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>
23 #include <afs/unified_afs.h>
24 #include <afs/vlserver.h>
25 #include <WINNT/afsreg.h>
27 osi_rwlock_t cm_connLock;
29 DWORD RDRtimeout = CM_CONN_DEFAULTRDRTIMEOUT;
30 unsigned short ConnDeadtimeout = CM_CONN_CONNDEADTIME;
31 unsigned short HardDeadtimeout = CM_CONN_HARDDEADTIME;
32 unsigned short IdleDeadtimeout = CM_CONN_IDLEDEADTIME;
33 unsigned short ReplicaIdleDeadtimeout = CM_CONN_IDLEDEADTIME_REP;
34 unsigned short NatPingInterval = CM_CONN_NATPINGINTERVAL;
36 #define LANMAN_WKS_PARAM_KEY "SYSTEM\\CurrentControlSet\\Services\\lanmanworkstation\\parameters"
37 #define LANMAN_WKS_SESSION_TIMEOUT "SessTimeout"
38 #define LANMAN_WKS_EXT_SESSION_TIMEOUT "ExtendedSessTimeout"
40 afs_uint32 cryptall = 0;
41 afs_uint32 cm_anonvldb = 0;
42 afs_uint32 rx_pmtu_discovery = 0;
44 void cm_PutConn(cm_conn_t *connp)
46 afs_int32 refCount = InterlockedDecrement(&connp->refCount);
47 osi_assertx(refCount >= 0, "cm_conn_t refcount underflow");
50 void cm_InitConn(void)
52 static osi_once_t once;
58 if (osi_Once(&once)) {
59 lock_InitializeRWLock(&cm_connLock, "connection global lock",
60 LOCK_HIERARCHY_CONN_GLOBAL);
62 /* keisa - read timeout value for lanmanworkstation service.
64 * http://support.microsoft.com:80/support/kb/articles/Q102/0/67.asp&NoWebContent=1
65 * the SessTimeout is a minimum timeout not a maximum timeout. Therefore,
66 * I believe that the default should not be short. Instead, we should wait until
67 * RX times out before reporting a timeout to the SMB client.
69 code = RegOpenKeyEx(HKEY_LOCAL_MACHINE, LANMAN_WKS_PARAM_KEY,
70 0, KEY_QUERY_VALUE, &parmKey);
71 if (code == ERROR_SUCCESS)
73 BOOL extTimeouts = msftSMBRedirectorSupportsExtendedTimeouts();
77 * The default value is 1000 seconds. However, this default
78 * will not apply to "\\AFS" unless "AFS" is listed in
79 * ServersWithExtendedSessTimeout which we will add when we
80 * create the ExtendedSessTimeout value in smb_Init()
82 dummyLen = sizeof(DWORD);
83 code = RegQueryValueEx(parmKey,
84 LANMAN_WKS_EXT_SESSION_TIMEOUT,
86 (BYTE *) &dwValue, &dummyLen);
87 if (code == ERROR_SUCCESS) {
89 afsi_log("lanmanworkstation : ExtSessTimeout %u", RDRtimeout);
92 if (!extTimeouts || code != ERROR_SUCCESS) {
93 dummyLen = sizeof(DWORD);
94 code = RegQueryValueEx(parmKey,
95 LANMAN_WKS_SESSION_TIMEOUT,
97 (BYTE *) &dwValue, &dummyLen);
98 if (code == ERROR_SUCCESS) {
100 afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
103 RegCloseKey(parmKey);
106 code = RegOpenKeyEx(HKEY_LOCAL_MACHINE, AFSREG_CLT_SVC_PARAM_SUBKEY,
107 0, KEY_QUERY_VALUE, &parmKey);
108 if (code == ERROR_SUCCESS) {
109 dummyLen = sizeof(DWORD);
110 code = RegQueryValueEx(parmKey, "ConnDeadTimeout", NULL, NULL,
111 (BYTE *) &dwValue, &dummyLen);
112 if (code == ERROR_SUCCESS) {
113 ConnDeadtimeout = (unsigned short)dwValue;
114 afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
116 dummyLen = sizeof(DWORD);
117 code = RegQueryValueEx(parmKey, "HardDeadTimeout", NULL, NULL,
118 (BYTE *) &dwValue, &dummyLen);
119 if (code == ERROR_SUCCESS) {
120 HardDeadtimeout = (unsigned short)dwValue;
121 afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
123 dummyLen = sizeof(DWORD);
124 code = RegQueryValueEx(parmKey, "IdleDeadTimeout", NULL, NULL,
125 (BYTE *) &dwValue, &dummyLen);
126 if (code == ERROR_SUCCESS) {
127 IdleDeadtimeout = (unsigned short)dwValue;
128 afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
130 dummyLen = sizeof(DWORD);
131 code = RegQueryValueEx(parmKey, "ReplicaIdleDeadTimeout", NULL, NULL,
132 (BYTE *) &dwValue, &dummyLen);
133 if (code == ERROR_SUCCESS) {
134 ReplicaIdleDeadtimeout = (unsigned short)dwValue;
135 afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
137 dummyLen = sizeof(DWORD);
138 code = RegQueryValueEx(parmKey, "NatPingInterval", NULL, NULL,
139 (BYTE *) &dwValue, &dummyLen);
140 if (code == ERROR_SUCCESS) {
141 NatPingInterval = (unsigned short)dwValue;
143 afsi_log("NatPingInterval is %d", NatPingInterval);
144 RegCloseKey(parmKey);
148 * If these values were not set via cpp macro or obtained
149 * from the registry, we use a value that is derived from
150 * the smb redirector session timeout (RDRtimeout).
