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 == VNOSERVICE || 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_VNOSERVICE;
615 format = "Server %s reported volume %d in cell %s as not in service.";
618 msgID = MSG_SERVER_REPORTS_VIO;
619 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
623 if (fidp) { /* File Server query */
625 /* Log server being unavailable for this volume */
626 sprintf(addr, "%d.%d.%d.%d",
627 ((serverp->addr.sin_addr.s_addr & 0xff)),
628 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
629 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
630 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
632 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
633 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
636 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
637 CM_GETVOL_FLAG_NO_LRU_UPDATE,
640 statep = cm_VolumeStateByID(volp, fidp->volume);
642 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
643 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
645 code = cm_ForceUpdateVolume(fidp, userp, reqp);
647 location_updated = 1;
649 /* Even if the update fails, there might still be another replica */
651 reqp->flags |= CM_REQ_VOLUME_UPDATED;
652 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
653 fidp->cell, fidp->volume, code);
657 cm_UpdateVolumeStatus(volp, statep->ID);
658 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
659 fidp->cell, fidp->volume, statep->state);
663 lock_ObtainRead(&cm_volumeLock);
665 lock_ReleaseRead(&cm_volumeLock);
670 * Mark server offline for this volume or delete the volume
671 * from the server list if it was moved or is not present.
673 if (!serversp || location_updated) {
674 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
676 serversp = *serverspp;
682 lock_ObtainWrite(&cm_serverLock);
683 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
684 if (tsrp->status == srv_deleted)
687 sprintf(addr, "%d.%d.%d.%d",
688 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
689 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
690 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
691 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
693 if (cm_ServerEqual(tsrp->server, serverp)) {
695 if (errorCode == VMOVED || errorCode == VNOVOL) {
696 osi_Log2(afsd_logp, "volume %d not present on server %s",
697 fidp->volume, osi_LogSaveString(afsd_logp,addr));
698 tsrp->status = srv_deleted;
700 cm_RemoveVolumeFromServer(serverp, fidp->volume);
702 osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
703 fidp->volume, osi_LogSaveString(afsd_logp,addr));
704 tsrp->status = srv_offline;
708 osi_Log3(afsd_logp, "volume %d exists on server %s with status %u",
709 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
712 lock_ReleaseWrite(&cm_serverLock);
714 /* Free the server list before cm_ForceUpdateVolume is called */
716 cm_FreeServerList(serverspp, 0);
723 } else if ( errorCode == VNOVNODE ) {
725 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
726 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
728 scp = cm_FindSCache(fidp);
730 cm_scache_t *pscp = NULL;
732 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
733 pscp = cm_FindSCacheParent(scp);
735 lock_ObtainWrite(&scp->rw);
736 scp->flags |= CM_SCACHEFLAG_DELETED;
737 lock_ObtainWrite(&cm_scacheLock);
738 cm_AdjustScacheLRU(scp);
739 cm_RemoveSCacheFromHashTable(scp);
740 lock_ReleaseWrite(&cm_scacheLock);
741 cm_LockMarkSCacheLost(scp);
742 lock_ReleaseWrite(&scp->rw);
744 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
745 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
746 cm_ReleaseSCache(scp);
749 if (cm_HaveCallback(pscp)) {
750 lock_ObtainWrite(&pscp->rw);
751 cm_DiscardSCache(pscp);
752 lock_ReleaseWrite(&pscp->rw);
755 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
756 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
759 cm_ReleaseSCache(pscp);
763 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
768 else if (errorCode == RX_CALL_TIMEOUT) {
769 /* RPC took longer than hardDeadTime or the server
770 * reported idle for longer than idleDeadTime
771 * don't mark server as down but don't retry
772 * this is to prevent the SMB session from timing out
773 * In addition, we log an event to the event log
777 sprintf(addr, "%d.%d.%d.%d",
778 ((serverp->addr.sin_addr.s_addr & 0xff)),
779 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
780 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
781 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
783 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
784 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
785 osi_LogSaveString(afsd_logp,addr));
786 reqp->tokenIdleErrorServp = serverp;
791 scp = cm_FindSCache(fidp);
793 if (cm_HaveCallback(scp)) {
794 lock_ObtainWrite(&scp->rw);
795 cm_DiscardSCache(scp);
796 lock_ReleaseWrite(&scp->rw);
799 * We really should notify the redirector that we discarded
800 * the status information but doing so in this case is not
801 * safe as it can result in a deadlock with extent release
805 cm_ReleaseSCache(scp);
809 if (!fidp) { /* vldb */
812 cm_volume_t *volp = cm_GetVolumeByFID(fidp);
814 if (fidp->volume == cm_GetROVolumeID(volp))
821 else if (errorCode == RX_MSGSIZE) {
823 * RPC failed because a transmitted rx packet
824 * may have grown larger than the path mtu.
