DEVEL15-windows-vlserver-updates-20080924

LICENSE MIT

if a vlserver returns a ubik error, skip the server and retry

(cherry picked from commit dd05bd3ba7be90a6348b144eb1c0a6ac1c5f43ba)

diff --git a/src/WINNT/afsd/cm_conn.c b/src/WINNT/afsd/cm_conn.c
index a09c41fffd523ddb3aaede73df6d464684829fa6..f89efbe4fcad55c720b1c23889e8a168e7650cea 100644 (file)
--- a/src/WINNT/afsd/cm_conn.c
+++ b/src/WINNT/afsd/cm_conn.c
@@ -141,12 +141,12 @@ static long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
     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);
     lock_ReleaseRead(&cm_volumeLock);
-    return 0;
+    return (*serversppp ? 0 : CM_ERROR_NOSUCHVOLUME);
 }
 
 /*
@@ -635,6 +635,19 @@ cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
             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 )
@@ -1126,7 +1139,7 @@ long cm_ConnFromVolume(struct cm_volume *volp, unsigned long volid, struct cm_us
 
     *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);

diff --git a/src/WINNT/afsd/cm_ioctl.c b/src/WINNT/afsd/cm_ioctl.c
index 46dcf80f5f5bb5ec26098c279a4bac22567c3d8e..41f7fd1e2abda44c13b605a83ace1145a38f34f8 100644 (file)
--- a/src/WINNT/afsd/cm_ioctl.c
+++ b/src/WINNT/afsd/cm_ioctl.c
@@ -908,7 +908,7 @@ cm_IoctlGetOwner(struct cm_ioctl *ioctlp, struct cm_user *userp, cm_scache_t *sc
 afs_int32 
 cm_IoctlWhereIs(struct cm_ioctl *ioctlp, struct cm_user *userp, cm_scache_t *scp, cm_req_t *reqp)
 {
-    afs_int32 code;
+    afs_int32 code = 0;
     cm_cell_t *cellp;
     cm_volume_t *tvp;
     cm_serverRef_t **tsrpp, *current;
@@ -952,16 +952,19 @@ cm_IoctlWhereIs(struct cm_ioctl *ioctlp, struct cm_user *userp, cm_scache_t *scp
        
         cp = ioctlp->outDatap;
         
-        tsrpp = cm_GetVolServers(tvp, volume);
-        lock_ObtainRead(&cm_serverLock);
-        for (current = *tsrpp; current; current = current->next) {
-            tsp = current->server;
-            memcpy(cp, (char *)&tsp->addr.sin_addr.s_addr, sizeof(long));
-            cp += sizeof(long);
+        tsrpp = cm_GetVolServers(tvp, volume, userp, reqp);
+        if (tsrpp == NULL) {
+            code = CM_ERROR_NOSUCHVOLUME;
+        } else {
+            lock_ObtainRead(&cm_serverLock);
+            for (current = *tsrpp; current; current = current->next) {
+                tsp = current->server;
+                memcpy(cp, (char *)&tsp->addr.sin_addr.s_addr, sizeof(long));
+                cp += sizeof(long);
+            }
+            lock_ReleaseRead(&cm_serverLock);
+            cm_FreeServerList(tsrpp, 0);
         }
-        lock_ReleaseRead(&cm_serverLock);
-        cm_FreeServerList(tsrpp, 0);
-
         /* still room for terminating NULL, add it on */
         volume = 0;    /* reuse vbl */
         memcpy(cp, (char *)&volume, sizeof(long));
@@ -970,7 +973,7 @@ cm_IoctlWhereIs(struct cm_ioctl *ioctlp, struct cm_user *userp, cm_scache_t *scp
         ioctlp->outDatap = cp;
         cm_PutVolume(tvp);
     }
-    return 0;
+    return code;
 }       
 
 /* 
@@ -3023,6 +3026,8 @@ cm_IoctlMemoryDump(struct cm_ioctl *ioctlp, struct cm_user *userp)
     cm_DumpSCache(hLogFile, cookie, 1);
     cm_DumpBufHashTable(hLogFile, cookie, 1);
     smb_DumpVCP(hLogFile, cookie, 1);
+    rx_DumpCalls(hLogFile, cookie);
+    rx_DumpPackets(hLogFile, cookie);
 
     CloseHandle(hLogFile);                          
   

diff --git a/src/WINNT/afsd/cm_volume.c b/src/WINNT/afsd/cm_volume.c
index 5a34d0c32b9be9c8046468f2dac105f192c4e8db..7d23c0a8981f6209d53b95da506848285702aea7 100644 (file)
--- a/src/WINNT/afsd/cm_volume.c
+++ b/src/WINNT/afsd/cm_volume.c
@@ -1009,11 +1009,13 @@ long cm_ForceUpdateVolume(cm_fid_t *fidp, cm_user_t *userp, cm_req_t *reqp)
 }
 
 /* find the appropriate servers from a volume */
-cm_serverRef_t **cm_GetVolServers(cm_volume_t *volp, afs_uint32 volume)
+cm_serverRef_t **cm_GetVolServers(cm_volume_t *volp, afs_uint32 volume, cm_user_t *userp, cm_req_t *reqp)
 {
     cm_serverRef_t **serverspp;
     cm_serverRef_t *current;
+    int firstTry = 1;
 
+  start:
     lock_ObtainWrite(&cm_serverLock);
 
     if (volume == volp->vol[RWVOL].ID)
@@ -1022,9 +1024,21 @@ cm_serverRef_t **cm_GetVolServers(cm_volume_t *volp, afs_uint32 volume)
         serverspp = &volp->vol[ROVOL].serversp;
     else if (volume == volp->vol[BACKVOL].ID)
         serverspp = &volp->vol[BACKVOL].serversp;
-    else 
-        osi_panic("bad volume ID in cm_GetVolServers", __FILE__, __LINE__);
-        
+    else {
+        lock_ReleaseWrite(&cm_serverLock);
+        if (firstTry) {
+            afs_int32 code;
+            firstTry = 0;
+            lock_ObtainWrite(&volp->rw);
+            volp->flags |= CM_VOLUMEFLAG_RESET;
+            code = cm_UpdateVolumeLocation(volp->cellp, userp, reqp, volp);
+            lock_ReleaseWrite(&volp->rw);
+            if (code == 0)
+                goto start;
+        }
+        return NULL;
+    }
+
     /* 
      * Increment the refCount on deleted items as well.
      * They will be freed by cm_FreeServerList when they get to zero 

diff --git a/src/WINNT/afsd/cm_volume.h b/src/WINNT/afsd/cm_volume.h
index 9a276392641f625dc0d592bbb786cfd63e0bf3c8..b32d72e0c3b25ddb35006c92796d4fa93f0be2aa 100644 (file)
--- a/src/WINNT/afsd/cm_volume.h
+++ b/src/WINNT/afsd/cm_volume.h
@@ -92,7 +92,8 @@ extern long cm_GetROVolumeID(cm_volume_t *volp);
 extern long cm_ForceUpdateVolume(struct cm_fid *fidp, cm_user_t *userp,
        cm_req_t *reqp);
 
-extern cm_serverRef_t **cm_GetVolServers(cm_volume_t *volp, afs_uint32 volume);
+extern cm_serverRef_t **cm_GetVolServers(cm_volume_t *volp, afs_uint32 volume, 
+                                         cm_user_t *userp, cm_req_t *reqp);
 
 extern void cm_ChangeRankVolume(cm_server_t *tsp);