152 * The UNIX cache manager uses 120 seconds for the hard dead
153 * timeout and 50 seconds for the connection and idle timeouts.
155 * We base our values on those while making sure we leave
156 * enough time for overhead.
158 * To further complicate matters we need to take into account
159 * file server hard dead timeouts as they affect the length
160 * of time it takes the file server to give up when attempting
161 * to break callbacks to unresponsive clients. The file
162 * server hard dead timeout is 120 seconds.
164 * For SMB, we have no choice but to timeout quickly. For
165 * the AFS redirector, we can wait.
168 afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
169 if (ConnDeadtimeout == 0) {
170 ConnDeadtimeout = (unsigned short) ((RDRtimeout / 2) < 50 ? (RDRtimeout / 2) : 50);
171 afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
173 if (HardDeadtimeout == 0) {
174 HardDeadtimeout = (unsigned short) (RDRtimeout > 125 ? 120 : (RDRtimeout - 5));
175 afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
177 if (IdleDeadtimeout == 0) {
178 IdleDeadtimeout = 10 * (unsigned short) HardDeadtimeout;
179 afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
181 if (ReplicaIdleDeadtimeout == 0) {
182 ReplicaIdleDeadtimeout = (unsigned short) HardDeadtimeout;
183 afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
186 if (ConnDeadtimeout == 0) {
187 ConnDeadtimeout = CM_CONN_IFS_CONNDEADTIME;
188 afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
190 if (HardDeadtimeout == 0) {
191 HardDeadtimeout = CM_CONN_IFS_HARDDEADTIME;
192 afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
194 if (IdleDeadtimeout == 0) {
195 IdleDeadtimeout = CM_CONN_IFS_IDLEDEADTIME;
196 afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
198 if (IdleDeadtimeout == 0) {
199 ReplicaIdleDeadtimeout = CM_CONN_IFS_IDLEDEADTIME_REP;
200 afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
207 void cm_InitReq(cm_req_t *reqp)
209 memset(reqp, 0, sizeof(cm_req_t));
210 reqp->startTime = GetTickCount();
213 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
214 struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
217 cm_volume_t *volp = NULL;
218 cm_vol_state_t *volstatep = NULL;
219 cm_cell_t *cellp = NULL;
223 return CM_ERROR_INVAL;
226 cellp = cm_FindCellByID(fidp->cell, 0);
228 return CM_ERROR_NOSUCHCELL;
230 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
234 volstatep = cm_VolumeStateByID(volp, fidp->volume);
235 *replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
236 *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp);
238 lock_ObtainRead(&cm_volumeLock);
240 lock_ReleaseRead(&cm_volumeLock);
241 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
245 cm_SetServerBusyStatus(cm_serverRef_t *serversp, cm_server_t *serverp)
247 cm_serverRef_t *tsrp;
249 lock_ObtainWrite(&cm_serverLock);
250 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
251 if (tsrp->status == srv_deleted)
253 if (cm_ServerEqual(tsrp->server, serverp) && tsrp->status == srv_not_busy) {
254 tsrp->status = srv_busy;
258 lock_ReleaseWrite(&cm_serverLock);
262 cm_ResetServerBusyStatus(cm_serverRef_t *serversp)
264 cm_serverRef_t *tsrp;
266 lock_ObtainWrite(&cm_serverLock);
267 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
268 if (tsrp->status == srv_deleted)
270 if (tsrp->status == srv_busy) {
271 tsrp->status = srv_not_busy;
274 lock_ReleaseWrite(&cm_serverLock);
278 * Analyze the error return from an RPC. Determine whether or not to retry,
279 * and if we're going to retry, determine whether failover is appropriate,
280 * and whether timed backoff is appropriate.
282 * If the error code is from cm_ConnFromFID() or friends, it will be a CM_ERROR code.
283 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
284 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
285 * be a security code (e.g. RXKADEXPIRED).
287 * If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
289 * For VLDB calls, fidp will be NULL and cellp will not be.
291 * volSyncp and/or cbrp may also be NULL.