825 * Force a retry and the Rx library will try
826 * with a smaller mtu size.
830 sprintf(addr, "%d.%d.%d.%d",
831 ((serverp->addr.sin_addr.s_addr & 0xff)),
832 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
833 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
834 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
836 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
837 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
838 osi_LogSaveString(afsd_logp,addr));
842 else if (errorCode == RX_CALL_BUSY) {
844 * RPC failed because the selected call channel
845 * is currently busy on the server. Unconditionally
846 * retry the request so an alternate call channel can be used.
849 sprintf(addr, "%d.%d.%d.%d",
850 ((serverp->addr.sin_addr.s_addr & 0xff)),
851 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
852 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
853 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
855 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
856 osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
857 osi_LogSaveString(afsd_logp,addr));
860 else if (errorCode == RX_CALL_IDLE) {
862 * RPC failed because the server failed to respond with data
863 * within the idle dead timeout period. This could be for a variety
865 * 1. The server could have a bad partition such as a failed
866 * disk or iSCSI target and all I/O to that partition is
867 * blocking on the server and will never complete.
869 * 2. The server vnode may be locked by another client request
870 * that is taking a very long time.
872 * 3. The server may have a very long queue of requests
873 * pending and is unable to process this request.
875 * 4. The server could be malicious and is performing a denial
876 * of service attack against the client.
878 * If this is a request against a .readonly with alternate sites
879 * the server should be marked down for this request and the
880 * client should fail over to another server. If this is a
881 * request against a single source, the client may retry once.
884 sprintf(addr, "%d.%d.%d.%d",
885 ((serverp->addr.sin_addr.s_addr & 0xff)),
886 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
887 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
888 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
891 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
892 CM_GETVOL_FLAG_NO_LRU_UPDATE,
895 statep = cm_VolumeStateByID(volp, fidp->volume);
898 replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
900 lock_ObtainRead(&cm_volumeLock);
902 lock_ReleaseRead(&cm_volumeLock);
907 scp = cm_FindSCache(fidp);
909 if (cm_HaveCallback(scp)) {
910 lock_ObtainWrite(&scp->rw);
911 cm_DiscardSCache(scp);
912 lock_ReleaseWrite(&scp->rw);
915 * We really should notify the redirector that we discarded
916 * the status information but doing so in this case is not
917 * safe as it can result in a deadlock with extent release
921 cm_ReleaseSCache(scp);
925 if (replicated && serverp) {
926 reqp->tokenIdleErrorServp = serverp;
927 reqp->tokenError = errorCode;
933 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
934 osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
935 osi_LogSaveString(afsd_logp,addr), retry);
937 else if (errorCode == RX_CALL_DEAD) {
938 /* mark server as down */
940 sprintf(addr, "%d.%d.%d.%d",
941 ((serverp->addr.sin_addr.s_addr & 0xff)),
942 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
943 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
944 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
946 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
947 osi_LogSaveString(afsd_logp,addr),
948 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
951 if ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
952 lock_ObtainMutex(&serverp->mx);
953 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
954 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
955 serverp->downTime = time(NULL);
957 lock_ReleaseMutex(&serverp->mx);
959 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
961 cm_ForceNewConnections(serverp);
966 scp = cm_FindSCache(fidp);
968 if (cm_HaveCallback(scp)) {
969 lock_ObtainWrite(&scp->rw);
970 cm_DiscardSCache(scp);
971 lock_ReleaseWrite(&scp->rw);
974 * We really should notify the redirector that we discarded
975 * the status information but doing so in this case is not
976 * safe as it can result in a deadlock with extent release
980 cm_ReleaseSCache(scp);
986 else if (errorCode >= -64 && errorCode < 0) {
987 /* mark server as down */
989 sprintf(addr, "%d.%d.%d.%d",
990 ((serverp->addr.sin_addr.s_addr & 0xff)),
991 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
992 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
993 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
995 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
997 osi_LogSaveString(afsd_logp,addr),
998 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1001 if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
1002 reqp->tokenIdleErrorServp = serverp;
1003 reqp->tokenError = errorCode;
1005 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1007 cm_ForceNewConnections(serverp);
1013 else if (errorCode == RXKADEXPIRED) {
1014 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1016 if (!dead_session) {
1017 lock_ObtainMutex(&userp->mx);
1018 ucellp = cm_GetUCell(userp, serverp->cellp);
1019 if (ucellp->ticketp) {
1020 free(ucellp->ticketp);
1021 ucellp->ticketp = NULL;
1023 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1025 lock_ReleaseMutex(&userp->mx);
1029 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1030 char * s = "unknown error";
1031 switch ( errorCode ) {
1032 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1033 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1034 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1035 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1036 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1037 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1038 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1039 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1040 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1041 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1042 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1043 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1044 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1046 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1050 reqp->tokenIdleErrorServp = serverp;
1051 reqp->tokenError = errorCode;
1054 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1056 * We received a ubik error. its possible that the server we are
1057 * communicating with has a corrupted database or is partitioned
1058 * from the rest of the servers and another server might be able
1059 * to answer our query. Therefore, we will retry the request
1060 * and force the use of another server.
