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 static long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
214 struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
217 cm_volume_t *volp = NULL;
218 cm_vol_state_t *volstatep = NULL;
219 cm_cell_t *cellp = NULL;
223 return CM_ERROR_INVAL;
226 cellp = cm_FindCellByID(fidp->cell, 0);
228 return CM_ERROR_NOSUCHCELL;
230 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
234 volstatep = cm_VolumeStateByID(volp, fidp->volume);
235 *replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
236 *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp);
238 lock_ObtainRead(&cm_volumeLock);
240 lock_ReleaseRead(&cm_volumeLock);
241 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
245 cm_SetServerBusyStatus(cm_serverRef_t *serversp, cm_server_t *serverp)
247 cm_serverRef_t *tsrp;
249 lock_ObtainWrite(&cm_serverLock);
250 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
251 if (tsrp->status == srv_deleted)
253 if (tsrp->server == serverp && tsrp->status == srv_not_busy) {
254 tsrp->status = srv_busy;
258 lock_ReleaseWrite(&cm_serverLock);
262 cm_ResetServerBusyStatus(cm_serverRef_t *serversp)
264 cm_serverRef_t *tsrp;
266 lock_ObtainWrite(&cm_serverLock);
267 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
268 if (tsrp->status == srv_deleted)
270 if (tsrp->status == srv_busy) {
271 tsrp->status = srv_not_busy;
274 lock_ReleaseWrite(&cm_serverLock);
278 * Analyze the error return from an RPC. Determine whether or not to retry,
279 * and if we're going to retry, determine whether failover is appropriate,
280 * and whether timed backoff is appropriate.
282 * If the error code is from cm_ConnFromFID() or friends, it will be a CM_ERROR code.
283 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
284 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
285 * be a security code (e.g. RXKADEXPIRED).
287 * If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
289 * For VLDB calls, fidp will be NULL.
291 * volSyncp and/or cbrp may also be NULL.
294 cm_Analyze(cm_conn_t *connp,
299 AFSVolSync *volSyncp,
300 cm_serverRef_t * serversp,
301 cm_callbackRequest_t *cbrp,
304 cm_server_t *serverp = NULL;
305 cm_serverRef_t **serverspp = NULL;
306 cm_serverRef_t *tsrp;
307 cm_cell_t *cellp = NULL;
309 cm_volume_t * volp = NULL;
310 cm_vol_state_t *statep = NULL;
311 cm_scache_t * scp = NULL;
312 afs_uint32 replicated;
314 int free_svr_list = 0;
316 long timeUsed, timeLeft;
318 char addr[16]="unknown";
320 int location_updated = 0;
324 osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
327 /* no locking required, since connp->serverp never changes after
329 dead_session = (userp->cellInfop == NULL);
331 serverp = connp->serverp;
333 /* Update callback pointer */
334 if (cbrp && serverp && errorCode == 0) {
336 if ( cbrp->serverp != serverp ) {
337 lock_ObtainWrite(&cm_serverLock);
338 cm_PutServerNoLock(cbrp->serverp);
339 cm_GetServerNoLock(serverp);
340 lock_ReleaseWrite(&cm_serverLock);
343 cm_GetServer(serverp);
345 cbrp->serverp = serverp;
348 /* if timeout - check that it did not exceed the HardDead timeout
351 /* timeleft - get it from reqp the same way as cm_ConnByMServers does */
352 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
353 if ( reqp->flags & CM_REQ_SOURCE_SMB )
354 timeLeft = HardDeadtimeout - timeUsed;
356 timeLeft = 0x0FFFFFFF;
358 /* get a pointer to the cell */
360 if (cellp == NULL && serverp)
361 cellp = serverp->cellp;
362 if (cellp == NULL && serversp) {
363 struct cm_serverRef * refp;
364 for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
365 if (refp->status == srv_deleted)
368 cellp = refp->server->cellp;
371 if (cellp == NULL && fidp) {
372 cellp = cm_FindCellByID(fidp->cell, 0);
376 if (errorCode == CM_ERROR_TIMEDOUT) {
377 if ( timeLeft > 5 ) {
379 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
384 else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
385 errorCode == UAEAGAIN || errorCode == EAGAIN) {
386 osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
387 if ( timeLeft > 5 ) {
393 /* if there is nosuchvolume, then we have a situation in which a
394 * previously known volume no longer has a set of servers
395 * associated with it. Either the volume has moved or
396 * the volume has been deleted. Try to find a new server list
397 * until the timeout period expires.
