contrib/bind-9.3/bin/named/client.c

   1 /*
   2  * Copyright (C) 2004, 2005  Internet Systems Consortium, Inc. ("ISC")
   3  * Copyright (C) 1999-2003  Internet Software Consortium.
   4  *
   5  * Permission to use, copy, modify, and distribute this software for any
   6  * purpose with or without fee is hereby granted, provided that the above
   7  * copyright notice and this permission notice appear in all copies.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
  10  * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
  11  * AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
  12  * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
  13  * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
  14  * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
  15  * PERFORMANCE OF THIS SOFTWARE.
  16  */
  17
  18 /* $Id: client.c,v 1.176.2.13.4.26 2005/07/27 02:53:14 marka Exp $ */
  19
  20 #include <config.h>
  21
  22 #include <isc/formatcheck.h>
  23 #include <isc/mutex.h>
  24 #include <isc/once.h>
  25 #include <isc/print.h>
  26 #include <isc/stdio.h>
  27 #include <isc/string.h>
  28 #include <isc/task.h>
  29 #include <isc/timer.h>
  30 #include <isc/util.h>
  31
  32 #include <dns/db.h>
  33 #include <dns/dispatch.h>
  34 #include <dns/events.h>
  35 #include <dns/message.h>
  36 #include <dns/rcode.h>
  37 #include <dns/resolver.h>
  38 #include <dns/rdata.h>
  39 #include <dns/rdataclass.h>
  40 #include <dns/rdatalist.h>
  41 #include <dns/rdataset.h>
  42 #include <dns/tsig.h>
  43 #include <dns/view.h>
  44 #include <dns/zone.h>
  45
  46 #include <named/interfacemgr.h>
  47 #include <named/log.h>
  48 #include <named/notify.h>
  49 #include <named/server.h>
  50 #include <named/update.h>
  51
  52 /***
  53  *** Client
  54  ***/
  55
  56 /*
  57  * Important note!
  58  *
  59  * All client state changes, other than that from idle to listening, occur
  60  * as a result of events.  This guarantees serialization and avoids the
  61  * need for locking.
  62  *
  63  * If a routine is ever created that allows someone other than the client's
  64  * task to change the client, then the client will have to be locked.
  65  */
  66
  67 #define NS_CLIENT_TRACE
  68 #ifdef NS_CLIENT_TRACE
  69 #define CTRACE(m)       ns_client_log(client, \
  70                                       NS_LOGCATEGORY_CLIENT, \
  71                                       NS_LOGMODULE_CLIENT, \
  72                                       ISC_LOG_DEBUG(3), \
  73                                       "%s", (m))
  74 #define MTRACE(m)       isc_log_write(ns_g_lctx, \
  75                                       NS_LOGCATEGORY_GENERAL, \
  76                                       NS_LOGMODULE_CLIENT, \
  77                                       ISC_LOG_DEBUG(3), \
  78                                       "clientmgr @%p: %s", manager, (m))
  79 #else
  80 #define CTRACE(m)       ((void)(m))
  81 #define MTRACE(m)       ((void)(m))
  82 #endif
  83
  84 #define TCP_CLIENT(c)   (((c)->attributes & NS_CLIENTATTR_TCP) != 0)
  85
  86 #define TCP_BUFFER_SIZE                 (65535 + 2)
  87 #define SEND_BUFFER_SIZE                4096
  88 #define RECV_BUFFER_SIZE                4096
  89
  90 struct ns_clientmgr {
  91         /* Unlocked. */
  92         unsigned int                    magic;
  93         isc_mem_t *                     mctx;
  94         isc_taskmgr_t *                 taskmgr;
  95         isc_timermgr_t *                timermgr;
  96         isc_mutex_t                     lock;
  97         /* Locked by lock. */
  98         isc_boolean_t                   exiting;
  99         client_list_t                   active;         /* Active clients */
 100         client_list_t                   recursing;      /* Recursing clients */
 101         client_list_t                   inactive;       /* To be recycled */
 102 };
 103
 104 #define MANAGER_MAGIC                   ISC_MAGIC('N', 'S', 'C', 'm')
 105 #define VALID_MANAGER(m)                ISC_MAGIC_VALID(m, MANAGER_MAGIC)
 106
 107 /*
 108  * Client object states.  Ordering is significant: higher-numbered
 109  * states are generally "more active", meaning that the client can
 110  * have more dynamically allocated data, outstanding events, etc.
 111  * In the list below, any such properties listed for state N
 112  * also apply to any state > N.
 113  *
 114  * To force the client into a less active state, set client->newstate
 115  * to that state and call exit_check().  This will cause any
 116  * activities defined for higher-numbered states to be aborted.
 117  */
 118
 119 #define NS_CLIENTSTATE_FREED    0
 120 /*
 121  * The client object no longer exists.
 122  */
 123
 124 #define NS_CLIENTSTATE_INACTIVE 1
 125 /*
 126  * The client object exists and has a task and timer.
 127  * Its "query" struct and sendbuf are initialized.
 128  * It is on the client manager's list of inactive clients.
 129  * It has a message and OPT, both in the reset state.
 130  */
 131
 132 #define NS_CLIENTSTATE_READY    2
 133 /*
 134  * The client object is either a TCP or a UDP one, and
 135  * it is associated with a network interface.  It is on the
 136  * client manager's list of active clients.
 137  *
 138  * If it is a TCP client object, it has a TCP listener socket
 139  * and an outstanding TCP listen request.
 140  *
 141  * If it is a UDP client object, it has a UDP listener socket
 142  * and an outstanding UDP receive request.
 143  */
 144
 145 #define NS_CLIENTSTATE_READING  3
 146 /*
 147  * The client object is a TCP client object that has received
 148  * a connection.  It has a tcpsocket, tcpmsg, TCP quota, and an
 149  * outstanding TCP read request.  This state is not used for
 150  * UDP client objects.
 151  */
 152
 153 #define NS_CLIENTSTATE_WORKING  4
 154 /*
 155  * The client object has received a request and is working
 156  * on it.  It has a view, and it may have any of a non-reset OPT,
 157  * recursion quota, and an outstanding write request.
 158  */
 159
 160 #define NS_CLIENTSTATE_MAX      9
 161 /*
 162  * Sentinel value used to indicate "no state".  When client->newstate
 163  * has this value, we are not attempting to exit the current state.
 164  * Must be greater than any valid state.
 165  */
 166
 167
 168 static void client_read(ns_client_t *client);
 169 static void client_accept(ns_client_t *client);
 170 static void client_udprecv(ns_client_t *client);
 171 static void clientmgr_destroy(ns_clientmgr_t *manager);
 172 static isc_boolean_t exit_check(ns_client_t *client);
 173 static void ns_client_endrequest(ns_client_t *client);
 174 static void ns_client_checkactive(ns_client_t *client);
 175 static void client_start(isc_task_t *task, isc_event_t *event);
 176 static void client_request(isc_task_t *task, isc_event_t *event);
 177 static void ns_client_dumpmessage(ns_client_t *client, const char *reason);
 178
 179 void
 180 ns_client_recursing(ns_client_t *client) {
 181         REQUIRE(NS_CLIENT_VALID(client));
 182
 183         LOCK(&client->manager->lock);
 184         ISC_LIST_UNLINK(*client->list, client, link);
 185         ISC_LIST_APPEND(client->manager->recursing, client, link);
 186         client->list = &client->manager->recursing;
 187         UNLOCK(&client->manager->lock);
 188 }
 189
 190 void
 191 ns_client_killoldestquery(ns_client_t *client) {
 192         ns_client_t *oldest;
 193         REQUIRE(NS_CLIENT_VALID(client));
 194
 195         LOCK(&client->manager->lock);
 196         oldest = ISC_LIST_HEAD(client->manager->recursing);
 197         if (oldest != NULL) {
 198                 ns_query_cancel(oldest);
 199                 ISC_LIST_UNLINK(*oldest->list, oldest, link);
 200                 ISC_LIST_APPEND(client->manager->active, oldest, link);
 201                 oldest->list = &client->manager->active;
 202         }
 203         UNLOCK(&client->manager->lock);
 204 }
 205
 206 void
 207 ns_client_settimeout(ns_client_t *client, unsigned int seconds) {
 208         isc_result_t result;
 209         isc_interval_t interval;
 210
 211         isc_interval_set(&interval, seconds, 0);
 212         result = isc_timer_reset(client->timer, isc_timertype_once, NULL,
 213                                  &interval, ISC_FALSE);
 214         client->timerset = ISC_TRUE;
 215         if (result != ISC_R_SUCCESS) {
 216                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
 217                               NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
 218                               "setting timeout: %s",
 219                               isc_result_totext(result));
 220                 /* Continue anyway. */
 221         }
 222 }
 223
 224 /*
 225  * Check for a deactivation or shutdown request and take appropriate
 226  * action.  Returns ISC_TRUE if either is in progress; in this case
 227  * the caller must no longer use the client object as it may have been
 228  * freed.
 229  */
 230 static isc_boolean_t
 231 exit_check(ns_client_t *client) {
 232         ns_clientmgr_t *locked_manager = NULL;
 233         ns_clientmgr_t *destroy_manager = NULL;
 234
 235         REQUIRE(NS_CLIENT_VALID(client));
 236
 237         if (client->state <= client->newstate)
 238                 return (ISC_FALSE); /* Business as usual. */
 239
 240         INSIST(client->newstate < NS_CLIENTSTATE_WORKING);
 241
 242         /*
 243          * We need to detach from the view early when shutting down
 244          * the server to break the following vicious circle:
 245          *
 246          *  - The resolver will not shut down until the view refcount is zero
 247          *  - The view refcount does not go to zero until all clients detach
 248          *  - The client does not detach from the view until references is zero
 249          *  - references does not go to zero until the resolver has shut down
 250          *
 251          * Keep the view attached until any outstanding updates complete.
 252          */
 253         if (client->nupdates == 0 &&
 254             client->newstate == NS_CLIENTSTATE_FREED && client->view != NULL)
 255                 dns_view_detach(&client->view);
 256
 257         if (client->state == NS_CLIENTSTATE_WORKING) {
 258                 INSIST(client->newstate <= NS_CLIENTSTATE_READING);
 259                 /*
 260                  * Let the update processing complete.
 261                  */
 262                 if (client->nupdates > 0)
 263                         return (ISC_TRUE);
 264                 /*
 265                  * We are trying to abort request processing.
 266                  */
 267                 if (client->nsends > 0) {
 268                         isc_socket_t *socket;
 269                         if (TCP_CLIENT(client))
 270                                 socket = client->tcpsocket;
 271                         else
 272                                 socket = client->udpsocket;
 273                         isc_socket_cancel(socket, client->task,
 274                                           ISC_SOCKCANCEL_SEND);
 275                 }
 276
 277                 if (! (client->nsends == 0 && client->nrecvs == 0 &&
 278                        client->references == 0))
 279                 {
 280                         /*
 281                          * Still waiting for I/O cancel completion.
 282                          * or lingering references.
