/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
unsigned short ConnDeadtimeout = CM_CONN_CONNDEADTIME;
unsigned short HardDeadtimeout = CM_CONN_HARDDEADTIME;
unsigned short IdleDeadtimeout = CM_CONN_IDLEDEADTIME;
+unsigned short ReplicaIdleDeadtimeout = CM_CONN_IDLEDEADTIME_REP;
unsigned short NatPingInterval = CM_CONN_NATPINGINTERVAL;
#define LANMAN_WKS_PARAM_KEY "SYSTEM\\CurrentControlSet\\Services\\lanmanworkstation\\parameters"
DWORD dwValue;
DWORD dummyLen;
HKEY parmKey;
-
+
if (osi_Once(&once)) {
lock_InitializeRWLock(&cm_connLock, "connection global lock",
LOCK_HIERARCHY_CONN_GLOBAL);
/* keisa - read timeout value for lanmanworkstation service.
- * jaltman - as per
+ * jaltman - as per
* http://support.microsoft.com:80/support/kb/articles/Q102/0/67.asp&NoWebContent=1
- * the SessTimeout is a minimum timeout not a maximum timeout. Therefore,
+ * the SessTimeout is a minimum timeout not a maximum timeout. Therefore,
* I believe that the default should not be short. Instead, we should wait until
* RX times out before reporting a timeout to the SMB client.
*/
BOOL extTimeouts = msftSMBRedirectorSupportsExtendedTimeouts();
if (extTimeouts) {
- /*
+ /*
* The default value is 1000 seconds. However, this default
- * will not apply to "\\AFS" unless "AFS" is listed in
+ * will not apply to "\\AFS" unless "AFS" is listed in
* ServersWithExtendedSessTimeout which we will add when we
* create the ExtendedSessTimeout value in smb_Init()
*/
dummyLen = sizeof(DWORD);
- code = RegQueryValueEx(parmKey,
+ code = RegQueryValueEx(parmKey,
LANMAN_WKS_EXT_SESSION_TIMEOUT,
NULL, NULL,
(BYTE *) &dwValue, &dummyLen);
RDRtimeout = dwValue;
afsi_log("lanmanworkstation : ExtSessTimeout %u", RDRtimeout);
}
- }
+ }
if (!extTimeouts || code != ERROR_SUCCESS) {
dummyLen = sizeof(DWORD);
- code = RegQueryValueEx(parmKey,
+ code = RegQueryValueEx(parmKey,
LANMAN_WKS_SESSION_TIMEOUT,
NULL, NULL,
(BYTE *) &dwValue, &dummyLen);
afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
}
dummyLen = sizeof(DWORD);
+ code = RegQueryValueEx(parmKey, "ReplicaIdleDeadTimeout", NULL, NULL,
+ (BYTE *) &dwValue, &dummyLen);
+ if (code == ERROR_SUCCESS) {
+ ReplicaIdleDeadtimeout = (unsigned short)dwValue;
+ afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
+ }
+ dummyLen = sizeof(DWORD);
code = RegQueryValueEx(parmKey, "NatPingInterval", NULL, NULL,
(BYTE *) &dwValue, &dummyLen);
if (code == ERROR_SUCCESS) {
RegCloseKey(parmKey);
}
- /*
- * If these values were not set via cpp macro or obtained
+ /*
+ * If these values were not set via cpp macro or obtained
* from the registry, we use a value that is derived from
* the smb redirector session timeout (RDRtimeout).
*
* The UNIX cache manager uses 120 seconds for the hard dead
* timeout and 50 seconds for the connection and idle timeouts.
*
- * We base our values on those while making sure we leave
- * enough time for overhead.
+ * We base our values on those while making sure we leave
+ * enough time for overhead.
+ *
+ * To further complicate matters we need to take into account
+ * file server hard dead timeouts as they affect the length
+ * of time it takes the file server to give up when attempting
+ * to break callbacks to unresponsive clients. The file
+ * server hard dead timeout is 120 seconds.
+ *
+ * For SMB, we have no choice but to timeout quickly. For
+ * the AFS redirector, we can wait.
*/
- if (ConnDeadtimeout == 0) {
- ConnDeadtimeout = (unsigned short) ((RDRtimeout / 2) < 50 ? (RDRtimeout / 2) : 50);
- afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
- }
- if (HardDeadtimeout == 0) {
- HardDeadtimeout = (unsigned short) (RDRtimeout > 125 ? 120 : (RDRtimeout - 5));
- afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
- }
- if (IdleDeadtimeout == 0) {
- IdleDeadtimeout = (unsigned short) ConnDeadtimeout;
- afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
+ if (smb_Enabled) {
+ afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
+ if (ConnDeadtimeout == 0) {
+ ConnDeadtimeout = (unsigned short) ((RDRtimeout / 2) < 50 ? (RDRtimeout / 2) : 50);
+ afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
+ }
+ if (HardDeadtimeout == 0) {
+ HardDeadtimeout = (unsigned short) (RDRtimeout > 125 ? 120 : (RDRtimeout - 5));
+ afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
+ }
+ if (IdleDeadtimeout == 0) {
+ IdleDeadtimeout = 10 * (unsigned short) HardDeadtimeout;
+ afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
+ }
+ if (ReplicaIdleDeadtimeout == 0) {
+ ReplicaIdleDeadtimeout = (unsigned short) HardDeadtimeout;
+ afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
+ }
+ } else {
+ if (ConnDeadtimeout == 0) {
+ ConnDeadtimeout = CM_CONN_IFS_CONNDEADTIME;
+ afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
+ }
+ if (HardDeadtimeout == 0) {
+ HardDeadtimeout = CM_CONN_IFS_HARDDEADTIME;
+ afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
+ }
+ if (IdleDeadtimeout == 0) {
+ IdleDeadtimeout = CM_CONN_IFS_IDLEDEADTIME;
+ afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
+ }
+ if (IdleDeadtimeout == 0) {
+ ReplicaIdleDeadtimeout = CM_CONN_IFS_IDLEDEADTIME_REP;
+ afsi_log("ReplicaIdleDeadTimeout is %d", ReplicaIdleDeadtimeout);
+ }
}
osi_EndOnce(&once);
}
reqp->startTime = GetTickCount();
}
-static long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
- struct cm_req *reqp, cm_serverRef_t ***serversppp)
+long cm_GetVolServerList(cm_volume_t *volp, afs_uint32 volid, struct cm_user *userp,
+ struct cm_req *reqp, afs_uint32 *replicated,
+ cm_serverRef_t ***serversppp)
+{
+ *serversppp = cm_GetVolServers(volp, volid, userp, reqp, replicated);
+ return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
+}
+
+long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
+ struct cm_req *reqp, afs_uint32 *replicated, cm_serverRef_t ***serversppp)
{
long code;
cm_volume_t *volp = NULL;
}
cellp = cm_FindCellByID(fidp->cell, 0);
- if (!cellp)
+ if (!cellp)
return CM_ERROR_NOSUCHCELL;
code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
- if (code)
+ if (code)
return code;
-
- *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp);
+
+ *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp, replicated);
lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
}
+static void
+cm_SetServerBusyStatus(cm_serverRef_t **serverspp, cm_server_t *serverp)
+{
+ cm_serverRef_t *tsrp;
+
+ lock_ObtainRead(&cm_serverLock);
+ for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
+ if (cm_ServerEqual(tsrp->server, serverp) && tsrp->status == srv_not_busy) {
+ tsrp->status = srv_busy;
+ break;
+ }
+ }
+ lock_ReleaseRead(&cm_serverLock);
+}
+
+static void
+cm_ResetServerBusyStatus(cm_serverRef_t **serverspp)
+{
+ cm_serverRef_t *tsrp;
+
+ lock_ObtainRead(&cm_serverLock);
+ for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
+ if (tsrp->status == srv_busy) {
+ tsrp->status = srv_not_busy;
+ }
+ }
+ lock_ReleaseRead(&cm_serverLock);
+}
+
/*
* Analyze the error return from an RPC. Determine whether or not to retry,
* and if we're going to retry, determine whether failover is appropriate,
*
* If the error code is from cm_ConnFromFID() or friends, connp will be NULL.
