1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
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 * All rights reserved.
9 *
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
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.
24 *
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
35 * SUCH DAMAGE.
36 *
37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95
38 */
39
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42
43 #include "opt_ddb.h"
44 #include "opt_ipsec.h"
45 #include "opt_inet.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
49 #include "opt_rss.h"
50
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.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>
61 #include <sys/smp.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/refcount.h>
68 #include <sys/jail.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
71
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75
76 #include <vm/uma.h>
77
78 #include <net/if.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>
84 #include <net/vnet.h>
85
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
89 #ifdef INET
90 #include <netinet/in_var.h>
91 #endif
92 #include <netinet/ip_var.h>
93 #include <netinet/tcp_var.h>
94 #ifdef TCPHPTS
95 #include <netinet/tcp_hpts.h>
96 #endif
97 #include <netinet/udp.h>
98 #include <netinet/udp_var.h>
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #endif /* INET6 */
105 #endif
106
107 #include <netipsec/ipsec_support.h>
108
109 #include <security/mac/mac_framework.h>
110
111 #define INPCBLBGROUP_SIZMIN 8
112 #define INPCBLBGROUP_SIZMAX 256
113
114 static struct callout ipport_tick_callout;
115
116 /*
117 * These configure the range of local port addresses assigned to
118 * "unspecified" outgoing connections/packets/whatever.
119 */
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */
126
127 /*
128 * Reserved ports accessible only to root. There are significant
129 * security considerations that must be accounted for when changing these,
130 * but the security benefits can be great. Please be careful.
131 */
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
134
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
142
143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount)
144
145 static void in_pcbremlists(struct inpcb *inp);
146 #ifdef INET
147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 struct in_addr faddr, u_int fport_arg,
149 struct in_addr laddr, u_int lport_arg,
150 int lookupflags, struct ifnet *ifp);
151
152 #define RANGECHK(var, min, max) \
153 if ((var) < (min)) { (var) = (min); } \
154 else if ((var) > (max)) { (var) = (max); }
155
156 static int
sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
158 {
159 int error;
160
161 error = sysctl_handle_int(oidp, arg1, arg2, req);
162 if (error == 0) {
163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
169 }
170 return (error);
171 }
172
173 #undef RANGECHK
174
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
176 "IP Ports");
177
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 &VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 CTLFLAG_VNET | CTLFLAG_RW,
206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 "allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(ipport_randomtime), 0,
211 "Minimum time to keep sequental port "
212 "allocation before switching to a random one");
213 #endif /* INET */
214
215 /*
216 * in_pcb.c: manage the Protocol Control Blocks.
217 *
218 * NOTE: It is assumed that most of these functions will be called with
219 * the pcbinfo lock held, and often, the inpcb lock held, as these utility
220 * functions often modify hash chains or addresses in pcbs.
221 */
222
223 static struct inpcblbgroup *
in_pcblbgroup_alloc(struct inpcblbgrouphead * hdr,u_char vflag,uint16_t port,const union in_dependaddr * addr,int size)224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
225 uint16_t port, const union in_dependaddr *addr, int size)
226 {
227 struct inpcblbgroup *grp;
228 size_t bytes;
229
230 bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
231 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
232 if (!grp)
233 return (NULL);
234 grp->il_vflag = vflag;
235 grp->il_lport = port;
236 grp->il_dependladdr = *addr;
237 grp->il_inpsiz = size;
238 CK_LIST_INSERT_HEAD(hdr, grp, il_list);
239 return (grp);
240 }
241
242 static void
in_pcblbgroup_free_deferred(epoch_context_t ctx)243 in_pcblbgroup_free_deferred(epoch_context_t ctx)
244 {
245 struct inpcblbgroup *grp;
246
247 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
248 free(grp, M_PCB);
249 }
250
251 static void
in_pcblbgroup_free(struct inpcblbgroup * grp)252 in_pcblbgroup_free(struct inpcblbgroup *grp)
253 {
254
255 CK_LIST_REMOVE(grp, il_list);
256 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
257 in_pcblbgroup_free_deferred);
258 }
259
260 static struct inpcblbgroup *
in_pcblbgroup_resize(struct inpcblbgrouphead * hdr,struct inpcblbgroup * old_grp,int size)261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
262 struct inpcblbgroup *old_grp, int size)
263 {
264 struct inpcblbgroup *grp;
265 int i;
266
267 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
268 old_grp->il_lport, &old_grp->il_dependladdr, size);
269 if (grp == NULL)
270 return (NULL);
271
272 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
273 ("invalid new local group size %d and old local group count %d",
274 grp->il_inpsiz, old_grp->il_inpcnt));
275
276 for (i = 0; i < old_grp->il_inpcnt; ++i)
277 grp->il_inp[i] = old_grp->il_inp[i];
278 grp->il_inpcnt = old_grp->il_inpcnt;
279 in_pcblbgroup_free(old_grp);
280 return (grp);
281 }
282
283 /*
284 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
285 * and shrink group if possible.
286 */
287 static void
in_pcblbgroup_reorder(struct inpcblbgrouphead * hdr,struct inpcblbgroup ** grpp,int i)288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
289 int i)
290 {
291 struct inpcblbgroup *grp, *new_grp;
292
293 grp = *grpp;
294 for (; i + 1 < grp->il_inpcnt; ++i)
295 grp->il_inp[i] = grp->il_inp[i + 1];
296 grp->il_inpcnt--;
297
298 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
299 grp->il_inpcnt <= grp->il_inpsiz / 4) {
300 /* Shrink this group. */
301 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
302 if (new_grp != NULL)
303 *grpp = new_grp;
304 }
305 }
306
307 /*
308 * Add PCB to load balance group for SO_REUSEPORT_LB option.
309 */
310 static int
in_pcbinslbgrouphash(struct inpcb * inp)311 in_pcbinslbgrouphash(struct inpcb *inp)
312 {
313 const static struct timeval interval = { 60, 0 };
314 static struct timeval lastprint;
315 struct inpcbinfo *pcbinfo;
316 struct inpcblbgrouphead *hdr;
317 struct inpcblbgroup *grp;
318 uint32_t idx;
319
320 pcbinfo = inp->inp_pcbinfo;
321
322 INP_WLOCK_ASSERT(inp);
323 INP_HASH_WLOCK_ASSERT(pcbinfo);
324
325 /*
326 * Don't allow jailed socket to join local group.
327 */
328 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
329 return (0);
330
331 #ifdef INET6
332 /*
333 * Don't allow IPv4 mapped INET6 wild socket.
334 */
335 if ((inp->inp_vflag & INP_IPV4) &&
336 inp->inp_laddr.s_addr == INADDR_ANY &&
337 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
338 return (0);
339 }
340 #endif
341
342 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
343 hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
344 CK_LIST_FOREACH(grp, hdr, il_list) {
345 if (grp->il_vflag == inp->inp_vflag &&
346 grp->il_lport == inp->inp_lport &&
347 memcmp(&grp->il_dependladdr,
348 &inp->inp_inc.inc_ie.ie_dependladdr,
349 sizeof(grp->il_dependladdr)) == 0)
350 break;
351 }
352 if (grp == NULL) {
353 /* Create new load balance group. */
354 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
355 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
356 INPCBLBGROUP_SIZMIN);
357 if (grp == NULL)
358 return (ENOBUFS);
359 } else if (grp->il_inpcnt == grp->il_inpsiz) {
360 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
361 if (ratecheck(&lastprint, &interval))
362 printf("lb group port %d, limit reached\n",
363 ntohs(grp->il_lport));
364 return (0);
365 }
366
367 /* Expand this local group. */
368 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
369 if (grp == NULL)
370 return (ENOBUFS);
371 }
372
373 KASSERT(grp->il_inpcnt < grp->il_inpsiz,
374 ("invalid local group size %d and count %d", grp->il_inpsiz,
375 grp->il_inpcnt));
376
377 grp->il_inp[grp->il_inpcnt] = inp;
378 grp->il_inpcnt++;
379 return (0);
380 }
381
382 /*
383 * Remove PCB from load balance group.
384 */
385 static void
in_pcbremlbgrouphash(struct inpcb * inp)386 in_pcbremlbgrouphash(struct inpcb *inp)
387 {
388 struct inpcbinfo *pcbinfo;
389 struct inpcblbgrouphead *hdr;
390 struct inpcblbgroup *grp;
391 int i;
392
393 pcbinfo = inp->inp_pcbinfo;
394
395 INP_WLOCK_ASSERT(inp);
396 INP_HASH_WLOCK_ASSERT(pcbinfo);
397
398 hdr = &pcbinfo->ipi_lbgrouphashbase[
399 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
400 CK_LIST_FOREACH(grp, hdr, il_list) {
401 for (i = 0; i < grp->il_inpcnt; ++i) {
402 if (grp->il_inp[i] != inp)
403 continue;
404
405 if (grp->il_inpcnt == 1) {
406 /* We are the last, free this local group. */
407 in_pcblbgroup_free(grp);
408 } else {
409 /* Pull up inpcbs, shrink group if possible. */
410 in_pcblbgroup_reorder(hdr, &grp, i);
411 }
412 return;
413 }
414 }
415 }
416
417 /*
418 * Different protocols initialize their inpcbs differently - giving
419 * different name to the lock. But they all are disposed the same.
420 */
421 static void
inpcb_fini(void * mem,int size)422 inpcb_fini(void *mem, int size)
423 {
424 struct inpcb *inp = mem;
425
426 INP_LOCK_DESTROY(inp);
427 }
428
429 /*
430 * Initialize an inpcbinfo -- we should be able to reduce the number of
431 * arguments in time.
432 */
433 void
in_pcbinfo_init(struct inpcbinfo * pcbinfo,const char * name,struct inpcbhead * listhead,int hash_nelements,int porthash_nelements,char * inpcbzone_name,uma_init inpcbzone_init,u_int hashfields)434 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
435 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
436 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
437 {
438
439 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
440
441 INP_INFO_LOCK_INIT(pcbinfo, name);
442 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */
443 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
444 #ifdef VIMAGE
445 pcbinfo->ipi_vnet = curvnet;
446 #endif
447 pcbinfo->ipi_listhead = listhead;
448 CK_LIST_INIT(pcbinfo->ipi_listhead);
449 pcbinfo->ipi_count = 0;
450 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
451 &pcbinfo->ipi_hashmask);
452 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
453 &pcbinfo->ipi_porthashmask);
454 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
455 &pcbinfo->ipi_lbgrouphashmask);
456 #ifdef PCBGROUP
457 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
458 #endif
459 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
460 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
461 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
462 uma_zone_set_warning(pcbinfo->ipi_zone,
463 "kern.ipc.maxsockets limit reached");
464 }
465
466 /*
467 * Destroy an inpcbinfo.
468 */
469 void
in_pcbinfo_destroy(struct inpcbinfo * pcbinfo)470 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
471 {
472
473 KASSERT(pcbinfo->ipi_count == 0,
474 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
475
476 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
477 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
478 pcbinfo->ipi_porthashmask);
479 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
480 pcbinfo->ipi_lbgrouphashmask);
481 #ifdef PCBGROUP
482 in_pcbgroup_destroy(pcbinfo);
483 #endif
484 uma_zdestroy(pcbinfo->ipi_zone);
485 INP_LIST_LOCK_DESTROY(pcbinfo);
486 INP_HASH_LOCK_DESTROY(pcbinfo);
487 INP_INFO_LOCK_DESTROY(pcbinfo);
488 }
489
490 /*
491 * Allocate a PCB and associate it with the socket.
492 * On success return with the PCB locked.
493 */
494 int
in_pcballoc(struct socket * so,struct inpcbinfo * pcbinfo)495 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
496 {
497 struct inpcb *inp;
498 int error;
499
500 #ifdef INVARIANTS
501 if (pcbinfo == &V_tcbinfo) {
502 INP_INFO_RLOCK_ASSERT(pcbinfo);
503 } else {
504 INP_INFO_WLOCK_ASSERT(pcbinfo);
505 }
506 #endif
507
508 error = 0;
509 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
510 if (inp == NULL)
511 return (ENOBUFS);
512 bzero(&inp->inp_start_zero, inp_zero_size);
513 inp->inp_pcbinfo = pcbinfo;
514 inp->inp_socket = so;
515 inp->inp_cred = crhold(so->so_cred);
516 inp->inp_inc.inc_fibnum = so->so_fibnum;
517 #ifdef MAC
518 error = mac_inpcb_init(inp, M_NOWAIT);
519 if (error != 0)
520 goto out;
521 mac_inpcb_create(so, inp);
522 #endif
523 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
524 error = ipsec_init_pcbpolicy(inp);
525 if (error != 0) {
526 #ifdef MAC
527 mac_inpcb_destroy(inp);
528 #endif
529 goto out;
530 }
531 #endif /*IPSEC*/
532 #ifdef INET6
533 if (INP_SOCKAF(so) == AF_INET6) {
534 inp->inp_vflag |= INP_IPV6PROTO;
535 if (V_ip6_v6only)
536 inp->inp_flags |= IN6P_IPV6_V6ONLY;
537 }
538 #endif
539 INP_WLOCK(inp);
540 INP_LIST_WLOCK(pcbinfo);
541 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
542 pcbinfo->ipi_count++;
543 so->so_pcb = (caddr_t)inp;
544 #ifdef INET6
545 if (V_ip6_auto_flowlabel)
546 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
547 #endif
548 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
549 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */
550
551 /*
552 * Routes in inpcb's can cache L2 as well; they are guaranteed
553 * to be cleaned up.