294 cm_Analyze(cm_conn_t *connp,
300 AFSVolSync *volSyncp,
301 cm_serverRef_t * serversp,
302 cm_callbackRequest_t *cbrp,
305 cm_server_t *serverp = NULL;
306 cm_serverRef_t **serverspp = NULL;
307 cm_serverRef_t *tsrp;
309 cm_volume_t * volp = NULL;
310 cm_vol_state_t *statep = NULL;
311 cm_scache_t * scp = NULL;
312 afs_uint32 replicated;
314 int free_svr_list = 0;
316 long timeUsed, timeLeft;
318 char addr[16]="unknown";
320 int location_updated = 0;
324 osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
327 /* no locking required, since connp->serverp never changes after
329 dead_session = (userp->cellInfop == NULL);
331 serverp = connp->serverp;
333 /* Update callback pointer */
334 if (cbrp && serverp && errorCode == 0) {
336 if ( cbrp->serverp != serverp ) {
337 lock_ObtainWrite(&cm_serverLock);
338 cm_PutServerNoLock(cbrp->serverp);
339 cm_GetServerNoLock(serverp);
340 lock_ReleaseWrite(&cm_serverLock);
343 cm_GetServer(serverp);
345 cbrp->serverp = serverp;
348 /* if timeout - check that it did not exceed the HardDead timeout
351 /* timeleft - get it from reqp the same way as cm_ConnByMServers does */
352 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
353 if ( reqp->flags & CM_REQ_SOURCE_SMB )
354 timeLeft = HardDeadtimeout - timeUsed;
356 timeLeft = 0x0FFFFFFF;
358 /* get a pointer to the cell */
360 if (cellp == NULL && serverp)
361 cellp = serverp->cellp;
362 if (cellp == NULL && serversp) {
363 struct cm_serverRef * refp;
364 for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
365 if (refp->status == srv_deleted)
368 cellp = refp->server->cellp;
371 if (cellp == NULL && fidp) {
372 cellp = cm_FindCellByID(fidp->cell, 0);
376 if (errorCode == CM_ERROR_TIMEDOUT) {
377 if ( timeLeft > 5 ) {
379 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
384 else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
385 errorCode == UAEAGAIN || errorCode == EAGAIN) {
386 osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
387 if ( timeLeft > 5 ) {
393 /* if there is nosuchvolume, then we have a situation in which a
394 * previously known volume no longer has a set of servers
395 * associated with it. Either the volume has moved or
396 * the volume has been deleted. Try to find a new server list
397 * until the timeout period expires.
399 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
400 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
402 * The VNOVOL or VL_NOENT error has already been translated
403 * to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
407 else if (errorCode == CM_ERROR_EMPTY) {
409 * The server list is empty (or all entries have been deleted).
410 * If fidp is NULL, this was a vlServer list and we can attempt
411 * to force a cell lookup. If fidp is not NULL, we can attempt
412 * to refresh the volume location list.
415 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
416 CM_GETVOL_FLAG_NO_LRU_UPDATE,
419 if (cm_UpdateVolumeLocation(cellp, userp, reqp, volp) == 0) {
420 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
422 if (!cm_IsServerListEmpty(*serverspp))
424 cm_FreeServerList(serverspp, 0);
428 lock_ObtainRead(&cm_volumeLock);
430 lock_ReleaseRead(&cm_volumeLock);
434 cm_cell_t * newCellp = cm_UpdateCell( cellp, 0);
439 else if (errorCode == CM_ERROR_ALLDOWN) {
440 /* Servers marked DOWN will be restored by the background daemon
441 * thread as they become available. The volume status is
442 * updated as the server state changes.
445 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
446 cellp->name, fidp->volume);
447 msgID = MSG_ALL_SERVERS_DOWN;
448 format = "All servers are unreachable when accessing cell %s volume %d.";
449 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
451 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
454 else if (errorCode == CM_ERROR_ALLOFFLINE) {
455 /* Volume instances marked offline will be restored by the
456 * background daemon thread as they become available
459 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
460 cellp->name, fidp->volume);
461 msgID = MSG_ALL_SERVERS_OFFLINE;
462 format = "All servers are offline when accessing cell %s volume %d.";
463 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
466 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
468 serversp = *serverspp;
472 cm_ResetServerBusyStatus(serversp);
474 cm_FreeServerList(serverspp, 0);
479 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
480 CM_GETVOL_FLAG_NO_LRU_UPDATE,
484 * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
485 * was called by cm_CheckOfflineVolumeState().
487 if (!(reqp->flags & CM_REQ_OFFLINE_VOL_CHK) && timeLeft > 7) {
490 /* cm_CheckOfflineVolume() resets the serverRef state */
491 if (cm_CheckOfflineVolume(volp, fidp->volume))
494 cm_UpdateVolumeStatus(volp, fidp->volume);
496 lock_ObtainRead(&cm_volumeLock);
498 lock_ReleaseRead(&cm_volumeLock);
502 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
505 else if (errorCode == CM_ERROR_ALLBUSY) {
506 /* Volumes that are busy cannot be determined to be non-busy
507 * without actually attempting to access them.