*
- * For VLDB calls, fidp will be NULL.
+ * For VLDB calls, fidp will be NULL and cellp will not be.
*
* volSyncp and/or cbrp may also be NULL.
*/
int
-cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
- struct cm_fid *fidp,
- AFSVolSync *volSyncp,
- cm_serverRef_t * serversp,
- cm_callbackRequest_t *cbrp, long errorCode)
+cm_Analyze(cm_conn_t *connp,
+ cm_user_t *userp,
+ cm_req_t *reqp,
+ struct cm_fid *fidp,
+ cm_cell_t *cellp,
+ afs_uint32 storeOp,
+ AFSFetchStatus *statusp,
+ AFSVolSync *volSyncp,
+ cm_serverRef_t ** vlServerspp,
+ cm_callbackRequest_t *cbrp,
+ long errorCode)
{
cm_server_t *serverp = NULL;
- cm_serverRef_t **serverspp = NULL;
+ cm_serverRef_t **volServerspp = NULL;
cm_serverRef_t *tsrp;
- cm_cell_t *cellp = NULL;
cm_ucell_t *ucellp;
cm_volume_t * volp = NULL;
cm_vol_state_t *statep = NULL;
+ cm_scache_t * scp = NULL;
+ afs_uint32 replicated = 0;
int retry = 0;
int free_svr_list = 0;
- int dead_session;
+ int dead_session = (userp->cellInfop == NULL);
long timeUsed, timeLeft;
- long code;
+ long code = 0;
char addr[16]="unknown";
int forcing_new = 0;
int location_updated = 0;
char *format;
DWORD msgID;
+ int invalid_status = 0;
osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
connp, errorCode);
/* no locking required, since connp->serverp never changes after
* creation */
- dead_session = (userp->cellInfop == NULL);
if (connp)
serverp = connp->serverp;
} else {
cm_GetServer(serverp);
}
- lock_ObtainWrite(&cm_callbackLock);
cbrp->serverp = serverp;
- lock_ReleaseWrite(&cm_callbackLock);
}
/* if timeout - check that it did not exceed the HardDead timeout
* and retry */
-
+
/* timeleft - get it from reqp the same way as cm_ConnByMServers does */
timeUsed = (GetTickCount() - reqp->startTime) / 1000;
- timeLeft = HardDeadtimeout - timeUsed;
+ if ( (reqp->flags & CM_REQ_SOURCE_SMB) ||
+ (serverp && serverp->type == CM_SERVER_VLDB))
+ timeLeft = HardDeadtimeout - timeUsed;
+ else
+ timeLeft = 0x0FFFFFFF;
+
+ /*
+ * Similar to the UNIX cache manager, if the AFSFetchStatus info
+ * returned by the file server is invalid, consider the response
+ * as being equivalent to VBUSY so that another file server can
+ * be queried if there is one. If there is no replica, then the
+ * request will fail.
+ */
+ if (errorCode == 0 && statusp && !cm_IsStatusValid(statusp)) {
+ invalid_status = 1;
+ errorCode = VBUSY;
+ }
/* get a pointer to the cell */
if (errorCode) {
if (cellp == NULL && serverp)
cellp = serverp->cellp;
- if (cellp == NULL && serversp) {
+ if (cellp == NULL && vlServerspp) {
struct cm_serverRef * refp;
- for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
+ lock_ObtainRead(&cm_serverLock);
+ for ( refp=*vlServerspp ; cellp == NULL && refp != NULL; refp=refp->next) {
if (refp->status == srv_deleted)
continue;
if ( refp->server )
cellp = refp->server->cellp;
}
- }
+ lock_ReleaseRead(&cm_serverLock);
+ }
if (cellp == NULL && fidp) {
cellp = cm_FindCellByID(fidp->cell, 0);
}
}
- if (errorCode == CM_ERROR_TIMEDOUT) {
+ if (errorCode == 0) {
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+ }
+ else if (errorCode == CM_ERROR_TIMEDOUT) {
+ osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_TIMEDOUT");
if ( timeLeft > 5 ) {
thrd_Sleep(3000);
cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, cellp);
}
}
+ else if (errorCode == CM_ERROR_RETRY) {
+ osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_RETRY");
+ retry = 1;
+ }
+
else if (errorCode == UAEWOULDBLOCK || errorCode == EWOULDBLOCK ||
errorCode == UAEAGAIN || errorCode == EAGAIN) {
osi_Log0(afsd_logp, "cm_Analyze passed EWOULDBLOCK or EAGAIN.");
thrd_Sleep(1000);
retry = 1;
}
+
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
}
- /* if there is nosuchvolume, then we have a situation in which a
- * previously known volume no longer has a set of servers
+ /* if there is nosuchvolume, then we have a situation in which a
+ * previously known volume no longer has a set of servers
* associated with it. Either the volume has moved or
* the volume has been deleted. Try to find a new server list
* until the timeout period expires.
*/
else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
- /*
+ /*
* The VNOVOL or VL_NOENT error has already been translated
* to CM_ERROR_NOSUCHVOLUME. There is nothing for us to do.
*/
}
+ else if (errorCode == CM_ERROR_EMPTY) {
+ /*
+ * The server list is empty (or all entries have been deleted).
+ * If fidp is NULL, this was a vlServer list and we can attempt
+ * to force a cell lookup. If fidp is not NULL, we can attempt
+ * to refresh the volume location list.
+ */
+ if (fidp) {
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ &volp);
+ if (code == 0) {
+ lock_ObtainWrite(&volp->rw);
+ if (cm_UpdateVolumeLocation(cellp, userp, reqp, volp) == 0) {
+ lock_ReleaseWrite(&volp->rw);
+ code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
+ if (code == 0) {
+ if (!cm_IsServerListEmpty(*volServerspp))
+ retry = 1;
+ cm_FreeServerList(volServerspp, 0);
+ }
+ } else {
+ lock_ReleaseWrite(&volp->rw);
+ }
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
+ } else {
+ cm_cell_t * newCellp = cm_UpdateCell( cellp, 0);
+ if (newCellp)
+ retry = 1;
+ }
+ }
else if (errorCode == CM_ERROR_ALLDOWN) {
/* Servers marked DOWN will be restored by the background daemon
- * thread as they become available. The volume status is
+ * thread as they become available. The volume status is
* updated as the server state changes.
