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 * Analyze the error return from an RPC. Determine whether or not to retry,
246 * and if we're going to retry, determine whether failover is appropriate,
247 * and whether timed backoff is appropriate.
249 * If the error code is from cm_ConnFromFID() or friends, it will be a CM_ERROR code.
250 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
251 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
252 * be a security code (e.g. RXKADEXPIRED).
254 * If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
256 * For VLDB calls, fidp will be NULL.
258 * volSyncp and/or cbrp may also be NULL.
261 cm_Analyze(cm_conn_t *connp,
266 AFSVolSync *volSyncp,
267 cm_serverRef_t * serversp,
268 cm_callbackRequest_t *cbrp,
271 cm_server_t *serverp = NULL;
272 cm_serverRef_t **serverspp = NULL;
273 cm_serverRef_t *tsrp;
274 cm_cell_t *cellp = NULL;
276 cm_volume_t * volp = NULL;
277 cm_vol_state_t *statep = NULL;
278 cm_scache_t * scp = NULL;
279 afs_uint32 replicated;
281 int free_svr_list = 0;
283 long timeUsed, timeLeft;
285 char addr[16]="unknown";
287 int location_updated = 0;
291 osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
294 /* no locking required, since connp->serverp never changes after
296 dead_session = (userp->cellInfop == NULL);
298 serverp = connp->serverp;
300 /* Update callback pointer */
301 if (cbrp && serverp && errorCode == 0) {
303 if ( cbrp->serverp != serverp ) {
304 lock_ObtainWrite(&cm_serverLock);
305 cm_PutServerNoLock(cbrp->serverp);
306 cm_GetServerNoLock(serverp);
307 lock_ReleaseWrite(&cm_serverLock);
310 cm_GetServer(serverp);
312 cbrp->serverp = serverp;
315 /* if timeout - check that it did not exceed the HardDead timeout
318 /* timeleft - get it from reqp the same way as cm_ConnByMServers does */
319 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
320 if ( reqp->flags & CM_REQ_SOURCE_SMB )
321 timeLeft = HardDeadtimeout - timeUsed;
323 timeLeft = 0x0FFFFFFF;
325 /* get a pointer to the cell */
327 if (cellp == NULL && serverp)
328 cellp = serverp->cellp;
329 if (cellp == NULL && serversp) {
330 struct cm_serverRef * refp;
331 for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
332 if (refp->status == srv_deleted)
335 cellp = refp->server->cellp;
338 if (cellp == NULL && fidp) {
339 cellp = cm_FindCellByID(fidp->cell, 0);
343 if (errorCode == CM_ERROR_TIMEDOUT) {
344 if ( timeLeft > 5 ) {
346 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
351 else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
352 errorCode == UAEAGAIN || errorCode == EAGAIN) {
353 osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
354 if ( timeLeft > 5 ) {
360 /* if there is nosuchvolume, then we have a situation in which a
361 * previously known volume no longer has a set of servers
362 * associated with it. Either the volume has moved or
363 * the volume has been deleted. Try to find a new server list
364 * until the timeout period expires.
366 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
367 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
369 * The VNOVOL or VL_NOENT error has already been translated
370 * to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
374 else if (errorCode == CM_ERROR_ALLDOWN) {
375 /* Servers marked DOWN will be restored by the background daemon
376 * thread as they become available. The volume status is
377 * updated as the server state changes.
380 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
381 cellp->name, fidp->volume);
382 msgID = MSG_ALL_SERVERS_DOWN;
383 format = "All servers are unreachable when accessing cell %s volume %d.";
384 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
386 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
390 else if (errorCode == CM_ERROR_ALLOFFLINE) {
391 /* Volume instances marked offline will be restored by the
392 * background daemon thread as they become available
395 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
396 cellp->name, fidp->volume);
397 msgID = MSG_ALL_SERVERS_OFFLINE;
398 format = "All servers are offline when accessing cell %s volume %d.";
399 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
401 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
402 CM_GETVOL_FLAG_NO_LRU_UPDATE,
406 * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
407 * was called by cm_CheckOfflineVolumeState().