 283                          */
 284                         return (ISC_TRUE);
 285                 }
 286                 /*
 287                  * I/O cancel is complete.  Burn down all state
 288                  * related to the current request.
 289                  */
 290                 ns_client_endrequest(client);
 291
 292                 client->state = NS_CLIENTSTATE_READING;
 293                 INSIST(client->recursionquota == NULL);
 294                 if (NS_CLIENTSTATE_READING == client->newstate) {
 295                         client_read(client);
 296                         client->newstate = NS_CLIENTSTATE_MAX;
 297                         return (ISC_TRUE); /* We're done. */
 298                 }
 299         }
 300
 301         if (client->state == NS_CLIENTSTATE_READING) {
 302                 /*
 303                  * We are trying to abort the current TCP connection,
 304                  * if any.
 305                  */
 306                 INSIST(client->recursionquota == NULL);
 307                 INSIST(client->newstate <= NS_CLIENTSTATE_READY);
 308                 if (client->nreads > 0)
 309                         dns_tcpmsg_cancelread(&client->tcpmsg);
 310                 if (! client->nreads == 0) {
 311                         /* Still waiting for read cancel completion. */
 312                         return (ISC_TRUE);
 313                 }
 314
 315                 if (client->tcpmsg_valid) {
 316                         dns_tcpmsg_invalidate(&client->tcpmsg);
 317                         client->tcpmsg_valid = ISC_FALSE;
 318                 }
 319                 if (client->tcpsocket != NULL) {
 320                         CTRACE("closetcp");
 321                         isc_socket_detach(&client->tcpsocket);
 322                 }
 323
 324                 if (client->tcpquota != NULL)
 325                         isc_quota_detach(&client->tcpquota);
 326
 327                 if (client->timerset) {
 328                         (void)isc_timer_reset(client->timer,
 329                                               isc_timertype_inactive,
 330                                               NULL, NULL, ISC_TRUE);
 331                         client->timerset = ISC_FALSE;
 332                 }
 333
 334                 client->peeraddr_valid = ISC_FALSE;
 335
 336                 client->state = NS_CLIENTSTATE_READY;
 337                 INSIST(client->recursionquota == NULL);
 338
 339                 /*
 340                  * Now the client is ready to accept a new TCP connection
 341                  * or UDP request, but we may have enough clients doing
 342                  * that already.  Check whether this client needs to remain
 343                  * active and force it to go inactive if not.
 344                  */
 345                 ns_client_checkactive(client);
 346
 347                 if (NS_CLIENTSTATE_READY == client->newstate) {
 348                         if (TCP_CLIENT(client)) {
 349                                 client_accept(client);
 350                         } else
 351                                 client_udprecv(client);
 352                         client->newstate = NS_CLIENTSTATE_MAX;
 353                         return (ISC_TRUE);
 354                 }
 355         }
 356
 357         if (client->state == NS_CLIENTSTATE_READY) {
 358                 INSIST(client->newstate <= NS_CLIENTSTATE_INACTIVE);
 359                 /*
 360                  * We are trying to enter the inactive state.
 361                  */
 362                 if (client->naccepts > 0)
 363                         isc_socket_cancel(client->tcplistener, client->task,
 364                                           ISC_SOCKCANCEL_ACCEPT);
 365
 366                 if (! (client->naccepts == 0)) {
 367                         /* Still waiting for accept cancel completion. */
 368                         return (ISC_TRUE);
 369                 }
 370                 /* Accept cancel is complete. */
 371
 372                 if (client->nrecvs > 0)
 373                         isc_socket_cancel(client->udpsocket, client->task,
 374                                           ISC_SOCKCANCEL_RECV);
 375                 if (! (client->nrecvs == 0)) {
 376                         /* Still waiting for recv cancel completion. */
 377                         return (ISC_TRUE);
 378                 }
 379                 /* Recv cancel is complete. */
 380
 381                 if (client->nctls > 0) {
 382                         /* Still waiting for control event to be delivered */
 383                         return (ISC_TRUE);
 384                 }
 385
 386                 /* Deactivate the client. */
 387                 if (client->interface)
 388                         ns_interface_detach(&client->interface);
 389
 390                 INSIST(client->naccepts == 0);
 391                 INSIST(client->recursionquota == NULL);
 392                 if (client->tcplistener != NULL)
 393                         isc_socket_detach(&client->tcplistener);
 394
 395                 if (client->udpsocket != NULL)
 396                         isc_socket_detach(&client->udpsocket);
 397
 398                 if (client->dispatch != NULL)
 399                         dns_dispatch_detach(&client->dispatch);
 400
 401                 client->attributes = 0;
 402                 client->mortal = ISC_FALSE;
 403
 404                 LOCK(&client->manager->lock);
 405                 /*
 406                  * Put the client on the inactive list.  If we are aiming for
 407                  * the "freed" state, it will be removed from the inactive
 408                  * list shortly, and we need to keep the manager locked until
 409                  * that has been done, lest the manager decide to reactivate
 410                  * the dying client inbetween.
 411                  */
 412                 locked_manager = client->manager;
 413                 ISC_LIST_UNLINK(*client->list, client, link);
 414                 ISC_LIST_APPEND(client->manager->inactive, client, link);
 415                 client->list = &client->manager->inactive;
 416                 client->state = NS_CLIENTSTATE_INACTIVE;
 417                 INSIST(client->recursionquota == NULL);
 418
 419                 if (client->state == client->newstate) {
 420                         client->newstate = NS_CLIENTSTATE_MAX;
 421                         goto unlock;
 422                 }
 423         }
 424
 425         if (client->state == NS_CLIENTSTATE_INACTIVE) {
 426                 INSIST(client->newstate == NS_CLIENTSTATE_FREED);
 427                 /*
 428                  * We are trying to free the client.
 429                  *
 430                  * When "shuttingdown" is true, either the task has received
 431                  * its shutdown event or no shutdown event has ever been
 432                  * set up.  Thus, we have no outstanding shutdown
 433                  * event at this point.
 434                  */
 435                 REQUIRE(client->state == NS_CLIENTSTATE_INACTIVE);
 436
 437                 INSIST(client->recursionquota == NULL);
 438
 439                 ns_query_free(client);
 440                 isc_mem_put(client->mctx, client->recvbuf, RECV_BUFFER_SIZE);
 441                 isc_event_free((isc_event_t **)&client->sendevent);
 442                 isc_event_free((isc_event_t **)&client->recvevent);
 443                 isc_timer_detach(&client->timer);
 444
 445                 if (client->tcpbuf != NULL)
 446                         isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
 447                 if (client->opt != NULL) {
 448                         INSIST(dns_rdataset_isassociated(client->opt));
 449                         dns_rdataset_disassociate(client->opt);
 450                         dns_message_puttemprdataset(client->message, &client->opt);
 451                 }
 452                 dns_message_destroy(&client->message);
 453                 if (client->manager != NULL) {
 454                         ns_clientmgr_t *manager = client->manager;
 455                         if (locked_manager == NULL) {
 456                                 LOCK(&manager->lock);
 457                                 locked_manager = manager;
 458                         }
 459                         ISC_LIST_UNLINK(*client->list, client, link);
 460                         client->list = NULL;
 461                         if (manager->exiting &&
 462                             ISC_LIST_EMPTY(manager->active) &&
 463                             ISC_LIST_EMPTY(manager->inactive) &&
 464                             ISC_LIST_EMPTY(manager->recursing))
 465                                 destroy_manager = manager;
 466                 }
 467                 /*
 468                  * Detaching the task must be done after unlinking from
 469                  * the manager's lists because the manager accesses
 470                  * client->task.
 471                  */
 472                 if (client->task != NULL)
 473                         isc_task_detach(&client->task);
 474
 475                 CTRACE("free");
 476                 client->magic = 0;
 477                 isc_mem_put(client->mctx, client, sizeof(*client));
 478
 479                 goto unlock;
 480         }
 481
 482  unlock:
 483         if (locked_manager != NULL) {
 484                 UNLOCK(&locked_manager->lock);
 485                 locked_manager = NULL;
 486         }
 487
 488         /*
 489          * Only now is it safe to destroy the client manager (if needed),
 490          * because we have accessed its lock for the last time.
 491          */
 492         if (destroy_manager != NULL)
 493                 clientmgr_destroy(destroy_manager);
 494
 495         return (ISC_TRUE);
 496 }
 497
 498 /*
 499  * The client's task has received the client's control event
 500  * as part of the startup process.
 501  */
 502 static void
 503 client_start(isc_task_t *task, isc_event_t *event) {
 504         ns_client_t *client = (ns_client_t *) event->ev_arg;
 505
 506         INSIST(task == client->task);
 507
 508         UNUSED(task);
 509
 510         INSIST(client->nctls == 1);
 511         client->nctls--;
 512
 513         if (exit_check(client))
 514                 return;
 515
 516         if (TCP_CLIENT(client)) {
 517                 client_accept(client);
 518         } else {
 519                 client_udprecv(client);
 520         }
 521 }
 522
 523
 524 /*
 525  * The client's task has received a shutdown event.
 526  */
 527 static void
 528 client_shutdown(isc_task_t *task, isc_event_t *event) {
 529         ns_client_t *client;
 530
 531         REQUIRE(event != NULL);
 532         REQUIRE(event->ev_type == ISC_TASKEVENT_SHUTDOWN);
 533         client = event->ev_arg;
 534         REQUIRE(NS_CLIENT_VALID(client));
 535         REQUIRE(task == client->task);
 536
 537         UNUSED(task);
 538
 539         CTRACE("shutdown");
 540
 541         isc_event_free(&event);
 542
 543         if (client->shutdown != NULL) {
 544                 (client->shutdown)(client->shutdown_arg, ISC_R_SHUTTINGDOWN);
 545                 client->shutdown = NULL;
 546                 client->shutdown_arg = NULL;
 547         }
 548
 549         client->newstate = NS_CLIENTSTATE_FREED;
 550         (void)exit_check(client);
 551 }
 552
 553 static void
 554 ns_client_endrequest(ns_client_t *client) {
 555         INSIST(client->naccepts == 0);
 556         INSIST(client->nreads == 0);
 557         INSIST(client->nsends == 0);
 558         INSIST(client->nrecvs == 0);
 559         INSIST(client->nupdates == 0);
 560         INSIST(client->state == NS_CLIENTSTATE_WORKING);
 561
 562         CTRACE("endrequest");
 563
 564         if (client->next != NULL) {
 565                 (client->next)(client);
 566                 client->next = NULL;
 567         }
 568
 569         if (client->view != NULL)
 570                 dns_view_detach(&client->view);
 571         if (client->opt != NULL) {
 572                 INSIST(dns_rdataset_isassociated(client->opt));
 573                 dns_rdataset_disassociate(client->opt);
 574                 dns_message_puttemprdataset(client->message, &client->opt);
 575         }
 576
 577         client->udpsize = 512;
 578         client->extflags = 0;
 579         dns_message_reset(client->message, DNS_MESSAGE_INTENTPARSE);
 580
 581         if (client->recursionquota != NULL)
 582                 isc_quota_detach(&client->recursionquota);
 583
 584         /*
 585          * Clear all client attributes that are specific to
 586          * the request; that's all except the TCP flag.
