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, cm_serverRef_t ***serversppp)
216 *serversppp = cm_GetVolServers(volp, volid, userp, reqp, replicated);
217 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
220 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
221 struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
224 cm_volume_t *volp = NULL;
225 cm_cell_t *cellp = NULL;
229 return CM_ERROR_INVAL;
232 cellp = cm_FindCellByID(fidp->cell, 0);
234 return CM_ERROR_NOSUCHCELL;
236 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
240 *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp, replicated);
242 lock_ObtainRead(&cm_volumeLock);
244 lock_ReleaseRead(&cm_volumeLock);
245 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
249 cm_SetServerBusyStatus(cm_serverRef_t *serversp, cm_server_t *serverp)
251 cm_serverRef_t *tsrp;
253 lock_ObtainWrite(&cm_serverLock);
254 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
255 if (tsrp->status == srv_deleted)
257 if (cm_ServerEqual(tsrp->server, serverp) && tsrp->status == srv_not_busy) {
258 tsrp->status = srv_busy;
262 lock_ReleaseWrite(&cm_serverLock);
266 cm_ResetServerBusyStatus(cm_serverRef_t *serversp)
268 cm_serverRef_t *tsrp;
270 lock_ObtainWrite(&cm_serverLock);
271 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
272 if (tsrp->status == srv_deleted)
274 if (tsrp->status == srv_busy) {
275 tsrp->status = srv_not_busy;
278 lock_ReleaseWrite(&cm_serverLock);
282 * Analyze the error return from an RPC. Determine whether or not to retry,
283 * and if we're going to retry, determine whether failover is appropriate,
284 * and whether timed backoff is appropriate.
286 * If the error code is from cm_ConnFromFID() or friends, it will be a CM_ERROR code.
287 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
288 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
289 * be a security code (e.g. RXKADEXPIRED).
291 * If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
293 * For VLDB calls, fidp will be NULL and cellp will not be.
295 * volSyncp and/or cbrp may also be NULL.
298 cm_Analyze(cm_conn_t *connp,
304 AFSFetchStatus *statusp,
305 AFSVolSync *volSyncp,
306 cm_serverRef_t * serversp,
307 cm_callbackRequest_t *cbrp,
310 cm_server_t *serverp = NULL;
311 cm_serverRef_t **serverspp = NULL;
312 cm_serverRef_t *tsrp;
314 cm_volume_t * volp = NULL;
315 cm_vol_state_t *statep = NULL;
316 cm_scache_t * scp = NULL;
317 afs_uint32 replicated = 0;
319 int free_svr_list = 0;
320 int dead_session = (userp->cellInfop == NULL);
321 long timeUsed, timeLeft;
323 char addr[16]="unknown";
325 int location_updated = 0;
328 int invalid_status = 0;
330 osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
333 /* no locking required, since connp->serverp never changes after
336 serverp = connp->serverp;
338 /* Update callback pointer */
339 if (cbrp && serverp && errorCode == 0) {
341 if ( cbrp->serverp != serverp ) {
342 lock_ObtainWrite(&cm_serverLock);
343 cm_PutServerNoLock(cbrp->serverp);
344 cm_GetServerNoLock(serverp);
345 lock_ReleaseWrite(&cm_serverLock);
348 cm_GetServer(serverp);
350 cbrp->serverp = serverp;
353 /* if timeout - check that it did not exceed the HardDead timeout
356 /* timeleft - get it from reqp the same way as cm_ConnByMServers does */
357 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
358 if ( (reqp->flags & CM_REQ_SOURCE_SMB) ||
359 (serverp && serverp->type == CM_SERVER_VLDB))
360 timeLeft = HardDeadtimeout - timeUsed;
362 timeLeft = 0x0FFFFFFF;
365 * Similar to the UNIX cache manager, if the AFSFetchStatus info
366 * returned by the file server is invalid, consider the response
367 * as being equivalent to VBUSY so that another file server can
368 * be queried if there is one. If there is no replica, then the
371 if (errorCode == 0 && statusp && !cm_IsStatusValid(statusp)) {
376 /* get a pointer to the cell */
378 if (cellp == NULL && serverp)
379 cellp = serverp->cellp;
380 if (cellp == NULL && serversp) {
381 struct cm_serverRef * refp;
382 for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
383 if (refp->status == srv_deleted)
386 cellp = refp->server->cellp;
389 if (cellp == NULL && fidp) {
390 cellp = cm_FindCellByID(fidp->cell, 0);
394 if (errorCode == CM_ERROR_TIMEDOUT) {
395 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_TIMEDOUT");
396 if ( timeLeft > 5 ) {
398 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
403 else if (errorCode == CM_ERROR_RETRY) {
404 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_RETRY");
408 else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
409 errorCode == UAEAGAIN || errorCode == EAGAIN) {
410 osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
411 if ( timeLeft > 5 ) {
417 /* if there is nosuchvolume, then we have a situation in which a
418 * previously known volume no longer has a set of servers
419 * associated with it. Either the volume has moved or
420 * the volume has been deleted. Try to find a new server list
421 * until the timeout period expires.