409 if (!(reqp->flags & CM_REQ_OFFLINE_VOL_CHK) && timeLeft > 7) {
412 /* cm_CheckOfflineVolume() resets the serverRef state */
413 if (cm_CheckOfflineVolume(volp, fidp->volume))
416 cm_UpdateVolumeStatus(volp, fidp->volume);
418 lock_ObtainRead(&cm_volumeLock);
420 lock_ReleaseRead(&cm_volumeLock);
424 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
427 else if (errorCode == CM_ERROR_ALLBUSY) {
428 /* Volumes that are busy cannot be determined to be non-busy
429 * without actually attempting to access them.
431 if (fidp) { /* File Server query */
432 osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
433 cellp->name, fidp->volume);
434 msgID = MSG_ALL_SERVERS_BUSY;
435 format = "All servers are busy when accessing cell %s volume %d.";
436 LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
438 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
439 CM_GETVOL_FLAG_NO_LRU_UPDATE,
445 statep = cm_VolumeStateByID(volp, fidp->volume);
446 if (statep->state != vl_offline &&
447 statep->state != vl_busy &&
448 statep->state != vl_unknown) {
452 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
454 serversp = *serverspp;
458 lock_ObtainWrite(&cm_serverLock);
459 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
460 if (tsrp->status == srv_deleted)
462 if (tsrp->status == srv_busy) {
463 tsrp->status = srv_not_busy;
466 lock_ReleaseWrite(&cm_serverLock);
468 cm_FreeServerList(serverspp, 0);
473 cm_UpdateVolumeStatus(volp, fidp->volume);
477 cm_UpdateVolumeStatus(volp, fidp->volume);
480 lock_ObtainRead(&cm_volumeLock);
482 lock_ReleaseRead(&cm_volumeLock);
485 } else { /* VL Server query */
486 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
492 lock_ObtainWrite(&cm_serverLock);
493 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
494 if (tsrp->status == srv_deleted)
496 if (tsrp->status == srv_busy) {
497 tsrp->status = srv_not_busy;
500 lock_ReleaseWrite(&cm_serverLock);
507 /* special codes: VBUSY and VRESTARTING */
508 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
509 if (!serversp && fidp) {
510 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
512 serversp = *serverspp;
517 switch ( errorCode ) {
519 msgID = MSG_SERVER_REPORTS_VBUSY;
520 format = "Server %s reported busy when accessing volume %d in cell %s.";
523 msgID = MSG_SERVER_REPORTS_VRESTARTING;
524 format = "Server %s reported restarting when accessing volume %d in cell %s.";
528 if (serverp && fidp) {
529 /* Log server being offline for this volume */
530 sprintf(addr, "%d.%d.%d.%d",
531 ((serverp->addr.sin_addr.s_addr & 0xff)),
532 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
533 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
534 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
536 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
537 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
540 lock_ObtainWrite(&cm_serverLock);
541 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
542 if (tsrp->status == srv_deleted)
544 if (tsrp->server == serverp && tsrp->status == srv_not_busy) {
545 tsrp->status = srv_busy;
546 if (fidp) { /* File Server query */
547 lock_ReleaseWrite(&cm_serverLock);
548 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
549 CM_GETVOL_FLAG_NO_LRU_UPDATE,
552 statep = cm_VolumeStateByID(volp, fidp->volume);
553 lock_ObtainWrite(&cm_serverLock);
558 lock_ReleaseWrite(&cm_serverLock);
561 cm_UpdateVolumeStatus(volp, statep->ID);
562 lock_ObtainRead(&cm_volumeLock);
564 lock_ReleaseRead(&cm_volumeLock);
569 cm_FreeServerList(serverspp, 0);
576 /* special codes: missing volumes */
577 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
578 errorCode == VSALVAGE || errorCode == VNOSERVICE || errorCode == VIO)
580 /* In case of timeout */
581 reqp->volumeError = errorCode;
583 switch ( errorCode ) {
585 msgID = MSG_SERVER_REPORTS_VNOVOL;
586 format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
589 msgID = MSG_SERVER_REPORTS_VMOVED;
590 format = "Server %s reported volume %d in cell %s as moved.";
593 msgID = MSG_SERVER_REPORTS_VOFFLINE;
594 format = "Server %s reported volume %d in cell %s as offline.";
597 msgID = MSG_SERVER_REPORTS_VSALVAGE;
598 format = "Server %s reported volume %d in cell %s as needs salvage.";
601 msgID = MSG_SERVER_REPORTS_VNOSERVICE;
602 format = "Server %s reported volume %d in cell %s as not in service.";
605 msgID = MSG_SERVER_REPORTS_VIO;
606 format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
610 if (fidp) { /* File Server query */
612 /* Log server being unavailable for this volume */
613 sprintf(addr, "%d.%d.%d.%d",
614 ((serverp->addr.sin_addr.s_addr & 0xff)),
615 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
616 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
