2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
10 #include <afsconfig.h>
11 #include <afs/param.h>
23 #include <afs/unified_afs.h>
24 #include <afs/vlserver.h>
25 #include <WINNT/afsreg.h>
27 osi_rwlock_t cm_connLock;
29 DWORD RDRtimeout = CM_CONN_DEFAULTRDRTIMEOUT;
30 unsigned short ConnDeadtimeout = CM_CONN_CONNDEADTIME;
31 unsigned short HardDeadtimeout = CM_CONN_HARDDEADTIME;
32 unsigned short IdleDeadtimeout = CM_CONN_IDLEDEADTIME;
33 unsigned short ReplicaIdleDeadtimeout = CM_CONN_IDLEDEADTIME_REP;
34 unsigned short NatPingInterval = CM_CONN_NATPINGINTERVAL;
36 #define LANMAN_WKS_PARAM_KEY "SYSTEM\\CurrentControlSet\\Services\\lanmanworkstation\\parameters"
37 #define LANMAN_WKS_SESSION_TIMEOUT "SessTimeout"
38 #define LANMAN_WKS_EXT_SESSION_TIMEOUT "ExtendedSessTimeout"
40 afs_uint32 cryptall = 0;
41 afs_uint32 cm_anonvldb = 0;
42 afs_uint32 rx_pmtu_discovery = 0;
44 void cm_PutConn(cm_conn_t *connp)
46 afs_int32 refCount = InterlockedDecrement(&connp->refCount);
47 osi_assertx(refCount >= 0, "cm_conn_t refcount underflow");
50 void cm_InitConn(void)
52 static osi_once_t once;
58 if (osi_Once(&once)) {
59 lock_InitializeRWLock(&cm_connLock, "connection global lock",
60 LOCK_HIERARCHY_CONN_GLOBAL);
62 /* keisa - read timeout value for lanmanworkstation service.
64 * http://support.microsoft.com:80/support/kb/articles/Q102/0/67.asp&NoWebContent=1
65 * the SessTimeout is a minimum timeout not a maximum timeout. Therefore,
66 * I believe that the default should not be short. Instead, we should wait until
67 * RX times out before reporting a timeout to the SMB client.
69 code = RegOpenKeyEx(HKEY_LOCAL_MACHINE, LANMAN_WKS_PARAM_KEY,
70 0, KEY_QUERY_VALUE, &parmKey);
71 if (code == ERROR_SUCCESS)
73 BOOL extTimeouts = msftSMBRedirectorSupportsExtendedTimeouts();
77 * The default value is 1000 seconds. However, this default
78 * will not apply to "\\AFS" unless "AFS" is listed in
79 * ServersWithExtendedSessTimeout which we will add when we
80 * create the ExtendedSessTimeout value in smb_Init()
82 dummyLen = sizeof(DWORD);
83 code = RegQueryValueEx(parmKey,
84 LANMAN_WKS_EXT_SESSION_TIMEOUT,
86 (BYTE *) &dwValue, &dummyLen);
87 if (code == ERROR_SUCCESS) {
89 afsi_log("lanmanworkstation : ExtSessTimeout %u", RDRtimeout);
92 if (!extTimeouts || code != ERROR_SUCCESS) {
93 dummyLen = sizeof(DWORD);
94 code = RegQueryValueEx(parmKey,
95 LANMAN_WKS_SESSION_TIMEOUT,
97 (BYTE *) &dwValue, &dummyLen);
98 if (code == ERROR_SUCCESS) {
100 afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
103 RegCloseKey(parmKey);
106 code = RegOpenKeyEx(HKEY_LOCAL_MACHINE, AFSREG_CLT_SVC_PARAM_SUBKEY,
107 0, KEY_QUERY_VALUE, &parmKey);
108 if (code == ERROR_SUCCESS) {
109 dummyLen = sizeof(DWORD);
110 code = RegQueryValueEx(parmKey, "ConnDeadTimeout", NULL, NULL,
111 (BYTE *) &dwValue, &dummyLen);
112 if (code == ERROR_SUCCESS) {
113 ConnDeadtimeout = (unsigned short)dwValue;
114 afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
116 dummyLen = sizeof(DWORD);
117 code = RegQueryValueEx(parmKey, "HardDeadTimeout", NULL, NULL,
118 (BYTE *) &dwValue, &dummyLen);
119 if (code == ERROR_SUCCESS) {
120 HardDeadtimeout = (unsigned short)dwValue;
121 afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
123 dummyLen = sizeof(DWORD);
124 code = RegQueryValueEx(parmKey, "IdleDeadTimeout", NULL, NULL,
125 (BYTE *) &dwValue, &dummyLen);
126 if (code == ERROR_SUCCESS) {
127 IdleDeadtimeout = (unsigned short)dwValue;
128 afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
130 dummyLen = sizeof(DWORD);
131 code = RegQueryValueEx(parmKey, "ReplicaIdleDeadTimeout", NULL, NULL,
132 (BYTE *) &dwValue, &dummyLen);
133 if (code == ERROR_SUCCESS) {
134 ReplicaIdleDeadtimeout = (unsigned short)dwValue;
135 afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
137 dummyLen = sizeof(DWORD);
138 code = RegQueryValueEx(parmKey, "NatPingInterval", NULL, NULL,
139 (BYTE *) &dwValue, &dummyLen);
140 if (code == ERROR_SUCCESS) {
141 NatPingInterval = (unsigned short)dwValue;
143 afsi_log("NatPingInterval is %d", NatPingInterval);
144 RegCloseKey(parmKey);
148 * If these values were not set via cpp macro or obtained
149 * from the registry, we use a value that is derived from
150 * the smb redirector session timeout (RDRtimeout).
