DWORD RDRtimeout = CM_CONN_DEFAULTRDRTIMEOUT;
unsigned short ConnDeadtimeout = CM_CONN_CONNDEADTIME;
unsigned short HardDeadtimeout = CM_CONN_HARDDEADTIME;
+unsigned short IdleDeadtimeout = CM_CONN_IDLEDEADTIME;
#define LANMAN_WKS_PARAM_KEY "SYSTEM\\CurrentControlSet\\Services\\lanmanworkstation\\parameters"
#define LANMAN_WKS_SESSION_TIMEOUT "SessTimeout"
+#define LANMAN_WKS_EXT_SESSION_TIMEOUT "ExtendedSessTimeout"
afs_uint32 cryptall = 0;
afs_uint32 cm_anonvldb = 0;
HKEY parmKey;
if (osi_Once(&once)) {
- lock_InitializeRWLock(&cm_connLock, "connection global lock");
+ lock_InitializeRWLock(&cm_connLock, "connection global lock",
+ LOCK_HIERARCHY_CONN_GLOBAL);
/* keisa - read timeout value for lanmanworkstation service.
* jaltman - as per
0, KEY_QUERY_VALUE, &parmKey);
if (code == ERROR_SUCCESS)
{
- dummyLen = sizeof(DWORD);
- code = RegQueryValueEx(parmKey, LANMAN_WKS_SESSION_TIMEOUT, NULL, NULL,
- (BYTE *) &dwValue, &dummyLen);
- if (code == ERROR_SUCCESS)
- RDRtimeout = dwValue;
+ BOOL extTimeouts = msftSMBRedirectorSupportsExtendedTimeouts();
+
+ if (extTimeouts) {
+ /*
+ * The default value is 1000 seconds. However, this default
+ * 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,
+ LANMAN_WKS_EXT_SESSION_TIMEOUT,
+ NULL, NULL,
+ (BYTE *) &dwValue, &dummyLen);
+ if (code == ERROR_SUCCESS) {
+ RDRtimeout = dwValue;
+ afsi_log("lanmanworkstation : ExtSessTimeout %u", RDRtimeout);
+ }
+ }
+ if (!extTimeouts || code != ERROR_SUCCESS) {
+ dummyLen = sizeof(DWORD);
+ code = RegQueryValueEx(parmKey,
+ LANMAN_WKS_SESSION_TIMEOUT,
+ NULL, NULL,
+ (BYTE *) &dwValue, &dummyLen);
+ if (code == ERROR_SUCCESS) {
+ RDRtimeout = dwValue;
+ afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
+ }
+ }
RegCloseKey(parmKey);
}
dummyLen = sizeof(DWORD);
code = RegQueryValueEx(parmKey, "ConnDeadTimeout", NULL, NULL,
(BYTE *) &dwValue, &dummyLen);
- if (code == ERROR_SUCCESS)
+ if (code == ERROR_SUCCESS) {
ConnDeadtimeout = (unsigned short)dwValue;
-
+ afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
+ }
dummyLen = sizeof(DWORD);
code = RegQueryValueEx(parmKey, "HardDeadTimeout", NULL, NULL,
(BYTE *) &dwValue, &dummyLen);
- if (code == ERROR_SUCCESS)
+ if (code == ERROR_SUCCESS) {
HardDeadtimeout = (unsigned short)dwValue;
- afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
- RegCloseKey(parmKey);
+ afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
+ }
+ dummyLen = sizeof(DWORD);
+ code = RegQueryValueEx(parmKey, "IdleDeadTimeout", NULL, NULL,
+ (BYTE *) &dwValue, &dummyLen);
+ if (code == ERROR_SUCCESS) {
+ IdleDeadtimeout = (unsigned short)dwValue;
+ afsi_log("IdleDeadTimeout is %d", IdleDeadtimeout);
+ }
+ RegCloseKey(parmKey);
}
- afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
- if (ConnDeadtimeout == 0)
- ConnDeadtimeout = (unsigned short) (RDRtimeout / 2);
- afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
- if (HardDeadtimeout == 0)
- HardDeadtimeout = (unsigned short) RDRtimeout;
- afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
-
+ /*
+ * 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.
