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>
26 osi_rwlock_t cm_volumeLock;
29 cm_ValidateVolume(void)
34 for (volp = cm_data.allVolumesp, count = 0; volp; volp=volp->allNextp, count++) {
35 if ( volp->magic != CM_VOLUME_MAGIC ) {
36 afsi_log("cm_ValidateVolume failure: volp->magic != CM_VOLUME_MAGIC");
37 fprintf(stderr, "cm_ValidateVolume failure: volp->magic != CM_VOLUME_MAGIC\n");
40 if ( volp->cellp && volp->cellp->magic != CM_CELL_MAGIC ) {
41 afsi_log("cm_ValidateVolume failure: volp->cellp->magic != CM_CELL_MAGIC");
42 fprintf(stderr, "cm_ValidateVolume failure: volp->cellp->magic != CM_CELL_MAGIC\n");
45 if ( volp->allNextp && volp->allNextp->magic != CM_VOLUME_MAGIC ) {
46 afsi_log("cm_ValidateVolume failure: volp->allNextp->magic != CM_VOLUME_MAGIC");
47 fprintf(stderr, "cm_ValidateVolume failure: volp->allNextp->magic != CM_VOLUME_MAGIC\n");
50 if ( count != 0 && volp == cm_data.allVolumesp ||
51 count > cm_data.maxVolumes ) {
52 afsi_log("cm_ValidateVolume failure: cm_data.allVolumep loop detected");
53 fprintf(stderr, "cm_ValidateVolume failure: cm_data.allVolumep loop detected\n");
58 if ( count != cm_data.currentVolumes ) {
59 afsi_log("cm_ValidateVolume failure: count != cm_data.currentVolumes");
60 fprintf(stderr, "cm_ValidateVolume failure: count != cm_data.currentVolumes\n");
68 cm_ShutdownVolume(void)
72 for (volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
74 for ( volType = RWVOL; volType < NUM_VOL_TYPES; volType++) {
75 if (volp->vol[volType].ID)
76 cm_VolumeStatusNotification(volp, volp->vol[volType].ID, volp->vol[volType].state, vl_alldown);
78 volp->cbExpiresRO = 0;
80 volp->cbServerpRO = NULL;
81 lock_FinalizeRWLock(&volp->rw);
87 void cm_InitVolume(int newFile, long maxVols)
89 static osi_once_t once;
91 if (osi_Once(&once)) {
92 lock_InitializeRWLock(&cm_volumeLock, "cm global volume lock", LOCK_HIERARCHY_VOLUME_GLOBAL);
95 cm_data.allVolumesp = NULL;
96 cm_data.currentVolumes = 0;
97 cm_data.maxVolumes = maxVols;
98 memset(cm_data.volumeNameHashTablep, 0, sizeof(cm_volume_t *) * cm_data.volumeHashTableSize);
99 memset(cm_data.volumeRWIDHashTablep, 0, sizeof(cm_volume_t *) * cm_data.volumeHashTableSize);
100 memset(cm_data.volumeROIDHashTablep, 0, sizeof(cm_volume_t *) * cm_data.volumeHashTableSize);
101 memset(cm_data.volumeBKIDHashTablep, 0, sizeof(cm_volume_t *) * cm_data.volumeHashTableSize);
102 cm_data.volumeLRUFirstp = cm_data.volumeLRULastp = NULL;
106 for (volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
109 lock_InitializeRWLock(&volp->rw, "cm_volume_t rwlock", LOCK_HIERARCHY_VOLUME);
110 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_RESET);
111 _InterlockedAnd(&volp->flags, ~CM_VOLUMEFLAG_UPDATING_VL);
112 volp->lastUpdateTime = 0;
113 for (volType = RWVOL; volType < NUM_VOL_TYPES; volType++) {
114 volp->vol[volType].state = vl_unknown;
115 volp->vol[volType].serversp = NULL;
116 if (volp->vol[volType].ID)
117 cm_VolumeStatusNotification(volp, volp->vol[volType].ID, vl_unknown, volp->vol[volType].state);
119 volp->cbExpiresRO = 0;
120 volp->cbIssuedRO = 0;
121 volp->cbServerpRO = NULL;
129 /* returns true if the id is a decimal integer, in which case we interpret it
130 * as an id. make the cache manager much simpler.
131 * Stolen from src/volser/vlprocs.c */
133 cm_VolNameIsID(char *aname)
136 while (tc = *aname++) {
137 if (tc > '9' || tc < '0')
145 * Update a volume. Caller holds a write lock on the volume (volp->rw).
148 * shadow / openafs / jhutz@CS.CMU.EDU {ANDREW.CMU.EDU} 01:38 (JHutz)
149 * Yes, we support multihomed fileservers.
150 * Since before we got the code from IBM.
151 * But to find out about multiple addresses on a multihomed server, you need
152 * to use VL_GetEntryByNameU and VL_GetAddrsU. If you use
153 * VL_GetEntryByNameO or VL_GetEntryByNameN, the vlserver just gives you one
154 * address per server.
155 * shadow / openafs / jhutz@CS.CMU.EDU {ANDREW.CMU.EDU} 01:39 (JHutz)
156 * see src/afs/afs_volume.c, paying particular attention to
157 * afs_NewVolumeByName, afs_SetupVolume, and InstallUVolumeEntry
158 * shadow / openafs / jaltman {ANDREW.CMU.EDU} 01:40 (Jeffrey Altman)
159 * thanks. The windows client calls the 0 versions.
160 * shadow / openafs / jhutz@CS.CMU.EDU {ANDREW.CMU.EDU} 01:51 (JHutz)
162 * By not using the N versions, you only get up to 8 sites instead of 13.
163 * By not using the U versions, you don't get to know about multihomed serve
164 * shadow / openafs / jhutz@CS.CMU.EDU {ANDREW.CMU.EDU} 01:52 (JHutz)
165 * Of course, you probably want to support the older versions for backward
166 * compatibility. If you do that, you need to call the newest interface
167 * first, and fall back to successively older versions if you get
171 cm_GetEntryByName( struct cm_cell *cellp, const char *name,
172 struct vldbentry *vldbEntryp,
173 struct nvldbentry *nvldbEntryp,
174 struct uvldbentry *uvldbEntryp,
182 struct rx_connection * rxconnp;
184 osi_Log2(afsd_logp, "CALL VL_GetEntryByName{UNO} name %s:%s",
185 osi_LogSaveString(afsd_logp,cellp->name),
186 osi_LogSaveString(afsd_logp,name));
189 code = cm_ConnByMServers(cellp->vlServersp, FALSE, userp, reqp, &connp);
193 rxconnp = cm_GetRxConn(connp);
194 code = VL_GetEntryByNameU(rxconnp, name, uvldbEntryp);
196 if ( code == RXGEN_OPCODE )
198 code = VL_GetEntryByNameN(rxconnp, name, nvldbEntryp);
201 if ( code == RXGEN_OPCODE ) {
202 code = VL_GetEntryByNameO(rxconnp, name, vldbEntryp);
205 rx_PutConnection(rxconnp);
206 } while (cm_Analyze(connp, userp, reqp, NULL, cellp, 0, NULL, cellp->vlServersp, NULL, code));
207 code = cm_MapVLRPCError(code, reqp);
209 osi_Log3(afsd_logp, "CALL VL_GetEntryByName{UNO} name %s:%s FAILURE, code 0x%x",
210 osi_LogSaveString(afsd_logp,cellp->name),
211 osi_LogSaveString(afsd_logp,name), code);
213 osi_Log2(afsd_logp, "CALL VL_GetEntryByName{UNO} name %s:%s SUCCESS",
214 osi_LogSaveString(afsd_logp,cellp->name),
215 osi_LogSaveString(afsd_logp,name));
220 cm_GetEntryByID( struct cm_cell *cellp, afs_uint32 id,
221 struct vldbentry *vldbEntryp,
222 struct nvldbentry *nvldbEntryp,
223 struct uvldbentry *uvldbEntryp,
231 StringCbPrintf(name, sizeof(name), "%u", id);
233 return cm_GetEntryByName(cellp, name, vldbEntryp, nvldbEntryp, uvldbEntryp, methodp, userp, reqp);
236 long cm_UpdateVolumeLocation(struct cm_cell *cellp, cm_user_t *userp, cm_req_t *reqp,
239 struct rx_connection *rxconnp;
243 cm_serverRef_t *tsrp;
245 struct sockaddr_in tsockAddr;
248 struct vldbentry vldbEntry;
249 struct nvldbentry nvldbEntry;
250 struct uvldbentry uvldbEntry;
255 enum volstatus rwNewstate = vl_online;
256 enum volstatus roNewstate = vl_online;
257 enum volstatus bkNewstate = vl_online;
258 #ifdef AFS_FREELANCE_CLIENT
264 lock_AssertWrite(&volp->rw);
267 * If the last volume update was in the last five
268 * minutes and it did not exist, then avoid the RPC
269 * and return No Such Volume immediately.