 587          */
 588         client->attributes &= NS_CLIENTATTR_TCP;
 589 }
 590
 591 static void
 592 ns_client_checkactive(ns_client_t *client) {
 593         if (client->mortal) {
 594                 /*
 595                  * This client object should normally go inactive
 596                  * at this point, but if we have fewer active client
 597                  * objects than  desired due to earlier quota exhaustion,
 598                  * keep it active to make up for the shortage.
 599                  */
 600                 isc_boolean_t need_another_client = ISC_FALSE;
 601                 if (TCP_CLIENT(client)) {
 602                         LOCK(&client->interface->lock);
 603                         if (client->interface->ntcpcurrent <
 604                             client->interface->ntcptarget)
 605                                 need_another_client = ISC_TRUE;
 606                         UNLOCK(&client->interface->lock);
 607                 } else {
 608                         /*
 609                          * The UDP client quota is enforced by making
 610                          * requests fail rather than by not listening
 611                          * for new ones.  Therefore, there is always a
 612                          * full set of UDP clients listening.
 613                          */
 614                 }
 615                 if (! need_another_client) {
 616                         /*
 617                          * We don't need this client object.  Recycle it.
 618                          */
 619                         if (client->newstate >= NS_CLIENTSTATE_INACTIVE)
 620                                 client->newstate = NS_CLIENTSTATE_INACTIVE;
 621                 }
 622         }
 623 }
 624
 625 void
 626 ns_client_next(ns_client_t *client, isc_result_t result) {
 627         int newstate;
 628
 629         REQUIRE(NS_CLIENT_VALID(client));
 630         REQUIRE(client->state == NS_CLIENTSTATE_WORKING ||
 631                 client->state == NS_CLIENTSTATE_READING);
 632
 633         CTRACE("next");
 634
 635         if (result != ISC_R_SUCCESS)
 636                 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
 637                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
 638                               "request failed: %s", isc_result_totext(result));
 639
 640         /*
 641          * An error processing a TCP request may have left
 642          * the connection out of sync.  To be safe, we always
 643          * sever the connection when result != ISC_R_SUCCESS.
 644          */
 645         if (result == ISC_R_SUCCESS && TCP_CLIENT(client))
 646                 newstate = NS_CLIENTSTATE_READING;
 647         else
 648                 newstate = NS_CLIENTSTATE_READY;
 649
 650         if (client->newstate > newstate)
 651                 client->newstate = newstate;
 652         (void)exit_check(client);
 653 }
 654
 655
 656 static void
 657 client_senddone(isc_task_t *task, isc_event_t *event) {
 658         ns_client_t *client;
 659         isc_socketevent_t *sevent = (isc_socketevent_t *) event;
 660
 661         REQUIRE(sevent != NULL);
 662         REQUIRE(sevent->ev_type == ISC_SOCKEVENT_SENDDONE);
 663         client = sevent->ev_arg;
 664         REQUIRE(NS_CLIENT_VALID(client));
 665         REQUIRE(task == client->task);
 666         REQUIRE(sevent == client->sendevent);
 667
 668         UNUSED(task);
 669
 670         CTRACE("senddone");
 671
 672         if (sevent->result != ISC_R_SUCCESS)
 673                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
 674                               NS_LOGMODULE_CLIENT, ISC_LOG_WARNING,
 675                               "error sending response: %s",
 676                               isc_result_totext(sevent->result));
 677
 678         INSIST(client->nsends > 0);
 679         client->nsends--;
 680
 681         if (client->tcpbuf != NULL) {
 682                 INSIST(TCP_CLIENT(client));
 683                 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
 684                 client->tcpbuf = NULL;
 685         }
 686
 687         if (exit_check(client))
 688                 return;
 689
 690         ns_client_next(client, ISC_R_SUCCESS);
 691 }
 692
 693 /*
 694  * We only want to fail with ISC_R_NOSPACE when called from
 695  * ns_client_sendraw() and not when called from ns_client_send(),
 696  * tcpbuffer is NULL when called from ns_client_sendraw() and
 697  * length != 0.  tcpbuffer != NULL when called from ns_client_send()
 698  * and length == 0.
 699  */
 700
 701 static isc_result_t
 702 client_allocsendbuf(ns_client_t *client, isc_buffer_t *buffer,
 703                     isc_buffer_t *tcpbuffer, isc_uint32_t length,
 704                     unsigned char *sendbuf, unsigned char **datap)
 705 {
 706         unsigned char *data;
 707         isc_uint32_t bufsize;
 708         isc_result_t result;
 709
 710         INSIST(datap != NULL);
 711         INSIST((tcpbuffer == NULL && length != 0) ||
 712                (tcpbuffer != NULL && length == 0));
 713
 714         if (TCP_CLIENT(client)) {
 715                 INSIST(client->tcpbuf == NULL);
 716                 if (length + 2 > TCP_BUFFER_SIZE) {
 717                         result = ISC_R_NOSPACE;
 718                         goto done;
 719                 }
 720                 client->tcpbuf = isc_mem_get(client->mctx, TCP_BUFFER_SIZE);
 721                 if (client->tcpbuf == NULL) {
 722                         result = ISC_R_NOMEMORY;
 723                         goto done;
 724                 }
 725                 data = client->tcpbuf;
 726                 if (tcpbuffer != NULL) {
 727                         isc_buffer_init(tcpbuffer, data, TCP_BUFFER_SIZE);
 728                         isc_buffer_init(buffer, data + 2, TCP_BUFFER_SIZE - 2);
 729                 } else {
 730                         isc_buffer_init(buffer, data, TCP_BUFFER_SIZE);
 731                         INSIST(length <= 0xffff);
 732                         isc_buffer_putuint16(buffer, (isc_uint16_t)length);
 733                 }
 734         } else {
 735                 data = sendbuf;
 736                 if (client->udpsize < SEND_BUFFER_SIZE)
 737                         bufsize = client->udpsize;
 738                 else
 739                         bufsize = SEND_BUFFER_SIZE;
 740                 if (length > bufsize) {
 741                         result = ISC_R_NOSPACE;
 742                         goto done;
 743                 }
 744                 isc_buffer_init(buffer, data, bufsize);
 745         }
 746         *datap = data;
 747         result = ISC_R_SUCCESS;
 748
 749  done:
 750         return (result);
 751 }
 752
 753 static isc_result_t
 754 client_sendpkg(ns_client_t *client, isc_buffer_t *buffer) {
 755         struct in6_pktinfo *pktinfo;
 756         isc_result_t result;
 757         isc_region_t r;
 758         isc_sockaddr_t *address;
 759         isc_socket_t *socket;
 760         isc_netaddr_t netaddr;
 761         int match;
 762         unsigned int sockflags = ISC_SOCKFLAG_IMMEDIATE;
 763
 764         if (TCP_CLIENT(client)) {
 765                 socket = client->tcpsocket;
 766                 address = NULL;
 767         } else {
 768                 socket = client->udpsocket;
 769                 address = &client->peeraddr;
 770
 771                 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
 772                 if (ns_g_server->blackholeacl != NULL &&
 773                     dns_acl_match(&netaddr, NULL,
 774                                   ns_g_server->blackholeacl,
 775                                   &ns_g_server->aclenv,
 776                                   &match, NULL) == ISC_R_SUCCESS &&
 777                     match > 0)
 778                         return (DNS_R_BLACKHOLED);
 779                 sockflags |= ISC_SOCKFLAG_NORETRY;
 780         }
 781
 782         if ((client->attributes & NS_CLIENTATTR_PKTINFO) != 0 &&
 783             (client->attributes & NS_CLIENTATTR_MULTICAST) == 0)
 784                 pktinfo = &client->pktinfo;
 785         else
 786                 pktinfo = NULL;
 787
 788         isc_buffer_usedregion(buffer, &r);
 789
 790         CTRACE("sendto");
 791
 792         result = isc_socket_sendto2(socket, &r, client->task,
 793                                     address, pktinfo,
 794                                     client->sendevent, sockflags);
 795         if (result == ISC_R_SUCCESS || result == ISC_R_INPROGRESS) {
 796                 client->nsends++;
 797                 if (result == ISC_R_SUCCESS)
 798                         client_senddone(client->task,
 799                                         (isc_event_t *)client->sendevent);
 800                 result = ISC_R_SUCCESS;
 801         }
 802         return (result);
 803 }
 804
 805 void
 806 ns_client_sendraw(ns_client_t *client, dns_message_t *message) {
 807         isc_result_t result;
 808         unsigned char *data;
 809         isc_buffer_t buffer;
 810         isc_region_t r;
 811         isc_region_t *mr;
 812         unsigned char sendbuf[SEND_BUFFER_SIZE];
 813
 814         REQUIRE(NS_CLIENT_VALID(client));
 815
 816         CTRACE("sendraw");
 817
 818         mr = dns_message_getrawmessage(message);
 819         if (mr == NULL) {
 820                 result = ISC_R_UNEXPECTEDEND;
 821                 goto done;
 822         }
 823
 824         result = client_allocsendbuf(client, &buffer, NULL, mr->length,
 825                                      sendbuf, &data);
 826         if (result != ISC_R_SUCCESS)
 827                 goto done;
 828
 829         /*
 830          * Copy message to buffer and fixup id.
 831          */
 832         isc_buffer_availableregion(&buffer, &r);
 833         result = isc_buffer_copyregion(&buffer, mr);
 834         if (result != ISC_R_SUCCESS)
 835                 goto done;
 836         r.base[0] = (client->message->id >> 8) & 0xff;
 837         r.base[1] = client->message->id & 0xff;
 838
 839         result = client_sendpkg(client, &buffer);
 840         if (result == ISC_R_SUCCESS)
 841                 return;
 842
 843  done:
 844         if (client->tcpbuf != NULL) {
 845                 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
 846                 client->tcpbuf = NULL;
 847         }
 848         ns_client_next(client, result);
 849 }
 850
 851 void
 852 ns_client_send(ns_client_t *client) {
 853         isc_result_t result;
 854         unsigned char *data;
 855         isc_buffer_t buffer;
 856         isc_buffer_t tcpbuffer;
 857         isc_region_t r;
 858         dns_compress_t cctx;
 859         isc_boolean_t cleanup_cctx = ISC_FALSE;
 860         unsigned char sendbuf[SEND_BUFFER_SIZE];
 861         unsigned int dnssec_opts;
 862         unsigned int preferred_glue;
 863
 864         REQUIRE(NS_CLIENT_VALID(client));
 865
 866         CTRACE("send");
 867
 868         if ((client->attributes & NS_CLIENTATTR_RA) != 0)
 869                 client->message->flags |= DNS_MESSAGEFLAG_RA;
 870
 871         if ((client->attributes & NS_CLIENTATTR_WANTDNSSEC) != 0)
 872                 dnssec_opts = 0;
 873         else
 874                 dnssec_opts = DNS_MESSAGERENDER_OMITDNSSEC;
 875
 876         preferred_glue = 0;
 877         if (client->view != NULL) {
 878                 if (client->view->preferred_glue == dns_rdatatype_a)
 879                         preferred_glue = DNS_MESSAGERENDER_PREFER_A;
 880                 else if (client->view->preferred_glue == dns_rdatatype_aaaa)
 881                         preferred_glue = DNS_MESSAGERENDER_PREFER_AAAA;
 882         }
 883
 884         /*
 885          * XXXRTH  The following doesn't deal with TCP buffer resizing.