399 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
400 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
402 * The VNOVOL or VL_NOENT error has already been translated
403 * to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
407 else if (errorCode == CM_ERROR_ALLDOWN) {
408 /* Servers marked DOWN will be restored by the background daemon
409 * thread as they become available. The volume status is
410 * updated as the server state changes.
413 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
414 cellp->name, fidp->volume);
415 msgID = MSG_ALL_SERVERS_DOWN;
416 format = "All servers are unreachable when accessing cell %s volume %d.";
417 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
419 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
423 else if (errorCode == CM_ERROR_ALLOFFLINE) {
424 /* Volume instances marked offline will be restored by the
425 * background daemon thread as they become available
428 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
429 cellp->name, fidp->volume);
430 msgID = MSG_ALL_SERVERS_OFFLINE;
431 format = "All servers are offline when accessing cell %s volume %d.";
432 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
435 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
437 serversp = *serverspp;
441 cm_ResetServerBusyStatus(serversp);
443 cm_FreeServerList(serverspp, 0);
448 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
449 CM_GETVOL_FLAG_NO_LRU_UPDATE,
453 * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
454 * was called by cm_CheckOfflineVolumeState().
456 if (!(reqp->flags & CM_REQ_OFFLINE_VOL_CHK) && timeLeft > 7) {
459 /* cm_CheckOfflineVolume() resets the serverRef state */
460 if (cm_CheckOfflineVolume(volp, fidp->volume))
463 cm_UpdateVolumeStatus(volp, fidp->volume);
465 lock_ObtainRead(&cm_volumeLock);
467 lock_ReleaseRead(&cm_volumeLock);
471 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
474 else if (errorCode == CM_ERROR_ALLBUSY) {
475 /* Volumes that are busy cannot be determined to be non-busy
476 * without actually attempting to access them.
478 if (fidp) { /* File Server query */
479 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
480 cellp->name, fidp->volume);
481 msgID = MSG_ALL_SERVERS_BUSY;
482 format = "All servers are busy when accessing cell %s volume %d.";
483 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
486 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
488 serversp = *serverspp;
492 cm_ResetServerBusyStatus(serversp);
494 cm_FreeServerList(serverspp, 0);
499 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
500 CM_GETVOL_FLAG_NO_LRU_UPDATE,
505 statep = cm_VolumeStateByID(volp, fidp->volume);
508 cm_UpdateVolumeStatus(volp, fidp->volume);
510 lock_ObtainRead(&cm_volumeLock);
512 lock_ReleaseRead(&cm_volumeLock);
515 } else { /* VL Server query */
516 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
522 cm_ResetServerBusyStatus(serversp);
529 /* special codes: VBUSY and VRESTARTING */
530 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
531 if (!serversp && fidp) {
532 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
534 serversp = *serverspp;
539 switch ( errorCode ) {
541 msgID = MSG_SERVER_REPORTS_VBUSY;
542 format = "Server %s reported busy when accessing volume %d in cell %s.";
545 msgID = MSG_SERVER_REPORTS_VRESTARTING;
546 format = "Server %s reported restarting when accessing volume %d in cell %s.";
550 if (serverp && fidp) {
551 /* Log server being offline for this volume */
552 sprintf(addr, "%d.%d.%d.%d",
553 ((serverp->addr.sin_addr.s_addr & 0xff)),
554 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
555 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
556 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
558 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
559 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
562 cm_SetServerBusyStatus(serversp, serverp);
564 if (fidp) { /* File Server query */
565 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
566 CM_GETVOL_FLAG_NO_LRU_UPDATE,
569 statep = cm_VolumeStateByID(volp, fidp->volume);
572 cm_UpdateVolumeStatus(volp, statep->ID);
574 lock_ObtainRead(&cm_volumeLock);
576 lock_ReleaseRead(&cm_volumeLock);
582 cm_FreeServerList(serverspp, 0);
589 /* special codes: missing volumes */