*/
if (fidp) {
osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
}
}
-
else if (errorCode == CM_ERROR_ALLOFFLINE) {
- /* Volume instances marked offline will be restored by the
- * background daemon thread as they become available
+ /* Volume instances marked offline will be restored by the
+ * background daemon thread as they become available
*/
if (fidp) {
osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
format = "All servers are offline when accessing cell %s volume %d.";
LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
- code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
if (code == 0) {
- if (timeLeft > 7) {
+ if (volServerspp == NULL) {
+ code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
+ if (code == 0)
+ free_svr_list = 1;
+ }
+ if (volServerspp)
+ cm_ResetServerBusyStatus(volServerspp);
+ if (free_svr_list) {
+ cm_FreeServerList(volServerspp, 0);
+ free_svr_list = 0;
+ volServerspp = NULL;
+ }
+
+ /*
+ * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
+ * was called by cm_CheckOfflineVolumeState().
+ */
+ if (!(reqp->flags & (CM_REQ_OFFLINE_VOL_CHK|CM_REQ_NORETRY)) &&
+ timeLeft > 7)
+ {
thrd_Sleep(5000);
/* cm_CheckOfflineVolume() resets the serverRef state */
}
}
else if (errorCode == CM_ERROR_ALLBUSY) {
- /* Volumes that are busy cannot be determined to be non-busy
+ /* Volumes that are busy cannot be determined to be non-busy
* without actually attempting to access them.
*/
if (fidp) { /* File Server query */
format = "All servers are busy when accessing cell %s volume %d.";
LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
- code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
if (code == 0) {
- if (timeLeft > 7) {
- thrd_Sleep(5000);
-
- statep = cm_VolumeStateByID(volp, fidp->volume);
- if (statep->state != vl_offline &&
- statep->state != vl_busy &&
- statep->state != vl_unknown) {
- retry = 1;
- } else {
- if (!serversp) {
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
- if (code == 0) {
- serversp = *serverspp;
- free_svr_list = 1;
- }
- }
- lock_ObtainWrite(&cm_serverLock);
- for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
- if (tsrp->status == srv_deleted)
- continue;
- if (tsrp->status == srv_busy) {
- tsrp->status = srv_not_busy;
- }
- }
- lock_ReleaseWrite(&cm_serverLock);
- if (free_svr_list) {
- cm_FreeServerList(serverspp, 0);
- serversp = NULL;
- free_svr_list = 0;
- }
+ if (volServerspp == NULL) {
+ code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
+ if (code == 0)
+ free_svr_list = 1;
+ }
+ if (volServerspp)
+ cm_ResetServerBusyStatus(volServerspp);
+ if (free_svr_list) {
+ cm_FreeServerList(volServerspp, 0);
+ free_svr_list = 0;
+ volServerspp = NULL;
+ }
- cm_UpdateVolumeStatus(volp, fidp->volume);
- retry = 1;
- }
- } else {
- cm_UpdateVolumeStatus(volp, fidp->volume);
+ /*
+ * retry all replicas for 5 minutes waiting 15 seconds
+ * between attempts.
+ */
+ if (timeLeft > 20 && !(reqp->flags & CM_REQ_NORETRY) &&
+ reqp->volbusyCount++ < 20)
+ {
+ thrd_Sleep(15000);
+ retry = 1;
}
+ cm_UpdateVolumeStatus(volp, fidp->volume);
lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
} else { /* VL Server query */
osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
- if (timeLeft > 7) {
+ if (timeLeft > 7 && !(reqp->flags & CM_REQ_NORETRY) && vlServerspp)
+ {
thrd_Sleep(5000);
- if (serversp) {
- lock_ObtainWrite(&cm_serverLock);
- for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
- if (tsrp->status == srv_deleted)
- continue;
- if (tsrp->status == srv_busy) {
- tsrp->status = srv_not_busy;
- }
- }
- lock_ReleaseWrite(&cm_serverLock);
- retry = 1;
- }
+ cm_ResetServerBusyStatus(vlServerspp);
+ retry = 1;
}
}
}
/* special codes: VBUSY and VRESTARTING */
else if (errorCode == VBUSY || errorCode == VRESTARTING) {
- if (!serversp && fidp) {
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
+ if (fidp) {
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ &volp);
if (code == 0) {
- serversp = *serverspp;
- free_svr_list = 1;
- }
- }
+ if (volServerspp == NULL) {
+ code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &volServerspp);
+ if (code == 0)
+ free_svr_list = 1;
+ }
- switch ( errorCode ) {
- case VBUSY:
- msgID = MSG_SERVER_REPORTS_VBUSY;
- format = "Server %s reported busy when accessing volume %d in cell %s.";
- break;
- case VRESTARTING:
- msgID = MSG_SERVER_REPORTS_VRESTARTING;
- format = "Server %s reported restarting when accessing volume %d in cell %s.";
- break;
+ statep = cm_VolumeStateByID(volp, fidp->volume);
+
+ if (statep)
+ cm_UpdateVolumeStatus(volp, statep->ID);
+
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
}
- if (serverp && fidp) {
+ if (serverp) {
/* Log server being offline for this volume */
sprintf(addr, "%d.%d.%d.%d",
- ((serverp->addr.sin_addr.s_addr & 0xff)),
- ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
- ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
- ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
-
- osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
- LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
- }
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
- lock_ObtainWrite(&cm_serverLock);
- for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
- if (tsrp->status == srv_deleted)
- continue;
- if (tsrp->server == serverp && tsrp->status == srv_not_busy) {
- tsrp->status = srv_busy;
- if (fidp) { /* File Server query */
- lock_ReleaseWrite(&cm_serverLock);
- code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
- &volp);
- if (code == 0)
- statep = cm_VolumeStateByID(volp, fidp->volume);
- lock_ObtainWrite(&cm_serverLock);
+ switch ( errorCode ) {
+ case VBUSY:
+ if (invalid_status) {
+ msgID = MSG_SERVER_REPLIED_BAD_STATUS;
+ format = "Server %s replied with bad status info when accessing volume %d in cell %s. Data discarded by cache manager.";
+ } else {
+ msgID = MSG_SERVER_REPORTS_VBUSY;
+ format = "Server %s reported busy when accessing volume %d in cell %s.";
}
break;
+ case VRESTARTING:
+ msgID = MSG_SERVER_REPORTS_VRESTARTING;
+ format = "Server %s reported restarting when accessing volume %d in cell %s.";
+ break;
}
- }
- lock_ReleaseWrite(&cm_serverLock);
-
- if (statep) {
- cm_UpdateVolumeStatus(volp, statep->ID);
- lock_ObtainRead(&cm_volumeLock);
- cm_PutVolume(volp);
- lock_ReleaseRead(&cm_volumeLock);
- volp = NULL;
+
+ osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
+
+ if (volServerspp)
+ cm_SetServerBusyStatus(volServerspp, serverp);
}
if (free_svr_list) {
- cm_FreeServerList(serverspp, 0);
- serversp = NULL;
+ cm_FreeServerList(volServerspp, 0);
+ volServerspp = NULL;
free_svr_list = 0;
}
retry = 1;
/* special codes: missing volumes */
else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
- errorCode == VSALVAGE || errorCode == VNOSERVICE || errorCode == VIO)
- {
+ errorCode == VSALVAGE || errorCode == VIO)
+ {
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
/* In case of timeout */
reqp->volumeError = errorCode;
msgID = MSG_SERVER_REPORTS_VSALVAGE;
format = "Server %s reported volume %d in cell %s as needs salvage.";
break;
- case VNOSERVICE:
- msgID = MSG_SERVER_REPORTS_VNOSERVICE;
- format = "Server %s reported volume %d in cell %s as not in service.";
- break;
case VIO:
msgID = MSG_SERVER_REPORTS_VIO;
format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
!(reqp->flags & CM_REQ_VOLUME_UPDATED))
{
+ LONG_PTR oldSum, newSum;
+
+ oldSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
+
code = cm_ForceUpdateVolume(fidp, userp, reqp);
- if (code == 0)
+ if (code == 0) {
location_updated = 1;
+ newSum = cm_ChecksumVolumeServerList(fidp, userp, reqp);
+ }
- /* Even if the update fails, there might still be another replica */
+ /*
+ * Even if the update fails, there might still be another replica.