509 if (fidp) { /* File Server query */
510 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
511 cellp->name, fidp->volume);
512 msgID = MSG_ALL_SERVERS_BUSY;
513 format = "All servers are busy when accessing cell %s volume %d.";
514 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
517 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
519 serversp = *serverspp;
523 cm_ResetServerBusyStatus(serversp);
525 cm_FreeServerList(serverspp, 0);
530 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
531 CM_GETVOL_FLAG_NO_LRU_UPDATE,
536 statep = cm_VolumeStateByID(volp, fidp->volume);
539 cm_UpdateVolumeStatus(volp, fidp->volume);
541 lock_ObtainRead(&cm_volumeLock);
543 lock_ReleaseRead(&cm_volumeLock);
546 } else { /* VL Server query */
547 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
553 cm_ResetServerBusyStatus(serversp);
560 /* special codes: VBUSY and VRESTARTING */
561 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
562 if (!serversp && fidp) {
563 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
565 serversp = *serverspp;
570 switch ( errorCode ) {
572 msgID = MSG_SERVER_REPORTS_VBUSY;
573 format = "Server %s reported busy when accessing volume %d in cell %s.";
576 msgID = MSG_SERVER_REPORTS_VRESTARTING;
577 format = "Server %s reported restarting when accessing volume %d in cell %s.";
581 if (serverp && fidp) {
582 /* Log server being offline for this volume */
583 sprintf(addr, "%d.%d.%d.%d",
584 ((serverp->addr.sin_addr.s_addr & 0xff)),
585 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
586 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
587 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
589 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
590 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
593 cm_SetServerBusyStatus(serversp, serverp);
595 if (fidp) { /* File Server query */
596 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
597 CM_GETVOL_FLAG_NO_LRU_UPDATE,
600 statep = cm_VolumeStateByID(volp, fidp->volume);
603 cm_UpdateVolumeStatus(volp, statep->ID);
605 lock_ObtainRead(&cm_volumeLock);
607 lock_ReleaseRead(&cm_volumeLock);
613 cm_FreeServerList(serverspp, 0);
620 /* special codes: missing volumes */
621 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
622 errorCode == VSALVAGE || errorCode == VIO)
624 /* In case of timeout */
625 reqp->volumeError = errorCode;
627 switch ( errorCode ) {
629 msgID = MSG_SERVER_REPORTS_VNOVOL;
630 format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
633 msgID = MSG_SERVER_REPORTS_VMOVED;
634 format = "Server %s reported volume %d in cell %s as moved.";
637 msgID = MSG_SERVER_REPORTS_VOFFLINE;
638 format = "Server %s reported volume %d in cell %s as offline.";
641 msgID = MSG_SERVER_REPORTS_VSALVAGE;
642 format = "Server %s reported volume %d in cell %s as needs salvage.";
645 msgID = MSG_SERVER_REPORTS_VIO;
646 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
650 if (fidp) { /* File Server query */
652 /* Log server being unavailable for this volume */
653 sprintf(addr, "%d.%d.%d.%d",
654 ((serverp->addr.sin_addr.s_addr & 0xff)),
655 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
656 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
657 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
659 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
660 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
663 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
664 CM_GETVOL_FLAG_NO_LRU_UPDATE,
667 statep = cm_VolumeStateByID(volp, fidp->volume);
669 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
670 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
672 LONG_PTR oldSum, newSum;
674 oldSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
676 code = cm_ForceUpdateVolume(fidp, userp, reqp);
678 location_updated = 1;
679 newSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
683 * Even if the update fails, there might still be another replica.
684 * If the volume location list changed, permit another update on
685 * a subsequent error.
687 if (code || oldSum == newSum)
688 reqp->flags |= CM_REQ_VOLUME_UPDATED;
690 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
691 fidp->cell, fidp->volume, code);
695 cm_UpdateVolumeStatus(volp, statep->ID);
696 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
697 fidp->cell, fidp->volume, statep->state);
701 lock_ObtainRead(&cm_volumeLock);
703 lock_ReleaseRead(&cm_volumeLock);
708 * Mark server offline for this volume or delete the volume
709 * from the server list if it was moved or is not present.
711 if (!serversp || location_updated) {
712 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
714 serversp = *serverspp;
720 lock_ObtainWrite(&cm_serverLock);
721 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
722 if (tsrp->status == srv_deleted)
725 sprintf(addr, "%d.%d.%d.%d",
726 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
727 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
728 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
729 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
731 if (cm_ServerEqual(tsrp->server, serverp)) {
735 osi_Log2(afsd_logp, "volume %u moved from server %s",
736 fidp->volume, osi_LogSaveString(afsd_logp,addr));
737 tsrp->status = srv_deleted;
739 cm_RemoveVolumeFromServer(serverp, fidp->volume);
743 * The 1.6.0 and 1.6.1 file servers send transient VNOVOL errors which
744 * are no indicative of the volume not being present. For example,
745 * VNOVOL can be sent during a transition to a VBUSY state prior to
746 * salvaging or when cloning a .backup volume instance. As a result
747 * the cache manager must attempt at least one retry when a VNOVOL is
748 * receive but there are no changes to the volume location information.