423 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
424 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
426 * The VNOVOL or VL_NOENT error has already been translated
427 * to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
431 else if (errorCode == CM_ERROR_EMPTY) {
433 * The server list is empty (or all entries have been deleted).
434 * If fidp is NULL, this was a vlServer list and we can attempt
435 * to force a cell lookup. If fidp is not NULL, we can attempt
436 * to refresh the volume location list.
439 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
440 CM_GETVOL_FLAG_NO_LRU_UPDATE,
443 lock_ObtainWrite(&volp->rw);
444 if (cm_UpdateVolumeLocation(cellp, userp, reqp, volp) == 0) {
445 lock_ReleaseWrite(&volp->rw);
446 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
448 if (!cm_IsServerListEmpty(*serverspp))
450 cm_FreeServerList(serverspp, 0);
453 lock_ReleaseWrite(&volp->rw);
455 lock_ObtainRead(&cm_volumeLock);
457 lock_ReleaseRead(&cm_volumeLock);
461 cm_cell_t * newCellp = cm_UpdateCell( cellp, 0);
466 else if (errorCode == CM_ERROR_ALLDOWN) {
467 /* Servers marked DOWN will be restored by the background daemon
468 * thread as they become available. The volume status is
469 * updated as the server state changes.
472 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
473 cellp->name, fidp->volume);
474 msgID = MSG_ALL_SERVERS_DOWN;
475 format = "All servers are unreachable when accessing cell %s volume %d.";
476 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
478 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
481 else if (errorCode == CM_ERROR_ALLOFFLINE) {
482 /* Volume instances marked offline will be restored by the
483 * background daemon thread as they become available
486 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
487 cellp->name, fidp->volume);
488 msgID = MSG_ALL_SERVERS_OFFLINE;
489 format = "All servers are offline when accessing cell %s volume %d.";
490 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
492 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
493 CM_GETVOL_FLAG_NO_LRU_UPDATE,
497 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
499 serversp = *serverspp;
503 cm_ResetServerBusyStatus(serversp);
505 cm_FreeServerList(serverspp, 0);
511 * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
512 * was called by cm_CheckOfflineVolumeState().
514 if (!(reqp->flags & CM_REQ_OFFLINE_VOL_CHK) && timeLeft > 7) {
517 /* cm_CheckOfflineVolume() resets the serverRef state */
518 if (cm_CheckOfflineVolume(volp, fidp->volume))
521 cm_UpdateVolumeStatus(volp, fidp->volume);
523 lock_ObtainRead(&cm_volumeLock);
525 lock_ReleaseRead(&cm_volumeLock);
529 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
532 else if (errorCode == CM_ERROR_ALLBUSY) {
533 /* Volumes that are busy cannot be determined to be non-busy
534 * without actually attempting to access them.