590 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
591 errorCode == VSALVAGE || errorCode == VIO)
593 /* In case of timeout */
594 reqp->volumeError = errorCode;
596 switch ( errorCode ) {
598 msgID = MSG_SERVER_REPORTS_VNOVOL;
599 format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
602 msgID = MSG_SERVER_REPORTS_VMOVED;
603 format = "Server %s reported volume %d in cell %s as moved.";
606 msgID = MSG_SERVER_REPORTS_VOFFLINE;
607 format = "Server %s reported volume %d in cell %s as offline.";
610 msgID = MSG_SERVER_REPORTS_VSALVAGE;
611 format = "Server %s reported volume %d in cell %s as needs salvage.";
614 msgID = MSG_SERVER_REPORTS_VIO;
615 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
619 if (fidp) { /* File Server query */
621 /* Log server being unavailable for this volume */
622 sprintf(addr, "%d.%d.%d.%d",
623 ((serverp->addr.sin_addr.s_addr & 0xff)),
624 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
625 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
626 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
628 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
629 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
632 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
633 CM_GETVOL_FLAG_NO_LRU_UPDATE,
636 statep = cm_VolumeStateByID(volp, fidp->volume);
638 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
639 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
641 code = cm_ForceUpdateVolume(fidp, userp, reqp);
643 location_updated = 1;
645 /* Even if the update fails, there might still be another replica */
647 reqp->flags |= CM_REQ_VOLUME_UPDATED;
648 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
649 fidp->cell, fidp->volume, code);
653 cm_UpdateVolumeStatus(volp, statep->ID);
654 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
655 fidp->cell, fidp->volume, statep->state);
659 lock_ObtainRead(&cm_volumeLock);
661 lock_ReleaseRead(&cm_volumeLock);
666 * Mark server offline for this volume or delete the volume
667 * from the server list if it was moved or is not present.
669 if (!serversp || location_updated) {
670 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
672 serversp = *serverspp;
678 lock_ObtainWrite(&cm_serverLock);
679 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
680 if (tsrp->status == srv_deleted)
683 sprintf(addr, "%d.%d.%d.%d",
684 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
685 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
686 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
687 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
689 if (cm_ServerEqual(tsrp->server, serverp)) {
691 if (errorCode == VMOVED || errorCode == VNOVOL) {
692 osi_Log2(afsd_logp, "volume %d not present on server %s",
693 fidp->volume, osi_LogSaveString(afsd_logp,addr));
694 tsrp->status = srv_deleted;
696 cm_RemoveVolumeFromServer(serverp, fidp->volume);
698 osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
699 fidp->volume, osi_LogSaveString(afsd_logp,addr));
700 tsrp->status = srv_offline;
704 osi_Log3(afsd_logp, "volume %d exists on server %s with status %u",
705 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
708 lock_ReleaseWrite(&cm_serverLock);
710 /* Free the server list before cm_ForceUpdateVolume is called */
712 cm_FreeServerList(serverspp, 0);
719 } else if ( errorCode == VNOVNODE ) {
721 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
722 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
724 scp = cm_FindSCache(fidp);
726 cm_scache_t *pscp = NULL;
728 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
729 pscp = cm_FindSCacheParent(scp);
731 lock_ObtainWrite(&scp->rw);
732 scp->flags |= CM_SCACHEFLAG_DELETED;
733 lock_ObtainWrite(&cm_scacheLock);
734 cm_AdjustScacheLRU(scp);
735 cm_RemoveSCacheFromHashTable(scp);
736 lock_ReleaseWrite(&cm_scacheLock);
737 cm_LockMarkSCacheLost(scp);
738 lock_ReleaseWrite(&scp->rw);
740 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
741 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
742 cm_ReleaseSCache(scp);
745 if (cm_HaveCallback(pscp)) {
746 lock_ObtainWrite(&pscp->rw);
747 cm_DiscardSCache(pscp);
748 lock_ReleaseWrite(&pscp->rw);
751 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
752 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
755 cm_ReleaseSCache(pscp);
