1 /* $NetBSD: route.c,v 1.234 2022/09/20 02:23:37 knakahara Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Kevin M. Lahey of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 35 * All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the project nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 */ 61 62 /* 63 * Copyright (c) 1980, 1986, 1991, 1993 64 * The Regents of the University of California. All rights reserved. 65 * 66 * Redistribution and use in source and binary forms, with or without 67 * modification, are permitted provided that the following conditions 68 * are met: 69 * 1. Redistributions of source code must retain the above copyright 70 * notice, this list of conditions and the following disclaimer. 71 * 2. Redistributions in binary form must reproduce the above copyright 72 * notice, this list of conditions and the following disclaimer in the 73 * documentation and/or other materials provided with the distribution. 74 * 3. Neither the name of the University nor the names of its contributors 75 * may be used to endorse or promote products derived from this software 76 * without specific prior written permission. 77 * 78 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 79 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 80 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 81 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 82 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 83 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 84 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 85 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 86 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 87 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 88 * SUCH DAMAGE. 89 * 90 * @(#)route.c 8.3 (Berkeley) 1/9/95 91 */ 92 93 #ifdef _KERNEL_OPT 94 #include "opt_inet.h" 95 #include "opt_route.h" 96 #include "opt_net_mpsafe.h" 97 #endif 98 99 #include <sys/cdefs.h> 100 __KERNEL_RCSID(0, "$NetBSD: route.c,v 1.234 2022/09/20 02:23:37 knakahara Exp $"); 101 102 #include <sys/param.h> 103 #ifdef RTFLUSH_DEBUG 104 #include <sys/sysctl.h> 105 #endif 106 #include <sys/systm.h> 107 #include <sys/callout.h> 108 #include <sys/proc.h> 109 #include <sys/mbuf.h> 110 #include <sys/socket.h> 111 #include <sys/socketvar.h> 112 #include <sys/domain.h> 113 #include <sys/kernel.h> 114 #include <sys/ioctl.h> 115 #include <sys/pool.h> 116 #include <sys/kauth.h> 117 #include <sys/workqueue.h> 118 #include <sys/syslog.h> 119 #include <sys/rwlock.h> 120 #include <sys/mutex.h> 121 #include <sys/cpu.h> 122 #include <sys/kmem.h> 123 124 #include <net/if.h> 125 #include <net/if_dl.h> 126 #include <net/route.h> 127 #if defined(INET) || defined(INET6) 128 #include <net/if_llatbl.h> 129 #endif 130 131 #include <netinet/in.h> 132 #include <netinet/in_var.h> 133 134 #define PRESERVED_RTF (RTF_UP | RTF_GATEWAY | RTF_HOST | RTF_DONE | RTF_MASK) 135 136 #ifdef RTFLUSH_DEBUG 137 #define rtcache_debug() __predict_false(_rtcache_debug) 138 #else /* RTFLUSH_DEBUG */ 139 #define rtcache_debug() 0 140 #endif /* RTFLUSH_DEBUG */ 141 142 #ifdef RT_DEBUG 143 #define RT_REFCNT_TRACE(rt) printf("%s:%d: rt=%p refcnt=%d\n", \ 144 __func__, __LINE__, (rt), (rt)->rt_refcnt) 145 #else 146 #define RT_REFCNT_TRACE(rt) do {} while (0) 147 #endif 148 149 #ifdef RT_DEBUG 150 #define dlog(level, fmt, args...) log(level, fmt, ##args) 151 #else 152 #define dlog(level, fmt, args...) do {} while (0) 153 #endif 154 155 struct rtstat rtstat; 156 157 static int rttrash; /* routes not in table but not freed */ 158 159 static struct pool rtentry_pool; 160 static struct pool rttimer_pool; 161 162 static struct callout rt_timer_ch; /* callout for rt_timer_timer() */ 163 static struct workqueue *rt_timer_wq; 164 static struct work rt_timer_wk; 165 166 static void rt_timer_init(void); 167 static void rt_timer_queue_remove_all(struct rttimer_queue *); 168 static void rt_timer_remove_all(struct rtentry *); 169 static void rt_timer_timer(void *); 170 171 /* 172 * Locking notes: 173 * - The routing table is protected by a global rwlock 174 * - API: RT_RLOCK and friends 175 * - rtcaches are NOT protected by the framework 176 * - Callers must guarantee a rtcache isn't accessed simultaneously 177 * - How the constraint is guaranteed in the wild 178 * - Protect a rtcache by a mutex (e.g., inp_route) 179 * - Make rtcache per-CPU and allow only accesses from softint 180 * (e.g., ipforward_rt_percpu) 181 * - References to a rtentry is managed by reference counting and psref 182 * - Reference counting is used for temporal reference when a rtentry 183 * is fetched from the routing table 184 * - psref is used for temporal reference when a rtentry is fetched 185 * from a rtcache 186 * - struct route (rtcache) has struct psref, so we cannot obtain 187 * a reference twice on the same struct route 188 * - Before destroying or updating a rtentry, we have to wait for 189 * all references left (see below for details) 190 * - APIs 191 * - An obtained rtentry via rtalloc1 or rtrequest* must be 192 * unreferenced by rt_unref 193 * - An obtained rtentry via rtcache_* must be unreferenced by 194 * rtcache_unref 195 * - TODO: once we get a lockless routing table, we should use only 196 * psref for rtentries 197 * - rtentry destruction 198 * - A rtentry is destroyed (freed) only when we call rtrequest(RTM_DELETE) 199 * - If a caller of rtrequest grabs a reference of a rtentry, the caller 200 * has a responsibility to destroy the rtentry by itself by calling 201 * rt_free 202 * - If not, rtrequest itself does that 203 * - If rt_free is called in softint, the actual destruction routine is 204 * deferred to a workqueue 205 * - rtentry update 206 * - When updating a rtentry, RTF_UPDATING flag is set 207 * - If a rtentry is set RTF_UPDATING, fetching the rtentry from 208 * the routing table or a rtcache results in either of the following 209 * cases: 210 * - if the caller runs in softint, the caller fails to fetch 211 * - otherwise, the caller waits for the update completed and retries 212 * to fetch (probably succeed to fetch for the second time) 213 * - rtcache invalidation 214 * - There is a global generation counter that is incremented when 215 * any routes have been added or deleted 216 * - When a rtcache caches a rtentry into itself, it also stores 217 * a snapshot of the generation counter 218 * - If the snapshot equals to the global counter, the cache is valid, 219 * otherwise the cache is invalidated 220 */ 221 222 /* 223 * Global lock for the routing table. 224 */ 225 static krwlock_t rt_lock __cacheline_aligned; 226 #ifdef NET_MPSAFE 227 #define RT_RLOCK() rw_enter(&rt_lock, RW_READER) 228 #define RT_WLOCK() rw_enter(&rt_lock, RW_WRITER) 229 #define RT_UNLOCK() rw_exit(&rt_lock) 230 #define RT_WLOCKED() rw_write_held(&rt_lock) 231 #define RT_ASSERT_WLOCK() KASSERT(rw_write_held(&rt_lock)) 232 #else 233 #define RT_RLOCK() do {} while (0) 234 #define RT_WLOCK() do {} while (0) 235 #define RT_UNLOCK() do {} while (0) 236 #define RT_WLOCKED() true 237 #define RT_ASSERT_WLOCK() do {} while (0) 238 #endif 239 240 static uint64_t rtcache_generation; 241 242 /* 243 * mutex and cv that are used to wait for references to a rtentry left 244 * before updating the rtentry. 245 */ 246 static struct { 247 kmutex_t lock; 248 kcondvar_t cv; 249 bool ongoing; 250 const struct lwp *lwp; 251 } rt_update_global __cacheline_aligned; 252 253 /* 254 * A workqueue and stuff that are used to defer the destruction routine 255 * of rtentries. 