536 if (fidp) { /* File Server query */
537 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
538 cellp->name, fidp->volume);
539 msgID = MSG_ALL_SERVERS_BUSY;
540 format = "All servers are busy when accessing cell %s volume %d.";
541 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
543 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
544 CM_GETVOL_FLAG_NO_LRU_UPDATE,
548 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
550 serversp = *serverspp;
554 cm_ResetServerBusyStatus(serversp);
556 cm_FreeServerList(serverspp, 0);
563 statep = cm_VolumeStateByID(volp, fidp->volume);
566 cm_UpdateVolumeStatus(volp, fidp->volume);
568 lock_ObtainRead(&cm_volumeLock);
570 lock_ReleaseRead(&cm_volumeLock);
573 } else { /* VL Server query */
574 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
580 cm_ResetServerBusyStatus(serversp);
587 /* special codes: VBUSY and VRESTARTING */
588 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
590 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
591 CM_GETVOL_FLAG_NO_LRU_UPDATE,
595 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
597 serversp = *serverspp;
602 statep = cm_VolumeStateByID(volp, fidp->volume);
605 cm_UpdateVolumeStatus(volp, statep->ID);
607 lock_ObtainRead(&cm_volumeLock);
609 lock_ReleaseRead(&cm_volumeLock);
615 /* Log server being offline for this volume */
616 sprintf(addr, "%d.%d.%d.%d",
617 ((serverp->addr.sin_addr.s_addr & 0xff)),
618 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
619 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
620 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
622 switch ( errorCode ) {
624 if (invalid_status) {
625 msgID = MSG_SERVER_REPLIED_BAD_STATUS;
626 format = "Server %s replied with bad status info when accessing volume %d in cell %s. Data discarded by cache manager.";
628 msgID = MSG_SERVER_REPORTS_VBUSY;
629 format = "Server %s reported busy when accessing volume %d in cell %s.";
633 msgID = MSG_SERVER_REPORTS_VRESTARTING;
634 format = "Server %s reported restarting when accessing volume %d in cell %s.";
638 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
639 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
641 cm_SetServerBusyStatus(serversp, serverp);
645 cm_FreeServerList(serverspp, 0);
652 /* special codes: missing volumes */
653 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
654 errorCode == VSALVAGE || errorCode == VIO)
656 /* In case of timeout */
657 reqp->volumeError = errorCode;
659 switch ( errorCode ) {
661 msgID = MSG_SERVER_REPORTS_VNOVOL;
662 format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
665 msgID = MSG_SERVER_REPORTS_VMOVED;
666 format = "Server %s reported volume %d in cell %s as moved.";
669 msgID = MSG_SERVER_REPORTS_VOFFLINE;
670 format = "Server %s reported volume %d in cell %s as offline.";
673 msgID = MSG_SERVER_REPORTS_VSALVAGE;
674 format = "Server %s reported volume %d in cell %s as needs salvage.";
677 msgID = MSG_SERVER_REPORTS_VIO;
678 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
682 if (fidp) { /* File Server query */
684 /* Log server being unavailable for this volume */
685 sprintf(addr, "%d.%d.%d.%d",
686 ((serverp->addr.sin_addr.s_addr & 0xff)),
687 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
688 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
689 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
691 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
692 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
695 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
696 CM_GETVOL_FLAG_NO_LRU_UPDATE,
699 statep = cm_VolumeStateByID(volp, fidp->volume);
701 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
702 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
704 LONG_PTR oldSum, newSum;
706 oldSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
708 code = cm_ForceUpdateVolume(fidp, userp, reqp);
710 location_updated = 1;
711 newSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
715 * Even if the update fails, there might still be another replica.
716 * If the volume location list changed, permit another update on
717 * a subsequent error.
719 if (code || oldSum == newSum)
720 reqp->flags |= CM_REQ_VOLUME_UPDATED;
722 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
723 fidp->cell, fidp->volume, code);
727 cm_UpdateVolumeStatus(volp, statep->ID);
728 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
729 fidp->cell, fidp->volume, statep->state);
733 lock_ObtainRead(&cm_volumeLock);
735 lock_ReleaseRead(&cm_volumeLock);
740 * Mark server offline for this volume or delete the volume
741 * from the server list if it was moved or is not present.
743 if (!serversp || location_updated) {
744 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
746 serversp = *serverspp;
752 lock_ObtainWrite(&cm_serverLock);
753 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
754 if (tsrp->status == srv_deleted)
757 sprintf(addr, "%d.%d.%d.%d",
758 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
759 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
760 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
761 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
763 if (cm_ServerEqual(tsrp->server, serverp)) {
767 osi_Log2(afsd_logp, "volume %u moved from server %s",
768 fidp->volume, osi_LogSaveString(afsd_logp,addr));
769 tsrp->status = srv_deleted;
771 cm_RemoveVolumeFromServer(serverp, fidp->volume);
775 * The 1.6.0 and 1.6.1 file servers send transient VNOVOL errors which
776 * are no indicative of the volume not being present. For example,
777 * VNOVOL can be sent during a transition to a VBUSY state prior to
778 * salvaging or when cloning a .backup volume instance. As a result
779 * the cache manager must attempt at least one retry when a VNOVOL is
780 * receive but there are no changes to the volume location information.