759 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
764 else if (errorCode == RX_CALL_TIMEOUT) {
765 /* RPC took longer than hardDeadTime or the server
766 * reported idle for longer than idleDeadTime
767 * don't mark server as down but don't retry
768 * this is to prevent the SMB session from timing out
769 * In addition, we log an event to the event log
773 sprintf(addr, "%d.%d.%d.%d",
774 ((serverp->addr.sin_addr.s_addr & 0xff)),
775 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
776 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
777 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
779 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
780 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
781 osi_LogSaveString(afsd_logp,addr));
782 reqp->tokenIdleErrorServp = serverp;
787 scp = cm_FindSCache(fidp);
789 if (cm_HaveCallback(scp)) {
790 lock_ObtainWrite(&scp->rw);
791 cm_DiscardSCache(scp);
792 lock_ReleaseWrite(&scp->rw);
795 * We really should notify the redirector that we discarded
796 * the status information but doing so in this case is not
797 * safe as it can result in a deadlock with extent release
801 cm_ReleaseSCache(scp);
805 if (!fidp) { /* vldb */
808 cm_volume_t *volp = cm_GetVolumeByFID(fidp);
810 if (fidp->volume == cm_GetROVolumeID(volp))
817 else if (errorCode == RX_MSGSIZE) {
819 * RPC failed because a transmitted rx packet
820 * may have grown larger than the path mtu.
821 * Force a retry and the Rx library will try
822 * with a smaller mtu size.
826 sprintf(addr, "%d.%d.%d.%d",
827 ((serverp->addr.sin_addr.s_addr & 0xff)),
828 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
829 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
830 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
832 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
833 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
834 osi_LogSaveString(afsd_logp,addr));
838 else if (errorCode == RX_CALL_BUSY) {
840 * RPC failed because the selected call channel
841 * is currently busy on the server. Unconditionally
842 * retry the request so an alternate call channel can be used.
845 sprintf(addr, "%d.%d.%d.%d",
846 ((serverp->addr.sin_addr.s_addr & 0xff)),
847 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
848 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
849 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
851 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
852 osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
853 osi_LogSaveString(afsd_logp,addr));
856 else if (errorCode == VNOSERVICE) {
858 * The server did not service the RPC.
859 * If this was a file server RPC it means that for at
860 * least the file server's idle dead timeout period the
861 * file server did not receive any new data packets from
864 * The RPC was not serviced so it can be retried and any
865 * existing status information is still valid.
869 sprintf(addr, "%d.%d.%d.%d",
870 ((serverp->addr.sin_addr.s_addr & 0xff)),
871 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
872 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
873 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
875 LogEvent(EVENTLOG_WARNING_TYPE, MSG_SERVER_REPORTS_VNOSERVICE, addr);
876 osi_Log1(afsd_logp, "Server %s reported volume %d in cell %s as not in service.",
877 osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
883 else if (errorCode == RX_CALL_IDLE) {
885 * RPC failed because the server failed to respond with data
886 * within the idle dead timeout period. This could be for a variety
888 * 1. The server could have a bad partition such as a failed
889 * disk or iSCSI target and all I/O to that partition is
890 * blocking on the server and will never complete.
892 * 2. The server vnode may be locked by another client request
893 * that is taking a very long time.
895 * 3. The server may have a very long queue of requests
896 * pending and is unable to process this request.
898 * 4. The server could be malicious and is performing a denial
899 * of service attack against the client.
901 * If this is a request against a .readonly with alternate sites
902 * the server should be marked down for this request and the
903 * client should fail over to another server. If this is a
904 * request against a single source, the client may retry once.