554 */
555 inp->inp_route.ro_flags = RT_LLE_CACHE;
556 INP_LIST_WUNLOCK(pcbinfo);
557 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
558 out:
559 if (error != 0) {
560 crfree(inp->inp_cred);
561 uma_zfree(pcbinfo->ipi_zone, inp);
562 }
563 #endif
564 return (error);
565 }
566
567 #ifdef INET
568 int
in_pcbbind(struct inpcb * inp,struct sockaddr * nam,struct ucred * cred)569 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
570 {
571 int anonport, error;
572
573 INP_WLOCK_ASSERT(inp);
574 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
575
576 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
577 return (EINVAL);
578 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
579 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
580 &inp->inp_lport, cred);
581 if (error)
582 return (error);
583 if (in_pcbinshash(inp) != 0) {
584 inp->inp_laddr.s_addr = INADDR_ANY;
585 inp->inp_lport = 0;
586 return (EAGAIN);
587 }
588 if (anonport)
589 inp->inp_flags |= INP_ANONPORT;
590 return (0);
591 }
592 #endif
593
594 /*
595 * Select a local port (number) to use.
596 */
597 #if defined(INET) || defined(INET6)
598 int
in_pcb_lport(struct inpcb * inp,struct in_addr * laddrp,u_short * lportp,struct ucred * cred,int lookupflags)599 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
600 struct ucred *cred, int lookupflags)
601 {
602 struct inpcbinfo *pcbinfo;
603 struct inpcb *tmpinp;
604 unsigned short *lastport;
605 int count, dorandom, error;
606 u_short aux, first, last, lport;
607 #ifdef INET
608 struct in_addr laddr;
609 #endif
610
611 pcbinfo = inp->inp_pcbinfo;
612
613 /*
614 * Because no actual state changes occur here, a global write lock on
615 * the pcbinfo isn't required.
616 */
617 INP_LOCK_ASSERT(inp);
618 INP_HASH_LOCK_ASSERT(pcbinfo);
619
620 if (inp->inp_flags & INP_HIGHPORT) {
621 first = V_ipport_hifirstauto; /* sysctl */
622 last = V_ipport_hilastauto;
623 lastport = &pcbinfo->ipi_lasthi;
624 } else if (inp->inp_flags & INP_LOWPORT) {
625 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
626 if (error)
627 return (error);
628 first = V_ipport_lowfirstauto; /* 1023 */
629 last = V_ipport_lowlastauto; /* 600 */
630 lastport = &pcbinfo->ipi_lastlow;
631 } else {
632 first = V_ipport_firstauto; /* sysctl */
633 last = V_ipport_lastauto;
634 lastport = &pcbinfo->ipi_lastport;
635 }
636 /*
637 * For UDP(-Lite), use random port allocation as long as the user
638 * allows it. For TCP (and as of yet unknown) connections,
639 * use random port allocation only if the user allows it AND
640 * ipport_tick() allows it.
641 */
642 if (V_ipport_randomized &&
643 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
644 pcbinfo == &V_ulitecbinfo))
645 dorandom = 1;
646 else
647 dorandom = 0;
648 /*
649 * It makes no sense to do random port allocation if
650 * we have the only port available.
651 */
652 if (first == last)
653 dorandom = 0;
654 /* Make sure to not include UDP(-Lite) packets in the count. */
655 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
656 V_ipport_tcpallocs++;
657 /*
658 * Instead of having two loops further down counting up or down
659 * make sure that first is always <= last and go with only one
660 * code path implementing all logic.
661 */
662 if (first > last) {
663 aux = first;
664 first = last;
665 last = aux;
666 }
667
668 #ifdef INET
669 /* Make the compiler happy. */
670 laddr.s_addr = 0;
671 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
672 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
673 __func__, inp));
674 laddr = *laddrp;
675 }
676 #endif
677 tmpinp = NULL; /* Make compiler happy. */
678 lport = *lportp;
679
680 if (dorandom)
681 *lastport = first + (arc4random() % (last - first));
682
683 count = last - first;
684
685 do {
686 if (count-- < 0) /* completely used? */
687 return (EADDRNOTAVAIL);
688 ++*lastport;
689 if (*lastport < first || *lastport > last)
690 *lastport = first;
691 lport = htons(*lastport);
692
693 #ifdef INET6
694 if ((inp->inp_vflag & INP_IPV6) != 0)
695 tmpinp = in6_pcblookup_local(pcbinfo,
696 &inp->in6p_laddr, lport, lookupflags, cred);
697 #endif
698 #if defined(INET) && defined(INET6)
699 else
700 #endif
701 #ifdef INET
702 tmpinp = in_pcblookup_local(pcbinfo, laddr,
703 lport, lookupflags, cred);
704 #endif
705 } while (tmpinp != NULL);
706
707 #ifdef INET
708 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
709 laddrp->s_addr = laddr.s_addr;
710 #endif
711 *lportp = lport;
712
713 return (0);
714 }
715
716 /*
717 * Return cached socket options.
718 */
719 int
inp_so_options(const struct inpcb * inp)720 inp_so_options(const struct inpcb *inp)
721 {
722 int so_options;
723
724 so_options = 0;
725
726 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
727 so_options |= SO_REUSEPORT_LB;
728 if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
729 so_options |= SO_REUSEPORT;
730 if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
731 so_options |= SO_REUSEADDR;
732 return (so_options);
733 }
734 #endif /* INET || INET6 */
735
736 /*
737 * Check if a new BINDMULTI socket is allowed to be created.
738 *
739 * ni points to the new inp.
740 * oi points to the exisitng inp.
741 *
742 * This checks whether the existing inp also has BINDMULTI and
743 * whether the credentials match.
744 */
745 int
in_pcbbind_check_bindmulti(const struct inpcb * ni,const struct inpcb * oi)746 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
747 {
748 /* Check permissions match */
749 if ((ni->inp_flags2 & INP_BINDMULTI) &&
750 (ni->inp_cred->cr_uid !=
751 oi->inp_cred->cr_uid))
752 return (0);
753
754 /* Check the existing inp has BINDMULTI set */
755 if ((ni->inp_flags2 & INP_BINDMULTI) &&
756 ((oi->inp_flags2 & INP_BINDMULTI) == 0))
757 return (0);
758
759 /*
760 * We're okay - either INP_BINDMULTI isn't set on ni, or
761 * it is and it matches the checks.
762 */
763 return (1);
764 }
765
766 #ifdef INET
767 /*
768 * Set up a bind operation on a PCB, performing port allocation
769 * as required, but do not actually modify the PCB. Callers can
770 * either complete the bind by setting inp_laddr/inp_lport and
771 * calling in_pcbinshash(), or they can just use the resulting
772 * port and address to authorise the sending of a once-off packet.
773 *
774 * On error, the values of *laddrp and *lportp are not changed.
775 */
776 int
in_pcbbind_setup(struct inpcb * inp,struct sockaddr * nam,in_addr_t * laddrp,u_short * lportp,struct ucred * cred)777 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
778 u_short *lportp, struct ucred *cred)
779 {
780 struct socket *so = inp->inp_socket;
781 struct sockaddr_in *sin;
782 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
783 struct in_addr laddr;
784 u_short lport = 0;
785 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
786 int error;
787
788 /*
789 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
790 * so that we don't have to add to the (already messy) code below.
791 */
792 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
793
794 /*
795 * No state changes, so read locks are sufficient here.
796 */
797 INP_LOCK_ASSERT(inp);
798 INP_HASH_LOCK_ASSERT(pcbinfo);
799
800 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
801 return (EADDRNOTAVAIL);
802 laddr.s_addr = *laddrp;
803 if (nam != NULL && laddr.s_addr != INADDR_ANY)
804 return (EINVAL);
805 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
806 lookupflags = INPLOOKUP_WILDCARD;
807 if (nam == NULL) {
808 if ((error = prison_local_ip4(cred, &laddr)) != 0)
809 return (error);
810 } else {
811 sin = (struct sockaddr_in *)nam;
812 if (nam->sa_len != sizeof (*sin))
813 return (EINVAL);
814 #ifdef notdef
815 /*
816 * We should check the family, but old programs
817 * incorrectly fail to initialize it.
818 */
819 if (sin->sin_family != AF_INET)
820 return (EAFNOSUPPORT);
821 #endif
822 error = prison_local_ip4(cred, &sin->sin_addr);
823 if (error)
824 return (error);
825 if (sin->sin_port != *lportp) {
826 /* Don't allow the port to change. */
827 if (*lportp != 0)
828 return (EINVAL);
829 lport = sin->sin_port;
830 }
831 /* NB: lport is left as 0 if the port isn't being changed. */
832 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
833 /*
834 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
835 * allow complete duplication of binding if
836 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
837 * and a multicast address is bound on both
838 * new and duplicated sockets.
839 */
840 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
841 reuseport = SO_REUSEADDR|SO_REUSEPORT;
842 /*
843 * XXX: How to deal with SO_REUSEPORT_LB here?
844 * Treat same as SO_REUSEPORT for now.
845 */
846 if ((so->so_options &
847 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
848 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
849 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
850 sin->sin_port = 0; /* yech... */
851 bzero(&sin->sin_zero, sizeof(sin->sin_zero));
852 /*
853 * Is the address a local IP address?
854 * If INP_BINDANY is set, then the socket may be bound
855 * to any endpoint address, local or not.
856 */
857 if ((inp->inp_flags & INP_BINDANY) == 0 &&
858 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
859 return (EADDRNOTAVAIL);
860 }
861 laddr = sin->sin_addr;
862 if (lport) {
863 struct inpcb *t;
864 struct tcptw *tw;
865
866 /* GROSS */
867 if (ntohs(lport) <= V_ipport_reservedhigh &&
868 ntohs(lport) >= V_ipport_reservedlow &&
869 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
870 0))
871 return (EACCES);
872 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
873 priv_check_cred(inp->inp_cred,
874 PRIV_NETINET_REUSEPORT, 0) != 0) {
875 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
876 lport, INPLOOKUP_WILDCARD, cred);
877 /*
878 * XXX
879 * This entire block sorely needs a rewrite.
880 */
881 if (t &&
882 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
883 ((t->inp_flags & INP_TIMEWAIT) == 0) &&
884 (so->so_type != SOCK_STREAM ||
885 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
886 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
887 ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
888 (t->inp_flags2 & INP_REUSEPORT) ||
889 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
890 (inp->inp_cred->cr_uid !=
891 t->inp_cred->cr_uid))
892 return (EADDRINUSE);
893
894 /*
895 * If the socket is a BINDMULTI socket, then
896 * the credentials need to match and the
897 * original socket also has to have been bound
898 * with BINDMULTI.
899 */
900 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
901 return (EADDRINUSE);
902 }
903 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
904 lport, lookupflags, cred);
905 if (t && (t->inp_flags & INP_TIMEWAIT)) {
906 /*
907 * XXXRW: If an incpb has had its timewait
908 * state recycled, we treat the address as
909 * being in use (for now). This is better
910 * than a panic, but not desirable.
911 */
912 tw = intotw(t);
913 if (tw == NULL ||
914 ((reuseport & tw->tw_so_options) == 0 &&
915 (reuseport_lb &
916 tw->tw_so_options) == 0)) {
917 return (EADDRINUSE);
918 }
919 } else if (t &&
920 ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
921 (reuseport & inp_so_options(t)) == 0 &&
922 (reuseport_lb & inp_so_options(t)) == 0) {
923 #ifdef INET6
924 if (ntohl(sin->sin_addr.s_addr) !=
925 INADDR_ANY ||
926 ntohl(t->inp_laddr.s_addr) !=
927 INADDR_ANY ||
928 (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
929 (t->inp_vflag & INP_IPV6PROTO) == 0)
930 #endif
931 return (EADDRINUSE);
932 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
933 return (EADDRINUSE);
934 }
935 }
936 }
937 if (*lportp != 0)
938 lport = *lportp;
939 if (lport == 0) {
940 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
941 if (error != 0)
942 return (error);
943
944 }
945 *laddrp = laddr.s_addr;
946 *lportp = lport;
947 return (0);
948 }
949
950 /*
951 * Connect from a socket to a specified address.
952 * Both address and port must be specified in argument sin.
953 * If don't have a local address for this socket yet,
954 * then pick one.