272 if ((volp->flags & CM_VOLUMEFLAG_NOEXIST) &&
273 (now < volp->lastUpdateTime + 600))
275 return CM_ERROR_NOSUCHVOLUME;
278 #ifdef AFS_FREELANCE_CLIENT
279 if (cellp->cellID == AFS_FAKE_ROOT_CELL_ID)
282 memset(&vldbEntry, 0, sizeof(vldbEntry));
283 vldbEntry.flags |= VLF_RWEXISTS;
284 vldbEntry.volumeId[0] = AFS_FAKE_ROOT_VOL_ID;
290 while (volp->flags & CM_VOLUMEFLAG_UPDATING_VL) {
291 osi_Log3(afsd_logp, "cm_UpdateVolumeLocation sleeping name %s:%s flags 0x%x",
292 volp->cellp->name, volp->namep, volp->flags);
293 osi_SleepW((LONG_PTR) &volp->flags, &volp->rw);
294 lock_ObtainWrite(&volp->rw);
295 osi_Log3(afsd_logp, "cm_UpdateVolumeLocation awake name %s:%s flags 0x%x",
296 volp->cellp->name, volp->namep, volp->flags);
297 if (!(volp->flags & CM_VOLUMEFLAG_RESET)) {
298 osi_Log3(afsd_logp, "cm_UpdateVolumeLocation nothing to do, waking others name %s:%s flags 0x%x",
299 volp->cellp->name, volp->namep, volp->flags);
300 osi_Wakeup((LONG_PTR) &volp->flags);
306 /* Do not query again if the last update attempt failed in the last 60 seconds */
307 if ((volp->flags & CM_VOLUMEFLAG_RESET) && (volp->lastUpdateTime > now - 60))
309 osi_Log3(afsd_logp, "cm_UpdateVolumeLocation unsuccessful update in last 60 seconds -- name %s:%s flags 0x%x",
310 volp->cellp->name, volp->namep, volp->flags);
311 return(CM_ERROR_ALLDOWN);
314 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_UPDATING_VL);
316 /* Do not hold the volume lock across the RPC calls */
317 lock_ReleaseWrite(&volp->rw);
319 if (cellp->flags & CM_CELLFLAG_VLSERVER_INVALID)
320 cm_UpdateCell(cellp, 0);
322 /* now we have volume structure locked and held; make RPC to fill it */
323 code = cm_GetEntryByName(cellp, volp->namep, &vldbEntry, &nvldbEntry,
325 &method, userp, reqp);
327 /* We can end up here with code == CM_ERROR_NOSUCHVOLUME if the base volume name
328 * does not exist and is not a numeric string but there might exist a .readonly volume.
329 * If the base name doesn't exist we will not care about the .backup that might be left
330 * behind since there should be no method to access it.
332 if (code == CM_ERROR_NOSUCHVOLUME &&
333 _atoi64(volp->namep) == 0 &&
334 volp->vol[RWVOL].ID == 0 &&
335 strlen(volp->namep) < (VL_MAXNAMELEN - 9)) {
336 char name[VL_MAXNAMELEN];
338 snprintf(name, VL_MAXNAMELEN, "%s.readonly", volp->namep);
340 /* now we have volume structure locked and held; make RPC to fill it */
341 code = cm_GetEntryByName(cellp, name, &vldbEntry, &nvldbEntry,
343 &method, userp, reqp);
347 * What if there was a volume rename? The volume name no longer exists but the
348 * volume id might. Try to refresh the volume location information based one
349 * of the readwrite or readonly volume id.
351 if (code == CM_ERROR_NOSUCHVOLUME) {
352 if (volp->vol[RWVOL].ID != 0) {
353 code = cm_GetEntryByID(cellp, volp->vol[RWVOL].ID, &vldbEntry, &nvldbEntry,
355 &method, userp, reqp);
356 } else if (volp->vol[ROVOL].ID != 0) {
357 code = cm_GetEntryByID(cellp, volp->vol[ROVOL].ID, &vldbEntry, &nvldbEntry,
359 &method, userp, reqp);
362 lock_ObtainWrite(&volp->rw);
371 afs_int32 serverNumber[NMAXNSERVERS];
372 afs_int32 serverFlags[NMAXNSERVERS];
373 afsUUID serverUUID[NMAXNSERVERS];
374 afs_int32 rwServers_alldown = 1;
375 afs_int32 roServers_alldown = 1;
376 afs_int32 bkServers_alldown = 1;
377 char name[VL_MAXNAMELEN];
379 #ifdef AFS_FREELANCE_CLIENT
381 rwServers_alldown = 0;
384 /* clear out old bindings */
385 for ( volType = RWVOL; volType < NUM_VOL_TYPES; volType++) {
386 if (volp->vol[volType].serversp)
387 cm_FreeServerList(&volp->vol[volType].serversp, CM_FREESERVERLIST_DELETE);
390 memset(serverUUID, 0, sizeof(serverUUID));
394 flags = vldbEntry.flags;
395 nServers = vldbEntry.nServers;
396 rwID = vldbEntry.volumeId[0];
397 roID = vldbEntry.volumeId[1];
398 bkID = vldbEntry.volumeId[2];
399 for ( i=0; i<nServers; i++ ) {
400 serverFlags[i] = vldbEntry.serverFlags[i];
401 serverNumber[i] = vldbEntry.serverNumber[i];
403 strncpy(name, vldbEntry.name, VL_MAXNAMELEN);
404 name[VL_MAXNAMELEN - 1] = '\0';
407 flags = nvldbEntry.flags;
408 nServers = nvldbEntry.nServers;
409 rwID = nvldbEntry.volumeId[0];
410 roID = nvldbEntry.volumeId[1];
411 bkID = nvldbEntry.volumeId[2];
412 for ( i=0; i<nServers; i++ ) {
413 serverFlags[i] = nvldbEntry.serverFlags[i];
414 serverNumber[i] = nvldbEntry.serverNumber[i];
416 strncpy(name, nvldbEntry.name, VL_MAXNAMELEN);
417 name[VL_MAXNAMELEN - 1] = '\0';
420 flags = uvldbEntry.flags;
421 nServers = uvldbEntry.nServers;
422 rwID = uvldbEntry.volumeId[0];
423 roID = uvldbEntry.volumeId[1];
424 bkID = uvldbEntry.volumeId[2];
425 for ( i=0, j=0; code == 0 && i<nServers && j<NMAXNSERVERS; i++ ) {
426 if ( !(uvldbEntry.serverFlags[i] & VLSF_UUID) ) {
427 serverFlags[j] = uvldbEntry.serverFlags[i];
428 serverNumber[j] = uvldbEntry.serverNumber[i].time_low;
431 afs_uint32 * addrp, nentries, code, unique;
433 ListAddrByAttributes attrs;
436 memset(&attrs, 0, sizeof(attrs));
437 attrs.Mask = VLADDR_UUID;
438 attrs.uuid = uvldbEntry.serverNumber[i];
439 memset(&uuid, 0, sizeof(uuid));
440 memset(&addrs, 0, sizeof(addrs));
443 code = cm_ConnByMServers(cellp->vlServersp, FALSE, userp, reqp, &connp);
447 rxconnp = cm_GetRxConn(connp);
448 code = VL_GetAddrsU(rxconnp, &attrs, &uuid, &unique, &nentries, &addrs);
449 rx_PutConnection(rxconnp);
450 } while (cm_Analyze(connp, userp, reqp, NULL, cellp, 0, NULL, cellp->vlServersp, NULL, code));
453 code = cm_MapVLRPCError(code, reqp);
454 osi_Log2(afsd_logp, "CALL VL_GetAddrsU serverNumber %u FAILURE, code 0x%x",
458 osi_Log1(afsd_logp, "CALL VL_GetAddrsU serverNumber %u SUCCESS", i);
460 addrp = addrs.bulkaddrs_val;
461 for (k = 0; k < nentries && j < NMAXNSERVERS; j++, k++) {
462 serverFlags[j] = uvldbEntry.serverFlags[i];
463 serverNumber[j] = addrp[k];
464 serverUUID[j] = uuid;
467 xdr_free((xdrproc_t) xdr_bulkaddrs, &addrs);
470 code = CM_ERROR_INVAL;
473 nServers = j; /* update the server count */
474 strncpy(name, uvldbEntry.name, VL_MAXNAMELEN);
475 name[VL_MAXNAMELEN - 1] = '\0';
479 /* decode the response */
480 lock_ObtainWrite(&cm_volumeLock);
481 if (!cm_VolNameIsID(volp->namep)) {
486 if (len >= 8 && strcmp(name + len - 7, ".backup") == 0) {
487 name[len - 7] = '\0';
488 } else if (len >= 10 && strcmp(name + len - 9, ".readonly") == 0) {
489 name[len - 9] = '\0';
492 osi_Log2(afsd_logp, "cm_UpdateVolume name %s -> %s",
493 osi_LogSaveString(afsd_logp,volp->namep), osi_LogSaveString(afsd_logp,name));
495 if (volp->qflags & CM_VOLUME_QFLAG_IN_HASH)
496 cm_RemoveVolumeFromNameHashTable(volp);
498 strcpy(volp->namep, name);
500 cm_AddVolumeToNameHashTable(volp);
503 if (flags & VLF_DFSFILESET) {
504 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_DFS_VOLUME);
505 osi_Log1(afsd_logp, "cm_UpdateVolume Volume Group '%s' is a DFS File Set. Correct behavior is not implemented.",
506 osi_LogSaveString(afsd_logp, volp->namep));
509 if (flags & VLF_RWEXISTS) {
510 if (volp->vol[RWVOL].ID != rwID) {
511 if (volp->vol[RWVOL].qflags & CM_VOLUME_QFLAG_IN_HASH)