+ * If the volume location list changed, permit another update on
+ * a subsequent error.
+ */
+ if (code || oldSum == newSum)
+ reqp->flags |= CM_REQ_VOLUME_UPDATED;
- reqp->flags |= CM_REQ_VOLUME_UPDATED;
osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
- fidp->cell, fidp->volume, code);
+ fidp->cell, fidp->volume, code);
}
if (statep) {
* Mark server offline for this volume or delete the volume
* from the server list if it was moved or is not present.
*/
- if (!serversp || location_updated) {
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
- if (code == 0) {
- serversp = *serverspp;
+ if (volServerspp == NULL || location_updated) {
+ code = cm_GetServerList(fidp, userp, reqp, &replicated, &volServerspp);
+ if (code == 0)
free_svr_list = 1;
- }
}
}
- lock_ObtainWrite(&cm_serverLock);
- for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
- if (tsrp->status == srv_deleted)
- continue;
- sprintf(addr, "%d.%d.%d.%d",
- ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
- ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
- ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
- ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
-
- if (cm_ServerEqual(tsrp->server, serverp)) {
- /* REDIRECT */
- if (errorCode == VMOVED || errorCode == VNOVOL) {
- osi_Log2(afsd_logp, "volume %d not present on server %s",
- fidp->volume, osi_LogSaveString(afsd_logp,addr));
- tsrp->status = srv_deleted;
- if (fidp)
- cm_RemoveVolumeFromServer(serverp, fidp->volume);
- } else {
- osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
- fidp->volume, osi_LogSaveString(afsd_logp,addr));
- tsrp->status = srv_offline;
+ if (volServerspp) {
+ lock_ObtainWrite(&cm_serverLock);
+ for (tsrp = *volServerspp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
+
+ sprintf(addr, "%d.%d.%d.%d",
+ ((tsrp->server->addr.sin_addr.s_addr & 0xff)),
+ ((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ if (cm_ServerEqual(tsrp->server, serverp)) {
+ /* REDIRECT */
+ switch (errorCode) {
+ case VMOVED:
+ osi_Log2(afsd_logp, "volume %u moved from server %s",
+ fidp->volume, osi_LogSaveString(afsd_logp,addr));
+ tsrp->status = srv_deleted;
+ if (fidp)
+ cm_RemoveVolumeFromServer(serverp, fidp->volume);
+ break;
+ case VNOVOL:
+ /*
+ * The 1.6.0 and 1.6.1 file servers send transient VNOVOL errors which
+ * are no indicative of the volume not being present. For example,
+ * VNOVOL can be sent during a transition to a VBUSY state prior to
+ * salvaging or when cloning a .backup volume instance. As a result
+ * the cache manager must attempt at least one retry when a VNOVOL is
+ * receive but there are no changes to the volume location information.
+ */
+ if (reqp->vnovolError > 0 && cm_ServerEqual(reqp->errorServp, serverp)) {
+ osi_Log2(afsd_logp, "volume %u not present on server %s",
+ fidp->volume, osi_LogSaveString(afsd_logp,addr));
+ tsrp->status = srv_deleted;
+ if (fidp)
+ cm_RemoveVolumeFromServer(serverp, fidp->volume);
+ } else {
+ osi_Log2(afsd_logp, "VNOVOL received for volume %u from server %s",
+ fidp->volume, osi_LogSaveString(afsd_logp,addr));
+ if (replicated) {
+ if (tsrp->status == srv_not_busy)
+ tsrp->status = srv_busy;
+ } else {
+ Sleep(2000);
+ }
+ }
+ break;
+ case VOFFLINE:
+ osi_Log2(afsd_logp, "VOFFLINE received for volume %u from server %s",
+ fidp->volume, osi_LogSaveString(afsd_logp,addr));
+ tsrp->status = srv_offline;
+ break;
+ default:
+ osi_Log3(afsd_logp, "volume %u exists on server %s with status %u",
+ fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
+ }
}
- /* break; */
- } else {
- osi_Log3(afsd_logp, "volume %d exists on server %s with status %u",
- fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
}
+ lock_ReleaseWrite(&cm_serverLock);
+ }
+
+ /* Remember that the VNOVOL error occurred */
+ if (errorCode == VNOVOL) {
+ reqp->errorServp = serverp;
+ reqp->vnovolError++;
+ }
+
+ /* Remember that the VIO error occurred */
+ if (errorCode == VIO) {
+ reqp->errorServp = serverp;
+ reqp->vioCount++;
}
- lock_ReleaseWrite(&cm_serverLock);
/* Free the server list before cm_ForceUpdateVolume is called */
if (free_svr_list) {
- cm_FreeServerList(serverspp, 0);
- serversp = NULL;
+ cm_FreeServerList(volServerspp, 0);
+ volServerspp = NULL;
free_svr_list = 0;
}
- if ( timeLeft > 2 )
+ if ( timeLeft > 2 && reqp->vioCount < 100)
retry = 1;
} else if ( errorCode == VNOVNODE ) {
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
if ( fidp ) {
- cm_scache_t * scp;
osi_Log4(afsd_logp, "cm_Analyze passed VNOVNODE cell %u vol %u vn %u uniq %u.",
fidp->cell, fidp->volume, fidp->vnode, fidp->unique);
pscp = cm_FindSCacheParent(scp);
lock_ObtainWrite(&scp->rw);
+ _InterlockedOr(&scp->flags, CM_SCACHEFLAG_DELETED);
lock_ObtainWrite(&cm_scacheLock);
- cm_RemoveSCacheFromHashTable(scp);
+ cm_AdjustScacheLRU(scp);
lock_ReleaseWrite(&cm_scacheLock);
cm_LockMarkSCacheLost(scp);
- scp->flags |= CM_SCACHEFLAG_DELETED;
lock_ReleaseWrite(&scp->rw);
+ if (RDR_Initialized)
+ RDR_InvalidateObject(scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique,
+ scp->fid.hash, scp->fileType, AFS_INVALIDATE_DELETED);
cm_ReleaseSCache(scp);
if (pscp) {
if (cm_HaveCallback(pscp)) {
