpcnt;
303 in_pcblbgroup_free(old_grp);
308 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
309 * and shrink group if possible.
312 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
315 struct inpcblbgroup *grp, *new_grp;
318 for (; i + 1 < grp->il_inpcnt; ++i)
319 grp->il_inp[i] = grp->il_inp[i + 1];
322 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
323 grp->il_inpcnt <= grp->il_inpsiz / 4) {
324 /* Shrink this group. */
325 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
332 * Add PCB to load balance group for SO_REUSEPORT_LB option.
335 in_pcbinslbgrouphash(struct inpcb *inp)
337 const static struct timeval interval = { 60, 0 };
338 static struct timeval lastprint;
339 struct inpcbinfo *pcbinfo;
340 struct inpcblbgrouphead *hdr;
341 struct inpcblbgroup *grp;
344 pcbinfo = inp->inp_pcbinfo;
346 INP_WLOCK_ASSERT(inp);
347 INP_HASH_WLOCK_ASSERT(pcbinfo);
350 * Don't allow jailed socket to join local group.
352 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
357 * Don't allow IPv4 mapped INET6 wild socket.
359 if ((inp->inp_vflag & INP_IPV4) &&
360 inp->inp_laddr.s_addr == INADDR_ANY &&
361 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
366 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
367 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
368 CK_LIST_FOREACH(grp, hdr, il_list) {
369 if (grp->il_vflag == inp->inp_vflag &&
370 grp->il_lport == inp->inp_lport &&
371 memcmp(&grp->il_dependladdr,
372 &inp->inp_inc.inc_ie.ie_dependladdr,
373 sizeof(grp->il_dependladdr)) == 0)
377 /* Create new load balance group. */
378 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
379 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
380 INPCBLBGROUP_SIZMIN);
383 } else if (grp->il_inpcnt == grp->il_inpsiz) {
384 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
385 if (ratecheck(&lastprint, &interval))
386 printf("lb group port %d, limit reached\n",
387 ntohs(grp->il_lport));
391 /* Expand this local group. */
392 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
397 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
398 ("invalid local group size %d and count %d", grp->il_inpsiz,
401 grp->il_inp[grp->il_inpcnt] = inp;
407 * Remove PCB from load balance group.
410 in_pcbremlbgrouphash(struct inpcb *inp)
412 struct inpcbinfo *pcbinfo;
413 struct inpcblbgrouphead *hdr;
414 struct inpcblbgroup *grp;
417 pcbinfo = inp->inp_pcbinfo;
419 INP_WLOCK_ASSERT(inp);
420 INP_HASH_WLOCK_ASSERT(pcbinfo);
422 hdr = &pcbinfo->ipi_lbgrouphashbase[
423 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
424 CK_LIST_FOREACH(grp, hdr, il_list) {
425 for (i = 0; i < grp->il_inpcnt; ++i) {
426 if (grp->il_inp[i] != inp)
429 if (grp->il_inpcnt == 1) {
430 /* We are the last, free this local group. */
431 in_pcblbgroup_free(grp);
433 /* Pull up inpcbs, shrink group if possible. */
434 in_pcblbgroup_reorder(hdr, &grp, i);
442 * Different protocols initialize their inpcbs differently - giving
443 * different name to the lock. But they all are disposed the same.
446 inpcb_fini(void *mem, int size)
448 struct inpcb *inp = mem;
450 INP_LOCK_DESTROY(inp);
454 * Initialize an inpcbinfo -- we should be able to reduce the number of
458 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
459 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
460 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
463 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
465 INP_INFO_LOCK_INIT(pcbinfo, name);
466 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
467 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
469 pcbinfo->ipi_vnet = curvnet;
471 pcbinfo->ipi_listhead = listhead;
472 CK_LIST_INIT(pcbinfo->ipi_listhead);
473 pcbinfo->ipi_count = 0;
474 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
475 &pcbinfo->ipi_hashmask);
476 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
477 &pcbinfo->ipi_porthashmask);
478 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
479 &pcbinfo->ipi_lbgrouphashmask);
481 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
483 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
484 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
485 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
486 uma_zone_set_warning(pcbinfo->ipi_zone,
487 "kern.ipc.maxsockets limit reached");
491 * Destroy an inpcbinfo.
494 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
497 KASSERT(pcbinfo->ipi_count == 0,
498 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
500 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
501 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
502 pcbinfo->ipi_porthashmask);
503 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
504 pcbinfo->ipi_lbgrouphashmask);
506 in_pcbgroup_destroy(pcbinfo);
508 uma_zdestroy(pcbinfo->ipi_zone);
509 INP_LIST_LOCK_DESTROY(pcbinfo);
510 INP_HASH_LOCK_DESTROY(pcbinfo);
511 INP_INFO_LOCK_DESTROY(pcbinfo);
515 * Allocate a PCB and associate it with the socket.
516 * On success return with the PCB locked.
519 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
525 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
528 bzero(&inp->inp_start_zero, inp_zero_size);
530 inp->inp_numa_domain = M_NODOM;
532 inp->inp_pcbinfo = pcbinfo;
533 inp->inp_socket = so;
534 inp->inp_cred = crhold(so->so_cred);
535 inp->inp_inc.inc_fibnum = so->so_fibnum;
537 error = mac_inpcb_init(inp, M_NOWAIT);
540 mac_inpcb_create(so, inp);
542 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
543 error = ipsec_init_pcbpolicy(inp);
546 mac_inpcb_destroy(inp);
552 if (INP_SOCKAF(so) == AF_INET6) {
553 inp->inp_vflag |= INP_IPV6PROTO;
555 inp->inp_flags |= IN6P_IPV6_V6ONLY;
559 INP_LIST_WLOCK(pcbinfo);
560 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
561 pcbinfo->ipi_count++;
562 so->so_pcb = (caddr_t)inp;
564 if (V_ip6_auto_flowlabel)
565 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
567 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
568 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
571 * Routes in inpcb's can cache L2 as well; they are guaranteed
574 inp->inp_route.ro_flags = RT_LLE_CACHE;
575 INP_LIST_WUNLOCK(pcbinfo);
576 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
579 crfree(inp->inp_cred);
580 uma_zfree(pcbinfo->ipi_zone, inp);
588 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
592 INP_WLOCK_ASSERT(inp);
593 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
595 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
597 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
598 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
599 &inp->inp_lport, cred);
602 if (in_pcbinshash(inp) != 0) {
603 inp->inp_laddr.s_addr = INADDR_ANY;
608 inp->inp_flags |= INP_ANONPORT;
613 #if defined(INET) || defined(INET6)
615 * Assign a local port like in_pcb_lport(), but also used with connect()
616 * and a foreign address and port. If fsa is non-NULL, choose a local port
617 * that is unused with those, otherwise one that is completely unused.
618 * lsa can be NULL for IPv6.
621 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
622 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
624 struct inpcbinfo *pcbinfo;
625 struct inpcb *tmpinp;
626 unsigned short *lastport;
627 int count, dorandom, error;
628 u_short aux, first, last, lport;
630 struct in_addr laddr, faddr;
633 struct in6_addr *laddr6, *faddr6;
636 pcbinfo = inp->inp_pcbinfo;
639 * Because no actual state changes occur here, a global write lock on
640 * the pcbinfo isn't required.
642 INP_LOCK_ASSERT(inp);
643 INP_HASH_LOCK_ASSERT(pcbinfo);
645 if (inp->inp_flags & INP_HIGHPORT) {
646 first = V_ipport_hifirstauto; /* sysctl */
647 last = V_ipport_hilastauto;
648 lastport = &pcbinfo->ipi_lasthi;
649 } else if (inp->inp_flags & INP_LOWPORT) {
650 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
653 first = V_ipport_lowfirstauto; /* 1023 */
654 last = V_ipport_lowlastauto; /* 600 */
655 lastport = &pcbinfo->ipi_lastlow;
657 first = V_ipport_firstauto; /* sysctl */
658 last = V_ipport_lastauto;
659 lastport = &pcbinfo->ipi_lastport;
662 * For UDP(-Lite), use random port allocation as long as the user
663 * allows it. For TCP (and as of yet unknown) connections,
664 * use random port allocation only if the user allows it AND
665 * ipport_tick() allows it.