152 * The UNIX cache manager uses 120 seconds for the hard dead
153 * timeout and 50 seconds for the connection and idle timeouts.
155 * We base our values on those while making sure we leave
156 * enough time for overhead.
158 * To further complicate matters we need to take into account
159 * file server hard dead timeouts as they affect the length
160 * of time it takes the file server to give up when attempting
161 * to break callbacks to unresponsive clients. The file
162 * server hard dead timeout is 120 seconds.
164 * For SMB, we have no choice but to timeout quickly. For
165 * the AFS redirector, we can wait.
168 afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
169 if (ConnDeadtimeout == 0) {
170 ConnDeadtimeout = (unsigned short) ((RDRtimeout / 2) < 50 ? (RDRtimeout / 2) : 50);
171 afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
173 if (HardDeadtimeout == 0) {
174 HardDeadtimeout = (unsigned short) (RDRtimeout > 125 ? 120 : (RDRtimeout - 5));
175 afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
177 if (IdleDeadtimeout == 0) {
178 IdleDeadtimeout = 10 * (unsigned short) HardDeadtimeout;
179 afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
181 if (ReplicaIdleDeadtimeout == 0) {
182 ReplicaIdleDeadtimeout = (unsigned short) HardDeadtimeout;
183 afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
186 if (ConnDeadtimeout == 0) {
187 ConnDeadtimeout = CM_CONN_IFS_CONNDEADTIME;
188 afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
190 if (HardDeadtimeout == 0) {
191 HardDeadtimeout = CM_CONN_IFS_HARDDEADTIME;
192 afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
194 if (IdleDeadtimeout == 0) {
195 IdleDeadtimeout = CM_CONN_IFS_IDLEDEADTIME;
196 afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
198 if (IdleDeadtimeout == 0) {
199 ReplicaIdleDeadtimeout = CM_CONN_IFS_IDLEDEADTIME_REP;
200 afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
207 void cm_InitReq(cm_req_t *reqp)
209 memset(reqp, 0, sizeof(cm_req_t));
210 reqp->startTime = GetTickCount();
213 long cm_GetVolServerList(cm_volume_t *volp, afs_uint32 volid, struct cm_user *userp,
214 struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
216 *serversppp = cm_GetVolServers(volp, volid, userp, reqp, replicated);
217 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
220 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
221 struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
224 cm_volume_t *volp = NULL;
225 cm_cell_t *cellp = NULL;
229 return CM_ERROR_INVAL;
232 cellp = cm_FindCellByID(fidp->cell, 0);
234 return CM_ERROR_NOSUCHCELL;
236 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
240 *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp, replicated);
242 lock_ObtainRead(&cm_volumeLock);
244 lock_ReleaseRead(&cm_volumeLock);
245 return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
249 cm_SetServerBusyStatus(cm_serverRef_t *serversp, cm_server_t *serverp)
251 cm_serverRef_t *tsrp;
253 lock_ObtainWrite(&cm_serverLock);
254 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
255 if (tsrp->status == srv_deleted)
257 if (cm_ServerEqual(tsrp->server, serverp) && tsrp->status == srv_not_busy) {
258 tsrp->status = srv_busy;
262 lock_ReleaseWrite(&cm_serverLock);
266 cm_ResetServerBusyStatus(cm_serverRef_t *serversp)
268 cm_serverRef_t *tsrp;
270 lock_ObtainWrite(&cm_serverLock);
271 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
272 if (tsrp->status == srv_deleted)
274 if (tsrp->status == srv_busy) {
275 tsrp->status = srv_not_busy;
278 lock_ReleaseWrite(&cm_serverLock);
282 * Analyze the error return from an RPC. Determine whether or not to retry,
283 * and if we're going to retry, determine whether failover is appropriate,
284 * and whether timed backoff is appropriate.
286 * If the error code is from cm_ConnFromFID() or friends, it will be a CM_ERROR code.
287 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
288 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
289 * be a security code (e.g. RXKADEXPIRED).
291 * If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
293 * For VLDB calls, fidp will be NULL and cellp will not be.
295 * volSyncp and/or cbrp may also be NULL.
