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>
15 #include <sys/types.h>
24 #include <netinet/in.h>
31 #include <afs/afsutil.h>
33 #define UBIK_INTERNALS
37 int (*ubik_CheckRXSecurityProc) (void *, struct rx_call *);
38 void *ubik_CheckRXSecurityRock;
40 static void printServerInfo(void);
43 * routines for handling requests remotely-submitted by the sync site. These are
44 * only write transactions (we don't propagate read trans), and there is at most one
45 * write transaction extant at any one time.
48 struct ubik_trans *ubik_currentTrans = 0;
51 ubik_CheckAuth(struct rx_call *acall)
54 if (ubik_CheckRXSecurityProc) {
55 code = (*ubik_CheckRXSecurityProc) (ubik_CheckRXSecurityRock, acall);
61 /* the rest of these guys handle remote execution of write
62 * transactions: this is the code executed on the other servers when a
63 * sync site is executing a write transaction.
66 SDISK_Begin(struct rx_call *rxcall, struct ubik_tid *atid)
70 if ((code = ubik_CheckAuth(rxcall))) {
74 urecovery_CheckTid(atid, 1);
75 code = udisk_begin(ubik_dbase, UBIK_WRITETRANS, &ubik_currentTrans);
76 if (!code && ubik_currentTrans) {
77 /* label this trans with the right trans id */
78 ubik_currentTrans->tid.epoch = atid->epoch;
79 ubik_currentTrans->tid.counter = atid->counter;
87 SDISK_Commit(struct rx_call *rxcall, struct ubik_tid *atid)
90 struct ubik_dbase *dbase;
92 if ((code = ubik_CheckAuth(rxcall))) {
96 if (!ubik_currentTrans) {
100 * sanity check to make sure only write trans appear here
102 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
106 dbase = ubik_currentTrans->dbase;
108 ObtainWriteLock(&dbase->cache_lock);
112 urecovery_CheckTid(atid, 0);
113 if (!ubik_currentTrans) {
115 ReleaseWriteLock(&dbase->cache_lock);
119 code = udisk_commit(ubik_currentTrans);
121 /* sync site should now match */
122 ubik_dbVersion = ubik_dbase->version;
125 ReleaseWriteLock(&dbase->cache_lock);
130 SDISK_ReleaseLocks(struct rx_call *rxcall, struct ubik_tid *atid)
132 struct ubik_dbase *dbase;
135 if ((code = ubik_CheckAuth(rxcall))) {
139 if (!ubik_currentTrans) {
142 /* sanity check to make sure only write trans appear here */
143 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
147 dbase = ubik_currentTrans->dbase;
149 urecovery_CheckTid(atid, 0);
150 if (!ubik_currentTrans) {
155 /* If the thread is not waiting for lock - ok to end it */
156 #if !defined(UBIK_PAUSE)
157 if (ubik_currentTrans->locktype != LOCKWAIT) {
158 #endif /* UBIK_PAUSE */
159 udisk_end(ubik_currentTrans);
160 #if !defined(UBIK_PAUSE)
162 #endif /* UBIK_PAUSE */
163 ubik_currentTrans = (struct ubik_trans *)0;
169 SDISK_Abort(struct rx_call *rxcall, struct ubik_tid *atid)
172 struct ubik_dbase *dbase;
174 if ((code = ubik_CheckAuth(rxcall))) {
178 if (!ubik_currentTrans) {
181 /* sanity check to make sure only write trans appear here */
182 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
186 dbase = ubik_currentTrans->dbase;
188 urecovery_CheckTid(atid, 0);
189 if (!ubik_currentTrans) {
194 code = udisk_abort(ubik_currentTrans);
195 /* If the thread is not waiting for lock - ok to end it */
196 #if !defined(UBIK_PAUSE)
197 if (ubik_currentTrans->locktype != LOCKWAIT) {
198 #endif /* UBIK_PAUSE */
199 udisk_end(ubik_currentTrans);
200 #if !defined(UBIK_PAUSE)
202 #endif /* UBIK_PAUSE */
203 ubik_currentTrans = (struct ubik_trans *)0;
208 /* apos and alen are not used */
210 SDISK_Lock(struct rx_call *rxcall, struct ubik_tid *atid,
211 afs_int32 afile, afs_int32 apos, afs_int32 alen, afs_int32 atype)
214 struct ubik_dbase *dbase;
215 struct ubik_trans *ubik_thisTrans;
217 if ((code = ubik_CheckAuth(rxcall))) {
220 if (!ubik_currentTrans) {
223 /* sanity check to make sure only write trans appear here */
224 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
230 dbase = ubik_currentTrans->dbase;
232 urecovery_CheckTid(atid, 0);
233 if (!ubik_currentTrans) {
238 ubik_thisTrans = ubik_currentTrans;
239 code = ulock_getLock(ubik_currentTrans, atype, 1);
241 /* While waiting, the transaction may have been ended/
242 * aborted from under us (urecovery_CheckTid). In that
243 * case, end the transaction here.