512 cm_RemoveVolumeFromIDHashTable(volp, RWVOL);
513 volp->vol[RWVOL].ID = rwID;
514 cm_AddVolumeToIDHashTable(volp, RWVOL);
517 if (volp->vol[RWVOL].qflags & CM_VOLUME_QFLAG_IN_HASH)
518 cm_RemoveVolumeFromIDHashTable(volp, RWVOL);
519 volp->vol[RWVOL].ID = 0;
521 if (flags & VLF_ROEXISTS) {
522 if (volp->vol[ROVOL].ID != roID) {
523 if (volp->vol[ROVOL].qflags & CM_VOLUME_QFLAG_IN_HASH)
524 cm_RemoveVolumeFromIDHashTable(volp, ROVOL);
525 volp->vol[ROVOL].ID = roID;
526 cm_AddVolumeToIDHashTable(volp, ROVOL);
529 if (volp->vol[ROVOL].qflags & CM_VOLUME_QFLAG_IN_HASH)
530 cm_RemoveVolumeFromIDHashTable(volp, ROVOL);
531 volp->vol[ROVOL].ID = 0;
533 if (flags & VLF_BACKEXISTS) {
534 if (volp->vol[BACKVOL].ID != bkID) {
535 if (volp->vol[BACKVOL].qflags & CM_VOLUME_QFLAG_IN_HASH)
536 cm_RemoveVolumeFromIDHashTable(volp, BACKVOL);
537 volp->vol[BACKVOL].ID = bkID;
538 cm_AddVolumeToIDHashTable(volp, BACKVOL);
541 if (volp->vol[BACKVOL].qflags & CM_VOLUME_QFLAG_IN_HASH)
542 cm_RemoveVolumeFromIDHashTable(volp, BACKVOL);
543 volp->vol[BACKVOL].ID = 0;
545 lock_ReleaseWrite(&cm_volumeLock);
547 /* See if the replica sites are mixed versions */
548 for (i=0; i<nServers; i++) {
549 if (serverFlags[i] & VLSF_NEWREPSITE) {
555 for (i=0; i<nServers; i++) {
556 /* create a server entry */
557 tflags = serverFlags[i];
558 if (tflags & VLSF_DONTUSE)
560 tsockAddr.sin_port = htons(7000);
561 tsockAddr.sin_family = AF_INET;
562 tempAddr = htonl(serverNumber[i]);
563 tsockAddr.sin_addr.s_addr = tempAddr;
564 tsp = cm_FindServer(&tsockAddr, CM_SERVER_FILE, FALSE);
565 if (tsp && (method == 2) && (tsp->flags & CM_SERVERFLAG_UUID)) {
567 * Check to see if the uuid of the server we know at this address
568 * matches the uuid of the server we are being told about by the
569 * vlserver. If not, ...?
571 if (!afs_uuid_equal(&serverUUID[i], &tsp->uuid)) {
572 char uuid1[128], uuid2[128];
575 afsUUID_to_string(&serverUUID[i], uuid1, sizeof(uuid1));
576 afsUUID_to_string(&tsp->uuid, uuid2, sizeof(uuid2));
577 afs_inet_ntoa_r(serverNumber[i], hoststr);
579 osi_Log3(afsd_logp, "cm_UpdateVolumeLocation UUIDs do not match! %s != %s (%s)",
580 osi_LogSaveString(afsd_logp, uuid1),
581 osi_LogSaveString(afsd_logp, uuid2),
582 osi_LogSaveString(afsd_logp, hoststr));
587 * cm_NewServer will probe the file server which in turn will
588 * update the state on the volume group object. Do not probe
589 * in this thread. It will block the thread and can result in
590 * a recursive call to cm_UpdateVolumeLocation().
592 lock_ReleaseWrite(&volp->rw);
593 tsp = cm_NewServer(&tsockAddr, CM_SERVER_FILE, cellp, &serverUUID[i], CM_FLAG_NOPROBE);
594 lock_ObtainWrite(&volp->rw);
596 osi_assertx(tsp != NULL, "null cm_server_t");
599 * if this server was created by fs setserverprefs
600 * then it won't have either a cell assignment or
605 if ( (method == 2) && !(tsp->flags & CM_SERVERFLAG_UUID) &&
606 !afs_uuid_is_nil(&serverUUID[i])) {
607 tsp->uuid = serverUUID[i];
608 _InterlockedOr(&tsp->flags, CM_SERVERFLAG_UUID);
611 /* and add it to the list(s). */
613 * Each call to cm_NewServerRef() increments the
614 * ref count of tsp. These reference will be dropped,
615 * if and when the volume is reset; see reset code
616 * earlier in this function.
618 if ((tflags & VLSF_RWVOL) && (flags & VLF_RWEXISTS)) {
619 tsrp = cm_NewServerRef(tsp, rwID);
620 cm_InsertServerList(&volp->vol[RWVOL].serversp, tsrp);
621 if (!(tsp->flags & CM_SERVERFLAG_DOWN))
622 rwServers_alldown = 0;
625 * If there are mixed versions of RO releases on the replica
626 * sites, skip the servers with the out of date versions.
628 if ((tflags & VLSF_ROVOL) && (flags & VLF_ROEXISTS) &&
629 (!isMixed || (tflags & VLSF_NEWREPSITE))) {
630 tsrp = cm_NewServerRef(tsp, roID);
631 cm_InsertServerList(&volp->vol[ROVOL].serversp, tsrp);
634 if (!(tsp->flags & CM_SERVERFLAG_DOWN))
635 roServers_alldown = 0;
637 /* We don't use VLSF_BACKVOL !?! */
638 /* Because only the backup on the server holding the RW
639 * volume can be valid. This check prevents errors if a
640 * RW is moved but the old backup is not removed.
642 if ((tflags & VLSF_RWVOL) && (flags & VLF_BACKEXISTS)) {
643 tsrp = cm_NewServerRef(tsp, bkID);
644 cm_InsertServerList(&volp->vol[BACKVOL].serversp, tsrp);
646 if (!(tsp->flags & CM_SERVERFLAG_DOWN))
647 bkServers_alldown = 0;
649 /* Drop the reference obtained by cm_FindServer() */
656 * If the first n servers have the same rank, then we
657 * randomly pick one among them and move it to the beginning.
658 * We don't bother to re-order the whole list because
659 * the rest of the list is used only if the first server is
660 * down. We only do this for the RO list; we assume the other
661 * lists are length 1.
664 cm_RandomizeServer(&volp->vol[ROVOL].serversp);
665 _InterlockedOr(&volp->vol[ROVOL].flags, CM_VOL_STATE_FLAG_REPLICATED);
668 _InterlockedAnd(&volp->vol[ROVOL].flags, ~CM_VOL_STATE_FLAG_REPLICATED);
671 rwNewstate = rwServers_alldown ? vl_alldown : vl_online;
672 roNewstate = roServers_alldown ? vl_alldown : vl_online;
673 bkNewstate = bkServers_alldown ? vl_alldown : vl_online;
675 _InterlockedAnd(&volp->flags, ~CM_VOLUMEFLAG_NOEXIST);
676 } else if (code == CM_ERROR_NOSUCHVOLUME || code == VL_NOENT || code == VL_BADNAME) {
677 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_NOEXIST);
679 rwNewstate = roNewstate = bkNewstate = vl_alldown;
682 if (volp->vol[RWVOL].state != rwNewstate) {
683 if (volp->vol[RWVOL].ID)
684 cm_VolumeStatusNotification(volp, volp->vol[RWVOL].ID, volp->vol[RWVOL].state, rwNewstate);
685 volp->vol[RWVOL].state = rwNewstate;
687 if (volp->vol[ROVOL].state != roNewstate) {
688 if (volp->vol[ROVOL].ID)
689 cm_VolumeStatusNotification(volp, volp->vol[ROVOL].ID, volp->vol[ROVOL].state, roNewstate);
690 volp->vol[ROVOL].state = roNewstate;
692 if (volp->vol[BACKVOL].state != bkNewstate) {
693 if (volp->vol[BACKVOL].ID)
694 cm_VolumeStatusNotification(volp, volp->vol[BACKVOL].ID, volp->vol[BACKVOL].state, bkNewstate);
695 volp->vol[BACKVOL].state = bkNewstate;
698 volp->lastUpdateTime = time(NULL);
700 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_RO_MIXED);
702 _InterlockedAnd(&volp->flags, ~CM_VOLUMEFLAG_RO_MIXED);
705 _InterlockedAnd(&volp->flags, ~CM_VOLUMEFLAG_RESET);
707 _InterlockedAnd(&volp->flags, ~CM_VOLUMEFLAG_UPDATING_VL);
708 osi_Log4(afsd_logp, "cm_UpdateVolumeLocation done, waking others name %s:%s flags 0x%x code 0x%x",
709 osi_LogSaveString(afsd_logp,volp->cellp->name),
710 osi_LogSaveString(afsd_logp,volp->namep), volp->flags, code);
711 osi_Wakeup((LONG_PTR) &volp->flags);
716 /* Requires read or write lock on cm_volumeLock */
717 void cm_GetVolume(cm_volume_t *volp)
719 InterlockedIncrement(&volp->refCount);
722 cm_volume_t *cm_GetVolumeByFID(cm_fid_t *fidp)
727 lock_ObtainRead(&cm_volumeLock);
728 hash = CM_VOLUME_ID_HASH(fidp->volume);
729 /* The volumeID can be any one of the three types. So we must
730 * search the hash table for all three types until we find it.