298 cm_Analyze(cm_conn_t *connp,
304 AFSVolSync *volSyncp,
305 cm_serverRef_t * serversp,
306 cm_callbackRequest_t *cbrp,
309 cm_server_t *serverp = NULL;
310 cm_serverRef_t **serverspp = NULL;
311 cm_serverRef_t *tsrp;
313 cm_volume_t * volp = NULL;
314 cm_vol_state_t *statep = NULL;
315 cm_scache_t * scp = NULL;
316 afs_uint32 replicated;
318 int free_svr_list = 0;
320 long timeUsed, timeLeft;
322 char addr[16]="unknown";
324 int location_updated = 0;
328 osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
331 /* no locking required, since connp->serverp never changes after
333 dead_session = (userp->cellInfop == NULL);
335 serverp = connp->serverp;
337 /* Update callback pointer */
338 if (cbrp && serverp && errorCode == 0) {
340 if ( cbrp->serverp != serverp ) {
341 lock_ObtainWrite(&cm_serverLock);
342 cm_PutServerNoLock(cbrp->serverp);
343 cm_GetServerNoLock(serverp);
344 lock_ReleaseWrite(&cm_serverLock);
347 cm_GetServer(serverp);
349 cbrp->serverp = serverp;
352 /* if timeout - check that it did not exceed the HardDead timeout
355 /* timeleft - get it from reqp the same way as cm_ConnByMServers does */
356 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
357 if ( reqp->flags & CM_REQ_SOURCE_SMB )
358 timeLeft = HardDeadtimeout - timeUsed;
360 timeLeft = 0x0FFFFFFF;
362 /* get a pointer to the cell */
364 if (cellp == NULL && serverp)
365 cellp = serverp->cellp;
366 if (cellp == NULL && serversp) {
367 struct cm_serverRef * refp;
368 for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
369 if (refp->status == srv_deleted)
372 cellp = refp->server->cellp;
375 if (cellp == NULL && fidp) {
376 cellp = cm_FindCellByID(fidp->cell, 0);
380 if (errorCode == CM_ERROR_TIMEDOUT) {
381 if ( timeLeft > 5 ) {
383 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
388 else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
389 errorCode == UAEAGAIN || errorCode == EAGAIN) {
390 osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
391 if ( timeLeft > 5 ) {
397 /* if there is nosuchvolume, then we have a situation in which a
398 * previously known volume no longer has a set of servers
399 * associated with it. Either the volume has moved or
400 * the volume has been deleted. Try to find a new server list
401 * until the timeout period expires.
403 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
404 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
406 * The VNOVOL or VL_NOENT error has already been translated
407 * to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
411 else if (errorCode == CM_ERROR_EMPTY) {
413 * The server list is empty (or all entries have been deleted).
414 * If fidp is NULL, this was a vlServer list and we can attempt
415 * to force a cell lookup. If fidp is not NULL, we can attempt
416 * to refresh the volume location list.
419 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
420 CM_GETVOL_FLAG_NO_LRU_UPDATE,
423 lock_ObtainWrite(&volp->rw);
424 if (cm_UpdateVolumeLocation(cellp, userp, reqp, volp) == 0) {
425 lock_ReleaseWrite(&volp->rw);
426 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
428 if (!cm_IsServerListEmpty(*serverspp))
430 cm_FreeServerList(serverspp, 0);
433 lock_ReleaseWrite(&volp->rw);
435 lock_ObtainRead(&cm_volumeLock);
437 lock_ReleaseRead(&cm_volumeLock);
441 cm_cell_t * newCellp = cm_UpdateCell( cellp, 0);
446 else if (errorCode == CM_ERROR_ALLDOWN) {
447 /* Servers marked DOWN will be restored by the background daemon
448 * thread as they become available. The volume status is
449 * updated as the server state changes.
452 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
453 cellp->name, fidp->volume);
454 msgID = MSG_ALL_SERVERS_DOWN;
455 format = "All servers are unreachable when accessing cell %s volume %d.";
456 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
458 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
461 else if (errorCode == CM_ERROR_ALLOFFLINE) {
462 /* Volume instances marked offline will be restored by the
463 * background daemon thread as they become available
466 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
467 cellp->name, fidp->volume);
468 msgID = MSG_ALL_SERVERS_OFFLINE;
469 format = "All servers are offline when accessing cell %s volume %d.";
470 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
472 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
473 CM_GETVOL_FLAG_NO_LRU_UPDATE,
477 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
479 serversp = *serverspp;
483 cm_ResetServerBusyStatus(serversp);
485 cm_FreeServerList(serverspp, 0);
491 * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
492 * was called by cm_CheckOfflineVolumeState().
494 if (!(reqp->flags & CM_REQ_OFFLINE_VOL_CHK) && timeLeft > 7) {
497 /* cm_CheckOfflineVolume() resets the serverRef state */
498 if (cm_CheckOfflineVolume(volp, fidp->volume))
501 cm_UpdateVolumeStatus(volp, fidp->volume);
503 lock_ObtainRead(&cm_volumeLock);
505 lock_ReleaseRead(&cm_volumeLock);
509 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
512 else if (errorCode == CM_ERROR_ALLBUSY) {
513 /* Volumes that are busy cannot be determined to be non-busy
514 * without actually attempting to access them.