1063 reqp->tokenIdleErrorServp = serverp;
1064 reqp->tokenError = errorCode;
1067 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1068 cm_ForceNewConnections(serverp);
1073 char * s = "unknown error";
1074 switch ( errorCode ) {
1075 case VSALVAGE : s = "VSALVAGE"; break;
1076 case VNOVNODE : s = "VNOVNODE"; break;
1077 case VNOVOL : s = "VNOVOL"; break;
1078 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1079 case VNOSERVICE : s = "VNOSERVICE"; break;
1080 case VOFFLINE : s = "VOFFLINE"; break;
1081 case VONLINE : s = "VONLINE"; break;
1082 case VDISKFULL : s = "VDISKFULL"; break;
1083 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1084 case VBUSY : s = "VBUSY"; break;
1085 case VMOVED : s = "VMOVED"; break;
1086 case VIO : s = "VIO"; break;
1087 case VRESTRICTED : s = "VRESTRICTED"; break;
1088 case VRESTARTING : s = "VRESTARTING"; break;
1089 case VREADONLY : s = "VREADONLY"; break;
1090 case EAGAIN : s = "EAGAIN"; break;
1091 case UAEAGAIN : s = "UAEAGAIN"; break;
1092 case EINVAL : s = "EINVAL"; break;
1093 case UAEINVAL : s = "UAEINVAL"; break;
1094 case EACCES : s = "EACCES"; break;
1095 case UAEACCES : s = "UAEACCES"; break;
1096 case ENOENT : s = "ENOENT"; break;
1097 case UAENOENT : s = "UAENOENT"; break;
1098 case EEXIST : s = "EEXIST"; break;
1099 case UAEEXIST : s = "UAEEXIST"; break;
1100 case VICECONNBAD : s = "VICECONNBAD"; break;
1101 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1102 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1103 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1104 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1105 case VL_IO : s = "VL_IO"; break;
1106 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1107 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1108 case VL_NOENT : s = "VL_NOENT"; break;
1109 case VL_EMPTY : s = "VL_EMPTY"; break;
1110 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1111 case VL_BADNAME : s = "VL_BADNAME"; break;
1112 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1113 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1114 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1115 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1116 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1117 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1118 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1119 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1120 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1121 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1122 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1123 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1124 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1125 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1126 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1127 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1128 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1129 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1130 case VL_PERM : s = "VL_PERM"; break;
1131 case VL_NOMEM : s = "VL_NOMEM"; break;
1132 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1133 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1134 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1135 case VL_BADMASK : s = "VL_BADMASK"; break;
1136 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1137 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1138 case CM_ERROR_TIMEDOUT : s = "CM_ERROR_TIMEDOUT"; break;
1139 case CM_ERROR_RETRY : s = "CM_ERROR_RETRY"; break;
1140 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1141 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1142 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1143 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1144 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1145 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1146 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1147 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1148 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1149 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1150 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1151 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1152 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1153 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1154 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1155 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1156 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1157 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1158 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1159 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1160 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1161 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1162 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1163 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1164 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1165 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1166 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1167 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1168 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1169 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1170 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1171 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1172 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1173 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1174 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1175 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1176 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1177 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1178 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1179 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1180 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1181 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1182 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1183 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1184 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1185 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1186 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1188 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1194 /* If not allowed to retry, don't */
1195 if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
1196 (errorCode != RX_MSGSIZE && errorCode != RX_CALL_BUSY))
1198 else if (retry && dead_session)
1201 /* drop this on the way out */
1206 * clear the volume updated flag if we succeed.