782 if (reqp->vnovolError > 0 && cm_ServerEqual(reqp->errorServp, serverp)) {
783 osi_Log2(afsd_logp, "volume %u not present on server %s",
784 fidp->volume, osi_LogSaveString(afsd_logp,addr));
785 tsrp->status = srv_deleted;
787 cm_RemoveVolumeFromServer(serverp, fidp->volume);
789 osi_Log2(afsd_logp, "VNOVOL received for volume %u from server %s",
790 fidp->volume, osi_LogSaveString(afsd_logp,addr));
792 if (tsrp->status == srv_not_busy)
793 tsrp->status = srv_busy;
800 osi_Log2(afsd_logp, "VOFFLINE received for volume %u from server %s",
801 fidp->volume, osi_LogSaveString(afsd_logp,addr));
802 tsrp->status = srv_offline;
805 osi_Log3(afsd_logp, "volume %u exists on server %s with status %u",
806 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
810 lock_ReleaseWrite(&cm_serverLock);
812 /* Remember that the VNOVOL error occurred */
813 if (errorCode == VNOVOL) {
814 reqp->errorServp = serverp;
818 /* Free the server list before cm_ForceUpdateVolume is called */
820 cm_FreeServerList(serverspp, 0);
827 } else if ( errorCode == VNOVNODE ) {
829 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
830 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
832 scp = cm_FindSCache(fidp);
834 cm_scache_t *pscp = NULL;
836 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
837 pscp = cm_FindSCacheParent(scp);
839 lock_ObtainWrite(&scp->rw);
840 scp->flags |= CM_SCACHEFLAG_DELETED;
841 lock_ObtainWrite(&cm_scacheLock);
842 cm_AdjustScacheLRU(scp);
843 cm_RemoveSCacheFromHashTable(scp);
844 lock_ReleaseWrite(&cm_scacheLock);
845 cm_LockMarkSCacheLost(scp);
846 lock_ReleaseWrite(&scp->rw);
848 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
849 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
850 cm_ReleaseSCache(scp);
853 if (cm_HaveCallback(pscp)) {
854 lock_ObtainWrite(&pscp->rw);
855 cm_DiscardSCache(pscp);
856 lock_ReleaseWrite(&pscp->rw);
859 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
860 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
863 cm_ReleaseSCache(pscp);
867 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
872 else if (errorCode == RX_CALL_TIMEOUT) {
873 /* RPC took longer than hardDeadTime or the server
874 * reported idle for longer than idleDeadTime
875 * don't mark server as down but don't retry
876 * this is to prevent the SMB session from timing out
877 * In addition, we log an event to the event log
881 sprintf(addr, "%d.%d.%d.%d",
882 ((serverp->addr.sin_addr.s_addr & 0xff)),
883 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
884 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
885 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
887 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
888 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
889 osi_LogSaveString(afsd_logp,addr));
890 reqp->errorServp = serverp;
895 scp = cm_FindSCache(fidp);
897 if (cm_HaveCallback(scp)) {
898 lock_ObtainWrite(&scp->rw);
899 cm_DiscardSCache(scp);
900 lock_ReleaseWrite(&scp->rw);
903 * We really should notify the redirector that we discarded
904 * the status information but doing so in this case is not
905 * safe as it can result in a deadlock with extent release
909 cm_ReleaseSCache(scp);
913 if (!fidp) { /* vldb */
916 cm_volume_t *volp = cm_GetVolumeByFID(fidp);
918 if (fidp->volume == cm_GetROVolumeID(volp))
925 else if (errorCode == RX_MSGSIZE) {
927 * RPC failed because a transmitted rx packet
928 * may have grown larger than the path mtu.
929 * Force a retry and the Rx library will try
930 * with a smaller mtu size.
934 sprintf(addr, "%d.%d.%d.%d",
935 ((serverp->addr.sin_addr.s_addr & 0xff)),
936 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
937 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
938 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
940 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
941 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
942 osi_LogSaveString(afsd_logp,addr));
946 else if (errorCode == RX_CALL_BUSY) {
948 * RPC failed because the selected call channel
949 * is currently busy on the server. Unconditionally
950 * retry the request so an alternate call channel can be used.