955 */
956 int
in_pcbconnect_mbuf(struct inpcb * inp,struct sockaddr * nam,struct ucred * cred,struct mbuf * m)957 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
958 struct ucred *cred, struct mbuf *m)
959 {
960 u_short lport, fport;
961 in_addr_t laddr, faddr;
962 int anonport, error;
963
964 INP_WLOCK_ASSERT(inp);
965 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
966
967 lport = inp->inp_lport;
968 laddr = inp->inp_laddr.s_addr;
969 anonport = (lport == 0);
970 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
971 NULL, cred);
972 if (error)
973 return (error);
974
975 /* Do the initial binding of the local address if required. */
976 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
977 inp->inp_lport = lport;
978 inp->inp_laddr.s_addr = laddr;
979 if (in_pcbinshash(inp) != 0) {
980 inp->inp_laddr.s_addr = INADDR_ANY;
981 inp->inp_lport = 0;
982 return (EAGAIN);
983 }
984 }
985
986 /* Commit the remaining changes. */
987 inp->inp_lport = lport;
988 inp->inp_laddr.s_addr = laddr;
989 inp->inp_faddr.s_addr = faddr;
990 inp->inp_fport = fport;
991 in_pcbrehash_mbuf(inp, m);
992
993 if (anonport)
994 inp->inp_flags |= INP_ANONPORT;
995 return (0);
996 }
997
998 int
in_pcbconnect(struct inpcb * inp,struct sockaddr * nam,struct ucred * cred)999 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1000 {
1001
1002 return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1003 }
1004
1005 /*
1006 * Do proper source address selection on an unbound socket in case
1007 * of connect. Take jails into account as well.
1008 */
1009 int
in_pcbladdr(struct inpcb * inp,struct in_addr * faddr,struct in_addr * laddr,struct ucred * cred)1010 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1011 struct ucred *cred)
1012 {
1013 struct ifaddr *ifa;
1014 struct sockaddr *sa;
1015 struct sockaddr_in *sin;
1016 struct route sro;
1017 int error;
1018
1019 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1020 /*
1021 * Bypass source address selection and use the primary jail IP
1022 * if requested.
1023 */
1024 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1025 return (0);
1026
1027 error = 0;
1028 bzero(&sro, sizeof(sro));
1029
1030 sin = (struct sockaddr_in *)&sro.ro_dst;
1031 sin->sin_family = AF_INET;
1032 sin->sin_len = sizeof(struct sockaddr_in);
1033 sin->sin_addr.s_addr = faddr->s_addr;
1034
1035 /*
1036 * If route is known our src addr is taken from the i/f,
1037 * else punt.
1038 *
1039 * Find out route to destination.
1040 */
1041 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1042 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1043
1044 /*
1045 * If we found a route, use the address corresponding to
1046 * the outgoing interface.
1047 *
1048 * Otherwise assume faddr is reachable on a directly connected
1049 * network and try to find a corresponding interface to take
1050 * the source address from.
1051 */
1052 NET_EPOCH_ENTER();
1053 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1054 struct in_ifaddr *ia;
1055 struct ifnet *ifp;
1056
1057 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1058 inp->inp_socket->so_fibnum));
1059 if (ia == NULL) {
1060 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1061 inp->inp_socket->so_fibnum));
1062
1063 }
1064 if (ia == NULL) {
1065 error = ENETUNREACH;
1066 goto done;
1067 }
1068
1069 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1070 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1071 goto done;
1072 }
1073
1074 ifp = ia->ia_ifp;
1075 ia = NULL;
1076 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1077
1078 sa = ifa->ifa_addr;
1079 if (sa->sa_family != AF_INET)
1080 continue;
1081 sin = (struct sockaddr_in *)sa;
1082 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1083 ia = (struct in_ifaddr *)ifa;
1084 break;
1085 }
1086 }
1087 if (ia != NULL) {
1088 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1089 goto done;
1090 }
1091
1092 /* 3. As a last resort return the 'default' jail address. */
1093 error = prison_get_ip4(cred, laddr);
1094 goto done;
1095 }
1096
1097 /*
1098 * If the outgoing interface on the route found is not
1099 * a loopback interface, use the address from that interface.
1100 * In case of jails do those three steps:
1101 * 1. check if the interface address belongs to the jail. If so use it.
1102 * 2. check if we have any address on the outgoing interface
1103 * belonging to this jail. If so use it.
1104 * 3. as a last resort return the 'default' jail address.
1105 */
1106 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1107 struct in_ifaddr *ia;
1108 struct ifnet *ifp;
1109
1110 /* If not jailed, use the default returned. */
1111 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1112 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1113 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1114 goto done;
1115 }
1116
1117 /* Jailed. */
1118 /* 1. Check if the iface address belongs to the jail. */
1119 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1120 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1121 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1122 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1123 goto done;
1124 }
1125
1126 /*
1127 * 2. Check if we have any address on the outgoing interface
1128 * belonging to this jail.
1129 */
1130 ia = NULL;
1131 ifp = sro.ro_rt->rt_ifp;
1132 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1133 sa = ifa->ifa_addr;
1134 if (sa->sa_family != AF_INET)
1135 continue;
1136 sin = (struct sockaddr_in *)sa;
1137 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1138 ia = (struct in_ifaddr *)ifa;
1139 break;
1140 }
1141 }
1142 if (ia != NULL) {
1143 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1144 goto done;
1145 }
1146
1147 /* 3. As a last resort return the 'default' jail address. */
1148 error = prison_get_ip4(cred, laddr);
1149 goto done;
1150 }
1151
1152 /*
1153 * The outgoing interface is marked with 'loopback net', so a route
1154 * to ourselves is here.
1155 * Try to find the interface of the destination address and then
1156 * take the address from there. That interface is not necessarily
1157 * a loopback interface.
1158 * In case of jails, check that it is an address of the jail
1159 * and if we cannot find, fall back to the 'default' jail address.
1160 */
1161 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1162 struct sockaddr_in sain;
1163 struct in_ifaddr *ia;
1164
1165 bzero(&sain, sizeof(struct sockaddr_in));
1166 sain.sin_family = AF_INET;
1167 sain.sin_len = sizeof(struct sockaddr_in);
1168 sain.sin_addr.s_addr = faddr->s_addr;
1169
1170 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1171 inp->inp_socket->so_fibnum));
1172 if (ia == NULL)
1173 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1174 inp->inp_socket->so_fibnum));
1175 if (ia == NULL)
1176 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1177
1178 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1179 if (ia == NULL) {
1180 error = ENETUNREACH;
1181 goto done;
1182 }
1183 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1184 goto done;
1185 }
1186
1187 /* Jailed. */
1188 if (ia != NULL) {
1189 struct ifnet *ifp;
1190
1191 ifp = ia->ia_ifp;
1192 ia = NULL;
1193 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1194 sa = ifa->ifa_addr;
1195 if (sa->sa_family != AF_INET)
1196 continue;
1197 sin = (struct sockaddr_in *)sa;
1198 if (prison_check_ip4(cred,
1199 &sin->sin_addr) == 0) {
1200 ia = (struct in_ifaddr *)ifa;
1201 break;
1202 }
1203 }
1204 if (ia != NULL) {
1205 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1206 goto done;
1207 }
1208 }
1209
1210 /* 3. As a last resort return the 'default' jail address. */
1211 error = prison_get_ip4(cred, laddr);
1212 goto done;
1213 }
1214
1215 done:
1216 NET_EPOCH_EXIT();
1217 if (sro.ro_rt != NULL)
1218 RTFREE(sro.ro_rt);
1219 return (error);
1220 }
1221
1222 /*
1223 * Set up for a connect from a socket to the specified address.
1224 * On entry, *laddrp and *lportp should contain the current local
1225 * address and port for the PCB; these are updated to the values
1226 * that should be placed in inp_laddr and inp_lport to complete
1227 * the connect.
1228 *
1229 * On success, *faddrp and *fportp will be set to the remote address
1230 * and port. These are not updated in the error case.
1231 *
1232 * If the operation fails because the connection already exists,
1233 * *oinpp will be set to the PCB of that connection so that the
1234 * caller can decide to override it. In all other cases, *oinpp
1235 * is set to NULL.
1236 */
1237 int
in_pcbconnect_setup(struct inpcb * inp,struct sockaddr * nam,in_addr_t * laddrp,u_short * lportp,in_addr_t * faddrp,u_short * fportp,struct inpcb ** oinpp,struct ucred * cred)1238 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1239 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1240 struct inpcb **oinpp, struct ucred *cred)
1241 {
1242 struct rm_priotracker in_ifa_tracker;
1243 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1244 struct in_ifaddr *ia;
1245 struct inpcb *oinp;
1246 struct in_addr laddr, faddr;
1247 u_short lport, fport;
1248 int error;
1249
1250 /*
1251 * Because a global state change doesn't actually occur here, a read
1252 * lock is sufficient.
1253 */
1254 INP_LOCK_ASSERT(inp);
1255 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1256
1257 if (oinpp != NULL)
1258 *oinpp = NULL;
1259 if (nam->sa_len != sizeof (*sin))
1260 return (EINVAL);
1261 if (sin->sin_family != AF_INET)
1262 return (EAFNOSUPPORT);
1263 if (sin->sin_port == 0)
1264 return (EADDRNOTAVAIL);
1265 laddr.s_addr = *laddrp;
1266 lport = *lportp;
1267 faddr = sin->sin_addr;
1268 fport = sin->sin_port;
1269
1270 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1271 /*
1272 * If the destination address is INADDR_ANY,
1273 * use the primary local address.
1274 * If the supplied address is INADDR_BROADCAST,
1275 * and the primary interface supports broadcast,
1276 * choose the broadcast address for that interface.
1277 */
1278 if (faddr.s_addr == INADDR_ANY) {
1279 IN_IFADDR_RLOCK(&in_ifa_tracker);
1280 faddr =
1281 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1282 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1283 if (cred != NULL &&
1284 (error = prison_get_ip4(cred, &faddr)) != 0)
1285 return (error);
1286 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1287 IN_IFADDR_RLOCK(&in_ifa_tracker);
1288 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1289 IFF_BROADCAST)
1290 faddr = satosin(&CK_STAILQ_FIRST(
1291 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1292 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1293 }
1294 }
1295 if (laddr.s_addr == INADDR_ANY) {
1296 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1297 /*
1298 * If the destination address is multicast and an outgoing
1299 * interface has been set as a multicast option, prefer the
1300 * address of that interface as our source address.
1301 */
1302 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1303 inp->inp_moptions != NULL) {
1304 struct ip_moptions *imo;
1305 struct ifnet *ifp;
1306
1307 imo = inp->inp_moptions;
1308 if (imo->imo_multicast_ifp != NULL) {
1309 ifp = imo->imo_multicast_ifp;
1310 IN_IFADDR_RLOCK(&in_ifa_tracker);
1311 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1312 if ((ia->ia_ifp == ifp) &&
1313 (cred == NULL ||
1314 prison_check_ip4(cred,
1315 &ia->ia_addr.sin_addr) == 0))
1316 break;
1317 }
1318 if (ia == NULL)
1319 error = EADDRNOTAVAIL;
1320 else {
1321 laddr = ia->ia_addr.sin_addr;
1322 error = 0;
1323 }
1324 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1325 }
1326 }
1327 if (error)
1328 return (error);
1329 }
1330 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1331 laddr, lport, 0, NULL);
1332 if (oinp != NULL) {
1333 if (oinpp != NULL)
1334 *oinpp = oinp;
1335 return (EADDRINUSE);
1336 }
1337 if (lport == 0) {
1338 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1339 cred);
1340 if (error)
1341 return (error);
1342 }
1343 *laddrp = laddr.s_addr;
1344 *lportp = lport;
1345 *faddrp = faddr.s_addr;
1346 *fportp = fport;
1347 return (0);
1348 }
1349
1350 void
in_pcbdisconnect(struct inpcb * inp)1351 in_pcbdisconnect(struct inpcb *inp)
1352 {
1353
1354 INP_WLOCK_ASSERT(inp);
1355 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1356
1357 inp->inp_faddr.s_addr = INADDR_ANY;
1358 inp->inp_fport = 0;
1359 in_pcbrehash(inp);
1360 }
1361 #endif /* INET */
1362
1363 /*
1364 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1365 * For most protocols, this will be invoked immediately prior to calling
1366 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the
1367 * socket, in which case in_pcbfree() is deferred.
1368 */
1369 void
in_pcbdetach(struct inpcb * inp)1370 in_pcbdetach(struct inpcb *inp)
1371 {
1372
1373 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1374
1375 #ifdef RATELIMIT
1376 if (inp->inp_snd_tag != NULL)
1377 in_pcbdetach_txrtlmt(inp);
1378 #endif
1379 inp->inp_socket->so_pcb = NULL;
1380 inp->inp_socket = NULL;
1381 }
1382
1383 /*
1384 * in_pcbref() bumps the reference count on an inpcb in order to maintain
1385 * stability of an inpcb pointer despite the inpcb lock being released. This
1386 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1387 * but where the inpcb lock may already held, or when acquiring a reference
1388 * via a pcbgroup.
1389 *
1390 * in_pcbref() should be used only to provide brief memory stability, and
1391 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1392 * garbage collect the inpcb if it has been in_pcbfree()'d from another
1393 * context. Until in_pcbrele() has returned that the inpcb is still valid,
1394 * lock and rele are the *only* safe operations that may be performed on the
1395 * inpcb.
1396 *
1397 * While the inpcb will not be freed, releasing the inpcb lock means that the
1398 * connection's state may change, so the caller should be careful to
1399 * revalidate any cached state on reacquiring the lock. Drop the reference
1400 * using in_pcbrele().