245 if (!code && (ubik_currentTrans != ubik_thisTrans)) {
246 udisk_end(ubik_thisTrans);
255 * \brief Write a vector of data
258 SDISK_WriteV(struct rx_call *rxcall, struct ubik_tid *atid,
259 iovec_wrt *io_vector, iovec_buf *io_buffer)
261 afs_int32 code, i, offset;
262 struct ubik_dbase *dbase;
263 struct ubik_iovec *iovec;
266 if ((code = ubik_CheckAuth(rxcall))) {
269 if (!ubik_currentTrans) {
272 /* sanity check to make sure only write trans appear here */
273 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
277 dbase = ubik_currentTrans->dbase;
279 urecovery_CheckTid(atid, 0);
280 if (!ubik_currentTrans) {
285 iovec = (struct ubik_iovec *)io_vector->iovec_wrt_val;
286 iobuf = (char *)io_buffer->iovec_buf_val;
287 for (i = 0, offset = 0; i < io_vector->iovec_wrt_len; i++) {
288 /* Sanity check for going off end of buffer */
289 if ((offset + iovec[i].length) > io_buffer->iovec_buf_len) {
293 udisk_write(ubik_currentTrans, iovec[i].file, &iobuf[offset],
294 iovec[i].position, iovec[i].length);
299 offset += iovec[i].length;
307 SDISK_Write(struct rx_call *rxcall, struct ubik_tid *atid,
308 afs_int32 afile, afs_int32 apos, bulkdata *adata)
311 struct ubik_dbase *dbase;
313 if ((code = ubik_CheckAuth(rxcall))) {
316 if (!ubik_currentTrans) {
319 /* sanity check to make sure only write trans appear here */
320 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
324 dbase = ubik_currentTrans->dbase;
326 urecovery_CheckTid(atid, 0);
327 if (!ubik_currentTrans) {
332 udisk_write(ubik_currentTrans, afile, adata->bulkdata_val, apos,
333 adata->bulkdata_len);
339 SDISK_Truncate(struct rx_call *rxcall, struct ubik_tid *atid,
340 afs_int32 afile, afs_int32 alen)
343 struct ubik_dbase *dbase;
345 if ((code = ubik_CheckAuth(rxcall))) {
348 if (!ubik_currentTrans) {
351 /* sanity check to make sure only write trans appear here */
352 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
356 dbase = ubik_currentTrans->dbase;
358 urecovery_CheckTid(atid, 0);
359 if (!ubik_currentTrans) {
363 code = udisk_truncate(ubik_currentTrans, afile, alen);
369 SDISK_GetVersion(struct rx_call *rxcall,
370 struct ubik_version *aversion)
374 if ((code = ubik_CheckAuth(rxcall))) {
379 * If we are the sync site, recovery shouldn't be running on any
380 * other site. We shouldn't be getting this RPC as long as we are
381 * the sync site. To prevent any unforseen activity, we should
382 * reject this RPC until we have recognized that we are not the
383 * sync site anymore, and/or if we have any pending WRITE
384 * transactions that have to complete. This way we can be assured
385 * that this RPC would not block any pending transactions that
386 * should either fail or pass. If we have recognized the fact that
387 * we are not the sync site any more, all write transactions would
388 * fail with UNOQUORUM anyway.
390 if (ubeacon_AmSyncSite()) {
395 code = (*ubik_dbase->getlabel) (ubik_dbase, 0, aversion);
398 /* tell other side there's no dbase */
400 aversion->counter = 0;
406 SDISK_GetFile(struct rx_call *rxcall, afs_int32 file,
407 struct ubik_version *version)
410 struct ubik_dbase *dbase;
412 struct ubik_stat ubikstat;
417 if ((code = ubik_CheckAuth(rxcall))) {
420 /* temporarily disabled because it causes problems for migration tool. Hey, it's just
421 * a sanity check, anyway.