516 if (fidp) { /* File Server query */
517 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
518 cellp->name, fidp->volume);
519 msgID = MSG_ALL_SERVERS_BUSY;
520 format = "All servers are busy when accessing cell %s volume %d.";
521 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
523 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
524 CM_GETVOL_FLAG_NO_LRU_UPDATE,
528 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
530 serversp = *serverspp;
534 cm_ResetServerBusyStatus(serversp);
536 cm_FreeServerList(serverspp, 0);
543 statep = cm_VolumeStateByID(volp, fidp->volume);
546 cm_UpdateVolumeStatus(volp, fidp->volume);
548 lock_ObtainRead(&cm_volumeLock);
550 lock_ReleaseRead(&cm_volumeLock);
553 } else { /* VL Server query */
554 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
560 cm_ResetServerBusyStatus(serversp);
567 /* special codes: VBUSY and VRESTARTING */
568 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
570 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
571 CM_GETVOL_FLAG_NO_LRU_UPDATE,
575 code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
577 serversp = *serverspp;
582 statep = cm_VolumeStateByID(volp, fidp->volume);
585 cm_UpdateVolumeStatus(volp, statep->ID);
587 lock_ObtainRead(&cm_volumeLock);
589 lock_ReleaseRead(&cm_volumeLock);
595 /* Log server being offline for this volume */
596 sprintf(addr, "%d.%d.%d.%d",
597 ((serverp->addr.sin_addr.s_addr & 0xff)),
598 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
599 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
600 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
602 switch ( errorCode ) {
604 msgID = MSG_SERVER_REPORTS_VBUSY;
605 format = "Server %s reported busy when accessing volume %d in cell %s.";
608 msgID = MSG_SERVER_REPORTS_VRESTARTING;
609 format = "Server %s reported restarting when accessing volume %d in cell %s.";
613 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
614 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
616 cm_SetServerBusyStatus(serversp, serverp);
620 cm_FreeServerList(serverspp, 0);
627 /* special codes: missing volumes */
628 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
629 errorCode == VSALVAGE || errorCode == VIO)
631 /* In case of timeout */
632 reqp->volumeError = errorCode;
634 switch ( errorCode ) {
636 msgID = MSG_SERVER_REPORTS_VNOVOL;
637 format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
640 msgID = MSG_SERVER_REPORTS_VMOVED;
641 format = "Server %s reported volume %d in cell %s as moved.";
644 msgID = MSG_SERVER_REPORTS_VOFFLINE;
645 format = "Server %s reported volume %d in cell %s as offline.";
648 msgID = MSG_SERVER_REPORTS_VSALVAGE;
649 format = "Server %s reported volume %d in cell %s as needs salvage.";
652 msgID = MSG_SERVER_REPORTS_VIO;
653 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
657 if (fidp) { /* File Server query */
659 /* Log server being unavailable for this volume */
660 sprintf(addr, "%d.%d.%d.%d",
661 ((serverp->addr.sin_addr.s_addr & 0xff)),
662 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
663 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
664 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
666 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
667 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
670 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
671 CM_GETVOL_FLAG_NO_LRU_UPDATE,
674 statep = cm_VolumeStateByID(volp, fidp->volume);
676 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
677 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
679 LONG_PTR oldSum, newSum;
681 oldSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
683 code = cm_ForceUpdateVolume(fidp, userp, reqp);
685 location_updated = 1;
686 newSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
690 * Even if the update fails, there might still be another replica.
691 * If the volume location list changed, permit another update on
692 * a subsequent error.
694 if (code || oldSum == newSum)
695 reqp->flags |= CM_REQ_VOLUME_UPDATED;
697 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
698 fidp->cell, fidp->volume, code);
702 cm_UpdateVolumeStatus(volp, statep->ID);
703 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
704 fidp->cell, fidp->volume, statep->state);
708 lock_ObtainRead(&cm_volumeLock);
710 lock_ReleaseRead(&cm_volumeLock);
715 * Mark server offline for this volume or delete the volume
716 * from the server list if it was moved or is not present.
718 if (!serversp || location_updated) {
719 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
721 serversp = *serverspp;
727 lock_ObtainWrite(&cm_serverLock);
728 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
729 if (tsrp->status == srv_deleted)
732 sprintf(addr, "%d.%d.%d.%d",
733 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
734 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
735 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
736 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
738 if (cm_ServerEqual(tsrp->server, serverp)) {
742 osi_Log2(afsd_logp, "volume %u moved from server %s",
743 fidp->volume, osi_LogSaveString(afsd_logp,addr));
744 tsrp->status = srv_deleted;
746 cm_RemoveVolumeFromServer(serverp, fidp->volume);
750 * The 1.6.0 and 1.6.1 file servers send transient VNOVOL errors which
751 * are no indicative of the volume not being present. For example,
752 * VNOVOL can be sent during a transition to a VBUSY state prior to
753 * salvaging or when cloning a .backup volume instance. As a result
754 * the cache manager must attempt at least one retry when a VNOVOL is
755 * receive but there are no changes to the volume location information.
