2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
10 #include <afsconfig.h>
11 #include <afs/param.h>
18 #include <afs/afsutil.h>
19 #include <afs/cellconfig.h>
21 #define UBIK_INTERNALS
26 * This module is responsible for determining when the system has
27 * recovered to the point that it can handle new transactions. It
28 * replays logs, polls to determine the current dbase after a crash,
29 * and distributes the new database to the others.
31 * The sync site associates a version number with each database. It
32 * broadcasts the version associated with its current dbase in every
33 * one of its beacon messages. When the sync site send a dbase to a
34 * server, it also sends the db's version. A non-sync site server can
35 * tell if it has the right dbase version by simply comparing the
36 * version from the beacon message \p uvote_dbVersion with the version
37 * associated with the database \p ubik_dbase->version. The sync site
38 * itself simply has one counter to keep track of all of this (again
39 * \p ubik_dbase->version).
41 * sync site: routine called when the sync site loses its quorum; this
42 * procedure is called "up" from the beacon package. It resyncs the
43 * dbase and nudges the recovery daemon to try to propagate out the
44 * changes. It also resets the recovery daemon's state, since
45 * recovery must potentially find a new dbase to propagate out. This
46 * routine should not do anything with variables used by non-sync site
51 * if this flag is set, then ubik will use only the primary address
52 * (the address specified in the CellServDB) to contact other
53 * ubik servers. Ubik recovery will not try opening connections
54 * to the alternate interface addresses.
56 int ubikPrimaryAddrOnly;
59 urecovery_ResetState(void)
62 #if !defined(AFS_PTHREAD_ENV)
63 /* No corresponding LWP_WaitProcess found anywhere for this -- klm */
64 LWP_NoYieldSignal(&urecovery_state);
72 * routine called when a non-sync site server goes down; restarts recovery
73 * process to send missing server the new db when it comes back up for
74 * non-sync site servers.
76 * \note This routine should not do anything with variables used by non-sync site servers.
79 urecovery_LostServer(struct ubik_server *ts)
81 ubeacon_ReinitServer(ts);
82 #if !defined(AFS_PTHREAD_ENV)
83 /* No corresponding LWP_WaitProcess found anywhere for this -- klm */
84 LWP_NoYieldSignal(&urecovery_state);
90 * return true iff we have a current database (called by both sync
91 * sites and non-sync sites) How do we determine this? If we're the
92 * sync site, we wait until recovery has finished fetching and
93 * re-labelling its dbase (it may still be trying to propagate it out
94 * to everyone else; that's THEIR problem). If we're not the sync
95 * site, then we must have a dbase labelled with the right version,
96 * and we must have a currently-good sync site.
99 urecovery_AllBetter(struct ubik_dbase *adbase, int areadAny)
103 ubik_dprint_25("allbetter checking\n");
108 if (ubik_dbase->version.epoch > 1)
109 rcode = 1; /* Happy with any good version of database */
112 /* Check if we're sync site and we've got the right data */
113 else if (ubeacon_AmSyncSite() && (urecovery_state & UBIK_RECHAVEDB)) {
117 /* next, check if we're aux site, and we've ever been sent the
118 * right data (note that if a dbase update fails, we won't think
119 * that the sync site is still the sync site, 'cause it won't talk
120 * to us until a timeout period has gone by. When we recover, we
121 * leave this clear until we get a new dbase */
122 else if ((uvote_GetSyncSite() && (vcmp(ubik_dbVersion, ubik_dbase->version) == 0))) { /* && order is important */
126 ubik_dprint_25("allbetter: returning %d\n", rcode);
131 * \brief abort all transactions on this database
134 urecovery_AbortAll(struct ubik_dbase *adbase)
136 struct ubik_trans *tt;
137 for (tt = adbase->activeTrans; tt; tt = tt->next) {
144 * \brief this routine aborts the current remote transaction, if any, if the tid is wrong
147 urecovery_CheckTid(struct ubik_tid *atid, int abortalways)
149 if (ubik_currentTrans) {
150 /* there is remote write trans, see if we match, see if this
151 * is a new transaction */
152 if (atid->epoch != ubik_currentTrans->tid.epoch
153 || atid->counter > ubik_currentTrans->tid.counter || abortalways) {
154 /* don't match, abort it */
155 /* If the thread is not waiting for lock - ok to end it */
156 if (ubik_currentTrans->locktype != LOCKWAIT) {
157 udisk_end(ubik_currentTrans);
159 ubik_currentTrans = (struct ubik_trans *)0;
168 * log format is defined here, and implicitly in disk.c
170 * 4 byte opcode, followed by parameters, each 4 bytes long. All integers
171 * are in logged in network standard byte order, in case we want to move logs
172 * from machine-to-machine someday.