750 if (reqp->vnovolError > 0 && cm_ServerEqual(reqp->errorServp, serverp)) {
751 osi_Log2(afsd_logp, "volume %u not present on server %s",
752 fidp->volume, osi_LogSaveString(afsd_logp,addr));
753 tsrp->status = srv_deleted;
755 cm_RemoveVolumeFromServer(serverp, fidp->volume);
757 osi_Log2(afsd_logp, "VNOVOL received for volume %u from server %s",
758 fidp->volume, osi_LogSaveString(afsd_logp,addr));
760 if (tsrp->status == srv_not_busy)
761 tsrp->status = srv_busy;
768 osi_Log3(afsd_logp, "volume %u exists on server %s with status %u",
769 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
773 lock_ReleaseWrite(&cm_serverLock);
775 /* Remember that the VNOVOL error occurred */
776 if (errorCode == VNOVOL) {
777 reqp->errorServp = serverp;
781 /* Free the server list before cm_ForceUpdateVolume is called */
783 cm_FreeServerList(serverspp, 0);
790 } else if ( errorCode == VNOVNODE ) {
792 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
793 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
795 scp = cm_FindSCache(fidp);
797 cm_scache_t *pscp = NULL;
799 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
800 pscp = cm_FindSCacheParent(scp);
802 lock_ObtainWrite(&scp->rw);
803 scp->flags |= CM_SCACHEFLAG_DELETED;
804 lock_ObtainWrite(&cm_scacheLock);
805 cm_AdjustScacheLRU(scp);
806 cm_RemoveSCacheFromHashTable(scp);
807 lock_ReleaseWrite(&cm_scacheLock);
808 cm_LockMarkSCacheLost(scp);
809 lock_ReleaseWrite(&scp->rw);
811 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
812 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
813 cm_ReleaseSCache(scp);
816 if (cm_HaveCallback(pscp)) {
817 lock_ObtainWrite(&pscp->rw);
818 cm_DiscardSCache(pscp);
819 lock_ReleaseWrite(&pscp->rw);
822 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
823 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
826 cm_ReleaseSCache(pscp);
830 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
835 else if (errorCode == RX_CALL_TIMEOUT) {
836 /* RPC took longer than hardDeadTime or the server
837 * reported idle for longer than idleDeadTime
838 * don't mark server as down but don't retry
839 * this is to prevent the SMB session from timing out
840 * In addition, we log an event to the event log
844 sprintf(addr, "%d.%d.%d.%d",
845 ((serverp->addr.sin_addr.s_addr & 0xff)),
846 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
847 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
848 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
850 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
851 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
852 osi_LogSaveString(afsd_logp,addr));
853 reqp->errorServp = serverp;
858 scp = cm_FindSCache(fidp);
860 if (cm_HaveCallback(scp)) {
861 lock_ObtainWrite(&scp->rw);
862 cm_DiscardSCache(scp);
863 lock_ReleaseWrite(&scp->rw);
866 * We really should notify the redirector that we discarded
867 * the status information but doing so in this case is not
868 * safe as it can result in a deadlock with extent release
872 cm_ReleaseSCache(scp);
876 if (!fidp) { /* vldb */
879 cm_volume_t *volp = cm_GetVolumeByFID(fidp);
881 if (fidp->volume == cm_GetROVolumeID(volp))
888 else if (errorCode == RX_MSGSIZE) {
890 * RPC failed because a transmitted rx packet
891 * may have grown larger than the path mtu.
892 * Force a retry and the Rx library will try
893 * with a smaller mtu size.
897 sprintf(addr, "%d.%d.%d.%d",
898 ((serverp->addr.sin_addr.s_addr & 0xff)),
899 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
900 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
901 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
903 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
904 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
905 osi_LogSaveString(afsd_logp,addr));
909 else if (errorCode == RX_CALL_BUSY) {
911 * RPC failed because the selected call channel
912 * is currently busy on the server. Unconditionally
913 * retry the request so an alternate call channel can be used.
916 sprintf(addr, "%d.%d.%d.%d",
917 ((serverp->addr.sin_addr.s_addr & 0xff)),
918 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
919 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
920 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
922 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
923 osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
924 osi_LogSaveString(afsd_logp,addr));
927 else if (errorCode == VNOSERVICE) {
929 * The server did not service the RPC.
930 * If this was a file server RPC it means that for at
931 * least the file server's idle dead timeout period the
932 * file server did not receive any new data packets from
935 * The RPC was not serviced so it can be retried and any
936 * existing status information is still valid.
940 sprintf(addr, "%d.%d.%d.%d",
941 ((serverp->addr.sin_addr.s_addr & 0xff)),
942 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
943 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
944 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
946 LogEvent(EVENTLOG_WARNING_TYPE, MSG_SERVER_REPORTS_VNOSERVICE,
947 addr, fidp->volume, cellp->name);
948 osi_Log3(afsd_logp, "Server %s reported rpc to volume %d in cell %s as not serviced.",
949 osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
955 else if (errorCode == RX_CALL_IDLE) {
957 * RPC failed because the server failed to respond with data
958 * within the idle dead timeout period. This could be for a variety
960 * 1. The server could have a bad partition such as a failed
961 * disk or iSCSI target and all I/O to that partition is
962 * blocking on the server and will never complete.
964 * 2. The server vnode may be locked by another client request
965 * that is taking a very long time.
967 * 3. The server may have a very long queue of requests
968 * pending and is unable to process this request.
970 * 4. The server could be malicious and is performing a denial
971 * of service attack against the client.
973 * If this is a request against a .readonly with alternate sites
974 * the server should be marked down for this request and the
975 * client should fail over to another server. If this is a
976 * request against a single source, the client may retry once.
