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"
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) {
*
* 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 (smb_Enabled) {
afsi_log("lanmanworkstation : SessTimeout %u", RDRtimeout);
afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
}
if (IdleDeadtimeout == 0) {
- IdleDeadtimeout = (unsigned short) ConnDeadtimeout;
+ 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;
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_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;
+ cm_vol_state_t *volstatep = NULL;
cm_cell_t *cellp = NULL;
if (!fidp) {
if (code)
return code;
+ volstatep = cm_VolumeStateByID(volp, fidp->volume);
+ *replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
*serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp);
lock_ObtainRead(&cm_volumeLock);
return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
}
+void
+cm_SetServerBusyStatus(cm_serverRef_t *serversp, cm_server_t *serverp)
+{
+ cm_serverRef_t *tsrp;
+
+ lock_ObtainWrite(&cm_serverLock);
+ for (tsrp = serversp; 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_ReleaseWrite(&cm_serverLock);
+}
+
+void
+cm_ResetServerBusyStatus(cm_serverRef_t *serversp)
+{
+ cm_serverRef_t *tsrp;
+
+ 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);
+}
+
/*
* 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,
* volSyncp and/or cbrp may also be NULL.
*/
int
-cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
+cm_Analyze(cm_conn_t *connp,
+ cm_user_t *userp,
+ cm_req_t *reqp,
struct cm_fid *fidp,
+ afs_uint32 storeOp,
AFSVolSync *volSyncp,
cm_serverRef_t * serversp,
- cm_callbackRequest_t *cbrp, long errorCode)
+ cm_callbackRequest_t *cbrp,
+ long errorCode)
{
cm_server_t *serverp = NULL;
cm_serverRef_t **serverspp = NULL;
cm_ucell_t *ucellp;
cm_volume_t * volp = NULL;
cm_vol_state_t *statep = NULL;
+ cm_scache_t * scp = NULL;
+ afs_uint32 replicated;
int retry = 0;
int free_svr_list = 0;
int dead_session;
} 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
format = "All servers are offline when accessing cell %s volume %d.";
LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
+ if (!serversp) {
+ code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
+ if (code == 0) {
+ serversp = *serverspp;
+ free_svr_list = 1;
+ }
+ }
+ cm_ResetServerBusyStatus(serversp);
+ if (free_svr_list) {
+ cm_FreeServerList(serverspp, 0);
+ free_svr_list = 0;
+ serversp = NULL;
+ }
+
code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
format = "All servers are busy when accessing cell %s volume %d.";
LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
+ if (!serversp) {
+ code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
+ if (code == 0) {
+ serversp = *serverspp;
+ free_svr_list = 1;
+ }
+ }
+ cm_ResetServerBusyStatus(serversp);
+ if (free_svr_list) {
+ cm_FreeServerList(serverspp, 0);
+ free_svr_list = 0;
+ serversp = NULL;
+ }
+
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;
- }
-
- cm_UpdateVolumeStatus(volp, fidp->volume);
- retry = 1;
- }
- } else {
- cm_UpdateVolumeStatus(volp, fidp->volume);
+ retry = 1;
}
+ cm_UpdateVolumeStatus(volp, fidp->volume);
lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
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);
+ cm_ResetServerBusyStatus(serversp);
retry = 1;
}
}
/* special codes: VBUSY and VRESTARTING */
else if (errorCode == VBUSY || errorCode == VRESTARTING) {
if (!serversp && fidp) {
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
+ code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
if (code == 0) {
serversp = *serverspp;
free_svr_list = 1;
LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
}
- 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);
- }
- break;
- }
- }
- lock_ReleaseWrite(&cm_serverLock);
+ cm_SetServerBusyStatus(serversp, serverp);
- if (statep) {
- cm_UpdateVolumeStatus(volp, statep->ID);
- lock_ObtainRead(&cm_volumeLock);
- cm_PutVolume(volp);
- lock_ReleaseRead(&cm_volumeLock);
- volp = NULL;
+ 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 (statep)
+ cm_UpdateVolumeStatus(volp, statep->ID);
+
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
}
if (free_svr_list) {
/* special codes: missing volumes */
else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
- errorCode == VSALVAGE || errorCode == VNOSERVICE || errorCode == VIO)
+ errorCode == VSALVAGE || errorCode == VIO)
{
/* 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) {
* from the server list if it was moved or is not present.