174 * Begin transaction: opcode \n
175 * Commit transaction: opcode, version (8 bytes) \n
176 * Truncate file: opcode, file number, length \n
177 * Abort transaction: opcode \n
178 * Write data: opcode, file, position, length, <length> data bytes \n
180 * A very simple routine, it just replays the log. Note that this is a new-value only log, which
181 * implies that no uncommitted data is written to the dbase: one writes data to the log, including
182 * the commit record, then we allow data to be written through to the dbase. In our particular
183 * implementation, once a transaction is done, we write out the pages to the database, so that
184 * our buffer package doesn't have to know about stable and uncommitted data in the memory buffers:
185 * any changed data while there is an uncommitted write transaction can be zapped during an
186 * abort and the remaining dbase on the disk is exactly the right dbase, without having to read
190 ReplayLog(struct ubik_dbase *adbase)
193 afs_int32 code, tpos;
195 afs_int32 len, thisSize, tfile, filePos;
197 afs_int32 syncFile = -1;
198 afs_int32 data[1024];
200 /* read the lock twice, once to see whether we have a transaction to deal
201 * with that committed, (theoretically, we should support more than one
202 * trans in the log at once, but not yet), and once replaying the
206 /* for now, assume that all ops in log pertain to one transaction; see if there's a commit */
209 (*adbase->read) (adbase, LOGFILE, (char *)&opcode, tpos,
211 if (code != sizeof(afs_int32))
213 opcode = ntohl(opcode);
214 if (opcode == LOGNEW) {
215 /* handle begin trans */
216 tpos += sizeof(afs_int32);
217 } else if (opcode == LOGABORT)
219 else if (opcode == LOGEND) {
222 } else if (opcode == LOGTRUNCATE) {
225 (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
226 2 * sizeof(afs_int32));
227 if (code != 2 * sizeof(afs_int32))
228 break; /* premature eof or io error */
229 tpos += 2 * sizeof(afs_int32);
230 } else if (opcode == LOGDATA) {
233 (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
234 3 * sizeof(afs_int32));
235 if (code != 3 * sizeof(afs_int32))
237 /* otherwise, skip over the data bytes, too */
238 tpos += ntohl(buffer[2]) + 3 * sizeof(afs_int32);
240 ubik_print("corrupt log opcode (%d) at position %d\n", opcode,
242 break; /* corrupt log! */
246 /* actually do the replay; log should go all the way through the commit record, since
247 * we just read it above. */
253 (*adbase->read) (adbase, LOGFILE, (char *)&opcode, tpos,
255 if (code != sizeof(afs_int32))
257 opcode = ntohl(opcode);
258 if (opcode == LOGNEW) {
259 /* handle begin trans */
260 tpos += sizeof(afs_int32);
261 } else if (opcode == LOGABORT)
262 panic("log abort\n");
263 else if (opcode == LOGEND) {
264 struct ubik_version version;
267 (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
268 2 * sizeof(afs_int32));
269 if (code != 2 * sizeof(afs_int32))
271 version.epoch = ntohl(buffer[0]);
272 version.counter = ntohl(buffer[1]);
273 code = (*adbase->setlabel) (adbase, 0, &version);
276 ubik_print("Successfully replayed log for interrupted "
277 "transaction; db version is now %ld.%ld\n",
278 (long) version.epoch, (long) version.counter);
280 break; /* all done now */
281 } else if (opcode == LOGTRUNCATE) {
284 (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
285 2 * sizeof(afs_int32));
286 if (code != 2 * sizeof(afs_int32))
287 break; /* premature eof or io error */
288 tpos += 2 * sizeof(afs_int32);
290 (*adbase->truncate) (adbase, ntohl(buffer[0]),