731 * We will search in the order of RO, RW, BK.
733 for ( volp = cm_data.volumeROIDHashTablep[hash]; volp; volp = volp->vol[ROVOL].nextp) {
734 if ( fidp->cell == volp->cellp->cellID && fidp->volume == volp->vol[ROVOL].ID )
739 for ( volp = cm_data.volumeRWIDHashTablep[hash]; volp; volp = volp->vol[RWVOL].nextp) {
740 if ( fidp->cell == volp->cellp->cellID && fidp->volume == volp->vol[RWVOL].ID )
746 for ( volp = cm_data.volumeBKIDHashTablep[hash]; volp; volp = volp->vol[BACKVOL].nextp) {
747 if ( fidp->cell == volp->cellp->cellID && fidp->volume == volp->vol[BACKVOL].ID )
752 /* hold the volume if we found it */
756 lock_ReleaseRead(&cm_volumeLock);
760 long cm_FindVolumeByID(cm_cell_t *cellp, afs_uint32 volumeID, cm_user_t *userp,
761 cm_req_t *reqp, afs_uint32 flags, cm_volume_t **outVolpp)
764 #ifdef SEARCH_ALL_VOLUMES
767 char volNameString[VL_MAXNAMELEN];
771 lock_ObtainRead(&cm_volumeLock);
772 #ifdef SEARCH_ALL_VOLUMES
773 for(volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
774 if (cellp == volp->cellp &&
775 ((unsigned) volumeID == volp->vol[RWVOL].ID ||
776 (unsigned) volumeID == volp->vol[ROVOL].ID ||
777 (unsigned) volumeID == volp->vol[BACKVOL].ID))
782 #endif /* SEARCH_ALL_VOLUMES */
784 hash = CM_VOLUME_ID_HASH(volumeID);
785 /* The volumeID can be any one of the three types. So we must
786 * search the hash table for all three types until we find it.
787 * We will search in the order of RO, RW, BK.
789 for ( volp = cm_data.volumeROIDHashTablep[hash]; volp; volp = volp->vol[ROVOL].nextp) {
790 if ( cellp == volp->cellp && volumeID == volp->vol[ROVOL].ID )
795 for ( volp = cm_data.volumeRWIDHashTablep[hash]; volp; volp = volp->vol[RWVOL].nextp) {
796 if ( cellp == volp->cellp && volumeID == volp->vol[RWVOL].ID )
802 for ( volp = cm_data.volumeBKIDHashTablep[hash]; volp; volp = volp->vol[BACKVOL].nextp) {
803 if ( cellp == volp->cellp && volumeID == volp->vol[BACKVOL].ID )
808 #ifdef SEARCH_ALL_VOLUMES
809 osi_assertx(volp == volp2, "unexpected cm_vol_t");
812 /* hold the volume if we found it */
816 lock_ReleaseRead(&cm_volumeLock);
820 lock_ObtainWrite(&volp->rw);
823 if ((volp->flags & CM_VOLUMEFLAG_RESET) && !(flags & CM_GETVOL_FLAG_NO_RESET)) {
824 code = cm_UpdateVolumeLocation(cellp, userp, reqp, volp);
826 lock_ReleaseWrite(&volp->rw);
830 if (!(flags & CM_GETVOL_FLAG_NO_LRU_UPDATE)) {
831 lock_ObtainWrite(&cm_volumeLock);
832 cm_AdjustVolumeLRU(volp);
833 lock_ReleaseWrite(&cm_volumeLock);
836 lock_ObtainRead(&cm_volumeLock);
838 lock_ReleaseRead(&cm_volumeLock);
843 /* otherwise, we didn't find it so consult the VLDB */
844 sprintf(volNameString, "%u", volumeID);
845 code = cm_FindVolumeByName(cellp, volNameString, userp, reqp,
846 flags | CM_GETVOL_FLAG_IGNORE_LINKED_CELL, outVolpp);
848 if (code == CM_ERROR_NOSUCHVOLUME && cellp->linkedName[0] &&
849 !(flags & CM_GETVOL_FLAG_IGNORE_LINKED_CELL)) {
850 cm_cell_t *linkedCellp = cm_GetCell(cellp->linkedName, flags);
853 code = cm_FindVolumeByID(linkedCellp, volumeID, userp, reqp,
854 flags | CM_GETVOL_FLAG_IGNORE_LINKED_CELL,
861 long cm_FindVolumeByName(struct cm_cell *cellp, char *volumeNamep,
862 struct cm_user *userp, struct cm_req *reqp,
863 afs_uint32 flags, cm_volume_t **outVolpp)
866 #ifdef SEARCH_ALL_VOLUMES
870 char name[VL_MAXNAMELEN];
875 strncpy(name, volumeNamep, VL_MAXNAMELEN);
876 name[VL_MAXNAMELEN-1] = '\0';
879 if (len >= 8 && strcmp(name + len - 7, ".backup") == 0) {
881 name[len - 7] = '\0';
882 } else if (len >= 10 && strcmp(name + len - 9, ".readonly") == 0) {
884 name[len - 9] = '\0';
889 lock_ObtainRead(&cm_volumeLock);
890 #ifdef SEARCH_ALL_VOLUMES
891 for (volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
892 if (cellp == volp->cellp && strcmp(name, volp->namep) == 0) {
897 #endif /* SEARCH_ALL_VOLUMES */
899 hash = CM_VOLUME_NAME_HASH(name);
900 for (volp = cm_data.volumeNameHashTablep[hash]; volp; volp = volp->nameNextp) {
901 if (cellp == volp->cellp && strcmp(name, volp->namep) == 0)
905 #ifdef SEARCH_ALL_VOLUMES
906 osi_assertx(volp2 == volp, "unexpected cm_vol_t");
909 if (!volp && (flags & CM_GETVOL_FLAG_CREATE)) {
911 /* otherwise, get from VLDB */
914 * Change to a write lock so that we have exclusive use of
915 * the first cm_volume_t with a refCount of 0 so that we
916 * have time to increment it.
918 lock_ConvertRToW(&cm_volumeLock);
920 if ( cm_data.currentVolumes >= cm_data.maxVolumes ) {
921 #ifdef RECYCLE_FROM_ALL_VOLUMES_LIST
922 for (volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
923 if ( volp->refCount == 0 ) {
924 /* There is one we can re-use */
929 for ( volp = cm_data.volumeLRULastp;
931 volp = (cm_volume_t *) osi_QPrev(&volp->q))
933 if ( volp->refCount == 0 ) {
934 /* There is one we can re-use */
940 osi_panic("Exceeded Max Volumes", __FILE__, __LINE__);
942 InterlockedIncrement(&volp->refCount);
943 lock_ReleaseWrite(&cm_volumeLock);
944 lock_ObtainWrite(&volp->rw);
945 lock_ObtainWrite(&cm_volumeLock);
947 osi_Log2(afsd_logp, "Recycling Volume %s:%s",
948 volp->cellp->name, volp->namep);
950 /* The volp is removed from the LRU queue in order to
951 * prevent two threads from attempting to recycle the
952 * same object. This volp must be re-inserted back into
953 * the LRU queue before this function exits.