1401 */
1402 void
in_pcbref(struct inpcb * inp)1403 in_pcbref(struct inpcb *inp)
1404 {
1405
1406 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1407
1408 refcount_acquire(&inp->inp_refcount);
1409 }
1410
1411 /*
1412 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1413 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1414 * return a flag indicating whether or not the inpcb remains valid. If it is
1415 * valid, we return with the inpcb lock held.
1416 *
1417 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1418 * reference on an inpcb. Historically more work was done here (actually, in
1419 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1420 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely
1421 * about memory stability (and continued use of the write lock).
1422 */
1423 int
in_pcbrele_rlocked(struct inpcb * inp)1424 in_pcbrele_rlocked(struct inpcb *inp)
1425 {
1426 struct inpcbinfo *pcbinfo;
1427
1428 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1429
1430 INP_RLOCK_ASSERT(inp);
1431
1432 if (refcount_release(&inp->inp_refcount) == 0) {
1433 /*
1434 * If the inpcb has been freed, let the caller know, even if
1435 * this isn't the last reference.
1436 */
1437 if (inp->inp_flags2 & INP_FREED) {
1438 INP_RUNLOCK(inp);
1439 return (1);
1440 }
1441 return (0);
1442 }
1443
1444 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1445 #ifdef TCPHPTS
1446 if (inp->inp_in_hpts || inp->inp_in_input) {
1447 struct tcp_hpts_entry *hpts;
1448 /*
1449 * We should not be on the hpts at
1450 * this point in any form. we must
1451 * get the lock to be sure.
1452 */
1453 hpts = tcp_hpts_lock(inp);
1454 if (inp->inp_in_hpts)
1455 panic("Hpts:%p inp:%p at free still on hpts",
1456 hpts, inp);
1457 mtx_unlock(&hpts->p_mtx);
1458 hpts = tcp_input_lock(inp);
1459 if (inp->inp_in_input)
1460 panic("Hpts:%p inp:%p at free still on input hpts",
1461 hpts, inp);
1462 mtx_unlock(&hpts->p_mtx);
1463 }
1464 #endif
1465 INP_RUNLOCK(inp);
1466 pcbinfo = inp->inp_pcbinfo;
1467 uma_zfree(pcbinfo->ipi_zone, inp);
1468 return (1);
1469 }
1470
1471 int
in_pcbrele_wlocked(struct inpcb * inp)1472 in_pcbrele_wlocked(struct inpcb *inp)
1473 {
1474 struct inpcbinfo *pcbinfo;
1475
1476 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1477
1478 INP_WLOCK_ASSERT(inp);
1479
1480 if (refcount_release(&inp->inp_refcount) == 0) {
1481 /*
1482 * If the inpcb has been freed, let the caller know, even if
1483 * this isn't the last reference.
1484 */
1485 if (inp->inp_flags2 & INP_FREED) {
1486 INP_WUNLOCK(inp);
1487 return (1);
1488 }
1489 return (0);
1490 }
1491
1492 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1493 #ifdef TCPHPTS
1494 if (inp->inp_in_hpts || inp->inp_in_input) {
1495 struct tcp_hpts_entry *hpts;
1496 /*
1497 * We should not be on the hpts at
1498 * this point in any form. we must
1499 * get the lock to be sure.
1500 */
1501 hpts = tcp_hpts_lock(inp);
1502 if (inp->inp_in_hpts)
1503 panic("Hpts:%p inp:%p at free still on hpts",
1504 hpts, inp);
1505 mtx_unlock(&hpts->p_mtx);
1506 hpts = tcp_input_lock(inp);
1507 if (inp->inp_in_input)
1508 panic("Hpts:%p inp:%p at free still on input hpts",
1509 hpts, inp);
1510 mtx_unlock(&hpts->p_mtx);
1511 }
1512 #endif
1513 INP_WUNLOCK(inp);
1514 pcbinfo = inp->inp_pcbinfo;
1515 uma_zfree(pcbinfo->ipi_zone, inp);
1516 return (1);
1517 }
1518
1519 /*
1520 * Temporary wrapper.
1521 */
1522 int
in_pcbrele(struct inpcb * inp)1523 in_pcbrele(struct inpcb *inp)
1524 {
1525
1526 return (in_pcbrele_wlocked(inp));
1527 }
1528
1529 void
in_pcblist_rele_rlocked(epoch_context_t ctx)1530 in_pcblist_rele_rlocked(epoch_context_t ctx)
1531 {
1532 struct in_pcblist *il;
1533 struct inpcb *inp;
1534 struct inpcbinfo *pcbinfo;
1535 int i, n;
1536
1537 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1538 pcbinfo = il->il_pcbinfo;
1539 n = il->il_count;
1540 INP_INFO_WLOCK(pcbinfo);
1541 for (i = 0; i < n; i++) {
1542 inp = il->il_inp_list[i];
1543 INP_RLOCK(inp);
1544 if (!in_pcbrele_rlocked(inp))
1545 INP_RUNLOCK(inp);
1546 }
1547 INP_INFO_WUNLOCK(pcbinfo);
1548 free(il, M_TEMP);
1549 }
1550
1551 static void
inpcbport_free(epoch_context_t ctx)1552 inpcbport_free(epoch_context_t ctx)
1553 {
1554 struct inpcbport *phd;
1555
1556 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1557 free(phd, M_PCB);
1558 }
1559
1560 static void
in_pcbfree_deferred(epoch_context_t ctx)1561 in_pcbfree_deferred(epoch_context_t ctx)
1562 {
1563 struct inpcb *inp;
1564 int released __unused;
1565
1566 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1567
1568 INP_WLOCK(inp);
1569 CURVNET_SET(inp->inp_vnet);
1570 #ifdef INET
1571 struct ip_moptions *imo = inp->inp_moptions;
1572 inp->inp_moptions = NULL;
1573 #endif
1574 /* XXXRW: Do as much as possible here. */
1575 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1576 if (inp->inp_sp != NULL)
1577 ipsec_delete_pcbpolicy(inp);
1578 #endif
1579 #ifdef INET6
1580 struct ip6_moptions *im6o = NULL;
1581 if (inp->inp_vflag & INP_IPV6PROTO) {
1582 ip6_freepcbopts(inp->in6p_outputopts);
1583 im6o = inp->in6p_moptions;
1584 inp->in6p_moptions = NULL;
1585 }
1586 #endif
1587 if (inp->inp_options)
1588 (void)m_free(inp->inp_options);
1589 inp->inp_vflag = 0;
1590 crfree(inp->inp_cred);
1591 #ifdef MAC
1592 mac_inpcb_destroy(inp);
1593 #endif
1594 released = in_pcbrele_wlocked(inp);
1595 MPASS(released);
1596 #ifdef INET6
1597 ip6_freemoptions(im6o);
1598 #endif
1599 #ifdef INET
1600 inp_freemoptions(imo);
1601 #endif
1602 CURVNET_RESTORE();
1603 }
1604
1605 /*
1606 * Unconditionally schedule an inpcb to be freed by decrementing its
1607 * reference count, which should occur only after the inpcb has been detached
1608 * from its socket. If another thread holds a temporary reference (acquired
1609 * using in_pcbref()) then the free is deferred until that reference is
1610 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all
1611 * work, including removal from global lists, is done in this context, where
1612 * the pcbinfo lock is held.
1613 */
1614 void
in_pcbfree(struct inpcb * inp)1615 in_pcbfree(struct inpcb *inp)
1616 {
1617 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1618
1619 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1620 KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1621 ("%s: called twice for pcb %p", __func__, inp));
1622 if (inp->inp_flags2 & INP_FREED) {
1623 INP_WUNLOCK(inp);
1624 return;
1625 }
1626
1627 #ifdef INVARIANTS
1628 if (pcbinfo == &V_tcbinfo) {
1629 INP_INFO_LOCK_ASSERT(pcbinfo);
1630 } else {
1631 INP_INFO_WLOCK_ASSERT(pcbinfo);
1632 }
1633 #endif
1634 INP_WLOCK_ASSERT(inp);
1635 INP_LIST_WLOCK(pcbinfo);
1636 in_pcbremlists(inp);
1637 INP_LIST_WUNLOCK(pcbinfo);
1638 RO_INVALIDATE_CACHE(&inp->inp_route);
1639 /* mark as destruction in progress */
1640 inp->inp_flags2 |= INP_FREED;
1641 INP_WUNLOCK(inp);
1642 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1643 }
1644
1645 /*
1646 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1647 * port reservation, and preventing it from being returned by inpcb lookups.
1648 *
1649 * It is used by TCP to mark an inpcb as unused and avoid future packet
1650 * delivery or event notification when a socket remains open but TCP has
1651 * closed. This might occur as a result of a shutdown()-initiated TCP close
1652 * or a RST on the wire, and allows the port binding to be reused while still
1653 * maintaining the invariant that so_pcb always points to a valid inpcb until
1654 * in_pcbdetach().
1655 *
1656 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1657 * in_pcbnotifyall() and in_pcbpurgeif0()?
1658 */
1659 void
in_pcbdrop(struct inpcb * inp)1660 in_pcbdrop(struct inpcb *inp)
1661 {
1662
1663 INP_WLOCK_ASSERT(inp);
1664 #ifdef INVARIANTS
1665 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1666 MPASS(inp->inp_refcount > 1);
1667 #endif
1668
1669 /*
1670 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1671 * the hash lock...?
1672 */
1673 inp->inp_flags |= INP_DROPPED;
1674 if (inp->inp_flags & INP_INHASHLIST) {
1675 struct inpcbport *phd = inp->inp_phd;
1676
1677 INP_HASH_WLOCK(inp->inp_pcbinfo);
1678 in_pcbremlbgrouphash(inp);
1679 CK_LIST_REMOVE(inp, inp_hash);
1680 CK_LIST_REMOVE(inp, inp_portlist);
1681 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1682 CK_LIST_REMOVE(phd, phd_hash);
1683 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1684 }
1685 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1686 inp->inp_flags &= ~INP_INHASHLIST;
1687 #ifdef PCBGROUP
1688 in_pcbgroup_remove(inp);
1689 #endif
1690 }
1691 }
1692
1693 #ifdef INET
1694 /*
1695 * Common routines to return the socket addresses associated with inpcbs.
1696 */
1697 struct sockaddr *
in_sockaddr(in_port_t port,struct in_addr * addr_p)1698 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1699 {
1700 struct sockaddr_in *sin;
1701
1702 sin = malloc(sizeof *sin, M_SONAME,
1703 M_WAITOK | M_ZERO);
1704 sin->sin_family = AF_INET;
1705 sin->sin_len = sizeof(*sin);
1706 sin->sin_addr = *addr_p;
1707 sin->sin_port = port;
1708
1709 return (struct sockaddr *)sin;
1710 }
1711
1712 int
in_getsockaddr(struct socket * so,struct sockaddr ** nam)1713 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1714 {
1715 struct inpcb *inp;
1716 struct in_addr addr;
1717 in_port_t port;
1718
1719 inp = sotoinpcb(so);
1720 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1721
1722 INP_RLOCK(inp);
1723 port = inp->inp_lport;
1724 addr = inp->inp_laddr;
1725 INP_RUNLOCK(inp);
1726
1727 *nam = in_sockaddr(port, &addr);
1728 return 0;
1729 }
1730
1731 int
in_getpeeraddr(struct socket * so,struct sockaddr ** nam)1732 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1733 {
1734 struct inpcb *inp;
1735 struct in_addr addr;
1736 in_port_t port;
1737
1738 inp = sotoinpcb(so);
1739 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1740
1741 INP_RLOCK(inp);
1742 port = inp->inp_fport;
1743 addr = inp->inp_faddr;
1744 INP_RUNLOCK(inp);
1745
1746 *nam = in_sockaddr(port, &addr);
1747 return 0;
1748 }
1749
1750 void
in_pcbnotifyall(struct inpcbinfo * pcbinfo,struct in_addr faddr,int errno,struct inpcb * (* notify)(struct inpcb *,int))1751 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1752 struct inpcb *(*notify)(struct inpcb *, int))
1753 {
1754 struct inpcb *inp, *inp_temp;
1755
1756 INP_INFO_WLOCK(pcbinfo);
1757 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1758 INP_WLOCK(inp);
1759 #ifdef INET6
1760 if ((inp->inp_vflag & INP_IPV4) == 0) {
1761 INP_WUNLOCK(inp);
1762 continue;
1763 }
1764 #endif
1765 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1766 inp->inp_socket == NULL) {
1767 INP_WUNLOCK(inp);
1768 continue;
1769 }
1770 if ((*notify)(inp, errno))
1771 INP_WUNLOCK(inp);
1772 }
1773 INP_INFO_WUNLOCK(pcbinfo);
1774 }
1775
1776 void
in_pcbpurgeif0(struct inpcbinfo * pcbinfo,struct ifnet * ifp)1777 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1778 {
1779 struct inpcb *inp;
1780 struct in_multi *inm;
1781 struct in_mfilter *imf;
1782 struct ip_moptions *imo;
1783
1784 INP_INFO_WLOCK(pcbinfo);
1785 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1786 INP_WLOCK(inp);
1787 imo = inp->inp_moptions;
1788 if ((inp->inp_vflag & INP_IPV4) &&
1789 imo != NULL) {
1790 /*
1791 * Unselect the outgoing interface if it is being
1792 * detached.