 886          */
 887         result = client_allocsendbuf(client, &buffer, &tcpbuffer, 0,
 888                                      sendbuf, &data);
 889         if (result != ISC_R_SUCCESS)
 890                 goto done;
 891
 892         result = dns_compress_init(&cctx, -1, client->mctx);
 893         if (result != ISC_R_SUCCESS)
 894                 goto done;
 895         cleanup_cctx = ISC_TRUE;
 896
 897         result = dns_message_renderbegin(client->message, &cctx, &buffer);
 898         if (result != ISC_R_SUCCESS)
 899                 goto done;
 900         if (client->opt != NULL) {
 901                 result = dns_message_setopt(client->message, client->opt);
 902                 /*
 903                  * XXXRTH dns_message_setopt() should probably do this...
 904                  */
 905                 client->opt = NULL;
 906                 if (result != ISC_R_SUCCESS)
 907                         goto done;
 908         }
 909         result = dns_message_rendersection(client->message,
 910                                            DNS_SECTION_QUESTION, 0);
 911         if (result == ISC_R_NOSPACE) {
 912                 client->message->flags |= DNS_MESSAGEFLAG_TC;
 913                 goto renderend;
 914         }
 915         if (result != ISC_R_SUCCESS)
 916                 goto done;
 917         result = dns_message_rendersection(client->message,
 918                                            DNS_SECTION_ANSWER,
 919                                            DNS_MESSAGERENDER_PARTIAL |
 920                                            dnssec_opts);
 921         if (result == ISC_R_NOSPACE) {
 922                 client->message->flags |= DNS_MESSAGEFLAG_TC;
 923                 goto renderend;
 924         }
 925         if (result != ISC_R_SUCCESS)
 926                 goto done;
 927         result = dns_message_rendersection(client->message,
 928                                            DNS_SECTION_AUTHORITY,
 929                                            DNS_MESSAGERENDER_PARTIAL |
 930                                            dnssec_opts);
 931         if (result == ISC_R_NOSPACE) {
 932                 client->message->flags |= DNS_MESSAGEFLAG_TC;
 933                 goto renderend;
 934         }
 935         if (result != ISC_R_SUCCESS)
 936                 goto done;
 937         result = dns_message_rendersection(client->message,
 938                                            DNS_SECTION_ADDITIONAL,
 939                                            preferred_glue | dnssec_opts);
 940         if (result != ISC_R_SUCCESS && result != ISC_R_NOSPACE)
 941                 goto done;
 942  renderend:
 943         result = dns_message_renderend(client->message);
 944
 945         if (result != ISC_R_SUCCESS)
 946                 goto done;
 947
 948         if (cleanup_cctx) {
 949                 dns_compress_invalidate(&cctx);
 950                 cleanup_cctx = ISC_FALSE;
 951         }
 952
 953         if (TCP_CLIENT(client)) {
 954                 isc_buffer_usedregion(&buffer, &r);
 955                 isc_buffer_putuint16(&tcpbuffer, (isc_uint16_t) r.length);
 956                 isc_buffer_add(&tcpbuffer, r.length);
 957                 result = client_sendpkg(client, &tcpbuffer);
 958         } else
 959                 result = client_sendpkg(client, &buffer);
 960         if (result == ISC_R_SUCCESS)
 961                 return;
 962
 963  done:
 964         if (client->tcpbuf != NULL) {
 965                 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
 966                 client->tcpbuf = NULL;
 967         }
 968
 969         if (cleanup_cctx)
 970                 dns_compress_invalidate(&cctx);
 971
 972         ns_client_next(client, result);
 973 }
 974
 975 void
 976 ns_client_error(ns_client_t *client, isc_result_t result) {
 977         dns_rcode_t rcode;
 978         dns_message_t *message;
 979
 980         REQUIRE(NS_CLIENT_VALID(client));
 981
 982         CTRACE("error");
 983
 984         message = client->message;
 985         rcode = dns_result_torcode(result);
 986
 987         /*
 988          * Message may be an in-progress reply that we had trouble
 989          * with, in which case QR will be set.  We need to clear QR before
 990          * calling dns_message_reply() to avoid triggering an assertion.
 991          */
 992         message->flags &= ~DNS_MESSAGEFLAG_QR;
 993         /*
 994          * AA and AD shouldn't be set.
 995          */
 996         message->flags &= ~(DNS_MESSAGEFLAG_AA | DNS_MESSAGEFLAG_AD);
 997         result = dns_message_reply(message, ISC_TRUE);
 998         if (result != ISC_R_SUCCESS) {
 999                 /*
1000                  * It could be that we've got a query with a good header,
1001                  * but a bad question section, so we try again with
1002                  * want_question_section set to ISC_FALSE.
1003                  */
1004                 result = dns_message_reply(message, ISC_FALSE);
1005                 if (result != ISC_R_SUCCESS) {
1006                         ns_client_next(client, result);
1007                         return;
1008                 }
1009         }
1010         message->rcode = rcode;
1011
1012         /*
1013          * FORMERR loop avoidance:  If we sent a FORMERR message
1014          * with the same ID to the same client less than two
1015          * seconds ago, assume that we are in an infinite error
1016          * packet dialog with a server for some protocol whose
1017          * error responses look enough like DNS queries to
1018          * elicit a FORMERR response.  Drop a packet to break
1019          * the loop.
1020          */
1021         if (rcode == dns_rcode_formerr) {
1022                 if (isc_sockaddr_equal(&client->peeraddr,
1023                                        &client->formerrcache.addr) &&
1024                     message->id == client->formerrcache.id &&
1025                     client->requesttime - client->formerrcache.time < 2) {
1026                         /* Drop packet. */
1027                         ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1028                                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1029                                       "possible error packet loop, "
1030                                       "FORMERR dropped");
1031                         ns_client_next(client, result);
1032                         return;
1033                 }
1034                 client->formerrcache.addr = client->peeraddr;
1035                 client->formerrcache.time = client->requesttime;
1036                 client->formerrcache.id = message->id;
1037         }
1038         ns_client_send(client);
1039 }
1040
1041 static inline isc_result_t
1042 client_addopt(ns_client_t *client) {
1043         dns_rdataset_t *rdataset;
1044         dns_rdatalist_t *rdatalist;
1045         dns_rdata_t *rdata;
1046         isc_result_t result;
1047         dns_view_t *view;
1048         dns_resolver_t *resolver;
1049         isc_uint16_t udpsize;
1050
1051         REQUIRE(client->opt == NULL);   /* XXXRTH free old. */
1052
1053         rdatalist = NULL;
1054         result = dns_message_gettemprdatalist(client->message, &rdatalist);
1055         if (result != ISC_R_SUCCESS)
1056                 return (result);
1057         rdata = NULL;
1058         result = dns_message_gettemprdata(client->message, &rdata);
1059         if (result != ISC_R_SUCCESS)
1060                 return (result);
1061         rdataset = NULL;
1062         result = dns_message_gettemprdataset(client->message, &rdataset);
1063         if (result != ISC_R_SUCCESS)
1064                 return (result);
1065         dns_rdataset_init(rdataset);
1066
1067         rdatalist->type = dns_rdatatype_opt;
1068         rdatalist->covers = 0;
1069
1070         /*
1071          * Set the maximum UDP buffer size.
1072          */
1073         view = client->view;
1074         resolver = (view != NULL) ? view->resolver : NULL;
1075         if (resolver != NULL)
1076                 udpsize = dns_resolver_getudpsize(resolver);
1077         else
1078                 udpsize = ns_g_udpsize;
1079         rdatalist->rdclass = udpsize;
1080
1081         /*
1082          * Set EXTENDED-RCODE, VERSION and Z to 0.
1083          */
1084         rdatalist->ttl = (client->extflags & DNS_MESSAGEEXTFLAG_REPLYPRESERVE);
1085
1086         /*
1087          * No ENDS options in the default case.
1088          */
1089         rdata->data = NULL;
1090         rdata->length = 0;
1091         rdata->rdclass = rdatalist->rdclass;
1092         rdata->type = rdatalist->type;
1093         rdata->flags = 0;
1094
1095         ISC_LIST_INIT(rdatalist->rdata);
1096         ISC_LIST_APPEND(rdatalist->rdata, rdata, link);
1097         RUNTIME_CHECK(dns_rdatalist_tordataset(rdatalist, rdataset)
1098                       == ISC_R_SUCCESS);
1099
1100         client->opt = rdataset;
1101
1102         return (ISC_R_SUCCESS);
1103 }
1104
1105 static inline isc_boolean_t
1106 allowed(isc_netaddr_t *addr, dns_name_t *signer, dns_acl_t *acl) {
1107         int match;
1108         isc_result_t result;
1109
1110         if (acl == NULL)
1111                 return (ISC_TRUE);
1112         result = dns_acl_match(addr, signer, acl, &ns_g_server->aclenv,
1113                                &match, NULL);
1114         if (result == ISC_R_SUCCESS && match > 0)
1115                 return (ISC_TRUE);
1116         return (ISC_FALSE);
1117 }
1118
1119 /*
1120  * Handle an incoming request event from the socket (UDP case)
1121  * or tcpmsg (TCP case).
1122  */
1123 static void
1124 client_request(isc_task_t *task, isc_event_t *event) {
1125         ns_client_t *client;
1126         isc_socketevent_t *sevent;
1127         isc_result_t result;
1128         isc_result_t sigresult = ISC_R_SUCCESS;
1129         isc_buffer_t *buffer;
1130         isc_buffer_t tbuffer;
1131         dns_view_t *view;
1132         dns_rdataset_t *opt;
1133         isc_boolean_t ra;       /* Recursion available. */
1134         isc_netaddr_t netaddr;
1135         isc_netaddr_t destaddr;
1136         int match;
1137         dns_messageid_t id;
1138         unsigned int flags;
1139         isc_boolean_t notimp;
1140
1141         REQUIRE(event != NULL);
1142         client = event->ev_arg;
1143         REQUIRE(NS_CLIENT_VALID(client));
1144         REQUIRE(task == client->task);
1145
1146         INSIST(client->recursionquota == NULL);
1147
1148         INSIST(client->state ==
1149                TCP_CLIENT(client) ?
1150                NS_CLIENTSTATE_READING :
1151                NS_CLIENTSTATE_READY);
1152
1153         if (event->ev_type == ISC_SOCKEVENT_RECVDONE) {
1154                 INSIST(!TCP_CLIENT(client));
1155                 sevent = (isc_socketevent_t *)event;
1156                 REQUIRE(sevent == client->recvevent);
1157                 isc_buffer_init(&tbuffer, sevent->region.base, sevent->n);
1158                 isc_buffer_add(&tbuffer, sevent->n);
1159                 buffer = &tbuffer;
1160                 result = sevent->result;
1161                 if (result == ISC_R_SUCCESS) {
1162                         client->peeraddr = sevent->address;
1163                         client->peeraddr_valid = ISC_TRUE;
1164                 }
1165                 if ((sevent->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0) {
1166                         client->attributes |= NS_CLIENTATTR_PKTINFO;
1167                         client->pktinfo = sevent->pktinfo;
1168                 }
1169                 if ((sevent->attributes & ISC_SOCKEVENTATTR_MULTICAST) != 0)
1170                         client->attributes |= NS_CLIENTATTR_MULTICAST;
1171                 client->nrecvs--;
1172         } else {
1173                 INSIST(TCP_CLIENT(client));
1174                 REQUIRE(event->ev_type == DNS_EVENT_TCPMSG);
1175                 REQUIRE(event->ev_sender == &client->tcpmsg);
1176                 buffer = &client->tcpmsg.buffer;
1177                 result = client->tcpmsg.result;
1178                 INSIST(client->nreads == 1);
1179                 /*
1180                  * client->peeraddr was set when the connection was accepted.