979 sprintf(addr, "%d.%d.%d.%d",
980 ((serverp->addr.sin_addr.s_addr & 0xff)),
981 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
982 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
983 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
986 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
987 CM_GETVOL_FLAG_NO_LRU_UPDATE,
990 statep = cm_VolumeStateByID(volp, fidp->volume);
993 replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
995 lock_ObtainRead(&cm_volumeLock);
997 lock_ReleaseRead(&cm_volumeLock);
1002 scp = cm_FindSCache(fidp);
1004 if (cm_HaveCallback(scp)) {
1005 lock_ObtainWrite(&scp->rw);
1006 cm_DiscardSCache(scp);
1007 lock_ReleaseWrite(&scp->rw);
1010 * We really should notify the redirector that we discarded
1011 * the status information but doing so in this case is not
1012 * safe as it can result in a deadlock with extent release
1016 cm_ReleaseSCache(scp);
1020 if (replicated && serverp) {
1021 reqp->errorServp = serverp;
1022 reqp->tokenError = errorCode;
1028 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
1029 osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
1030 osi_LogSaveString(afsd_logp,addr), retry);
1032 else if (errorCode == RX_CALL_DEAD) {
1033 /* mark server as down */
1035 sprintf(addr, "%d.%d.%d.%d",
1036 ((serverp->addr.sin_addr.s_addr & 0xff)),
1037 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1038 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1039 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1041 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
1042 osi_LogSaveString(afsd_logp,addr),
1043 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1046 if ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
1047 lock_ObtainMutex(&serverp->mx);
1048 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
1049 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
1050 serverp->downTime = time(NULL);
1052 lock_ReleaseMutex(&serverp->mx);
1054 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1056 cm_ForceNewConnections(serverp);
1060 if (fidp && storeOp)
1061 scp = cm_FindSCache(fidp);
1063 if (cm_HaveCallback(scp)) {
1064 lock_ObtainWrite(&scp->rw);
1065 cm_DiscardSCache(scp);
1066 lock_ReleaseWrite(&scp->rw);
1069 * We really should notify the redirector that we discarded
1070 * the status information but doing so in this case is not
1071 * safe as it can result in a deadlock with extent release
1075 cm_ReleaseSCache(scp);
1081 else if (errorCode >= -64 && errorCode < 0) {
1082 /* mark server as down */
1084 sprintf(addr, "%d.%d.%d.%d",
1085 ((serverp->addr.sin_addr.s_addr & 0xff)),
1086 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1087 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1088 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1090 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
1092 osi_LogSaveString(afsd_logp,addr),
1093 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1096 if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
1097 reqp->errorServp = serverp;
1098 reqp->tokenError = errorCode;
1100 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1102 cm_ForceNewConnections(serverp);
1108 else if (errorCode == RXKADEXPIRED) {
1109 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1111 if (!dead_session) {
1112 lock_ObtainMutex(&userp->mx);
1113 ucellp = cm_GetUCell(userp, cellp);
1114 if (ucellp->ticketp) {
1115 free(ucellp->ticketp);
1116 ucellp->ticketp = NULL;
1118 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1120 lock_ReleaseMutex(&userp->mx);
1124 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1125 char * s = "unknown error";
1126 switch ( errorCode ) {
1127 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1128 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1129 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1130 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1131 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1132 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1133 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1134 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1135 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1136 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1137 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1138 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1139 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1141 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1145 reqp->errorServp = serverp;
1146 reqp->tokenError = errorCode;
1149 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1151 * We received a ubik error. its possible that the server we are
1152 * communicating with has a corrupted database or is partitioned
1153 * from the rest of the servers and another server might be able
1154 * to answer our query. Therefore, we will retry the request
1155 * and force the use of another server.
1158 reqp->errorServp = serverp;
1159 reqp->tokenError = errorCode;
1162 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1163 cm_ForceNewConnections(serverp);
1168 char * s = "unknown error";
1169 switch ( errorCode ) {
1170 case VSALVAGE : s = "VSALVAGE"; break;
1171 case VNOVNODE : s = "VNOVNODE"; break;
1172 case VNOVOL : s = "VNOVOL"; break;
1173 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1174 case VNOSERVICE : s = "VNOSERVICE"; break;
1175 case VOFFLINE : s = "VOFFLINE"; break;
1176 case VONLINE : s = "VONLINE"; break;
1177 case VDISKFULL : s = "VDISKFULL"; break;
1178 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1179 case VBUSY : s = "VBUSY"; break;
1180 case VMOVED : s = "VMOVED"; break;
1181 case VIO : s = "VIO"; break;
1182 case VRESTRICTED : s = "VRESTRICTED"; break;
1183 case VRESTARTING : s = "VRESTARTING"; break;
1184 case VREADONLY : s = "VREADONLY"; break;
1185 case EAGAIN : s = "EAGAIN"; break;
1186 case UAEAGAIN : s = "UAEAGAIN"; break;
1187 case EINVAL : s = "EINVAL"; break;
1188 case UAEINVAL : s = "UAEINVAL"; break;
1189 case EACCES : s = "EACCES"; break;
1190 case UAEACCES : s = "UAEACCES"; break;
1191 case ENOENT : s = "ENOENT"; break;
1192 case UAENOENT : s = "UAENOENT"; break;
1193 case EEXIST : s = "EEXIST"; break;
1194 case UAEEXIST : s = "UAEEXIST"; break;
1195 case VICECONNBAD : s = "VICECONNBAD"; break;
1196 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1197 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1198 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1199 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1200 case VL_IO : s = "VL_IO"; break;
1201 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1202 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1203 case VL_NOENT : s = "VL_NOENT"; break;
1204 case VL_EMPTY : s = "VL_EMPTY"; break;
1205 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1206 case VL_BADNAME : s = "VL_BADNAME"; break;