422 if (ubeacon_AmSyncSite()) {
428 code = (*dbase->stat) (dbase, file, &ubikstat);
433 length = ubikstat.size;
434 tlen = htonl(length);
435 code = rx_Write(rxcall, (char *)&tlen, sizeof(afs_int32));
436 if (code != sizeof(afs_int32)) {
438 ubik_dprint("Rx-write length error=%d\n", code);
443 tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
444 code = (*dbase->read) (dbase, file, tbuffer, offset, tlen);
447 ubik_dprint("read failed error=%d\n", code);
450 code = rx_Write(rxcall, tbuffer, tlen);
453 ubik_dprint("Rx-write length error=%d\n", code);
459 code = (*dbase->getlabel) (dbase, file, version); /* return the dbase, too */
465 SDISK_SendFile(struct rx_call *rxcall, afs_int32 file,
466 afs_int32 length, struct ubik_version *avers)
469 struct ubik_dbase *dbase = NULL;
472 struct ubik_version tversion;
474 struct rx_peer *tpeer;
475 struct rx_connection *tconn;
476 afs_uint32 otherHost = 0;
478 #ifndef OLD_URECOVERY
485 /* send the file back to the requester */
487 if ((code = ubik_CheckAuth(rxcall))) {
491 /* next, we do a sanity check to see if the guy sending us the database is
492 * the guy we think is the sync site. It turns out that we might not have
493 * decided yet that someone's the sync site, but they could have enough
494 * votes from others to be sync site anyway, and could send us the database
495 * in advance of getting our votes. This is fine, what we're really trying
496 * to check is that some authenticated bogon isn't sending a random database
497 * into another configuration. This could happen on a bad configuration
498 * screwup. Thus, we only object if we're sure we know who the sync site
499 * is, and it ain't the guy talking to us.
501 offset = uvote_GetSyncSite();
502 tconn = rx_ConnectionOf(rxcall);
503 tpeer = rx_PeerOf(tconn);
504 otherHost = ubikGetPrimaryInterfaceAddr(rx_HostOf(tpeer));
505 if (offset && offset != otherHost) {
506 /* we *know* this is the wrong guy */
514 /* abort any active trans that may scribble over the database */
515 urecovery_AbortAll(dbase);
517 ubik_print("Ubik: Synchronize database with server %s\n",
518 afs_inet_ntoa_r(otherHost, hoststr));
522 (*dbase->truncate) (dbase, file, 0); /* truncate first */
523 tversion.counter = 0;
527 tversion.epoch = 0; /* start off by labelling in-transit db as invalid */
528 (*dbase->setlabel) (dbase, file, &tversion); /* setlabel does sync */
529 #ifndef OLD_URECOVERY
530 afs_snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP", ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
531 fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
536 code = lseek(fd, HDRSIZE, 0);
537 if (code != HDRSIZE) {
543 memcpy(&ubik_dbase->version, &tversion, sizeof(struct ubik_version));
545 tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
546 #if !defined(OLD_URECOVERY) && !defined(AFS_PTHREAD_ENV)
550 code = rx_Read(rxcall, tbuffer, tlen);
553 ubik_dprint("Rx-read length error=%d\n", code);
559 code = (*dbase->write) (dbase, file, tbuffer, offset, tlen);
561 code = write(fd, tbuffer, tlen);
566 ubik_dprint("write failed error=%d\n", code);
574 #ifndef OLD_URECOVERY
580 /* sync data first, then write label and resync (resync done by setlabel call).
581 * This way, good label is only on good database. */
583 (*ubik_dbase->sync) (dbase, file);
585 afs_snprintf(tbuffer, sizeof(tbuffer), "%s.DB%s%d", ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
587 afs_snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD", ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
588 code = unlink(pbuffer);
590 code = rename(tbuffer, pbuffer);
591 afs_snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP", ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
594 code = rename(pbuffer, tbuffer);
596 (*ubik_dbase->open) (ubik_dbase, file);
598 code = (*ubik_dbase->setlabel) (dbase, file, avers);
599 #ifndef OLD_URECOVERY
602 afs_snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD", ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
606 memcpy(&ubik_dbase->version, avers, sizeof(struct ubik_version));
607 udisk_Invalidate(dbase, file); /* new dbase, flush disk buffers */
608 #ifdef AFS_PTHREAD_ENV
609 assert(pthread_cond_broadcast(&dbase->version_cond) == 0);
611 LWP_NoYieldSignal(&dbase->version);
616 #ifndef OLD_URECOVERY
618 /* Failed to sync. Allow reads again for now. */
620 tversion.epoch = epoch;
621 (*dbase->setlabel) (dbase, file, &tversion);
625 ("Ubik: Synchronize database with server %s failed (error = %d)\n",
626 afs_inet_ntoa_r(otherHost, hoststr), code);
628 ubik_print("Ubik: Synchronize database completed\n");
635 SDISK_Probe(struct rx_call *rxcall)
641 * \brief Update remote machines addresses in my server list
643 * Send back my addresses to caller of this RPC
644 * \return zero on success, else 1.