667 if (V_ipport_randomized &&
668 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
669 pcbinfo == &V_ulitecbinfo))
674 * It makes no sense to do random port allocation if
675 * we have the only port available.
679 /* Make sure to not include UDP(-Lite) packets in the count. */
680 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
681 V_ipport_tcpallocs++;
683 * Instead of having two loops further down counting up or down
684 * make sure that first is always <= last and go with only one
685 * code path implementing all logic.
694 laddr.s_addr = INADDR_ANY;
695 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
697 laddr = ((struct sockaddr_in *)lsa)->sin_addr;
699 faddr = ((struct sockaddr_in *)fsa)->sin_addr;
704 if ((inp->inp_vflag & INP_IPV6) != 0) {
706 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
708 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
716 *lastport = first + (arc4random() % (last - first));
718 count = last - first;
721 if (count-- < 0) /* completely used? */
722 return (EADDRNOTAVAIL);
724 if (*lastport < first || *lastport > last)
726 lport = htons(*lastport);
731 if (lsa->sa_family == AF_INET) {
732 tmpinp = in_pcblookup_hash_locked(pcbinfo,
733 faddr, fport, laddr, lport, lookupflags,
738 if (lsa->sa_family == AF_INET6) {
739 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
740 faddr6, fport, laddr6, lport, lookupflags,
746 if ((inp->inp_vflag & INP_IPV6) != 0)
747 tmpinp = in6_pcblookup_local(pcbinfo,
748 &inp->in6p_laddr, lport, lookupflags, cred);
750 #if defined(INET) && defined(INET6)
754 tmpinp = in_pcblookup_local(pcbinfo, laddr,
755 lport, lookupflags, cred);
758 } while (tmpinp != NULL);
766 * Select a local port (number) to use.
769 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
770 struct ucred *cred, int lookupflags)
772 struct sockaddr_in laddr;
775 bzero(&laddr, sizeof(laddr));
776 laddr.sin_family = AF_INET;
777 laddr.sin_addr = *laddrp;
779 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
780 NULL, lportp, NULL, 0, cred, lookupflags));
784 * Return cached socket options.
787 inp_so_options(const struct inpcb *inp)
793 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
794 so_options |= SO_REUSEPORT_LB;
795 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
796 so_options |= SO_REUSEPORT;
797 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
798 so_options |= SO_REUSEADDR;
801 #endif /* INET || INET6 */
804 * Check if a new BINDMULTI socket is allowed to be created.
806 * ni points to the new inp.
807 * oi points to the exisitng inp.
809 * This checks whether the existing inp also has BINDMULTI and
810 * whether the credentials match.
813 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
815 /* Check permissions match */
816 if ((ni->inp_flags2 & INP_BINDMULTI) &&
817 (ni->inp_cred->cr_uid !=
818 oi->inp_cred->cr_uid))
821 /* Check the existing inp has BINDMULTI set */
822 if ((ni->inp_flags2 & INP_BINDMULTI) &&
823 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
827 * We're okay - either INP_BINDMULTI isn't set on ni, or
828 * it is and it matches the checks.
835 * Set up a bind operation on a PCB, performing port allocation
836 * as required, but do not actually modify the PCB. Callers can
837 * either complete the bind by setting inp_laddr/inp_lport and
838 * calling in_pcbinshash(), or they can just use the resulting
839 * port and address to authorise the sending of a once-off packet.
841 * On error, the values of *laddrp and *lportp are not changed.
844 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
845 u_short *lportp, struct ucred *cred)
847 struct socket *so = inp->inp_socket;
848 struct sockaddr_in *sin;
849 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
850 struct in_addr laddr;
852 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
856 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
857 * so that we don't have to add to the (already messy) code below.
859 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
862 * No state changes, so read locks are sufficient here.
864 INP_LOCK_ASSERT(inp);
865 INP_HASH_LOCK_ASSERT(pcbinfo);
867 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
868 return (EADDRNOTAVAIL);
869 laddr.s_addr = *laddrp;
870 if (nam != NULL && laddr.s_addr != INADDR_ANY)
872 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
873 lookupflags = INPLOOKUP_WILDCARD;
875 if ((error = prison_local_ip4(cred, &laddr)) != 0)
878 sin = (struct sockaddr_in *)nam;
879 if (nam->sa_len != sizeof (*sin))
883 * We should check the family, but old programs
884 * incorrectly fail to initialize it.
886 if (sin->sin_family != AF_INET)
887 return (EAFNOSUPPORT);
889 error = prison_local_ip4(cred, &sin->sin_addr);
892 if (sin->sin_port != *lportp) {
893 /* Don't allow the port to change. */
896 lport = sin->sin_port;
898 /* NB: lport is left as 0 if the port isn't being changed. */
899 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
901 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
902 * allow complete duplication of binding if
903 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
904 * and a multicast address is bound on both
905 * new and duplicated sockets.
907 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
908 reuseport = SO_REUSEADDR|SO_REUSEPORT;
910 * XXX: How to deal with SO_REUSEPORT_LB here?
911 * Treat same as SO_REUSEPORT for now.
913 if ((so->so_options &
914 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
915 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
916 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
917 sin->sin_port = 0; /* yech... */
918 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
920 * Is the address a local IP address?
921 * If INP_BINDANY is set, then the socket may be bound
922 * to any endpoint address, local or not.
924 if ((inp->inp_flags & INP_BINDANY) == 0 &&
925 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
926 return (EADDRNOTAVAIL);
928 laddr = sin->sin_addr;
934 if (ntohs(lport) <= V_ipport_reservedhigh &&
935 ntohs(lport) >= V_ipport_reservedlow &&
936 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
938 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
939 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
940 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
941 lport, INPLOOKUP_WILDCARD, cred);
944 * This entire block sorely needs a rewrite.
947 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
948 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
949 (so->so_type != SOCK_STREAM ||
950 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
951 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
952 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
953 (t->inp_flags2 & INP_REUSEPORT) ||
954 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
955 (inp->inp_cred->cr_uid !=
956 t->inp_cred->cr_uid))
960 * If the socket is a BINDMULTI socket, then
961 * the credentials need to match and the
962 * original socket also has to have been bound
965 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
968 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
969 lport, lookupflags, cred);
970 if (t && (t->inp_flags & INP_TIMEWAIT)) {
972 * XXXRW: If an incpb has had its timewait
973 * state recycled, we treat the address as
974 * being in use (for now). This is better
975 * than a panic, but not desirable.
979 ((reuseport & tw->tw_so_options) == 0 &&
981 tw->tw_so_options) == 0)) {
985 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
986 (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)
997 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1005 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1010 *laddrp = laddr.s_addr;
1016 * Connect from a socket to a specified address.
1017 * Both address and port must be specified in argument sin.
1018 * If don't have a local address for this socket yet,
1022 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
1023 struct ucred *cred, struct mbuf *m, bool rehash)
1025 u_short lport, fport;
1026 in_addr_t laddr, faddr;
1027 int anonport, error;
1029 INP_WLOCK_ASSERT(inp);
1030 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1032 lport = inp->inp_lport;
1033 laddr = inp->inp_laddr.s_addr;
1034 anonport = (lport == 0);
1035 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &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 KASSERT(rehash == true,
1043 ("Rehashing required for unbound inps"));
1044 inp->inp_lport = lport;
1045 inp->inp_laddr.s_addr = laddr;
1046 if (in_pcbinshash(inp) != 0) {
1047 inp->inp_laddr.s_addr = INADDR_ANY;
1053 /* Commit the remaining changes. */
1054 inp->inp_lport = lport;
1055 inp->inp_laddr.s_addr = laddr;
1056 inp->inp_faddr.s_addr = faddr;
1057 inp->inp_fport = fport;
1059 in_pcbrehash_mbuf(inp, m);
1061 in_pcbinshash_mbuf(inp, m);
1065 inp->inp_flags |= INP_ANONPORT;
1070 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1073 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1077 * Do proper source address selection on an unbound socket in case
1078 * of connect. Take jails into account as well.