1207 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1208 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1209 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1210 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1211 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1212 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1213 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1214 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1215 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1216 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1217 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1218 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1219 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1220 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1221 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1222 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1223 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1224 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1225 case VL_PERM : s = "VL_PERM"; break;
1226 case VL_NOMEM : s = "VL_NOMEM"; break;
1227 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1228 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1229 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1230 case VL_BADMASK : s = "VL_BADMASK"; break;
1231 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1232 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1233 case CM_ERROR_TIMEDOUT : s = "CM_ERROR_TIMEDOUT"; break;
1234 case CM_ERROR_RETRY : s = "CM_ERROR_RETRY"; break;
1235 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1236 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1237 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1238 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1239 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1240 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1241 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1242 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1243 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1244 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1245 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1246 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1247 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1248 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1249 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1250 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1251 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1252 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1253 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1254 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1255 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1256 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1257 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1258 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1259 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1260 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1261 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1262 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1263 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1264 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1265 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1266 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1267 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1268 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1269 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1270 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1271 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1272 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1273 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1274 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1275 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1276 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1277 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1278 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1279 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1280 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1281 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1283 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1289 /* If not allowed to retry, don't */
1290 if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
1291 (errorCode != RX_MSGSIZE && errorCode != RX_CALL_BUSY))
1293 else if (retry && dead_session)
1296 /* drop this on the way out */
1301 * clear the volume updated flag if we succeed.
1302 * this way the flag will not prevent a subsequent volume
1303 * from being updated if necessary.
1307 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1311 errorCode != VBUSY &&
1312 errorCode != VRESTARTING &&
1313 errorCode != CM_ERROR_ALLBUSY)
1315 cm_ResetServerBusyStatus(serversp);
1318 /* retry until we fail to find a connection */
1322 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1323 cm_req_t *reqp, cm_conn_t **connpp)
1326 cm_serverRef_t *tsrp;
1328 long firstError = 0;
1329 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1, allDeleted = 1;
1330 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1331 long timeUsed, timeLeft, hardTimeLeft;
1335 if (serversp == NULL) {
1336 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_EMPTY);
1337 return CM_ERROR_EMPTY;
1340 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1341 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1343 /* leave 5 seconds margin of safety */
1344 timeLeft = ConnDeadtimeout - timeUsed - 5;
1345 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1348 lock_ObtainRead(&cm_serverLock);
1349 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1350 if (tsrp->status == srv_deleted)
1356 if (reqp->errorServp) {
1358 * search the list until we find the server
1359 * that failed last time. When we find it
1360 * clear the error, skip it and try the next one
1363 if (tsp == reqp->errorServp)
1364 reqp->errorServp = NULL;
1368 cm_GetServerNoLock(tsp);
1369 lock_ReleaseRead(&cm_serverLock);
1370 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1372 if (tsrp->status == srv_busy) {
1375 } else if (tsrp->status == srv_offline) {
1381 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1382 if (code == 0) { /* cm_CBS only returns 0 */
1384 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1385 /* Set RPC timeout */
1386 if (timeLeft > ConnDeadtimeout)
1387 timeLeft = ConnDeadtimeout;
1389 if (hardTimeLeft > HardDeadtimeout)
1390 hardTimeLeft = HardDeadtimeout;
1392 lock_ObtainMutex(&(*connpp)->mx);
1393 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1394 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1395 lock_ReleaseMutex(&(*connpp)->mx);
1400 /* therefore, this code is never executed */
1401 if (firstError == 0)
1405 lock_ObtainRead(&cm_serverLock);
1406 cm_PutServerNoLock(tsp);
1409 lock_ReleaseRead(&cm_serverLock);
1411 if (firstError == 0) {
1413 firstError = CM_ERROR_EMPTY;
1414 } else if (allDown) {
1415 firstError = (reqp->tokenError ? reqp->tokenError :
1416 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1418 * if we experienced either a token error or and idle dead time error
1419 * and now all of the servers are down, we have either tried them
1420 * all or lost connectivity. Clear the error we are returning so
1421 * we will not return it indefinitely if the request is retried.