256 */ 257 static struct { 258 struct workqueue *wq; 259 struct work wk; 260 kmutex_t lock; 261 SLIST_HEAD(, rtentry) queue; 262 bool enqueued; 263 } rt_free_global __cacheline_aligned; 264 265 /* psref for rtentry */ 266 static struct psref_class *rt_psref_class __read_mostly; 267 268 #ifdef RTFLUSH_DEBUG 269 static int _rtcache_debug = 0; 270 #endif /* RTFLUSH_DEBUG */ 271 272 static kauth_listener_t route_listener; 273 274 static int rtdeletemsg(struct rtentry *); 275 276 static void rt_maskedcopy(const struct sockaddr *, 277 struct sockaddr *, const struct sockaddr *); 278 279 static void rtcache_invalidate(void); 280 281 static void rt_ref(struct rtentry *); 282 283 static struct rtentry * 284 rtalloc1_locked(const struct sockaddr *, int, bool, bool); 285 286 static struct ifaddr *rt_getifa(struct rt_addrinfo *, struct psref *); 287 static struct ifnet *rt_getifp(struct rt_addrinfo *, struct psref *); 288 static struct ifaddr *ifa_ifwithroute_psref(int, const struct sockaddr *, 289 const struct sockaddr *, struct psref *); 290 291 static void rtcache_ref(struct rtentry *, struct route *); 292 293 #ifdef NET_MPSAFE 294 static void rt_update_wait(void); 295 #endif 296 297 static bool rt_wait_ok(void); 298 static void rt_wait_refcnt(const char *, struct rtentry *, int); 299 static void rt_wait_psref(struct rtentry *); 300 301 #ifdef DDB 302 static void db_print_sa(const struct sockaddr *); 303 static void db_print_ifa(struct ifaddr *); 304 static int db_show_rtentry(struct rtentry *, void *); 305 #endif 306 307 #ifdef RTFLUSH_DEBUG 308 static void sysctl_net_rtcache_setup(struct sysctllog **); 309 static void 310 sysctl_net_rtcache_setup(struct sysctllog **clog) 311 { 312 const struct sysctlnode *rnode; 313 314 if (sysctl_createv(clog, 0, NULL, &rnode, CTLFLAG_PERMANENT, 315 CTLTYPE_NODE, 316 "rtcache", SYSCTL_DESCR("Route cache related settings"), 317 NULL, 0, NULL, 0, CTL_NET, CTL_CREATE, CTL_EOL) != 0) 318 return; 319 if (sysctl_createv(clog, 0, &rnode, &rnode, 320 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_INT, 321 "debug", SYSCTL_DESCR("Debug route caches"), 322 NULL, 0, &_rtcache_debug, 0, CTL_CREATE, CTL_EOL) != 0) 323 return; 324 } 325 #endif /* RTFLUSH_DEBUG */ 326 327 static inline void 328 rt_destroy(struct rtentry *rt) 329 { 330 if (rt->_rt_key != NULL) 331 sockaddr_free(rt->_rt_key); 332 if (rt->rt_gateway != NULL) 333 sockaddr_free(rt->rt_gateway); 334 if (rt_gettag(rt) != NULL) 335 sockaddr_free(rt_gettag(rt)); 336 rt->_rt_key = rt->rt_gateway = rt->rt_tag = NULL; 337 } 338 339 static inline const struct sockaddr * 340 rt_setkey(struct rtentry *rt, const struct sockaddr *key, int flags) 341 { 342 if (rt->_rt_key == key) 343 goto out; 344 345 if (rt->_rt_key != NULL) 346 sockaddr_free(rt->_rt_key); 347 rt->_rt_key = sockaddr_dup(key, flags); 348 out: 349 rt->rt_nodes->rn_key = (const char *)rt->_rt_key; 350 return rt->_rt_key; 351 } 352 353 struct ifaddr * 354 rt_get_ifa(struct rtentry *rt) 355 { 356 struct ifaddr *ifa; 357 358 ifa = rt->rt_ifa; 359 if (ifa->ifa_getifa == NULL) 360 return ifa; 361 #if 0 362 else if (ifa->ifa_seqno != NULL && *ifa->ifa_seqno == rt->rt_ifa_seqno) 363 return ifa; 364 #endif 365 else { 366 ifa = (*ifa->ifa_getifa)(ifa, rt_getkey(rt)); 367 if (ifa == NULL) 368 return NULL; 369 rt_replace_ifa(rt, ifa); 370 return ifa; 371 } 372 } 373 374 static void 375 rt_set_ifa1(struct rtentry *rt, struct ifaddr *ifa) 376 { 377 rt->rt_ifa = ifa; 378 if (ifa->ifa_seqno != NULL) 379 rt->rt_ifa_seqno = *ifa->ifa_seqno; 380 } 381 382 /* 383 * Is this route the connected route for the ifa? 384 */ 385 static int 386 rt_ifa_connected(const struct rtentry *rt, const struct ifaddr *ifa) 387 { 388 const struct sockaddr *key, *dst, *odst; 389 struct sockaddr_storage maskeddst; 390 391 key = rt_getkey(rt); 392 dst = rt->rt_flags & RTF_HOST ? ifa->ifa_dstaddr : ifa->ifa_addr; 393 if (dst == NULL || 394 dst->sa_family != key->sa_family || 395 dst->sa_len != key->sa_len) 396 return 0; 397 if ((rt->rt_flags & RTF_HOST) == 0 && ifa->ifa_netmask) { 398 odst = dst; 399 dst = (struct sockaddr *)&maskeddst; 400 rt_maskedcopy(odst, (struct sockaddr *)&maskeddst, 401 ifa->ifa_netmask); 402 } 403 return (memcmp(dst, key, dst->sa_len) == 0); 404 } 405 406 void 407 rt_replace_ifa(struct rtentry *rt, struct ifaddr *ifa) 408 { 409 struct ifaddr *old; 410 411 if (rt->rt_ifa == ifa) 412 return; 413 414 if (rt->rt_ifa != ifa && 415 rt->rt_ifa->ifa_flags & IFA_ROUTE && 416 rt_ifa_connected(rt, rt->rt_ifa)) 417 { 418 RT_DPRINTF("rt->_rt_key = %p, ifa = %p, " 419 "replace deleted IFA_ROUTE\n", 420 (void *)rt->_rt_key, (void *)rt->rt_ifa); 421 rt->rt_ifa->ifa_flags &= ~IFA_ROUTE; 422 if (rt_ifa_connected(rt, ifa)) { 423 RT_DPRINTF("rt->_rt_key = %p, ifa = %p, " 424 "replace added IFA_ROUTE\n", 425 (void *)rt->_rt_key, (void *)ifa); 426 ifa->ifa_flags |= IFA_ROUTE; 427 } 428 } 429 430 ifaref(ifa); 431 old = rt->rt_ifa; 432 rt_set_ifa1(rt, ifa); 433 ifafree(old); 434 } 435 436 static void 437 rt_set_ifa(struct rtentry *rt, struct ifaddr *ifa) 438 { 439 ifaref(ifa); 440 rt_set_ifa1(rt, ifa); 441 } 442 443 static int 444 route_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 445 void *arg0, void *arg1, void *arg2, void *arg3) 446 { 447 struct rt_msghdr *rtm; 448 int result; 449 450 result = KAUTH_RESULT_DEFER; 451 rtm = arg1; 452 453 if (action != KAUTH_NETWORK_ROUTE) 454 return result; 455 456 if (rtm->rtm_type == RTM_GET) 457 result = KAUTH_RESULT_ALLOW; 458 459 return result; 460 } 461 462 static void rt_free_work(struct work *, void *); 463 464 void 465 rt_init(void) 466 { 467 int error; 468 469 #ifdef RTFLUSH_DEBUG 470 sysctl_net_rtcache_setup(NULL); 471 #endif 472 473 mutex_init(&rt_free_global.lock, MUTEX_DEFAULT, IPL_SOFTNET); 474 SLIST_INIT(&rt_free_global.queue); 475 rt_free_global.enqueued = false; 476 477 rt_psref_class = psref_class_create("rtentry", IPL_SOFTNET); 478 479 error = workqueue_create(&rt_free_global.wq, "rt_free", 480 rt_free_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE); 481 if (error) 482 panic("%s: workqueue_create failed (%d)\n", __func__, error); 483 484 mutex_init(&rt_update_global.lock, MUTEX_DEFAULT, IPL_SOFTNET); 485 cv_init(&rt_update_global.cv, "rt_update"); 486 487 pool_init(&rtentry_pool, sizeof(struct rtentry), 0, 0, 0, "rtentpl", 488 NULL, IPL_SOFTNET); 489 pool_init(&rttimer_pool, sizeof(struct rttimer), 0, 0, 0, "rttmrpl", 490 NULL, IPL_SOFTNET); 491 492 rn_init(); /* initialize all zeroes, all ones, mask table */ 493 rtbl_init(); 494 495 route_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 496 route_listener_cb, NULL); 497 } 498 499 static void 500 rtcache_invalidate(void) 501 { 502 503 RT_ASSERT_WLOCK(); 504 505 if (rtcache_debug()) 506 printf("%s: enter\n", __func__); 507 508 rtcache_generation++; 509 } 510 511 #ifdef RT_DEBUG 512 static void 513 dump_rt(const struct rtentry *rt) 514 { 515 char buf[512]; 516 517 log(LOG_DEBUG, "rt: "); 518 log(LOG_DEBUG, "p=%p ", rt); 519 if (rt->_rt_key == NULL) { 520 log(LOG_DEBUG, "dst=(NULL) "); 521 } else { 522 sockaddr_format(rt->_rt_key, buf, sizeof(buf)); 523 log(LOG_DEBUG, "dst=%s ", buf); 524 } 525 if (rt->rt_gateway == NULL) { 526 log(LOG_DEBUG, "gw=(NULL) "); 527 } else { 528 sockaddr_format(rt->_rt_key, buf, sizeof(buf)); 529 log(LOG_DEBUG, "gw=%s ", buf); 530 } 531 log(LOG_DEBUG, "flags=%x ", rt->rt_flags); 532 if (rt->rt_ifp == NULL) { 533 log(LOG_DEBUG, "if=(NULL) "); 534 } else { 535 log(LOG_DEBUG, "if=%s ", rt->rt_ifp->if_xname); 536 } 537 log(LOG_DEBUG, "\n"); 538 } 539 #endif /* RT_DEBUG */ 540 541 /* 542 * Packet routing routines. If success, refcnt of a returned rtentry 543 * will be incremented. The caller has to rtfree it by itself. 544 */ 545 struct rtentry * 546 rtalloc1_locked(const struct sockaddr *dst, int report, bool wait_ok, 547 bool wlock) 548 { 549 rtbl_t *rtbl; 550 struct rtentry *rt; 551 int s; 552 553 #ifdef NET_MPSAFE 554 retry: 555 #endif 556 s = splsoftnet(); 557 rtbl = rt_gettable(dst->sa_family); 558 if (rtbl == NULL) 559 goto miss; 560 561 rt = rt_matchaddr(rtbl, dst); 562 if (rt == NULL) 563 goto miss; 564 565 if (!ISSET(rt->rt_flags, RTF_UP)) 566 goto miss; 567 568 #ifdef NET_MPSAFE 569 if (ISSET(rt->rt_flags, RTF_UPDATING) && 570 /* XXX updater should be always able to acquire */ 571 curlwp != rt_update_global.lwp) { 572 if (!wait_ok || !rt_wait_ok()) 573 goto miss; 574 RT_UNLOCK(); 575 splx(s); 576 577 /* We can wait until the update is complete */ 578 rt_update_wait(); 579 580 if (wlock) 581 RT_WLOCK(); 582 else 583 RT_RLOCK(); 584 goto retry; 585 } 586 #endif /* NET_MPSAFE */ 587 588 rt_ref(rt); 589 RT_REFCNT_TRACE(rt); 590 591 splx(s); 592 return rt; 593 miss: 594 rtstat.rts_unreach++; 595 if (report) { 596 struct rt_addrinfo info; 597 598 memset(&info, 0, sizeof(info)); 599 info.rti_info[RTAX_DST] = dst; 600 rt_missmsg(RTM_MISS, &info, 0, 0); 601 } 602 splx(s); 603 return NULL; 604 } 605 606 struct rtentry * 607 rtalloc1(const struct sockaddr *dst, int report) 608 { 609 struct