1081 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1085 struct sockaddr *sa;
1086 struct sockaddr_in *sin, dst;
1087 struct nhop_object *nh;
1091 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1093 * Bypass source address selection and use the primary jail IP
1096 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1102 bzero(&dst, sizeof(dst));
1104 sin->sin_family = AF_INET;
1105 sin->sin_len = sizeof(struct sockaddr_in);
1106 sin->sin_addr.s_addr = faddr->s_addr;
1109 * If route is known our src addr is taken from the i/f,
1112 * Find out route to destination.
1114 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1115 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1119 * If we found a route, use the address corresponding to
1120 * the outgoing interface.
1122 * Otherwise assume faddr is reachable on a directly connected
1123 * network and try to find a corresponding interface to take
1124 * the source address from.
1126 if (nh == NULL || nh->nh_ifp == NULL) {
1127 struct in_ifaddr *ia;
1130 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1131 inp->inp_socket->so_fibnum));
1133 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1134 inp->inp_socket->so_fibnum));
1138 error = ENETUNREACH;
1142 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1143 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1149 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1152 if (sa->sa_family != AF_INET)
1154 sin = (struct sockaddr_in *)sa;
1155 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1156 ia = (struct in_ifaddr *)ifa;
1161 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1165 /* 3. As a last resort return the 'default' jail address. */
1166 error = prison_get_ip4(cred, laddr);
1171 * If the outgoing interface on the route found is not
1172 * a loopback interface, use the address from that interface.
1173 * In case of jails do those three steps:
1174 * 1. check if the interface address belongs to the jail. If so use it.
1175 * 2. check if we have any address on the outgoing interface
1176 * belonging to this jail. If so use it.
1177 * 3. as a last resort return the 'default' jail address.
1179 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1180 struct in_ifaddr *ia;
1183 /* If not jailed, use the default returned. */
1184 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1185 ia = (struct in_ifaddr *)nh->nh_ifa;
1186 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1191 /* 1. Check if the iface address belongs to the jail. */
1192 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1193 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1194 ia = (struct in_ifaddr *)nh->nh_ifa;
1195 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1200 * 2. Check if we have any address on the outgoing interface
1201 * belonging to this jail.
1205 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1207 if (sa->sa_family != AF_INET)
1209 sin = (struct sockaddr_in *)sa;
1210 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1211 ia = (struct in_ifaddr *)ifa;
1216 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1220 /* 3. As a last resort return the 'default' jail address. */
1221 error = prison_get_ip4(cred, laddr);
1226 * The outgoing interface is marked with 'loopback net', so a route
1227 * to ourselves is here.
1228 * Try to find the interface of the destination address and then
1229 * take the address from there. That interface is not necessarily
1230 * a loopback interface.
1231 * In case of jails, check that it is an address of the jail
1232 * and if we cannot find, fall back to the 'default' jail address.
1234 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1235 struct in_ifaddr *ia;
1237 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1238 inp->inp_socket->so_fibnum));
1240 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1241 inp->inp_socket->so_fibnum));
1243 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1245 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1247 error = ENETUNREACH;
1250 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1260 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1262 if (sa->sa_family != AF_INET)
1264 sin = (struct sockaddr_in *)sa;
1265 if (prison_check_ip4(cred,
1266 &sin->sin_addr) == 0) {
1267 ia = (struct in_ifaddr *)ifa;
1272 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1277 /* 3. As a last resort return the 'default' jail address. */
1278 error = prison_get_ip4(cred, laddr);
1287 * Set up for a connect from a socket to the specified address.
1288 * On entry, *laddrp and *lportp should contain the current local
1289 * address and port for the PCB; these are updated to the values
1290 * that should be placed in inp_laddr and inp_lport to complete
1293 * On success, *faddrp and *fportp will be set to the remote address
1294 * and port. These are not updated in the error case.
1296 * If the operation fails because the connection already exists,
1297 * *oinpp will be set to the PCB of that connection so that the
1298 * caller can decide to override it. In all other cases, *oinpp
1302 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1303 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1304 struct inpcb **oinpp, struct ucred *cred)
1306 struct rm_priotracker in_ifa_tracker;
1307 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1308 struct in_ifaddr *ia;
1310 struct in_addr laddr, faddr;
1311 u_short lport, fport;
1315 * Because a global state change doesn't actually occur here, a read
1316 * lock is sufficient.
1319 INP_LOCK_ASSERT(inp);
1320 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1324 if (nam->sa_len != sizeof (*sin))
1326 if (sin->sin_family != AF_INET)
1327 return (EAFNOSUPPORT);
1328 if (sin->sin_port == 0)
1329 return (EADDRNOTAVAIL);
1330 laddr.s_addr = *laddrp;
1332 faddr = sin->sin_addr;
1333 fport = sin->sin_port;
1335 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1337 * If the destination address is INADDR_ANY,
1338 * use the primary local address.
1339 * If the supplied address is INADDR_BROADCAST,
1340 * and the primary interface supports broadcast,
1341 * choose the broadcast address for that interface.
1343 if (faddr.s_addr == INADDR_ANY) {
1344 IN_IFADDR_RLOCK(&in_ifa_tracker);
1346 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1347 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1349 (error = prison_get_ip4(cred, &faddr)) != 0)
1351 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1352 IN_IFADDR_RLOCK(&in_ifa_tracker);
1353 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1355 faddr = satosin(&CK_STAILQ_FIRST(
1356 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1357 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1360 if (laddr.s_addr == INADDR_ANY) {
1361 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1363 * If the destination address is multicast and an outgoing
1364 * interface has been set as a multicast option, prefer the
1365 * address of that interface as our source address.
1367 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1368 inp->inp_moptions != NULL) {
1369 struct ip_moptions *imo;
1372 imo = inp->inp_moptions;
1373 if (imo->imo_multicast_ifp != NULL) {
1374 ifp = imo->imo_multicast_ifp;
1375 IN_IFADDR_RLOCK(&in_ifa_tracker);
1376 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1377 if ((ia->ia_ifp == ifp) &&
1379 prison_check_ip4(cred,
1380 &ia->ia_addr.sin_addr) == 0))
1384 error = EADDRNOTAVAIL;
1386 laddr = ia->ia_addr.sin_addr;
1389 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1396 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1397 fport, laddr, lport, 0, NULL);
1401 return (EADDRINUSE);
1404 struct sockaddr_in lsin, fsin;
1406 bzero(&lsin, sizeof(lsin));
1407 bzero(&fsin, sizeof(fsin));
1408 lsin.sin_family = AF_INET;
1409 lsin.sin_addr = laddr;
1410 fsin.sin_family = AF_INET;
1411 fsin.sin_addr = faddr;
1412 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1413 &lport, (struct sockaddr *)& fsin, fport, cred,
1414 INPLOOKUP_WILDCARD);
1418 *laddrp = laddr.s_addr;
1420 *faddrp = faddr.s_addr;
1426 in_pcbdisconnect(struct inpcb *inp)
1429 INP_WLOCK_ASSERT(inp);
1430 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1432 inp->inp_faddr.s_addr = INADDR_ANY;
1439 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1440 * For most protocols, this will be invoked immediately prior to calling
1441 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1442 * socket, in which case in_pcbfree() is deferred.
1445 in_pcbdetach(struct inpcb *inp)
1448 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1451 if (inp->inp_snd_tag != NULL)
1452 in_pcbdetach_txrtlmt(inp);
1454 inp->inp_socket->so_pcb = NULL;
1455 inp->inp_socket = NULL;
1459 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1460 * stability of an inpcb pointer despite the inpcb lock being released. This
1461 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1462 * but where the inpcb lock may already held, or when acquiring a reference
1465 * in_pcbref() should be used only to provide brief memory stability, and
1466 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1467 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1468 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1469 * lock and rele are the *only* safe operations that may be performed on the
1472 * While the inpcb will not be freed, releasing the inpcb lock means that the
1473 * connection's state may change, so the caller should be careful to
1474 * revalidate any cached state on reacquiring the lock. Drop the reference
1475 * using in_pcbrele().