+ */
+ 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);
+ }
osi_EndOnce(&once);
}
}
if (!cellp)
return CM_ERROR_NOSUCHCELL;
- code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp, CM_GETVOL_FLAG_CREATE, &volp);
if (code)
return code;
- *serversppp = cm_GetVolServers(volp, fidp->volume);
+ *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp);
+ lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
- return 0;
+ lock_ReleaseRead(&cm_volumeLock);
+ return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
}
/*
int dead_session;
long timeUsed, timeLeft;
long code;
- char addr[16];
+ char addr[16]="unknown";
int forcing_new = 0;
osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
if (cellp == NULL && serversp) {
struct cm_serverRef * refp;
for ( refp=serversp ; cellp == NULL && refp != NULL; refp=refp->next) {
+ if (refp->status == srv_deleted)
+ continue;
if ( refp->server )
cellp = refp->server->cellp;
}
*/
else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
- if (timeLeft > 7) {
- thrd_Sleep(5000);
-
- retry = 1;
-
- if (fidp != NULL) /* Not a VLDB call */
- cm_ForceUpdateVolume(fidp, userp, reqp);
- }
+ /*
+ * 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_ALLDOWN) {
osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN.");
/* Servers marked DOWN will be restored by the background daemon
- * thread as they become available.
+ * thread as they become available. The volume status is
+ * updated as the server state changes.
*/
}
/* Volume instances marked offline will be restored by the
* background daemon thread as they become available
*/
- if (timeLeft > 7 && fidp) {
- thrd_Sleep(5000);
-
- code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
- &volp);
+ if (fidp) {
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ &volp);
if (code == 0) {
- if (fidp->volume == volp->rw.ID)
- statep = &volp->rw;
- else if (fidp->volume == volp->ro.ID)
- statep = &volp->ro;
- else if (fidp->volume == volp->bk.ID)
- statep = &volp->bk;
-
- if (statep->state != vl_offline && statep->state != vl_unknown) {
- retry = 1;
- } else {
- if (cm_CheckOfflineVolume(volp, statep->ID))
+ if (timeLeft > 7) {
+ thrd_Sleep(5000);
+
+ /* cm_CheckOfflineVolume() resets the serverRef state */
+ if (cm_CheckOfflineVolume(volp, fidp->volume))
retry = 1;
+ } else {
+ cm_UpdateVolumeStatus(volp, fidp->volume);
}
-
+ lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
}
- }
+ }
}
else if (errorCode == CM_ERROR_ALLBUSY) {
/* Volumes that are busy cannot be determined to be non-busy
* without actually attempting to access them.
*/
osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY.");
- if (timeLeft > 7) {
-
- thrd_Sleep(5000);
-
- if (fidp) { /* File Server query */
- code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
- &volp);
- if (code == 0) {
- if (fidp->volume == volp->rw.ID)
- statep = &volp->rw;
- else if (fidp->volume == volp->ro.ID)
- statep = &volp->ro;
- else if (fidp->volume == volp->bk.ID)
- statep = &volp->bk;
+ if (fidp) { /* File Server query */
+ 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) {
+ statep->state != vl_busy &&
+ statep->state != vl_unknown) {
retry = 1;
} else {
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);
if (free_svr_list) {
- cm_FreeServerList(&serversp, 0);
- *serverspp = serversp;
+ cm_FreeServerList(serverspp, 0);
+ serverspp = NULL;
+ serversp = NULL;
+ free_svr_list = 0;
}
- cm_UpdateVolumeStatus(volp, statep->ID);
+ cm_UpdateVolumeStatus(volp, fidp->volume);
retry = 1;
}
-
- cm_PutVolume(volp);
+ } else {
+ cm_UpdateVolumeStatus(volp, fidp->volume);
}
- } else { /* VL Server query */
+
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
+ } else { /* VL Server query */
+ if (timeLeft > 7) {
+ 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_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 */
- code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp,
+ lock_ReleaseWrite(&cm_serverLock);
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
- if (code == 0) {
- if (fidp->volume == volp->rw.ID)
- statep = &volp->rw;
- else if (fidp->volume == volp->ro.ID)