1793 */
1794 if (imo->imo_multicast_ifp == ifp)
1795 imo->imo_multicast_ifp = NULL;
1796
1797 /*
1798 * Drop multicast group membership if we joined
1799 * through the interface being detached.
1800 *
1801 * XXX This can all be deferred to an epoch_call
1802 */
1803 restart:
1804 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1805 if ((inm = imf->imf_inm) == NULL)
1806 continue;
1807 if (inm->inm_ifp != ifp)
1808 continue;
1809 ip_mfilter_remove(&imo->imo_head, imf);
1810 IN_MULTI_LOCK_ASSERT();
1811 in_leavegroup_locked(inm, NULL);
1812 ip_mfilter_free(imf);
1813 goto restart;
1814 }
1815 }
1816 INP_WUNLOCK(inp);
1817 }
1818 INP_INFO_WUNLOCK(pcbinfo);
1819 }
1820
1821 /*
1822 * Lookup a PCB based on the local address and port. Caller must hold the
1823 * hash lock. No inpcb locks or references are acquired.
1824 */
1825 #define INP_LOOKUP_MAPPED_PCB_COST 3
1826 struct inpcb *
in_pcblookup_local(struct inpcbinfo * pcbinfo,struct in_addr laddr,u_short lport,int lookupflags,struct ucred * cred)1827 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1828 u_short lport, int lookupflags, struct ucred *cred)
1829 {
1830 struct inpcb *inp;
1831 #ifdef INET6
1832 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1833 #else
1834 int matchwild = 3;
1835 #endif
1836 int wildcard;
1837
1838 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1839 ("%s: invalid lookup flags %d", __func__, lookupflags));
1840
1841 INP_HASH_LOCK_ASSERT(pcbinfo);
1842
1843 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1844 struct inpcbhead *head;
1845 /*
1846 * Look for an unconnected (wildcard foreign addr) PCB that
1847 * matches the local address and port we're looking for.
1848 */
1849 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1850 0, pcbinfo->ipi_hashmask)];
1851 CK_LIST_FOREACH(inp, head, inp_hash) {
1852 #ifdef INET6
1853 /* XXX inp locking */
1854 if ((inp->inp_vflag & INP_IPV4) == 0)
1855 continue;
1856 #endif
1857 if (inp->inp_faddr.s_addr == INADDR_ANY &&
1858 inp->inp_laddr.s_addr == laddr.s_addr &&
1859 inp->inp_lport == lport) {
1860 /*
1861 * Found?
1862 */
1863 if (cred == NULL ||
1864 prison_equal_ip4(cred->cr_prison,
1865 inp->inp_cred->cr_prison))
1866 return (inp);
1867 }
1868 }
1869 /*
1870 * Not found.
1871 */
1872 return (NULL);
1873 } else {
1874 struct inpcbporthead *porthash;
1875 struct inpcbport *phd;
1876 struct inpcb *match = NULL;
1877 /*
1878 * Best fit PCB lookup.
1879 *
1880 * First see if this local port is in use by looking on the
1881 * port hash list.
1882 */
1883 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1884 pcbinfo->ipi_porthashmask)];
1885 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1886 if (phd->phd_port == lport)
1887 break;
1888 }
1889 if (phd != NULL) {
1890 /*
1891 * Port is in use by one or more PCBs. Look for best
1892 * fit.
1893 */
1894 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1895 wildcard = 0;
1896 if (cred != NULL &&
1897 !prison_equal_ip4(inp->inp_cred->cr_prison,
1898 cred->cr_prison))
1899 continue;
1900 #ifdef INET6
1901 /* XXX inp locking */
1902 if ((inp->inp_vflag & INP_IPV4) == 0)
1903 continue;
1904 /*
1905 * We never select the PCB that has
1906 * INP_IPV6 flag and is bound to :: if
1907 * we have another PCB which is bound
1908 * to 0.0.0.0. If a PCB has the
1909 * INP_IPV6 flag, then we set its cost
1910 * higher than IPv4 only PCBs.
1911 *
1912 * Note that the case only happens
1913 * when a socket is bound to ::, under
1914 * the condition that the use of the
1915 * mapped address is allowed.
1916 */
1917 if ((inp->inp_vflag & INP_IPV6) != 0)
1918 wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1919 #endif
1920 if (inp->inp_faddr.s_addr != INADDR_ANY)
1921 wildcard++;
1922 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1923 if (laddr.s_addr == INADDR_ANY)
1924 wildcard++;
1925 else if (inp->inp_laddr.s_addr != laddr.s_addr)
1926 continue;
1927 } else {
1928 if (laddr.s_addr != INADDR_ANY)
1929 wildcard++;
1930 }
1931 if (wildcard < matchwild) {
1932 match = inp;
1933 matchwild = wildcard;
1934 if (matchwild == 0)
1935 break;
1936 }
1937 }
1938 }
1939 return (match);
1940 }
1941 }
1942 #undef INP_LOOKUP_MAPPED_PCB_COST
1943
1944 static struct inpcb *
in_pcblookup_lbgroup(const struct inpcbinfo * pcbinfo,const struct in_addr * laddr,uint16_t lport,const struct in_addr * faddr,uint16_t fport,int lookupflags)1945 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1946 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1947 uint16_t fport, int lookupflags)
1948 {
1949 struct inpcb *local_wild;
1950 const struct inpcblbgrouphead *hdr;
1951 struct inpcblbgroup *grp;
1952 uint32_t idx;
1953
1954 INP_HASH_LOCK_ASSERT(pcbinfo);
1955
1956 hdr = &pcbinfo->ipi_lbgrouphashbase[
1957 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1958
1959 /*
1960 * Order of socket selection:
1961 * 1. non-wild.
1962 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1963 *
1964 * NOTE:
1965 * - Load balanced group does not contain jailed sockets
1966 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1967 */
1968 local_wild = NULL;
1969 CK_LIST_FOREACH(grp, hdr, il_list) {
1970 #ifdef INET6
1971 if (!(grp->il_vflag & INP_IPV4))
1972 continue;
1973 #endif
1974 if (grp->il_lport != lport)
1975 continue;
1976
1977 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1978 grp->il_inpcnt;
1979 if (grp->il_laddr.s_addr == laddr->s_addr)
1980 return (grp->il_inp[idx]);
1981 if (grp->il_laddr.s_addr == INADDR_ANY &&
1982 (lookupflags & INPLOOKUP_WILDCARD) != 0)
1983 local_wild = grp->il_inp[idx];
1984 }
1985 return (local_wild);
1986 }
1987
1988 #ifdef PCBGROUP
1989 /*
1990 * Lookup PCB in hash list, using pcbgroup tables.
1991 */
1992 static struct inpcb *
in_pcblookup_group(struct inpcbinfo * pcbinfo,struct inpcbgroup * pcbgroup,struct in_addr faddr,u_int fport_arg,struct in_addr laddr,u_int lport_arg,int lookupflags,struct ifnet * ifp)1993 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1994 struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1995 u_int lport_arg, int lookupflags, struct ifnet *ifp)
1996 {
1997 struct inpcbhead *head;
1998 struct inpcb *inp, *tmpinp;
1999 u_short fport = fport_arg, lport = lport_arg;
2000 bool locked;
2001
2002 /*
2003 * First look for an exact match.
2004 */
2005 tmpinp = NULL;
2006 INP_GROUP_LOCK(pcbgroup);
2007 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2008 pcbgroup->ipg_hashmask)];
2009 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2010 #ifdef INET6
2011 /* XXX inp locking */
2012 if ((inp->inp_vflag & INP_IPV4) == 0)
2013 continue;
2014 #endif
2015 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2016 inp->inp_laddr.s_addr == laddr.s_addr &&
2017 inp->inp_fport == fport &&
2018 inp->inp_lport == lport) {
2019 /*
2020 * XXX We should be able to directly return
2021 * the inp here, without any checks.
2022 * Well unless both bound with SO_REUSEPORT?
2023 */
2024 if (prison_flag(inp->inp_cred, PR_IP4))
2025 goto found;
2026 if (tmpinp == NULL)
2027 tmpinp = inp;
2028 }
2029 }
2030 if (tmpinp != NULL) {
2031 inp = tmpinp;
2032 goto found;
2033 }
2034
2035 #ifdef RSS
2036 /*
2037 * For incoming connections, we may wish to do a wildcard
2038 * match for an RSS-local socket.
2039 */
2040 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2041 struct inpcb *local_wild = NULL, *local_exact = NULL;
2042 #ifdef INET6
2043 struct inpcb *local_wild_mapped = NULL;
2044 #endif
2045 struct inpcb *jail_wild = NULL;
2046 struct inpcbhead *head;
2047 int injail;
2048
2049 /*
2050 * Order of socket selection - we always prefer jails.
2051 * 1. jailed, non-wild.
2052 * 2. jailed, wild.
2053 * 3. non-jailed, non-wild.
2054 * 4. non-jailed, wild.
2055 */
2056
2057 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2058 lport, 0, pcbgroup->ipg_hashmask)];
2059 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2060 #ifdef INET6
2061 /* XXX inp locking */
2062 if ((inp->inp_vflag & INP_IPV4) == 0)
2063 continue;
2064 #endif
2065 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2066 inp->inp_lport != lport)
2067 continue;
2068
2069 injail = prison_flag(inp->inp_cred, PR_IP4);
2070 if (injail) {
2071 if (prison_check_ip4(inp->inp_cred,
2072 &laddr) != 0)
2073 continue;
2074 } else {
2075 if (local_exact != NULL)
2076 continue;
2077 }
2078
2079 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2080 if (injail)
2081 goto found;
2082 else
2083 local_exact = inp;
2084 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2085 #ifdef INET6
2086 /* XXX inp locking, NULL check */
2087 if (inp->inp_vflag & INP_IPV6PROTO)
2088 local_wild_mapped = inp;
2089 else
2090 #endif
2091 if (injail)
2092 jail_wild = inp;
2093 else
2094 local_wild = inp;
2095 }
2096 } /* LIST_FOREACH */
2097
2098 inp = jail_wild;
2099 if (inp == NULL)
2100 inp = local_exact;
2101 if (inp == NULL)
2102 inp = local_wild;
2103 #ifdef INET6
2104 if (inp == NULL)
2105 inp = local_wild_mapped;
2106 #endif
2107 if (inp != NULL)
2108 goto found;
2109 }
2110 #endif
2111
2112 /*
2113 * Then look for a wildcard match, if requested.
2114 */
2115 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2116 struct inpcb *local_wild = NULL, *local_exact = NULL;
2117 #ifdef INET6
2118 struct inpcb *local_wild_mapped = NULL;
2119 #endif
2120 struct inpcb *jail_wild = NULL;
2121 struct inpcbhead *head;
2122 int injail;
2123
2124 /*
2125 * Order of socket selection - we always prefer jails.
2126 * 1. jailed, non-wild.
2127 * 2. jailed, wild.
2128 * 3. non-jailed, non-wild.
2129 * 4. non-jailed, wild.
2130 */
2131 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2132 0, pcbinfo->ipi_wildmask)];
2133 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2134 #ifdef INET6
2135 /* XXX inp locking */
2136 if ((inp->inp_vflag & INP_IPV4) == 0)
2137 continue;
2138 #endif
2139 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2140 inp->inp_lport != lport)
2141 continue;
2142
2143 injail = prison_flag(inp->inp_cred, PR_IP4);
2144 if (injail) {
2145 if (prison_check_ip4(inp->inp_cred,
2146 &laddr) != 0)
2147 continue;
2148 } else {
2149 if (local_exact != NULL)
2150 continue;
2151 }
2152
2153 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2154 if (injail)
2155 goto found;
2156 else
2157 local_exact = inp;
2158 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2159 #ifdef INET6
2160 /* XXX inp locking, NULL check */
2161 if (inp->inp_vflag & INP_IPV6PROTO)
2162 local_wild_mapped = inp;
2163 else
2164 #endif
2165 if (injail)
2166 jail_wild = inp;
2167 else
2168 local_wild = inp;
2169 }
2170 } /* LIST_FOREACH */
2171 inp = jail_wild;
2172 if (inp == NULL)
2173 inp = local_exact;
2174 if (inp == NULL)
2175 inp = local_wild;
2176 #ifdef INET6
2177 if (inp == NULL)
2178 inp = local_wild_mapped;
2179 #endif
2180 if (inp != NULL)
2181 goto found;
2182 } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2183 INP_GROUP_UNLOCK(pcbgroup);
2184 return (NULL);
2185
2186 found:
2187 if (lookupflags & INPLOOKUP_WLOCKPCB)
2188 locked = INP_TRY_WLOCK(inp);
2189 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2190 locked = INP_TRY_RLOCK(inp);
2191 else
2192 panic("%s: locking bug", __func__);
2193 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2194 if (lookupflags & INPLOOKUP_WLOCKPCB)
2195 INP_WUNLOCK(inp);
2196 else
2197 INP_RUNLOCK(inp);
2198 return (NULL);
2199 } else if (!locked)
2200 in_pcbref(inp);
2201 INP_GROUP_UNLOCK(pcbgroup);
2202 if (!locked) {
2203 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2204 INP_WLOCK(inp);
2205 if (in_pcbrele_wlocked(inp))
2206 return (NULL);
2207 } else {
2208 INP_RLOCK(inp);
2209 if (in_pcbrele_rlocked(inp))
2210 return (NULL);
2211 }
2212 }
2213 #ifdef INVARIANTS
2214 if (lookupflags & INPLOOKUP_WLOCKPCB)
2215 INP_WLOCK_ASSERT(inp);
2216 else
2217 INP_RLOCK_ASSERT(inp);
2218 #endif
2219 return (inp);
2220 }
2221 #endif /* PCBGROUP */
2222
2223 /*
2224 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes
2225 * that the caller has locked the hash list, and will not perform any further
2226 * locking or reference operations on either the hash list or the connection.