907 sprintf(addr, "%d.%d.%d.%d",
908 ((serverp->addr.sin_addr.s_addr & 0xff)),
909 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
910 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
911 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
914 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
915 CM_GETVOL_FLAG_NO_LRU_UPDATE,
918 statep = cm_VolumeStateByID(volp, fidp->volume);
921 replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
923 lock_ObtainRead(&cm_volumeLock);
925 lock_ReleaseRead(&cm_volumeLock);
930 scp = cm_FindSCache(fidp);
932 if (cm_HaveCallback(scp)) {
933 lock_ObtainWrite(&scp->rw);
934 cm_DiscardSCache(scp);
935 lock_ReleaseWrite(&scp->rw);
938 * We really should notify the redirector that we discarded
939 * the status information but doing so in this case is not
940 * safe as it can result in a deadlock with extent release
944 cm_ReleaseSCache(scp);
948 if (replicated && serverp) {
949 reqp->tokenIdleErrorServp = serverp;
950 reqp->tokenError = errorCode;
956 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
957 osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
958 osi_LogSaveString(afsd_logp,addr), retry);
960 else if (errorCode == RX_CALL_DEAD) {
961 /* mark server as down */
963 sprintf(addr, "%d.%d.%d.%d",
964 ((serverp->addr.sin_addr.s_addr & 0xff)),
965 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
966 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
967 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
969 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
970 osi_LogSaveString(afsd_logp,addr),
971 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
974 if ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
975 lock_ObtainMutex(&serverp->mx);
976 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
977 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
978 serverp->downTime = time(NULL);
980 lock_ReleaseMutex(&serverp->mx);
982 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
984 cm_ForceNewConnections(serverp);
989 scp = cm_FindSCache(fidp);
991 if (cm_HaveCallback(scp)) {
992 lock_ObtainWrite(&scp->rw);
993 cm_DiscardSCache(scp);
994 lock_ReleaseWrite(&scp->rw);
997 * We really should notify the redirector that we discarded
998 * the status information but doing so in this case is not
999 * safe as it can result in a deadlock with extent release
1003 cm_ReleaseSCache(scp);
1009 else if (errorCode >= -64 && errorCode < 0) {
1010 /* mark server as down */
1012 sprintf(addr, "%d.%d.%d.%d",
1013 ((serverp->addr.sin_addr.s_addr & 0xff)),
1014 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1015 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1016 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1018 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
1020 osi_LogSaveString(afsd_logp,addr),
1021 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1024 if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
1025 reqp->tokenIdleErrorServp = serverp;
1026 reqp->tokenError = errorCode;
1028 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1030 cm_ForceNewConnections(serverp);
1036 else if (errorCode == RXKADEXPIRED) {
1037 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1039 if (!dead_session) {
1040 lock_ObtainMutex(&userp->mx);
1041 ucellp = cm_GetUCell(userp, serverp->cellp);
1042 if (ucellp->ticketp) {
1043 free(ucellp->ticketp);
1044 ucellp->ticketp = NULL;
1046 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1048 lock_ReleaseMutex(&userp->mx);
1052 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1053 char * s = "unknown error";
1054 switch ( errorCode ) {
1055 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1056 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1057 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1058 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1059 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1060 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1061 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1062 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1063 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1064 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1065 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1066 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1067 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1069 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1073 reqp->tokenIdleErrorServp = serverp;
1074 reqp->tokenError = errorCode;
1077 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1079 * We received a ubik error. its possible that the server we are
1080 * communicating with has a corrupted database or is partitioned
1081 * from the rest of the servers and another server might be able
1082 * to answer our query. Therefore, we will retry the request
1083 * and force the use of another server.