1181                  */
1182                 client->nreads--;
1183         }
1184
1185         if (exit_check(client))
1186                 goto cleanup;
1187         client->state = client->newstate = NS_CLIENTSTATE_WORKING;
1188
1189         isc_task_getcurrenttime(task, &client->requesttime);
1190         client->now = client->requesttime;
1191
1192         if (result != ISC_R_SUCCESS) {
1193                 if (TCP_CLIENT(client)) {
1194                         ns_client_next(client, result);
1195                 } else {
1196                         if  (result != ISC_R_CANCELED)
1197                                 isc_log_write(ns_g_lctx, NS_LOGCATEGORY_CLIENT,
1198                                               NS_LOGMODULE_CLIENT,
1199                                               ISC_LOG_ERROR,
1200                                               "UDP client handler shutting "
1201                                               "down due to fatal receive "
1202                                               "error: %s",
1203                                               isc_result_totext(result));
1204                         isc_task_shutdown(client->task);
1205                 }
1206                 goto cleanup;
1207         }
1208
1209         isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
1210
1211         ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1212                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1213                       "%s request",
1214                       TCP_CLIENT(client) ? "TCP" : "UDP");
1215
1216         /*
1217          * Check the blackhole ACL for UDP only, since TCP is done in
1218          * client_newconn.
1219          */
1220         if (!TCP_CLIENT(client)) {
1221
1222                 if (ns_g_server->blackholeacl != NULL &&
1223                     dns_acl_match(&netaddr, NULL, ns_g_server->blackholeacl,
1224                                   &ns_g_server->aclenv,
1225                                   &match, NULL) == ISC_R_SUCCESS &&
1226                     match > 0)
1227                 {
1228                         ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1229                                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1230                                       "blackholed UDP datagram");
1231                         ns_client_next(client, ISC_R_SUCCESS);
1232                         goto cleanup;
1233                 }
1234         }
1235
1236         /*
1237          * Silently drop multicast requests for the present.
1238          * XXXMPA look at when/if mDNS spec stabilizes.
1239          */
1240         if ((client->attributes & NS_CLIENTATTR_MULTICAST) != 0) {
1241                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1242                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(2),
1243                               "dropping multicast request");
1244                 ns_client_next(client, DNS_R_REFUSED);
1245         }
1246
1247         result = dns_message_peekheader(buffer, &id, &flags);
1248         if (result != ISC_R_SUCCESS) {
1249                 /*
1250                  * There isn't enough header to determine whether
1251                  * this was a request or a response.  Drop it.
1252                  */
1253                 ns_client_next(client, result);
1254                 goto cleanup;
1255         }
1256
1257         /*
1258          * The client object handles requests, not responses.
1259          * If this is a UDP response, forward it to the dispatcher.
1260          * If it's a TCP response, discard it here.
1261          */
1262         if ((flags & DNS_MESSAGEFLAG_QR) != 0) {
1263                 if (TCP_CLIENT(client)) {
1264                         CTRACE("unexpected response");
1265                         ns_client_next(client, DNS_R_FORMERR);
1266                         goto cleanup;
1267                 } else {
1268                         dns_dispatch_importrecv(client->dispatch, event);
1269                         ns_client_next(client, ISC_R_SUCCESS);
1270                         goto cleanup;
1271                 }
1272         }
1273
1274         /*
1275          * It's a request.  Parse it.
1276          */
1277         result = dns_message_parse(client->message, buffer, 0);
1278         if (result != ISC_R_SUCCESS) {
1279                 /*
1280                  * Parsing the request failed.  Send a response
1281                  * (typically FORMERR or SERVFAIL).
1282                  */
1283                 ns_client_error(client, result);
1284                 goto cleanup;
1285         }
1286
1287         switch (client->message->opcode) {
1288         case dns_opcode_query:
1289         case dns_opcode_update:
1290         case dns_opcode_notify:
1291                 notimp = ISC_FALSE;
1292                 break;
1293         case dns_opcode_iquery:
1294         default:
1295                 notimp = ISC_TRUE;
1296                 break;
1297         }
1298
1299         client->message->rcode = dns_rcode_noerror;
1300
1301         /* RFC1123 section 6.1.3.2 */
1302         if ((client->attributes & NS_CLIENTATTR_MULTICAST) != 0)
1303                 client->message->flags &= ~DNS_MESSAGEFLAG_RD;
1304
1305         /*
1306          * Deal with EDNS.
1307          */
1308         opt = dns_message_getopt(client->message);
1309         if (opt != NULL) {
1310                 unsigned int version;
1311
1312                 /*
1313                  * Set the client's UDP buffer size.
1314                  */
1315                 client->udpsize = opt->rdclass;
1316
1317                 /*
1318                  * If the requested UDP buffer size is less than 512,
1319                  * ignore it and use 512.
1320                  */
1321                 if (client->udpsize < 512)
1322                         client->udpsize = 512;
1323
1324                 /*
1325                  * Get the flags out of the OPT record.
1326                  */
1327                 client->extflags = (isc_uint16_t)(opt->ttl & 0xFFFF);
1328
1329                 /*
1330                  * Create an OPT for our reply.
1331                  */
1332                 result = client_addopt(client);
1333                 if (result != ISC_R_SUCCESS) {
1334                         ns_client_error(client, result);
1335                         goto cleanup;
1336                 }
1337
1338                 /*
1339                  * Do we understand this version of ENDS?
1340                  *
1341                  * XXXRTH need library support for this!
1342                  */
1343                 version = (opt->ttl & 0x00FF0000) >> 16;
1344                 if (version != 0) {
1345                         ns_client_error(client, DNS_R_BADVERS);
1346                         goto cleanup;
1347                 }
1348         }
1349
1350         if (client->message->rdclass == 0) {
1351                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1352                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1353                               "message class could not be determined");
1354                 ns_client_dumpmessage(client,
1355                                       "message class could not be determined");
1356                 ns_client_error(client, notimp ? DNS_R_NOTIMP : DNS_R_FORMERR);
1357                 goto cleanup;
1358         }
1359
1360         /*
1361          * Determine the destination address.  If the receiving interface is
1362          * bound to a specific address, we simply use it regardless of the
1363          * address family.  All IPv4 queries should fall into this case.
1364          * Otherwise, if this is a TCP query, get the address from the
1365          * receiving socket (this needs a system call and can be heavy).
1366          * For IPv6 UDP queries, we get this from the pktinfo structure (if
1367          * supported).
1368          * If all the attempts fail (this can happen due to memory shortage,
1369          * etc), we regard this as an error for safety.
1370          */
1371         if ((client->interface->flags & NS_INTERFACEFLAG_ANYADDR) == 0)
1372                 isc_netaddr_fromsockaddr(&destaddr, &client->interface->addr);
1373         else {
1374                 result = ISC_R_FAILURE;
1375
1376                 if (TCP_CLIENT(client)) {
1377                         isc_sockaddr_t destsockaddr;
1378
1379                         result = isc_socket_getsockname(client->tcpsocket,
1380                                                         &destsockaddr);
1381                         if (result == ISC_R_SUCCESS)
1382                                 isc_netaddr_fromsockaddr(&destaddr,
1383                                                          &destsockaddr);
1384                 }
1385                 if (result != ISC_R_SUCCESS &&
1386                     client->interface->addr.type.sa.sa_family == AF_INET6 &&
1387                     (client->attributes & NS_CLIENTATTR_PKTINFO) != 0) {
1388                         isc_uint32_t zone = 0;
1389
1390                         /*
1391                          * XXXJT technically, we should convert the receiving
1392                          * interface ID to a proper scope zone ID.  However,
1393                          * due to the fact there is no standard API for this,
1394                          * we only handle link-local addresses and use the
1395                          * interface index as link ID.  Despite the assumption,
1396                          * it should cover most typical cases.
1397                          */
1398                         if (IN6_IS_ADDR_LINKLOCAL(&client->pktinfo.ipi6_addr))
1399                                 zone = (isc_uint32_t)client->pktinfo.ipi6_ifindex;
1400
1401                         isc_netaddr_fromin6(&destaddr,
1402                                             &client->pktinfo.ipi6_addr);
1403                         isc_netaddr_setzone(&destaddr, zone);
1404                         result = ISC_R_SUCCESS;
1405                 }
1406                 if (result != ISC_R_SUCCESS) {
1407                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1408                                          "failed to get request's "
1409                                          "destination: %s",
1410                                          isc_result_totext(result));
1411                         goto cleanup;
1412                 }
1413         }
1414
1415         /*
1416          * Find a view that matches the client's source address.
1417          */
1418         for (view = ISC_LIST_HEAD(ns_g_server->viewlist);
1419              view != NULL;
1420              view = ISC_LIST_NEXT(view, link)) {
1421                 if (client->message->rdclass == view->rdclass ||
1422                     client->message->rdclass == dns_rdataclass_any)
1423                 {
1424                         dns_name_t *tsig = NULL;
1425                         sigresult = dns_message_rechecksig(client->message,
1426                                                            view);
1427                         if (sigresult == ISC_R_SUCCESS)
1428                                 tsig = client->message->tsigname;
1429
1430                         if (allowed(&netaddr, tsig, view->matchclients) &&
1431                             allowed(&destaddr, tsig, view->matchdestinations) &&
1432                             !((client->message->flags & DNS_MESSAGEFLAG_RD)
1433                               == 0 && view->matchrecursiveonly))
1434                         {
1435                                 dns_view_attach(view, &client->view);
1436                                 break;
1437                         }
1438                 }
1439         }
1440
1441         if (view == NULL) {
1442                 char classname[DNS_RDATACLASS_FORMATSIZE];
1443
1444                 /*
1445                  * Do a dummy TSIG verification attempt so that the
1446                  * response will have a TSIG if the query did, as
1447                  * required by RFC2845.