955 if (volp->qflags & CM_VOLUME_QFLAG_IN_LRU_QUEUE)
956 cm_RemoveVolumeFromLRU(volp);
957 if (volp->qflags & CM_VOLUME_QFLAG_IN_HASH)
958 cm_RemoveVolumeFromNameHashTable(volp);
960 for ( volType = RWVOL; volType < NUM_VOL_TYPES; volType++) {
961 if (volp->vol[volType].qflags & CM_VOLUME_QFLAG_IN_HASH)
962 cm_RemoveVolumeFromIDHashTable(volp, volType);
963 if (volp->vol[volType].ID)
964 cm_VolumeStatusNotification(volp, volp->vol[volType].ID, volp->vol[volType].state, vl_unknown);
965 volp->vol[volType].ID = 0;
966 cm_SetFid(&volp->vol[volType].dotdotFid, 0, 0, 0, 0);
967 lock_ReleaseWrite(&cm_volumeLock);
968 cm_FreeServerList(&volp->vol[volType].serversp, CM_FREESERVERLIST_DELETE);
969 lock_ObtainWrite(&cm_volumeLock);
972 volp = &cm_data.volumeBaseAddress[cm_data.currentVolumes++];
973 memset(volp, 0, sizeof(cm_volume_t));
974 volp->magic = CM_VOLUME_MAGIC;
975 volp->allNextp = cm_data.allVolumesp;
976 cm_data.allVolumesp = volp;
977 lock_InitializeRWLock(&volp->rw, "cm_volume_t rwlock", LOCK_HIERARCHY_VOLUME);
978 lock_ReleaseWrite(&cm_volumeLock);
979 lock_ObtainWrite(&volp->rw);
980 lock_ObtainWrite(&cm_volumeLock);
981 volp->refCount = 1; /* starts off held */
984 strncpy(volp->namep, name, VL_MAXNAMELEN);
985 volp->namep[VL_MAXNAMELEN-1] = '\0';
986 volp->flags = CM_VOLUMEFLAG_RESET;
987 volp->lastUpdateTime = 0;
989 for ( volType = RWVOL; volType < NUM_VOL_TYPES; volType++) {
990 volp->vol[volType].state = vl_unknown;
991 volp->vol[volType].nextp = NULL;
992 volp->vol[volType].flags = 0;
994 volp->cbExpiresRO = 0;
995 volp->cbIssuedRO = 0;
996 volp->cbServerpRO = NULL;
997 volp->creationDateRO = 0;
998 cm_AddVolumeToNameHashTable(volp);
999 lock_ReleaseWrite(&cm_volumeLock);
1004 lock_ReleaseRead(&cm_volumeLock);
1007 return CM_ERROR_NOSUCHVOLUME;
1009 lock_ObtainWrite(&volp->rw);
1012 /* if we get here we are holding the mutex */
1013 if ((volp->flags & CM_VOLUMEFLAG_RESET) && !(flags & CM_GETVOL_FLAG_NO_RESET)) {
1014 code = cm_UpdateVolumeLocation(cellp, userp, reqp, volp);
1016 lock_ReleaseWrite(&volp->rw);
1018 if (code == 0 && (type == BACKVOL && volp->vol[BACKVOL].ID == 0 ||
1019 type == ROVOL && volp->vol[ROVOL].ID == 0))
1020 code = CM_ERROR_NOSUCHVOLUME;
1025 lock_ObtainWrite(&cm_volumeLock);
1026 if (!(volp->qflags & CM_VOLUME_QFLAG_IN_LRU_QUEUE) ||
1027 (flags & CM_GETVOL_FLAG_NO_LRU_UPDATE))
1028 cm_AdjustVolumeLRU(volp);
1029 lock_ReleaseWrite(&cm_volumeLock);
1032 * do not return it to the caller but do insert it in the LRU
1033 * otherwise it will be lost
1035 lock_ObtainWrite(&cm_volumeLock);
1036 if (!(volp->qflags & CM_VOLUME_QFLAG_IN_LRU_QUEUE) ||
1037 (flags & CM_GETVOL_FLAG_NO_LRU_UPDATE))
1038 cm_AdjustVolumeLRU(volp);
1040 lock_ReleaseWrite(&cm_volumeLock);
1043 if (code == CM_ERROR_NOSUCHVOLUME && cellp->linkedName[0] &&
1044 !(flags & CM_GETVOL_FLAG_IGNORE_LINKED_CELL)) {
1045 cm_cell_t *linkedCellp = cm_GetCell(cellp->linkedName, flags);
1048 code = cm_FindVolumeByName(linkedCellp, volumeNamep, userp, reqp,
1049 flags | CM_GETVOL_FLAG_IGNORE_LINKED_CELL,
1056 * Only call this function in response to a VNOVOL or VMOVED error
1057 * from a file server. Do not call it in response to CM_ERROR_NOSUCHVOLUME
1058 * as that can lead to recursive calls.
1060 long cm_ForceUpdateVolume(cm_fid_t *fidp, cm_user_t *userp, cm_req_t *reqp)
1064 #ifdef SEARCH_ALL_VOLUMES
1071 return CM_ERROR_INVAL;
1073 cellp = cm_FindCellByID(fidp->cell, 0);
1075 return CM_ERROR_NOSUCHCELL;
1077 /* search for the volume */
1078 lock_ObtainRead(&cm_volumeLock);
1079 #ifdef SEARCH_ALL_VOLUMES
1080 for(volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
1081 if (cellp == volp->cellp &&
1082 (fidp->volume == volp->vol[RWVOL].ID ||
1083 fidp->volume == volp->vol[ROVOL].ID ||
1084 fidp->volume == volp->vol[BACKVOL].ID))
1087 #endif /* SEARCH_ALL_VOLUMES */
1089 hash = CM_VOLUME_ID_HASH(fidp->volume);
1090 /* The volumeID can be any one of the three types. So we must
1091 * search the hash table for all three types until we find it.
1092 * We will search in the order of RO, RW, BK.
1094 for ( volp = cm_data.volumeROIDHashTablep[hash]; volp; volp = volp->vol[ROVOL].nextp) {
1095 if ( cellp == volp->cellp && fidp->volume == volp->vol[ROVOL].ID )
1099 /* try RW volumes */
1100 for ( volp = cm_data.volumeRWIDHashTablep[hash]; volp; volp = volp->vol[RWVOL].nextp) {
1101 if ( cellp == volp->cellp && fidp->volume == volp->vol[RWVOL].ID )
1106 /* try BK volumes */
1107 for ( volp = cm_data.volumeBKIDHashTablep[hash]; volp; volp = volp->vol[BACKVOL].nextp) {
1108 if ( cellp == volp->cellp && fidp->volume == volp->vol[BACKVOL].ID )
1113 #ifdef SEARCH_ALL_VOLUMES
1114 osi_assertx(volp == volp2, "unexpected cm_vol_t");
1116 /* hold the volume if we found it */
1120 lock_ReleaseRead(&cm_volumeLock);
1123 return CM_ERROR_NOSUCHVOLUME;
1126 cm_data.mountRootGen = time(NULL);
1127 lock_ObtainWrite(&volp->rw);
1128 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_RESET);
1129 volp->lastUpdateTime = 0;
1131 code = cm_UpdateVolumeLocation(cellp, userp, reqp, volp);
1132 lock_ReleaseWrite(&volp->rw);
1134 lock_ObtainRead(&cm_volumeLock);
1136 lock_ReleaseRead(&cm_volumeLock);
1141 /* find the appropriate servers from a volume */
1142 cm_serverRef_t **cm_GetVolServers(cm_volume_t *volp, afs_uint32 volid, cm_user_t *userp, cm_req_t *reqp, afs_uint32 *replicated)
1144 cm_serverRef_t **serverspp;
1145 cm_serverRef_t *current;
1147 cm_vol_state_t *volstatep = NULL;
1150 volstatep = cm_VolumeStateByID(volp, volid);
1152 lock_ObtainWrite(&cm_serverLock);
1155 *replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1156 serverspp = &volstatep->serversp;
1158 lock_ReleaseWrite(&cm_serverLock);
1162 lock_ObtainWrite(&volp->rw);
1163 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_RESET);
1164 volp->lastUpdateTime = 0;
1165 code = cm_UpdateVolumeLocation(volp->cellp, userp, reqp, volp);
1166 lock_ReleaseWrite(&volp->rw);
1174 * Increment the refCount on deleted items as well.
