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_GetVolServerList(cm_volume_t *volp, afs_uint32 volid, struct cm_user *userp,
214 struct cm_req *reqp, afs_uint32 *replicated,
215 cm_serverRef_t ***serversppp)
217 *serversppp = cm_GetVolServers(volp, volid, userp, reqp, replicated);
218 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
221 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
222 struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
225 cm_volume_t *volp = NULL;
226 cm_cell_t *cellp = NULL;
230 return CM_ERROR_INVAL;
233 cellp = cm_FindCellByID(fidp->cell, 0);
235 return CM_ERROR_NOSUCHCELL;
237 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
241 *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp, replicated);
243 lock_ObtainRead(&cm_volumeLock);
245 lock_ReleaseRead(&cm_volumeLock);
246 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
250 cm_SetServerBusyStatus(cm_serverRef_t **serverspp, cm_server_t *serverp)
252 cm_serverRef_t *tsrp;
254 lock_ObtainRead(&cm_serverLock);
255 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
256 if (tsrp->status == srv_deleted)
258 if (cm_ServerEqual(tsrp->server, serverp) && tsrp->status == srv_not_busy) {
259 tsrp->status = srv_busy;
263 lock_ReleaseRead(&cm_serverLock);
267 cm_ResetServerBusyStatus(cm_serverRef_t **serverspp)
269 cm_serverRef_t *tsrp;
271 lock_ObtainRead(&cm_serverLock);
272 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
273 if (tsrp->status == srv_deleted)
275 if (tsrp->status == srv_busy) {
276 tsrp->status = srv_not_busy;
279 lock_ReleaseRead(&cm_serverLock);
283 isNetworkError(int code)
285 if (code >= -64 && code < 0)
291 * Analyze the error return from an RPC. Determine whether or not to retry,
292 * and if we're going to retry, determine whether failover is appropriate,
293 * and whether timed backoff is appropriate.
295 * If the error code is from cm_ConnFromFID() or friends, it will be a CM_ERROR code.
296 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
297 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
298 * be a security code (e.g. RXKADEXPIRED).
300 * If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
302 * For VLDB calls, fidp will be NULL and cellp will not be.
304 * volSyncp and/or cbrp may also be NULL.
307 cm_Analyze(cm_conn_t *connp,
313 AFSFetchStatus *statusp,
314 AFSVolSync *volSyncp,
315 cm_serverRef_t ** vlServerspp,
316 cm_callbackRequest_t *cbrp,
319 cm_server_t *serverp = NULL;
320 cm_serverRef_t **volServerspp = NULL;
321 cm_serverRef_t *tsrp;
323 cm_volume_t * volp = NULL;
324 cm_vol_state_t *statep = NULL;
325 cm_scache_t * scp = NULL;
326 afs_uint32 replicated = 0;
328 int free_svr_list = 0;
329 int dead_session = (userp->cellInfop == NULL);
330 long timeUsed, timeLeft;
332 char addr[16]="unknown";
334 int location_updated = 0;
337 int invalid_status = 0;
339 osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
342 /* no locking required, since connp->serverp never changes after
345 serverp = connp->serverp;
347 /* Update callback pointer */
348 if (cbrp && serverp && errorCode == 0) {
350 if ( cbrp->serverp != serverp ) {
351 lock_ObtainWrite(&cm_serverLock);
352 cm_PutServerNoLock(cbrp->serverp);
353 cm_GetServerNoLock(serverp);
354 lock_ReleaseWrite(&cm_serverLock);
357 cm_GetServer(serverp);
359 cbrp->serverp = serverp;
362 /* if timeout - check that it did not exceed the HardDead timeout
365 /* timeleft - get it from reqp the same way as cm_ConnByMServers does */
366 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
367 if ( (reqp->flags & CM_REQ_SOURCE_SMB) ||
368 (serverp && serverp->type == CM_SERVER_VLDB))
369 timeLeft = HardDeadtimeout - timeUsed;
371 timeLeft = 0x0FFFFFFF;
374 * Similar to the UNIX cache manager, if the AFSFetchStatus info
375 * returned by the file server is invalid, consider the response
376 * as being equivalent to VBUSY so that another file server can
377 * be queried if there is one. If there is no replica, then the
380 if (errorCode == 0 && statusp && !cm_IsStatusValid(statusp)) {
385 /* get a pointer to the cell */
387 if (cellp == NULL && serverp)
388 cellp = serverp->cellp;
389 if (cellp == NULL && vlServerspp) {
390 struct cm_serverRef * refp;
391 lock_ObtainRead(&cm_serverLock);
392 for ( refp=*vlServerspp ; cellp == NULL && refp != NULL; refp=refp->next) {
393 if (refp->status == srv_deleted)
396 cellp = refp->server->cellp;
398 lock_ReleaseRead(&cm_serverLock);
400 if (cellp == NULL && fidp) {
401 cellp = cm_FindCellByID(fidp->cell, 0);
405 if (errorCode == 0) {
407 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
409 else if (errorCode == CM_ERROR_TIMEDOUT) {
410 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_TIMEDOUT");
411 if ( timeLeft > 5 ) {
413 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
418 else if (errorCode == CM_ERROR_RETRY) {
419 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_RETRY");
423 else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
424 errorCode == UAEAGAIN || errorCode == EAGAIN) {
425 osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
426 if ( timeLeft > 5 ) {
432 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
435 /* if there is nosuchvolume, then we have a situation in which a
436 * previously known volume no longer has a set of servers
437 * associated with it. Either the volume has moved or
438 * the volume has been deleted. Try to find a new server list
439 * until the timeout period expires.