*/
if (!serversp || location_updated) {
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
+ code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
if (code == 0) {
serversp = *serverspp;
free_svr_list = 1;
if (cm_ServerEqual(tsrp->server, serverp)) {
/* REDIRECT */
- if (errorCode == VMOVED || errorCode == VNOVOL) {
- osi_Log2(afsd_logp, "volume %d not present on server %s",
+ 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);
- } else {
- osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
- fidp->volume, osi_LogSaveString(afsd_logp,addr));
- tsrp->status = srv_offline;
+ 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) {
+ cm_SetServerBusyStatus(serversp, serverp);
+ } else {
+ Sleep(2000);
+ }
+ }
+ 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++;
+ }
+
/* Free the server list before cm_ForceUpdateVolume is called */
if (free_svr_list) {
cm_FreeServerList(serverspp, 0);
retry = 1;
} else if ( errorCode == VNOVNODE ) {
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);
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;
retry = 1;
}
- else if (errorCode == CM_RX_RETRY_BUSY_CALL) {
+ else if (errorCode == RX_CALL_BUSY) {
/*
* RPC failed because the selected call channel
* is currently busy on the server. Unconditionally
osi_LogSaveString(afsd_logp,addr));
retry = 1;
}
- else if (errorCode >= -64 && errorCode < 0) {
+ 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 (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);
+ }
+
+ if (timeLeft > 2)
+ retry = 1;
+ }
+ 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 (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;
+
+ if (timeLeft > 2)
+ retry = 1;
+ }
+
+ 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 ((reqp->flags & CM_REQ_NEW_CONN_FORCED)) {
+ lock_ObtainMutex(&serverp->mx);
if (!(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);
+ }
+ }
+
+ 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 (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);
}
- lock_ReleaseMutex(&serverp->mx);
- cm_ForceNewConnections(serverp);
}
if ( timeLeft > 2 )
retry = 1;
errorCode, s);
if (serverp) {
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->tokenError = errorCode;
retry = 1;
}
* and force the use of another server.
*/
if (serverp) {
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->tokenError = errorCode;
retry = 1;
}
/* If not allowed to retry, don't */
if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
- (errorCode != RX_MSGSIZE && errorCode != CM_RX_RETRY_BUSY_CALL))
+ (errorCode != RX_MSGSIZE && errorCode != RX_CALL_BUSY))
retry = 0;
else if (retry && dead_session)
retry = 0;
reqp->flags &= ~CM_REQ_VOLUME_UPDATED;
}
+ if ( serversp &&
+ errorCode != VBUSY &&
+ errorCode != VRESTARTING &&
+ errorCode != CM_ERROR_ALLBUSY)
+ {
+ cm_ResetServerBusyStatus(serversp);
+ }
+
/* 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;
continue;
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 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
}
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;
secIndex);
rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
- rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
/*
- * 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) {
+ 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
rxs_Release(secObjp); /* Decrement the initial refCount */
}
-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;
lock_ObtainMutex(&userp->mx);
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;
}
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);
lock_ReleaseWrite(&cm_connLock);
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);
}
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;
{
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;
+ cm_vol_state_t * volstatep;
*connpp = NULL;
+ volstatep = cm_VolumeStateByID(volp, volid);
+ replicated = (volstatep->flags & CM_VOL_STATE_FLAG_REPLICATED);
serverspp = cm_GetVolServers(volp, volid, userp, reqp);
- code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
+ code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
cm_FreeServerList(serverspp, 0);
return code;
}
git://git.openafs.org / openafs.git / blobdiff