lock_ObtainWrite(&pscp->rw);
cm_DiscardSCache(pscp);
- lock_ReleaseWrite(&pscp->rw);
- }
- cm_ReleaseSCache(pscp);
+ lock_ReleaseWrite(&pscp->rw);
+
+ if (RDR_Initialized)
+ RDR_InvalidateObject(pscp->fid.cell, pscp->fid.volume, pscp->fid.vnode, pscp->fid.unique,
+ pscp->fid.hash, pscp->fileType, AFS_INVALIDATE_EXPIRED);
+
+ }
+ cm_ReleaseSCache(pscp);
}
}
} else {
* reported idle for longer than idleDeadTime
* don't mark server as down but don't retry
* this is to prevent the SMB session from timing out
- * In addition, we log an event to the event log
+ * In addition, we log an event to the event log
*/
if (serverp) {
- sprintf(addr, "%d.%d.%d.%d",
+ sprintf(addr, "%d.%d.%d.%d",
((serverp->addr.sin_addr.s_addr & 0xff)),
((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
- ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime or idleDeadtime exceeded addr[%s]",
osi_LogSaveString(afsd_logp,addr));
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->idleError++;
}
+ if (fidp && storeOp)
+ scp = cm_FindSCache(fidp);
+ if (scp) {
+ if (cm_HaveCallback(scp)) {
+ lock_ObtainWrite(&scp->rw);
+ cm_DiscardSCache(scp);
+ lock_ReleaseWrite(&scp->rw);
+
+ /*
+ * We really should notify the redirector that we discarded
+ * the status information but doing so in this case is not
+ * safe as it can result in a deadlock with extent release
+ * processing.
+ */
+ }
+ cm_ReleaseSCache(scp);
+ }
+
if (timeLeft > 2) {
if (!fidp) { /* vldb */
retry = 1;
} else { /* file */
- cm_volume_t *volp = cm_GetVolumeByFID(fidp);
+ cm_volume_t *volp = cm_FindVolumeByFID(fidp, userp, reqp);
if (volp) {
if (fidp->volume == cm_GetROVolumeID(volp))
retry = 1;
* with a smaller mtu size.
*/
- if (serverp) {
+ if (serverp)
sprintf(addr, "%d.%d.%d.%d",
((serverp->addr.sin_addr.s_addr & 0xff)),
((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
- LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
- osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
- osi_LogSaveString(afsd_logp,addr));
+ LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
+ osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
+ osi_LogSaveString(afsd_logp,addr));
+
+ retry = 2;
+ }
+ else if (errorCode == RX_CALL_BUSY) {
+ /*
+ * RPC failed because the selected call channel
+ * is currently busy on the server. Unconditionally
+ * retry the request so an alternate call channel can be used.
+ */
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
+ if (serverp)
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
+ osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
+ osi_LogSaveString(afsd_logp,addr));
+ retry = 2;
+ }
+ else if (errorCode == VNOSERVICE) {
+ /*
+ * The server did not service the RPC.
+ * If this was a file server RPC it means that for at
+ * least the file server's idle dead timeout period the
+ * file server did not receive any new data packets from
+ * client.
+ *
+ * The RPC was not serviced so it can be retried and any
+ * existing status information is still valid.
+ */
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
+ if (fidp) {
+ if (serverp)
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ LogEvent(EVENTLOG_WARNING_TYPE, MSG_SERVER_REPORTS_VNOSERVICE,
+ addr, fidp->volume, cellp->name);
+ osi_Log3(afsd_logp, "Server %s reported rpc to volume %d in cell %s as not serviced.",
+ osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
}
- retry = 1;
+ retry = 2;
}
- else if (errorCode >= -64 && errorCode < 0) {
+ else if (errorCode == RX_CALL_IDLE) {
+ /*
+ * RPC failed because the server failed to respond with data
+ * within the idle dead timeout period. This could be for a variety
+ * of reasons:
+ * 1. The server could have a bad partition such as a failed
+ * disk or iSCSI target and all I/O to that partition is
+ * blocking on the server and will never complete.
+ *
+ * 2. The server vnode may be locked by another client request
+ * that is taking a very long time.
+ *
+ * 3. The server may have a very long queue of requests
+ * pending and is unable to process this request.
+ *
+ * 4. The server could be malicious and is performing a denial
+ * of service attack against the client.
+ *
+ * If this is a request against a .readonly with alternate sites
+ * the server should be marked down for this request and the
+ * client should fail over to another server. If this is a
+ * request against a single source, the client may retry once.
+ */
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
+ if (serverp)
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ if (fidp) {
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ &volp);
+ if (code == 0) {
+ statep = cm_VolumeStateByID(volp, fidp->volume);
+
+ if (statep)
+ replicated = (statep->flags & CM_VOL_STATE_FLAG_REPLICATED);
+
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
+
+ if (storeOp)
+ scp = cm_FindSCache(fidp);
+ if (scp) {
+ if (cm_HaveCallback(scp)) {
+ lock_ObtainWrite(&scp->rw);
+ cm_DiscardSCache(scp);
+ lock_ReleaseWrite(&scp->rw);
+
+ /*
+ * We really should notify the redirector that we discarded
+ * the status information but doing so in this case is not
+ * safe as it can result in a deadlock with extent release
+ * processing.