441 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
442 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
444 * The VNOVOL or VL_NOENT error has already been translated
445 * to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
449 else if (errorCode == CM_ERROR_EMPTY) {
451 * The server list is empty (or all entries have been deleted).
452 * If fidp is NULL, this was a vlServer list and we can attempt
453 * to force a cell lookup. If fidp is not NULL, we can attempt
454 * to refresh the volume location list.
457 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
458 CM_GETVOL_FLAG_NO_LRU_UPDATE,
461 lock_ObtainWrite(&volp->rw);
462 if (cm_UpdateVolumeLocation(cellp, userp, reqp, volp) == 0) {
463 lock_ReleaseWrite(&volp->rw);
464 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
466 if (!cm_IsServerListEmpty(*volServerspp))
468 cm_FreeServerList(volServerspp, 0);
471 lock_ReleaseWrite(&volp->rw);
473 lock_ObtainRead(&cm_volumeLock);
475 lock_ReleaseRead(&cm_volumeLock);
479 cm_cell_t * newCellp = cm_UpdateCell( cellp, 0);
484 else if (errorCode == CM_ERROR_ALLDOWN) {
485 /* Servers marked DOWN will be restored by the background daemon
486 * thread as they become available. The volume status is
487 * updated as the server state changes.
490 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
491 cellp->name, fidp->volume);
492 msgID = MSG_ALL_SERVERS_DOWN;
493 format = "All servers are unreachable when accessing cell %s volume %d.";
494 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
496 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
499 else if (errorCode == CM_ERROR_ALLOFFLINE) {
500 /* Volume instances marked offline will be restored by the
501 * background daemon thread as they become available
504 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
505 cellp->name, fidp->volume);
506 msgID = MSG_ALL_SERVERS_OFFLINE;
507 format = "All servers are offline when accessing cell %s volume %d.";
508 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
510 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
511 CM_GETVOL_FLAG_NO_LRU_UPDATE,
514 if (volServerspp == NULL) {
515 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
520 cm_ResetServerBusyStatus(volServerspp);
522 cm_FreeServerList(volServerspp, 0);
528 * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
529 * was called by cm_CheckOfflineVolumeState().
531 if (!(reqp->flags & (CM_REQ_OFFLINE_VOL_CHK|CM_REQ_NORETRY)) &&
536 /* cm_CheckOfflineVolume() resets the serverRef state */
537 if (cm_CheckOfflineVolume(volp, fidp->volume))
540 cm_UpdateVolumeStatus(volp, fidp->volume);
542 lock_ObtainRead(&cm_volumeLock);
544 lock_ReleaseRead(&cm_volumeLock);
548 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
551 else if (errorCode == CM_ERROR_ALLBUSY) {
552 /* Volumes that are busy cannot be determined to be non-busy
553 * without actually attempting to access them.