294 } else if (opcode == LOGDATA) {
297 (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
298 3 * sizeof(afs_int32));
299 if (code != 3 * sizeof(afs_int32))
301 tpos += 3 * sizeof(afs_int32);
302 /* otherwise, skip over the data bytes, too */
303 len = ntohl(buffer[2]); /* total number of bytes to copy */
304 filePos = ntohl(buffer[1]);
305 tfile = ntohl(buffer[0]);
306 /* try to minimize file syncs */
307 if (syncFile != tfile) {
309 code = (*adbase->sync) (adbase, syncFile);
317 thisSize = (len > sizeof(data) ? sizeof(data) : len);
318 /* copy sizeof(data) buffer bytes at a time */
320 (*adbase->read) (adbase, LOGFILE, (char *)data, tpos,
322 if (code != thisSize)
325 (*adbase->write) (adbase, tfile, (char *)data, filePos,
327 if (code != thisSize)
334 ubik_print("corrupt log opcode (%d) at position %d\n",
336 break; /* corrupt log! */
341 code = (*adbase->sync) (adbase, syncFile);
345 ubik_print("Log read error on pass 2\n");
350 /* now truncate the log, we're done with it */
351 code = (*adbase->truncate) (adbase, LOGFILE, 0);
356 * Called at initialization to figure out version of the dbase we really have.
358 * This routine is called after replaying the log; it reads the restored labels.
361 InitializeDB(struct ubik_dbase *adbase)
365 code = (*adbase->getlabel) (adbase, 0, &adbase->version);
367 /* try setting the label to a new value */
368 adbase->version.epoch = 1; /* value for newly-initialized db */
369 adbase->version.counter = 1;
370 code = (*adbase->setlabel) (adbase, 0, &adbase->version);
372 /* failed, try to set it back */
373 adbase->version.epoch = 0;
374 adbase->version.counter = 0;
375 (*adbase->setlabel) (adbase, 0, &adbase->version);
377 #ifdef AFS_PTHREAD_ENV
378 CV_BROADCAST(&adbase->version_cond);
380 LWP_NoYieldSignal(&adbase->version);
387 * \brief initialize the local ubik_dbase
389 * We replay the logs and then read the resulting file to figure out what version we've really got.
392 urecovery_Initialize(struct ubik_dbase *adbase)
396 code = ReplayLog(adbase);
399 code = InitializeDB(adbase);
404 * \brief Main interaction loop for the recovery manager
406 * The recovery light-weight process only runs when you're the
407 * synchronization site. It performs the following tasks, if and only
408 * if the prerequisite tasks have been performed successfully (it
409 * keeps track of which ones have been performed in its bit map,
410 * \p urecovery_state).
412 * First, it is responsible for probing that all servers are up. This
413 * is the only operation that must be performed even if this is not
414 * yet the sync site, since otherwise this site may not notice that
415 * enough other machines are running to even elect this guy to be the
418 * After that, the recovery process does nothing until the beacon and
419 * voting modules manage to get this site elected sync site.
421 * After becoming sync site, recovery first attempts to find the best
422 * database available in the network (it must do this in order to
423 * ensure finding the latest committed data). After finding the right
424 * database, it must fetch this dbase to the sync site.
426 * After fetching the dbase, it relabels it with a new version number,
427 * to ensure that everyone recognizes this dbase as the most recent
430 * One the dbase has been relabelled, this machine can start handling
431 * requests. However, the recovery module still has one more task:
432 * propagating the dbase out to everyone who is up in the network.