2227 */
2228 static struct inpcb *
in_pcblookup_hash_locked(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport_arg,struct in_addr laddr,u_int lport_arg,int lookupflags,struct ifnet * ifp)2229 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2230 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2231 struct ifnet *ifp)
2232 {
2233 struct inpcbhead *head;
2234 struct inpcb *inp, *tmpinp;
2235 u_short fport = fport_arg, lport = lport_arg;
2236
2237 #ifdef INVARIANTS
2238 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2239 ("%s: invalid lookup flags %d", __func__, lookupflags));
2240 if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2241 MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2242 #endif
2243 /*
2244 * First look for an exact match.
2245 */
2246 tmpinp = NULL;
2247 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2248 pcbinfo->ipi_hashmask)];
2249 CK_LIST_FOREACH(inp, head, inp_hash) {
2250 #ifdef INET6
2251 /* XXX inp locking */
2252 if ((inp->inp_vflag & INP_IPV4) == 0)
2253 continue;
2254 #endif
2255 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2256 inp->inp_laddr.s_addr == laddr.s_addr &&
2257 inp->inp_fport == fport &&
2258 inp->inp_lport == lport) {
2259 /*
2260 * XXX We should be able to directly return
2261 * the inp here, without any checks.
2262 * Well unless both bound with SO_REUSEPORT?
2263 */
2264 if (prison_flag(inp->inp_cred, PR_IP4))
2265 return (inp);
2266 if (tmpinp == NULL)
2267 tmpinp = inp;
2268 }
2269 }
2270 if (tmpinp != NULL)
2271 return (tmpinp);
2272
2273 /*
2274 * Then look in lb group (for wildcard match).
2275 */
2276 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2277 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2278 fport, lookupflags);
2279 if (inp != NULL)
2280 return (inp);
2281 }
2282
2283 /*
2284 * Then look for a wildcard match, if requested.
2285 */
2286 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2287 struct inpcb *local_wild = NULL, *local_exact = NULL;
2288 #ifdef INET6
2289 struct inpcb *local_wild_mapped = NULL;
2290 #endif
2291 struct inpcb *jail_wild = NULL;
2292 int injail;
2293
2294 /*
2295 * Order of socket selection - we always prefer jails.
2296 * 1. jailed, non-wild.
2297 * 2. jailed, wild.
2298 * 3. non-jailed, non-wild.
2299 * 4. non-jailed, wild.
2300 */
2301
2302 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2303 0, pcbinfo->ipi_hashmask)];
2304 CK_LIST_FOREACH(inp, head, inp_hash) {
2305 #ifdef INET6
2306 /* XXX inp locking */
2307 if ((inp->inp_vflag & INP_IPV4) == 0)
2308 continue;
2309 #endif
2310 if (inp->inp_faddr.s_addr != INADDR_ANY ||
2311 inp->inp_lport != lport)
2312 continue;
2313
2314 injail = prison_flag(inp->inp_cred, PR_IP4);
2315 if (injail) {
2316 if (prison_check_ip4(inp->inp_cred,
2317 &laddr) != 0)
2318 continue;
2319 } else {
2320 if (local_exact != NULL)
2321 continue;
2322 }
2323
2324 if (inp->inp_laddr.s_addr == laddr.s_addr) {
2325 if (injail)
2326 return (inp);
2327 else
2328 local_exact = inp;
2329 } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2330 #ifdef INET6
2331 /* XXX inp locking, NULL check */
2332 if (inp->inp_vflag & INP_IPV6PROTO)
2333 local_wild_mapped = inp;
2334 else
2335 #endif
2336 if (injail)
2337 jail_wild = inp;
2338 else
2339 local_wild = inp;
2340 }
2341 } /* LIST_FOREACH */
2342 if (jail_wild != NULL)
2343 return (jail_wild);
2344 if (local_exact != NULL)
2345 return (local_exact);
2346 if (local_wild != NULL)
2347 return (local_wild);
2348 #ifdef INET6
2349 if (local_wild_mapped != NULL)
2350 return (local_wild_mapped);
2351 #endif
2352 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2353
2354 return (NULL);
2355 }
2356
2357 /*
2358 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the
2359 * hash list lock, and will return the inpcb locked (i.e., requires
2360 * INPLOOKUP_LOCKPCB).
2361 */
2362 static struct inpcb *
in_pcblookup_hash(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport,struct in_addr laddr,u_int lport,int lookupflags,struct ifnet * ifp)2363 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2364 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2365 struct ifnet *ifp)
2366 {
2367 struct inpcb *inp;
2368
2369 INP_HASH_RLOCK(pcbinfo);
2370 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2371 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2372 if (inp != NULL) {
2373 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2374 INP_WLOCK(inp);
2375 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2376 INP_WUNLOCK(inp);
2377 inp = NULL;
2378 }
2379 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2380 INP_RLOCK(inp);
2381 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2382 INP_RUNLOCK(inp);
2383 inp = NULL;
2384 }
2385 } else
2386 panic("%s: locking bug", __func__);
2387 #ifdef INVARIANTS
2388 if (inp != NULL) {
2389 if (lookupflags & INPLOOKUP_WLOCKPCB)
2390 INP_WLOCK_ASSERT(inp);
2391 else
2392 INP_RLOCK_ASSERT(inp);
2393 }
2394 #endif
2395 }
2396 INP_HASH_RUNLOCK(pcbinfo);
2397 return (inp);
2398 }
2399
2400 /*
2401 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2402 * from which a pre-calculated hash value may be extracted.
2403 *
2404 * Possibly more of this logic should be in in_pcbgroup.c.
2405 */
2406 struct inpcb *
in_pcblookup(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport,struct in_addr laddr,u_int lport,int lookupflags,struct ifnet * ifp)2407 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2408 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2409 {
2410 #if defined(PCBGROUP) && !defined(RSS)
2411 struct inpcbgroup *pcbgroup;
2412 #endif
2413
2414 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2415 ("%s: invalid lookup flags %d", __func__, lookupflags));
2416 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2417 ("%s: LOCKPCB not set", __func__));
2418
2419 /*
2420 * When not using RSS, use connection groups in preference to the
2421 * reservation table when looking up 4-tuples. When using RSS, just
2422 * use the reservation table, due to the cost of the Toeplitz hash
2423 * in software.
2424 *
2425 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2426 * we could be doing RSS with a non-Toeplitz hash that is affordable
2427 * in software.
2428 */
2429 #if defined(PCBGROUP) && !defined(RSS)
2430 if (in_pcbgroup_enabled(pcbinfo)) {
2431 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2432 fport);
2433 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2434 laddr, lport, lookupflags, ifp));
2435 }
2436 #endif
2437 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2438 lookupflags, ifp));
2439 }
2440
2441 struct inpcb *
in_pcblookup_mbuf(struct inpcbinfo * pcbinfo,struct in_addr faddr,u_int fport,struct in_addr laddr,u_int lport,int lookupflags,struct ifnet * ifp,struct mbuf * m)2442 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2443 u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2444 struct ifnet *ifp, struct mbuf *m)
2445 {
2446 #ifdef PCBGROUP
2447 struct inpcbgroup *pcbgroup;
2448 #endif
2449
2450 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2451 ("%s: invalid lookup flags %d", __func__, lookupflags));
2452 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2453 ("%s: LOCKPCB not set", __func__));
2454
2455 #ifdef PCBGROUP
2456 /*
2457 * If we can use a hardware-generated hash to look up the connection
2458 * group, use that connection group to find the inpcb. Otherwise
2459 * fall back on a software hash -- or the reservation table if we're
2460 * using RSS.
2461 *
2462 * XXXRW: As above, that policy belongs in the pcbgroup code.
2463 */
2464 if (in_pcbgroup_enabled(pcbinfo) &&
2465 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2466 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2467 m->m_pkthdr.flowid);
2468 if (pcbgroup != NULL)
2469 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2470 fport, laddr, lport, lookupflags, ifp));
2471 #ifndef RSS
2472 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2473 fport);
2474 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2475 laddr, lport, lookupflags, ifp));
2476 #endif
2477 }
2478 #endif
2479 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2480 lookupflags, ifp));
2481 }
2482 #endif /* INET */
2483
2484 /*
2485 * Insert PCB onto various hash lists.
2486 */
2487 static int
in_pcbinshash_internal(struct inpcb * inp,int do_pcbgroup_update)2488 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2489 {
2490 struct inpcbhead *pcbhash;
2491 struct inpcbporthead *pcbporthash;
2492 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2493 struct inpcbport *phd;
2494 u_int32_t hashkey_faddr;
2495 int so_options;
2496
2497 INP_WLOCK_ASSERT(inp);
2498 INP_HASH_WLOCK_ASSERT(pcbinfo);
2499
2500 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2501 ("in_pcbinshash: INP_INHASHLIST"));
2502
2503 #ifdef INET6
2504 if (inp->inp_vflag & INP_IPV6)
2505 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2506 else
2507 #endif
2508 hashkey_faddr = inp->inp_faddr.s_addr;
2509
2510 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2511 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2512
2513 pcbporthash = &pcbinfo->ipi_porthashbase[
2514 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2515
2516 /*
2517 * Add entry to load balance group.
2518 * Only do this if SO_REUSEPORT_LB is set.
2519 */
2520 so_options = inp_so_options(inp);
2521 if (so_options & SO_REUSEPORT_LB) {
2522 int ret = in_pcbinslbgrouphash(inp);
2523 if (ret) {
2524 /* pcb lb group malloc fail (ret=ENOBUFS). */
2525 return (ret);
2526 }
2527 }
2528
2529 /*
2530 * Go through port list and look for a head for this lport.
2531 */
2532 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2533 if (phd->phd_port == inp->inp_lport)
2534 break;
2535 }
2536 /*
2537 * If none exists, malloc one and tack it on.
2538 */
2539 if (phd == NULL) {
2540 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2541 if (phd == NULL) {
2542 return (ENOBUFS); /* XXX */
2543 }
2544 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2545 phd->phd_port = inp->inp_lport;
2546 CK_LIST_INIT(&phd->phd_pcblist);
2547 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2548 }
2549 inp->inp_phd = phd;
2550 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2551 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2552 inp->inp_flags |= INP_INHASHLIST;
2553 #ifdef PCBGROUP
2554 if (do_pcbgroup_update)
2555 in_pcbgroup_update(inp);
2556 #endif
2557 return (0);
2558 }
2559
2560 /*
2561 * For now, there are two public interfaces to insert an inpcb into the hash
2562 * lists -- one that does update pcbgroups, and one that doesn't. The latter
2563 * is used only in the TCP syncache, where in_pcbinshash is called before the
2564 * full 4-tuple is set for the inpcb, and we don't want to install in the
2565 * pcbgroup until later.
2566 *
2567 * XXXRW: This seems like a misfeature. in_pcbinshash should always update
2568 * connection groups, and partially initialised inpcbs should not be exposed
2569 * to either reservation hash tables or pcbgroups.
2570 */
2571 int
in_pcbinshash(struct inpcb * inp)2572 in_pcbinshash(struct inpcb *inp)
2573 {
2574
2575 return (in_pcbinshash_internal(inp, 1));
2576 }
2577
2578 int
in_pcbinshash_nopcbgroup(struct inpcb * inp)2579 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2580 {
2581
2582 return (in_pcbinshash_internal(inp, 0));
2583 }
2584
2585 /*
2586 * Move PCB to the proper hash bucket when { faddr, fport } have been
2587 * changed. NOTE: This does not handle the case of the lport changing (the
2588 * hashed port list would have to be updated as well), so the lport must
2589 * not change after in_pcbinshash() has been called.