953 sprintf(addr, "%d.%d.%d.%d",
954 ((serverp->addr.sin_addr.s_addr & 0xff)),
955 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
956 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
957 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
959 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
960 osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
961 osi_LogSaveString(afsd_logp,addr));
964 else if (errorCode == VNOSERVICE) {
966 * The server did not service the RPC.
967 * If this was a file server RPC it means that for at
968 * least the file server's idle dead timeout period the
969 * file server did not receive any new data packets from
972 * The RPC was not serviced so it can be retried and any
973 * existing status information is still valid.
977 sprintf(addr, "%d.%d.%d.%d",
978 ((serverp->addr.sin_addr.s_addr & 0xff)),
979 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
980 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
981 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
983 LogEvent(EVENTLOG_WARNING_TYPE, MSG_SERVER_REPORTS_VNOSERVICE,
984 addr, fidp->volume, cellp->name);
985 osi_Log3(afsd_logp, "Server %s reported rpc to volume %d in cell %s as not serviced.",
986 osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
992 else if (errorCode == RX_CALL_IDLE) {
994 * RPC failed because the server failed to respond with data
995 * within the idle dead timeout period. This could be for a variety
997 * 1. The server could have a bad partition such as a failed
998 * disk or iSCSI target and all I/O to that partition is
999 * blocking on the server and will never complete.
1001 * 2. The server vnode may be locked by another client request
1002 * that is taking a very long time.
1004 * 3. The server may have a very long queue of requests
1005 * pending and is unable to process this request.
1007 * 4. The server could be malicious and is performing a denial
1008 * of service attack against the client.
1010 * If this is a request against a .readonly with alternate sites
1011 * the server should be marked down for this request and the
1012 * client should fail over to another server. If this is a
1013 * request against a single source, the client may retry once.
1016 sprintf(addr, "%d.%d.%d.%d",
1017 ((serverp->addr.sin_addr.s_addr & 0xff)),
1018 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1019 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1020 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1023 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
1024 CM_GETVOL_FLAG_NO_LRU_UPDATE,
1027 statep = cm_VolumeStateByID(volp, fidp->volume);
1030 replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1032 lock_ObtainRead(&cm_volumeLock);
1034 lock_ReleaseRead(&cm_volumeLock);
1039 scp = cm_FindSCache(fidp);
1041 if (cm_HaveCallback(scp)) {
1042 lock_ObtainWrite(&scp->rw);
1043 cm_DiscardSCache(scp);
1044 lock_ReleaseWrite(&scp->rw);
1047 * We really should notify the redirector that we discarded
1048 * the status information but doing so in this case is not
1049 * safe as it can result in a deadlock with extent release
1053 cm_ReleaseSCache(scp);
1057 if (replicated && serverp) {
1058 reqp->errorServp = serverp;
1059 reqp->tokenError = errorCode;
1065 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
1066 osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
1067 osi_LogSaveString(afsd_logp,addr), retry);
1069 else if (errorCode == RX_CALL_DEAD) {
1070 /* mark server as down */
1072 sprintf(addr, "%d.%d.%d.%d",
1073 ((serverp->addr.sin_addr.s_addr & 0xff)),
1074 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1075 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1076 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1078 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
1079 osi_LogSaveString(afsd_logp,addr),
1080 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1083 if ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
1084 lock_ObtainMutex(&serverp->mx);
1085 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
1086 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
1087 serverp->downTime = time(NULL);
1089 lock_ReleaseMutex(&serverp->mx);
1091 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1093 cm_ForceNewConnections(serverp);
1097 if (fidp && storeOp)
1098 scp = cm_FindSCache(fidp);
1100 if (cm_HaveCallback(scp)) {
1101 lock_ObtainWrite(&scp->rw);
1102 cm_DiscardSCache(scp);
1103 lock_ReleaseWrite(&scp->rw);
1106 * We really should notify the redirector that we discarded
1107 * the status information but doing so in this case is not
1108 * safe as it can result in a deadlock with extent release
1112 cm_ReleaseSCache(scp);
1118 else if (errorCode >= -64 && errorCode < 0) {
1119 /* mark server as down */
1121 sprintf(addr, "%d.%d.%d.%d",
1122 ((serverp->addr.sin_addr.s_addr & 0xff)),
1123 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1124 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1125 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1127 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
1129 osi_LogSaveString(afsd_logp,addr),
1130 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1133 if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
1134 reqp->errorServp = serverp;
1135 reqp->tokenError = errorCode;
1137 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1139 cm_ForceNewConnections(serverp);
1146 else if (errorCode == RXKADEXPIRED) {
1147 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1149 if (!dead_session) {
1150 lock_ObtainMutex(&userp->mx);
1151 ucellp = cm_GetUCell(userp, cellp);
1152 if (ucellp->ticketp) {
1153 free(ucellp->ticketp);
1154 ucellp->ticketp = NULL;
1156 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1158 lock_ReleaseMutex(&userp->mx);
1162 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1163 char * s = "unknown error";
1164 switch ( errorCode ) {
1165 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1166 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1167 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1168 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1169 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1170 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1171 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1172 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1173 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1174 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1175 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1176 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1177 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1179 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1183 reqp->errorServp = serverp;
1184 reqp->tokenError = errorCode;
1187 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1189 * We received a ubik error. its possible that the server we are
1190 * communicating with has a corrupted database or is partitioned
1191 * from the rest of the servers and another server might be able
1192 * to answer our query. Therefore, we will retry the request
1193 * and force the use of another server.