1207 * this way the flag will not prevent a subsequent volume
1208 * from being updated if necessary.
1212 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1216 errorCode != VBUSY &&
1217 errorCode != VRESTARTING &&
1218 errorCode != CM_ERROR_ALLBUSY)
1220 cm_ResetServerBusyStatus(serversp);
1223 /* retry until we fail to find a connection */
1227 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1228 cm_req_t *reqp, cm_conn_t **connpp)
1231 cm_serverRef_t *tsrp;
1233 long firstError = 0;
1234 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1235 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1236 long timeUsed, timeLeft, hardTimeLeft;
1240 if (serversp == NULL) {
1241 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_ALLDOWN);
1242 return CM_ERROR_ALLDOWN;
1245 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1246 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1248 /* leave 5 seconds margin of safety */
1249 timeLeft = ConnDeadtimeout - timeUsed - 5;
1250 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1253 lock_ObtainRead(&cm_serverLock);
1254 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1255 if (tsrp->status == srv_deleted)
1259 if (reqp->tokenIdleErrorServp) {
1261 * search the list until we find the server
1262 * that failed last time. When we find it
1263 * clear the error, skip it and try the next one
1266 if (tsp == reqp->tokenIdleErrorServp)
1267 reqp->tokenIdleErrorServp = NULL;
1271 cm_GetServerNoLock(tsp);
1272 lock_ReleaseRead(&cm_serverLock);
1273 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1275 if (tsrp->status == srv_busy) {
1278 } else if (tsrp->status == srv_offline) {
1284 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1285 if (code == 0) { /* cm_CBS only returns 0 */
1287 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1288 /* Set RPC timeout */
1289 if (timeLeft > ConnDeadtimeout)
1290 timeLeft = ConnDeadtimeout;
1292 if (hardTimeLeft > HardDeadtimeout)
1293 hardTimeLeft = HardDeadtimeout;
1295 lock_ObtainMutex(&(*connpp)->mx);
1296 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1297 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1298 lock_ReleaseMutex(&(*connpp)->mx);
1303 /* therefore, this code is never executed */
1304 if (firstError == 0)
1308 lock_ObtainRead(&cm_serverLock);
1309 cm_PutServerNoLock(tsp);
1312 lock_ReleaseRead(&cm_serverLock);
1314 if (firstError == 0) {
1316 firstError = (reqp->tokenError ? reqp->tokenError :
1317 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1319 * if we experienced either a token error or and idle dead time error
1320 * and now all of the servers are down, we have either tried them
1321 * all or lost connectivity. Clear the error we are returning so
1322 * we will not return it indefinitely if the request is retried.
1324 reqp->idleError = reqp->tokenError = 0;
1325 } else if (allBusy) {
1326 firstError = CM_ERROR_ALLBUSY;
1327 } else if (allOffline || (someBusy && someOffline)) {
1328 firstError = CM_ERROR_ALLOFFLINE;
1330 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1331 firstError = CM_ERROR_TIMEDOUT;
1335 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1339 /* called with a held server to GC all bad connections hanging off of the server */
1340 void cm_GCConnections(cm_server_t *serverp)
1346 lock_ObtainWrite(&cm_connLock);
1347 lcpp = &serverp->connsp;
1348 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1350 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1351 /* do the deletion of this guy */
1352 cm_PutServer(tcp->serverp);
1353 cm_ReleaseUser(userp);
1355 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1356 rx_DestroyConnection(tcp->rxconnp);
1357 lock_FinalizeMutex(&tcp->mx);
1361 /* just advance to the next */
1365 lock_ReleaseWrite(&cm_connLock);
1368 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1369 cm_server_t *serverp, afs_uint32 replicated)
1371 unsigned short port;
1374 struct rx_securityClass *secObjp;
1376 port = serverp->addr.sin_port;
1377 switch (serverp->type) {
1378 case CM_SERVER_VLDB:
1383 case CM_SERVER_FILE:
1389 osi_panic("unknown server type", __FILE__, __LINE__);
1392 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1396 tcp->cryptlevel = rxkad_clear;
1399 tcp->cryptlevel = rxkad_auth;
1402 tcp->cryptlevel = rxkad_crypt;
1404 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1405 &ucellp->sessionKey, ucellp->kvno,
1406 ucellp->ticketLen, ucellp->ticketp);
1410 tcp->cryptlevel = rxkad_clear;
1411 secObjp = rxnull_NewClientSecurityObject();
1413 osi_assertx(secObjp != NULL, "null rx_securityClass");
1414 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1419 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1420 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1423 * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
1426 tcp->flags &= CM_CONN_FLAG_REPLICATION;
1427 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1429 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1433 * Let the Rx library know that we can auto-retry if an
1434 * RX_MSGSIZE error is returned.