rtentry *rt; 610 611 RT_RLOCK(); 612 rt = rtalloc1_locked(dst, report, true, false); 613 RT_UNLOCK(); 614 615 return rt; 616 } 617 618 static void 619 rt_ref(struct rtentry *rt) 620 { 621 622 KASSERTMSG(rt->rt_refcnt >= 0, "rt_refcnt=%d", rt->rt_refcnt); 623 atomic_inc_uint(&rt->rt_refcnt); 624 } 625 626 void 627 rt_unref(struct rtentry *rt) 628 { 629 630 KASSERT(rt != NULL); 631 KASSERTMSG(rt->rt_refcnt > 0, "refcnt=%d", rt->rt_refcnt); 632 633 atomic_dec_uint(&rt->rt_refcnt); 634 if (!ISSET(rt->rt_flags, RTF_UP) || ISSET(rt->rt_flags, RTF_UPDATING)) { 635 mutex_enter(&rt_free_global.lock); 636 cv_broadcast(&rt->rt_cv); 637 mutex_exit(&rt_free_global.lock); 638 } 639 } 640 641 static bool 642 rt_wait_ok(void) 643 { 644 645 KASSERT(!cpu_intr_p()); 646 return !cpu_softintr_p(); 647 } 648 649 void 650 rt_wait_refcnt(const char *title, struct rtentry *rt, int cnt) 651 { 652 mutex_enter(&rt_free_global.lock); 653 while (rt->rt_refcnt > cnt) { 654 dlog(LOG_DEBUG, "%s: %s waiting (refcnt=%d)\n", 655 __func__, title, rt->rt_refcnt); 656 cv_wait(&rt->rt_cv, &rt_free_global.lock); 657 dlog(LOG_DEBUG, "%s: %s waited (refcnt=%d)\n", 658 __func__, title, rt->rt_refcnt); 659 } 660 mutex_exit(&rt_free_global.lock); 661 } 662 663 void 664 rt_wait_psref(struct rtentry *rt) 665 { 666 667 psref_target_destroy(&rt->rt_psref, rt_psref_class); 668 psref_target_init(&rt->rt_psref, rt_psref_class); 669 } 670 671 static void 672 _rt_free(struct rtentry *rt) 673 { 674 struct ifaddr *ifa; 675 676 /* 677 * Need to avoid a deadlock on rt_wait_refcnt of update 678 * and a conflict on psref_target_destroy of update. 679 */ 680 #ifdef NET_MPSAFE 681 rt_update_wait(); 682 #endif 683 684 RT_REFCNT_TRACE(rt); 685 KASSERTMSG(rt->rt_refcnt >= 0, "refcnt=%d", rt->rt_refcnt); 686 rt_wait_refcnt("free", rt, 0); 687 #ifdef NET_MPSAFE 688 psref_target_destroy(&rt->rt_psref, rt_psref_class); 689 #endif 690 691 rt_assert_inactive(rt); 692 rttrash--; 693 ifa = rt->rt_ifa; 694 rt->rt_ifa = NULL; 695 ifafree(ifa); 696 rt->rt_ifp = NULL; 697 cv_destroy(&rt->rt_cv); 698 rt_destroy(rt); 699 pool_put(&rtentry_pool, rt); 700 } 701 702 static void 703 rt_free_work(struct work *wk, void *arg) 704 { 705 706 for (;;) { 707 struct rtentry *rt; 708 709 mutex_enter(&rt_free_global.lock); 710 if ((rt = SLIST_FIRST(&rt_free_global.queue)) == NULL) { 711 rt_free_global.enqueued = false; 712 mutex_exit(&rt_free_global.lock); 713 return; 714 } 715 SLIST_REMOVE_HEAD(&rt_free_global.queue, rt_free); 716 mutex_exit(&rt_free_global.lock); 717 atomic_dec_uint(&rt->rt_refcnt); 718 _rt_free(rt); 719 } 720 } 721 722 void 723 rt_free(struct rtentry *rt) 724 { 725 726 KASSERTMSG(rt->rt_refcnt > 0, "rt_refcnt=%d", rt->rt_refcnt); 727 if (rt_wait_ok()) { 728 atomic_dec_uint(&rt->rt_refcnt); 729 _rt_free(rt); 730 return; 731 } 732 733 mutex_enter(&rt_free_global.lock); 734 /* No need to add a reference here. */ 735 SLIST_INSERT_HEAD(&rt_free_global.queue, rt, rt_free); 736 if (!rt_free_global.enqueued) { 737 workqueue_enqueue(rt_free_global.wq, &rt_free_global.wk, NULL); 738 rt_free_global.enqueued = true; 739 } 740 mutex_exit(&rt_free_global.lock); 741 } 742 743 #ifdef NET_MPSAFE 744 static void 745 rt_update_wait(void) 746 { 747 748 mutex_enter(&rt_update_global.lock); 749 while (rt_update_global.ongoing) { 750 dlog(LOG_DEBUG, "%s: waiting lwp=%p\n", __func__, curlwp); 751 cv_wait(&rt_update_global.cv, &rt_update_global.lock); 752 dlog(LOG_DEBUG, "%s: waited lwp=%p\n", __func__, curlwp); 753 } 754 mutex_exit(&rt_update_global.lock); 755 } 756 #endif 757 758 int 759 rt_update_prepare(struct rtentry *rt) 760 { 761 762 dlog(LOG_DEBUG, "%s: updating rt=%p lwp=%p\n", __func__, rt, curlwp); 763 764 RT_WLOCK(); 765 /* If the entry is being destroyed, don't proceed the update. */ 766 if (!ISSET(rt->rt_flags, RTF_UP)) { 767 RT_UNLOCK(); 768 return ESRCH; 769 } 770 rt->rt_flags |= RTF_UPDATING; 771 RT_UNLOCK(); 772 773 mutex_enter(&rt_update_global.lock); 774 while (rt_update_global.ongoing) { 775 dlog(LOG_DEBUG, "%s: waiting ongoing updating rt=%p lwp=%p\n", 776 __func__, rt, curlwp); 777 cv_wait(&rt_update_global.cv, &rt_update_global.lock); 778 dlog(LOG_DEBUG, "%s: waited ongoing updating rt=%p lwp=%p\n", 779 __func__, rt, curlwp); 780 } 781 rt_update_global.ongoing = true; 782 /* XXX need it to avoid rt_update_wait by updater itself. */ 783 rt_update_global.lwp = curlwp; 784 mutex_exit(&rt_update_global.lock); 785 786 rt_wait_refcnt("update", rt, 1); 787 rt_wait_psref(rt); 788 789 return 0; 790 } 791 792 void 793 rt_update_finish(struct rtentry *rt) 794 { 795 796 RT_WLOCK(); 797 rt->rt_flags &= ~RTF_UPDATING; 798 RT_UNLOCK(); 799 800 mutex_enter(&rt_update_global.lock); 801 rt_update_global.ongoing = false; 802 rt_update_global.lwp = NULL; 803 cv_broadcast(&rt_update_global.cv); 804 mutex_exit(&rt_update_global.lock); 805 806 dlog(LOG_DEBUG, "%s: updated rt=%p lwp=%p\n", __func__, rt, curlwp); 807 } 808 809 /* 810 * Force a routing table entry to the specified 811 * destination to go through the given gateway. 812 * Normally called as a result of a routing redirect 813 * message from the network layer. 814 * 815 * N.B.: must be called at splsoftnet 816 */ 817 void 818 rtredirect(const struct sockaddr *dst, const struct sockaddr *gateway, 819 const struct sockaddr *netmask, int flags, const struct sockaddr *src, 820 struct rtentry **rtp) 821 { 822 struct rtentry *rt; 823 int error = 0; 824 uint64_t *stat = NULL; 825 struct rt_addrinfo info; 826 struct ifaddr *ifa; 827 struct psref psref; 828 829 /* verify the gateway is directly reachable */ 830 if ((ifa = ifa_ifwithnet_psref(gateway, &psref)) == NULL) { 831 error = ENETUNREACH; 832 goto out; 833 } 834 rt = rtalloc1(dst, 0); 835 /* 836 * If the redirect isn't from our current router for this dst, 837 * it's either old or wrong. If it redirects us to ourselves, 838 * we have a routing loop, perhaps as a result of an interface 839 * going down recently. 840 */ 841 if (!(flags & RTF_DONE) && rt && 842 (sockaddr_cmp(src, rt->rt_gateway) != 0 || rt->rt_ifa != ifa)) 843 error = EINVAL; 844 else { 845 int s = pserialize_read_enter(); 846 struct ifaddr *_ifa; 847 848 _ifa = ifa_ifwithaddr(gateway); 849 if (_ifa != NULL) 850 error = EHOSTUNREACH; 851 pserialize_read_exit(s); 852 } 853 if (error) 854 goto done; 855 /* 856 * Create a new entry if we just got back a wildcard entry 857 * or the lookup failed. This is necessary for hosts 858 * which use routing redirects generated by smart gateways 859 * to dynamically build the routing tables. 860 */ 861 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2)) 862 goto create; 863 /* 864 * Don't listen to the redirect if it's 865 * for a route to an interface. 866 */ 867 if (rt->rt_flags & RTF_GATEWAY) { 868 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) { 869 /* 870 * Changing from route to net => route to host. 871 * Create new route, rather than smashing route to net. 872 */ 873 create: 874 if (rt != NULL) 875 rt_unref(rt); 876 flags |= RTF_GATEWAY | RTF_DYNAMIC; 877 memset(&info, 0, sizeof(info)); 878 info.rti_info[RTAX_DST] = dst; 879 info.rti_info[RTAX_GATEWAY] = gateway; 880 info.rti_info[RTAX_NETMASK] = netmask; 881 info.rti_ifa = ifa; 882 info.rti_flags = flags; 883 rt = NULL; 884 error = rtrequest1(RTM_ADD, &info, &rt); 885 if (rt != NULL) 886 flags = rt->rt_flags; 887 if (error == 0) 888 rt_newmsg_dynamic(RTM_ADD, rt); 889 stat = &rtstat.rts_dynamic; 890 } else { 891 /* 892 * Smash the current notion of the gateway to 893 * this destination. Should check about netmask!!! 894 */ 895 #ifdef NET_MPSAFE 896 KASSERT(!cpu_softintr_p()); 897 898 error = rt_update_prepare(rt); 899 if (error == 0) { 900 #endif 901 RT_WLOCK(); 902 error = rt_setgate(rt, gateway); 903 if (error == 0) { 904 rt->rt_flags |= RTF_MODIFIED; 905 flags |= RTF_MODIFIED; 906 } 907 RT_UNLOCK(); 908 #ifdef NET_MPSAFE 909 rt_update_finish(rt); 910 } else { 911 /* 912 * If error != 0, the rtentry is being 913 * destroyed, so doing nothing doesn't 914 * matter. 