647 SDISK_UpdateInterfaceAddr(struct rx_call *rxcall,
648 UbikInterfaceAddr *inAddr,
649 UbikInterfaceAddr *outAddr)
651 struct ubik_server *ts, *tmp;
652 afs_uint32 remoteAddr; /* in net byte order */
653 int i, j, found = 0, probableMatch = 0;
656 /* copy the output parameters */
657 for (i = 0; i < UBIK_MAX_INTERFACE_ADDR; i++)
658 outAddr->hostAddr[i] = ntohl(ubik_host[i]);
660 remoteAddr = htonl(inAddr->hostAddr[0]);
661 for (ts = ubik_servers; ts; ts = ts->next)
662 if (ts->addr[0] == remoteAddr) { /* both in net byte order */
668 /* verify that all addresses in the incoming RPC are
669 ** not part of other server entries in my CellServDB
671 for (i = 0; !found && (i < UBIK_MAX_INTERFACE_ADDR)
672 && inAddr->hostAddr[i]; i++) {
673 remoteAddr = htonl(inAddr->hostAddr[i]);
674 for (tmp = ubik_servers; (!found && tmp); tmp = tmp->next) {
675 if (ts == tmp) /* this is my server */
677 for (j = 0; (j < UBIK_MAX_INTERFACE_ADDR) && tmp->addr[j];
679 if (remoteAddr == tmp->addr[j]) {
687 /* if (probableMatch) */
688 /* inconsistent addresses in CellServDB */
689 if (!probableMatch || found) {
690 ubik_print("Inconsistent Cell Info from server: ");
691 for (i = 0; i < UBIK_MAX_INTERFACE_ADDR && inAddr->hostAddr[i]; i++)
692 ubik_print("%s ", afs_inet_ntoa_r(htonl(inAddr->hostAddr[i]), hoststr));
700 /* update our data structures */
701 for (i = 1; i < UBIK_MAX_INTERFACE_ADDR; i++)
702 ts->addr[i] = htonl(inAddr->hostAddr[i]);
704 ubik_print("ubik: A Remote Server has addresses: ");
705 for (i = 0; i < UBIK_MAX_INTERFACE_ADDR && ts->addr[i]; i++)
706 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[i], hoststr));
713 printServerInfo(void)
715 struct ubik_server *ts;
719 ubik_print("Local CellServDB:");
720 for (ts = ubik_servers; ts; ts = ts->next, j++) {
721 ubik_print("Server %d: ", j);
722 for (i = 0; (i < UBIK_MAX_INTERFACE_ADDR) && ts->addr[i]; i++)
723 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[i], hoststr));
729 SDISK_SetVersion(struct rx_call *rxcall, struct ubik_tid *atid,
730 struct ubik_version *oldversionp,
731 struct ubik_version *newversionp)
734 struct ubik_dbase *dbase;
736 if ((code = ubik_CheckAuth(rxcall))) {
740 if (!ubik_currentTrans) {
743 /* sanity check to make sure only write trans appear here */
744 if (ubik_currentTrans->type != UBIK_WRITETRANS) {
748 /* Should not get this for the sync site */
749 if (ubeacon_AmSyncSite()) {
753 dbase = ubik_currentTrans->dbase;
755 urecovery_CheckTid(atid, 0);
756 if (!ubik_currentTrans) {
761 /* Set the label if its version matches the sync-site's */
762 if ((oldversionp->epoch == ubik_dbVersion.epoch)
763 && (oldversionp->counter == ubik_dbVersion.counter)) {
764 code = (*dbase->setlabel) (ubik_dbase, 0, newversionp);
766 ubik_dbase->version = *newversionp;
767 ubik_dbVersion = *newversionp;