555 if (fidp) { /* File Server query */
556 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
557 cellp->name, fidp->volume);
558 msgID = MSG_ALL_SERVERS_BUSY;
559 format = "All servers are busy when accessing cell %s volume %d.";
560 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
562 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
563 CM_GETVOL_FLAG_NO_LRU_UPDATE,
566 if (volServerspp == NULL) {
567 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
572 cm_ResetServerBusyStatus(volServerspp);
574 cm_FreeServerList(volServerspp, 0);
580 * retry all replicas for 5 minutes waiting 15 seconds
583 if (timeLeft > 20 && !(reqp->flags & CM_REQ_NORETRY) &&
584 reqp->volbusyCount++ < 20)
589 cm_UpdateVolumeStatus(volp, fidp->volume);
591 lock_ObtainRead(&cm_volumeLock);
593 lock_ReleaseRead(&cm_volumeLock);
596 } else { /* VL Server query */
597 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
599 if (timeLeft > 7 && !(reqp->flags & CM_REQ_NORETRY) && vlServerspp)
603 cm_ResetServerBusyStatus(vlServerspp);
609 /* special codes: VBUSY and VRESTARTING */
610 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
612 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
615 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
616 CM_GETVOL_FLAG_NO_LRU_UPDATE,
619 if (volServerspp == NULL) {
620 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
625 statep = cm_VolumeStateByID(volp, fidp->volume);
628 cm_UpdateVolumeStatus(volp, statep->ID);
630 lock_ObtainRead(&cm_volumeLock);
632 lock_ReleaseRead(&cm_volumeLock);
638 /* Log server being offline for this volume */
639 sprintf(addr, "%d.%d.%d.%d",
640 ((serverp->addr.sin_addr.s_addr & 0xff)),
641 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
642 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
643 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
645 switch ( errorCode ) {
647 if (invalid_status) {
648 msgID = MSG_SERVER_REPLIED_BAD_STATUS;
649 format = "Server %s replied with bad status info when accessing volume %d in cell %s. Data discarded by cache manager.";
651 msgID = MSG_SERVER_REPORTS_VBUSY;
652 format = "Server %s reported busy when accessing volume %d in cell %s.";
656 msgID = MSG_SERVER_REPORTS_VRESTARTING;
657 format = "Server %s reported restarting when accessing volume %d in cell %s.";
661 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
662 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
665 cm_SetServerBusyStatus(volServerspp, serverp);
669 cm_FreeServerList(volServerspp, 0);
676 /* special codes: missing volumes */
677 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
678 errorCode == VSALVAGE || errorCode == VIO)
681 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
683 /* In case of timeout */
684 reqp->volumeError = errorCode;
686 switch ( errorCode ) {
688 msgID = MSG_SERVER_REPORTS_VNOVOL;
689 format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
692 msgID = MSG_SERVER_REPORTS_VMOVED;
693 format = "Server %s reported volume %d in cell %s as moved.";
696 msgID = MSG_SERVER_REPORTS_VOFFLINE;
697 format = "Server %s reported volume %d in cell %s as offline.";
700 msgID = MSG_SERVER_REPORTS_VSALVAGE;
701 format = "Server %s reported volume %d in cell %s as needs salvage.";
704 msgID = MSG_SERVER_REPORTS_VIO;
705 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
709 if (fidp) { /* File Server query */
711 /* Log server being unavailable for this volume */
712 sprintf(addr, "%d.%d.%d.%d",
713 ((serverp->addr.sin_addr.s_addr & 0xff)),
714 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
715 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
716 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
718 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
719 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
722 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
723 CM_GETVOL_FLAG_NO_LRU_UPDATE,
726 statep = cm_VolumeStateByID(volp, fidp->volume);
728 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
729 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
731 LONG_PTR oldSum, newSum;
733 oldSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
735 code = cm_ForceUpdateVolume(fidp, userp, reqp);
737 location_updated = 1;
738 newSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
742 * Even if the update fails, there might still be another replica.
743 * If the volume location list changed, permit another update on
744 * a subsequent error.
746 if (code || oldSum == newSum)
747 reqp->flags |= CM_REQ_VOLUME_UPDATED;
749 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
750 fidp->cell, fidp->volume, code);
754 cm_UpdateVolumeStatus(volp, statep->ID);
755 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
756 fidp->cell, fidp->volume, statep->state);
760 lock_ObtainRead(&cm_volumeLock);
762 lock_ReleaseRead(&cm_volumeLock);
767 * Mark server offline for this volume or delete the volume
768 * from the server list if it was moved or is not present.