757 if (reqp->vnovolError > 0 && cm_ServerEqual(reqp->errorServp, serverp)) {
758 osi_Log2(afsd_logp, "volume %u not present on server %s",
759 fidp->volume, osi_LogSaveString(afsd_logp,addr));
760 tsrp->status = srv_deleted;
762 cm_RemoveVolumeFromServer(serverp, fidp->volume);
764 osi_Log2(afsd_logp, "VNOVOL received for volume %u from server %s",
765 fidp->volume, osi_LogSaveString(afsd_logp,addr));
767 if (tsrp->status == srv_not_busy)
768 tsrp->status = srv_busy;
775 osi_Log2(afsd_logp, "VOFFLINE received for volume %u from server %s",
776 fidp->volume, osi_LogSaveString(afsd_logp,addr));
777 tsrp->status = srv_offline;
780 osi_Log3(afsd_logp, "volume %u exists on server %s with status %u",
781 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
785 lock_ReleaseWrite(&cm_serverLock);
787 /* Remember that the VNOVOL error occurred */
788 if (errorCode == VNOVOL) {
789 reqp->errorServp = serverp;
793 /* Free the server list before cm_ForceUpdateVolume is called */
795 cm_FreeServerList(serverspp, 0);
802 } else if ( errorCode == VNOVNODE ) {
804 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
805 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
807 scp = cm_FindSCache(fidp);
809 cm_scache_t *pscp = NULL;
811 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
812 pscp = cm_FindSCacheParent(scp);
814 lock_ObtainWrite(&scp->rw);
815 scp->flags |= CM_SCACHEFLAG_DELETED;
816 lock_ObtainWrite(&cm_scacheLock);
817 cm_AdjustScacheLRU(scp);
818 cm_RemoveSCacheFromHashTable(scp);
819 lock_ReleaseWrite(&cm_scacheLock);
820 cm_LockMarkSCacheLost(scp);
821 lock_ReleaseWrite(&scp->rw);
823 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
824 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
825 cm_ReleaseSCache(scp);
828 if (cm_HaveCallback(pscp)) {
829 lock_ObtainWrite(&pscp->rw);
830 cm_DiscardSCache(pscp);
831 lock_ReleaseWrite(&pscp->rw);
834 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
835 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
838 cm_ReleaseSCache(pscp);
842 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
847 else if (errorCode == RX_CALL_TIMEOUT) {
848 /* RPC took longer than hardDeadTime or the server
849 * reported idle for longer than idleDeadTime
850 * don't mark server as down but don't retry
851 * this is to prevent the SMB session from timing out
852 * In addition, we log an event to the event log
856 sprintf(addr, "%d.%d.%d.%d",
857 ((serverp->addr.sin_addr.s_addr & 0xff)),
858 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
859 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
860 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
862 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
863 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
864 osi_LogSaveString(afsd_logp,addr));
865 reqp->errorServp = serverp;
870 scp = cm_FindSCache(fidp);
872 if (cm_HaveCallback(scp)) {
873 lock_ObtainWrite(&scp->rw);
874 cm_DiscardSCache(scp);
875 lock_ReleaseWrite(&scp->rw);
878 * We really should notify the redirector that we discarded
879 * the status information but doing so in this case is not
880 * safe as it can result in a deadlock with extent release
884 cm_ReleaseSCache(scp);
888 if (!fidp) { /* vldb */
891 cm_volume_t *volp = cm_GetVolumeByFID(fidp);
893 if (fidp->volume == cm_GetROVolumeID(volp))
900 else if (errorCode == RX_MSGSIZE) {
902 * RPC failed because a transmitted rx packet
903 * may have grown larger than the path mtu.
904 * Force a retry and the Rx library will try
905 * with a smaller mtu size.
909 sprintf(addr, "%d.%d.%d.%d",
910 ((serverp->addr.sin_addr.s_addr & 0xff)),
911 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
912 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
913 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
915 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
916 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
917 osi_LogSaveString(afsd_logp,addr));
921 else if (errorCode == RX_CALL_BUSY) {
923 * RPC failed because the selected call channel
924 * is currently busy on the server. Unconditionally
925 * retry the request so an alternate call channel can be used.
928 sprintf(addr, "%d.%d.%d.%d",
929 ((serverp->addr.sin_addr.s_addr & 0xff)),
930 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
931 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
932 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
934 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
935 osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
936 osi_LogSaveString(afsd_logp,addr));
939 else if (errorCode == VNOSERVICE) {
941 * The server did not service the RPC.
942 * If this was a file server RPC it means that for at
943 * least the file server's idle dead timeout period the
944 * file server did not receive any new data packets from
947 * The RPC was not serviced so it can be retried and any
948 * existing status information is still valid.
952 sprintf(addr, "%d.%d.%d.%d",
953 ((serverp->addr.sin_addr.s_addr & 0xff)),
954 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
955 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
956 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
958 LogEvent(EVENTLOG_WARNING_TYPE, MSG_SERVER_REPORTS_VNOSERVICE,
959 addr, fidp->volume, cellp->name);
960 osi_Log3(afsd_logp, "Server %s reported rpc to volume %d in cell %s as not serviced.",
961 osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
967 else if (errorCode == RX_CALL_IDLE) {
969 * RPC failed because the server failed to respond with data
970 * within the idle dead timeout period. This could be for a variety
972 * 1. The server could have a bad partition such as a failed
973 * disk or iSCSI target and all I/O to that partition is
974 * blocking on the server and will never complete.
976 * 2. The server vnode may be locked by another client request
977 * that is taking a very long time.
979 * 3. The server may have a very long queue of requests
980 * pending and is unable to process this request.
982 * 4. The server could be malicious and is performing a denial
983 * of service attack against the client.
985 * If this is a request against a .readonly with alternate sites
986 * the server should be marked down for this request and the
987 * client should fail over to another server. If this is a
988 * request against a single source, the client may retry once.