617 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
619 osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
620 LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
623 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
624 CM_GETVOL_FLAG_NO_LRU_UPDATE,
627 statep = cm_VolumeStateByID(volp, fidp->volume);
629 if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
630 !(reqp->flags & CM_REQ_VOLUME_UPDATED))
632 code = cm_ForceUpdateVolume(fidp, userp, reqp);
634 location_updated = 1;
636 /* Even if the update fails, there might still be another replica */
638 reqp->flags |= CM_REQ_VOLUME_UPDATED;
639 osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
640 fidp->cell, fidp->volume, code);
644 cm_UpdateVolumeStatus(volp, statep->ID);
645 osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
646 fidp->cell, fidp->volume, statep->state);
650 lock_ObtainRead(&cm_volumeLock);
652 lock_ReleaseRead(&cm_volumeLock);
657 * Mark server offline for this volume or delete the volume
658 * from the server list if it was moved or is not present.
660 if (!serversp || location_updated) {
661 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
663 serversp = *serverspp;
669 lock_ObtainWrite(&cm_serverLock);
670 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
671 if (tsrp->status == srv_deleted)
674 sprintf(addr, "%d.%d.%d.%d",
675 ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
676 ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
677 ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
678 ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
680 if (cm_ServerEqual(tsrp->server, serverp)) {
682 if (errorCode == VMOVED || errorCode == VNOVOL) {
683 osi_Log2(afsd_logp, "volume %d not present on server %s",
684 fidp->volume, osi_LogSaveString(afsd_logp,addr));
685 tsrp->status = srv_deleted;
687 cm_RemoveVolumeFromServer(serverp, fidp->volume);
689 osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
690 fidp->volume, osi_LogSaveString(afsd_logp,addr));
691 tsrp->status = srv_offline;
695 osi_Log3(afsd_logp, "volume %d exists on server %s with status %u",
696 fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
699 lock_ReleaseWrite(&cm_serverLock);
701 /* Free the server list before cm_ForceUpdateVolume is called */
703 cm_FreeServerList(serverspp, 0);
710 } else if ( errorCode == VNOVNODE ) {
712 osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
713 fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
715 scp = cm_FindSCache(fidp);
717 cm_scache_t *pscp = NULL;
719 if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
720 pscp = cm_FindSCacheParent(scp);
722 lock_ObtainWrite(&scp->rw);
723 scp->flags |= CM_SCACHEFLAG_DELETED;
724 lock_ObtainWrite(&cm_scacheLock);
725 cm_AdjustScacheLRU(scp);
726 cm_RemoveSCacheFromHashTable(scp);
727 lock_ReleaseWrite(&cm_scacheLock);
728 cm_LockMarkSCacheLost(scp);
729 lock_ReleaseWrite(&scp->rw);
731 RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
732 scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
733 cm_ReleaseSCache(scp);
736 if (cm_HaveCallback(pscp)) {
737 lock_ObtainWrite(&pscp->rw);
738 cm_DiscardSCache(pscp);
739 lock_ReleaseWrite(&pscp->rw);
742 RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
743 pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
746 cm_ReleaseSCache(pscp);
750 osi_Log0(afsd_logp, "cm_Analyze passed VNOVNODE unknown fid.");
755 else if (errorCode == RX_CALL_TIMEOUT) {
756 /* RPC took longer than hardDeadTime or the server
757 * reported idle for longer than idleDeadTime
758 * don't mark server as down but don't retry
759 * this is to prevent the SMB session from timing out
760 * In addition, we log an event to the event log
764 sprintf(addr, "%d.%d.%d.%d",
765 ((serverp->addr.sin_addr.s_addr & 0xff)),
766 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
767 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
768 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
770 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
771 osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
772 osi_LogSaveString(afsd_logp,addr));
773 reqp->tokenIdleErrorServp = serverp;
778 scp = cm_FindSCache(fidp);
780 if (cm_HaveCallback(scp)) {
781 lock_ObtainWrite(&scp->rw);
782 cm_DiscardSCache(scp);
783 lock_ReleaseWrite(&scp->rw);
786 * We really should notify the redirector that we discarded
787 * the status information but doing so in this case is not
788 * safe as it can result in a deadlock with extent release
792 cm_ReleaseSCache(scp);
796 if (!fidp) { /* vldb */
799 cm_volume_t *volp = cm_GetVolumeByFID(fidp);
801 if (fidp->volume == cm_GetROVolumeID(volp))
808 else if (errorCode == RX_MSGSIZE) {
810 * RPC failed because a transmitted rx packet
811 * may have grown larger than the path mtu.