770 if (volServerspp == NULL || location_updated) {
771 code = cm_GetServerList(fidp, userp, reqp, &replicated, &volServerspp);
779 lock_ObtainWrite(&cm_serverLock);
780 for (tsrp = *volServerspp; tsrp; tsrp=tsrp->next) {
781 if (tsrp->status == srv_deleted)
784 sprintf(addr, "%d.%d.%d.%d",
785 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
786 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
787 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
788 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
790 if (cm_ServerEqual(tsrp->server, serverp)) {
794 osi_Log2(afsd_logp, "volume %u moved from server %s",
795 fidp->volume, osi_LogSaveString(afsd_logp,addr));
796 tsrp->status = srv_deleted;
798 cm_RemoveVolumeFromServer(serverp, fidp->volume);
802 * The 1.6.0 and 1.6.1 file servers send transient VNOVOL errors which
803 * are no indicative of the volume not being present. For example,
804 * VNOVOL can be sent during a transition to a VBUSY state prior to
805 * salvaging or when cloning a .backup volume instance. As a result
806 * the cache manager must attempt at least one retry when a VNOVOL is
807 * receive but there are no changes to the volume location information.
809 if (reqp->vnovolError > 0 && cm_ServerEqual(reqp->errorServp, serverp)) {
810 osi_Log2(afsd_logp, "volume %u not present on server %s",
811 fidp->volume, osi_LogSaveString(afsd_logp,addr));
812 tsrp->status = srv_deleted;
814 cm_RemoveVolumeFromServer(serverp, fidp->volume);
816 osi_Log2(afsd_logp, "VNOVOL received for volume %u from server %s",
817 fidp->volume, osi_LogSaveString(afsd_logp,addr));
819 if (tsrp->status == srv_not_busy)
820 tsrp->status = srv_busy;
827 osi_Log2(afsd_logp, "VOFFLINE received for volume %u from server %s",
828 fidp->volume, osi_LogSaveString(afsd_logp,addr));
829 tsrp->status = srv_offline;
832 osi_Log3(afsd_logp, "volume %u exists on server %s with status %u",
833 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
837 lock_ReleaseWrite(&cm_serverLock);
840 /* Remember that the VNOVOL error occurred */
841 if (errorCode == VNOVOL) {
842 reqp->errorServp = serverp;
846 /* Remember that the VIO error occurred */
847 if (errorCode == VIO) {
848 reqp->errorServp = serverp;
852 /* Free the server list before cm_ForceUpdateVolume is called */
854 cm_FreeServerList(volServerspp, 0);
859 if ( timeLeft > 2 && reqp->vioCount < 100)
861 } else if ( errorCode == VNOVNODE ) {
863 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
866 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
867 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
869 scp = cm_FindSCache(fidp);
871 cm_scache_t *pscp = NULL;
873 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
874 pscp = cm_FindSCacheParent(scp);
876 lock_ObtainWrite(&scp->rw);
877 _InterlockedOr(&scp->flags, CM_SCACHEFLAG_DELETED);
878 lock_ObtainWrite(&cm_scacheLock);
879 cm_AdjustScacheLRU(scp);
880 lock_ReleaseWrite(&cm_scacheLock);
881 cm_LockMarkSCacheLost(scp);
882 lock_ReleaseWrite(&scp->rw);
884 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
885 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
886 cm_ReleaseSCache(scp);
889 if (cm_HaveCallback(pscp)) {
890 lock_ObtainWrite(&pscp->rw);
891 cm_DiscardSCache(pscp);
892 lock_ReleaseWrite(&pscp->rw);
895 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
896 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
899 cm_ReleaseSCache(pscp);
903 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
908 else if (errorCode == RX_CALL_TIMEOUT) {
909 /* RPC took longer than hardDeadTime or the server
910 * reported idle for longer than idleDeadTime
911 * don't mark server as down but don't retry
912 * this is to prevent the SMB session from timing out
913 * In addition, we log an event to the event log
917 sprintf(addr, "%d.%d.%d.%d",
918 ((serverp->addr.sin_addr.s_addr & 0xff)),
919 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
920 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
921 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
923 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
924 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
925 osi_LogSaveString(afsd_logp,addr));
926 reqp->errorServp = serverp;
931 scp = cm_FindSCache(fidp);
933 if (cm_HaveCallback(scp)) {
934 lock_ObtainWrite(&scp->rw);
935 cm_DiscardSCache(scp);
936 lock_ReleaseWrite(&scp->rw);
939 * We really should notify the redirector that we discarded
940 * the status information but doing so in this case is not
941 * safe as it can result in a deadlock with extent release
945 cm_ReleaseSCache(scp);
949 if (!fidp) { /* vldb */
952 cm_volume_t *volp = cm_FindVolumeByFID(fidp, userp, reqp);
954 if (fidp->volume == cm_GetROVolumeID(volp))
961 else if (errorCode == RX_MSGSIZE) {
963 * RPC failed because a transmitted rx packet
964 * may have grown larger than the path mtu.