1175 * They will be freed by cm_FreeServerList when they get to zero
1177 for (current = *serverspp; current; current = current->next)
1178 cm_GetServerRef(current, TRUE);
1180 lock_ReleaseWrite(&cm_serverLock);
1185 void cm_PutVolume(cm_volume_t *volp)
1187 afs_int32 refCount = InterlockedDecrement(&volp->refCount);
1188 osi_assertx(refCount >= 0, "cm_volume_t refCount underflow has occurred");
1191 /* return the read-only volume, if there is one, or the read-write volume if
1194 long cm_GetROVolumeID(cm_volume_t *volp)
1198 lock_ObtainRead(&volp->rw);
1199 if (volp->vol[ROVOL].ID && !cm_IsServerListEmpty(volp->vol[ROVOL].serversp))
1200 id = volp->vol[ROVOL].ID;
1202 id = volp->vol[RWVOL].ID;
1203 lock_ReleaseRead(&volp->rw);
1208 void cm_RefreshVolumes(int lifetime)
1216 /* force mount point target updates */
1217 if (cm_data.mountRootGen + lifetime <= now)
1218 cm_data.mountRootGen = now;
1221 * force a re-loading of volume data from the vldb
1222 * if the lifetime for the cached data has expired
1224 lock_ObtainRead(&cm_volumeLock);
1225 for (volp = cm_data.allVolumesp; volp; volp=volp->allNextp) {
1226 InterlockedIncrement(&volp->refCount);
1227 lock_ReleaseRead(&cm_volumeLock);
1229 if (!(volp->flags & CM_VOLUMEFLAG_RESET)) {
1230 lock_ObtainWrite(&volp->rw);
1231 if (volp->flags & CM_VOLUMEFLAG_RO_MIXED) {
1232 if (volp->lastUpdateTime + 300 <= now) {
1233 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_RESET);
1234 volp->lastUpdateTime = 0;
1237 if (volp->lastUpdateTime + lifetime <= now) {
1238 _InterlockedOr(&volp->flags, CM_VOLUMEFLAG_RESET);
1239 volp->lastUpdateTime = 0;
1242 lock_ReleaseWrite(&volp->rw);
1245 lock_ObtainRead(&cm_volumeLock);
1246 refCount = InterlockedDecrement(&volp->refCount);
1247 osi_assertx(refCount >= 0, "cm_volume_t refCount underflow");
1249 lock_ReleaseRead(&cm_volumeLock);
1253 cm_CheckOfflineVolumeState(cm_volume_t *volp, cm_vol_state_t *statep, afs_uint32 volID,
1254 afs_uint32 *onlinep, afs_uint32 *volumeUpdatedp)
1258 AFSFetchVolumeStatus volStat;
1263 struct rx_connection * rxconnp;
1265 char offLineMsg[256];
1267 long alldown, alldeleted;
1268 cm_serverRef_t *serversp;
1270 cm_scache_t *vscp = NULL;
1273 OfflineMsg = offLineMsg;
1276 if (statep->ID != 0 && (!volID || volID == statep->ID)) {
1277 /* create fid for volume root so that VNOVOL and VMOVED errors can be processed */
1278 cm_SetFid(&vfid, volp->cellp->cellID, statep->ID, 1, 1);
1280 if (!statep->serversp && !(*volumeUpdatedp)) {
1282 code = cm_UpdateVolumeLocation(volp->cellp, cm_rootUserp, &req, volp);
1283 *volumeUpdatedp = 1;
1286 lock_ObtainRead(&cm_serverLock);
1287 if (statep->serversp) {
1290 for (serversp = statep->serversp; serversp; serversp = serversp->next) {
1291 if (serversp->status == srv_deleted)
1296 if (!(serversp->server->flags & CM_SERVERFLAG_DOWN))
1299 if (serversp->status == srv_busy || serversp->status == srv_offline)
1300 serversp->status = srv_not_busy;
1302 lock_ReleaseRead(&cm_serverLock);
1304 if (alldeleted && !(*volumeUpdatedp)) {
1306 code = cm_UpdateVolumeLocation(volp->cellp, cm_rootUserp, &req, volp);
1307 *volumeUpdatedp = 1;
1310 if (statep->state == vl_busy || statep->state == vl_offline || statep->state == vl_unknown ||
1311 (!alldown && statep->state == vl_alldown)) {
1313 req.flags |= CM_REQ_OFFLINE_VOL_CHK;
1314 lock_ReleaseWrite(&volp->rw);
1316 code = cm_GetSCache(&vfid, NULL, &vscp, cm_rootUserp, &req);
1318 lock_ObtainWrite(&vscp->rw);
1319 code = cm_SyncOp(vscp, NULL, cm_rootUserp, &req, PRSFS_READ,
1320 CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
1321 lock_ReleaseWrite(&vscp->rw);
1324 code = cm_ConnFromVolume(volp, statep->ID, cm_rootUserp, &req, &connp);
1328 rxconnp = cm_GetRxConn(connp);
1329 code = RXAFS_GetVolumeStatus(rxconnp, statep->ID,
1330 &volStat, &Name, &OfflineMsg, &MOTD);
1331 rx_PutConnection(rxconnp);
1332 } while (cm_Analyze(connp, cm_rootUserp, &req, &vfid, NULL, 0, NULL, NULL, NULL, code));
1333 code = cm_MapRPCError(code, &req);
1336 lock_ObtainWrite(&vscp->rw);
1337 cm_SyncOpDone(vscp, NULL, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
1338 lock_ReleaseWrite(&vscp->rw);
1339 cm_ReleaseSCache(vscp);
1341 lock_ObtainWrite(&volp->rw);
1342 if (code == 0 && volStat.Online) {
1343 cm_VolumeStatusNotification(volp, statep->ID, statep->state, vl_online);
1344 statep->state = vl_online;
1346 } else if (code == CM_ERROR_NOACCESS) {
1347 cm_VolumeStatusNotification(volp, statep->ID, statep->state, vl_unknown);
1348 statep->state = vl_unknown;
1351 } else if (alldown && statep->state != vl_alldown) {
1352 cm_VolumeStatusNotification(volp, statep->ID, statep->state, vl_alldown);
1353 statep->state = vl_alldown;
1356 lock_ReleaseRead(&cm_serverLock);
1357 if (statep->state != vl_alldown) {
1358 cm_VolumeStatusNotification(volp, statep->ID, statep->state, vl_alldown);
1359 statep->state = vl_alldown;
1365 /* The return code is 0 if the volume is not online and
1366 * 1 if the volume is online
1369 cm_CheckOfflineVolume(cm_volume_t *volp, afs_uint32 volID)
1373 afs_uint32 online = 0;
1374 afs_uint32 volumeUpdated = 0;
1376 lock_ObtainWrite(&volp->rw);
1378 if (volp->flags & CM_VOLUMEFLAG_RESET) {
1380 code = cm_UpdateVolumeLocation(volp->cellp, cm_rootUserp, &req, volp);
1384 cm_CheckOfflineVolumeState(volp, &volp->vol[RWVOL], volID, &online, &volumeUpdated);
1385 cm_CheckOfflineVolumeState(volp, &volp->vol[ROVOL], volID, &online, &volumeUpdated);
1386 cm_CheckOfflineVolumeState(volp, &volp->vol[BACKVOL], volID, &online, &volumeUpdated);
1388 lock_ReleaseWrite(&volp->rw);
1394 * called from the Daemon thread.
1395 * when checking the offline status, check those of the most recently used volumes first.
1397 void cm_CheckOfflineVolumes(void)
1401 extern int daemon_ShutdownFlag;
1402 extern int powerStateSuspended;
1404 lock_ObtainRead(&cm_volumeLock);
1405 for (volp = cm_data.volumeLRULastp;
1406 volp && !daemon_ShutdownFlag && !powerStateSuspended;
1407 volp=(cm_volume_t *) osi_QPrev(&volp->q)) {
1409 * Skip volume entries that did not exist last time
1410 * the vldb was queried. For those entries wait until
1411 * the next actual request is received for the volume
1412 * before checking its state.