812 * Force a retry and the Rx library will try
813 * with a smaller mtu size.
817 sprintf(addr, "%d.%d.%d.%d",
818 ((serverp->addr.sin_addr.s_addr & 0xff)),
819 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
820 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
821 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
823 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
824 osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
825 osi_LogSaveString(afsd_logp,addr));
829 else if (errorCode == RX_CALL_BUSY) {
831 * RPC failed because the selected call channel
832 * is currently busy on the server. Unconditionally
833 * retry the request so an alternate call channel can be used.
836 sprintf(addr, "%d.%d.%d.%d",
837 ((serverp->addr.sin_addr.s_addr & 0xff)),
838 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
839 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
840 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
842 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
843 osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
844 osi_LogSaveString(afsd_logp,addr));
847 else if (errorCode == RX_CALL_IDLE) {
849 * RPC failed because the server failed to respond with data
850 * within the idle dead timeout period. This could be for a variety
852 * 1. The server could have a bad partition such as a failed
853 * disk or iSCSI target and all I/O to that partition is
854 * blocking on the server and will never complete.
856 * 2. The server vnode may be locked by another client request
857 * that is taking a very long time.
859 * 3. The server may have a very long queue of requests
860 * pending and is unable to process this request.
862 * 4. The server could be malicious and is performing a denial
863 * of service attack against the client.
865 * If this is a request against a .readonly with alternate sites
866 * the server should be marked down for this request and the
867 * client should fail over to another server. If this is a
868 * request against a single source, the client may retry once.
871 sprintf(addr, "%d.%d.%d.%d",
872 ((serverp->addr.sin_addr.s_addr & 0xff)),
873 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
874 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
875 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
878 code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
879 CM_GETVOL_FLAG_NO_LRU_UPDATE,
882 statep = cm_VolumeStateByID(volp, fidp->volume);
885 replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
887 lock_ObtainRead(&cm_volumeLock);
889 lock_ReleaseRead(&cm_volumeLock);
894 scp = cm_FindSCache(fidp);
896 if (cm_HaveCallback(scp)) {
897 lock_ObtainWrite(&scp->rw);
898 cm_DiscardSCache(scp);
899 lock_ReleaseWrite(&scp->rw);
902 * We really should notify the redirector that we discarded
903 * the status information but doing so in this case is not
904 * safe as it can result in a deadlock with extent release
908 cm_ReleaseSCache(scp);
912 if (replicated && serverp) {
913 reqp->tokenIdleErrorServp = serverp;
914 reqp->tokenError = errorCode;
920 LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
921 osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
922 osi_LogSaveString(afsd_logp,addr), retry);
924 else if (errorCode == RX_CALL_DEAD) {
925 /* mark server as down */
927 sprintf(addr, "%d.%d.%d.%d",
928 ((serverp->addr.sin_addr.s_addr & 0xff)),
929 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
930 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
931 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
933 osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
934 osi_LogSaveString(afsd_logp,addr),
935 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
938 if ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
939 lock_ObtainMutex(&serverp->mx);
940 if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
941 _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
942 serverp->downTime = time(NULL);
944 lock_ReleaseMutex(&serverp->mx);
946 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
948 cm_ForceNewConnections(serverp);
953 scp = cm_FindSCache(fidp);
955 if (cm_HaveCallback(scp)) {
956 lock_ObtainWrite(&scp->rw);
957 cm_DiscardSCache(scp);
958 lock_ReleaseWrite(&scp->rw);
961 * We really should notify the redirector that we discarded
962 * the status information but doing so in this case is not
963 * safe as it can result in a deadlock with extent release
967 cm_ReleaseSCache(scp);
973 else if (errorCode >= -64 && errorCode < 0) {
974 /* mark server as down */
976 sprintf(addr, "%d.%d.%d.%d",
977 ((serverp->addr.sin_addr.s_addr & 0xff)),