965 * Force a retry and the Rx library will try
966 * with a smaller mtu size.
970 sprintf(addr, "%d.%d.%d.%d",
971 ((serverp->addr.sin_addr.s_addr & 0xff)),
972 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
973 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
974 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
976 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
977 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
978 osi_LogSaveString(afsd_logp,addr));
982 else if (errorCode == VNOSERVICE) {
984 * The server did not service the RPC.
985 * If this was a file server RPC it means that for at
986 * least the file server's idle dead timeout period the
987 * file server did not receive any new data packets from
990 * The RPC was not serviced so it can be retried and any
991 * existing status information is still valid.
994 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
998 sprintf(addr, "%d.%d.%d.%d",
999 ((serverp->addr.sin_addr.s_addr & 0xff)),
1000 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1001 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1002 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1004 LogEvent(EVENTLOG_WARNING_TYPE, MSG_SERVER_REPORTS_VNOSERVICE,
1005 addr, fidp->volume, cellp->name);
1006 osi_Log3(afsd_logp, "Server %s reported rpc to volume %d in cell %s as not serviced.",
1007 osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
1012 else if (errorCode == RX_CALL_DEAD) {
1013 /* mark server as down */
1015 sprintf(addr, "%d.%d.%d.%d",
1016 ((serverp->addr.sin_addr.s_addr & 0xff)),
1017 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1018 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1019 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1021 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
1022 osi_LogSaveString(afsd_logp,addr),
1023 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1026 if ((connp->flags & CM_CONN_FLAG_NEW) ||
1027 (reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
1028 lock_ObtainMutex(&serverp->mx);
1029 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
1030 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
1031 serverp->downTime = time(NULL);
1033 lock_ReleaseMutex(&serverp->mx);
1035 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1037 cm_ForceNewConnections(serverp);
1041 if (fidp && storeOp)
1042 scp = cm_FindSCache(fidp);
1044 if (cm_HaveCallback(scp)) {
1045 lock_ObtainWrite(&scp->rw);
1046 cm_DiscardSCache(scp);
1047 lock_ReleaseWrite(&scp->rw);
1050 * We really should notify the redirector that we discarded
1051 * the status information but doing so in this case is not
1052 * safe as it can result in a deadlock with extent release
1056 cm_ReleaseSCache(scp);
1062 else if (isNetworkError(errorCode)) {
1064 * any other rx error - mark server as down.
1066 * The most likely error to receive is RX_INVALID_OPERATION which
1067 * can be received as a result of an unrecognized rx service id or
1068 * an unrecognized security class. Failing to mark the server down
1069 * when a service id or security class that is expected to be
1070 * recognized is not will result in an infinite retry loop. Instead
1071 * mark the server down. If we are ever able to successfully
1072 * communicate with it again it will be marked up via a probe.
1074 * The most common scenario for this behavior would be an AuriStor
1075 * file server being replaced by an AFS3 file server or a reflected
1079 sprintf(addr, "%d.%d.%d.%d",
1080 ((serverp->addr.sin_addr.s_addr & 0xff)),
1081 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1082 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1083 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1085 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
1087 osi_LogSaveString(afsd_logp,addr),
1088 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1091 if ((connp->flags & CM_CONN_FLAG_NEW) ||
1092 (reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
1093 lock_ObtainMutex(&serverp->mx);
1094 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
1095 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
1096 serverp->downTime = time(NULL);
1098 lock_ReleaseMutex(&serverp->mx);
1100 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1102 cm_ForceNewConnections(serverp);
1109 else if (errorCode == RXKADEXPIRED) {
1110 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1112 if (!dead_session) {
1113 lock_ObtainMutex(&userp->mx);
1114 ucellp = cm_GetUCell(userp, cellp);
1115 if (ucellp->ticketp) {
1116 free(ucellp->ticketp);
1117 ucellp->ticketp = NULL;
1119 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1121 lock_ReleaseMutex(&userp->mx);
1123 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1125 cm_ForceNewConnections(serverp);
1130 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1131 char * s = "unknown error";
1132 switch ( errorCode ) {
1133 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1134 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1135 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1136 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1137 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1138 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1139 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1140 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1141 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1142 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1143 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1144 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1145 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1147 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1151 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
1154 reqp->errorServp = serverp;
1155 reqp->tokenError = errorCode;
1158 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1160 * We received a ubik error. its possible that the server we are
1161 * communicating with has a corrupted database or is partitioned
1162 * from the rest of the servers and another server might be able
1163 * to answer our query. Therefore, we will retry the request
1164 * and force the use of another server.