1448                  */
1449                 isc_buffer_t b;
1450                 isc_region_t *r;
1451
1452                 dns_message_resetsig(client->message);
1453
1454                 r = dns_message_getrawmessage(client->message);
1455                 isc_buffer_init(&b, r->base, r->length);
1456                 isc_buffer_add(&b, r->length);
1457                 (void)dns_tsig_verify(&b, client->message, NULL, NULL);
1458
1459                 dns_rdataclass_format(client->message->rdclass, classname,
1460                                       sizeof(classname));
1461                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1462                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1463                               "no matching view in class '%s'", classname);
1464                 ns_client_dumpmessage(client, "no matching view in class");
1465                 ns_client_error(client, notimp ? DNS_R_NOTIMP : DNS_R_REFUSED);
1466                 goto cleanup;
1467         }
1468
1469         ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1470                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(5),
1471                       "using view '%s'", view->name);
1472
1473         /*
1474          * Check for a signature.  We log bad signatures regardless of
1475          * whether they ultimately cause the request to be rejected or
1476          * not.  We do not log the lack of a signature unless we are
1477          * debugging.
1478          */
1479         client->signer = NULL;
1480         dns_name_init(&client->signername, NULL);
1481         result = dns_message_signer(client->message, &client->signername);
1482         if (result == ISC_R_SUCCESS) {
1483                 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1484                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1485                               "request has valid signature");
1486                 client->signer = &client->signername;
1487         } else if (result == ISC_R_NOTFOUND) {
1488                 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1489                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1490                               "request is not signed");
1491         } else if (result == DNS_R_NOIDENTITY) {
1492                 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1493                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1494                               "request is signed by a nonauthoritative key");
1495         } else {
1496                 char tsigrcode[64];
1497                 isc_buffer_t b;
1498                 dns_name_t *name = NULL;
1499
1500                 isc_buffer_init(&b, tsigrcode, sizeof(tsigrcode) - 1);
1501                 RUNTIME_CHECK(dns_tsigrcode_totext(client->message->tsigstatus,
1502                                                    &b) == ISC_R_SUCCESS);
1503                 tsigrcode[isc_buffer_usedlength(&b)] = '\0';
1504                 /* There is a signature, but it is bad. */
1505                 if (dns_message_gettsig(client->message, &name) != NULL) {
1506                         char namebuf[DNS_NAME_FORMATSIZE];
1507                         dns_name_format(name, namebuf, sizeof(namebuf));
1508                         ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1509                                       NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
1510                                       "request has invalid signature: "
1511                                       "TSIG %s: %s (%s)", namebuf,
1512                                       isc_result_totext(result), tsigrcode);
1513                 } else {
1514                         ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1515                                       NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
1516                                       "request has invalid signature: %s (%s)",
1517                                       isc_result_totext(result), tsigrcode);
1518                 }
1519                 /*
1520                  * Accept update messages signed by unknown keys so that
1521                  * update forwarding works transparently through slaves
1522                  * that don't have all the same keys as the master.
1523                  */
1524                 if (!(client->message->tsigstatus == dns_tsigerror_badkey &&
1525                       client->message->opcode == dns_opcode_update)) {
1526                         ns_client_error(client, sigresult);
1527                         goto cleanup;
1528                 }
1529         }
1530
1531         /*
1532          * Decide whether recursive service is available to this client.
1533          * We do this here rather than in the query code so that we can
1534          * set the RA bit correctly on all kinds of responses, not just
1535          * responses to ordinary queries.
1536          */
1537         ra = ISC_FALSE;
1538         if (client->view->resolver != NULL &&
1539             client->view->recursion == ISC_TRUE &&
1540             ns_client_checkaclsilent(client, client->view->recursionacl,
1541                                      ISC_TRUE) == ISC_R_SUCCESS)
1542                 ra = ISC_TRUE;
1543
1544         if (ra == ISC_TRUE)
1545                 client->attributes |= NS_CLIENTATTR_RA;
1546
1547         ns_client_log(client, DNS_LOGCATEGORY_SECURITY, NS_LOGMODULE_CLIENT,
1548                       ISC_LOG_DEBUG(3), ra ? "recursion available" :
1549                                              "recursion not available");
1550
1551         /*
1552          * Dispatch the request.
1553          */
1554         switch (client->message->opcode) {
1555         case dns_opcode_query:
1556                 CTRACE("query");
1557                 ns_query_start(client);
1558                 break;
1559         case dns_opcode_update:
1560                 CTRACE("update");
1561                 ns_client_settimeout(client, 60);
1562                 ns_update_start(client, sigresult);
1563                 break;
1564         case dns_opcode_notify:
1565                 CTRACE("notify");
1566                 ns_client_settimeout(client, 60);
1567                 ns_notify_start(client);
1568                 break;
1569         case dns_opcode_iquery:
1570                 CTRACE("iquery");
1571                 ns_client_error(client, DNS_R_NOTIMP);
1572                 break;
1573         default:
1574                 CTRACE("unknown opcode");
1575                 ns_client_error(client, DNS_R_NOTIMP);
1576         }
1577
1578  cleanup:
1579         return;
1580 }
1581
1582 static void
1583 client_timeout(isc_task_t *task, isc_event_t *event) {
1584         ns_client_t *client;
1585
1586         REQUIRE(event != NULL);
1587         REQUIRE(event->ev_type == ISC_TIMEREVENT_LIFE ||
1588                 event->ev_type == ISC_TIMEREVENT_IDLE);
1589         client = event->ev_arg;
1590         REQUIRE(NS_CLIENT_VALID(client));
1591         REQUIRE(task == client->task);
1592         REQUIRE(client->timer != NULL);
1593
1594         UNUSED(task);
1595
1596         CTRACE("timeout");
1597
1598         isc_event_free(&event);
1599
1600         if (client->shutdown != NULL) {
1601                 (client->shutdown)(client->shutdown_arg, ISC_R_TIMEDOUT);
1602                 client->shutdown = NULL;
1603                 client->shutdown_arg = NULL;
1604         }
1605
1606         if (client->newstate > NS_CLIENTSTATE_READY)
1607                 client->newstate = NS_CLIENTSTATE_READY;
1608         (void)exit_check(client);
1609 }
1610
1611 static isc_result_t
1612 client_create(ns_clientmgr_t *manager, ns_client_t **clientp) {
1613         ns_client_t *client;
1614         isc_result_t result;
1615
1616         /*
1617          * Caller must be holding the manager lock.
1618          *
1619          * Note: creating a client does not add the client to the
1620          * manager's client list or set the client's manager pointer.
1621          * The caller is responsible for that.
1622          */
1623
1624         REQUIRE(clientp != NULL && *clientp == NULL);
1625
1626         client = isc_mem_get(manager->mctx, sizeof(*client));
1627         if (client == NULL)
1628                 return (ISC_R_NOMEMORY);
1629
1630         client->task = NULL;
1631         result = isc_task_create(manager->taskmgr, 0, &client->task);
1632         if (result != ISC_R_SUCCESS)
1633                 goto cleanup_client;
1634         isc_task_setname(client->task, "client", client);
1635
1636         client->timer = NULL;
1637         result = isc_timer_create(manager->timermgr, isc_timertype_inactive,
1638                                   NULL, NULL, client->task, client_timeout,
1639                                   client, &client->timer);
1640         if (result != ISC_R_SUCCESS)
1641                 goto cleanup_task;
1642         client->timerset = ISC_FALSE;
1643
1644         client->message = NULL;
1645         result = dns_message_create(manager->mctx, DNS_MESSAGE_INTENTPARSE,
1646                                     &client->message);
1647         if (result != ISC_R_SUCCESS)
1648                 goto cleanup_timer;
1649
1650         /* XXXRTH  Hardwired constants */
1651
1652         client->sendevent = (isc_socketevent_t *)
1653                             isc_event_allocate(manager->mctx, client,
1654                                                ISC_SOCKEVENT_SENDDONE,
1655                                                client_senddone, client,
1656                                                sizeof(isc_socketevent_t));
1657         if (client->sendevent == NULL) {
1658                 result = ISC_R_NOMEMORY;
1659                 goto cleanup_message;
1660         }
1661
1662         client->recvbuf = isc_mem_get(manager->mctx, RECV_BUFFER_SIZE);
1663         if  (client->recvbuf == NULL) {
1664                 result = ISC_R_NOMEMORY;
1665                 goto cleanup_sendevent;
1666         }
1667
1668         client->recvevent = (isc_socketevent_t *)
1669                             isc_event_allocate(manager->mctx, client,
1670                                                ISC_SOCKEVENT_RECVDONE,
1671                                                client_request, client,
1672                                                sizeof(isc_socketevent_t));
1673         if (client->recvevent == NULL) {
1674                 result = ISC_R_NOMEMORY;
1675                 goto cleanup_recvbuf;
1676         }
1677
1678         client->magic = NS_CLIENT_MAGIC;
1679         client->mctx = manager->mctx;
1680         client->manager = NULL;
1681         client->state = NS_CLIENTSTATE_INACTIVE;
1682         client->newstate = NS_CLIENTSTATE_MAX;
1683         client->naccepts = 0;
1684         client->nreads = 0;
1685         client->nsends = 0;
1686         client->nrecvs = 0;
1687         client->nupdates = 0;
1688         client->nctls = 0;
1689         client->references = 0;
1690         client->attributes = 0;
1691         client->view = NULL;
1692         client->dispatch = NULL;
1693         client->udpsocket = NULL;
1694         client->tcplistener = NULL;
1695         client->tcpsocket = NULL;
1696         client->tcpmsg_valid = ISC_FALSE;
1697         client->tcpbuf = NULL;
1698         client->opt = NULL;
1699         client->udpsize = 512;
1700         client->extflags = 0;
1701         client->next = NULL;
1702         client->shutdown = NULL;
1703         client->shutdown_arg = NULL;
1704         dns_name_init(&client->signername, NULL);
1705         client->mortal = ISC_FALSE;
1706         client->tcpquota = NULL;
1707         client->recursionquota = NULL;
1708         client->interface = NULL;
1709         client->peeraddr_valid = ISC_FALSE;
1710         ISC_EVENT_INIT(&client->ctlevent, sizeof(client->ctlevent), 0, NULL,
1711                        NS_EVENT_CLIENTCONTROL, client_start, client, client,
1712                        NULL, NULL);
1713         /*
1714          * Initialize FORMERR cache to sentinel value that will not match
1715          * any actual FORMERR response.
1716          */
1717         isc_sockaddr_any(&client->formerrcache.addr);
1718         client->formerrcache.time = 0;
1719         client->formerrcache.id = 0;
1720         ISC_LINK_INIT(client, link);
1721         client->list = NULL;
1722
1723         /*
1724          * We call the init routines for the various kinds of client here,
1725          * after we have created an otherwise valid client, because some
1726          * of them call routines that REQUIRE(NS_CLIENT_VALID(client)).