+ */
+ }
+ cm_ReleaseSCache(scp);
+ }
+ }
+
+ if (replicated && serverp) {
+ reqp->errorServp = serverp;
+ reqp->tokenError = errorCode;
+
+ retry = 2;
+ }
+
+ LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_IDLE_DEAD_TIMEOUT, addr, retry);
+ osi_Log2(afsd_logp, "cm_Analyze: RPC failed due to idle dead timeout addr[%s] retry=%u",
+ osi_LogSaveString(afsd_logp,addr), retry);
+ }
+ else if (errorCode == RX_CALL_DEAD) {
/* mark server as down */
if (serverp)
sprintf(addr, "%d.%d.%d.%d",
((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
- if (errorCode == RX_CALL_DEAD)
- osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
- osi_LogSaveString(afsd_logp,addr),
- (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
- else
- osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
- errorCode,
- osi_LogSaveString(afsd_logp,addr),
- (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
+ osi_Log2(afsd_logp, "cm_Analyze: Rx Call Dead addr[%s] forcedNew[%s]",
+ osi_LogSaveString(afsd_logp,addr),
+ (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
if (serverp) {
- lock_ObtainMutex(&serverp->mx);
- if (errorCode == RX_CALL_DEAD &&
- (reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
+ if ((connp->flags & CM_CONN_FLAG_NEW) ||
+ (reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
+ lock_ObtainMutex(&serverp->mx);
if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
- serverp->flags |= CM_SERVERFLAG_DOWN;
+ _InterlockedOr(&serverp->flags, CM_SERVERFLAG_DOWN);
serverp->downTime = time(NULL);
}
+ lock_ReleaseMutex(&serverp->mx);
} else {
- if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
- reqp->tokenIdleErrorServp = serverp;
- reqp->tokenError = errorCode;
- } else {
- reqp->flags |= CM_REQ_NEW_CONN_FORCED;
- forcing_new = 1;
- }
+ reqp->flags |= CM_REQ_NEW_CONN_FORCED;
+ forcing_new = 1;
+ cm_ForceNewConnections(serverp);
}
- lock_ReleaseMutex(&serverp->mx);
- cm_ForceNewConnections(serverp);
}
+
+ if (fidp && storeOp)
+ scp = cm_FindSCache(fidp);
+ if (scp) {
+ if (cm_HaveCallback(scp)) {
+ lock_ObtainWrite(&scp->rw);
+ cm_DiscardSCache(scp);
+ lock_ReleaseWrite(&scp->rw);
+
+ /*
+ * We really should notify the redirector that we discarded
+ * the status information but doing so in this case is not
+ * safe as it can result in a deadlock with extent release
+ * processing.
+ */
+ }
+ cm_ReleaseSCache(scp);
+ }
+
if ( timeLeft > 2 )
retry = 1;
}
+ else if (errorCode >= -64 && errorCode < 0) {
+ /* mark server as down */
+ if (serverp)
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ osi_Log3(afsd_logp, "cm_Analyze: Rx Misc Error[%d] addr[%s] forcedNew[%s]",
+ errorCode,
+ osi_LogSaveString(afsd_logp,addr),
+ (reqp->flags & CM_REQ_NEW_CONN_FORCED ? "yes" : "no"));
+
+ if (serverp) {
+ if ((connp->flags & CM_CONN_FLAG_NEW) ||
+ (reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
+ reqp->errorServp = serverp;
+ reqp->tokenError = errorCode;
+ } else {
+ reqp->flags |= CM_REQ_NEW_CONN_FORCED;
+ forcing_new = 1;
+ cm_ForceNewConnections(serverp);
+ }
+
+ if ( timeLeft > 2 )
+ retry = 1;
+ }
+ }
else if (errorCode == RXKADEXPIRED) {
osi_Log1(afsd_logp, "cm_Analyze: rxkad error code 0x%x (RXKADEXPIRED)",
errorCode);
if (!dead_session) {
lock_ObtainMutex(&userp->mx);
- ucellp = cm_GetUCell(userp, serverp->cellp);
+ ucellp = cm_GetUCell(userp, cellp);
if (ucellp->ticketp) {
free(ucellp->ticketp);
ucellp->ticketp = NULL;
}
- ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
+ _InterlockedAnd(&ucellp->flags, ~CM_UCELLFLAG_RXKAD);
ucellp->gen++;
lock_ReleaseMutex(&userp->mx);
- if ( timeLeft > 2 )
- retry = 1;
+
+ reqp->flags |= CM_REQ_NEW_CONN_FORCED;
+ forcing_new = 1;
+ cm_ForceNewConnections(serverp);
+
+ if ( timeLeft > 2 )
+ retry = 2;
}
} else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
char * s = "unknown error";
osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
errorCode, s);
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
if (serverp) {
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->tokenError = errorCode;
retry = 1;
}
* to answer our query. Therefore, we will retry the request
* and force the use of another server.
*/
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
if (serverp) {
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->tokenError = errorCode;
retry = 1;
}
} else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
- cm_ForceNewConnections(serverp);
- if ( timeLeft > 2 )
- retry = 1;
+ reqp->flags |= CM_REQ_NEW_CONN_FORCED;
+ forcing_new = 1;
+ cm_ForceNewConnections(serverp);
+ retry = 2;
} else {
+ if (connp)
+ _InterlockedAnd(&connp->flags, ~CM_CONN_FLAG_NEW);
+
if (errorCode) {
char * s = "unknown error";
switch ( errorCode ) {
case VL_INDEXERANGE : s = "VL_INDEXERANGE"; break;
case VL_MULTIPADDR : s = "VL_MULTIPADDR"; break;
case VL_BADMASK : s = "VL_BADMASK"; break;
- case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
- case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
- case CM_ERROR_TIMEDOUT : s = "CM_ERROR_TIMEDOUT"; break;
- case CM_ERROR_RETRY : s = "CM_ERROR_RETRY"; break;
- case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
- case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
- case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
- case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
- case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
- case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
- case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
- case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
- case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
- case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
- case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
- case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
- case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
- case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
- case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
- case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
- case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
- case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
- case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
- case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
- case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
- case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
- case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
- case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
- case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
- case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
- case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