915 */ 916 } 917 #endif 918 stat = &rtstat.rts_newgateway; 919 } 920 } else 921 error = EHOSTUNREACH; 922 done: 923 if (rt) { 924 if (rtp != NULL && !error) 925 *rtp = rt; 926 else 927 rt_unref(rt); 928 } 929 out: 930 if (error) 931 rtstat.rts_badredirect++; 932 else if (stat != NULL) 933 (*stat)++; 934 memset(&info, 0, sizeof(info)); 935 info.rti_info[RTAX_DST] = dst; 936 info.rti_info[RTAX_GATEWAY] = gateway; 937 info.rti_info[RTAX_NETMASK] = netmask; 938 info.rti_info[RTAX_AUTHOR] = src; 939 rt_missmsg(RTM_REDIRECT, &info, flags, error); 940 ifa_release(ifa, &psref); 941 } 942 943 /* 944 * Delete a route and generate a message. 945 * It doesn't free a passed rt. 946 */ 947 static int 948 rtdeletemsg(struct rtentry *rt) 949 { 950 int error; 951 struct rt_addrinfo info; 952 struct rtentry *retrt; 953 954 /* 955 * Request the new route so that the entry is not actually 956 * deleted. That will allow the information being reported to 957 * be accurate (and consistent with route_output()). 958 */ 959 memset(&info, 0, sizeof(info)); 960 info.rti_info[RTAX_DST] = rt_getkey(rt); 961 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 962 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 963 info.rti_flags = rt->rt_flags; 964 error = rtrequest1(RTM_DELETE, &info, &retrt); 965 966 rt_missmsg(RTM_DELETE, &info, info.rti_flags, error); 967 968 return error; 969 } 970 971 static struct ifaddr * 972 ifa_ifwithroute_psref(int flags, const struct sockaddr *dst, 973 const struct sockaddr *gateway, struct psref *psref) 974 { 975 struct ifaddr *ifa = NULL; 976 977 if ((flags & RTF_GATEWAY) == 0) { 978 /* 979 * If we are adding a route to an interface, 980 * and the interface is a pt to pt link 981 * we should search for the destination 982 * as our clue to the interface. Otherwise 983 * we can use the local address. 984 */ 985 if ((flags & RTF_HOST) && gateway->sa_family != AF_LINK) 986 ifa = ifa_ifwithdstaddr_psref(dst, psref); 987 if (ifa == NULL) 988 ifa = ifa_ifwithaddr_psref(gateway, psref); 989 } else { 990 /* 991 * If we are adding a route to a remote net 992 * or host, the gateway may still be on the 993 * other end of a pt to pt link. 994 */ 995 ifa = ifa_ifwithdstaddr_psref(gateway, psref); 996 } 997 if (ifa == NULL) 998 ifa = ifa_ifwithnet_psref(gateway, psref); 999 if (ifa == NULL) { 1000 int s; 1001 struct rtentry *rt; 1002 1003 rt = rtalloc1_locked(gateway, 0, true, true); 1004 if (rt == NULL) 1005 return NULL; 1006 if (rt->rt_flags & RTF_GATEWAY) { 1007 rt_unref(rt); 1008 return NULL; 1009 } 1010 /* 1011 * Just in case. May not need to do this workaround. 1012 * Revisit when working on rtentry MP-ification. 1013 */ 1014 s = pserialize_read_enter(); 1015 IFADDR_READER_FOREACH(ifa, rt->rt_ifp) { 1016 if (ifa == rt->rt_ifa) 1017 break; 1018 } 1019 if (ifa != NULL) 1020 ifa_acquire(ifa, psref); 1021 pserialize_read_exit(s); 1022 rt_unref(rt); 1023 if (ifa == NULL) 1024 return NULL; 1025 } 1026 if (ifa->ifa_addr->sa_family != dst->sa_family) { 1027 struct ifaddr *nifa; 1028 int s; 1029 1030 s = pserialize_read_enter(); 1031 nifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 1032 if (nifa != NULL) { 1033 ifa_release(ifa, psref); 1034 ifa_acquire(nifa, psref); 1035 ifa = nifa; 1036 } 1037 pserialize_read_exit(s); 1038 } 1039 return ifa; 1040 } 1041 1042 /* 1043 * If it suceeds and ret_nrt isn't NULL, refcnt of ret_nrt is incremented. 1044 * The caller has to rtfree it by itself. 1045 */ 1046 int 1047 rtrequest(int req, const struct sockaddr *dst, const struct sockaddr *gateway, 1048 const struct sockaddr *netmask, int flags, struct rtentry **ret_nrt) 1049 { 1050 struct rt_addrinfo info; 1051 1052 memset(&info, 0, sizeof(info)); 1053 info.rti_flags = flags; 1054 info.rti_info[RTAX_DST] = dst; 1055 info.rti_info[RTAX_GATEWAY] = gateway; 1056 info.rti_info[RTAX_NETMASK] = netmask; 1057 return rtrequest1(req, &info, ret_nrt); 1058 } 1059 1060 static struct ifnet * 1061 rt_getifp(struct rt_addrinfo *info, struct psref *psref) 1062 { 1063 const struct sockaddr *ifpaddr = info->rti_info[RTAX_IFP]; 1064 1065 if (info->rti_ifp != NULL) 1066 return NULL; 1067 /* 1068 * ifp may be specified by sockaddr_dl when protocol address 1069 * is ambiguous 1070 */ 1071 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 1072 struct ifaddr *ifa; 1073 int s = pserialize_read_enter(); 1074 1075 ifa = ifa_ifwithnet(ifpaddr); 1076 if (ifa != NULL) 1077 info->rti_ifp = if_get_byindex(ifa->ifa_ifp->if_index, 1078 psref); 1079 pserialize_read_exit(s); 1080 } 1081 1082 return info->rti_ifp; 1083 } 1084 1085 static struct ifaddr * 1086 rt_getifa(struct rt_addrinfo *info, struct psref *psref) 1087 { 1088 struct ifaddr *ifa = NULL; 1089 const struct sockaddr *dst = info->rti_info[RTAX_DST]; 1090 const struct sockaddr *gateway = info->rti_info[RTAX_GATEWAY]; 1091 const struct sockaddr *ifaaddr = info->rti_info[RTAX_IFA]; 1092 int flags = info->rti_flags; 1093 const struct sockaddr *sa; 1094 1095 if (info->rti_ifa == NULL && ifaaddr != NULL) { 1096 ifa = ifa_ifwithaddr_psref(ifaaddr, psref); 1097 if (ifa != NULL) 1098 goto got; 1099 } 1100 1101 sa = ifaaddr != NULL ? ifaaddr : 1102 (gateway != NULL ? gateway : dst); 1103 if (sa != NULL && info->rti_ifp != NULL) 1104 ifa = ifaof_ifpforaddr_psref(sa, info->rti_ifp, psref); 1105 else if (dst != NULL && gateway != NULL) 1106 ifa = ifa_ifwithroute_psref(flags, dst, gateway, psref); 1107 else if (sa != NULL) 1108 ifa = ifa_ifwithroute_psref(flags, sa, sa, psref); 1109 if (ifa == NULL) 1110 return NULL; 1111 got: 1112 if (ifa->ifa_getifa != NULL) { 1113 /* FIXME ifa_getifa is NOMPSAFE */ 1114 ifa = (*ifa->ifa_getifa)(ifa, dst); 1115 if (ifa == NULL) 1116 return NULL; 1117 ifa_acquire(ifa, psref); 1118 } 1119 info->rti_ifa = ifa; 1120 if (info->rti_ifp == NULL) 1121 info->rti_ifp = ifa->ifa_ifp; 1122 return ifa; 1123 } 1124 1125 /* 1126 * If it suceeds and ret_nrt isn't NULL, refcnt of ret_nrt is incremented. 1127 * The caller has to rtfree it by itself. 1128 */ 1129 int 1130 rtrequest1(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt) 1131 { 1132 int s = splsoftnet(), ss; 1133 int error = 0, rc; 1134 struct rtentry *rt; 1135 rtbl_t *rtbl; 1136 struct ifaddr *ifa = NULL; 1137 struct sockaddr_storage maskeddst; 1138 const struct sockaddr *dst = info->rti_info[RTAX_DST]; 1139 const struct sockaddr *gateway = info->rti_info[RTAX_GATEWAY]; 1140 const struct sockaddr *netmask = info->rti_info[RTAX_NETMASK]; 1141 int flags = info->rti_flags; 1142 struct psref psref_ifp, psref_ifa; 1143 int bound = 0; 1144 struct ifnet *ifp = NULL; 1145 bool need_to_release_ifa = true; 1146 bool need_unlock = true; 1147 #define senderr(x) { error = x ; goto bad; } 1148 1149 RT_WLOCK(); 1150 1151 bound = curlwp_bind(); 1152 if ((rtbl = rt_gettable(dst->sa_family)) == NULL) 1153 senderr(ESRCH); 1154 if (flags & RTF_HOST) 1155 netmask = NULL; 1156 switch (req) { 1157 case RTM_DELETE: 1158 if (netmask) { 1159 rt_maskedcopy(dst, (struct sockaddr *)&maskeddst, 1160 netmask); 1161 dst = (struct sockaddr *)&maskeddst; 1162 } 1163 if ((rt = rt_lookup(rtbl, dst, netmask)) == NULL) 1164 senderr(ESRCH); 1165 if ((rt = rt_deladdr(rtbl, dst, netmask)) == NULL) 1166 senderr(ESRCH); 1167 rt->rt_flags &= ~RTF_UP; 1168 ifa = rt->rt_ifa; 1169 if (ifa->ifa_flags & IFA_ROUTE && 1170 rt_ifa_connected(rt, ifa)) { 1171 RT_DPRINTF("rt->_rt_key = %p, ifa = %p, " 1172 "deleted IFA_ROUTE\n", 1173 (void *)rt->_rt_key, (void *)ifa); 1174 ifa->ifa_flags &= ~IFA_ROUTE; 1175 } 1176 if (ifa->ifa_rtrequest) 1177 ifa->ifa_rtrequest(RTM_DELETE, rt, info); 1178 ifa = NULL; 1179 rttrash++; 1180 if (ret_nrt) { 1181 *ret_nrt = rt; 1182 rt_ref(rt); 1183 RT_REFCNT_TRACE(rt); 1184 } 1185 rtcache_invalidate(); 1186 RT_UNLOCK(); 1187 need_unlock = false; 1188 rt_timer_remove_all(rt); 1189 #if defined(INET) || defined(INET6) 1190 if (netmask != NULL) 1191 lltable_prefix_free(dst->sa_family, dst, netmask, 0); 1192 #endif 1193 if (ret_nrt == NULL) { 1194 /* Adjust the refcount */ 1195 rt_ref(rt); 1196 RT_REFCNT_TRACE(rt); 1197 rt_free(rt); 1198 } 1199 break; 1200 1201 case RTM_ADD: 1202 if (info->rti_ifa == NULL) { 1203 ifp = rt_getifp(info, &psref_ifp); 1204 ifa = rt_getifa(info, &psref_ifa); 1205 if (ifa == NULL) 1206 senderr(ENETUNREACH); 1207 } else { 1208 /* Caller should have a reference of ifa */ 1209 ifa = info->rti_ifa; 1210 need_to_release_ifa = false; 1211 } 1212 rt = pool_get(&rtentry_pool, PR_NOWAIT); 1213 if (rt == NULL) 1214 senderr(ENOBUFS); 1215 memset(rt, 0, sizeof(*rt)); 1216 rt->rt_flags = RTF_UP | (flags & ~RTF_DONTCHANGEIFA); 1217 LIST_INIT(&rt->rt_timer); 1218 1219 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1220 if (netmask) { 1221 rt_maskedcopy(dst, (struct sockaddr *)&maskeddst, 1222 netmask); 1223 rt_setkey(rt, (struct sockaddr *)&maskeddst, M_NOWAIT); 1224 } else { 1225 rt_setkey(rt, dst, M_NOWAIT); 1226 } 1227 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1228 if (rt_getkey(rt) == NULL || 1229 rt_setgate(rt, gateway) != 0) { 1230 pool_put(&rtentry_pool, rt); 1231 senderr(ENOBUFS); 1232 } 1233 1234 rt_set_ifa(rt, ifa); 1235 if (info->rti_info[RTAX_TAG] != NULL) { 1236 const struct sockaddr *tag; 1237 tag = rt_settag(rt, info->rti_info[RTAX_TAG]); 1238 if (tag == NULL) 1239 senderr(ENOBUFS); 1240 } 1241 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1242 1243 ss = pserialize_read_enter(); 1244 if (info->rti_info[RTAX_IFP] != NULL) { 1245 struct ifaddr *ifa2; 1246 ifa2 = ifa_ifwithnet(info->rti_info[RTAX_IFP]); 1247 if (ifa2 != NULL) 1248 rt->rt_ifp = ifa2->ifa_ifp; 1249 else 1250 rt->rt_ifp = ifa->ifa_ifp; 1251 } else 1252 rt->rt_ifp = ifa->ifa_ifp; 1253 pserialize_read_exit(ss); 1254 cv_init(&rt->rt_cv, "rtentry"); 1255 psref_target_init(&rt->rt_psref, rt_psref_class); 1256 1257 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1258 rc = rt_addaddr(rtbl, rt, netmask); 1259 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1260 if (rc != 0) { 1261 ifafree(ifa); /* for rt_set_ifa above */ 1262 cv_destroy(&rt->rt_cv); 1263 rt_destroy(rt); 1264 pool_put(&rtentry_pool, rt); 1265 senderr(rc); 1266 } 1267 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1268 if (ifa->ifa_rtrequest) 1269 ifa->ifa_rtrequest(req, rt, info); 1270 if (need_to_release_ifa) 1271 ifa_release(ifa, &psref_ifa); 1272 ifa = NULL; 1273 if_put(ifp, &psref_ifp); 1274 ifp = NULL; 1275 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1276 if (ret_nrt) { 1277 *ret_nrt = rt; 1278 rt_ref(rt); 1279 RT_REFCNT_TRACE(rt); 1280 } 1281 rtcache_invalidate(); 1282 RT_UNLOCK(); 1283 need_unlock = false; 1284 break; 1285 case RTM_GET: 1286 if (netmask != NULL) { 1287 rt_maskedcopy(dst, (struct sockaddr *)&maskeddst, 1288 netmask); 1289 dst = (struct sockaddr *)&maskeddst; 1290 } 1291 if ((rt = rt_lookup(rtbl, dst, netmask)) == NULL) 1292 senderr(ESRCH); 1293 if (ret_nrt != NULL) { 1294 *ret_nrt = rt; 1295 rt_ref(rt); 1296 RT_REFCNT_TRACE(rt); 1297 } 1298 break; 1299 } 1300 bad: 1301 if (need_to_release_ifa) 1302 ifa_release(ifa, &psref_ifa); 1303 if_put(ifp, &psref_ifp); 1304 curlwp_bindx(bound); 1305 if (need_unlock) 1306 RT_UNLOCK(); 1307 splx(s); 1308 return error; 1309 } 1310 1311 int 1312 rt_setgate(struct rtentry *rt, const struct sockaddr *gate) 1313 { 1314 struct sockaddr *new, *old; 1315 1316 KASSERT(RT_WLOCKED()); 1317 KASSERT(rt->_rt_key != NULL); 1318 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1319 1320 new = sockaddr_dup(gate, M_ZERO | M_NOWAIT); 1321 if (new == NULL) 1322 return ENOMEM; 1323 1324 old = rt->rt_gateway; 1325 rt->rt_gateway = new; 1326 if (old != NULL) 1327 sockaddr_free(old); 1328 1329 KASSERT(rt->_rt_key != NULL); 1330 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1331 1332 if (rt->rt_flags & RTF_GATEWAY) { 1333 struct rtentry *gwrt; 1334 1335 gwrt = rtalloc1_locked(gate, 1, false, true); 1336 /* 1337 * If we switched gateways, grab the MTU from the new 1338 * gateway route if the current MTU, if the current MTU is 1339 * greater than the MTU of gateway. 1340 * Note that, if the MTU of gateway is 0, we will reset the 1341 * MTU of the route to run PMTUD again from scratch. XXX 1342 */ 1343 if (gwrt != NULL) { 1344 KASSERT(gwrt->_rt_key != NULL); 1345 RT_DPRINTF("gwrt->_rt_key = %p\n", gwrt->_rt_key); 1346 if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 && 1347 rt->rt_rmx.rmx_mtu && 1348 rt->rt_rmx.rmx_mtu > gwrt->rt_rmx.rmx_mtu) { 1349 rt->rt_rmx.rmx_mtu = gwrt->rt_rmx.rmx_mtu; 1350 } 1351 rt_unref(gwrt); 1352 } 1353 } 1354 KASSERT(rt->_rt_key != NULL); 1355 RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key); 1356 return 0; 1357 } 1358 1359 static struct ifaddr * 1360 rt_update_get_ifa(const struct rt_addrinfo *info, const struct rtentry *rt, 1361 struct ifnet **ifp, struct psref *psref_ifp, struct psref *psref) 1362 { 1363 struct ifaddr *ifa = NULL; 1364 1365 *ifp = NULL; 1366 if (info->rti_info[RTAX_IFP] != NULL) { 1367 ifa = ifa_ifwithnet_psref(info->rti_info[RTAX_IFP], psref); 1368 if (ifa == NULL) 1369 goto next; 1370 *ifp = ifa->ifa_ifp; 1371 if_acquire(*ifp, psref_ifp); 1372 if (info->rti_info[RTAX_IFA] == NULL && 1373 info->rti_info[RTAX_GATEWAY] == NULL) 1374 goto out; 1375 ifa_release(ifa, psref); 1376 if (info->rti_info[RTAX_IFA] == NULL) { 1377 /* route change <dst> <gw> -ifp <if> */ 1378 ifa = ifaof_ifpforaddr_psref( 1379 info->rti_info[RTAX_GATEWAY], *ifp, psref); 1380 } else { 1381 /* route change <dst> -ifp <if> -ifa <addr> */ 1382 ifa = ifa_ifwithaddr_psref(info->rti_info[RTAX_IFA], 1383 psref); 1384 if (ifa != NULL) 1385 goto out; 1386 ifa = ifaof_ifpforaddr_psref(info->rti_info[RTAX_IFA], 1387 *ifp, psref); 1388 } 1389 goto out; 1390 } 1391 next: 1392 if (info->rti_info[RTAX_IFA] != NULL) { 1393 /* route change <dst> <gw> -ifa <addr> */ 1394 ifa = ifa_ifwithaddr_psref(info->rti_info[RTAX_IFA], psref); 1395 if (ifa != NULL) 1396 goto out; 1397 } 1398 if (info->rti_info[RTAX_GATEWAY] != NULL) { 1399 /* route change <dst> <gw> */ 1400 ifa = ifa_ifwithroute_psref(rt->rt_flags, rt_getkey(rt), 1401 info->rti_info[RTAX_GATEWAY], psref); 1402 } 1403 out: 1404 if (ifa != NULL && *ifp == NULL) { 1405 *ifp = ifa->ifa_ifp; 1406 if_acquire(*ifp, psref_ifp); 1407 } 1408 if (ifa == NULL && *ifp != NULL) { 1409 if_put(*ifp, psref_ifp); 1410 *ifp = NULL; 1411 } 1412 return ifa; 1413 } 1414 1415 int 1416 rt_update(struct rtentry *rt, struct rt_addrinfo *info, void *rtm) 1417 { 1418 int error = 0; 1419 struct ifnet *ifp = NULL, *new_ifp = NULL; 1420 struct ifaddr *ifa = NULL, *new_ifa; 1421 struct psref psref_ifa, psref_new_ifa, psref_ifp, psref_new_ifp; 1422 bool newgw, ifp_changed = false; 1423 1424 RT_WLOCK(); 1425 /* 1426 * New gateway could require new ifaddr, ifp; 1427 * flags may also be different; ifp may be specified 1428 * by ll sockaddr when protocol address is ambiguous 1429 */ 1430 newgw = info->rti_info[RTAX_GATEWAY] != NULL && 1431 sockaddr_cmp(info->rti_info[RTAX_GATEWAY], rt->rt_gateway) != 0; 1432 1433 if (newgw || info->rti_info[RTAX_IFP] != NULL || 1434 info->rti_info[RTAX_IFA] != NULL) { 1435 ifp = rt_getifp(info, &psref_ifp); 1436 /* info refers ifp so we need to keep a reference */ 1437 ifa = rt_getifa(info, &psref_ifa); 1438 if (ifa == NULL) { 1439 error = ENETUNREACH; 1440 goto out; 1441 } 1442 } 1443 if (newgw) { 1444 error = rt_setgate(rt, info->rti_info[RTAX_GATEWAY]); 1445 if (error != 0) 1446 goto out; 1447 } 1448 if (info->rti_info[RTAX_TAG]) { 1449 const struct sockaddr *tag; 1450 tag = rt_settag(rt, info->rti_info[RTAX_TAG]); 1451 if (tag == NULL) { 1452 error = ENOBUFS; 1453 goto out; 1454 } 1455 } 1456 /* 1457 * New gateway could require new ifaddr, ifp; 1458 * flags may also be different; ifp may be specified 1459 * by ll sockaddr when protocol address is ambiguous 1460 */ 1461 new_ifa = rt_update_get_ifa(info, rt, &new_ifp, &psref_new_ifp, 1462 &psref_new_ifa); 1463 if (new_ifa != NULL) { 1464 ifa_release(ifa, &psref_ifa); 1465 ifa = new_ifa; 1466 } 1467 if (ifa) { 1468 struct ifaddr *oifa = rt->rt_ifa; 1469 if (oifa != ifa && !ifa_is_destroying(ifa) && 1470 new_ifp != NULL && !if_is_deactivated(new_ifp)) { 1471 if (oifa && oifa->ifa_rtrequest) 1472 oifa->ifa_rtrequest(RTM_DELETE, rt, info); 1473 rt_replace_ifa(rt, ifa); 1474 rt->rt_ifp = new_ifp; 1475 ifp_changed = true; 1476 } 1477 if (new_ifa == NULL) 1478 ifa_release(ifa, &psref_ifa); 1479 /* To avoid ifa_release below */ 1480 ifa = NULL; 1481 } 1482 ifa_release(new_ifa, &psref_new_ifa); 1483 if (new_ifp && rt->rt_ifp != new_ifp && !if_is_deactivated(new_ifp)) { 1484 rt->rt_ifp = new_ifp; 1485 ifp_changed = true; 1486 } 1487 rt_setmetrics(rtm, rt); 1488 if (rt->rt_flags != info->rti_flags) { 1489 rt->rt_flags = (info->rti_flags & ~PRESERVED_RTF) | 1490 (rt->rt_flags & PRESERVED_RTF); 1491 } 1492 if (rt->rt_ifa->ifa_rtrequest) 1493 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info); 1494 #if defined(INET) || defined(INET6) 1495 if (ifp_changed && rt_mask(rt) != NULL) 1496 lltable_prefix_free(rt_getkey(rt)->sa_family, rt_getkey(rt), 1497 rt_mask(rt), 0); 1498 #else 1499 (void)ifp_changed; /* XXX gcc */ 1500 #endif 1501 out: 1502 ifa_release(ifa, &psref_ifa); 1503 if_put(new_ifp, &psref_new_ifp); 1504 if_put(ifp, &psref_ifp); 1505 1506 RT_UNLOCK(); 1507 1508 return error; 1509 } 1510 1511 static void 1512 rt_maskedcopy(const struct sockaddr *src, struct sockaddr *dst, 1513 const struct sockaddr *netmask) 1514 { 1515 const char *netmaskp = &netmask->sa_data[0], 1516 *srcp = &src->sa_data[0]; 1517 char *dstp = &dst->sa_data[0]; 1518 const char *maskend = (char *)dst + MIN(netmask->sa_len, src->sa_len); 1519 const char *srcend = (char *)dst + src->sa_len; 1520 1521 dst->sa_len = src->sa_len; 1522 dst->sa_family = src->sa_family; 1523 1524 while (dstp < maskend) 1525 *dstp++ = *srcp++ & *netmaskp++; 1526 if (dstp < srcend) 1527 memset(dstp, 0, (size_t)(srcend - dstp)); 1528 } 1529 1530 /* 1531 * Inform the routing socket of a route change. 