1727          */
1728         result = ns_query_init(client);
1729         if (result != ISC_R_SUCCESS)
1730                 goto cleanup_recvevent;
1731
1732         result = isc_task_onshutdown(client->task, client_shutdown, client);
1733         if (result != ISC_R_SUCCESS)
1734                 goto cleanup_query;
1735
1736         CTRACE("create");
1737
1738         *clientp = client;
1739
1740         return (ISC_R_SUCCESS);
1741
1742  cleanup_query:
1743         ns_query_free(client);
1744
1745  cleanup_recvevent:
1746         isc_event_free((isc_event_t **)&client->recvevent);
1747
1748  cleanup_recvbuf:
1749         isc_mem_put(manager->mctx, client->recvbuf, RECV_BUFFER_SIZE);
1750
1751  cleanup_sendevent:
1752         isc_event_free((isc_event_t **)&client->sendevent);
1753
1754         client->magic = 0;
1755
1756  cleanup_message:
1757         dns_message_destroy(&client->message);
1758
1759  cleanup_timer:
1760         isc_timer_detach(&client->timer);
1761
1762  cleanup_task:
1763         isc_task_detach(&client->task);
1764
1765  cleanup_client:
1766         isc_mem_put(manager->mctx, client, sizeof(*client));
1767
1768         return (result);
1769 }
1770
1771 static void
1772 client_read(ns_client_t *client) {
1773         isc_result_t result;
1774
1775         CTRACE("read");
1776
1777         result = dns_tcpmsg_readmessage(&client->tcpmsg, client->task,
1778                                         client_request, client);
1779         if (result != ISC_R_SUCCESS)
1780                 goto fail;
1781
1782         /*
1783          * Set a timeout to limit the amount of time we will wait
1784          * for a request on this TCP connection.
1785          */
1786         ns_client_settimeout(client, 30);
1787
1788         client->state = client->newstate = NS_CLIENTSTATE_READING;
1789         INSIST(client->nreads == 0);
1790         INSIST(client->recursionquota == NULL);
1791         client->nreads++;
1792
1793         return;
1794  fail:
1795         ns_client_next(client, result);
1796 }
1797
1798 static void
1799 client_newconn(isc_task_t *task, isc_event_t *event) {
1800         ns_client_t *client = event->ev_arg;
1801         isc_socket_newconnev_t *nevent = (isc_socket_newconnev_t *)event;
1802         isc_result_t result;
1803
1804         REQUIRE(event->ev_type == ISC_SOCKEVENT_NEWCONN);
1805         REQUIRE(NS_CLIENT_VALID(client));
1806         REQUIRE(client->task == task);
1807
1808         UNUSED(task);
1809
1810         INSIST(client->state == NS_CLIENTSTATE_READY);
1811
1812         INSIST(client->naccepts == 1);
1813         client->naccepts--;
1814
1815         LOCK(&client->interface->lock);
1816         INSIST(client->interface->ntcpcurrent > 0);
1817         client->interface->ntcpcurrent--;
1818         UNLOCK(&client->interface->lock);
1819
1820         /*
1821          * We must take ownership of the new socket before the exit
1822          * check to make sure it gets destroyed if we decide to exit.
1823          */
1824         if (nevent->result == ISC_R_SUCCESS) {
1825                 client->tcpsocket = nevent->newsocket;
1826                 client->state = NS_CLIENTSTATE_READING;
1827                 INSIST(client->recursionquota == NULL);
1828
1829                 (void)isc_socket_getpeername(client->tcpsocket,
1830                                              &client->peeraddr);
1831                 client->peeraddr_valid = ISC_TRUE;
1832                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1833                            NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1834                            "new TCP connection");
1835         } else {
1836                 /*
1837                  * XXXRTH  What should we do?  We're trying to accept but
1838                  *         it didn't work.  If we just give up, then TCP
1839                  *         service may eventually stop.
1840                  *
1841                  *         For now, we just go idle.
1842                  *
1843                  *         Going idle is probably the right thing if the
1844                  *         I/O was canceled.
1845                  */
1846                 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1847                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1848                               "accept failed: %s",
1849                               isc_result_totext(nevent->result));
1850         }
1851
1852         if (exit_check(client))
1853                 goto freeevent;
1854
1855         if (nevent->result == ISC_R_SUCCESS) {
1856                 int match;
1857                 isc_netaddr_t netaddr;
1858
1859                 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
1860
1861                 if (ns_g_server->blackholeacl != NULL &&
1862                     dns_acl_match(&netaddr, NULL,
1863                                   ns_g_server->blackholeacl,
1864                                   &ns_g_server->aclenv,
1865                                   &match, NULL) == ISC_R_SUCCESS &&
1866                     match > 0)
1867                 {
1868                         ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1869                                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1870                                       "blackholed connection attempt");
1871                         client->newstate = NS_CLIENTSTATE_READY;
1872                         (void)exit_check(client);
1873                         goto freeevent;
1874                 }
1875
1876                 INSIST(client->tcpmsg_valid == ISC_FALSE);
1877                 dns_tcpmsg_init(client->mctx, client->tcpsocket,
1878                                 &client->tcpmsg);
1879                 client->tcpmsg_valid = ISC_TRUE;
1880
1881                 /*
1882                  * Let a new client take our place immediately, before
1883                  * we wait for a request packet.  If we don't,
1884                  * telnetting to port 53 (once per CPU) will
1885                  * deny service to legititmate TCP clients.
1886                  */
1887                 result = isc_quota_attach(&ns_g_server->tcpquota,
1888                                           &client->tcpquota);
1889                 if (result == ISC_R_SUCCESS)
1890                         result = ns_client_replace(client);
1891                 if (result != ISC_R_SUCCESS) {
1892                         ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1893                                       NS_LOGMODULE_CLIENT, ISC_LOG_WARNING,
1894                                       "no more TCP clients: %s",
1895                                       isc_result_totext(result));
1896                 }
1897
1898                 client_read(client);
1899         }
1900
1901  freeevent:
1902         isc_event_free(&event);
1903 }
1904
1905 static void
1906 client_accept(ns_client_t *client) {
1907         isc_result_t result;
1908
1909         CTRACE("accept");
1910
1911         result = isc_socket_accept(client->tcplistener, client->task,
1912                                    client_newconn, client);
1913         if (result != ISC_R_SUCCESS) {
1914                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1915                                  "isc_socket_accept() failed: %s",
1916                                  isc_result_totext(result));
1917                 /*
1918                  * XXXRTH  What should we do?  We're trying to accept but
1919                  *         it didn't work.  If we just give up, then TCP
1920                  *         service may eventually stop.
1921                  *
1922                  *         For now, we just go idle.
1923                  */
1924                 return;
1925         }
1926         INSIST(client->naccepts == 0);
1927         client->naccepts++;
1928         LOCK(&client->interface->lock);
1929         client->interface->ntcpcurrent++;
1930         UNLOCK(&client->interface->lock);
1931 }
1932
1933 static void
1934 client_udprecv(ns_client_t *client) {
1935         isc_result_t result;
1936         isc_region_t r;
1937
1938         CTRACE("udprecv");
1939
1940         r.base = client->recvbuf;
1941         r.length = RECV_BUFFER_SIZE;
1942         result = isc_socket_recv2(client->udpsocket, &r, 1,
1943                                   client->task, client->recvevent, 0);
1944         if (result != ISC_R_SUCCESS) {
1945                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1946                                  "isc_socket_recv2() failed: %s",
1947                                  isc_result_totext(result));
1948                 /*
1949                  * This cannot happen in the current implementation, since
1950                  * isc_socket_recv2() cannot fail if flags == 0.
1951                  *
1952                  * If this does fail, we just go idle.
1953                  */
1954                 return;
1955         }
1956         INSIST(client->nrecvs == 0);
1957         client->nrecvs++;
1958 }
1959
1960 void
1961 ns_client_attach(ns_client_t *source, ns_client_t **targetp) {
1962         REQUIRE(NS_CLIENT_VALID(source));
1963         REQUIRE(targetp != NULL && *targetp == NULL);
1964
1965         source->references++;
1966         ns_client_log(source, NS_LOGCATEGORY_CLIENT,
1967                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1968                       "ns_client_attach: ref = %d", source->references);
1969         *targetp = source;
1970 }
1971
1972 void
1973 ns_client_detach(ns_client_t **clientp) {
1974         ns_client_t *client = *clientp;
1975
1976         client->references--;
1977         INSIST(client->references >= 0);
1978         *clientp = NULL;
1979         ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1980                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1981                       "ns_client_detach: ref = %d", client->references);
1982         (void)exit_check(client);
1983 }
1984
1985 isc_boolean_t
1986 ns_client_shuttingdown(ns_client_t *client) {
1987         return (ISC_TF(client->newstate == NS_CLIENTSTATE_FREED));
1988 }
1989
1990 isc_result_t
1991 ns_client_replace(ns_client_t *client) {
1992         isc_result_t result;
1993
1994         CTRACE("replace");
1995
1996         result = ns_clientmgr_createclients(client->manager,
1997                                             1, client->interface,
1998                                             (TCP_CLIENT(client) ?
1999                                              ISC_TRUE : ISC_FALSE));
2000         if (result != ISC_R_SUCCESS)
2001                 return (result);
2002
2003         /*
2004          * The responsibility for listening for new requests is hereby
2005          * transferred to the new client.  Therefore, the old client
2006          * should refrain from listening for any more requests.