435 urecovery_Interact(void *dummy)
437 afs_int32 code, tcode;
438 struct ubik_server *bestServer = NULL;
439 struct ubik_server *ts;
440 int dbok, doingRPC, now;
441 afs_int32 lastProbeTime;
442 /* if we're the sync site, the best db version we've found yet */
443 static struct ubik_version bestDBVersion;
444 struct ubik_version tversion;
446 int length, tlen, offset, file, nbytes;
447 struct rx_call *rxcall;
449 struct ubik_stat ubikstat;
450 struct in_addr inAddr;
456 /* otherwise, begin interaction */
460 /* Run through this loop every 4 seconds */
463 #ifdef AFS_PTHREAD_ENV
464 select(0, 0, 0, 0, &tv);
466 IOMGR_Select(0, 0, 0, 0, &tv);
469 ubik_dprint("recovery running in state %x\n", urecovery_state);
471 /* Every 30 seconds, check all the down servers and mark them
472 * as up if they respond. When a server comes up or found to
473 * not be current, then re-find the the best database and
476 if ((now = FT_ApproxTime()) > 30 + lastProbeTime) {
478 #ifdef AFS_PTHREAD_ENV
482 for (ts = ubik_servers, doingRPC = 0; ts; ts = ts->next) {
488 urecovery_state &= ~UBIK_RECFOUNDDB;
490 } else if (!ts->currentDB) {
491 urecovery_state &= ~UBIK_RECFOUNDDB;
495 #ifdef AFS_PTHREAD_ENV
500 now = FT_ApproxTime();
504 /* Mark whether we are the sync site */
505 if (!ubeacon_AmSyncSite()) {
506 urecovery_state &= ~UBIK_RECSYNCSITE;
507 continue; /* nothing to do */
509 urecovery_state |= UBIK_RECSYNCSITE;
511 /* If a server has just come up or if we have not found the
512 * most current database, then go find the most current db.
514 if (!(urecovery_state & UBIK_RECFOUNDDB)) {
515 bestServer = (struct ubik_server *)0;
516 bestDBVersion.epoch = 0;
517 bestDBVersion.counter = 0;
518 for (ts = ubik_servers; ts; ts = ts->next) {
520 continue; /* don't bother with these guys */
523 code = DISK_GetVersion(ts->disk_rxcid, &ts->version);
525 /* perhaps this is the best version */
526 if (vcmp(ts->version, bestDBVersion) > 0) {
527 /* new best version */
528 bestDBVersion = ts->version;
533 /* take into consideration our version. Remember if we,
534 * the sync site, have the best version. Also note that
535 * we may need to send the best version out.
537 if (vcmp(ubik_dbase->version, bestDBVersion) >= 0) {
538 bestDBVersion = ubik_dbase->version;
539 bestServer = (struct ubik_server *)0;
540 urecovery_state |= UBIK_RECHAVEDB;
542 /* Clear the flag only when we know we have to retrieve
543 * the db. Because urecovery_AllBetter() looks at it.
545 urecovery_state &= ~UBIK_RECHAVEDB;
547 urecovery_state |= UBIK_RECFOUNDDB;
548 urecovery_state &= ~UBIK_RECSENTDB;
550 if (!(urecovery_state & UBIK_RECFOUNDDB))
551 continue; /* not ready */
553 /* If we, the sync site, do not have the best db version, then
554 * go and get it from the server that does.