1532 */ 1533 void 1534 rt_newmsg(const int cmd, const struct rtentry *rt) 1535 { 1536 struct rt_addrinfo info; 1537 1538 memset((void *)&info, 0, sizeof(info)); 1539 info.rti_info[RTAX_DST] = rt_getkey(rt); 1540 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1541 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 1542 if (rt->rt_ifp) { 1543 info.rti_info[RTAX_IFP] = rt->rt_ifp->if_dl->ifa_addr; 1544 info.rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr; 1545 } 1546 1547 rt_missmsg(cmd, &info, rt->rt_flags, 0); 1548 } 1549 1550 /* 1551 * Inform the routing socket of a route change for RTF_DYNAMIC. 1552 */ 1553 void 1554 rt_newmsg_dynamic(const int cmd, const struct rtentry *rt) 1555 { 1556 struct rt_addrinfo info; 1557 struct sockaddr *gateway = rt->rt_gateway; 1558 1559 if (gateway == NULL) 1560 return; 1561 1562 switch(gateway->sa_family) { 1563 #ifdef INET 1564 case AF_INET: { 1565 extern bool icmp_dynamic_rt_msg; 1566 if (!icmp_dynamic_rt_msg) 1567 return; 1568 break; 1569 } 1570 #endif 1571 #ifdef INET6 1572 case AF_INET6: { 1573 extern bool icmp6_dynamic_rt_msg; 1574 if (!icmp6_dynamic_rt_msg) 1575 return; 1576 break; 1577 } 1578 #endif 1579 default: 1580 return; 1581 } 1582 1583 memset((void *)&info, 0, sizeof(info)); 1584 info.rti_info[RTAX_DST] = rt_getkey(rt); 1585 info.rti_info[RTAX_GATEWAY] = gateway; 1586 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 1587 if (rt->rt_ifp) { 1588 info.rti_info[RTAX_IFP] = rt->rt_ifp->if_dl->ifa_addr; 1589 info.rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr; 1590 } 1591 1592 rt_missmsg(cmd, &info, rt->rt_flags, 0); 1593 } 1594 1595 /* 1596 * Set up or tear down a routing table entry, normally 1597 * for an interface. 1598 */ 1599 int 1600 rtinit(struct ifaddr *ifa, int cmd, int flags) 1601 { 1602 struct rtentry *rt; 1603 struct sockaddr *dst, *odst; 1604 struct sockaddr_storage maskeddst; 1605 struct rtentry *nrt = NULL; 1606 int error; 1607 struct rt_addrinfo info; 1608 1609 dst = flags & RTF_HOST ? ifa->ifa_dstaddr : ifa->ifa_addr; 1610 if (cmd == RTM_DELETE) { 1611 if ((flags & RTF_HOST) == 0 && ifa->ifa_netmask) { 1612 /* Delete subnet route for this interface */ 1613 odst = dst; 1614 dst = (struct sockaddr *)&maskeddst; 1615 rt_maskedcopy(odst, dst, ifa->ifa_netmask); 1616 } 1617 if ((rt = rtalloc1(dst, 0)) != NULL) { 1618 if (rt->rt_ifa != ifa) { 1619 rt_unref(rt); 1620 return (flags & RTF_HOST) ? EHOSTUNREACH 1621 : ENETUNREACH; 1622 } 1623 rt_unref(rt); 1624 } 1625 } 1626 memset(&info, 0, sizeof(info)); 1627 info.rti_ifa = ifa; 1628 info.rti_flags = flags | ifa->ifa_flags | RTF_DONTCHANGEIFA; 1629 info.rti_info[RTAX_DST] = dst; 1630 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1631 1632 /* 1633 * XXX here, it seems that we are assuming that ifa_netmask is NULL 1634 * for RTF_HOST. bsdi4 passes NULL explicitly (via intermediate 1635 * variable) when RTF_HOST is 1. still not sure if i can safely 1636 * change it to meet bsdi4 behavior. 1637 */ 1638 if (cmd != RTM_LLINFO_UPD) 1639 info.rti_info[RTAX_NETMASK] = ifa->ifa_netmask; 1640 error = rtrequest1((cmd == RTM_LLINFO_UPD) ? RTM_GET : cmd, &info, 1641 &nrt); 1642 if (error != 0) 1643 return error; 1644 1645 rt = nrt; 1646 RT_REFCNT_TRACE(rt); 1647 switch (cmd) { 1648 case RTM_DELETE: 1649 rt_newmsg(cmd, rt); 1650 rt_free(rt); 1651 break; 1652 case RTM_LLINFO_UPD: 1653 if (cmd == RTM_LLINFO_UPD && ifa->ifa_rtrequest != NULL) 1654 ifa->ifa_rtrequest(RTM_LLINFO_UPD, rt, &info); 1655 rt_newmsg(RTM_CHANGE, rt); 1656 rt_unref(rt); 1657 break; 1658 case RTM_ADD: 1659 KASSERT(rt->rt_ifa == ifa); 1660 rt_newmsg(cmd, rt); 1661 rt_unref(rt); 1662 RT_REFCNT_TRACE(rt); 1663 break; 1664 } 1665 return error; 1666 } 1667 1668 /* 1669 * Create a local route entry for the address. 1670 * Announce the addition of the address and the route to the routing socket. 1671 */ 1672 int 1673 rt_ifa_addlocal(struct ifaddr *ifa) 1674 { 1675 struct rtentry *rt; 1676 int e; 1677 1678 /* If there is no loopback entry, allocate one. */ 1679 rt = rtalloc1(ifa->ifa_addr, 0); 1680 #ifdef RT_DEBUG 1681 if (rt != NULL) 1682 dump_rt(rt); 1683 #endif 1684 if (rt == NULL || (rt->rt_flags & RTF_HOST) == 0 || 1685 (rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) 1686 { 1687 struct rt_addrinfo info; 1688 struct rtentry *nrt; 1689 1690 memset(&info, 0, sizeof(info)); 1691 info.rti_flags = RTF_HOST | RTF_LOCAL | RTF_DONTCHANGEIFA; 1692 info.rti_info[RTAX_DST] = ifa->ifa_addr; 1693 info.rti_info[RTAX_GATEWAY] = 1694 (const struct sockaddr *)ifa->ifa_ifp->if_sadl; 1695 info.rti_ifa = ifa; 1696 nrt = NULL; 1697 e = rtrequest1(RTM_ADD, &info, &nrt); 1698 rt_addrmsg_rt(RTM_ADD, ifa, e, nrt); 1699 if (nrt != NULL) { 1700 KASSERT(nrt->rt_ifa == ifa); 1701 #ifdef RT_DEBUG 1702 dump_rt(nrt); 1703 #endif 1704 rt_unref(nrt); 1705 RT_REFCNT_TRACE(nrt); 1706 } 1707 } else { 1708 e = 0; 1709 rt_addrmsg(RTM_NEWADDR, ifa); 1710 } 1711 if (rt != NULL) 1712 rt_unref(rt); 1713 return e; 1714 } 1715 1716 /* 1717 * Remove the local route entry for the address. 1718 * Announce the removal of the address and the route to the routing socket. 1719 */ 1720 int 1721 rt_ifa_remlocal(struct ifaddr *ifa, struct ifaddr *alt_ifa) 1722 { 1723 struct rtentry *rt; 1724 int e = 0; 1725 1726 rt = rtalloc1(ifa->ifa_addr, 0); 1727 1728 /* 1729 * Before deleting, check if a corresponding loopbacked 1730 * host route surely exists. With this check, we can avoid 1731 * deleting an interface direct route whose destination is 1732 * the same as the address being removed. This can happen 1733 * when removing a subnet-router anycast address on an 1734 * interface attached to a shared medium. 1735 */ 1736 if (rt != NULL && 1737 (rt->rt_flags & RTF_HOST) && 1738 (rt->rt_ifp->if_flags & IFF_LOOPBACK)) 1739 { 1740 /* If we cannot replace the route's ifaddr with the equivalent 1741 * ifaddr of another interface, I believe it is safest to 1742 * delete the route. 1743 */ 1744 if (alt_ifa == NULL) { 1745 e = rtdeletemsg(rt); 1746 if (e == 0) { 1747 rt_unref(rt); 1748 rt_free(rt); 1749 rt = NULL; 1750 } 1751 rt_addrmsg(RTM_DELADDR, ifa); 1752 } else { 1753 #ifdef NET_MPSAFE 1754 int error = rt_update_prepare(rt); 1755 if (error == 0) { 1756 rt_replace_ifa(rt, alt_ifa); 1757 rt_update_finish(rt); 1758 } else { 1759 /* 1760 * If error != 0, the rtentry is being 1761 * destroyed, so doing nothing doesn't 1762 * matter. 1763 */ 1764 } 1765 #else 1766 rt_replace_ifa(rt, alt_ifa); 1767 #endif 1768 rt_newmsg(RTM_CHANGE, rt); 1769 } 1770 } else 1771 rt_addrmsg(RTM_DELADDR, ifa); 1772 if (rt != NULL) 1773 rt_unref(rt); 1774 return e; 1775 } 1776 1777 /* 1778 * Route timer routines. These routes allow functions to be called 1779 * for various routes at any time. This is useful in supporting 1780 * path MTU discovery and redirect route deletion. 1781 * 1782 * This is similar to some BSDI internal functions, but it provides 1783 * for multiple queues for efficiency's sake... 1784 */ 1785 1786 LIST_HEAD(, rttimer_queue) rttimer_queue_head; 1787 static int rt_init_done = 0; 1788 1789 /* 1790 * Some subtle order problems with domain initialization mean that 1791 * we cannot count on this being run from rt_init before various 1792 * protocol initializations are done. Therefore, we make sure 1793 * that this is run when the first queue is added... 