978 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
979 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
980 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
982 osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
984 osi_LogSaveString(afsd_logp,addr),
985 (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
988 if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
989 reqp->tokenIdleErrorServp = serverp;
990 reqp->tokenError = errorCode;
992 reqp->flags |= CM_REQ_NEW_CONN_FORCED;
994 cm_ForceNewConnections(serverp);
1000 else if (errorCode == RXKADEXPIRED) {
1001 osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
1003 if (!dead_session) {
1004 lock_ObtainMutex(&userp->mx);
1005 ucellp = cm_GetUCell(userp, serverp->cellp);
1006 if (ucellp->ticketp) {
1007 free(ucellp->ticketp);
1008 ucellp->ticketp = NULL;
1010 _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
1012 lock_ReleaseMutex(&userp->mx);
1016 } else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
1017 char * s = "unknown error";
1018 switch ( errorCode ) {
1019 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
1020 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
1021 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
1022 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
1023 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
1024 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
1025 case RXKADBADKEY : s = "RXKADBADKEY"; break;
1026 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
1027 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
1028 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
1029 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
1030 case RXKADDATALEN : s = "RXKADDATALEN"; break;
1031 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
1033 osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
1037 reqp->tokenIdleErrorServp = serverp;
1038 reqp->tokenError = errorCode;
1041 } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
1043 * We received a ubik error. its possible that the server we are
1044 * communicating with has a corrupted database or is partitioned
1045 * from the rest of the servers and another server might be able
1046 * to answer our query. Therefore, we will retry the request
1047 * and force the use of another server.
1050 reqp->tokenIdleErrorServp = serverp;
1051 reqp->tokenError = errorCode;
1054 } else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
1055 cm_ForceNewConnections(serverp);
1060 char * s = "unknown error";
1061 switch ( errorCode ) {
1062 case VSALVAGE : s = "VSALVAGE"; break;
1063 case VNOVNODE : s = "VNOVNODE"; break;
1064 case VNOVOL : s = "VNOVOL"; break;
1065 case VVOLEXISTS : s = "VVOLEXISTS"; break;
1066 case VNOSERVICE : s = "VNOSERVICE"; break;
1067 case VOFFLINE : s = "VOFFLINE"; break;
1068 case VONLINE : s = "VONLINE"; break;
1069 case VDISKFULL : s = "VDISKFULL"; break;
1070 case VOVERQUOTA : s = "VOVERQUOTA"; break;
1071 case VBUSY : s = "VBUSY"; break;
1072 case VMOVED : s = "VMOVED"; break;
1073 case VIO : s = "VIO"; break;
1074 case VRESTRICTED : s = "VRESTRICTED"; break;
1075 case VRESTARTING : s = "VRESTARTING"; break;
1076 case VREADONLY : s = "VREADONLY"; break;
1077 case EAGAIN : s = "EAGAIN"; break;
1078 case UAEAGAIN : s = "UAEAGAIN"; break;
1079 case EINVAL : s = "EINVAL"; break;
1080 case UAEINVAL : s = "UAEINVAL"; break;
1081 case EACCES : s = "EACCES"; break;
1082 case UAEACCES : s = "UAEACCES"; break;
1083 case ENOENT : s = "ENOENT"; break;
1084 case UAENOENT : s = "UAENOENT"; break;
1085 case EEXIST : s = "EEXIST"; break;
1086 case UAEEXIST : s = "UAEEXIST"; break;
1087 case VICECONNBAD : s = "VICECONNBAD"; break;
1088 case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
1089 case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
1090 case UAEWOULDBLOCK : s = "UAEWOULDBLOCK"; break;
1091 case VL_IDEXIST : s = "VL_IDEXIST"; break;
1092 case VL_IO : s = "VL_IO"; break;
1093 case VL_NAMEEXIST : s = "VL_NAMEEXIST"; break;
1094 case VL_CREATEFAIL : s = "VL_CREATEFAIL"; break;
1095 case VL_NOENT : s = "VL_NOENT"; break;
1096 case VL_EMPTY : s = "VL_EMPTY"; break;
1097 case VL_ENTDELETED : s = "VL_ENTDELETED"; break;
1098 case VL_BADNAME : s = "VL_BADNAME"; break;
1099 case VL_BADINDEX : s = "VL_BADINDEX"; break;
1100 case VL_BADVOLTYPE : s = "VL_BADVOLTYPE"; break;
1101 case VL_BADSERVER : s = "VL_BADSERVER"; break;
1102 case VL_BADPARTITION : s = "VL_BADPARTITION"; break;
1103 case VL_REPSFULL : s = "VL_REPSFULL"; break;
1104 case VL_NOREPSERVER : s = "VL_NOREPSERVER"; break;