- statep = &volp->ro;
- else if (fidp->volume == volp->bk.ID)
- statep = &volp->bk;
- }
+ if (code == 0)
+ statep = cm_VolumeStateByID(volp, fidp->volume);
+ lock_ObtainWrite(&cm_serverLock);
}
break;
}
if (statep) {
cm_UpdateVolumeStatus(volp, statep->ID);
+ lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
}
if (free_svr_list) {
- cm_FreeServerList(&serversp, 0);
- *serverspp = serversp;
+ cm_FreeServerList(serverspp, 0);
+ serverspp = NULL;
+ serversp = NULL;
+ free_svr_list = 0;
}
retry = 1;
}
else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
errorCode == VSALVAGE || errorCode == VNOSERVICE || errorCode == VIO)
{
- char addr[16];
char *format;
DWORD msgID;
+
+ /* In case of timeout */
+ reqp->volumeError = errorCode;
+
switch ( errorCode ) {
case VNOVOL:
msgID = MSG_SERVER_REPORTS_VNOVOL;
- format = "Server %s reported volume %d as not attached.";
+ format = "Server %s reported volume %d as not attached (does not exist).";
break;
case VMOVED:
msgID = MSG_SERVER_REPORTS_VMOVED;
LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume);
}
- /* Mark server offline for this volume */
+ /*
+ * Mark server offline for this volume or delete the volume
+ * from the server list if it was moved or is not present.
+ */
if (!serversp && fidp) {
code = cm_GetServerList(fidp, userp, reqp, &serverspp);
if (code == 0) {
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 (tsrp->server == serverp) {
/* REDIRECT */
- if (errorCode == VMOVED) {
+ 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 (fidp) { /* File Server query */
- code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
- &volp);
- if (code == 0) {
- if (fidp->volume == volp->rw.ID)
- statep = &volp->rw;
- else if (fidp->volume == volp->ro.ID)
- statep = &volp->ro;
- else if (fidp->volume == volp->bk.ID)
- statep = &volp->bk;
- }
- }
+ /* 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);
- if (fidp && errorCode == VMOVED)
- cm_ForceUpdateVolume(fidp, userp, reqp);
-
- if (statep) {
- cm_UpdateVolumeStatus(volp, statep->ID);
- cm_PutVolume(volp);
+ /* Free the server list before cm_ForceUpdateVolume is called */
+ if (free_svr_list) {
+ cm_FreeServerList(serverspp, 0);
+ serverspp = NULL;
+ serversp = NULL;
+ free_svr_list = 0;
}
- if (free_svr_list) {
- cm_FreeServerList(&serversp, 0);
- *serverspp = serversp;
+ if (fidp) { /* File Server query */
+ 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 ((errorCode == VMOVED || errorCode == VNOVOL) &&
+ !(reqp->flags & CM_REQ_VOLUME_UPDATED))
+ {
+ code = cm_ForceUpdateVolume(fidp, userp, reqp);
+ if (code)
+ timeLeft = 0; /* prevent a retry on failure */
+ else
+ 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);
+ }
+
+ if (statep) {
+ cm_UpdateVolumeStatus(volp, statep->ID);
+ osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
+ fidp->cell, fidp->volume, statep->state);
+ }
+
+ if (volp) {
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
}
+
if ( timeLeft > 2 )
retry = 1;
} else if ( errorCode == VNOVNODE ) {
if (scp->fileType != CM_SCACHETYPE_DIRECTORY)
pscp = cm_FindSCacheParent(scp);
-
lock_ObtainWrite(&scp->rw);
lock_ObtainWrite(&cm_scacheLock);
cm_RemoveSCacheFromHashTable(scp);
/* RX codes */
else if (errorCode == RX_CALL_TIMEOUT) {
- /* server took longer than hardDeadTime
+ /* RPC took longer than hardDeadTime or the server
+ * 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
*/
- /* 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));
-
- LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_HARD_DEAD_TIME_EXCEEDED, addr);
-
- retry = 0;
- osi_Log1(afsd_logp, "cm_Analyze: hardDeadTime exceeded addr[%s]",
- osi_LogSaveString(afsd_logp,addr));
+ 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_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->idleError++;
+
+ if (timeLeft > 2) {
+ if (!fidp) { /* vldb */
+ retry = 1;
+ } else { /* file */
+ cm_volume_t *volp = cm_GetVolumeByFID(fidp);
+ if (volp) {
+ if (fidp->volume == cm_GetROVolumeID(volp))
+ retry = 1;
+ cm_PutVolume(volp);
+ }
+ }
+ }
+ }
}
else if (errorCode >= -64 && errorCode < 0) {