1478 in_pcbref(struct inpcb *inp)
1481 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1483 refcount_acquire(&inp->inp_refcount);
1487 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1488 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1489 * return a flag indicating whether or not the inpcb remains valid. If it is
1490 * valid, we return with the inpcb lock held.
1492 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1493 * reference on an inpcb. Historically more work was done here (actually, in
1494 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1495 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1496 * about memory stability (and continued use of the write lock).
1499 in_pcbrele_rlocked(struct inpcb *inp)
1501 struct inpcbinfo *pcbinfo;
1503 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1505 INP_RLOCK_ASSERT(inp);
1507 if (refcount_release(&inp->inp_refcount) == 0) {
1509 * If the inpcb has been freed, let the caller know, even if
1510 * this isn't the last reference.
1512 if (inp->inp_flags2 & INP_FREED) {
1519 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1521 if (inp->inp_in_hpts || inp->inp_in_input) {
1522 struct tcp_hpts_entry *hpts;
1524 * We should not be on the hpts at
1525 * this point in any form. we must
1526 * get the lock to be sure.
1528 hpts = tcp_hpts_lock(inp);
1529 if (inp->inp_in_hpts)
1530 panic("Hpts:%p inp:%p at free still on hpts",
1532 mtx_unlock(&hpts->p_mtx);
1533 hpts = tcp_input_lock(inp);
1534 if (inp->inp_in_input)
1535 panic("Hpts:%p inp:%p at free still on input hpts",
1537 mtx_unlock(&hpts->p_mtx);
1541 pcbinfo = inp->inp_pcbinfo;
1542 uma_zfree(pcbinfo->ipi_zone, inp);
1547 in_pcbrele_wlocked(struct inpcb *inp)
1549 struct inpcbinfo *pcbinfo;
1551 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1553 INP_WLOCK_ASSERT(inp);
1555 if (refcount_release(&inp->inp_refcount) == 0) {
1557 * If the inpcb has been freed, let the caller know, even if
1558 * this isn't the last reference.
1560 if (inp->inp_flags2 & INP_FREED) {
1567 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1569 if (inp->inp_in_hpts || inp->inp_in_input) {
1570 struct tcp_hpts_entry *hpts;
1572 * We should not be on the hpts at
1573 * this point in any form. we must
1574 * get the lock to be sure.
1576 hpts = tcp_hpts_lock(inp);
1577 if (inp->inp_in_hpts)
1578 panic("Hpts:%p inp:%p at free still on hpts",
1580 mtx_unlock(&hpts->p_mtx);
1581 hpts = tcp_input_lock(inp);
1582 if (inp->inp_in_input)
1583 panic("Hpts:%p inp:%p at free still on input hpts",
1585 mtx_unlock(&hpts->p_mtx);
1589 pcbinfo = inp->inp_pcbinfo;
1590 uma_zfree(pcbinfo->ipi_zone, inp);
1595 * Temporary wrapper.
1598 in_pcbrele(struct inpcb *inp)
1601 return (in_pcbrele_wlocked(inp));
1605 in_pcblist_rele_rlocked(epoch_context_t ctx)
1607 struct in_pcblist *il;
1609 struct inpcbinfo *pcbinfo;
1612 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1613 pcbinfo = il->il_pcbinfo;
1615 INP_INFO_WLOCK(pcbinfo);
1616 for (i = 0; i < n; i++) {
1617 inp = il->il_inp_list[i];
1619 if (!in_pcbrele_rlocked(inp))
1622 INP_INFO_WUNLOCK(pcbinfo);
1627 inpcbport_free(epoch_context_t ctx)
1629 struct inpcbport *phd;
1631 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1636 in_pcbfree_deferred(epoch_context_t ctx)
1639 int released __unused;
1641 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1644 CURVNET_SET(inp->inp_vnet);
1646 struct ip_moptions *imo = inp->inp_moptions;
1647 inp->inp_moptions = NULL;
1649 /* XXXRW: Do as much as possible here. */
1650 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1651 if (inp->inp_sp != NULL)
1652 ipsec_delete_pcbpolicy(inp);
1655 struct ip6_moptions *im6o = NULL;
1656 if (inp->inp_vflag & INP_IPV6PROTO) {
1657 ip6_freepcbopts(inp->in6p_outputopts);
1658 im6o = inp->in6p_moptions;
1659 inp->in6p_moptions = NULL;
1662 if (inp->inp_options)
1663 (void)m_free(inp->inp_options);
1665 crfree(inp->inp_cred);
1667 mac_inpcb_destroy(inp);
1669 released = in_pcbrele_wlocked(inp);
1672 ip6_freemoptions(im6o);
1675 inp_freemoptions(imo);
1681 * Unconditionally schedule an inpcb to be freed by decrementing its
1682 * reference count, which should occur only after the inpcb has been detached
1683 * from its socket. If another thread holds a temporary reference (acquired
1684 * using in_pcbref()) then the free is deferred until that reference is
1685 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1686 * work, including removal from global lists, is done in this context, where
1687 * the pcbinfo lock is held.
1690 in_pcbfree(struct inpcb *inp)
1692 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1694 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1695 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1696 ("%s: called twice for pcb %p", __func__, inp));
1697 if (inp->inp_flags2 & INP_FREED) {
1702 INP_WLOCK_ASSERT(inp);
1703 INP_LIST_WLOCK(pcbinfo);
1704 in_pcbremlists(inp);
1705 INP_LIST_WUNLOCK(pcbinfo);
1706 RO_INVALIDATE_CACHE(&inp->inp_route);
1707 /* mark as destruction in progress */
1708 inp->inp_flags2 |= INP_FREED;
1710 NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx);
1714 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1715 * port reservation, and preventing it from being returned by inpcb lookups.
1717 * It is used by TCP to mark an inpcb as unused and avoid future packet
1718 * delivery or event notification when a socket remains open but TCP has
1719 * closed. This might occur as a result of a shutdown()-initiated TCP close
1720 * or a RST on the wire, and allows the port binding to be reused while still
1721 * maintaining the invariant that so_pcb always points to a valid inpcb until
1724 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1725 * in_pcbnotifyall() and in_pcbpurgeif0()?
1728 in_pcbdrop(struct inpcb *inp)
1731 INP_WLOCK_ASSERT(inp);
1733 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1734 MPASS(inp->inp_refcount > 1);
1738 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1741 inp->inp_flags |= INP_DROPPED;
1742 if (inp->inp_flags & INP_INHASHLIST) {
1743 struct inpcbport *phd = inp->inp_phd;
1745 INP_HASH_WLOCK(inp->inp_pcbinfo);
1746 in_pcbremlbgrouphash(inp);
1747 CK_LIST_REMOVE(inp, inp_hash);
1748 CK_LIST_REMOVE(inp, inp_portlist);
1749 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1750 CK_LIST_REMOVE(phd, phd_hash);
1751 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
1753 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1754 inp->inp_flags &= ~INP_INHASHLIST;
1756 in_pcbgroup_remove(inp);
1763 * Common routines to return the socket addresses associated with inpcbs.
1766 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1768 struct sockaddr_in *sin;
1770 sin = malloc(sizeof *sin, M_SONAME,
1772 sin->sin_family = AF_INET;
1773 sin->sin_len = sizeof(*sin);
1774 sin->sin_addr = *addr_p;
1775 sin->sin_port = port;
1777 return (struct sockaddr *)sin;
1781 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1784 struct in_addr addr;
1787 inp = sotoinpcb(so);
1788 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1791 port = inp->inp_lport;
1792 addr = inp->inp_laddr;
1795 *nam = in_sockaddr(port, &addr);
1800 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1803 struct in_addr addr;
1806 inp = sotoinpcb(so);
1807 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1810 port = inp->inp_fport;
1811 addr = inp->inp_faddr;
1814 *nam = in_sockaddr(port, &addr);
1819 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1820 struct inpcb *(*notify)(struct inpcb *, int))
1822 struct inpcb *inp, *inp_temp;
1824 INP_INFO_WLOCK(pcbinfo);
1825 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1828 if ((inp->inp_vflag & INP_IPV4) == 0) {
1833 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1834 inp->inp_socket == NULL) {
1838 if ((*notify)(inp, errno))
1841 INP_INFO_WUNLOCK(pcbinfo);
1845 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1848 struct in_multi *inm;
1849 struct in_mfilter *imf;
1850 struct ip_moptions *imo;
1852 INP_INFO_WLOCK(pcbinfo);
1853 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1855 imo = inp->inp_moptions;
1856 if ((inp->inp_vflag & INP_IPV4) &&
1859 * Unselect the outgoing interface if it is being
1862 if (imo->imo_multicast_ifp == ifp)
1863 imo->imo_multicast_ifp = NULL;
1866 * Drop multicast group membership if we joined
1867 * through the interface being detached.