1105 case VL_DUPREPSERVER : s = "VL_DUPREPSERVER"; break;
1106 case VL_RWNOTFOUND : s = "VL_RWNOTFOUND"; break;
1107 case VL_BADREFCOUNT : s = "VL_BADREFCOUNT"; break;
1108 case VL_SIZEEXCEEDED : s = "VL_SIZEEXCEEDED"; break;
1109 case VL_BADENTRY : s = "VL_BADENTRY"; break;
1110 case VL_BADVOLIDBUMP : s = "VL_BADVOLIDBUMP"; break;
1111 case VL_IDALREADYHASHED: s = "VL_IDALREADYHASHED"; break;
1112 case VL_ENTRYLOCKED : s = "VL_ENTRYLOCKED"; break;
1113 case VL_BADVOLOPER : s = "VL_BADVOLOPER"; break;
1114 case VL_BADRELLOCKTYPE : s = "VL_BADRELLOCKTYPE"; break;
1115 case VL_RERELEASE : s = "VL_RERELEASE"; break;
1116 case VL_BADSERVERFLAG : s = "VL_BADSERVERFLAG"; break;
1117 case VL_PERM : s = "VL_PERM"; break;
1118 case VL_NOMEM : s = "VL_NOMEM"; break;
1119 case VL_BADVERSION : s = "VL_BADVERSION"; break;
1120 case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
1121 case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
1122 case VL_BADMASK : s = "VL_BADMASK"; break;
1123 case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
1124 case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
1125 case CM_ERROR_TIMEDOUT : s = "CM_ERROR_TIMEDOUT"; break;
1126 case CM_ERROR_RETRY : s = "CM_ERROR_RETRY"; break;
1127 case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
1128 case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
1129 case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
1130 case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
1131 case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
1132 case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
1133 case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
1134 case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
1135 case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
1136 case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
1137 case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
1138 case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
1139 case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
1140 case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
1141 case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
1142 case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
1143 case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
1144 case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
1145 case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
1146 case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
1147 case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
1148 case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
1149 case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
1150 case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
1151 case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
1152 case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
1153 case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
1154 case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
1155 case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
1156 case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
1157 case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
1158 case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
1159 case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
1160 case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
1161 case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
1162 case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
1163 case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
1164 case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
1165 case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
1166 case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
1167 case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
1168 case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
1169 case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
1170 case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
1171 case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
1172 case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
1173 case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
1175 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
1181 /* If not allowed to retry, don't */
1182 if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
1183 (errorCode != RX_MSGSIZE && errorCode != RX_CALL_BUSY))
1185 else if (retry && dead_session)
1188 /* drop this on the way out */
1193 * clear the volume updated flag if we succeed.
1194 * this way the flag will not prevent a subsequent volume
1195 * from being updated if necessary.