1423 reqp->idleError = reqp->tokenError = 0;
1424 } else if (allBusy) {
1425 firstError = CM_ERROR_ALLBUSY;
1426 } else if (allOffline || (someBusy && someOffline)) {
1427 firstError = CM_ERROR_ALLOFFLINE;
1429 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1430 firstError = CM_ERROR_TIMEDOUT;
1434 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1438 /* called with a held server to GC all bad connections hanging off of the server */
1439 void cm_GCConnections(cm_server_t *serverp)
1445 lock_ObtainWrite(&cm_connLock);
1446 lcpp = &serverp->connsp;
1447 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1449 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1450 /* do the deletion of this guy */
1451 cm_PutServer(tcp->serverp);
1452 cm_ReleaseUser(userp);
1454 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1455 rx_DestroyConnection(tcp->rxconnp);
1456 lock_FinalizeMutex(&tcp->mx);
1460 /* just advance to the next */
1464 lock_ReleaseWrite(&cm_connLock);
1467 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1468 cm_server_t *serverp, afs_uint32 replicated)
1470 unsigned short port;
1473 struct rx_securityClass *secObjp;
1475 port = serverp->addr.sin_port;
1476 switch (serverp->type) {
1477 case CM_SERVER_VLDB:
1482 case CM_SERVER_FILE:
1488 osi_panic("unknown server type", __FILE__, __LINE__);
1491 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1495 tcp->cryptlevel = rxkad_clear;
1498 tcp->cryptlevel = rxkad_auth;
1501 tcp->cryptlevel = rxkad_crypt;
1503 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1504 &ucellp->sessionKey, ucellp->kvno,
1505 ucellp->ticketLen, ucellp->ticketp);
1509 tcp->cryptlevel = rxkad_clear;
1510 secObjp = rxnull_NewClientSecurityObject();
1512 osi_assertx(secObjp != NULL, "null rx_securityClass");
1513 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1518 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1519 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1522 * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
1525 tcp->flags &= CM_CONN_FLAG_REPLICATION;
1526 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1528 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1532 * Let the Rx library know that we can auto-retry if an
1533 * RX_MSGSIZE error is returned.
1535 if (rx_pmtu_discovery)
1536 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1539 * Attempt to limit NAT pings to the anonymous file server connections.
1540 * Only file servers implement client callbacks and we only need one ping
1541 * to be sent to each server.
1543 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1544 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1545 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1548 tcp->ucgen = ucellp->gen;
1550 rxs_Release(secObjp); /* Decrement the initial refCount */
1553 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1560 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1561 userp = cm_rootUserp;
1563 lock_ObtainMutex(&userp->mx);
1564 lock_ObtainRead(&cm_connLock);
1565 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1566 if (tcp->userp == userp &&
1567 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1568 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1572 /* find ucell structure */
1573 ucellp = cm_GetUCell(userp, serverp->cellp);
1575 lock_ConvertRToW(&cm_connLock);
1576 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1577 if (tcp->userp == userp)
1581 lock_ReleaseWrite(&cm_connLock);
1584 cm_GetServer(serverp);
1585 tcp = malloc(sizeof(*tcp));
1586 memset(tcp, 0, sizeof(*tcp));
1587 tcp->nextp = serverp->connsp;
1588 serverp->connsp = tcp;
1591 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1592 lock_ObtainMutex(&tcp->mx);
1593 tcp->serverp = serverp;
1594 tcp->cryptlevel = rxkad_clear;
1595 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1597 lock_ReleaseMutex(&tcp->mx);
1598 lock_ReleaseWrite(&cm_connLock);
1600 lock_ReleaseRead(&cm_connLock);
1602 InterlockedIncrement(&tcp->refCount);
1604 lock_ObtainMutex(&tcp->mx);
1605 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1606 (tcp->ucgen < ucellp->gen) ||
1607 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1609 if (tcp->ucgen < ucellp->gen)
1610 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1612 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1613 tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
1614 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1615 rx_DestroyConnection(tcp->rxconnp);
1616 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1618 lock_ReleaseMutex(&tcp->mx);
1620 lock_ReleaseMutex(&userp->mx);
1622 /* return this pointer to our caller */
1623 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1629 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1633 cm_serverRef_t **serverspp;
1634 cm_serverRef_t *tsrp;
1636 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1637 afs_uint32 replicated;
1641 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1645 lock_ObtainRead(&cm_serverLock);
1646 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1647 if (tsrp->status == srv_deleted)
1650 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1652 if (tsrp->status == srv_busy) {
1655 } else if (tsrp->status == srv_offline) {
1664 lock_ReleaseRead(&cm_serverLock);
1665 cm_FreeServerList(serverspp, 0);
1671 else if (allOffline || (someBusy && someOffline))
1678 * The returned cm_conn_t ** object is released in the subsequent call
1681 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1685 cm_serverRef_t **serverspp;
1686 afs_uint32 replicated;
1690 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1694 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1695 cm_FreeServerList(serverspp, 0);
1700 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1704 cm_serverRef_t **serverspp;
1705 afs_uint32 replicated;
1706 cm_vol_state_t * volstatep;
1710 volstatep = cm_VolumeStateByID(volp, volid);
1711 replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1712 serverspp = cm_GetVolServers(volp, volid, userp, reqp);
1714 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1715 cm_FreeServerList(serverspp, 0);
1720 extern struct rx_connection *
1721 cm_GetRxConn(cm_conn_t *connp)
1723 struct rx_connection * rxconnp;
1724 lock_ObtainMutex(&connp->mx);
1725 rxconnp = connp->rxconnp;
1726 rx_GetConnection(rxconnp);
1727 lock_ReleaseMutex(&connp->mx);
1731 void cm_ForceNewConnections(cm_server_t *serverp)
1735 lock_ObtainWrite(&cm_connLock);
1736 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1737 lock_ObtainMutex(&tcp->mx);
1738 tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
1739 lock_ReleaseMutex(&tcp->mx);
1741 lock_ReleaseWrite(&cm_connLock);