1086 reqp->tokenIdleErrorServp = serverp;
1087 reqp->tokenError = errorCode;
1090 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1091 cm_ForceNewConnections(serverp);
1096 char * s = "unknown error";
1097 switch ( errorCode ) {
1098 case VSALVAGE : s = "VSALVAGE"; break;
1099 case VNOVNODE : s = "VNOVNODE"; break;
1100 case VNOVOL : s = "VNOVOL"; break;
1101 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1102 case VNOSERVICE : s = "VNOSERVICE"; break;
1103 case VOFFLINE : s = "VOFFLINE"; break;
1104 case VONLINE : s = "VONLINE"; break;
1105 case VDISKFULL : s = "VDISKFULL"; break;
1106 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1107 case VBUSY : s = "VBUSY"; break;
1108 case VMOVED : s = "VMOVED"; break;
1109 case VIO : s = "VIO"; break;
1110 case VRESTRICTED : s = "VRESTRICTED"; break;
1111 case VRESTARTING : s = "VRESTARTING"; break;
1112 case VREADONLY : s = "VREADONLY"; break;
1113 case EAGAIN : s = "EAGAIN"; break;
1114 case UAEAGAIN : s = "UAEAGAIN"; break;
1115 case EINVAL : s = "EINVAL"; break;
1116 case UAEINVAL : s = "UAEINVAL"; break;
1117 case EACCES : s = "EACCES"; break;
1118 case UAEACCES : s = "UAEACCES"; break;
1119 case ENOENT : s = "ENOENT"; break;
1120 case UAENOENT : s = "UAENOENT"; break;
1121 case EEXIST : s = "EEXIST"; break;
1122 case UAEEXIST : s = "UAEEXIST"; break;
1123 case VICECONNBAD : s = "VICECONNBAD"; break;
1124 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1125 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1126 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1127 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1128 case VL_IO : s = "VL_IO"; break;
1129 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1130 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1131 case VL_NOENT : s = "VL_NOENT"; break;
1132 case VL_EMPTY : s = "VL_EMPTY"; break;
1133 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1134 case VL_BADNAME : s = "VL_BADNAME"; break;
1135 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1136 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1137 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1138 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1139 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1140 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1141 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1142 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1143 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1144 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1145 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1146 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1147 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1148 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1149 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1150 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1151 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1152 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1153 case VL_PERM : s = "VL_PERM"; break;
1154 case VL_NOMEM : s = "VL_NOMEM"; break;
1155 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1156 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1157 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1158 case VL_BADMASK : s = "VL_BADMASK"; break;
1159 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1160 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1161 case CM_ERROR_TIMEDOUT : s = "CM_ERROR_TIMEDOUT"; break;
1162 case CM_ERROR_RETRY : s = "CM_ERROR_RETRY"; break;
1163 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1164 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1165 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1166 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1167 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1168 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1169 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1170 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1171 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1172 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1173 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1174 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1175 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1176 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1177 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1178 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1179 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1180 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1181 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1182 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1183 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1184 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1185 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1186 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1187 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1188 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1189 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1190 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1191 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1192 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1193 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1194 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1195 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1196 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1197 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1198 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1199 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1200 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1201 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1202 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1203 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1204 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1205 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1206 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1207 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1208 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1209 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1211 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1217 /* If not allowed to retry, don't */
1218 if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
1219 (errorCode != RX_MSGSIZE && errorCode != RX_CALL_BUSY))
1221 else if (retry && dead_session)
1224 /* drop this on the way out */
1229 * clear the volume updated flag if we succeed.
1230 * this way the flag will not prevent a subsequent volume
1231 * from being updated if necessary.