1167 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
1170 reqp->errorServp = serverp;
1171 reqp->tokenError = errorCode;
1174 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1175 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1177 cm_ForceNewConnections(serverp);
1181 _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
1184 char * s = "unknown error";
1185 switch ( errorCode ) {
1186 case VSALVAGE : s = "VSALVAGE"; break;
1187 case VNOVNODE : s = "VNOVNODE"; break;
1188 case VNOVOL : s = "VNOVOL"; break;
1189 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1190 case VNOSERVICE : s = "VNOSERVICE"; break;
1191 case VOFFLINE : s = "VOFFLINE"; break;
1192 case VONLINE : s = "VONLINE"; break;
1193 case VDISKFULL : s = "VDISKFULL"; break;
1194 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1195 case VBUSY : s = "VBUSY"; break;
1196 case VMOVED : s = "VMOVED"; break;
1197 case VIO : s = "VIO"; break;
1198 case VRESTRICTED : s = "VRESTRICTED"; break;
1199 case VRESTARTING : s = "VRESTARTING"; break;
1200 case VREADONLY : s = "VREADONLY"; break;
1201 case EAGAIN : s = "EAGAIN"; break;
1202 case UAEAGAIN : s = "UAEAGAIN"; break;
1203 case EINVAL : s = "EINVAL"; break;
1204 case UAEINVAL : s = "UAEINVAL"; break;
1205 case EACCES : s = "EACCES"; break;
1206 case UAEACCES : s = "UAEACCES"; break;
1207 case ENOENT : s = "ENOENT"; break;
1208 case UAENOENT : s = "UAENOENT"; break;
1209 case EEXIST : s = "EEXIST"; break;
1210 case UAEEXIST : s = "UAEEXIST"; break;
1211 case VICECONNBAD : s = "VICECONNBAD"; break;
1212 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1213 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1214 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1215 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1216 case VL_IO : s = "VL_IO"; break;
1217 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1218 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1219 case VL_NOENT : s = "VL_NOENT"; break;
1220 case VL_EMPTY : s = "VL_EMPTY"; break;
1221 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1222 case VL_BADNAME : s = "VL_BADNAME"; break;
1223 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1224 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1225 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1226 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1227 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1228 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1229 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1230 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1231 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1232 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1233 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1234 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1235 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1236 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1237 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1238 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1239 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1240 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1241 case VL_PERM : s = "VL_PERM"; break;
1242 case VL_NOMEM : s = "VL_NOMEM"; break;
1243 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1244 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1245 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1246 case VL_BADMASK : s = "VL_BADMASK"; break;
1247 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1248 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1249 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1250 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1251 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1252 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1253 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1254 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1255 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1256 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1257 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1258 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1259 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1260 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1261 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1262 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1263 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1264 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1265 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1266 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1267 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1268 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1269 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1270 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1271 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1272 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1273 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1274 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1275 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1276 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1277 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1278 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1279 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1280 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1281 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1282 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1283 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1284 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1285 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1286 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1287 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1288 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1289 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1290 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1291 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1292 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1293 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1294 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1295 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1296 case UAEIO : s = "UAEIO"; break;
1297 case EIO : s = "EIO"; break;
1299 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1305 /* If not allowed to retry, don't */
1308 else if (!forcing_new && (retry < 2) && (reqp->flags & CM_REQ_NORETRY))
1311 /* drop this on the way out */
1317 * clear the volume updated flag if we succeed.
1318 * this way the flag will not prevent a subsequent volume
1319 * from being updated if necessary.