1869 * XXX This can all be deferred to an epoch_call
1872 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1873 if ((inm = imf->imf_inm) == NULL)
1875 if (inm->inm_ifp != ifp)
1877 ip_mfilter_remove(&imo->imo_head, imf);
1878 IN_MULTI_LOCK_ASSERT();
1879 in_leavegroup_locked(inm, NULL);
1880 ip_mfilter_free(imf);
1886 INP_INFO_WUNLOCK(pcbinfo);
1890 * Lookup a PCB based on the local address and port. Caller must hold the
1891 * hash lock. No inpcb locks or references are acquired.
1893 #define INP_LOOKUP_MAPPED_PCB_COST 3
1895 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1896 u_short lport, int lookupflags, struct ucred *cred)
1900 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1906 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1907 ("%s: invalid lookup flags %d", __func__, lookupflags));
1909 INP_HASH_LOCK_ASSERT(pcbinfo);
1911 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1912 struct inpcbhead *head;
1914 * Look for an unconnected (wildcard foreign addr) PCB that
1915 * matches the local address and port we're looking for.
1917 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1918 0, pcbinfo->ipi_hashmask)];
1919 CK_LIST_FOREACH(inp, head, inp_hash) {
1921 /* XXX inp locking */
1922 if ((inp->inp_vflag & INP_IPV4) == 0)
1925 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1926 inp->inp_laddr.s_addr == laddr.s_addr &&
1927 inp->inp_lport == lport) {
1932 prison_equal_ip4(cred->cr_prison,
1933 inp->inp_cred->cr_prison))
1942 struct inpcbporthead *porthash;
1943 struct inpcbport *phd;
1944 struct inpcb *match = NULL;
1946 * Best fit PCB lookup.
1948 * First see if this local port is in use by looking on the
1951 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1952 pcbinfo->ipi_porthashmask)];
1953 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1954 if (phd->phd_port == lport)
1959 * Port is in use by one or more PCBs. Look for best
1962 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1965 !prison_equal_ip4(inp->inp_cred->cr_prison,
1969 /* XXX inp locking */
1970 if ((inp->inp_vflag & INP_IPV4) == 0)
1973 * We never select the PCB that has
1974 * INP_IPV6 flag and is bound to :: if
1975 * we have another PCB which is bound
1976 * to 0.0.0.0. If a PCB has the
1977 * INP_IPV6 flag, then we set its cost
1978 * higher than IPv4 only PCBs.
1980 * Note that the case only happens
1981 * when a socket is bound to ::, under
1982 * the condition that the use of the
1983 * mapped address is allowed.
1985 if ((inp->inp_vflag & INP_IPV6) != 0)
1986 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1988 if (inp->inp_faddr.s_addr != INADDR_ANY)
1990 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1991 if (laddr.s_addr == INADDR_ANY)
1993 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1996 if (laddr.s_addr != INADDR_ANY)
1999 if (wildcard < matchwild) {
2001 matchwild = wildcard;
2010 #undef INP_LOOKUP_MAPPED_PCB_COST
2012 static struct inpcb *
2013 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2014 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
2015 uint16_t fport, int lookupflags)
2017 struct inpcb *local_wild;
2018 const struct inpcblbgrouphead *hdr;
2019 struct inpcblbgroup *grp;
2022 INP_HASH_LOCK_ASSERT(pcbinfo);
2024 hdr = &pcbinfo->ipi_lbgrouphashbase[
2025 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2028 * Order of socket selection:
2030 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
2033 * - Load balanced group does not contain jailed sockets
2034 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
2037 CK_LIST_FOREACH(grp, hdr, il_list) {
2039 if (!(grp->il_vflag & INP_IPV4))
2042 if (grp->il_lport != lport)
2045 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
2047 if (grp->il_laddr.s_addr == laddr->s_addr)
2048 return (grp->il_inp[idx]);
2049 if (grp->il_laddr.s_addr == INADDR_ANY &&
2050 (lookupflags & INPLOOKUP_WILDCARD) != 0)
2051 local_wild = grp->il_inp[idx];
2053 return (local_wild);
2058 * Lookup PCB in hash list, using pcbgroup tables.
2060 static struct inpcb *
2061 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2062 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2063 u_int lport_arg, int lookupflags, struct ifnet *ifp)
2065 struct inpcbhead *head;
2066 struct inpcb *inp, *tmpinp;
2067 u_short fport = fport_arg, lport = lport_arg;
2071 * First look for an exact match.
2074 INP_GROUP_LOCK(pcbgroup);
2075 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2076 pcbgroup->ipg_hashmask)];
2077 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2079 /* XXX inp locking */
2080 if ((inp->inp_vflag & INP_IPV4) == 0)
2083 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2084 inp->inp_laddr.s_addr == laddr.s_addr &&
2085 inp->inp_fport == fport &&
2086 inp->inp_lport == lport) {
2088 * XXX We should be able to directly return
2089 * the inp here, without any checks.
2090 * Well unless both bound with SO_REUSEPORT?
2092 if (prison_flag(inp->inp_cred, PR_IP4))
2098 if (tmpinp != NULL) {
2105 * For incoming connections, we may wish to do a wildcard
2106 * match for an RSS-local socket.
2108 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2109 struct inpcb *local_wild = NULL, *local_exact = NULL;
2111 struct inpcb *local_wild_mapped = NULL;
2113 struct inpcb *jail_wild = NULL;
2114 struct inpcbhead *head;
2118 * Order of socket selection - we always prefer jails.
2119 * 1. jailed, non-wild.
2121 * 3. non-jailed, non-wild.
2122 * 4. non-jailed, wild.
2125 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2126 lport, 0, pcbgroup->ipg_hashmask)];
2127 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2129 /* XXX inp locking */
2130 if ((inp->inp_vflag & INP_IPV4) == 0)
2133 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2134 inp->inp_lport != lport)
2137 injail = prison_flag(inp->inp_cred, PR_IP4);
2139 if (prison_check_ip4(inp->inp_cred,
2143 if (local_exact != NULL)
2147 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2152 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2154 /* XXX inp locking, NULL check */
2155 if (inp->inp_vflag & INP_IPV6PROTO)
2156 local_wild_mapped = inp;
2164 } /* LIST_FOREACH */
2173 inp = local_wild_mapped;
2181 * Then look for a wildcard match, if requested.
2183 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2184 struct inpcb *local_wild = NULL, *local_exact = NULL;
2186 struct inpcb *local_wild_mapped = NULL;
2188 struct inpcb *jail_wild = NULL;
2189 struct inpcbhead *head;
2193 * Order of socket selection - we always prefer jails.
2194 * 1. jailed, non-wild.
2196 * 3. non-jailed, non-wild.
2197 * 4. non-jailed, wild.
2199 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2200 0, pcbinfo->ipi_wildmask)];
2201 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2203 /* XXX inp locking */
2204 if ((inp->inp_vflag & INP_IPV4) == 0)
2207 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2208 inp->inp_lport != lport)
2211 injail = prison_flag(inp->inp_cred, PR_IP4);
2213 if (prison_check_ip4(inp->inp_cred,
2217 if (local_exact != NULL)
2221 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2226 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2228 /* XXX inp locking, NULL check */
2229 if (inp->inp_vflag & INP_IPV6PROTO)
2230 local_wild_mapped = inp;
2238 } /* LIST_FOREACH */
2246 inp = local_wild_mapped;
2250 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2251 INP_GROUP_UNLOCK(pcbgroup);
2255 if (lookupflags & INPLOOKUP_WLOCKPCB)
2256 locked = INP_TRY_WLOCK(inp);
2257 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2258 locked = INP_TRY_RLOCK(inp);
2260 panic("%s: locking bug", __func__);
2261 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2262 if (lookupflags & INPLOOKUP_WLOCKPCB)
2269 INP_GROUP_UNLOCK(pcbgroup);
2271 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2273 if (in_pcbrele_wlocked(inp))
2277 if (in_pcbrele_rlocked(inp))
2282 if (lookupflags & INPLOOKUP_WLOCKPCB)
2283 INP_WLOCK_ASSERT(inp);
2285 INP_RLOCK_ASSERT(inp);
2289 #endif /* PCBGROUP */
2292 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2293 * that the caller has locked the hash list, and will not perform any further
2294 * locking or reference operations on either the hash list or the connection.