- case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
- case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
- case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
- case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
- case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
- case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
- case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
- case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
- case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
- case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
- case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
- case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
- case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
- case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
- case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
- case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
- case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
- case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
- case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
- case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
+ case CM_ERROR_NOSUCHCELL : s = "CM_ERROR_NOSUCHCELL"; break;
+ case CM_ERROR_NOSUCHVOLUME : s = "CM_ERROR_NOSUCHVOLUME"; break;
+ case CM_ERROR_NOACCESS : s = "CM_ERROR_NOACCESS"; break;
+ case CM_ERROR_NOSUCHFILE : s = "CM_ERROR_NOSUCHFILE"; break;
+ case CM_ERROR_STOPNOW : s = "CM_ERROR_STOPNOW"; break;
+ case CM_ERROR_TOOBIG : s = "CM_ERROR_TOOBIG"; break;
+ case CM_ERROR_INVAL : s = "CM_ERROR_INVAL"; break;
+ case CM_ERROR_BADFD : s = "CM_ERROR_BADFD"; break;
+ case CM_ERROR_BADFDOP : s = "CM_ERROR_BADFDOP"; break;
+ case CM_ERROR_EXISTS : s = "CM_ERROR_EXISTS"; break;
+ case CM_ERROR_CROSSDEVLINK : s = "CM_ERROR_CROSSDEVLINK"; break;
+ case CM_ERROR_BADOP : s = "CM_ERROR_BADOP"; break;
+ case CM_ERROR_BADPASSWORD : s = "CM_ERROR_BADPASSWORD"; break;
+ case CM_ERROR_NOTDIR : s = "CM_ERROR_NOTDIR"; break;
+ case CM_ERROR_ISDIR : s = "CM_ERROR_ISDIR"; break;
+ case CM_ERROR_READONLY : s = "CM_ERROR_READONLY"; break;
+ case CM_ERROR_WOULDBLOCK : s = "CM_ERROR_WOULDBLOCK"; break;
+ case CM_ERROR_QUOTA : s = "CM_ERROR_QUOTA"; break;
+ case CM_ERROR_SPACE : s = "CM_ERROR_SPACE"; break;
+ case CM_ERROR_BADSHARENAME : s = "CM_ERROR_BADSHARENAME"; break;
+ case CM_ERROR_BADTID : s = "CM_ERROR_BADTID"; break;
+ case CM_ERROR_UNKNOWN : s = "CM_ERROR_UNKNOWN"; break;
+ case CM_ERROR_NOMORETOKENS : s = "CM_ERROR_NOMORETOKENS"; break;
+ case CM_ERROR_NOTEMPTY : s = "CM_ERROR_NOTEMPTY"; break;
+ case CM_ERROR_USESTD : s = "CM_ERROR_USESTD"; break;
+ case CM_ERROR_REMOTECONN : s = "CM_ERROR_REMOTECONN"; break;
+ case CM_ERROR_ATSYS : s = "CM_ERROR_ATSYS"; break;
+ case CM_ERROR_NOSUCHPATH : s = "CM_ERROR_NOSUCHPATH"; break;
+ case CM_ERROR_CLOCKSKEW : s = "CM_ERROR_CLOCKSKEW"; break;
+ case CM_ERROR_BADSMB : s = "CM_ERROR_BADSMB"; break;
+ case CM_ERROR_ALLBUSY : s = "CM_ERROR_ALLBUSY"; break;
+ case CM_ERROR_NOFILES : s = "CM_ERROR_NOFILES"; break;
+ case CM_ERROR_PARTIALWRITE : s = "CM_ERROR_PARTIALWRITE"; break;
+ case CM_ERROR_NOIPC : s = "CM_ERROR_NOIPC"; break;
+ case CM_ERROR_BADNTFILENAME : s = "CM_ERROR_BADNTFILENAME"; break;
+ case CM_ERROR_BUFFERTOOSMALL : s = "CM_ERROR_BUFFERTOOSMALL"; break;
+ case CM_ERROR_RENAME_IDENTICAL : s = "CM_ERROR_RENAME_IDENTICAL"; break;
+ case CM_ERROR_ALLOFFLINE : s = "CM_ERROR_ALLOFFLINE"; break;
+ case CM_ERROR_AMBIGUOUS_FILENAME: s = "CM_ERROR_AMBIGUOUS_FILENAME"; break;
+ case CM_ERROR_BADLOGONTYPE : s = "CM_ERROR_BADLOGONTYPE"; break;
+ case CM_ERROR_GSSCONTINUE : s = "CM_ERROR_GSSCONTINUE"; break;
+ case CM_ERROR_TIDIPC : s = "CM_ERROR_TIDIPC"; break;
+ case CM_ERROR_TOO_MANY_SYMLINKS : s = "CM_ERROR_TOO_MANY_SYMLINKS"; break;
+ case CM_ERROR_PATH_NOT_COVERED : s = "CM_ERROR_PATH_NOT_COVERED"; break;
+ case CM_ERROR_LOCK_CONFLICT : s = "CM_ERROR_LOCK_CONFLICT"; break;
+ case CM_ERROR_SHARING_VIOLATION : s = "CM_ERROR_SHARING_VIOLATION"; break;
+ case CM_ERROR_ALLDOWN : s = "CM_ERROR_ALLDOWN"; break;
+ case CM_ERROR_TOOFEWBUFS : s = "CM_ERROR_TOOFEWBUFS"; break;
+ case CM_ERROR_TOOMANYBUFS : s = "CM_ERROR_TOOMANYBUFS"; break;
+ case UAEIO : s = "UAEIO"; break;
+ case EIO : s = "EIO"; break;
}
- osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
+ osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
errorCode, s);
retry = 0;
}
}
/* If not allowed to retry, don't */
- if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) && (errorCode != RX_MSGSIZE))
- retry = 0;
- else if (retry && dead_session)
+ if (dead_session ||
+ !forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
+ !(errorCode > -64 && errorCode <= RX_INVALID_OPERATION))
retry = 0;
/* drop this on the way out */
if (connp)
cm_PutConn(connp);
- /*
+ /*
+
* clear the volume updated flag if we succeed.
- * this way the flag will not prevent a subsequent volume
+ * this way the flag will not prevent a subsequent volume
* from being updated if necessary.
*/
if (errorCode == 0)
reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
}
+ if ( vlServerspp &&
+ errorCode != VBUSY &&
+ errorCode != VRESTARTING &&
+ errorCode != CM_ERROR_ALLBUSY)
+ {
+ cm_ResetServerBusyStatus(vlServerspp);
+ }
+
/* retry until we fail to find a connection */
return retry;
}
-long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
- cm_req_t *reqp, cm_conn_t **connpp)
+long cm_ConnByMServers(cm_serverRef_t *serversp, afs_uint32 replicated, cm_user_t *usersp,
+ cm_req_t *reqp, cm_conn_t **connpp)
{
long code;
cm_serverRef_t *tsrp;
cm_server_t *tsp;
long firstError = 0;
- int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
+ int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1, allDeleted = 1;
#ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
long timeUsed, timeLeft, hardTimeLeft;
#endif
*connpp = NULL;
if (serversp == NULL) {
- osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_ALLDOWN);
- return CM_ERROR_ALLDOWN;
+ osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", CM_ERROR_EMPTY);
+ return CM_ERROR_EMPTY;
}
#ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
timeUsed = (GetTickCount() - reqp->startTime) / 1000;
-
+
/* leave 5 seconds margin of safety */
timeLeft = ConnDeadtimeout - timeUsed - 5;
hardTimeLeft = HardDeadtimeout - timeUsed - 5;
if (tsrp->status == srv_deleted)
continue;
+ allDeleted = 0;
+
tsp = tsrp->server;
- if (reqp->tokenIdleErrorServp) {
- /*
+ if (reqp->errorServp) {
+ /*
* search the list until we find the server
* that failed last time. When we find it
- * clear the error, skip it and try the one
+ * clear the error, skip it and try the next one
* in the list.