2590 */
2591 void
in_pcbrehash_mbuf(struct inpcb * inp,struct mbuf * m)2592 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2593 {
2594 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2595 struct inpcbhead *head;
2596 u_int32_t hashkey_faddr;
2597
2598 INP_WLOCK_ASSERT(inp);
2599 INP_HASH_WLOCK_ASSERT(pcbinfo);
2600
2601 KASSERT(inp->inp_flags & INP_INHASHLIST,
2602 ("in_pcbrehash: !INP_INHASHLIST"));
2603
2604 #ifdef INET6
2605 if (inp->inp_vflag & INP_IPV6)
2606 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2607 else
2608 #endif
2609 hashkey_faddr = inp->inp_faddr.s_addr;
2610
2611 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2612 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2613
2614 CK_LIST_REMOVE(inp, inp_hash);
2615 CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2616
2617 #ifdef PCBGROUP
2618 if (m != NULL)
2619 in_pcbgroup_update_mbuf(inp, m);
2620 else
2621 in_pcbgroup_update(inp);
2622 #endif
2623 }
2624
2625 void
in_pcbrehash(struct inpcb * inp)2626 in_pcbrehash(struct inpcb *inp)
2627 {
2628
2629 in_pcbrehash_mbuf(inp, NULL);
2630 }
2631
2632 /*
2633 * Remove PCB from various lists.
2634 */
2635 static void
in_pcbremlists(struct inpcb * inp)2636 in_pcbremlists(struct inpcb *inp)
2637 {
2638 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2639
2640 #ifdef INVARIANTS
2641 if (pcbinfo == &V_tcbinfo) {
2642 INP_INFO_RLOCK_ASSERT(pcbinfo);
2643 } else {
2644 INP_INFO_WLOCK_ASSERT(pcbinfo);
2645 }
2646 #endif
2647
2648 INP_WLOCK_ASSERT(inp);
2649 INP_LIST_WLOCK_ASSERT(pcbinfo);
2650
2651 inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2652 if (inp->inp_flags & INP_INHASHLIST) {
2653 struct inpcbport *phd = inp->inp_phd;
2654
2655 INP_HASH_WLOCK(pcbinfo);
2656
2657 /* XXX: Only do if SO_REUSEPORT_LB set? */
2658 in_pcbremlbgrouphash(inp);
2659
2660 CK_LIST_REMOVE(inp, inp_hash);
2661 CK_LIST_REMOVE(inp, inp_portlist);
2662 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2663 CK_LIST_REMOVE(phd, phd_hash);
2664 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2665 }
2666 INP_HASH_WUNLOCK(pcbinfo);
2667 inp->inp_flags &= ~INP_INHASHLIST;
2668 }
2669 CK_LIST_REMOVE(inp, inp_list);
2670 pcbinfo->ipi_count--;
2671 #ifdef PCBGROUP
2672 in_pcbgroup_remove(inp);
2673 #endif
2674 }
2675
2676 /*
2677 * Check for alternatives when higher level complains
2678 * about service problems. For now, invalidate cached
2679 * routing information. If the route was created dynamically
2680 * (by a redirect), time to try a default gateway again.
2681 */
2682 void
in_losing(struct inpcb * inp)2683 in_losing(struct inpcb *inp)
2684 {
2685
2686 RO_INVALIDATE_CACHE(&inp->inp_route);
2687 return;
2688 }
2689
2690 /*
2691 * A set label operation has occurred at the socket layer, propagate the
2692 * label change into the in_pcb for the socket.
2693 */
2694 void
in_pcbsosetlabel(struct socket * so)2695 in_pcbsosetlabel(struct socket *so)
2696 {
2697 #ifdef MAC
2698 struct inpcb *inp;
2699
2700 inp = sotoinpcb(so);
2701 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2702
2703 INP_WLOCK(inp);
2704 SOCK_LOCK(so);
2705 mac_inpcb_sosetlabel(so, inp);
2706 SOCK_UNLOCK(so);
2707 INP_WUNLOCK(inp);
2708 #endif
2709 }
2710
2711 /*
2712 * ipport_tick runs once per second, determining if random port allocation
2713 * should be continued. If more than ipport_randomcps ports have been
2714 * allocated in the last second, then we return to sequential port
2715 * allocation. We return to random allocation only once we drop below
2716 * ipport_randomcps for at least ipport_randomtime seconds.
2717 */
2718 static void
ipport_tick(void * xtp)2719 ipport_tick(void *xtp)
2720 {
2721 VNET_ITERATOR_DECL(vnet_iter);
2722
2723 VNET_LIST_RLOCK_NOSLEEP();
2724 VNET_FOREACH(vnet_iter) {
2725 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2726 if (V_ipport_tcpallocs <=
2727 V_ipport_tcplastcount + V_ipport_randomcps) {
2728 if (V_ipport_stoprandom > 0)
2729 V_ipport_stoprandom--;
2730 } else
2731 V_ipport_stoprandom = V_ipport_randomtime;
2732 V_ipport_tcplastcount = V_ipport_tcpallocs;
2733 CURVNET_RESTORE();
2734 }
2735 VNET_LIST_RUNLOCK_NOSLEEP();
2736 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2737 }
2738
2739 static void
ip_fini(void * xtp)2740 ip_fini(void *xtp)
2741 {
2742
2743 callout_stop(&ipport_tick_callout);
2744 }
2745
2746 /*
2747 * The ipport_callout should start running at about the time we attach the
2748 * inet or inet6 domains.
2749 */
2750 static void
ipport_tick_init(const void * unused __unused)2751 ipport_tick_init(const void *unused __unused)
2752 {
2753
2754 /* Start ipport_tick. */
2755 callout_init(&ipport_tick_callout, 1);
2756 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2757 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2758 SHUTDOWN_PRI_DEFAULT);
2759 }
2760 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2761 ipport_tick_init, NULL);
2762
2763 void
inp_wlock(struct inpcb * inp)2764 inp_wlock(struct inpcb *inp)
2765 {
2766
2767 INP_WLOCK(inp);
2768 }
2769
2770 void
inp_wunlock(struct inpcb * inp)2771 inp_wunlock(struct inpcb *inp)
2772 {
2773
2774 INP_WUNLOCK(inp);
2775 }
2776
2777 void
inp_rlock(struct inpcb * inp)2778 inp_rlock(struct inpcb *inp)
2779 {
2780
2781 INP_RLOCK(inp);
2782 }
2783
2784 void
inp_runlock(struct inpcb * inp)2785 inp_runlock(struct inpcb *inp)
2786 {
2787
2788 INP_RUNLOCK(inp);
2789 }
2790
2791 #ifdef INVARIANT_SUPPORT
2792 void
inp_lock_assert(struct inpcb * inp)2793 inp_lock_assert(struct inpcb *inp)
2794 {
2795
2796 INP_WLOCK_ASSERT(inp);
2797 }
2798
2799 void
inp_unlock_assert(struct inpcb * inp)2800 inp_unlock_assert(struct inpcb *inp)
2801 {
2802
2803 INP_UNLOCK_ASSERT(inp);
2804 }
2805 #endif
2806
2807 void
inp_apply_all(void (* func)(struct inpcb *,void *),void * arg)2808 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2809 {
2810 struct inpcb *inp;
2811
2812 INP_INFO_WLOCK(&V_tcbinfo);
2813 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2814 INP_WLOCK(inp);
2815 func(inp, arg);
2816 INP_WUNLOCK(inp);
2817 }
2818 INP_INFO_WUNLOCK(&V_tcbinfo);
2819 }
2820
2821 struct socket *
inp_inpcbtosocket(struct inpcb * inp)2822 inp_inpcbtosocket(struct inpcb *inp)
2823 {
2824
2825 INP_WLOCK_ASSERT(inp);
2826 return (inp->inp_socket);
2827 }
2828
2829 struct tcpcb *
inp_inpcbtotcpcb(struct inpcb * inp)2830 inp_inpcbtotcpcb(struct inpcb *inp)
2831 {
2832
2833 INP_WLOCK_ASSERT(inp);
2834 return ((struct tcpcb *)inp->inp_ppcb);
2835 }
2836
2837 int
inp_ip_tos_get(const struct inpcb * inp)2838 inp_ip_tos_get(const struct inpcb *inp)
2839 {
2840
2841 return (inp->inp_ip_tos);
2842 }
2843
2844 void
inp_ip_tos_set(struct inpcb * inp,int val)2845 inp_ip_tos_set(struct inpcb *inp, int val)
2846 {
2847
2848 inp->inp_ip_tos = val;
2849 }
2850
2851 void
inp_4tuple_get(struct inpcb * inp,uint32_t * laddr,uint16_t * lp,uint32_t * faddr,uint16_t * fp)2852 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2853 uint32_t *faddr, uint16_t *fp)
2854 {
2855
2856 INP_LOCK_ASSERT(inp);
2857 *laddr = inp->inp_laddr.s_addr;
2858 *faddr = inp->inp_faddr.s_addr;
2859 *lp = inp->inp_lport;
2860 *fp = inp->inp_fport;
2861 }
2862
2863 struct inpcb *
so_sotoinpcb(struct socket * so)2864 so_sotoinpcb(struct socket *so)
2865 {
2866
2867 return (sotoinpcb(so));
2868 }
2869
2870 struct tcpcb *
so_sototcpcb(struct socket * so)2871 so_sototcpcb(struct socket *so)
2872 {
2873
2874 return (sototcpcb(so));
2875 }
2876
2877 /*
2878 * Create an external-format (``xinpcb'') structure using the information in
2879 * the kernel-format in_pcb structure pointed to by inp. This is done to
2880 * reduce the spew of irrelevant information over this interface, to isolate
2881 * user code from changes in the kernel structure, and potentially to provide
2882 * information-hiding if we decide that some of this information should be
2883 * hidden from users.