/* mark server as down */
- lock_ObtainMutex(&serverp->mx);
- if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
- if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
- serverp->flags |= CM_SERVERFLAG_DOWN;
- serverp->downTime = time(NULL);
+ 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 (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"));
+
+ if (serverp) {
+ lock_ObtainMutex(&serverp->mx);
+ if (errorCode == RX_CALL_DEAD &&
+ (reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
+ if (!(serverp->flags & CM_SERVERFLAG_DOWN)) {
+ serverp->flags |= CM_SERVERFLAG_DOWN;
+ serverp->downTime = time(NULL);
+ }
+ } 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;
+ }
}
- } else {
- reqp->flags |= CM_REQ_NEW_CONN_FORCED;
- forcing_new = 1;
- }
- lock_ReleaseMutex(&serverp->mx);
- cm_ForceNewConnections(serverp);
+ lock_ReleaseMutex(&serverp->mx);
+ 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);
retry = 1;
}
} else if (errorCode >= ERROR_TABLE_BASE_RXK && errorCode < ERROR_TABLE_BASE_RXK + 256) {
- reqp->tokenErrorServp = serverp;
- reqp->tokenError = errorCode;
- retry = 1;
+ char * s = "unknown error";
+ switch ( errorCode ) {
+ case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
+ case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
+ case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
+ case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
+ case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
+ case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
+ case RXKADBADKEY : s = "RXKADBADKEY"; break;
+ case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
+ case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
+ case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
+ case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
+ case RXKADDATALEN : s = "RXKADDATALEN"; break;
+ case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
+ }
+ osi_Log2(afsd_logp, "cm_Analyze: rxkad error code 0x%x (%s)",
+ errorCode, s);
+
+ if (serverp) {
+ reqp->tokenIdleErrorServp = serverp;
+ reqp->tokenError = errorCode;
+ retry = 1;
+ }
+ } else if (errorCode >= ERROR_TABLE_BASE_U && errorCode < ERROR_TABLE_BASE_U + 256) {
+ /*
+ * We received a ubik error. its possible that the server we are
+ * communicating with has a corrupted database or is partitioned
+ * from the rest of the servers and another server might be able
+ * to answer our query. Therefore, we will retry the request
+ * and force the use of another server.
+ */
+ if (serverp) {
+ reqp->tokenIdleErrorServp = serverp;
+ reqp->tokenError = errorCode;
+ retry = 1;
+ }
} else if (errorCode == VICECONNBAD || errorCode == VICETOKENDEAD) {
cm_ForceNewConnections(serverp);
if ( timeLeft > 2 )
if (errorCode) {
char * s = "unknown error";
switch ( errorCode ) {
- case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
- case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
- case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
- case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
- case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
- case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
- case RXKADBADKEY : s = "RXKADBADKEY"; break;
- case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
- case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
- case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
- case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
- case RXKADDATALEN : s = "RXKADDATALEN"; break;
- case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
case VSALVAGE : s = "VSALVAGE"; break;
case VNOVNODE : s = "VNOVNODE"; break;
case VNOVOL : s = "VNOVOL"; break;
case UAEACCES : s = "UAEACCES"; break;
case ENOENT : s = "ENOENT"; break;
case UAENOENT : s = "UAENOENT"; break;
+ case EEXIST : s = "EEXIST"; break;
+ case UAEEXIST : s = "UAEEXIST"; break;
case VICECONNBAD : s = "VICECONNBAD"; break;
case VICETOKENDEAD : s = "VICETOKENDEAD"; break;
case WSAEWOULDBLOCK : s = "WSAEWOULDBLOCK"; break;
if (connp)
cm_PutConn(connp);
+ /*
+ * clear the volume updated flag if we succeed.