2296 static struct inpcb *
2297 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2298 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2301 struct inpcbhead *head;
2302 struct inpcb *inp, *tmpinp;
2303 u_short fport = fport_arg, lport = lport_arg;
2305 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2306 ("%s: invalid lookup flags %d", __func__, lookupflags));
2307 INP_HASH_LOCK_ASSERT(pcbinfo);
2310 * First look for an exact match.
2313 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2314 pcbinfo->ipi_hashmask)];
2315 CK_LIST_FOREACH(inp, head, inp_hash) {
2317 /* XXX inp locking */
2318 if ((inp->inp_vflag & INP_IPV4) == 0)
2321 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2322 inp->inp_laddr.s_addr == laddr.s_addr &&
2323 inp->inp_fport == fport &&
2324 inp->inp_lport == lport) {
2326 * XXX We should be able to directly return
2327 * the inp here, without any checks.
2328 * Well unless both bound with SO_REUSEPORT?
2330 if (prison_flag(inp->inp_cred, PR_IP4))
2340 * Then look in lb group (for wildcard match).
2342 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2343 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2344 fport, lookupflags);
2350 * Then look for a wildcard match, if requested.
2352 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2353 struct inpcb *local_wild = NULL, *local_exact = NULL;
2355 struct inpcb *local_wild_mapped = NULL;
2357 struct inpcb *jail_wild = NULL;
2361 * Order of socket selection - we always prefer jails.
2362 * 1. jailed, non-wild.
2364 * 3. non-jailed, non-wild.
2365 * 4. non-jailed, wild.
2368 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2369 0, pcbinfo->ipi_hashmask)];
2370 CK_LIST_FOREACH(inp, head, inp_hash) {
2372 /* XXX inp locking */
2373 if ((inp->inp_vflag & INP_IPV4) == 0)
2376 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2377 inp->inp_lport != lport)
2380 injail = prison_flag(inp->inp_cred, PR_IP4);
2382 if (prison_check_ip4(inp->inp_cred,
2386 if (local_exact != NULL)
2390 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2395 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2397 /* XXX inp locking, NULL check */
2398 if (inp->inp_vflag & INP_IPV6PROTO)
2399 local_wild_mapped = inp;
2407 } /* LIST_FOREACH */
2408 if (jail_wild != NULL)
2410 if (local_exact != NULL)
2411 return (local_exact);
2412 if (local_wild != NULL)
2413 return (local_wild);
2415 if (local_wild_mapped != NULL)
2416 return (local_wild_mapped);
2418 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2424 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2425 * hash list lock, and will return the inpcb locked (i.e., requires
2426 * INPLOOKUP_LOCKPCB).
2428 static struct inpcb *
2429 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2430 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2435 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2436 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2438 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2440 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2444 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2446 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2451 panic("%s: locking bug", __func__);
2454 if (lookupflags & INPLOOKUP_WLOCKPCB)
2455 INP_WLOCK_ASSERT(inp);
2457 INP_RLOCK_ASSERT(inp);
2466 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2467 * from which a pre-calculated hash value may be extracted.
2469 * Possibly more of this logic should be in in_pcbgroup.c.
2472 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2473 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2475 #if defined(PCBGROUP) && !defined(RSS)
2476 struct inpcbgroup *pcbgroup;
2479 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2480 ("%s: invalid lookup flags %d", __func__, lookupflags));
2481 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2482 ("%s: LOCKPCB not set", __func__));
2485 * When not using RSS, use connection groups in preference to the
2486 * reservation table when looking up 4-tuples. When using RSS, just
2487 * use the reservation table, due to the cost of the Toeplitz hash
2490 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2491 * we could be doing RSS with a non-Toeplitz hash that is affordable
2494 #if defined(PCBGROUP) && !defined(RSS)
2495 if (in_pcbgroup_enabled(pcbinfo)) {
2496 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2498 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2499 laddr, lport, lookupflags, ifp));
2502 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2507 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2508 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2509 struct ifnet *ifp, struct mbuf *m)
2512 struct inpcbgroup *pcbgroup;
2515 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2516 ("%s: invalid lookup flags %d", __func__, lookupflags));
2517 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2518 ("%s: LOCKPCB not set", __func__));
2522 * If we can use a hardware-generated hash to look up the connection
2523 * group, use that connection group to find the inpcb. Otherwise
2524 * fall back on a software hash -- or the reservation table if we're
2527 * XXXRW: As above, that policy belongs in the pcbgroup code.
2529 if (in_pcbgroup_enabled(pcbinfo) &&
2530 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2531 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2532 m->m_pkthdr.flowid);
2533 if (pcbgroup != NULL)
2534 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2535 fport, laddr, lport, lookupflags, ifp));
2537 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2539 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2540 laddr, lport, lookupflags, ifp));
2544 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2550 * Insert PCB onto various hash lists.
2553 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2555 struct inpcbhead *pcbhash;
2556 struct inpcbporthead *pcbporthash;
2557 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2558 struct inpcbport *phd;
2559 u_int32_t hashkey_faddr;
2562 INP_WLOCK_ASSERT(inp);
2563 INP_HASH_WLOCK_ASSERT(pcbinfo);
2565 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2566 ("in_pcbinshash: INP_INHASHLIST"));
2569 if (inp->inp_vflag & INP_IPV6)
2570 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2573 hashkey_faddr = inp->inp_faddr.s_addr;
2575 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2576 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2578 pcbporthash = &pcbinfo->ipi_porthashbase[
2579 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2582 * Add entry to load balance group.
2583 * Only do this if SO_REUSEPORT_LB is set.
2585 so_options = inp_so_options(inp);
2586 if (so_options & SO_REUSEPORT_LB) {
2587 int ret = in_pcbinslbgrouphash(inp);
2589 /* pcb lb group malloc fail (ret=ENOBUFS). */
2595 * Go through port list and look for a head for this lport.
2597 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2598 if (phd->phd_port == inp->inp_lport)
2602 * If none exists, malloc one and tack it on.
2605 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2607 return (ENOBUFS); /* XXX */
2609 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2610 phd->phd_port = inp->inp_lport;
2611 CK_LIST_INIT(&phd->phd_pcblist);
2612 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2615 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2616 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2617 inp->inp_flags |= INP_INHASHLIST;
2620 in_pcbgroup_update_mbuf(inp, m);
2622 in_pcbgroup_update(inp);
2629 in_pcbinshash(struct inpcb *inp)
2632 return (in_pcbinshash_internal(inp, NULL));
2636 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2639 return (in_pcbinshash_internal(inp, m));
2643 * Move PCB to the proper hash bucket when { faddr, fport } have been
2644 * changed. NOTE: This does not handle the case of the lport changing (the
2645 * hashed port list would have to be updated as well), so the lport must
2646 * not change after in_pcbinshash() has been called.
2649 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2651 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2652 struct inpcbhead *head;
2653 u_int32_t hashkey_faddr;
2655 INP_WLOCK_ASSERT(inp);
2656 INP_HASH_WLOCK_ASSERT(pcbinfo);
2658 KASSERT(inp->inp_flags & INP_INHASHLIST,
2659 ("in_pcbrehash: !INP_INHASHLIST"));
2662 if (inp->inp_vflag & INP_IPV6)
2663 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2666 hashkey_faddr = inp->inp_faddr.s_addr;
2668 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2669 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2671 CK_LIST_REMOVE(inp, inp_hash);
2672 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2676 in_pcbgroup_update_mbuf(inp, m);
2678 in_pcbgroup_update(inp);
2683 in_pcbrehash(struct inpcb *inp)
2686 in_pcbrehash_mbuf(inp, NULL);
2690 * Remove PCB from various lists.