991 sprintf(addr, "%d.%d.%d.%d",
992 ((serverp->addr.sin_addr.s_addr & 0xff)),
993 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
994 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
995 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
998 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
999 CM_GETVOL_FLAG_NO_LRU_UPDATE,
1002 statep = cm_VolumeStateByID(volp, fidp->volume);
1005 replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1007 lock_ObtainRead(&cm_volumeLock);
1009 lock_ReleaseRead(&cm_volumeLock);
1014 scp = cm_FindSCache(fidp);
1016 if (cm_HaveCallback(scp)) {
1017 lock_ObtainWrite(&scp->rw);
1018 cm_DiscardSCache(scp);
1019 lock_ReleaseWrite(&scp->rw);
1022 * We really should notify the redirector that we discarded
1023 * the status information but doing so in this case is not
1024 * safe as it can result in a deadlock with extent release
1028 cm_ReleaseSCache(scp);
1032 if (replicated && serverp) {
1033 reqp->errorServp = serverp;
1034 reqp->tokenError = errorCode;
1040 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
1041 osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
1042 osi_LogSaveString(afsd_logp,addr), retry);
1044 else if (errorCode == RX_CALL_DEAD) {
1045 /* mark server as down */
1047 sprintf(addr, "%d.%d.%d.%d",
1048 ((serverp->addr.sin_addr.s_addr & 0xff)),
1049 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1050 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1051 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1053 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
1054 osi_LogSaveString(afsd_logp,addr),
1055 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1058 if ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
1059 lock_ObtainMutex(&serverp->mx);
1060 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
1061 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
1062 serverp->downTime = time(NULL);
1064 lock_ReleaseMutex(&serverp->mx);
1066 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1068 cm_ForceNewConnections(serverp);
1072 if (fidp && storeOp)
1073 scp = cm_FindSCache(fidp);
1075 if (cm_HaveCallback(scp)) {
1076 lock_ObtainWrite(&scp->rw);
1077 cm_DiscardSCache(scp);
1078 lock_ReleaseWrite(&scp->rw);
1081 * We really should notify the redirector that we discarded
1082 * the status information but doing so in this case is not
1083 * safe as it can result in a deadlock with extent release
1087 cm_ReleaseSCache(scp);
1093 else if (errorCode >= -64 && errorCode < 0) {
1094 /* mark server as down */
1096 sprintf(addr, "%d.%d.%d.%d",
1097 ((serverp->addr.sin_addr.s_addr & 0xff)),
1098 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
1099 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
1100 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
1102 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
1104 osi_LogSaveString(afsd_logp,addr),
1105 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
1108 if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
1109 reqp->errorServp = serverp;
1110 reqp->tokenError = errorCode;
1112 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
1114 cm_ForceNewConnections(serverp);
1120 else if (errorCode == RXKADEXPIRED) {
1121 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1123 if (!dead_session) {
1124 lock_ObtainMutex(&userp->mx);
1125 ucellp = cm_GetUCell(userp, cellp);
1126 if (ucellp->ticketp) {
1127 free(ucellp->ticketp);
1128 ucellp->ticketp = NULL;
1130 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1132 lock_ReleaseMutex(&userp->mx);
1136 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1137 char * s = "unknown error";
1138 switch ( errorCode ) {
1139 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1140 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1141 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1142 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1143 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1144 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1145 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1146 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1147 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1148 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1149 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1150 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1151 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1153 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1157 reqp->errorServp = serverp;
1158 reqp->tokenError = errorCode;
1161 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1163 * We received a ubik error. its possible that the server we are
1164 * communicating with has a corrupted database or is partitioned
1165 * from the rest of the servers and another server might be able
1166 * to answer our query. Therefore, we will retry the request
1167 * and force the use of another server.
1170 reqp->errorServp = serverp;
1171 reqp->tokenError = errorCode;
1174 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1175 cm_ForceNewConnections(serverp);
1180 char * s = "unknown error";
1181 switch ( errorCode ) {
1182 case VSALVAGE : s = "VSALVAGE"; break;
1183 case VNOVNODE : s = "VNOVNODE"; break;
1184 case VNOVOL : s = "VNOVOL"; break;
1185 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1186 case VNOSERVICE : s = "VNOSERVICE"; break;
1187 case VOFFLINE : s = "VOFFLINE"; break;
1188 case VONLINE : s = "VONLINE"; break;
1189 case VDISKFULL : s = "VDISKFULL"; break;
1190 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1191 case VBUSY : s = "VBUSY"; break;
1192 case VMOVED : s = "VMOVED"; break;
1193 case VIO : s = "VIO"; break;
1194 case VRESTRICTED : s = "VRESTRICTED"; break;
1195 case VRESTARTING : s = "VRESTARTING"; break;
1196 case VREADONLY : s = "VREADONLY"; break;
1197 case EAGAIN : s = "EAGAIN"; break;
1198 case UAEAGAIN : s = "UAEAGAIN"; break;
1199 case EINVAL : s = "EINVAL"; break;
1200 case UAEINVAL : s = "UAEINVAL"; break;
1201 case EACCES : s = "EACCES"; break;
1202 case UAEACCES : s = "UAEACCES"; break;
1203 case ENOENT : s = "ENOENT"; break;
1204 case UAENOENT : s = "UAENOENT"; break;
1205 case EEXIST : s = "EEXIST"; break;
1206 case UAEEXIST : s = "UAEEXIST"; break;
1207 case VICECONNBAD : s = "VICECONNBAD"; break;
1208 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1209 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1210 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1211 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1212 case VL_IO : s = "VL_IO"; break;
1213 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1214 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1215 case VL_NOENT : s = "VL_NOENT"; break;
1216 case VL_EMPTY : s = "VL_EMPTY"; break;
1217 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1218 case VL_BADNAME : s = "VL_BADNAME"; break;