+ * this way the flag will not prevent a subsequent volume
+ * from being updated if necessary.
+ */
+ if (errorCode == 0)
+ {
+ reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
+ }
+
/* retry until we fail to find a connection */
return retry;
}
cm_server_t *tsp;
long firstError = 0;
int someBusy = 0, someOffline = 0, allOffline = 1, allBusy = 1, allDown = 1;
+#ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
long timeUsed, timeLeft, hardTimeLeft;
-
+#endif
*connpp = NULL;
if (serversp == NULL) {
return CM_ERROR_ALLDOWN;
}
+#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;
+#endif
lock_ObtainRead(&cm_serverLock);
for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
+
tsp = tsrp->server;
- if (reqp->tokenErrorServp) {
+ if (reqp->tokenIdleErrorServp) {
/*
* 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
* in the list.
*/
- if (tsp == reqp->tokenErrorServp)
- reqp->tokenErrorServp = NULL;
+ if (tsp == reqp->tokenIdleErrorServp)
+ reqp->tokenIdleErrorServp = NULL;
continue;
}
- cm_GetServerNoLock(tsp);
- lock_ReleaseRead(&cm_serverLock);
- if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
- allDown = 0;
- if (tsrp->status == srv_deleted) {
- /* skip this entry. no longer valid. */;
- } else if (tsrp->status == srv_busy) {
- allOffline = 0;
- someBusy = 1;
- } else if (tsrp->status == srv_offline) {
- allBusy = 0;
- someOffline = 1;
- } else {
- allOffline = 0;
- allBusy = 0;
- code = cm_ConnByServer(tsp, usersp, connpp);
- if (code == 0) { /* cm_CBS only returns 0 */
- cm_PutServer(tsp);
- /* Set RPC timeout */
- if (timeLeft > ConnDeadtimeout)
- timeLeft = ConnDeadtimeout;
-
- if (hardTimeLeft > HardDeadtimeout)
- hardTimeLeft = HardDeadtimeout;
-
- lock_ObtainMutex(&(*connpp)->mx);
- rx_SetConnDeadTime((*connpp)->callp, timeLeft);
- rx_SetConnHardDeadTime((*connpp)->callp, (u_short) hardTimeLeft);
- lock_ReleaseMutex(&(*connpp)->mx);
- return 0;
+ if (tsp) {
+ cm_GetServerNoLock(tsp);
+ lock_ReleaseRead(&cm_serverLock);
+ if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
+ allDown = 0;
+ if (tsrp->status == srv_busy) {
+ allOffline = 0;
+ someBusy = 1;
+ } else if (tsrp->status == srv_offline) {
+ allBusy = 0;
+ someOffline = 1;
+ } else {
+ allOffline = 0;
+ allBusy = 0;
+ code = cm_ConnByServer(tsp, usersp, connpp);
+ if (code == 0) { /* cm_CBS only returns 0 */
+ cm_PutServer(tsp);
+#ifdef SET_RX_TIMEOUTS_TO_TIMELEFT
+ /* Set RPC timeout */
+ if (timeLeft > ConnDeadtimeout)
+ timeLeft = ConnDeadtimeout;
+
+ if (hardTimeLeft > HardDeadtimeout)
+ hardTimeLeft = HardDeadtimeout;
+
+ lock_ObtainMutex(&(*connpp)->mx);
+ rx_SetConnDeadTime((*connpp)->rxconnp, timeLeft);
+ rx_SetConnHardDeadTime((*connpp)->rxconnp, (u_short) hardTimeLeft);
+ lock_ReleaseMutex(&(*connpp)->mx);
+#endif
+ return 0;
+ }
+
+ /* therefore, this code is never executed */
+ if (firstError == 0)
+ firstError = code;
}
-
- /* therefore, this code is never executed */
- if (firstError == 0)
- firstError = code;
}
+ lock_ObtainRead(&cm_serverLock);
+ cm_PutServerNoLock(tsp);
}
- lock_ObtainRead(&cm_serverLock);
- cm_PutServerNoLock(tsp);
}
lock_ReleaseRead(&cm_serverLock);
if (firstError == 0) {
if (allDown)