1414 if ((volp->qflags & CM_VOLUME_QFLAG_IN_HASH) &&
1415 !(volp->flags & CM_VOLUMEFLAG_NOEXIST)) {
1416 InterlockedIncrement(&volp->refCount);
1417 lock_ReleaseRead(&cm_volumeLock);
1418 cm_CheckOfflineVolume(volp, 0);
1419 lock_ObtainRead(&cm_volumeLock);
1420 refCount = InterlockedDecrement(&volp->refCount);
1421 osi_assertx(refCount >= 0, "cm_volume_t refCount underflow");
1424 lock_ReleaseRead(&cm_volumeLock);
1429 cm_UpdateVolumeStatusInt(cm_volume_t *volp, struct cm_vol_state *statep)
1431 enum volstatus newStatus;
1432 cm_serverRef_t *tsrp;
1434 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1437 if (!volp || !statep) {
1444 lock_ObtainWrite(&cm_serverLock);
1445 for (tsrp = statep->serversp; tsrp; tsrp=tsrp->next) {
1447 sprintf(addr, "%d.%d.%d.%d",
1448 ((tsp->addr.sin_addr.s_addr & 0xff)),
1449 ((tsp->addr.sin_addr.s_addr & 0xff00)>> 8),
1450 ((tsp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1451 ((tsp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1453 if (tsrp->status == srv_deleted) {
1454 osi_Log2(afsd_logp, "cm_UpdateVolumeStatusInt volume %d server reference %s deleted",
1455 statep->ID, osi_LogSaveString(afsd_logp,addr));
1459 cm_GetServerNoLock(tsp);
1460 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1462 if (tsrp->status == srv_busy) {
1463 osi_Log2(afsd_logp, "cm_UpdateVolumeStatusInt volume %d server reference %s busy",
1464 statep->ID, osi_LogSaveString(afsd_logp,addr));
1467 } else if (tsrp->status == srv_offline) {
1468 osi_Log2(afsd_logp, "cm_UpdateVolumeStatusInt volume %d server reference %s offline",
1469 statep->ID, osi_LogSaveString(afsd_logp,addr));
1473 osi_Log2(afsd_logp, "cm_UpdateVolumeStatusInt volume %d server reference %s online",
1474 statep->ID, osi_LogSaveString(afsd_logp,addr));
1479 osi_Log2(afsd_logp, "cm_UpdateVolumeStatusInt volume %d server reference %s down",
1480 statep->ID, osi_LogSaveString(afsd_logp,addr));
1482 cm_PutServerNoLock(tsp);
1485 lock_ReleaseWrite(&cm_serverLock);
1487 osi_Log5(afsd_logp, "cm_UpdateVolumeStatusInt allDown %d allBusy %d someBusy %d someOffline %d allOffline %d",
1488 allDown, allBusy, someBusy, someOffline, allOffline);
1491 newStatus = vl_alldown;
1492 else if (allBusy || (someBusy && someOffline))
1493 newStatus = vl_busy;
1494 else if (allOffline)
1495 newStatus = vl_offline;
1497 newStatus = vl_online;
1499 if (statep->ID && statep->state != newStatus)
1500 cm_VolumeStatusNotification(volp, statep->ID, statep->state, newStatus);
1502 statep->state = newStatus;
1506 cm_UpdateVolumeStatus(cm_volume_t *volp, afs_uint32 volID)
1509 if (volp->vol[RWVOL].ID == volID) {
1510 cm_UpdateVolumeStatusInt(volp, &volp->vol[RWVOL]);
1511 } else if (volp->vol[ROVOL].ID == volID) {
1512 cm_UpdateVolumeStatusInt(volp, &volp->vol[ROVOL]);
1513 } else if (volp->vol[BACKVOL].ID == volID) {
1514 cm_UpdateVolumeStatusInt(volp, &volp->vol[BACKVOL]);
1517 * If we are called with volID == 0 then something has gone wrong.
1518 * Most likely a race occurred in the server volume list maintenance.
1519 * Since we don't know which volume's status should be updated,
1520 * just update all of them that are known to exist. Better to be
1521 * correct than fast.
1524 for ( volType = RWVOL; volType < NUM_VOL_TYPES; volType++) {
1525 if (volp->vol[volType].ID != 0)
1526 cm_UpdateVolumeStatusInt(volp, &volp->vol[volType]);
1532 ** Finds all volumes that reside on this server and reorders their
1533 ** RO list according to the changed rank of server.
1535 void cm_ChangeRankVolume(cm_server_t *tsp)
1541 /* find volumes which might have RO copy on server*/
1542 lock_ObtainRead(&cm_volumeLock);
1543 for(volp = cm_data.allVolumesp; volp; volp=volp->allNextp)
1545 code = 1 ; /* assume that list is unchanged */
1546 InterlockedIncrement(&volp->refCount);
1547 lock_ReleaseRead(&cm_volumeLock);
1548 lock_ObtainWrite(&volp->rw);
1550 if ((tsp->cellp==volp->cellp) && (volp->vol[ROVOL].serversp))
1551 code =cm_ChangeRankServer(&volp->vol[ROVOL].serversp, tsp);
1553 /* this volume list was changed */
1555 cm_RandomizeServer(&volp->vol[ROVOL].serversp);
1557 lock_ReleaseWrite(&volp->rw);
1558 lock_ObtainRead(&cm_volumeLock);
1559 refCount = InterlockedDecrement(&volp->refCount);
1560 osi_assertx(refCount >= 0, "cm_volume_t refCount underflow");
1562 lock_ReleaseRead(&cm_volumeLock);
1565 /* dump all volumes that have reference count > 0 to a file.
1566 * cookie is used to identify this batch for easy parsing,
1567 * and it a string provided by a caller
1569 int cm_DumpVolumes(FILE *outputFile, char *cookie, int lock)
1576 lock_ObtainRead(&cm_scacheLock);
1577 lock_ObtainRead(&cm_volumeLock);
1580 sprintf(output, "%s - dumping volumes - cm_data.currentVolumes=%d, cm_data.maxVolumes=%d\r\n",
1581 cookie, cm_data.currentVolumes, cm_data.maxVolumes);
1582 WriteFile(outputFile, output, (DWORD)strlen(output), &zilch, NULL);
1584 for (volp = cm_data.allVolumesp; volp; volp=volp->allNextp)
1587 char *srvStr = NULL;
1588 afs_uint32 srvStrRpc = TRUE;
1592 if (volp->cbServerpRO) {
1593 if (!((volp->cbServerpRO->flags & CM_SERVERFLAG_UUID) &&
1594 UuidToString((UUID *)&volp->cbServerpRO->uuid, &srvStr) == RPC_S_OK)) {
1595 srvStr = malloc(16);
1597 afs_inet_ntoa_r(volp->cbServerpRO->addr.sin_addr.s_addr, srvStr);
1601 if (volp->cbExpiresRO) {
1602 t = volp->cbExpiresRO;
1606 cbt[strlen(cbt)-1] = '\0';
1609 if (volp->creationDateRO) {
1610 t = volp->creationDateRO;
1613 cdrot = strdup(cdrot);
1614 cdrot[strlen(cdrot)-1] = '\0';
1619 "%s - volp=0x%p cell=%s name=%s rwID=%u roID=%u bkID=%u flags=0x%x:%x "
1620 "cbServerpRO='%s' cbExpiresRO='%s' creationDateRO='%s' refCount=%u\r\n",
1621 cookie, volp, volp->cellp->name, volp->namep, volp->vol[RWVOL].ID,
1622 volp->vol[ROVOL].ID, volp->vol[BACKVOL].ID, volp->flags, volp->qflags,
1623 srvStr ? srvStr : "<none>", cbt ? cbt : "<none>", cdrot ? cdrot : "<none>",
1625 WriteFile(outputFile, output, (DWORD)strlen(output), &zilch, NULL);
1629 RpcStringFree(&srvStr);
1638 sprintf(output, "%s - Done dumping volumes.\r\n", cookie);
1639 WriteFile(outputFile, output, (DWORD)strlen(output), &zilch, NULL);
1642 lock_ReleaseRead(&cm_volumeLock);
1643 lock_ReleaseRead(&cm_scacheLock);
1650 * String hash function used by SDBM project.