*/
- if (tsp == reqp->tokenIdleErrorServp)
- reqp->tokenIdleErrorServp = NULL;
+ if (tsp == reqp->errorServp)
+ reqp->errorServp = NULL;
continue;
}
if (tsp) {
} else {
allOffline = 0;
allBusy = 0;
- code = cm_ConnByServer(tsp, usersp, connpp);
+ code = cm_ConnByServer(tsp, usersp, replicated, connpp);
if (code == 0) { /* cm_CBS only returns 0 */
cm_PutServer(tsp);
#ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
if (timeLeft > ConnDeadtimeout)
timeLeft = ConnDeadtimeout;
- if (hardTimeLeft > HardDeadtimeout)
+ if (hardTimeLeft > HardDeadtimeout)
hardTimeLeft = HardDeadtimeout;
lock_ObtainMutex(&(*connpp)->mx);
lock_ObtainRead(&cm_serverLock);
cm_PutServerNoLock(tsp);
}
- }
+ }
lock_ReleaseRead(&cm_serverLock);
if (firstError == 0) {
- if (allDown) {
- firstError = (reqp->tokenError ? reqp->tokenError :
+ if (allDeleted) {
+ firstError = CM_ERROR_EMPTY;
+ } else if (allDown) {
+ firstError = (reqp->tokenError ? reqp->tokenError :
(reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
/*
* if we experienced either a token error or and idle dead time error
}
static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
- cm_server_t *serverp)
+ cm_server_t *serverp, afs_uint32 replicated)
{
unsigned short port;
int serviceID;
}
secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
&ucellp->sessionKey, ucellp->kvno,
- ucellp->ticketLen, ucellp->ticketp);
+ ucellp->ticketLen, ucellp->ticketp);
} else {
/* normal auth */
secIndex = 0;
secObjp,
secIndex);
rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
- rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
- rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
+ if (smb_Enabled || tcp->serverp->type == CM_SERVER_VLDB)
+ rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
/*
- * Register the error to be returned on an idle dead timeout
+ * Setting idle dead timeout to a non-zero value activates RX_CALL_IDLE errors
*/
- rx_SetServerConnIdleDeadErr(tcp->rxconnp, RX_CALL_DEAD);
+ if (replicated) {
+ _InterlockedOr(&tcp->flags, CM_CONN_FLAG_REPLICATION);
+ rx_SetConnIdleDeadTime(tcp->rxconnp, ReplicaIdleDeadtimeout);
+ } else {
+ rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
+ }
/*
* Let the Rx library know that we can auto-retry if an
* Only file servers implement client callbacks and we only need one ping
* to be sent to each server.
*/
- if (NatPingInterval && serverp->type == CM_SERVER_FILE && secIndex == 0)
+ if (NatPingInterval && serverp->type == CM_SERVER_FILE &&
+ (ucellp->flags & CM_UCELLFLAG_ROOTUSER)) {
rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
+ }
tcp->ucgen = ucellp->gen;
if (secObjp)
rxs_Release(secObjp); /* Decrement the initial refCount */
+
+ _InterlockedAnd(&tcp->flags, ~CM_CONN_FLAG_FORCE_NEW);
+ _InterlockedOr(&tcp->flags, CM_CONN_FLAG_NEW);
}
-long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
+long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, afs_uint32 replicated, cm_conn_t **connpp)
{
cm_conn_t *tcp;
cm_ucell_t *ucellp;
userp = cm_rootUserp;
lock_ObtainMutex(&userp->mx);
+ /* find ucell structure */
+ ucellp = cm_GetUCell(userp, serverp->cellp);
+
lock_ObtainRead(&cm_connLock);
for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
- if (tcp->userp == userp)
+ if (tcp->userp == userp &&
+ (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
+ !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
break;
}
-
- /* find ucell structure */
- ucellp = cm_GetUCell(userp, serverp->cellp);
+
if (!tcp) {
lock_ConvertRToW(&cm_connLock);
for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
- if (tcp->userp == userp)
- break;
+ if (tcp->userp == userp &&
+ (replicated && (tcp->flags & CM_CONN_FLAG_REPLICATION) ||
+ !replicated && !(tcp->flags & CM_CONN_FLAG_REPLICATION)))
+ break;
}
if (tcp) {
+ InterlockedIncrement(&tcp->refCount);
lock_ReleaseWrite(&cm_connLock);
goto haveconn;
}
cm_GetServer(serverp);
tcp = malloc(sizeof(*tcp));
memset(tcp, 0, sizeof(*tcp));
- tcp->nextp = serverp->connsp;
- serverp->connsp = tcp;
cm_HoldUser(userp);
tcp->userp = userp;
lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
- lock_ObtainMutex(&tcp->mx);
tcp->serverp = serverp;
tcp->cryptlevel = rxkad_clear;
- cm_NewRXConnection(tcp, ucellp, serverp);
+ cm_NewRXConnection(tcp, ucellp, serverp, replicated);
tcp->refCount = 1;
- lock_ReleaseMutex(&tcp->mx);
+ tcp->nextp = serverp->connsp;
+ serverp->connsp = tcp;
lock_ReleaseWrite(&cm_connLock);
+ lock_ReleaseMutex(&userp->mx);
} else {
+ InterlockedIncrement(&tcp->refCount);
lock_ReleaseRead(&cm_connLock);
haveconn:
- InterlockedIncrement(&tcp->refCount);
+ lock_ReleaseMutex(&userp->mx);
lock_ObtainMutex(&tcp->mx);
if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
else
osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
- tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
rx_DestroyConnection(tcp->rxconnp);
- cm_NewRXConnection(tcp, ucellp, serverp);
+ cm_NewRXConnection(tcp, ucellp, serverp, replicated);
}
lock_ReleaseMutex(&tcp->mx);
}
- lock_ReleaseMutex(&userp->mx);
/* return this pointer to our caller */
osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%p", tcp);
cm_serverRef_t *tsrp;
cm_server_t *tsp;
int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
+ afs_uint32 replicated;
cm_InitReq(&req);
- code = cm_GetServerList(fidp, userp, &req, &serverspp);
+ code = cm_GetServerList(fidp, userp, &req, &replicated, &serverspp);
if (code)
return 0;
allBusy = 0;
}
}
- }
+ }
lock_ReleaseRead(&cm_serverLock);
cm_FreeServerList(serverspp, 0);
if (allDown)
return 0;
- else if (allBusy)
+ else if (allBusy)
return 0;
else if (allOffline || (someBusy && someOffline))
return 0;
return 1;
}
-/*
+/*
* The returned cm_conn_t ** object is released in the subsequent call
- * to cm_Analyze().
+ * to cm_Analyze().
*/
long cm_ConnFromFID(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
cm_conn_t **connpp)
{
long code;
cm_serverRef_t **serverspp;
+ afs_uint32 replicated;
*connpp = NULL;
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
- if (code) {
+ code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
+ if (code)
return code;
- }
- code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
+ code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
cm_FreeServerList(serverspp, 0);
return code;
}
{
long code;
cm_serverRef_t **serverspp;
+ afs_uint32 replicated;
*connpp = NULL;
- serverspp = cm_GetVolServers(volp, volid, userp, reqp);
+ code = cm_GetVolServerList(volp, volid, userp, reqp, &replicated, &serverspp);
+ if (code)
+ return code;
- code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
+ code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
cm_FreeServerList(serverspp, 0);
return code;
}
{
cm_conn_t *tcp;
+ if (serverp == NULL)
+ return;
+
lock_ObtainWrite(&cm_connLock);
for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
lock_ObtainMutex(&tcp->mx);
- tcp->flags |= CM_CONN_FLAG_FORCE_NEW;
+ _InterlockedOr(&tcp->flags, CM_CONN_FLAG_FORCE_NEW);
lock_ReleaseMutex(&tcp->mx);
}
lock_ReleaseWrite(&cm_connLock);