1199 reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
1202 /* retry until we fail to find a connection */
1206 long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
1207 cm_req_t *reqp, cm_conn_t **connpp)
1210 cm_serverRef_t *tsrp;
1212 long firstError = 0;
1213 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1214 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1215 long timeUsed, timeLeft, hardTimeLeft;
1219 if (serversp == NULL) {
1220 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_ALLDOWN);
1221 return CM_ERROR_ALLDOWN;
1224 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1225 timeUsed = (GetTickCount() - reqp->startTime) / 1000;
1227 /* leave 5 seconds margin of safety */
1228 timeLeft = ConnDeadtimeout - timeUsed - 5;
1229 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
1232 lock_ObtainRead(&cm_serverLock);
1233 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
1234 if (tsrp->status == srv_deleted)
1238 if (reqp->tokenIdleErrorServp) {
1240 * search the list until we find the server
1241 * that failed last time. When we find it
1242 * clear the error, skip it and try the next one
1245 if (tsp == reqp->tokenIdleErrorServp)
1246 reqp->tokenIdleErrorServp = NULL;
1250 cm_GetServerNoLock(tsp);
1251 lock_ReleaseRead(&cm_serverLock);
1252 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1254 if (tsrp->status == srv_busy) {
1257 } else if (tsrp->status == srv_offline) {
1263 code = cm_ConnByServer(tsp, usersp, replicated, connpp);
1264 if (code == 0) { /* cm_CBS only returns 0 */
1266 #ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
1267 /* Set RPC timeout */
1268 if (timeLeft > ConnDeadtimeout)
1269 timeLeft = ConnDeadtimeout;
1271 if (hardTimeLeft > HardDeadtimeout)
1272 hardTimeLeft = HardDeadtimeout;
1274 lock_ObtainMutex(&(*connpp)->mx);
1275 rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
1276 rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
1277 lock_ReleaseMutex(&(*connpp)->mx);
1282 /* therefore, this code is never executed */
1283 if (firstError == 0)
1287 lock_ObtainRead(&cm_serverLock);
1288 cm_PutServerNoLock(tsp);
1291 lock_ReleaseRead(&cm_serverLock);
1293 if (firstError == 0) {
1295 firstError = (reqp->tokenError ? reqp->tokenError :
1296 (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
1298 * if we experienced either a token error or and idle dead time error
1299 * and now all of the servers are down, we have either tried them
1300 * all or lost connectivity. Clear the error we are returning so
1301 * we will not return it indefinitely if the request is retried.
1303 reqp->idleError = reqp->tokenError = 0;
1304 } else if (allBusy) {
1305 firstError = CM_ERROR_ALLBUSY;
1306 } else if (allOffline || (someBusy && someOffline)) {
1307 firstError = CM_ERROR_ALLOFFLINE;
1309 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
1310 firstError = CM_ERROR_TIMEDOUT;
1314 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
1318 /* called with a held server to GC all bad connections hanging off of the server */
1319 void cm_GCConnections(cm_server_t *serverp)
1325 lock_ObtainWrite(&cm_connLock);
1326 lcpp = &serverp->connsp;
1327 for (tcp = *lcpp; tcp; tcp = *lcpp) {
1329 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
1330 /* do the deletion of this guy */
1331 cm_PutServer(tcp->serverp);
1332 cm_ReleaseUser(userp);
1334 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1335 rx_DestroyConnection(tcp->rxconnp);
1336 lock_FinalizeMutex(&tcp->mx);
1340 /* just advance to the next */
1344 lock_ReleaseWrite(&cm_connLock);
1347 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
1348 cm_server_t *serverp, afs_uint32 replicated)
1350 unsigned short port;
1353 struct rx_securityClass *secObjp;
1355 port = serverp->addr.sin_port;
1356 switch (serverp->type) {
1357 case CM_SERVER_VLDB:
1362 case CM_SERVER_FILE:
1368 osi_panic("unknown server type", __FILE__, __LINE__);
1371 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
1375 tcp->cryptlevel = rxkad_clear;
1378 tcp->cryptlevel = rxkad_auth;
1381 tcp->cryptlevel = rxkad_crypt;
1383 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
1384 &ucellp->sessionKey, ucellp->kvno,
1385 ucellp->ticketLen, ucellp->ticketp);
1389 tcp->cryptlevel = rxkad_clear;
1390 secObjp = rxnull_NewClientSecurityObject();
1392 osi_assertx(secObjp != NULL, "null rx_securityClass");
1393 tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
1398 rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
1399 rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
1402 * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
1405 tcp->flags &= CM_CONN_FLAG_REPLICATION;
1406 rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
1408 rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
1412 * Let the Rx library know that we can auto-retry if an
1413 * RX_MSGSIZE error is returned.