1651 * It was chosen because it is fast and provides
1654 afs_uint32 SDBMHash(const char * str)
1656 afs_uint32 hash = 0;
1662 for(i = 0, len = strlen(str); i < len; i++)
1664 hash = str[i] + (hash << 6) + (hash << 16) - hash;
1667 return (hash & 0x7FFFFFFF);
1670 /* call with volume write-locked and mutex held */
1671 void cm_AddVolumeToNameHashTable(cm_volume_t *volp)
1675 if (volp->qflags & CM_VOLUME_QFLAG_IN_HASH)
1678 i = CM_VOLUME_NAME_HASH(volp->namep);
1680 volp->nameNextp = cm_data.volumeNameHashTablep[i];
1681 cm_data.volumeNameHashTablep[i] = volp;
1682 _InterlockedOr(&volp->qflags, CM_VOLUME_QFLAG_IN_HASH);
1685 /* call with volume write-locked and mutex held */
1686 void cm_RemoveVolumeFromNameHashTable(cm_volume_t *volp)
1688 cm_volume_t **lvolpp;
1692 if (volp->qflags & CM_VOLUME_QFLAG_IN_HASH) {
1693 /* hash it out first */
1694 i = CM_VOLUME_NAME_HASH(volp->namep);
1695 for (lvolpp = &cm_data.volumeNameHashTablep[i], tvolp = cm_data.volumeNameHashTablep[i];
1697 lvolpp = &tvolp->nameNextp, tvolp = tvolp->nameNextp) {
1698 if (tvolp == volp) {
1699 *lvolpp = volp->nameNextp;
1700 _InterlockedAnd(&volp->qflags, ~CM_VOLUME_QFLAG_IN_HASH);
1701 volp->nameNextp = NULL;
1708 /* call with volume write-locked and mutex held */
1709 void cm_AddVolumeToIDHashTable(cm_volume_t *volp, afs_uint32 volType)
1712 struct cm_vol_state * statep;
1714 statep = cm_VolumeStateByType(volp, volType);
1716 if (statep->qflags & CM_VOLUME_QFLAG_IN_HASH)
1719 i = CM_VOLUME_ID_HASH(statep->ID);
1723 statep->nextp = cm_data.volumeRWIDHashTablep[i];
1724 cm_data.volumeRWIDHashTablep[i] = volp;
1727 statep->nextp = cm_data.volumeROIDHashTablep[i];
1728 cm_data.volumeROIDHashTablep[i] = volp;
1731 statep->nextp = cm_data.volumeBKIDHashTablep[i];
1732 cm_data.volumeBKIDHashTablep[i] = volp;
1735 _InterlockedOr(&statep->qflags, CM_VOLUME_QFLAG_IN_HASH);
1739 /* call with volume write-locked and mutex held */
1740 void cm_RemoveVolumeFromIDHashTable(cm_volume_t *volp, afs_uint32 volType)
1742 cm_volume_t **lvolpp;
1744 struct cm_vol_state * statep;
1747 statep = cm_VolumeStateByType(volp, volType);
1749 if (statep->qflags & CM_VOLUME_QFLAG_IN_HASH) {
1750 /* hash it out first */
1751 i = CM_VOLUME_ID_HASH(statep->ID);
1755 lvolpp = &cm_data.volumeRWIDHashTablep[i];
1756 tvolp = cm_data.volumeRWIDHashTablep[i];
1759 lvolpp = &cm_data.volumeROIDHashTablep[i];
1760 tvolp = cm_data.volumeROIDHashTablep[i];
1763 lvolpp = &cm_data.volumeBKIDHashTablep[i];
1764 tvolp = cm_data.volumeBKIDHashTablep[i];
1767 osi_assertx(0, "invalid volume type");
1770 if (tvolp == volp) {
1771 *lvolpp = statep->nextp;
1772 _InterlockedAnd(&statep->qflags, ~CM_VOLUME_QFLAG_IN_HASH);
1773 statep->nextp = NULL;
1777 lvolpp = &tvolp->vol[volType].nextp;
1778 tvolp = tvolp->vol[volType].nextp;
1783 /* must be called with cm_volumeLock write-locked! */
1784 void cm_AdjustVolumeLRU(cm_volume_t *volp)
1786 lock_AssertWrite(&cm_volumeLock);
1788 if (volp == cm_data.volumeLRUFirstp)
1791 if (volp->qflags & CM_VOLUME_QFLAG_IN_LRU_QUEUE)
1792 osi_QRemoveHT((osi_queue_t **) &cm_data.volumeLRUFirstp, (osi_queue_t **) &cm_data.volumeLRULastp, &volp->q);
1793 osi_QAddH((osi_queue_t **) &cm_data.volumeLRUFirstp, (osi_queue_t **) &cm_data.volumeLRULastp, &volp->q);
1794 _InterlockedOr(&volp->qflags, CM_VOLUME_QFLAG_IN_LRU_QUEUE);
1796 osi_assertx(cm_data.volumeLRULastp != NULL, "null cm_data.volumeLRULastp");
1799 /* must be called with cm_volumeLock write-locked! */
1800 void cm_MoveVolumeToLRULast(cm_volume_t *volp)
1802 lock_AssertWrite(&cm_volumeLock);
1804 if (volp == cm_data.volumeLRULastp)
1807 if (volp->qflags & CM_VOLUME_QFLAG_IN_LRU_QUEUE)
1808 osi_QRemoveHT((osi_queue_t **) &cm_data.volumeLRUFirstp, (osi_queue_t **) &cm_data.volumeLRULastp, &volp->q);
1809 osi_QAddT((osi_queue_t **) &cm_data.volumeLRUFirstp, (osi_queue_t **) &cm_data.volumeLRULastp, &volp->q);
1810 _InterlockedOr(&volp->qflags, CM_VOLUME_QFLAG_IN_LRU_QUEUE);
1812 osi_assertx(cm_data.volumeLRULastp != NULL, "null cm_data.volumeLRULastp");
1815 /* must be called with cm_volumeLock write-locked! */
1816 void cm_RemoveVolumeFromLRU(cm_volume_t *volp)
1818 lock_AssertWrite(&cm_volumeLock);
1820 if (volp->qflags & CM_VOLUME_QFLAG_IN_LRU_QUEUE) {
1821 osi_QRemoveHT((osi_queue_t **) &cm_data.volumeLRUFirstp, (osi_queue_t **) &cm_data.volumeLRULastp, &volp->q);
1822 _InterlockedAnd(&volp->qflags, ~CM_VOLUME_QFLAG_IN_LRU_QUEUE);
1825 osi_assertx(cm_data.volumeLRULastp != NULL, "null cm_data.volumeLRULastp");
1828 static char * volstatus_str(enum volstatus vs)
1844 void cm_VolumeStatusNotification(cm_volume_t * volp, afs_uint32 volID, enum volstatus old, enum volstatus new)
1846 char volstr[CELL_MAXNAMELEN + VL_MAXNAMELEN]="";
1849 if (volID == volp->vol[RWVOL].ID)
1851 else if (volID == volp->vol[ROVOL].ID)
1853 else if (volID == volp->vol[BACKVOL].ID)
1857 snprintf(volstr, sizeof(volstr), "%s:%s%s", volp->cellp->name, volp->namep, ext);
1859 osi_Log4(afsd_logp, "VolumeStatusNotification: %-48s [%10u] (%s -> %s)",
1860 osi_LogSaveString(afsd_logp, volstr), volID, volstatus_str(old), volstatus_str(new));
1862 cm_VolStatus_Change_Notification(volp->cellp->cellID, volID, new);
1865 enum volstatus cm_GetVolumeStatus(cm_volume_t *volp, afs_uint32 volID)
1867 cm_vol_state_t * statep = cm_VolumeStateByID(volp, volID);
1869 return statep->state;
1874 /* Renew .readonly volume callbacks that are more than
1875 * 30 minutes old. (A volume callback is issued for 2 hours.)
1878 cm_VolumeRenewROCallbacks(void)
1881 time_t minexp = time(NULL) + 90 * 60;
1882 extern int daemon_ShutdownFlag;
1883 extern int powerStateSuspended;
1885 lock_ObtainRead(&cm_volumeLock);
1886 for (volp = cm_data.allVolumesp;
1887 volp && !daemon_ShutdownFlag && !powerStateSuspended;
1888 volp=volp->allNextp) {
1889 if ( volp->cbExpiresRO > 0 && volp->cbExpiresRO < minexp) {
1894 cm_SetFid(&fid, volp->cellp->cellID, volp->vol[ROVOL].ID, 1, 1);
1898 lock_ReleaseRead(&cm_volumeLock);
1899 if (cm_GetSCache(&fid, NULL, &scp, cm_rootUserp, &req) == 0) {
1900 lock_ObtainWrite(&scp->rw);
1901 cm_GetCallback(scp, cm_rootUserp, &req, 1);
1902 lock_ReleaseWrite(&scp->rw);
1903 cm_ReleaseSCache(scp);
1905 lock_ObtainRead(&cm_volumeLock);
1908 lock_ReleaseRead(&cm_volumeLock);
1912 cm_VolumeStateByType(cm_volume_t *volp, afs_uint32 volType)
1914 return &volp->vol[volType];
1918 cm_VolumeStateByID(cm_volume_t *volp, afs_uint32 id)
1920 cm_vol_state_t * statep = NULL;
1922 if (id == volp->vol[RWVOL].ID)
1923 statep = &volp->vol[RWVOL];
1924 else if (id == volp->vol[ROVOL].ID)
1925 statep = &volp->vol[ROVOL];
1926 else if (id == volp->vol[BACKVOL].ID)
1927 statep = &volp->vol[BACKVOL];
1933 cm_VolumeStateByName(cm_volume_t *volp, char *volname)
1935 size_t len = strlen(volname);
1936 cm_vol_state_t *statep;
1938 if (cm_stricmp_utf8N(".readonly", &volname[len-9]) == 0)
1939 statep = &volp->vol[ROVOL];
1940 else if (cm_stricmp_utf8N(".backup", &volname[len-7]) == 0)
1941 statep = &volp->vol[BACKVOL];
1943 statep = &volp->vol[RWVOL];
1949 cm_VolumeType(cm_volume_t *volp, afs_uint32 id)
1951 if (id == volp->vol[RWVOL].ID)
1953 else if (id == volp->vol[ROVOL].ID)
1955 else if (id == volp->vol[BACKVOL].ID)
1962 cm_ChecksumVolumeServerList(struct cm_fid *fidp, cm_user_t *userp, cm_req_t *reqp)
1966 afs_uint32 replicated;
1967 cm_serverRef_t **serverspp;
1969 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1971 cksum = cm_ChecksumServerList(*serverspp);
1972 cm_FreeServerList(serverspp, 0);
1978 cm_IsVolumeReplicated(cm_fid_t *fidp)
1980 afs_int32 replicated = 0;
1982 cm_vol_state_t * volstatep;
1984 volp = cm_GetVolumeByFID(fidp);
1986 volstatep = cm_VolumeStateByID(volp, fidp->volume);
1987 replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);