2007          */
2008         client->mortal = ISC_TRUE;
2009
2010         return (ISC_R_SUCCESS);
2011 }
2012
2013 /***
2014  *** Client Manager
2015  ***/
2016
2017 static void
2018 clientmgr_destroy(ns_clientmgr_t *manager) {
2019         REQUIRE(ISC_LIST_EMPTY(manager->active));
2020         REQUIRE(ISC_LIST_EMPTY(manager->inactive));
2021         REQUIRE(ISC_LIST_EMPTY(manager->recursing));
2022
2023         MTRACE("clientmgr_destroy");
2024
2025         DESTROYLOCK(&manager->lock);
2026         manager->magic = 0;
2027         isc_mem_put(manager->mctx, manager, sizeof(*manager));
2028 }
2029
2030 isc_result_t
2031 ns_clientmgr_create(isc_mem_t *mctx, isc_taskmgr_t *taskmgr,
2032                     isc_timermgr_t *timermgr, ns_clientmgr_t **managerp)
2033 {
2034         ns_clientmgr_t *manager;
2035         isc_result_t result;
2036
2037         manager = isc_mem_get(mctx, sizeof(*manager));
2038         if (manager == NULL)
2039                 return (ISC_R_NOMEMORY);
2040
2041         result = isc_mutex_init(&manager->lock);
2042         if (result != ISC_R_SUCCESS)
2043                 goto cleanup_manager;
2044
2045         manager->mctx = mctx;
2046         manager->taskmgr = taskmgr;
2047         manager->timermgr = timermgr;
2048         manager->exiting = ISC_FALSE;
2049         ISC_LIST_INIT(manager->active);
2050         ISC_LIST_INIT(manager->inactive);
2051         ISC_LIST_INIT(manager->recursing);
2052         manager->magic = MANAGER_MAGIC;
2053
2054         MTRACE("create");
2055
2056         *managerp = manager;
2057
2058         return (ISC_R_SUCCESS);
2059
2060  cleanup_manager:
2061         isc_mem_put(manager->mctx, manager, sizeof(*manager));
2062
2063         return (result);
2064 }
2065
2066 void
2067 ns_clientmgr_destroy(ns_clientmgr_t **managerp) {
2068         ns_clientmgr_t *manager;
2069         ns_client_t *client;
2070         isc_boolean_t need_destroy = ISC_FALSE;
2071
2072         REQUIRE(managerp != NULL);
2073         manager = *managerp;
2074         REQUIRE(VALID_MANAGER(manager));
2075
2076         MTRACE("destroy");
2077
2078         LOCK(&manager->lock);
2079
2080         manager->exiting = ISC_TRUE;
2081
2082         for (client = ISC_LIST_HEAD(manager->recursing);
2083              client != NULL;
2084              client = ISC_LIST_NEXT(client, link))
2085                 isc_task_shutdown(client->task);
2086
2087         for (client = ISC_LIST_HEAD(manager->active);
2088              client != NULL;
2089              client = ISC_LIST_NEXT(client, link))
2090                 isc_task_shutdown(client->task);
2091
2092         for (client = ISC_LIST_HEAD(manager->inactive);
2093              client != NULL;
2094              client = ISC_LIST_NEXT(client, link))
2095                 isc_task_shutdown(client->task);
2096
2097         if (ISC_LIST_EMPTY(manager->active) &&
2098             ISC_LIST_EMPTY(manager->inactive) &&
2099             ISC_LIST_EMPTY(manager->recursing))
2100                 need_destroy = ISC_TRUE;
2101
2102         UNLOCK(&manager->lock);
2103
2104         if (need_destroy)
2105                 clientmgr_destroy(manager);
2106
2107         *managerp = NULL;
2108 }
2109
2110 isc_result_t
2111 ns_clientmgr_createclients(ns_clientmgr_t *manager, unsigned int n,
2112                            ns_interface_t *ifp, isc_boolean_t tcp)
2113 {
2114         isc_result_t result = ISC_R_SUCCESS;
2115         unsigned int i;
2116         ns_client_t *client;
2117
2118         REQUIRE(VALID_MANAGER(manager));
2119         REQUIRE(n > 0);
2120
2121         MTRACE("createclients");
2122
2123         /*
2124          * We MUST lock the manager lock for the entire client creation
2125          * process.  If we didn't do this, then a client could get a
2126          * shutdown event and disappear out from under us.
2127          */
2128
2129         LOCK(&manager->lock);
2130
2131         for (i = 0; i < n; i++) {
2132                 isc_event_t *ev;
2133                 /*
2134                  * Allocate a client.  First try to get a recycled one;
2135                  * if that fails, make a new one.
2136                  */
2137                 client = ISC_LIST_HEAD(manager->inactive);
2138                 if (client != NULL) {
2139                         MTRACE("recycle");
2140                         ISC_LIST_UNLINK(manager->inactive, client, link);
2141                         client->list = NULL;
2142                 } else {
2143                         MTRACE("create new");
2144                         result = client_create(manager, &client);
2145                         if (result != ISC_R_SUCCESS)
2146                                 break;
2147                 }
2148
2149                 ns_interface_attach(ifp, &client->interface);
2150                 client->state = NS_CLIENTSTATE_READY;
2151                 INSIST(client->recursionquota == NULL);
2152
2153                 if (tcp) {
2154                         client->attributes |= NS_CLIENTATTR_TCP;
2155                         isc_socket_attach(ifp->tcpsocket,
2156                                           &client->tcplistener);
2157                 } else {
2158                         isc_socket_t *sock;
2159
2160                         dns_dispatch_attach(ifp->udpdispatch,
2161                                             &client->dispatch);
2162                         sock = dns_dispatch_getsocket(client->dispatch);
2163                         isc_socket_attach(sock, &client->udpsocket);
2164                 }
2165                 client->manager = manager;
2166                 ISC_LIST_APPEND(manager->active, client, link);
2167                 client->list = &manager->active;
2168
2169                 INSIST(client->nctls == 0);
2170                 client->nctls++;
2171                 ev = &client->ctlevent;
2172                 isc_task_send(client->task, &ev);
2173         }
2174         if (i != 0) {
2175                 /*
2176                  * We managed to create at least one client, so we
2177                  * declare victory.
2178                  */
2179                 result = ISC_R_SUCCESS;
2180         }
2181
2182         UNLOCK(&manager->lock);
2183
2184         return (result);
2185 }
2186
2187 isc_sockaddr_t *
2188 ns_client_getsockaddr(ns_client_t *client) {
2189         return (&client->peeraddr);
2190 }
2191
2192 isc_result_t
2193 ns_client_checkaclsilent(ns_client_t *client, dns_acl_t *acl,
2194                          isc_boolean_t default_allow)
2195 {
2196         isc_result_t result;
2197         int match;
2198         isc_netaddr_t netaddr;
2199
2200         if (acl == NULL) {
2201                 if (default_allow)
2202                         goto allow;
2203                 else
2204                         goto deny;
2205         }
2206
2207         isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
2208
2209         result = dns_acl_match(&netaddr, client->signer, acl,
2210                                &ns_g_server->aclenv,
2211                                &match, NULL);
2212         if (result != ISC_R_SUCCESS)
2213                 goto deny; /* Internal error, already logged. */
2214         if (match > 0)
2215                 goto allow;
2216         goto deny; /* Negative match or no match. */
2217
2218  allow:
2219         return (ISC_R_SUCCESS);
2220
2221  deny:
2222         return (DNS_R_REFUSED);
2223 }
2224
2225 isc_result_t
2226 ns_client_checkacl(ns_client_t *client,
2227                    const char *opname, dns_acl_t *acl,
2228                    isc_boolean_t default_allow, int log_level)
2229 {
2230         isc_result_t result =
2231                 ns_client_checkaclsilent(client, acl, default_allow);
2232
2233         if (result == ISC_R_SUCCESS)
2234                 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2235                               NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2236                               "%s approved", opname);
2237         else
2238                 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2239                               NS_LOGMODULE_CLIENT,
2240                               log_level, "%s denied", opname);
2241         return (result);
2242 }
2243
2244 static void
2245 ns_client_name(ns_client_t *client, char *peerbuf, size_t len) {
2246         if (client->peeraddr_valid)
2247                 isc_sockaddr_format(&client->peeraddr, peerbuf, len);
2248         else
2249                 snprintf(peerbuf, len, "@%p", client);
2250 }
2251
2252 void
2253 ns_client_logv(ns_client_t *client, isc_logcategory_t *category,
2254            isc_logmodule_t *module, int level, const char *fmt, va_list ap)
2255 {
2256         char msgbuf[2048];
2257         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
2258         const char *name = "";
2259         const char *sep = "";
2260
2261         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
2262         ns_client_name(client, peerbuf, sizeof(peerbuf));
2263         if (client->view != NULL && strcmp(client->view->name, "_bind") != 0 &&
2264             strcmp(client->view->name, "_default") != 0) {
2265                 name = client->view->name;
2266                 sep = ": view ";
2267         }
2268
2269         isc_log_write(ns_g_lctx, category, module, level,
2270                       "client %s%s%s: %s", peerbuf, sep, name, msgbuf);
2271 }
2272
2273 void
2274 ns_client_log(ns_client_t *client, isc_logcategory_t *category,
2275            isc_logmodule_t *module, int level, const char *fmt, ...)
2276 {
2277         va_list ap;
2278
2279         if (! isc_log_wouldlog(ns_g_lctx, level))
2280                 return;
2281
2282         va_start(ap, fmt);
2283         ns_client_logv(client, category, module, level, fmt, ap);
2284         va_end(ap);
2285 }
2286
2287 void
2288 ns_client_aclmsg(const char *msg, dns_name_t *name, dns_rdatatype_t type,
2289                  dns_rdataclass_t rdclass, char *buf, size_t len)
2290 {
2291         char namebuf[DNS_NAME_FORMATSIZE];
2292         char typebuf[DNS_RDATATYPE_FORMATSIZE];
2293         char classbuf[DNS_RDATACLASS_FORMATSIZE];
2294
2295         dns_name_format(name, namebuf, sizeof(namebuf));
2296         dns_rdatatype_format(type, typebuf, sizeof(typebuf));
2297         dns_rdataclass_format(rdclass, classbuf, sizeof(classbuf));
2298         (void)snprintf(buf, len, "%s '%s/%s/%s'", msg, namebuf, typebuf,
2299                        classbuf);
2300 }
2301
2302 static void
2303 ns_client_dumpmessage(ns_client_t *client, const char *reason) {
2304         isc_buffer_t buffer;
2305         char *buf = NULL;
2306         int len = 1024;
2307         isc_result_t result;
2308
2309         /*
2310          * Note that these are multiline debug messages.  We want a newline
2311          * to appear in the log after each message.
2312          */
2313
2314         do {
2315                 buf = isc_mem_get(client->mctx, len);
2316                 if (buf == NULL)
2317                         break;
2318                 isc_buffer_init(&buffer, buf, len);
2319                 result = dns_message_totext(client->message,
2320                                             &dns_master_style_debug,
2321                                             0, &buffer);
2322                 if (result == ISC_R_NOSPACE) {
2323                         isc_mem_put(client->mctx, buf, len);
2324                         len += 1024;
2325                 } else if (result == ISC_R_SUCCESS)
2326                         ns_client_log(client, NS_LOGCATEGORY_UNMATCHED,
2327                                       NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
2328                                       "%s\n%.*s", reason,
2329                                        (int)isc_buffer_usedlength(&buffer),
2330                                        buf);
2331         } while (result == ISC_R_NOSPACE);
2332
2333         if (buf != NULL)
2334                 isc_mem_put(client->mctx, buf, len);
2335 }
2336
2337 void
2338 ns_client_dumprecursing(FILE *f, ns_clientmgr_t *manager) {
2339         ns_client_t *client;
2340         char namebuf[DNS_NAME_FORMATSIZE];
2341         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
2342         const char *name;
2343         const char *sep;
2344
2345         REQUIRE(VALID_MANAGER(manager));
2346
2347         LOCK(&manager->lock);
2348         client = ISC_LIST_HEAD(manager->recursing);
2349         while (client != NULL) {
2350                 ns_client_name(client, peerbuf, sizeof(peerbuf));
2351                 if (client->view != NULL &&
2352                     strcmp(client->view->name, "_bind") != 0 &&
2353                     strcmp(client->view->name, "_default") != 0) {
2354                         name = client->view->name;
2355                         sep = ": view ";
2356                 } else {
2357                         name = "";
2358                         sep = "";
2359                 }
2360                 dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
2361                 fprintf(f, "; client %s%s%s: '%s' requesttime %d\n",
2362                         peerbuf, sep, name, namebuf, client->requesttime);
2363                 client = ISC_LIST_NEXT(client, link);
2364         }
2365         UNLOCK(&manager->lock);
2366 }