1415 if (rx_pmtu_discovery)
1416 rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
1419 * Attempt to limit NAT pings to the anonymous file server connections.
1420 * Only file servers implement client callbacks and we only need one ping
1421 * to be sent to each server.
1423 if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
1424 (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
1425 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
1428 tcp->ucgen = ucellp->gen;
1430 rxs_Release(secObjp); /* Decrement the initial refCount */
1433 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
1440 if (cm_anonvldb && serverp->type == CM_SERVER_VLDB)
1441 userp = cm_rootUserp;
1443 lock_ObtainMutex(&userp->mx);
1444 lock_ObtainRead(&cm_connLock);
1445 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1446 if (tcp->userp == userp &&
1447 (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
1448 !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
1452 /* find ucell structure */
1453 ucellp = cm_GetUCell(userp, serverp->cellp);
1455 lock_ConvertRToW(&cm_connLock);
1456 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1457 if (tcp->userp == userp)
1461 lock_ReleaseWrite(&cm_connLock);
1464 cm_GetServer(serverp);
1465 tcp = malloc(sizeof(*tcp));
1466 memset(tcp, 0, sizeof(*tcp));
1467 tcp->nextp = serverp->connsp;
1468 serverp->connsp = tcp;
1471 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
1472 lock_ObtainMutex(&tcp->mx);
1473 tcp->serverp = serverp;
1474 tcp->cryptlevel = rxkad_clear;
1475 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1477 lock_ReleaseMutex(&tcp->mx);
1478 lock_ReleaseWrite(&cm_connLock);
1480 lock_ReleaseRead(&cm_connLock);
1482 InterlockedIncrement(&tcp->refCount);
1484 lock_ObtainMutex(&tcp->mx);
1485 if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
1486 (tcp->ucgen < ucellp->gen) ||
1487 (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
1489 if (tcp->ucgen < ucellp->gen)
1490 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
1492 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
1493 tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
1494 rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
1495 rx_DestroyConnection(tcp->rxconnp);
1496 cm_NewRXConnection(tcp, ucellp, serverp, replicated);
1498 lock_ReleaseMutex(&tcp->mx);
1500 lock_ReleaseMutex(&userp->mx);
1502 /* return this pointer to our caller */
1503 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
1509 long cm_ServerAvailable(struct cm_fid *fidp, struct cm_user *userp)
1513 cm_serverRef_t **serverspp;
1514 cm_serverRef_t *tsrp;
1516 int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
1517 afs_uint32 replicated;
1521 code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
1525 lock_ObtainRead(&cm_serverLock);
1526 for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
1527 if (tsrp->status == srv_deleted)
1530 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
1532 if (tsrp->status == srv_busy) {
1535 } else if (tsrp->status == srv_offline) {
1544 lock_ReleaseRead(&cm_serverLock);
1545 cm_FreeServerList(serverspp, 0);
1551 else if (allOffline || (someBusy && someOffline))
1558 * The returned cm_conn_t ** object is released in the subsequent call
1561 long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
1565 cm_serverRef_t **serverspp;
1566 afs_uint32 replicated;
1570 code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
1574 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1575 cm_FreeServerList(serverspp, 0);
1580 long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_user *userp, cm_req_t *reqp,
1584 cm_serverRef_t **serverspp;
1585 afs_uint32 replicated;
1586 cm_vol_state_t * volstatep;
1590 volstatep = cm_VolumeStateByID(volp, volid);
1591 replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
1592 serverspp = cm_GetVolServers(volp, volid, userp, reqp);
1594 code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
1595 cm_FreeServerList(serverspp, 0);
1600 extern struct rx_connection *
1601 cm_GetRxConn(cm_conn_t *connp)
1603 struct rx_connection * rxconnp;
1604 lock_ObtainMutex(&connp->mx);
1605 rxconnp = connp->rxconnp;
1606 rx_GetConnection(rxconnp);
1607 lock_ReleaseMutex(&connp->mx);
1611 void cm_ForceNewConnections(cm_server_t *serverp)
1615 lock_ObtainWrite(&cm_connLock);
1616 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
1617 lock_ObtainMutex(&tcp->mx);
1618 tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
1619 lock_ReleaseMutex(&tcp->mx);
1621 lock_ReleaseWrite(&cm_connLock);