allserversdown-20040723

When all servers associated with a volume get into the CM_SERVERFLAG_DOWN
state, the error CM_ERROR_ALLOFFLINE would be returned by cm_ConnByMServers.
cm_Analyze was then supposed to be pausing for 5 seconds and then reset the
volume information.  Unfortunately, although it called cm_ForceUpdateVolume,
cm_ForceUpdateVolume does not reset the CM_SERVERFLAG_DOWN state on the
servers.  Therefore, when cm_ForceUpdateVolume calls cm_ConnByMServers it
would be given a CM_ERROR_ALLOFFLINE.  In other words, there was no way
out of the state.

cm_Analyze will now reset the CM_SERVERFLAG_DOWN as well as setting the
server status to not_busy after its 5 second wait.  This will allow
cm_ForceUpdateVolume to actually reset the volume information, and
refresh it if servers for the volume are newly accessible.

Also, added some missing locking calls in cm_Analyze.

In cm_ConnByMServers, change the error reporting to return CM_ERROR_ALLBUSY
only when all servers are busy or down; return CM_ERROR_ALLOFFLINE when all
servers are down; and only return CM_ERROR_NOSUCHVOLUME if the server list
for the volume is empty.  In all other cases return CM_ERROR_TIMEDOUT.

diff --git a/src/WINNT/afsd/cm_conn.c b/src/WINNT/afsd/cm_conn.c
index b3e22c0..ec0876a 100644 (file)
--- a/src/WINNT/afsd/cm_conn.c
+++ b/src/WINNT/afsd/cm_conn.c
@@ -201,17 +201,33 @@ cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
            osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
            thrd_Sleep(5000);
            /* cm_ForceUpdateVolume marks all servers as non_busy */
-           cm_ForceUpdateVolume(fidp, userp, reqp);
+               /* No it doesn't.  It won't do anything if all of the 
+                * the servers are marked as DOWN.  So clear the DOWN
+                * flag and reset the busy state as well.
+                */
+               cm_GetServerList(fidp, userp, reqp, &serversp);
+               lock_ObtainWrite(&cm_serverLock);
+               for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
+               tsrp->server->flags &= ~CM_SERVERFLAG_DOWN;
+                       if (tsrp->status == busy)
+                               tsrp->status = not_busy;
+               }
+        lock_ReleaseWrite(&cm_serverLock);
+
+        if (fidp != NULL)
+            cm_ForceUpdateVolume(fidp, userp, reqp);
            retry = 1;
        }
 
        /* if all servers are busy, mark them non-busy and start over */
        if (errorCode == CM_ERROR_ALLBUSY) {
                cm_GetServerList(fidp, userp, reqp, &serversp);
+               lock_ObtainWrite(&cm_serverLock);
                for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
                        if (tsrp->status == busy)
                                tsrp->status = not_busy;
                }
+        lock_ReleaseWrite(&cm_serverLock);
                thrd_Sleep(5000);
                retry = 1;
        }
@@ -219,6 +235,7 @@ cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
        /* special codes:  VBUSY and VRESTARTING */
        if (errorCode == VBUSY || errorCode == VRESTARTING) {
                cm_GetServerList(fidp, userp, reqp, &serversp);
+               lock_ObtainWrite(&cm_serverLock);
                for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
                        if (tsrp->server == serverp
                            && tsrp->status == not_busy) {
@@ -226,6 +243,7 @@ cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
                                break;
                        }
                }
+        lock_ReleaseWrite(&cm_serverLock);
                retry = 1;
        }
 
@@ -328,7 +346,7 @@ long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
        cm_serverRef_t *tsrp;
         cm_server_t *tsp;
         long firstError = 0;
-       int someBusy = 0, someOffline = 0, allDown = 1;
+       int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
        long timeUsed, timeLeft, hardTimeLeft;
 #ifdef DJGPP
         struct timeval now;
@@ -360,6 +378,7 @@ long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
             else if (tsrp->status == offline)
                 someOffline = 1;
             else {
+                               allBusy = 0;
                 code = cm_ConnByServer(tsp, usersp, connpp);
                 if (code == 0) {
                     cm_PutServer(tsp);
@@ -389,16 +408,17 @@ long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
 
        lock_ReleaseWrite(&cm_serverLock);
        if (firstError == 0) {
-               if (someBusy) 
+               if (allBusy) 
                        firstError = CM_ERROR_ALLBUSY;
-               else if (someOffline) 
+               else if (allDown) 
                        firstError = CM_ERROR_ALLOFFLINE;
-               else if (!allDown && serversp) 
-                       firstError = CM_ERROR_TIMEDOUT;
-               /* Only return CM_ERROR_NOSUCHVOLUME if there are no
-                  servers for this volume */
-               else 
+               else if (serversp == NULL) 
+                       /* Only return CM_ERROR_NOSUCHVOLUME if there are no
+                        * servers for this volume 
+                        */
                        firstError = CM_ERROR_NOSUCHVOLUME;
+               else
+                       firstError = CM_ERROR_TIMEDOUT;
        }
        osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
     return firstError;