1794 */ 1795 1796 static void rt_timer_work(struct work *, void *); 1797 1798 static void 1799 rt_timer_init(void) 1800 { 1801 int error; 1802 1803 assert(rt_init_done == 0); 1804 1805 /* XXX should be in rt_init */ 1806 rw_init(&rt_lock); 1807 1808 LIST_INIT(&rttimer_queue_head); 1809 callout_init(&rt_timer_ch, CALLOUT_MPSAFE); 1810 error = workqueue_create(&rt_timer_wq, "rt_timer", 1811 rt_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE); 1812 if (error) 1813 panic("%s: workqueue_create failed (%d)\n", __func__, error); 1814 callout_reset(&rt_timer_ch, hz, rt_timer_timer, NULL); 1815 rt_init_done = 1; 1816 } 1817 1818 struct rttimer_queue * 1819 rt_timer_queue_create(u_int timeout) 1820 { 1821 struct rttimer_queue *rtq; 1822 1823 if (rt_init_done == 0) 1824 rt_timer_init(); 1825 1826 R_Malloc(rtq, struct rttimer_queue *, sizeof *rtq); 1827 if (rtq == NULL) 1828 return NULL; 1829 memset(rtq, 0, sizeof(*rtq)); 1830 1831 rtq->rtq_timeout = timeout; 1832 TAILQ_INIT(&rtq->rtq_head); 1833 RT_WLOCK(); 1834 LIST_INSERT_HEAD(&rttimer_queue_head, rtq, rtq_link); 1835 RT_UNLOCK(); 1836 1837 return rtq; 1838 } 1839 1840 void 1841 rt_timer_queue_change(struct rttimer_queue *rtq, long timeout) 1842 { 1843 1844 rtq->rtq_timeout = timeout; 1845 } 1846 1847 static void 1848 rt_timer_queue_remove_all(struct rttimer_queue *rtq) 1849 { 1850 struct rttimer *r; 1851 1852 RT_ASSERT_WLOCK(); 1853 1854 while ((r = TAILQ_FIRST(&rtq->rtq_head)) != NULL) { 1855 LIST_REMOVE(r, rtt_link); 1856 TAILQ_REMOVE(&rtq->rtq_head, r, rtt_next); 1857 rt_ref(r->rtt_rt); /* XXX */ 1858 RT_REFCNT_TRACE(r->rtt_rt); 1859 RT_UNLOCK(); 1860 (*r->rtt_func)(r->rtt_rt, r); 1861 pool_put(&rttimer_pool, r); 1862 RT_WLOCK(); 1863 if (rtq->rtq_count > 0) 1864 rtq->rtq_count--; 1865 else 1866 printf("rt_timer_queue_remove_all: " 1867 "rtq_count reached 0\n"); 1868 } 1869 } 1870 1871 void 1872 rt_timer_queue_destroy(struct rttimer_queue *rtq) 1873 { 1874 1875 RT_WLOCK(); 1876 rt_timer_queue_remove_all(rtq); 1877 LIST_REMOVE(rtq, rtq_link); 1878 RT_UNLOCK(); 1879 1880 /* 1881 * Caller is responsible for freeing the rttimer_queue structure. 1882 */ 1883 } 1884 1885 unsigned long 1886 rt_timer_count(struct rttimer_queue *rtq) 1887 { 1888 return rtq->rtq_count; 1889 } 1890 1891 static void 1892 rt_timer_remove_all(struct rtentry *rt) 1893 { 1894 struct rttimer *r; 1895 1896 RT_WLOCK(); 1897 while ((r = LIST_FIRST(&rt->rt_timer)) != NULL) { 1898 LIST_REMOVE(r, rtt_link); 1899 TAILQ_REMOVE(&r->rtt_queue->rtq_head, r, rtt_next); 1900 if (r->rtt_queue->rtq_count > 0) 1901 r->rtt_queue->rtq_count--; 1902 else 1903 printf("rt_timer_remove_all: rtq_count reached 0\n"); 1904 pool_put(&rttimer_pool, r); 1905 } 1906 RT_UNLOCK(); 1907 } 1908 1909 int 1910 rt_timer_add(struct rtentry *rt, 1911 void (*func)(struct rtentry *, struct rttimer *), 1912 struct rttimer_queue *queue) 1913 { 1914 struct rttimer *r; 1915 1916 KASSERT(func != NULL); 1917 RT_WLOCK(); 1918 /* 1919 * If there's already a timer with this action, destroy it before 1920 * we add a new one. 1921 */ 1922 LIST_FOREACH(r, &rt->rt_timer, rtt_link) { 1923 if (r->rtt_func == func) 1924 break; 1925 } 1926 if (r != NULL) { 1927 LIST_REMOVE(r, rtt_link); 1928 TAILQ_REMOVE(&r->rtt_queue->rtq_head, r, rtt_next); 1929 if (r->rtt_queue->rtq_count > 0) 1930 r->rtt_queue->rtq_count--; 1931 else 1932 printf("rt_timer_add: rtq_count reached 0\n"); 1933 } else { 1934 r = pool_get(&rttimer_pool, PR_NOWAIT); 1935 if (r == NULL) { 1936 RT_UNLOCK(); 1937 return ENOBUFS; 1938 } 1939 } 1940 1941 memset(r, 0, sizeof(*r)); 1942 1943 r->rtt_rt = rt; 1944 r->rtt_time = time_uptime; 1945 r->rtt_func = func; 1946 r->rtt_queue = queue; 1947 LIST_INSERT_HEAD(&rt->rt_timer, r, rtt_link); 1948 TAILQ_INSERT_TAIL(&queue->rtq_head, r, rtt_next); 1949 r->rtt_queue->rtq_count++; 1950 1951 RT_UNLOCK(); 1952 1953 return 0; 1954 } 1955 1956 static void 1957 rt_timer_work(struct work *wk, void *arg) 1958 { 1959 struct rttimer_queue *rtq; 1960 struct rttimer *r; 1961 1962 RT_WLOCK(); 1963 LIST_FOREACH(rtq, &rttimer_queue_head, rtq_link) { 1964 while ((r = TAILQ_FIRST(&rtq->rtq_head)) != NULL && 1965 (r->rtt_time + rtq->rtq_timeout) < time_uptime) { 1966 LIST_REMOVE(r, rtt_link); 1967 TAILQ_REMOVE(&rtq->rtq_head, r, rtt_next); 1968 /* 1969 * Take a reference to avoid the rtentry is freed 1970 * accidentally after RT_UNLOCK. The callback 1971 * (rtt_func) must rt_unref it by itself. 1972 */ 1973 rt_ref(r->rtt_rt); 1974 RT_REFCNT_TRACE(r->rtt_rt); 1975 RT_UNLOCK(); 1976 (*r->rtt_func)(r->rtt_rt, r); 1977 pool_put(&rttimer_pool, r); 1978 RT_WLOCK(); 1979 if (rtq->rtq_count > 0) 1980 rtq->rtq_count--; 1981 else 1982 printf("rt_timer_timer: rtq_count reached 0\n"); 1983 } 1984 } 1985 RT_UNLOCK(); 1986 1987 callout_reset(&rt_timer_ch, hz, rt_timer_timer, NULL); 1988 } 1989 1990 static void 1991 rt_timer_timer(void *arg) 1992 { 1993 1994 workqueue_enqueue(rt_timer_wq, &rt_timer_wk, NULL); 1995 } 1996 1997 static struct rtentry * 1998 _rtcache_init(struct route *ro, int flag) 1999 { 2000 struct rtentry *rt; 2001 2002 rtcache_invariants(ro); 2003 KASSERT(ro->_ro_rt == NULL); 2004 2005 if (rtcache_getdst(ro) == NULL) 2006 return NULL; 2007 rt = rtalloc1(rtcache_getdst(ro), flag); 2008 if (rt != NULL) { 2009 RT_RLOCK(); 2010 if (ISSET(rt->rt_flags, RTF_UP)) { 2011 ro->_ro_rt = rt; 2012 ro->ro_rtcache_generation = rtcache_generation; 2013 rtcache_ref(rt, ro); 2014 } 2015 RT_UNLOCK(); 2016 rt_unref(rt); 2017 } 2018 2019 rtcache_invariants(ro); 2020 return ro->_ro_rt; 2021 } 2022 2023 struct rtentry * 2024 rtcache_init(struct route *ro) 2025 { 2026 2027 return _rtcache_init(ro, 1); 2028 } 2029 2030 struct rtentry * 2031 rtcache_init_noclone(struct route *ro) 2032 { 2033 2034 return _rtcache_init(ro, 0); 2035 } 2036 2037 struct rtentry * 2038 rtcache_update(struct route *ro, int clone) 2039 { 2040 2041 ro->_ro_rt = NULL; 2042 return _rtcache_init(ro, clone); 2043 } 2044 2045 void 2046 rtcache_copy(struct route *new_ro, struct route *old_ro) 2047 { 2048 struct rtentry *rt; 2049 int ret; 2050 2051 KASSERT(new_ro != old_ro); 2052 rtcache_invariants(new_ro); 2053 rtcache_invariants(old_ro); 2054 2055 rt = rtcache_validate(old_ro); 2056 2057 if (rtcache_getdst(old_ro) == NULL) 2058 goto out; 2059 ret = rtcache_setdst(new_ro, rtcache_getdst(old_ro)); 2060 if (ret != 0) 2061 goto out; 2062 2063 RT_RLOCK(); 2064 new_ro->_ro_rt = rt; 2065 new_ro->ro_rtcache_generation = rtcache_generation; 2066 RT_UNLOCK(); 2067 rtcache_invariants(new_ro); 2068 out: 2069 rtcache_unref(rt, old_ro); 2070 return; 2071 } 2072 2073 #if defined(RT_DEBUG) && defined(NET_MPSAFE) 2074 static void 2075 rtcache_trace(const char *func, struct rtentry *rt, struct route *ro) 2076 { 2077 char dst[64]; 2078 2079 sockaddr_format(ro->ro_sa, dst, 64); 2080 printf("trace: %s:\tdst=%s cpu=%d lwp=%p psref=%p target=%p\n", func, dst, 2081 cpu_index(curcpu()), curlwp, &ro->ro_psref, &rt->rt_psref); 2082 } 2083 #define RTCACHE_PSREF_TRACE(rt, ro) rtcache_trace(__func__, (rt), (ro)) 2084 #else 2085 #define RTCACHE_PSREF_TRACE(rt, ro) do {} while (0) 2086 #endif 2087 2088 static void 2089 rtcache_ref(struct rtentry *rt, struct route *ro) 2090 { 2091 2092 KASSERT(rt != NULL); 2093 2094 #ifdef NET_MPSAFE 2095 RTCACHE_PSREF_TRACE(rt, ro); 2096 ro->ro_bound = curlwp_bind(); 2097 /* XXX Use a real caller's address */ 2098 PSREF_DEBUG_FILL_RETURN_ADDRESS(&ro->ro_psref); 2099 psref_acquire(&ro->ro_psref, &rt->rt_psref, rt_psref_class); 2100 #endif 2101 } 2102 2103 void 2104 rtcache_unref(struct rtentry *rt, struct route *ro) 2105 { 2106 2107 if (rt == NULL) 2108 return; 2109 2110 #ifdef NET_MPSAFE 2111 psref_release(&ro->ro_psref, &rt->rt_psref, rt_psref_class); 2112 curlwp_bindx(ro->ro_bound); 2113 RTCACHE_PSREF_TRACE(rt, ro); 2114 #endif 2115 } 2116 2117 struct rtentry * 2118 rtcache_validate(struct route *ro) 2119 { 2120 struct rtentry *rt = NULL; 2121 2122 #ifdef NET_MPSAFE 2123 retry: 2124 #endif 2125 rtcache_invariants(ro); 2126 RT_RLOCK(); 2127 if (ro->ro_rtcache_generation != rtcache_generation) { 2128 /* The cache is invalidated */ 2129 rt = NULL; 2130 goto out; 2131 } 2132 2133 rt = ro->_ro_rt; 2134 if (rt == NULL) 2135 goto out; 2136 2137 if ((rt->rt_flags & RTF_UP) == 0) { 2138 rt = NULL; 2139 goto out; 2140 } 2141 #ifdef NET_MPSAFE 2142 if (ISSET(rt->rt_flags, RTF_UPDATING)) { 2143 if (rt_wait_ok()) { 2144 RT_UNLOCK(); 2145 2146 /* We can wait until the update is complete */ 2147 rt_update_wait(); 2148 goto retry; 2149 } else { 2150 rt = NULL; 2151 } 2152 } else 2153 #endif 2154 rtcache_ref(rt, ro); 2155 out: 2156 RT_UNLOCK(); 2157 return rt; 2158 } 2159 2160 struct rtentry * 2161 rtcache_lookup2(struct route *ro, const struct sockaddr *dst, 2162 int clone, int *hitp) 2163 { 2164 const struct sockaddr *odst; 2165 struct rtentry *rt = NULL; 2166 2167 odst = rtcache_getdst(ro); 2168 if (odst == NULL) 2169 goto miss; 2170 2171 if (sockaddr_cmp(odst, dst) != 0) { 2172 rtcache_free(ro); 2173 goto miss; 2174 } 2175 2176 rt = rtcache_validate(ro); 2177 if (rt == NULL) { 2178 ro->_ro_rt = NULL; 2179 goto miss; 2180 } 2181 2182 rtcache_invariants(ro); 2183 2184 if (hitp != NULL) 2185 *hitp = 1; 2186 return rt; 2187 miss: 2188 if (hitp != NULL) 2189 *hitp = 0; 2190 if (rtcache_setdst(ro, dst) == 0) 2191 rt = _rtcache_init(ro, clone); 2192 2193 rtcache_invariants(ro); 2194 2195 return rt; 2196 } 2197 2198 void 2199 rtcache_free(struct route *ro) 2200 { 2201 2202 ro->_ro_rt = NULL; 2203 if (ro->ro_sa != NULL) { 2204 sockaddr_free(ro->ro_sa); 2205 ro->ro_sa = NULL; 2206 } 2207 rtcache_invariants(ro); 2208 } 2209 2210 int 2211 rtcache_setdst(struct route *ro, const struct sockaddr *sa) 2212 { 2213 KASSERT(sa != NULL); 2214 2215 rtcache_invariants(ro); 2216 if (ro->ro_sa != NULL) { 2217 if (ro->ro_sa->sa_family == sa->sa_family) { 2218 ro->_ro_rt = NULL; 2219 sockaddr_copy(ro->ro_sa, ro->ro_sa->sa_len, sa); 2220 rtcache_invariants(ro); 2221 return 0; 2222 } 2223 /* free ro_sa, wrong family */ 2224 rtcache_free(ro); 2225 } 2226 2227 KASSERT(ro->_ro_rt == NULL); 2228 2229 if ((ro->ro_sa = sockaddr_dup(sa, M_ZERO | M_NOWAIT)) == NULL) { 2230 rtcache_invariants(ro); 2231 return ENOMEM; 2232 } 2233 rtcache_invariants(ro); 2234 return 0; 2235 } 2236 2237 static void 2238 rtcache_percpu_init_cpu(void *p, void *arg __unused, struct cpu_info *ci __unused) 2239 { 2240 struct route **rop = p; 2241 2242 /* 2243 * We can't have struct route as percpu data because it can be destroyed 2244 * over a memory enlargement processing of percpu. 2245 */ 2246 *rop = kmem_zalloc(sizeof(**rop), KM_SLEEP); 2247 } 2248 2249 percpu_t * 2250 rtcache_percpu_alloc(void) 2251 { 2252 2253 return percpu_create(sizeof(struct route *), 2254 rtcache_percpu_init_cpu, NULL, NULL); 2255 } 2256 2257 const struct sockaddr * 2258 rt_settag(struct rtentry *rt, const struct sockaddr *tag) 2259 { 2260 if (rt->rt_tag != tag) { 2261 if (rt->rt_tag != NULL) 2262 sockaddr_free(rt->rt_tag); 2263 rt->rt_tag = sockaddr_dup(tag, M_ZERO | M_NOWAIT); 2264 } 2265 return rt->rt_tag; 2266 } 2267 2268 struct sockaddr * 2269 rt_gettag(const struct rtentry *rt) 2270 { 2271 return rt->rt_tag; 2272 } 2273 2274 int 2275 rt_check_reject_route(const struct rtentry *rt, const struct ifnet *ifp) 2276 { 2277 2278 if ((rt->rt_flags & RTF_REJECT) != 0) { 2279 /* Mimic looutput */ 2280 if (ifp->if_flags & IFF_LOOPBACK) 2281 return (rt->rt_flags & RTF_HOST) ? 2282 EHOSTUNREACH : ENETUNREACH; 2283 else if (rt->rt_rmx.rmx_expire == 0 || 2284 time_uptime < rt->rt_rmx.rmx_expire) 2285 return (rt->rt_flags & RTF_GATEWAY) ? 2286 EHOSTUNREACH : EHOSTDOWN; 2287 } 2288 2289 return 0; 2290 } 2291 2292 void 2293 rt_delete_matched_entries(sa_family_t family, int (*f)(struct rtentry *, void *), 2294 void *v, bool notify) 2295 { 2296 2297 for (;;) { 2298 int s; 2299 int error; 2300 struct rtentry *rt, *retrt = NULL; 2301 2302 RT_RLOCK(); 2303 s = splsoftnet(); 2304 rt = rtbl_search_matched_entry(family, f, v); 2305 if (rt == NULL) { 2306 splx(s); 2307 RT_UNLOCK(); 2308 return; 2309 } 2310 rt_ref(rt); 2311 RT_REFCNT_TRACE(rt); 2312 splx(s); 2313 RT_UNLOCK(); 2314 2315 error = rtrequest(RTM_DELETE, rt_getkey(rt), rt->rt_gateway, 2316 rt_mask(rt), rt->rt_flags, &retrt); 2317 if (error == 0) { 2318 KASSERT(retrt == rt); 2319 KASSERT((retrt->rt_flags & RTF_UP) == 0); 2320 if (notify) 2321 rt_newmsg(RTM_DELETE, retrt); 2322 retrt->rt_ifp = NULL; 2323 rt_unref(rt); 2324 RT_REFCNT_TRACE(rt); 2325 rt_free(retrt); 2326 } else if (error == ESRCH) { 2327 /* Someone deleted the entry already. */ 2328 rt_unref(rt); 2329 RT_REFCNT_TRACE(rt); 2330 } else { 2331 log(LOG_ERR, "%s: unable to delete rtentry @ %p, " 2332 "error = %d\n", rt->rt_ifp->if_xname, rt, error); 2333 /* XXX how to treat this case? */ 2334 } 2335 } 2336 } 2337 2338 static int 2339 rt_walktree_locked(sa_family_t family, int (*f)(struct rtentry *, void *), 2340 void *v) 2341 { 2342 2343 return rtbl_walktree(family, f, v); 2344 } 2345 2346 void 2347 rt_replace_ifa_matched_entries(sa_family_t family, 2348 int (*f)(struct rtentry *, void *), void *v, struct ifaddr *ifa) 2349 { 2350 2351 for (;;) { 2352 int s; 2353 #ifdef NET_MPSAFE 2354 int error; 2355 #endif 2356 struct rtentry *rt; 2357 2358 RT_RLOCK(); 2359 s = splsoftnet(); 2360 rt = rtbl_search_matched_entry(family, f, v); 2361 if (rt == NULL) { 2362 splx(s); 2363 RT_UNLOCK(); 2364 return; 2365 } 2366 rt_ref(rt); 2367 RT_REFCNT_TRACE(rt); 2368 splx(s); 2369 RT_UNLOCK(); 2370 2371 #ifdef NET_MPSAFE 2372 error = rt_update_prepare(rt); 2373 if (error == 0) { 2374 rt_replace_ifa(rt, ifa); 2375 rt_update_finish(rt); 2376 rt_newmsg(RTM_CHANGE, rt); 2377 } else { 2378 /* 2379 * If error != 0, the rtentry is being 2380 * destroyed, so doing nothing doesn't 2381 * matter. 2382 */ 2383 } 2384 #else 2385 rt_replace_ifa(rt, ifa); 2386 rt_newmsg(RTM_CHANGE, rt); 2387 #endif 2388 rt_unref(rt); 2389 RT_REFCNT_TRACE(rt); 2390 } 2391 } 2392 2393 int 2394 rt_walktree(sa_family_t family, int (*f)(struct rtentry *, void *), void *v) 2395 { 2396 int error; 2397 2398 RT_RLOCK(); 2399 error = rt_walktree_locked(family, f, v); 2400 RT_UNLOCK(); 2401 2402 return error; 2403 } 2404 2405 #ifdef DDB 2406 2407 #include <machine/db_machdep.h> 2408 #include <ddb/db_interface.h> 2409 #include <ddb/db_output.h> 2410 2411 #define rt_expire rt_rmx.rmx_expire 2412 2413 static void 2414 db_print_sa(const struct sockaddr *sa) 2415 { 2416 int len; 2417 const u_char *p; 2418 2419 if (sa == NULL) { 2420 db_printf("[NULL]"); 2421 return; 2422 } 2423 2424 p = (const u_char *)sa; 2425 len = sa->sa_len; 2426 db_printf("["); 2427 while (len > 0) { 2428 db_printf("%d", *p); 2429 p++; len--; 2430 if (len) db_printf(","); 2431 } 2432 db_printf("]\n"); 2433 } 2434 2435 static void 2436 db_print_ifa(struct ifaddr *ifa) 2437 { 2438 if (ifa == NULL) 2439 return; 2440 db_printf(" ifa_addr="); 2441 db_print_sa(ifa->ifa_addr); 2442 db_printf(" ifa_dsta="); 2443 db_print_sa(ifa->ifa_dstaddr); 2444 db_printf(" ifa_mask="); 2445 db_print_sa(ifa->ifa_netmask); 2446 db_printf(" flags=0x%x,refcnt=%d,metric=%d\n", 2447 ifa->ifa_flags, 2448 ifa->ifa_refcnt, 2449 ifa->ifa_metric); 2450 } 2451 2452 /* 2453 * Function to pass to rt_walktree(). 2454 * Return non-zero error to abort walk. 2455 */ 2456 static int 2457 db_show_rtentry(struct rtentry *rt, void *w) 2458 { 2459 db_printf("rtentry=%p", rt); 2460 2461 db_printf(" flags=0x%x refcnt=%d use=%"PRId64" expire=%"PRId64"\n", 2462 rt->rt_flags, rt->rt_refcnt, 2463 rt->rt_use, (uint64_t)rt->rt_expire); 2464 2465 db_printf(" key="); db_print_sa(rt_getkey(rt)); 2466 db_printf(" mask="); db_print_sa(rt_mask(rt)); 2467 db_printf(" gw="); db_print_sa(rt->rt_gateway); 2468 2469 db_printf(" ifp=%p ", rt->rt_ifp); 2470 if (rt->rt_ifp) 2471 db_printf("(%s)", rt->rt_ifp->if_xname); 2472 else 2473 db_printf("(NULL)"); 2474 2475 db_printf(" ifa=%p\n", rt->rt_ifa); 2476 db_print_ifa(rt->rt_ifa); 2477 2478 db_printf(" gwroute=%p llinfo=%p\n", 2479 rt->rt_gwroute, rt->rt_llinfo); 2480 2481 return 0; 2482 } 2483 2484 /* 2485 * Function to print all the route trees. 2486 * Use this from ddb: "show routes" 2487 */ 2488 void 2489 db_show_routes(db_expr_t addr, bool have_addr, 2490 db_expr_t count, const char *modif) 2491 { 2492 2493 /* Taking RT_LOCK will fail if LOCKDEBUG is enabled. */ 2494 rt_walktree_locked(AF_INET, db_show_rtentry, NULL); 2495 } 2496 #endif 2497