2693 in_pcbremlists(struct inpcb *inp)
2695 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2697 INP_WLOCK_ASSERT(inp);
2698 INP_LIST_WLOCK_ASSERT(pcbinfo);
2700 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2701 if (inp->inp_flags & INP_INHASHLIST) {
2702 struct inpcbport *phd = inp->inp_phd;
2704 INP_HASH_WLOCK(pcbinfo);
2706 /* XXX: Only do if SO_REUSEPORT_LB set? */
2707 in_pcbremlbgrouphash(inp);
2709 CK_LIST_REMOVE(inp, inp_hash);
2710 CK_LIST_REMOVE(inp, inp_portlist);
2711 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2712 CK_LIST_REMOVE(phd, phd_hash);
2713 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
2715 INP_HASH_WUNLOCK(pcbinfo);
2716 inp->inp_flags &= ~INP_INHASHLIST;
2718 CK_LIST_REMOVE(inp, inp_list);
2719 pcbinfo->ipi_count--;
2721 in_pcbgroup_remove(inp);
2726 * Check for alternatives when higher level complains
2727 * about service problems. For now, invalidate cached
2728 * routing information. If the route was created dynamically
2729 * (by a redirect), time to try a default gateway again.
2732 in_losing(struct inpcb *inp)
2735 RO_INVALIDATE_CACHE(&inp->inp_route);
2740 * A set label operation has occurred at the socket layer, propagate the
2741 * label change into the in_pcb for the socket.
2744 in_pcbsosetlabel(struct socket *so)
2749 inp = sotoinpcb(so);
2750 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2754 mac_inpcb_sosetlabel(so, inp);
2761 * ipport_tick runs once per second, determining if random port allocation
2762 * should be continued. If more than ipport_randomcps ports have been
2763 * allocated in the last second, then we return to sequential port
2764 * allocation. We return to random allocation only once we drop below
2765 * ipport_randomcps for at least ipport_randomtime seconds.
2768 ipport_tick(void *xtp)
2770 VNET_ITERATOR_DECL(vnet_iter);
2772 VNET_LIST_RLOCK_NOSLEEP();
2773 VNET_FOREACH(vnet_iter) {
2774 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2775 if (V_ipport_tcpallocs <=
2776 V_ipport_tcplastcount + V_ipport_randomcps) {
2777 if (V_ipport_stoprandom > 0)
2778 V_ipport_stoprandom--;
2780 V_ipport_stoprandom = V_ipport_randomtime;
2781 V_ipport_tcplastcount = V_ipport_tcpallocs;
2784 VNET_LIST_RUNLOCK_NOSLEEP();
2785 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2792 callout_stop(&ipport_tick_callout);
2796 * The ipport_callout should start running at about the time we attach the
2797 * inet or inet6 domains.
2800 ipport_tick_init(const void *unused __unused)
2803 /* Start ipport_tick. */
2804 callout_init(&ipport_tick_callout, 1);
2805 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2806 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2807 SHUTDOWN_PRI_DEFAULT);
2809 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2810 ipport_tick_init, NULL);
2813 inp_wlock(struct inpcb *inp)
2820 inp_wunlock(struct inpcb *inp)
2827 inp_rlock(struct inpcb *inp)
2834 inp_runlock(struct inpcb *inp)
2840 #ifdef INVARIANT_SUPPORT
2842 inp_lock_assert(struct inpcb *inp)
2845 INP_WLOCK_ASSERT(inp);
2849 inp_unlock_assert(struct inpcb *inp)
2852 INP_UNLOCK_ASSERT(inp);
2857 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2861 INP_INFO_WLOCK(&V_tcbinfo);
2862 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2867 INP_INFO_WUNLOCK(&V_tcbinfo);
2871 inp_inpcbtosocket(struct inpcb *inp)
2874 INP_WLOCK_ASSERT(inp);
2875 return (inp->inp_socket);
2879 inp_inpcbtotcpcb(struct inpcb *inp)
2882 INP_WLOCK_ASSERT(inp);
2883 return ((struct tcpcb *)inp->inp_ppcb);
2887 inp_ip_tos_get(const struct inpcb *inp)
2890 return (inp->inp_ip_tos);
2894 inp_ip_tos_set(struct inpcb *inp, int val)
2897 inp->inp_ip_tos = val;
2901 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2902 uint32_t *faddr, uint16_t *fp)
2905 INP_LOCK_ASSERT(inp);
2906 *laddr = inp->inp_laddr.s_addr;
2907 *faddr = inp->inp_faddr.s_addr;
2908 *lp = inp->inp_lport;
2909 *fp = inp->inp_fport;
2913 so_sotoinpcb(struct socket *so)
2916 return (sotoinpcb(so));
2920 so_sototcpcb(struct socket *so)
2923 return (sototcpcb(so));
2927 * Create an external-format (``xinpcb'') structure using the information in
2928 * the kernel-format in_pcb structure pointed to by inp. This is done to
2929 * reduce the spew of irrelevant information over this interface, to isolate
2930 * user code from changes in the kernel structure, and potentially to provide
2931 * information-hiding if we decide that some of this information should be
2932 * hidden from users.
2935 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2938 bzero(xi, sizeof(*xi));
2939 xi->xi_len = sizeof(struct xinpcb);
2940 if (inp->inp_socket)
2941 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2942 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2943 xi->inp_gencnt = inp->inp_gencnt;
2944 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2945 xi->inp_flow = inp->inp_flow;
2946 xi->inp_flowid = inp->inp_flowid;
2947 xi->inp_flowtype = inp->inp_flowtype;
2948 xi->inp_flags = inp->inp_flags;
2949 xi->inp_flags2 = inp->inp_flags2;
2950 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2951 xi->in6p_cksum = inp->in6p_cksum;
2952 xi->in6p_hops = inp->in6p_hops;
2953 xi->inp_ip_tos = inp->inp_ip_tos;
2954 xi->inp_vflag = inp->inp_vflag;
2955 xi->inp_ip_ttl = inp->inp_ip_ttl;
2956 xi->inp_ip_p = inp->inp_ip_p;
2957 xi->inp_ip_minttl = inp->inp_ip_minttl;
2962 db_print_indent(int indent)
2966 for (i = 0; i < indent; i++)
2971 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2973 char faddr_str[48], laddr_str[48];
2975 db_print_indent(indent);
2976 db_printf("%s at %p\n", name, inc);
2981 if (inc->inc_flags & INC_ISIPV6) {
2983 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2984 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2989 inet_ntoa_r(inc->inc_laddr, laddr_str);
2990 inet_ntoa_r(inc->inc_faddr, faddr_str);
2992 db_print_indent(indent);
2993 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2994 ntohs(inc->inc_lport));
2995 db_print_indent(indent);
2996 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2997 ntohs(inc->inc_fport));
3001 db_print_inpflags(int inp_flags)
3006 if (inp_flags & INP_RECVOPTS) {
3007 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
3010 if (inp_flags & INP_RECVRETOPTS) {
3011 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
3014 if (inp_flags & INP_RECVDSTADDR) {
3015 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
3018 if (inp_flags & INP_ORIGDSTADDR) {
3019 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
3022 if (inp_flags & INP_HDRINCL) {
3023 db_printf("%sINP_HDRINCL", comma ? ", " : "");
3026 if (inp_flags & INP_HIGHPORT) {
3027 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3030 if (inp_flags & INP_LOWPORT) {
3031 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3034 if (inp_flags & INP_ANONPORT) {
3035 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3038 if (inp_flags & INP_RECVIF) {
3039 db_printf("%sINP_RECVIF", comma ? ", " : "");
3042 if (inp_flags & INP_MTUDISC) {
3043 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3046 if (inp_flags & INP_RECVTTL) {
3047 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3050 if (inp_flags & INP_DONTFRAG) {
3051 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3054 if (inp_flags & INP_RECVTOS) {
3055 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3058 if (inp_flags & IN6P_IPV6_V6ONLY) {
3059 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3062 if (inp_flags & IN6P_PKTINFO) {
3063 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3066 if (inp_flags & IN6P_HOPLIMIT) {
3067 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3070 if (inp_flags & IN6P_HOPOPTS) {
3071 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3074 if (inp_flags & IN6P_DSTOPTS) {
3075 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3078 if (inp_flags & IN6P_RTHDR) {
3079 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3082 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3083 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3086 if (inp_flags & IN6P_TCLASS) {
3087 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3090 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3091 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3094 if (inp_flags & INP_TIMEWAIT) {
