reqp->startTime = GetTickCount();
}
-static long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
+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;
- 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);
+ *serversppp = cm_GetVolServers(volp, fidp->volume, userp, reqp, replicated);
lock_ObtainRead(&cm_volumeLock);
cm_PutVolume(volp);
for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
if (tsrp->status == srv_deleted)
continue;
- if (tsrp->server == serverp && tsrp->status == srv_not_busy) {
+ if (cm_ServerEqual(tsrp->server, serverp) && tsrp->status == srv_not_busy) {
tsrp->status = srv_busy;
break;
}
*
* 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.
*/
cm_user_t *userp,
cm_req_t *reqp,
struct cm_fid *fidp,
+ cm_cell_t *cellp,
afs_uint32 storeOp,
AFSVolSync *volSyncp,
cm_serverRef_t * serversp,
cm_server_t *serverp = NULL;
cm_serverRef_t **serverspp = 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;
*/
}
+ 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) {
+ if (cm_UpdateVolumeLocation(cellp, userp, reqp, volp) == 0) {
+ code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
+ if (code == 0) {
+ if (!cm_IsServerListEmpty(*serverspp))
+ retry = 1;
+ cm_FreeServerList(serverspp, 0);
+ }
+ }
+
+ 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
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
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);
if (code == 0) {
+ if (!serversp) {
+ code = cm_GetVolServerList(volp, fidp->volume, 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;
+ }
+
/*
* Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
* was called by cm_CheckOfflineVolumeState().
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 (!serversp) {
+ code = cm_GetVolServerList(volp, fidp->volume, 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;
+ }
+
if (timeLeft > 7) {
thrd_Sleep(5000);
statep = cm_VolumeStateByID(volp, fidp->volume);
/* special codes: VBUSY and VRESTARTING */
else if (errorCode == VBUSY || errorCode == VRESTARTING) {
- if (!serversp && fidp) {
- code = cm_GetServerList(fidp, userp, reqp, &replicated, &serverspp);
- if (code == 0) {
- serversp = *serverspp;
- 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;
- }
-
- if (serverp && fidp) {
- /* 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);
- }
-
- cm_SetServerBusyStatus(serversp, serverp);
-
- if (fidp) { /* File Server query */
+ if (fidp) {
code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
if (code == 0) {
+ if (!serversp) {
+ code = cm_GetVolServerList(volp, fidp->volume, userp, reqp, &replicated, &serverspp);
+ if (code == 0) {
+ serversp = *serverspp;
+ free_svr_list = 1;
+ }
+ }
+
statep = cm_VolumeStateByID(volp, fidp->volume);
if (statep)
}
}
+ 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));
+
+ 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;
+ }
+
+ osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
+
+ cm_SetServerBusyStatus(serversp, serverp);
+ }
+
if (free_svr_list) {
cm_FreeServerList(serverspp, 0);
serversp = NULL;
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) {
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) {
+ if (tsrp->status == srv_not_busy)
+ tsrp->status = srv_busy;
+ } 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);
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 (replicated && serverp) {
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->tokenError = errorCode;
if (timeLeft > 2)
if (serverp) {
if (reqp->flags & CM_REQ_NEW_CONN_FORCED) {
- reqp->tokenIdleErrorServp = serverp;
+ reqp->errorServp = serverp;
reqp->tokenError = errorCode;
} else {
reqp->flags |= CM_REQ_NEW_CONN_FORCED;
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;
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;
}
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
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 next one
* in the list.
*/
- if (tsp == reqp->tokenIdleErrorServp)
- reqp->tokenIdleErrorServp = NULL;
+ if (tsp == reqp->errorServp)
+ reqp->errorServp = NULL;
continue;
}
if (tsp) {
lock_ReleaseRead(&cm_serverLock);
if (firstError == 0) {
- if (allDown) {
+ if (allDeleted) {
+ firstError = CM_ERROR_EMPTY;
+ } else if (allDown) {
firstError = (reqp->tokenError ? reqp->tokenError :
(reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
/*
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, replicated);
tcp->refCount = 1;
- lock_ReleaseMutex(&tcp->mx);
+ tcp->nextp = serverp->connsp;
+ serverp->connsp = tcp;
lock_ReleaseWrite(&cm_connLock);
+ lock_ReleaseMutex(&userp->mx);
} else {
lock_ReleaseRead(&cm_connLock);
haveconn:
+ lock_ReleaseMutex(&userp->mx);
InterlockedIncrement(&tcp->refCount);
lock_ObtainMutex(&tcp->mx);
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;
+ tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
rx_DestroyConnection(tcp->rxconnp);
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);
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);
+ serverspp = cm_GetVolServers(volp, volid, userp, reqp, &replicated);
code = cm_ConnByMServers(*serverspp, replicated, userp, reqp, connpp);
cm_FreeServerList(serverspp, 0);
git://git.openafs.org / openafs.git / blobdiff