1323 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1327 errorCode != VBUSY &&
1328 errorCode != VRESTARTING &&
1329 errorCode != CM_ERROR_ALLBUSY)
1331 cm_ResetServerBusyStatus(vlServerspp);
1334 /* retry until we fail to find a connection */
1338 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1339 cm_req_t *reqp, cm_conn_t **connpp)
1342 cm_serverRef_t *tsrp;
1344 long firstError = 0;
1345 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1, allDeleted = 1;
1346 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1347 long timeUsed, timeLeft, hardTimeLeft;
1351 if (serversp == NULL) {
1352 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_EMPTY);
1353 return CM_ERROR_EMPTY;
1356 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1357 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1359 /* leave 5 seconds margin of safety */
1360 timeLeft = ConnDeadtimeout - timeUsed - 5;
1361 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1364 lock_ObtainRead(&cm_serverLock);
1365 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1366 if (tsrp->status == srv_deleted)
1372 if (reqp->errorServp) {
1374 * search the list until we find the server
1375 * that failed last time. When we find it
1376 * clear the error, skip it and try the next one
1379 if (tsp == reqp->errorServp)
1380 reqp->errorServp = NULL;
1384 cm_GetServerNoLock(tsp);
1385 lock_ReleaseRead(&cm_serverLock);
1386 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1388 if (tsrp->status == srv_busy) {
1391 } else if (tsrp->status == srv_offline) {
1397 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1398 if (code == 0) { /* cm_CBS only returns 0 */
1400 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1401 /* Set RPC timeout */
1402 if (timeLeft > ConnDeadtimeout)
1403 timeLeft = ConnDeadtimeout;
1405 if (hardTimeLeft > HardDeadtimeout)
1406 hardTimeLeft = HardDeadtimeout;
1408 lock_ObtainMutex(&(*connpp)->mx);
1409 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1410 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1411 lock_ReleaseMutex(&(*connpp)->mx);
1416 /* therefore, this code is never executed */
1417 if (firstError == 0)
1421 lock_ObtainRead(&cm_serverLock);
1422 cm_PutServerNoLock(tsp);
1425 lock_ReleaseRead(&cm_serverLock);
1427 if (firstError == 0) {
1429 firstError = CM_ERROR_EMPTY;
1430 } else if (allDown) {
1431 firstError = (reqp->tokenError ? reqp->tokenError :
1432 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1434 * if we experienced either a token error or and idle dead time error
1435 * and now all of the servers are down, we have either tried them
1436 * all or lost connectivity. Clear the error we are returning so
1437 * we will not return it indefinitely if the request is retried.
1439 reqp->idleError = reqp->tokenError = 0;
1440 } else if (allBusy) {
1441 firstError = CM_ERROR_ALLBUSY;
1442 } else if (allOffline || (someBusy && someOffline)) {
1443 firstError = CM_ERROR_ALLOFFLINE;
1445 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1446 firstError = CM_ERROR_TIMEDOUT;
1450 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1454 /* called with a held server to GC all bad connections hanging off of the server */
1455 void cm_GCConnections(cm_server_t *serverp)
1461 lock_ObtainWrite(&cm_connLock);
1462 lcpp = &serverp->connsp;
1463 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1465 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1466 /* do the deletion of this guy */
1467 cm_PutServer(tcp->serverp);
1468 cm_ReleaseUser(userp);
1470 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1471 rx_DestroyConnection(tcp->rxconnp);
1472 lock_FinalizeMutex(&tcp->mx);
1476 /* just advance to the next */
1480 lock_ReleaseWrite(&cm_connLock);
1483 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1484 cm_server_t *serverp, afs_uint32 replicated)
1486 unsigned short port;
1489 struct rx_securityClass *secObjp;
1491 port = serverp->addr.sin_port;
1492 switch (serverp->type) {
1493 case CM_SERVER_VLDB:
1498 case CM_SERVER_FILE:
1504 osi_panic("unknown server type", __FILE__, __LINE__);
1507 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1511 tcp->cryptlevel = rxkad_clear;
1514 tcp->cryptlevel = rxkad_auth;
1517 tcp->cryptlevel = rxkad_crypt;
1519 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1520 &ucellp->sessionKey, ucellp->kvno,
1521 ucellp->ticketLen, ucellp->ticketp);
1525 tcp->cryptlevel = rxkad_clear;
1526 secObjp = rxnull_NewClientSecurityObject();
1528 osi_assertx(secObjp != NULL, "null rx_securityClass");
1529 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1534 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1535 if (smb_Enabled || tcp->serverp->type == CM_SERVER_VLDB)
1536 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1539 * Setting idle dead timeout to a non-zero value activates RX_CALL_TIMEOUT
1540 * errors if the call is idle for a certain amount of time.