1235 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1239 errorCode != VBUSY &&
1240 errorCode != VRESTARTING &&
1241 errorCode != CM_ERROR_ALLBUSY)
1243 cm_ResetServerBusyStatus(serversp);
1246 /* retry until we fail to find a connection */
1250 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1251 cm_req_t *reqp, cm_conn_t **connpp)
1254 cm_serverRef_t *tsrp;
1256 long firstError = 0;
1257 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1258 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1259 long timeUsed, timeLeft, hardTimeLeft;
1263 if (serversp == NULL) {
1264 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_ALLDOWN);
1265 return CM_ERROR_ALLDOWN;
1268 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1269 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1271 /* leave 5 seconds margin of safety */
1272 timeLeft = ConnDeadtimeout - timeUsed - 5;
1273 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1276 lock_ObtainRead(&cm_serverLock);
1277 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1278 if (tsrp->status == srv_deleted)
1282 if (reqp->tokenIdleErrorServp) {
1284 * search the list until we find the server
1285 * that failed last time. When we find it
1286 * clear the error, skip it and try the next one
1289 if (tsp == reqp->tokenIdleErrorServp)
1290 reqp->tokenIdleErrorServp = NULL;
1294 cm_GetServerNoLock(tsp);
1295 lock_ReleaseRead(&cm_serverLock);
1296 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1298 if (tsrp->status == srv_busy) {
1301 } else if (tsrp->status == srv_offline) {
1307 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1308 if (code == 0) { /* cm_CBS only returns 0 */
1310 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1311 /* Set RPC timeout */
1312 if (timeLeft > ConnDeadtimeout)
1313 timeLeft = ConnDeadtimeout;
1315 if (hardTimeLeft > HardDeadtimeout)
1316 hardTimeLeft = HardDeadtimeout;
1318 lock_ObtainMutex(&(*connpp)->mx);
1319 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1320 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1321 lock_ReleaseMutex(&(*connpp)->mx);
1326 /* therefore, this code is never executed */
1327 if (firstError == 0)
1331 lock_ObtainRead(&cm_serverLock);
1332 cm_PutServerNoLock(tsp);
1335 lock_ReleaseRead(&cm_serverLock);
1337 if (firstError == 0) {
1339 firstError = (reqp->tokenError ? reqp->tokenError :
1340 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1342 * if we experienced either a token error or and idle dead time error
1343 * and now all of the servers are down, we have either tried them
1344 * all or lost connectivity. Clear the error we are returning so
1345 * we will not return it indefinitely if the request is retried.
1347 reqp->idleError = reqp->tokenError = 0;
1348 } else if (allBusy) {
1349 firstError = CM_ERROR_ALLBUSY;
1350 } else if (allOffline || (someBusy && someOffline)) {
1351 firstError = CM_ERROR_ALLOFFLINE;
1353 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1354 firstError = CM_ERROR_TIMEDOUT;
1358 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1362 /* called with a held server to GC all bad connections hanging off of the server */
1363 void cm_GCConnections(cm_server_t *serverp)
1369 lock_ObtainWrite(&cm_connLock);
1370 lcpp = &serverp->connsp;
1371 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1373 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1374 /* do the deletion of this guy */
1375 cm_PutServer(tcp->serverp);
1376 cm_ReleaseUser(userp);
1378 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1379 rx_DestroyConnection(tcp->rxconnp);
1380 lock_FinalizeMutex(&tcp->mx);
1384 /* just advance to the next */
1388 lock_ReleaseWrite(&cm_connLock);
1391 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1392 cm_server_t *serverp, afs_uint32 replicated)
1394 unsigned short port;
1397 struct rx_securityClass *secObjp;
1399 port = serverp->addr.sin_port;
1400 switch (serverp->type) {
1401 case CM_SERVER_VLDB:
1406 case CM_SERVER_FILE:
1412 osi_panic("unknown server type", __FILE__, __LINE__);
1415 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1419 tcp->cryptlevel = rxkad_clear;
1422 tcp->cryptlevel = rxkad_auth;
1425 tcp->cryptlevel = rxkad_crypt;
1427 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1428 &ucellp->sessionKey, ucellp->kvno,
1429 ucellp->ticketLen, ucellp->ticketp);
1433 tcp->cryptlevel = rxkad_clear;
1434 secObjp = rxnull_NewClientSecurityObject();
1436 osi_assertx(secObjp != NULL, "null rx_securityClass");
1437 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1442 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1443 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1446 * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
1449 tcp->flags &= CM_CONN_FLAG_REPLICATION;
1450 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1452 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1456 * Let the Rx library know that we can auto-retry if an
1457 * RX_MSGSIZE error is returned.