556 if ((urecovery_state & UBIK_RECHAVEDB) || !bestServer) {
557 urecovery_state |= UBIK_RECHAVEDB;
559 /* we don't have the best version; we should fetch it. */
561 urecovery_AbortAll(ubik_dbase);
563 /* Rx code to do the Bulk fetch */
566 rxcall = rx_NewCall(bestServer->disk_rxcid);
568 ubik_print("Ubik: Synchronize database with server %s\n",
569 afs_inet_ntoa_r(bestServer->addr[0], hoststr));
571 code = StartDISK_GetFile(rxcall, file);
573 ubik_dprint("StartDiskGetFile failed=%d\n", code);
576 nbytes = rx_Read(rxcall, (char *)&length, sizeof(afs_int32));
577 length = ntohl(length);
578 if (nbytes != sizeof(afs_int32)) {
579 ubik_dprint("Rx-read length error=%d\n", code = BULK_ERROR);
584 /* give invalid label during file transit */
586 code = (*ubik_dbase->setlabel) (ubik_dbase, file, &tversion);
588 ubik_dprint("setlabel io error=%d\n", code);
591 snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
592 ubik_dbase->pathName, (file<0)?"SYS":"",
593 (file<0)?-file:file);
594 fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
599 code = lseek(fd, HDRSIZE, 0);
600 if (code != HDRSIZE) {
607 tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
608 #ifndef AFS_PTHREAD_ENV
612 nbytes = rx_Read(rxcall, tbuffer, tlen);
613 if (nbytes != tlen) {
614 ubik_dprint("Rx-read bulk error=%d\n", code = BULK_ERROR);
619 nbytes = write(fd, tbuffer, tlen);
621 if (nbytes != tlen) {
632 code = EndDISK_GetFile(rxcall, &tversion);
634 tcode = rx_EndCall(rxcall, code);
638 /* we got a new file, set up its header */
639 urecovery_state |= UBIK_RECHAVEDB;
640 memcpy(&ubik_dbase->version, &tversion,
641 sizeof(struct ubik_version));
642 snprintf(tbuffer, sizeof(tbuffer), "%s.DB%s%d",
643 ubik_dbase->pathName, (file<0)?"SYS":"",
644 (file<0)?-file:file);
646 snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
647 ubik_dbase->pathName, (file<0)?"SYS":"",
648 (file<0)?-file:file);
649 code = unlink(pbuffer);
651 code = rename(tbuffer, pbuffer);
652 snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
653 ubik_dbase->pathName, (file<0)?"SYS":"",
654 (file<0)?-file:file);
657 code = rename(pbuffer, tbuffer);
659 (*ubik_dbase->open) (ubik_dbase, file);
660 /* after data is good, sync disk with correct label */
662 (*ubik_dbase->setlabel) (ubik_dbase, 0,
663 &ubik_dbase->version);
666 snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
667 ubik_dbase->pathName, (file<0)?"SYS":"",
668 (file<0)?-file:file);
675 * We will effectively invalidate the old data forever now.
676 * Unclear if we *should* but we do.
678 ubik_dbase->version.epoch = 0;
679 ubik_dbase->version.counter = 0;
680 ubik_print("Ubik: Synchronize database failed (error = %d)\n",
683 ubik_print("Ubik: Synchronize database completed\n");
684 urecovery_state |= UBIK_RECHAVEDB;
686 udisk_Invalidate(ubik_dbase, 0); /* data has changed */
687 #ifdef AFS_PTHREAD_ENV
688 CV_BROADCAST(&ubik_dbase->version_cond);
690 LWP_NoYieldSignal(&ubik_dbase->version);
694 if (!(urecovery_state & UBIK_RECHAVEDB))
695 continue; /* not ready */
697 /* If the database was newly initialized, then when we establish quorum, write
698 * a new label. This allows urecovery_AllBetter() to allow access for reads.
699 * Setting it to 2 also allows another site to come along with a newer
700 * database and overwrite this one.
702 if (ubik_dbase->version.epoch == 1) {
704 urecovery_AbortAll(ubik_dbase);
706 ubik_dbase->version.epoch = ubik_epochTime;
707 ubik_dbase->version.counter = 1;
709 (*ubik_dbase->setlabel) (ubik_dbase, 0, &ubik_dbase->version);
710 udisk_Invalidate(ubik_dbase, 0); /* data may have changed */
711 #ifdef AFS_PTHREAD_ENV
712 CV_BROADCAST(&ubik_dbase->version_cond);
714 LWP_NoYieldSignal(&ubik_dbase->version);
719 /* Check the other sites and send the database to them if they
720 * do not have the current db.
722 if (!(urecovery_state & UBIK_RECSENTDB)) {
723 /* now propagate out new version to everyone else */
724 dbok = 1; /* start off assuming they all worked */
728 * Check if a write transaction is in progress. We can't send the
729 * db when a write is in progress here because the db would be
730 * obsolete as soon as it goes there. Also, ops after the begin
731 * trans would reach the recepient and wouldn't find a transaction
732 * pending there. Frankly, I don't think it's possible to get past
733 * the write-lock above if there is a write transaction in progress,
734 * but then, it won't hurt to check, will it?