2884 */
2885 void
in_pcbtoxinpcb(const struct inpcb * inp,struct xinpcb * xi)2886 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2887 {
2888
2889 bzero(xi, sizeof(*xi));
2890 xi->xi_len = sizeof(struct xinpcb);
2891 if (inp->inp_socket)
2892 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2893 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2894 xi->inp_gencnt = inp->inp_gencnt;
2895 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2896 xi->inp_flow = inp->inp_flow;
2897 xi->inp_flowid = inp->inp_flowid;
2898 xi->inp_flowtype = inp->inp_flowtype;
2899 xi->inp_flags = inp->inp_flags;
2900 xi->inp_flags2 = inp->inp_flags2;
2901 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2902 xi->in6p_cksum = inp->in6p_cksum;
2903 xi->in6p_hops = inp->in6p_hops;
2904 xi->inp_ip_tos = inp->inp_ip_tos;
2905 xi->inp_vflag = inp->inp_vflag;
2906 xi->inp_ip_ttl = inp->inp_ip_ttl;
2907 xi->inp_ip_p = inp->inp_ip_p;
2908 xi->inp_ip_minttl = inp->inp_ip_minttl;
2909 }
2910
2911 #ifdef DDB
2912 static void
db_print_indent(int indent)2913 db_print_indent(int indent)
2914 {
2915 int i;
2916
2917 for (i = 0; i < indent; i++)
2918 db_printf(" ");
2919 }
2920
2921 static void
db_print_inconninfo(struct in_conninfo * inc,const char * name,int indent)2922 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2923 {
2924 char faddr_str[48], laddr_str[48];
2925
2926 db_print_indent(indent);
2927 db_printf("%s at %p\n", name, inc);
2928
2929 indent += 2;
2930
2931 #ifdef INET6
2932 if (inc->inc_flags & INC_ISIPV6) {
2933 /* IPv6. */
2934 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2935 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2936 } else
2937 #endif
2938 {
2939 /* IPv4. */
2940 inet_ntoa_r(inc->inc_laddr, laddr_str);
2941 inet_ntoa_r(inc->inc_faddr, faddr_str);
2942 }
2943 db_print_indent(indent);
2944 db_printf("inc_laddr %s inc_lport %u\n", laddr_str,
2945 ntohs(inc->inc_lport));
2946 db_print_indent(indent);
2947 db_printf("inc_faddr %s inc_fport %u\n", faddr_str,
2948 ntohs(inc->inc_fport));
2949 }
2950
2951 static void
db_print_inpflags(int inp_flags)2952 db_print_inpflags(int inp_flags)
2953 {
2954 int comma;
2955
2956 comma = 0;
2957 if (inp_flags & INP_RECVOPTS) {
2958 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2959 comma = 1;
2960 }
2961 if (inp_flags & INP_RECVRETOPTS) {
2962 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2963 comma = 1;
2964 }
2965 if (inp_flags & INP_RECVDSTADDR) {
2966 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2967 comma = 1;
2968 }
2969 if (inp_flags & INP_ORIGDSTADDR) {
2970 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2971 comma = 1;
2972 }
2973 if (inp_flags & INP_HDRINCL) {
2974 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2975 comma = 1;
2976 }
2977 if (inp_flags & INP_HIGHPORT) {
2978 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2979 comma = 1;
2980 }
2981 if (inp_flags & INP_LOWPORT) {
2982 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2983 comma = 1;
2984 }
2985 if (inp_flags & INP_ANONPORT) {
2986 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2987 comma = 1;
2988 }
2989 if (inp_flags & INP_RECVIF) {
2990 db_printf("%sINP_RECVIF", comma ? ", " : "");
2991 comma = 1;
2992 }
2993 if (inp_flags & INP_MTUDISC) {
2994 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2995 comma = 1;
2996 }
2997 if (inp_flags & INP_RECVTTL) {
2998 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2999 comma = 1;
3000 }
3001 if (inp_flags & INP_DONTFRAG) {
3002 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3003 comma = 1;
3004 }
3005 if (inp_flags & INP_RECVTOS) {
3006 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3007 comma = 1;
3008 }
3009 if (inp_flags & IN6P_IPV6_V6ONLY) {
3010 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3011 comma = 1;
3012 }
3013 if (inp_flags & IN6P_PKTINFO) {
3014 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3015 comma = 1;
3016 }
3017 if (inp_flags & IN6P_HOPLIMIT) {
3018 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3019 comma = 1;
3020 }
3021 if (inp_flags & IN6P_HOPOPTS) {
3022 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3023 comma = 1;
3024 }
3025 if (inp_flags & IN6P_DSTOPTS) {
3026 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3027 comma = 1;
3028 }
3029 if (inp_flags & IN6P_RTHDR) {
3030 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3031 comma = 1;
3032 }
3033 if (inp_flags & IN6P_RTHDRDSTOPTS) {
3034 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3035 comma = 1;
3036 }
3037 if (inp_flags & IN6P_TCLASS) {
3038 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3039 comma = 1;
3040 }
3041 if (inp_flags & IN6P_AUTOFLOWLABEL) {
3042 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3043 comma = 1;
3044 }
3045 if (inp_flags & INP_TIMEWAIT) {
3046 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3047 comma = 1;
3048 }
3049 if (inp_flags & INP_ONESBCAST) {
3050 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3051 comma = 1;
3052 }
3053 if (inp_flags & INP_DROPPED) {
3054 db_printf("%sINP_DROPPED", comma ? ", " : "");
3055 comma = 1;
3056 }
3057 if (inp_flags & INP_SOCKREF) {
3058 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3059 comma = 1;
3060 }
3061 if (inp_flags & IN6P_RFC2292) {
3062 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3063 comma = 1;
3064 }
3065 if (inp_flags & IN6P_MTU) {
3066 db_printf("IN6P_MTU%s", comma ? ", " : "");
3067 comma = 1;
3068 }
3069 }
3070
3071 static void
db_print_inpvflag(u_char inp_vflag)3072 db_print_inpvflag(u_char inp_vflag)
3073 {
3074 int comma;
3075
3076 comma = 0;
3077 if (inp_vflag & INP_IPV4) {
3078 db_printf("%sINP_IPV4", comma ? ", " : "");
3079 comma = 1;
3080 }
3081 if (inp_vflag & INP_IPV6) {
3082 db_printf("%sINP_IPV6", comma ? ", " : "");
3083 comma = 1;
3084 }
3085 if (inp_vflag & INP_IPV6PROTO) {
3086 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3087 comma = 1;
3088 }
3089 }
3090
3091 static void
db_print_inpcb(struct inpcb * inp,const char * name,int indent)3092 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3093 {
3094
3095 db_print_indent(indent);
3096 db_printf("%s at %p\n", name, inp);
3097
3098 indent += 2;
3099
3100 db_print_indent(indent);
3101 db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3102
3103 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3104
3105 db_print_indent(indent);
3106 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n",
3107 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3108
3109 db_print_indent(indent);
3110 db_printf("inp_label: %p inp_flags: 0x%x (",
3111 inp->inp_label, inp->inp_flags);
3112 db_print_inpflags(inp->inp_flags);
3113 db_printf(")\n");
3114
3115 db_print_indent(indent);
3116 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp,
3117 inp->inp_vflag);
3118 db_print_inpvflag(inp->inp_vflag);
3119 db_printf(")\n");
3120
3121 db_print_indent(indent);
3122 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n",
3123 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3124
3125 db_print_indent(indent);
3126 #ifdef INET6
3127 if (inp->inp_vflag & INP_IPV6) {
3128 db_printf("in6p_options: %p in6p_outputopts: %p "
3129 "in6p_moptions: %p\n", inp->in6p_options,
3130 inp->in6p_outputopts, inp->in6p_moptions);
3131 db_printf("in6p_icmp6filt: %p in6p_cksum %d "
3132 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3133 inp->in6p_hops);
3134 } else
3135 #endif
3136 {
3137 db_printf("inp_ip_tos: %d inp_ip_options: %p "
3138 "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3139 inp->inp_options, inp->inp_moptions);
3140 }
3141
3142 db_print_indent(indent);
3143 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd,
3144 (uintmax_t)inp->inp_gencnt);
3145 }
3146
DB_SHOW_COMMAND(inpcb,db_show_inpcb)3147 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3148 {
3149 struct inpcb *inp;
3150
3151 if (!have_addr) {
3152 db_printf("usage: show inpcb <addr>\n");
3153 return;
3154 }
3155 inp = (struct inpcb *)addr;
3156
3157 db_print_inpcb(inp, "inpcb", 0);
3158 }
3159 #endif /* DDB */
3160
3161 #ifdef RATELIMIT
3162 /*
3163 * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3164 * if any.
3165 */
3166 int
in_pcbmodify_txrtlmt(struct inpcb * inp,uint32_t max_pacing_rate)3167 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3168 {
3169 union if_snd_tag_modify_params params = {
3170 .rate_limit.max_rate = max_pacing_rate,
3171 };
3172 struct m_snd_tag *mst;
3173 struct ifnet *ifp;
3174 int error;
3175
3176 mst = inp->inp_snd_tag;
3177 if (mst == NULL)
3178 return (EINVAL);
3179
3180 ifp = mst->ifp;
3181 if (ifp == NULL)
3182 return (EINVAL);
3183
3184 if (ifp->if_snd_tag_modify == NULL) {
3185 error = EOPNOTSUPP;
3186 } else {
3187 error = ifp->if_snd_tag_modify(mst, ¶ms);
3188 }
3189 return (error);
3190 }
3191
3192 /*
3193 * Query existing TX rate limit based on the existing
3194 * "inp->inp_snd_tag", if any.
3195 */
3196 int
in_pcbquery_txrtlmt(struct inpcb * inp,uint32_t * p_max_pacing_rate)3197 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3198 {
3199 union if_snd_tag_query_params params = { };
3200 struct m_snd_tag *mst;
3201 struct ifnet *ifp;
3202 int error;
3203
3204 mst = inp->inp_snd_tag;
3205 if (mst == NULL)
3206 return (EINVAL);
3207
3208 ifp = mst->ifp;
3209 if (ifp == NULL)
3210 return (EINVAL);
3211
3212 if (ifp->if_snd_tag_query == NULL) {
3213 error = EOPNOTSUPP;
3214 } else {
3215 error = ifp->if_snd_tag_query(mst, ¶ms);
3216 if (error == 0 && p_max_pacing_rate != NULL)
3217 *p_max_pacing_rate = params.rate_limit.max_rate;
3218 }
3219 return (error);
3220 }
3221
3222 /*
3223 * Query existing TX queue level based on the existing
3224 * "inp->inp_snd_tag", if any.
3225 */
3226 int
in_pcbquery_txrlevel(struct inpcb * inp,uint32_t * p_txqueue_level)3227 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3228 {
3229 union if_snd_tag_query_params params = { };
3230 struct m_snd_tag *mst;
3231 struct ifnet *ifp;
3232 int error;
3233
3234 mst = inp->inp_snd_tag;
3235 if (mst == NULL)
3236 return (EINVAL);
3237
3238 ifp = mst->ifp;
3239 if (ifp == NULL)
3240 return (EINVAL);
3241
3242 if (ifp->if_snd_tag_query == NULL)
3243 return (EOPNOTSUPP);
3244
3245 error = ifp->if_snd_tag_query(mst, ¶ms);
3246 if (error == 0 && p_txqueue_level != NULL)
3247 *p_txqueue_level = params.rate_limit.queue_level;
3248 return (error);
3249 }
3250
3251 /*
3252 * Allocate a new TX rate limit send tag from the network interface
3253 * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3254 */
3255 int
in_pcbattach_txrtlmt(struct inpcb * inp,struct ifnet * ifp,uint32_t flowtype,uint32_t flowid,uint32_t max_pacing_rate)3256 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3257 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3258 {
3259 union if_snd_tag_alloc_params params = {
3260 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3261 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3262 .rate_limit.hdr.flowid = flowid,
3263 .rate_limit.hdr.flowtype = flowtype,
3264 .rate_limit.max_rate = max_pacing_rate,
3265 };
3266 int error;
3267
3268 INP_WLOCK_ASSERT(inp);
3269
3270 if (inp->inp_snd_tag != NULL)
3271 return (EINVAL);
3272
3273 if (ifp->if_snd_tag_alloc == NULL) {
3274 error = EOPNOTSUPP;
3275 } else {
3276 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag);
3277
3278 /*
3279 * At success increment the refcount on
3280 * the send tag's network interface:
3281 */
3282 if (error == 0)
3283 if_ref(inp->inp_snd_tag->ifp);
3284 }
3285 return (error);
3286 }
3287
3288 /*
3289 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3290 * if any:
3291 */
3292 void
in_pcbdetach_txrtlmt(struct inpcb * inp)3293 in_pcbdetach_txrtlmt(struct inpcb *inp)
3294 {
3295 struct m_snd_tag *mst;
3296 struct ifnet *ifp;
3297
3298 INP_WLOCK_ASSERT(inp);
3299
3300 mst = inp->inp_snd_tag;
3301 inp->inp_snd_tag = NULL;
3302
3303 if (mst == NULL)
3304 return;
3305
3306 ifp = mst->ifp;
3307 if (ifp == NULL)
3308 return;
3309
3310 /*
3311 * If the device was detached while we still had reference(s)
3312 * on the ifp, we assume if_snd_tag_free() was replaced with
3313 * stubs.
3314 */
3315 ifp->if_snd_tag_free(mst);
3316
3317 /* release reference count on network interface */
3318 if_rele(ifp);
3319 }
3320
3321 /*
3322 * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3323 * is set in the fast path and will attach/detach/modify the TX rate
3324 * limit send tag based on the socket's so_max_pacing_rate value.
3325 */
3326 void
in_pcboutput_txrtlmt(struct inpcb * inp,struct ifnet * ifp,struct mbuf * mb)3327 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3328 {
3329 struct socket *socket;
3330 uint32_t max_pacing_rate;
3331 bool did_upgrade;
3332 int error;
3333
3334 if (inp == NULL)
3335 return;
3336
3337 socket = inp->inp_socket;
3338 if (socket == NULL)
3339 return;
3340
3341 if (!INP_WLOCKED(inp)) {
3342 /*
3343 * NOTE: If the write locking fails, we need to bail
3344 * out and use the non-ratelimited ring for the
3345 * transmit until there is a new chance to get the
3346 * write lock.
3347 */
3348 if (!INP_TRY_UPGRADE(inp))
3349 return;
3350 did_upgrade = 1;
3351 } else {
3352 did_upgrade = 0;
3353 }
3354
3355 /*
3356 * NOTE: The so_max_pacing_rate value is read unlocked,
3357 * because atomic updates are not required since the variable
3358 * is checked at every mbuf we send. It is assumed that the
3359 * variable read itself will be atomic.
3360 */
3361 max_pacing_rate = socket->so_max_pacing_rate;
3362
3363 /*
3364 * NOTE: When attaching to a network interface a reference is
3365 * made to ensure the network interface doesn't go away until
3366 * all ratelimit connections are gone. The network interface
3367 * pointers compared below represent valid network interfaces,
3368 * except when comparing towards NULL.
3369 */
3370 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3371 error = 0;
3372 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3373 if (inp->inp_snd_tag != NULL)
3374 in_pcbdetach_txrtlmt(inp);
3375 error = 0;
3376 } else if (inp->inp_snd_tag == NULL) {
3377 /*
3378 * In order to utilize packet pacing with RSS, we need
3379 * to wait until there is a valid RSS hash before we
3380 * can proceed:
3381 */
3382 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3383 error = EAGAIN;
3384 } else {
3385 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3386 mb->m_pkthdr.flowid, max_pacing_rate);
3387 }
3388 } else {
3389 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3390 }
3391 if (error == 0 || error == EOPNOTSUPP)
3392 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3393 if (did_upgrade)
3394 INP_DOWNGRADE(inp);
3395 }
3396
3397 /*
3398 * Track route changes for TX rate limiting.
3399 */
3400 void
in_pcboutput_eagain(struct inpcb * inp)3401 in_pcboutput_eagain(struct inpcb *inp)
3402 {
3403 bool did_upgrade;
3404
3405 if (inp == NULL)
3406 return;
3407
3408 if (inp->inp_snd_tag == NULL)
3409 return;
3410
3411 if (!INP_WLOCKED(inp)) {
3412 /*
3413 * NOTE: If the write locking fails, we need to bail
3414 * out and use the non-ratelimited ring for the
3415 * transmit until there is a new chance to get the
3416 * write lock.
3417 */
3418 if (!INP_TRY_UPGRADE(inp))
3419 return;
3420 did_upgrade = 1;
3421 } else {
3422 did_upgrade = 0;
3423 }
3424
3425 /* detach rate limiting */
3426 in_pcbdetach_txrtlmt(inp);
3427
3428 /* make sure new mbuf send tag allocation is made */
3429 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3430
3431 if (did_upgrade)
3432 INP_DOWNGRADE(inp);
3433 }
3434 #endif /* RATELIMIT */
3435