1219 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1220 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1221 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1222 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1223 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1224 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1225 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1226 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1227 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1228 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1229 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1230 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1231 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1232 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1233 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1234 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1235 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1236 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1237 case VL_PERM : s = "VL_PERM"; break;
1238 case VL_NOMEM : s = "VL_NOMEM"; break;
1239 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1240 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1241 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1242 case VL_BADMASK : s = "VL_BADMASK"; break;
1243 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1244 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1245 case CM_ERROR_TIMEDOUT : s = "CM_ERROR_TIMEDOUT"; break;
1246 case CM_ERROR_RETRY : s = "CM_ERROR_RETRY"; break;
1247 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1248 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1249 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1250 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1251 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1252 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1253 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1254 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1255 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1256 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1257 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1258 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1259 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1260 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1261 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1262 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1263 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1264 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1265 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1266 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1267 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1268 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1269 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1270 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1271 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1272 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1273 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1274 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1275 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1276 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1277 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1278 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1279 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1280 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1281 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1282 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1283 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1284 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1285 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1286 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1287 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1288 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1289 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1290 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1291 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1292 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1293 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1294 case UAEIO : s = "UAEIO"; break;
1295 case EIO : s = "EIO"; break;
1297 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1303 /* If not allowed to retry, don't */
1304 if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
1305 (errorCode != RX_MSGSIZE && errorCode != RX_CALL_BUSY))
1307 else if (retry && dead_session)
1310 /* drop this on the way out */
1315 * clear the volume updated flag if we succeed.
1316 * this way the flag will not prevent a subsequent volume
1317 * from being updated if necessary.
1321 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1325 errorCode != VBUSY &&
1326 errorCode != VRESTARTING &&
1327 errorCode != CM_ERROR_ALLBUSY)
1329 cm_ResetServerBusyStatus(serversp);
1332 /* retry until we fail to find a connection */
1336 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1337 cm_req_t *reqp, cm_conn_t **connpp)
1340 cm_serverRef_t *tsrp;
1342 long firstError = 0;
1343 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1, allDeleted = 1;
1344 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1345 long timeUsed, timeLeft, hardTimeLeft;
1349 if (serversp == NULL) {
1350 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_EMPTY);
1351 return CM_ERROR_EMPTY;
1354 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1355 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1357 /* leave 5 seconds margin of safety */
1358 timeLeft = ConnDeadtimeout - timeUsed - 5;
1359 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1362 lock_ObtainRead(&cm_serverLock);
1363 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1364 if (tsrp->status == srv_deleted)
1370 if (reqp->errorServp) {
1372 * search the list until we find the server
1373 * that failed last time. When we find it
1374 * clear the error, skip it and try the next one
1377 if (tsp == reqp->errorServp)
1378 reqp->errorServp = NULL;
1382 cm_GetServerNoLock(tsp);
1383 lock_ReleaseRead(&cm_serverLock);
1384 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1386 if (tsrp->status == srv_busy) {
1389 } else if (tsrp->status == srv_offline) {
1395 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1396 if (code == 0) { /* cm_CBS only returns 0 */
1398 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1399 /* Set RPC timeout */
1400 if (timeLeft > ConnDeadtimeout)
1401 timeLeft = ConnDeadtimeout;
1403 if (hardTimeLeft > HardDeadtimeout)
1404 hardTimeLeft = HardDeadtimeout;
1406 lock_ObtainMutex(&(*connpp)->mx);
1407 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1408 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1409 lock_ReleaseMutex(&(*connpp)->mx);
1414 /* therefore, this code is never executed */
1415 if (firstError == 0)
1419 lock_ObtainRead(&cm_serverLock);
1420 cm_PutServerNoLock(tsp);
1423 lock_ReleaseRead(&cm_serverLock);
1425 if (firstError == 0) {
1427 firstError = CM_ERROR_EMPTY;
1428 } else if (allDown) {
1429 firstError = (reqp->tokenError ? reqp->tokenError :
1430 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1432 * if we experienced either a token error or and idle dead time error
1433 * and now all of the servers are down, we have either tried them
1434 * all or lost connectivity. Clear the error we are returning so
1435 * we will not return it indefinitely if the request is retried.