1459 if (rx_pmtu_discovery)
1460 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1463 * Attempt to limit NAT pings to the anonymous file server connections.
1464 * Only file servers implement client callbacks and we only need one ping
1465 * to be sent to each server.
1467 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1468 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1469 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1472 tcp->ucgen = ucellp->gen;
1474 rxs_Release(secObjp); /* Decrement the initial refCount */
1477 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1484 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1485 userp = cm_rootUserp;
1487 lock_ObtainMutex(&userp->mx);
1488 lock_ObtainRead(&cm_connLock);
1489 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1490 if (tcp->userp == userp &&
1491 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1492 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1496 /* find ucell structure */
1497 ucellp = cm_GetUCell(userp, serverp->cellp);
1499 lock_ConvertRToW(&cm_connLock);
1500 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1501 if (tcp->userp == userp)
1505 lock_ReleaseWrite(&cm_connLock);
1508 cm_GetServer(serverp);
1509 tcp = malloc(sizeof(*tcp));
1510 memset(tcp, 0, sizeof(*tcp));
1511 tcp->nextp = serverp->connsp;
1512 serverp->connsp = tcp;
1515 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1516 lock_ObtainMutex(&tcp->mx);
1517 tcp->serverp = serverp;
1518 tcp->cryptlevel = rxkad_clear;
1519 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1521 lock_ReleaseMutex(&tcp->mx);
1522 lock_ReleaseWrite(&cm_connLock);
1524 lock_ReleaseRead(&cm_connLock);
1526 InterlockedIncrement(&tcp->refCount);
1528 lock_ObtainMutex(&tcp->mx);
1529 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1530 (tcp->ucgen < ucellp->gen) ||
1531 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1533 if (tcp->ucgen < ucellp->gen)
1534 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1536 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1537 tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
1538 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1539 rx_DestroyConnection(tcp->rxconnp);
1540 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1542 lock_ReleaseMutex(&tcp->mx);
1544 lock_ReleaseMutex(&userp->mx);
1546 /* return this pointer to our caller */
1547 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1553 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1557 cm_serverRef_t **serverspp;
1558 cm_serverRef_t *tsrp;
1560 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1561 afs_uint32 replicated;
1565 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1569 lock_ObtainRead(&cm_serverLock);
1570 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1571 if (tsrp->status == srv_deleted)
1574 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1576 if (tsrp->status == srv_busy) {
1579 } else if (tsrp->status == srv_offline) {
1588 lock_ReleaseRead(&cm_serverLock);
1589 cm_FreeServerList(serverspp, 0);
1595 else if (allOffline || (someBusy && someOffline))
1602 * The returned cm_conn_t ** object is released in the subsequent call
1605 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1609 cm_serverRef_t **serverspp;
1610 afs_uint32 replicated;
1614 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1618 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1619 cm_FreeServerList(serverspp, 0);
1624 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1628 cm_serverRef_t **serverspp;
1629 afs_uint32 replicated;
1630 cm_vol_state_t * volstatep;
1634 volstatep = cm_VolumeStateByID(volp, volid);
1635 replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1636 serverspp = cm_GetVolServers(volp, volid, userp, reqp);
1638 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1639 cm_FreeServerList(serverspp, 0);
1644 extern struct rx_connection *
1645 cm_GetRxConn(cm_conn_t *connp)
1647 struct rx_connection * rxconnp;
1648 lock_ObtainMutex(&connp->mx);
1649 rxconnp = connp->rxconnp;
1650 rx_GetConnection(rxconnp);
1651 lock_ReleaseMutex(&connp->mx);
1655 void cm_ForceNewConnections(cm_server_t *serverp)
1659 lock_ObtainWrite(&cm_connLock);
1660 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1661 lock_ObtainMutex(&tcp->mx);
1662 tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
1663 lock_ReleaseMutex(&tcp->mx);
1665 lock_ReleaseWrite(&cm_connLock);