736 if (ubik_dbase->flags & DBWRITING) {
741 while ((ubik_dbase->flags & DBWRITING) && (safety < 500)) {
743 /* sleep for a little while */
744 #ifdef AFS_PTHREAD_ENV
745 select(0, 0, 0, 0, &tv);
747 IOMGR_Select(0, 0, 0, 0, &tv);
755 for (ts = ubik_servers; ts; ts = ts->next) {
756 inAddr.s_addr = ts->addr[0];
758 ubik_dprint("recovery cannot send version to %s\n",
759 afs_inet_ntoa_r(inAddr.s_addr, hoststr));
763 ubik_dprint("recovery sending version to %s\n",
764 afs_inet_ntoa_r(inAddr.s_addr, hoststr));
765 if (vcmp(ts->version, ubik_dbase->version) != 0) {
766 ubik_dprint("recovery stating local database\n");
768 /* Rx code to do the Bulk Store */
769 code = (*ubik_dbase->stat) (ubik_dbase, 0, &ubikstat);
771 length = ubikstat.size;
773 rxcall = rx_NewCall(ts->disk_rxcid);
775 StartDISK_SendFile(rxcall, file, length,
776 &ubik_dbase->version);
778 ubik_dprint("StartDiskSendFile failed=%d\n",
785 sizeof(tbuffer) ? sizeof(tbuffer) : length);
787 (*ubik_dbase->read) (ubik_dbase, file,
788 tbuffer, offset, tlen);
789 if (nbytes != tlen) {
790 ubik_dprint("Local disk read error=%d\n",
794 nbytes = rx_Write(rxcall, tbuffer, tlen);
795 if (nbytes != tlen) {
796 ubik_dprint("Rx-write bulk error=%d\n", code =
803 code = EndDISK_SendFile(rxcall);
805 code = rx_EndCall(rxcall, code);
808 /* we set a new file, process its header */
809 ts->version = ubik_dbase->version;
814 /* mark file up to date */
820 urecovery_state |= UBIK_RECSENTDB;
827 * \brief send a Probe to all the network address of this server
829 * \return 0 if success, else return 1
832 DoProbe(struct ubik_server *server)
834 struct rx_connection *conns[UBIK_MAX_INTERFACE_ADDR];
835 struct rx_connection *connSuccess = 0;
836 int i, j, success_i = -1;
840 extern afs_int32 ubikSecIndex;
841 extern struct rx_securityClass *ubikSecClass;
843 for (i = 0; (addr = server->addr[i]) && (i < UBIK_MAX_INTERFACE_ADDR);
846 rx_NewConnection(addr, ubik_callPortal, DISK_SERVICE_ID,
847 ubikSecClass, ubikSecIndex);
849 /* user requirement to use only the primary interface */
850 if (ubikPrimaryAddrOnly) {
855 osi_Assert(i); /* at least one interface address for this server */
857 #ifdef AFS_PTHREAD_ENV
863 if (!multi_error) { /* first success */
870 #ifdef AFS_PTHREAD_ENV
874 if (success_i >= 0) {
875 addr = server->addr[success_i]; /* successful interface addr */
877 if (server->disk_rxcid) /* destroy existing conn */
878 rx_DestroyConnection(server->disk_rxcid);
879 if (server->vote_rxcid)
880 rx_DestroyConnection(server->vote_rxcid);
882 /* make new connections */
883 server->disk_rxcid = conns[success_i];
884 server->vote_rxcid = rx_NewConnection(addr, ubik_callPortal,
885 VOTE_SERVICE_ID, ubikSecClass,
888 connSuccess = conns[success_i];
889 strcpy(buffer, afs_inet_ntoa_r(server->addr[0], hoststr));
891 ubik_print("ubik:server %s is back up: will be contacted through %s\n",
892 buffer, afs_inet_ntoa_r(addr, hoststr));
895 /* Destroy all connections except the one on which we succeeded */
896 for (j = 0; j < i; j++)
897 if (conns[j] != connSuccess)
898 rx_DestroyConnection(conns[j]);
901 ubik_dprint("ubik:server %s still down\n",
902 afs_inet_ntoa_r(server->addr[0], hoststr));
905 return 0; /* success */
907 return 1; /* failure */