- firstError = (reqp->tokenError ? reqp->tokenError : CM_ERROR_ALLDOWN);
+ firstError = (reqp->tokenError ? reqp->tokenError :
+ (reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
else if (allBusy)
firstError = CM_ERROR_ALLBUSY;
else if (allOffline || (someBusy && someOffline))
cm_PutServer(tcp->serverp);
cm_ReleaseUser(userp);
*lcpp = tcp->nextp;
- rx_DestroyConnection(tcp->callp);
+ rx_DestroyConnection(tcp->rxconnp);
lock_FinalizeMutex(&tcp->mx);
free(tcp);
}
}
if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
secIndex = 2;
- if (cryptall) {
- tcp->cryptlevel = rxkad_crypt;
- } else {
+ switch (cryptall) {
+ case 0:
tcp->cryptlevel = rxkad_clear;
+ break;
+ case 2:
+ tcp->cryptlevel = rxkad_auth;
+ break;
+ default:
+ tcp->cryptlevel = rxkad_crypt;
}
secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
&ucellp->sessionKey, ucellp->kvno,
secObjp = rxnull_NewClientSecurityObject();
}
osi_assertx(secObjp != NULL, "null rx_securityClass");
- tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
+ tcp->rxconnp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
port,
serviceID,
secObjp,
secIndex);
- rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
- rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
+ rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
+ rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
+ rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
tcp->ucgen = ucellp->gen;
if (secObjp)
rxs_Release(secObjp); /* Decrement the initial refCount */
serverp->connsp = tcp;
cm_HoldUser(userp);
tcp->userp = userp;
- lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
+ lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex", LOCK_HIERARCHY_CONN);
lock_ObtainMutex(&tcp->mx);
tcp->serverp = serverp;
tcp->cryptlevel = rxkad_clear;
lock_ObtainMutex(&tcp->mx);
if ((tcp->flags & CM_CONN_FLAG_FORCE_NEW) ||
(tcp->ucgen < ucellp->gen) ||
- (tcp->cryptlevel != (cryptall ? (ucellp->flags & CM_UCELLFLAG_RXKAD ? rxkad_crypt : rxkad_clear) : rxkad_clear)))
+ (tcp->cryptlevel != (ucellp->flags & CM_UCELLFLAG_RXKAD ? (cryptall == 1 ? rxkad_crypt : (cryptall == 2 ? rxkad_auth : rxkad_clear)) : rxkad_clear)))
{
if (tcp->ucgen < ucellp->gen)
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_DestroyConnection(tcp->callp);
+ rx_DestroyConnection(tcp->rxconnp);
cm_NewRXConnection(tcp, ucellp, serverp);
}
lock_ReleaseMutex(&tcp->mx);
lock_ObtainRead(&cm_serverLock);
for (tsrp = *serverspp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
tsp = tsrp->server;
if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
allDown = 0;
*connpp = NULL;
- serverspp = cm_GetVolServers(volp, volid);
+ serverspp = cm_GetVolServers(volp, volid, userp, reqp);
code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
cm_FreeServerList(serverspp, 0);
extern struct rx_connection *
cm_GetRxConn(cm_conn_t *connp)
{
- struct rx_connection * rxconn;
+ struct rx_connection * rxconnp;
lock_ObtainMutex(&connp->mx);
- rxconn = connp->callp;
- rx_GetConnection(rxconn);
+ rxconnp = connp->rxconnp;
+ rx_GetConnection(rxconnp);
lock_ReleaseMutex(&connp->mx);
- return rxconn;
+ return rxconnp;
}
void cm_ForceNewConnections(cm_server_t *serverp)
git://git.openafs.org / openafs.git / blobdiff