1196 reqp->errorServp = serverp;
1197 reqp->tokenError = errorCode;
1200 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1201 cm_ForceNewConnections(serverp);
1206 char * s = "unknown error";
1207 switch ( errorCode ) {
1208 case VSALVAGE : s = "VSALVAGE"; break;
1209 case VNOVNODE : s = "VNOVNODE"; break;
1210 case VNOVOL : s = "VNOVOL"; break;
1211 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1212 case VNOSERVICE : s = "VNOSERVICE"; break;
1213 case VOFFLINE : s = "VOFFLINE"; break;
1214 case VONLINE : s = "VONLINE"; break;
1215 case VDISKFULL : s = "VDISKFULL"; break;
1216 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1217 case VBUSY : s = "VBUSY"; break;
1218 case VMOVED : s = "VMOVED"; break;
1219 case VIO : s = "VIO"; break;
1220 case VRESTRICTED : s = "VRESTRICTED"; break;
1221 case VRESTARTING : s = "VRESTARTING"; break;
1222 case VREADONLY : s = "VREADONLY"; break;
1223 case EAGAIN : s = "EAGAIN"; break;
1224 case UAEAGAIN : s = "UAEAGAIN"; break;
1225 case EINVAL : s = "EINVAL"; break;
1226 case UAEINVAL : s = "UAEINVAL"; break;
1227 case EACCES : s = "EACCES"; break;
1228 case UAEACCES : s = "UAEACCES"; break;
1229 case ENOENT : s = "ENOENT"; break;
1230 case UAENOENT : s = "UAENOENT"; break;
1231 case EEXIST : s = "EEXIST"; break;
1232 case UAEEXIST : s = "UAEEXIST"; break;
1233 case VICECONNBAD : s = "VICECONNBAD"; break;
1234 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1235 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1236 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1237 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1238 case VL_IO : s = "VL_IO"; break;
1239 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1240 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1241 case VL_NOENT : s = "VL_NOENT"; break;
1242 case VL_EMPTY : s = "VL_EMPTY"; break;
1243 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1244 case VL_BADNAME : s = "VL_BADNAME"; break;
1245 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1246 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1247 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1248 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1249 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1250 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1251 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1252 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1253 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1254 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1255 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1256 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1257 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1258 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1259 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1260 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1261 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1262 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1263 case VL_PERM : s = "VL_PERM"; break;
1264 case VL_NOMEM : s = "VL_NOMEM"; break;
1265 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1266 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1267 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1268 case VL_BADMASK : s = "VL_BADMASK"; break;
1269 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1270 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1271 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1272 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1273 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1274 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1275 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1276 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1277 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1278 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1279 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1280 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1281 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1282 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1283 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1284 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1285 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1286 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1287 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1288 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1289 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1290 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1291 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1292 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1293 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1294 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1295 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1296 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1297 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1298 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1299 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1300 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1301 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1302 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1303 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1304 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1305 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1306 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1307 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1308 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1309 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1310 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1311 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1312 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1313 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1314 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1315 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1316 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1317 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1318 case UAEIO : s = "UAEIO"; break;
1319 case EIO : s = "EIO"; break;
1321 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1327 /* If not allowed to retry, don't */
1329 !forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
1330 !(errorCode > -64 && errorCode <= RX_INVALID_OPERATION))
1333 /* drop this on the way out */
1338 * clear the volume updated flag if we succeed.