1436 if (rx_pmtu_discovery)
1437 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1440 * Attempt to limit NAT pings to the anonymous file server connections.
1441 * Only file servers implement client callbacks and we only need one ping
1442 * to be sent to each server.
1444 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1445 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1446 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1449 tcp->ucgen = ucellp->gen;
1451 rxs_Release(secObjp); /* Decrement the initial refCount */
1454 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1461 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1462 userp = cm_rootUserp;
1464 lock_ObtainMutex(&userp->mx);
1465 lock_ObtainRead(&cm_connLock);
1466 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1467 if (tcp->userp == userp &&
1468 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1469 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1473 /* find ucell structure */
1474 ucellp = cm_GetUCell(userp, serverp->cellp);
1476 lock_ConvertRToW(&cm_connLock);
1477 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1478 if (tcp->userp == userp)
1482 lock_ReleaseWrite(&cm_connLock);
1485 cm_GetServer(serverp);
1486 tcp = malloc(sizeof(*tcp));
1487 memset(tcp, 0, sizeof(*tcp));
1488 tcp->nextp = serverp->connsp;
1489 serverp->connsp = tcp;
1492 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1493 lock_ObtainMutex(&tcp->mx);
1494 tcp->serverp = serverp;
1495 tcp->cryptlevel = rxkad_clear;
1496 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1498 lock_ReleaseMutex(&tcp->mx);
1499 lock_ReleaseWrite(&cm_connLock);
1501 lock_ReleaseRead(&cm_connLock);
1503 InterlockedIncrement(&tcp->refCount);
1505 lock_ObtainMutex(&tcp->mx);
1506 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1507 (tcp->ucgen < ucellp->gen) ||
1508 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1510 if (tcp->ucgen < ucellp->gen)
1511 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1513 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1514 tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
1515 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1516 rx_DestroyConnection(tcp->rxconnp);
1517 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1519 lock_ReleaseMutex(&tcp->mx);
1521 lock_ReleaseMutex(&userp->mx);
1523 /* return this pointer to our caller */
1524 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1530 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1534 cm_serverRef_t **serverspp;
1535 cm_serverRef_t *tsrp;
1537 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1538 afs_uint32 replicated;
1542 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1546 lock_ObtainRead(&cm_serverLock);
1547 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1548 if (tsrp->status == srv_deleted)
1551 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1553 if (tsrp->status == srv_busy) {
1556 } else if (tsrp->status == srv_offline) {
1565 lock_ReleaseRead(&cm_serverLock);
1566 cm_FreeServerList(serverspp, 0);
1572 else if (allOffline || (someBusy && someOffline))
1579 * The returned cm_conn_t ** object is released in the subsequent call
1582 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1586 cm_serverRef_t **serverspp;
1587 afs_uint32 replicated;
1591 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1595 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1596 cm_FreeServerList(serverspp, 0);
1601 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1605 cm_serverRef_t **serverspp;
1606 afs_uint32 replicated;
1607 cm_vol_state_t * volstatep;
1611 volstatep = cm_VolumeStateByID(volp, volid);
1612 replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1613 serverspp = cm_GetVolServers(volp, volid, userp, reqp);
1615 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1616 cm_FreeServerList(serverspp, 0);
1621 extern struct rx_connection *
1622 cm_GetRxConn(cm_conn_t *connp)
1624 struct rx_connection * rxconnp;
1625 lock_ObtainMutex(&connp->mx);
1626 rxconnp = connp->rxconnp;
1627 rx_GetConnection(rxconnp);
1628 lock_ReleaseMutex(&connp->mx);
1632 void cm_ForceNewConnections(cm_server_t *serverp)
1636 lock_ObtainWrite(&cm_connLock);
1637 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1638 lock_ObtainMutex(&tcp->mx);
1639 tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
1640 lock_ReleaseMutex(&tcp->mx);
1642 lock_ReleaseWrite(&cm_connLock);