1437 reqp->idleError = reqp->tokenError = 0;
1438 } else if (allBusy) {
1439 firstError = CM_ERROR_ALLBUSY;
1440 } else if (allOffline || (someBusy && someOffline)) {
1441 firstError = CM_ERROR_ALLOFFLINE;
1443 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1444 firstError = CM_ERROR_TIMEDOUT;
1448 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1452 /* called with a held server to GC all bad connections hanging off of the server */
1453 void cm_GCConnections(cm_server_t *serverp)
1459 lock_ObtainWrite(&cm_connLock);
1460 lcpp = &serverp->connsp;
1461 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1463 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1464 /* do the deletion of this guy */
1465 cm_PutServer(tcp->serverp);
1466 cm_ReleaseUser(userp);
1468 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1469 rx_DestroyConnection(tcp->rxconnp);
1470 lock_FinalizeMutex(&tcp->mx);
1474 /* just advance to the next */
1478 lock_ReleaseWrite(&cm_connLock);
1481 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1482 cm_server_t *serverp, afs_uint32 replicated)
1484 unsigned short port;
1487 struct rx_securityClass *secObjp;
1489 port = serverp->addr.sin_port;
1490 switch (serverp->type) {
1491 case CM_SERVER_VLDB:
1496 case CM_SERVER_FILE:
1502 osi_panic("unknown server type", __FILE__, __LINE__);
1505 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1509 tcp->cryptlevel = rxkad_clear;
1512 tcp->cryptlevel = rxkad_auth;
1515 tcp->cryptlevel = rxkad_crypt;
1517 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1518 &ucellp->sessionKey, ucellp->kvno,
1519 ucellp->ticketLen, ucellp->ticketp);
1523 tcp->cryptlevel = rxkad_clear;
1524 secObjp = rxnull_NewClientSecurityObject();
1526 osi_assertx(secObjp != NULL, "null rx_securityClass");
1527 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1532 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1533 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1536 * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
1539 tcp->flags &= CM_CONN_FLAG_REPLICATION;
1540 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1542 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1546 * Let the Rx library know that we can auto-retry if an
1547 * RX_MSGSIZE error is returned.
1549 if (rx_pmtu_discovery)
1550 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1553 * Attempt to limit NAT pings to the anonymous file server connections.
1554 * Only file servers implement client callbacks and we only need one ping
1555 * to be sent to each server.
1557 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1558 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1559 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1562 tcp->ucgen = ucellp->gen;
1564 rxs_Release(secObjp); /* Decrement the initial refCount */
1567 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1574 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1575 userp = cm_rootUserp;
1577 lock_ObtainMutex(&userp->mx);
1578 lock_ObtainRead(&cm_connLock);
1579 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1580 if (tcp->userp == userp &&
1581 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1582 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1586 /* find ucell structure */
1587 ucellp = cm_GetUCell(userp, serverp->cellp);
1589 lock_ConvertRToW(&cm_connLock);
1590 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1591 if (tcp->userp == userp)
1595 InterlockedIncrement(&tcp->refCount);
1596 lock_ReleaseWrite(&cm_connLock);
1599 cm_GetServer(serverp);
1600 tcp = malloc(sizeof(*tcp));
1601 memset(tcp, 0, sizeof(*tcp));
1604 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1605 tcp->serverp = serverp;
1606 tcp->cryptlevel = rxkad_clear;
1607 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1609 tcp->nextp = serverp->connsp;
1610 serverp->connsp = tcp;
1611 lock_ReleaseWrite(&cm_connLock);
1612 lock_ReleaseMutex(&userp->mx);
1614 InterlockedIncrement(&tcp->refCount);
1615 lock_ReleaseRead(&cm_connLock);
1617 lock_ReleaseMutex(&userp->mx);
1619 lock_ObtainMutex(&tcp->mx);
1620 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1621 (tcp->ucgen < ucellp->gen) ||
1622 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1624 if (tcp->ucgen < ucellp->gen)
1625 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1627 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1628 tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
1629 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1630 rx_DestroyConnection(tcp->rxconnp);
1631 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1633 lock_ReleaseMutex(&tcp->mx);
1636 /* return this pointer to our caller */
1637 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1643 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1647 cm_serverRef_t **serverspp;
1648 cm_serverRef_t *tsrp;
1650 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1651 afs_uint32 replicated;
1655 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1659 lock_ObtainRead(&cm_serverLock);
1660 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1661 if (tsrp->status == srv_deleted)
1664 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1666 if (tsrp->status == srv_busy) {
1669 } else if (tsrp->status == srv_offline) {
1678 lock_ReleaseRead(&cm_serverLock);
1679 cm_FreeServerList(serverspp, 0);
1685 else if (allOffline || (someBusy && someOffline))
1692 * The returned cm_conn_t ** object is released in the subsequent call
1695 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1699 cm_serverRef_t **serverspp;
1700 afs_uint32 replicated;
1704 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1708 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1709 cm_FreeServerList(serverspp, 0);
1714 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1718 cm_serverRef_t **serverspp;
1719 afs_uint32 replicated;
1723 code = cm_GetVolServerList(volp, volid, userp, reqp, &replicated, &serverspp);
1727 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1728 cm_FreeServerList(serverspp, 0);
1733 extern struct rx_connection *
1734 cm_GetRxConn(cm_conn_t *connp)
1736 struct rx_connection * rxconnp;
1737 lock_ObtainMutex(&connp->mx);
1738 rxconnp = connp->rxconnp;
1739 rx_GetConnection(rxconnp);
1740 lock_ReleaseMutex(&connp->mx);
1744 void cm_ForceNewConnections(cm_server_t *serverp)
1748 lock_ObtainWrite(&cm_connLock);
1749 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1750 lock_ObtainMutex(&tcp->mx);
1751 tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
1752 lock_ReleaseMutex(&tcp->mx);
1754 lock_ReleaseWrite(&cm_connLock);