1339 * this way the flag will not prevent a subsequent volume
1340 * from being updated if necessary.
1344 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1348 errorCode != VBUSY &&
1349 errorCode != VRESTARTING &&
1350 errorCode != CM_ERROR_ALLBUSY)
1352 cm_ResetServerBusyStatus(serversp);
1355 /* retry until we fail to find a connection */
1359 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1360 cm_req_t *reqp, cm_conn_t **connpp)
1363 cm_serverRef_t *tsrp;
1365 long firstError = 0;
1366 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1, allDeleted = 1;
1367 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1368 long timeUsed, timeLeft, hardTimeLeft;
1372 if (serversp == NULL) {
1373 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_EMPTY);
1374 return CM_ERROR_EMPTY;
1377 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1378 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1380 /* leave 5 seconds margin of safety */
1381 timeLeft = ConnDeadtimeout - timeUsed - 5;
1382 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1385 lock_ObtainRead(&cm_serverLock);
1386 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1387 if (tsrp->status == srv_deleted)
1393 if (reqp->errorServp) {
1395 * search the list until we find the server
1396 * that failed last time. When we find it
1397 * clear the error, skip it and try the next one
1400 if (tsp == reqp->errorServp)
1401 reqp->errorServp = NULL;
1405 cm_GetServerNoLock(tsp);
1406 lock_ReleaseRead(&cm_serverLock);
1407 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1409 if (tsrp->status == srv_busy) {
1412 } else if (tsrp->status == srv_offline) {
1418 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1419 if (code == 0) { /* cm_CBS only returns 0 */
1421 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1422 /* Set RPC timeout */
1423 if (timeLeft > ConnDeadtimeout)
1424 timeLeft = ConnDeadtimeout;
1426 if (hardTimeLeft > HardDeadtimeout)
1427 hardTimeLeft = HardDeadtimeout;
1429 lock_ObtainMutex(&(*connpp)->mx);
1430 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1431 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1432 lock_ReleaseMutex(&(*connpp)->mx);
1437 /* therefore, this code is never executed */
1438 if (firstError == 0)
1442 lock_ObtainRead(&cm_serverLock);
1443 cm_PutServerNoLock(tsp);
1446 lock_ReleaseRead(&cm_serverLock);
1448 if (firstError == 0) {
1450 firstError = CM_ERROR_EMPTY;
1451 } else if (allDown) {
1452 firstError = (reqp->tokenError ? reqp->tokenError :
1453 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1455 * if we experienced either a token error or and idle dead time error
1456 * and now all of the servers are down, we have either tried them
1457 * all or lost connectivity. Clear the error we are returning so
1458 * we will not return it indefinitely if the request is retried.
1460 reqp->idleError = reqp->tokenError = 0;
1461 } else if (allBusy) {
1462 firstError = CM_ERROR_ALLBUSY;
1463 } else if (allOffline || (someBusy && someOffline)) {
1464 firstError = CM_ERROR_ALLOFFLINE;
1466 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1467 firstError = CM_ERROR_TIMEDOUT;
1471 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1475 /* called with a held server to GC all bad connections hanging off of the server */
1476 void cm_GCConnections(cm_server_t *serverp)
1482 lock_ObtainWrite(&cm_connLock);
1483 lcpp = &serverp->connsp;
1484 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1486 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1487 /* do the deletion of this guy */
1488 cm_PutServer(tcp->serverp);
1489 cm_ReleaseUser(userp);
1491 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1492 rx_DestroyConnection(tcp->rxconnp);
1493 lock_FinalizeMutex(&tcp->mx);
1497 /* just advance to the next */
1501 lock_ReleaseWrite(&cm_connLock);
1504 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1505 cm_server_t *serverp, afs_uint32 replicated)
1507 unsigned short port;
1510 struct rx_securityClass *secObjp;
1512 port = serverp->addr.sin_port;
1513 switch (serverp->type) {
1514 case CM_SERVER_VLDB:
1519 case CM_SERVER_FILE:
1525 osi_panic("unknown server type", __FILE__, __LINE__);
1528 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1532 tcp->cryptlevel = rxkad_clear;
1535 tcp->cryptlevel = rxkad_auth;
1538 tcp->cryptlevel = rxkad_crypt;
1540 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1541 &ucellp->sessionKey, ucellp->kvno,
1542 ucellp->ticketLen, ucellp->ticketp);
1546 tcp->cryptlevel = rxkad_clear;
1547 secObjp = rxnull_NewClientSecurityObject();
1549 osi_assertx(secObjp != NULL, "null rx_securityClass");
1550 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1555 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1556 if (smb_Enabled || tcp->serverp->type == CM_SERVER_VLDB)
1557 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1560 * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
1563 tcp->flags &= CM_CONN_FLAG_REPLICATION;
1564 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1566 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1570 * Let the Rx library know that we can auto-retry if an
1571 * RX_MSGSIZE error is returned.