1543 _InterlockedOr(&tcp->flags, CM_CONN_FLAG_REPLICATION);
1544 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1546 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1550 * Let the Rx library know that we can auto-retry if an
1551 * RX_MSGSIZE error is returned.
1553 if (rx_pmtu_discovery)
1554 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1557 * Attempt to limit NAT pings to the anonymous file server connections.
1558 * Only file servers implement client callbacks and we only need one ping
1559 * to be sent to each server.
1561 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1562 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1563 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1566 tcp->ucgen = ucellp->gen;
1568 rxs_Release(secObjp); /* Decrement the initial refCount */
1570 _InterlockedAnd(&tcp->flags, ~CM_CONN_FLAG_FORCE_NEW);
1571 _InterlockedOr(&tcp->flags, CM_CONN_FLAG_NEW);
1574 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1581 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1582 userp = cm_rootUserp;
1584 lock_ObtainMutex(&userp->mx);
1585 /* find ucell structure */
1586 ucellp = cm_GetUCell(userp, serverp->cellp);
1588 lock_ObtainRead(&cm_connLock);
1589 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1590 if (tcp->userp == userp &&
1591 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1592 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1597 lock_ConvertRToW(&cm_connLock);
1598 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1599 if (tcp->userp == userp &&
1600 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1601 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1605 InterlockedIncrement(&tcp->refCount);
1606 lock_ReleaseWrite(&cm_connLock);
1609 cm_GetServer(serverp);
1610 tcp = malloc(sizeof(*tcp));
1611 memset(tcp, 0, sizeof(*tcp));
1614 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1615 tcp->serverp = serverp;
1616 tcp->cryptlevel = rxkad_clear;
1617 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1619 tcp->nextp = serverp->connsp;
1620 serverp->connsp = tcp;
1621 lock_ReleaseWrite(&cm_connLock);
1622 lock_ReleaseMutex(&userp->mx);
1624 InterlockedIncrement(&tcp->refCount);
1625 lock_ReleaseRead(&cm_connLock);
1627 lock_ReleaseMutex(&userp->mx);
1629 lock_ObtainMutex(&tcp->mx);
1630 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1631 (tcp->ucgen < ucellp->gen) ||
1632 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1634 if (tcp->ucgen < ucellp->gen)
1635 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1637 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1638 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1639 rx_DestroyConnection(tcp->rxconnp);
1640 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1642 lock_ReleaseMutex(&tcp->mx);
1645 /* return this pointer to our caller */
1646 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1652 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1656 cm_serverRef_t **serverspp;
1657 cm_serverRef_t *tsrp;
1659 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1660 afs_uint32 replicated;
1664 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1668 lock_ObtainRead(&cm_serverLock);
1669 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1670 if (tsrp->status == srv_deleted)
1673 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1675 if (tsrp->status == srv_busy) {
1678 } else if (tsrp->status == srv_offline) {
1687 lock_ReleaseRead(&cm_serverLock);
1688 cm_FreeServerList(serverspp, 0);
1694 else if (allOffline || (someBusy && someOffline))
1701 * The returned cm_conn_t ** object is released in the subsequent call
1704 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1708 cm_serverRef_t **serverspp;
1709 afs_uint32 replicated;
1713 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1717 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1718 cm_FreeServerList(serverspp, 0);
1723 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1727 cm_serverRef_t **serverspp;
1728 afs_uint32 replicated;
1732 code = cm_GetVolServerList(volp, volid, userp, reqp, &replicated, &serverspp);
1736 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1737 cm_FreeServerList(serverspp, 0);
1742 extern struct rx_connection *
1743 cm_GetRxConn(cm_conn_t *connp)
1745 struct rx_connection * rxconnp;
1746 lock_ObtainMutex(&connp->mx);
1747 rxconnp = connp->rxconnp;
1748 rx_GetConnection(rxconnp);
1749 lock_ReleaseMutex(&connp->mx);
1753 void cm_ForceNewConnections(cm_server_t *serverp)
1757 if (serverp == NULL)
1760 lock_ObtainWrite(&cm_connLock);
1761 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1762 lock_ObtainMutex(&tcp->mx);
1763 _InterlockedOr(&tcp->flags, CM_CONN_FLAG_FORCE_NEW);
1764 lock_ReleaseMutex(&tcp->mx);
1766 lock_ReleaseWrite(&cm_connLock);