3095 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3098 if (inp_flags & INP_ONESBCAST) {
3099 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3102 if (inp_flags & INP_DROPPED) {
3103 db_printf("%sINP_DROPPED", comma ? ", " : "");
3106 if (inp_flags & INP_SOCKREF) {
3107 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3110 if (inp_flags & IN6P_RFC2292) {
3111 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3114 if (inp_flags & IN6P_MTU) {
3115 db_printf("IN6P_MTU%s", comma ? ", " : "");
3121 db_print_inpvflag(u_char inp_vflag)
3126 if (inp_vflag & INP_IPV4) {
3127 db_printf("%sINP_IPV4", comma ? ", " : "");
3130 if (inp_vflag & INP_IPV6) {
3131 db_printf("%sINP_IPV6", comma ? ", " : "");
3134 if (inp_vflag & INP_IPV6PROTO) {
3135 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3141 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3144 db_print_indent(indent);
3145 db_printf("%s at %p\n", name, inp);
3149 db_print_indent(indent);
3150 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3152 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3154 db_print_indent(indent);
3155 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3156 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3158 db_print_indent(indent);
3159 db_printf("inp_label: %p inp_flags: 0x%x (",
3160 inp->inp_label, inp->inp_flags);
3161 db_print_inpflags(inp->inp_flags);
3164 db_print_indent(indent);
3165 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3167 db_print_inpvflag(inp->inp_vflag);
3170 db_print_indent(indent);
3171 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3172 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3174 db_print_indent(indent);
3176 if (inp->inp_vflag & INP_IPV6) {
3177 db_printf("in6p_options: %p in6p_outputopts: %p "
3178 "in6p_moptions: %p\n", inp->in6p_options,
3179 inp->in6p_outputopts, inp->in6p_moptions);
3180 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3181 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3186 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3187 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3188 inp->inp_options, inp->inp_moptions);
3191 db_print_indent(indent);
3192 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3193 (uintmax_t)inp->inp_gencnt);
3196 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3201 db_printf("usage: show inpcb <addr>\n");
3204 inp = (struct inpcb *)addr;
3206 db_print_inpcb(inp, "inpcb", 0);
3212 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3216 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3218 union if_snd_tag_modify_params params = {
3219 .rate_limit.max_rate = max_pacing_rate,
3220 .rate_limit.flags = M_NOWAIT,
3222 struct m_snd_tag *mst;
3226 mst = inp->inp_snd_tag;
3234 if (ifp->if_snd_tag_modify == NULL) {
3237 error = ifp->if_snd_tag_modify(mst, ¶ms);
3243 * Query existing TX rate limit based on the existing
3244 * "inp->inp_snd_tag", if any.
3247 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3249 union if_snd_tag_query_params params = { };
3250 struct m_snd_tag *mst;
3254 mst = inp->inp_snd_tag;
3262 if (ifp->if_snd_tag_query == NULL) {
3265 error = ifp->if_snd_tag_query(mst, ¶ms);
3266 if (error == 0 && p_max_pacing_rate != NULL)
3267 *p_max_pacing_rate = params.rate_limit.max_rate;
3273 * Query existing TX queue level based on the existing
3274 * "inp->inp_snd_tag", if any.
3277 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3279 union if_snd_tag_query_params params = { };
3280 struct m_snd_tag *mst;
3284 mst = inp->inp_snd_tag;
3292 if (ifp->if_snd_tag_query == NULL)
3293 return (EOPNOTSUPP);
3295 error = ifp->if_snd_tag_query(mst, ¶ms);
3296 if (error == 0 && p_txqueue_level != NULL)
3297 *p_txqueue_level = params.rate_limit.queue_level;
3302 * Allocate a new TX rate limit send tag from the network interface
3303 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3306 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3307 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3310 union if_snd_tag_alloc_params params = {
3311 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3312 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3313 .rate_limit.hdr.flowid = flowid,
3314 .rate_limit.hdr.flowtype = flowtype,
3315 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3316 .rate_limit.max_rate = max_pacing_rate,
3317 .rate_limit.flags = M_NOWAIT,
3321 INP_WLOCK_ASSERT(inp);
3326 if (ifp->if_snd_tag_alloc == NULL) {
3329 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3333 counter_u64_add(rate_limit_set_ok, 1);
3334 counter_u64_add(rate_limit_active, 1);
3336 counter_u64_add(rate_limit_alloc_fail, 1);
3343 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3349 * If the device was detached while we still had reference(s)
3350 * on the ifp, we assume if_snd_tag_free() was replaced with
3353 ifp->if_snd_tag_free(mst);
3355 /* release reference count on network interface */
3358 counter_u64_add(rate_limit_active, -1);
3363 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3367 in_pcbdetach_txrtlmt(struct inpcb *inp)
3369 struct m_snd_tag *mst;
3371 INP_WLOCK_ASSERT(inp);
3373 mst = inp->inp_snd_tag;
3374 inp->inp_snd_tag = NULL;
3379 m_snd_tag_rele(mst);
3383 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3388 * If the existing send tag is for the wrong interface due to
3389 * a route change, first drop the existing tag. Set the
3390 * CHANGED flag so that we will keep trying to allocate a new
3391 * tag if we fail to allocate one this time.
3393 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3394 in_pcbdetach_txrtlmt(inp);
3395 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3399 * NOTE: When attaching to a network interface a reference is
3400 * made to ensure the network interface doesn't go away until
3401 * all ratelimit connections are gone. The network interface
3402 * pointers compared below represent valid network interfaces,
3403 * except when comparing towards NULL.
3405 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3407 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3408 if (inp->inp_snd_tag != NULL)
3409 in_pcbdetach_txrtlmt(inp);
3411 } else if (inp->inp_snd_tag == NULL) {
3413 * In order to utilize packet pacing with RSS, we need
3414 * to wait until there is a valid RSS hash before we
3417 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3420 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3421 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3424 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3426 if (error == 0 || error == EOPNOTSUPP)
3427 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3433 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3434 * is set in the fast path and will attach/detach/modify the TX rate
3435 * limit send tag based on the socket's so_max_pacing_rate value.
3438 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3440 struct socket *socket;
3441 uint32_t max_pacing_rate;
3448 socket = inp->inp_socket;
3452 if (!INP_WLOCKED(inp)) {
3454 * NOTE: If the write locking fails, we need to bail
3455 * out and use the non-ratelimited ring for the
3456 * transmit until there is a new chance to get the
3459 if (!INP_TRY_UPGRADE(inp))
3467 * NOTE: The so_max_pacing_rate value is read unlocked,
3468 * because atomic updates are not required since the variable
3469 * is checked at every mbuf we send. It is assumed that the
3470 * variable read itself will be atomic.
3472 max_pacing_rate = socket->so_max_pacing_rate;
3474 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3481 * Track route changes for TX rate limiting.
3484 in_pcboutput_eagain(struct inpcb *inp)
3491 if (inp->inp_snd_tag == NULL)
3494 if (!INP_WLOCKED(inp)) {
3496 * NOTE: If the write locking fails, we need to bail
3497 * out and use the non-ratelimited ring for the
3498 * transmit until there is a new chance to get the
3501 if (!INP_TRY_UPGRADE(inp))
3508 /* detach rate limiting */
3509 in_pcbdetach_txrtlmt(inp);
3511 /* make sure new mbuf send tag allocation is made */
3512 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3522 rate_limit_active = counter_u64_alloc(M_WAITOK);
3523 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3524 rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3527 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3529 #endif /* RATELIMIT */