1573 if (rx_pmtu_discovery)
1574 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1577 * Attempt to limit NAT pings to the anonymous file server connections.
1578 * Only file servers implement client callbacks and we only need one ping
1579 * to be sent to each server.
1581 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1582 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1583 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1586 tcp->ucgen = ucellp->gen;
1588 rxs_Release(secObjp); /* Decrement the initial refCount */
1591 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1598 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1599 userp = cm_rootUserp;
1601 lock_ObtainMutex(&userp->mx);
1602 lock_ObtainRead(&cm_connLock);
1603 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1604 if (tcp->userp == userp &&
1605 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1606 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1610 /* find ucell structure */
1611 ucellp = cm_GetUCell(userp, serverp->cellp);
1613 lock_ConvertRToW(&cm_connLock);
1614 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1615 if (tcp->userp == userp)
1619 InterlockedIncrement(&tcp->refCount);
1620 lock_ReleaseWrite(&cm_connLock);
1623 cm_GetServer(serverp);
1624 tcp = malloc(sizeof(*tcp));
1625 memset(tcp, 0, sizeof(*tcp));
1628 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1629 tcp->serverp = serverp;
1630 tcp->cryptlevel = rxkad_clear;
1631 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1633 tcp->nextp = serverp->connsp;
1634 serverp->connsp = tcp;
1635 lock_ReleaseWrite(&cm_connLock);
1636 lock_ReleaseMutex(&userp->mx);
1638 InterlockedIncrement(&tcp->refCount);
1639 lock_ReleaseRead(&cm_connLock);
1641 lock_ReleaseMutex(&userp->mx);
1643 lock_ObtainMutex(&tcp->mx);
1644 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1645 (tcp->ucgen < ucellp->gen) ||
1646 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1648 if (tcp->ucgen < ucellp->gen)
1649 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1651 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1652 tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
1653 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1654 rx_DestroyConnection(tcp->rxconnp);
1655 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1657 lock_ReleaseMutex(&tcp->mx);
1660 /* return this pointer to our caller */
1661 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1667 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1671 cm_serverRef_t **serverspp;
1672 cm_serverRef_t *tsrp;
1674 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1675 afs_uint32 replicated;
1679 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1683 lock_ObtainRead(&cm_serverLock);
1684 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1685 if (tsrp->status == srv_deleted)
1688 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1690 if (tsrp->status == srv_busy) {
1693 } else if (tsrp->status == srv_offline) {
1702 lock_ReleaseRead(&cm_serverLock);
1703 cm_FreeServerList(serverspp, 0);
1709 else if (allOffline || (someBusy && someOffline))
1716 * The returned cm_conn_t ** object is released in the subsequent call
1719 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1723 cm_serverRef_t **serverspp;
1724 afs_uint32 replicated;
1728 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1732 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1733 cm_FreeServerList(serverspp, 0);
1738 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1742 cm_serverRef_t **serverspp;
1743 afs_uint32 replicated;
1747 code = cm_GetVolServerList(volp, volid, userp, reqp, &replicated, &serverspp);
1751 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1752 cm_FreeServerList(serverspp, 0);
1757 extern struct rx_connection *
1758 cm_GetRxConn(cm_conn_t *connp)
1760 struct rx_connection * rxconnp;
1761 lock_ObtainMutex(&connp->mx);
1762 rxconnp = connp->rxconnp;
1763 rx_GetConnection(rxconnp);
1764 lock_ReleaseMutex(&connp->mx);
1768 void cm_ForceNewConnections(cm_server_t *serverp)
1772 lock_ObtainWrite(&cm_connLock);
1773 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1774 lock_ObtainMutex(&tcp->mx);
1775 tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
1776 lock_ReleaseMutex(&tcp->mx);
1778 lock_ReleaseWrite(&cm_connLock);