DAFS: avoid saving fileserver state when panicing

[openafs.git] / src / viced / viced.c

diff --git a/src/viced/viced.c b/src/viced/viced.c
index ad5b134..80c3a94 100644 (file)
--- a/src/viced/viced.c
+++ b/src/viced/viced.c
@@ -177,6 +177,16 @@ int SawLock;
 #endif
 time_t StartTime;
 
+/**
+ * seconds to wait until forcing a panic during ShutDownAndCore(PANIC)
+ * in case we get stuck.
+ */
+#ifdef AFS_DEMAND_ATTACH_FS
+static int panic_timeout = 2 * 60;
+#else
+static int panic_timeout = 30 * 60;
+#endif
+
 int rxpackets = 150;           /* 100 */
 int nSmallVns = 400;           /* 200 */
 int large = 400;               /* 200 */
@@ -772,12 +782,36 @@ CheckSignal(void *unused)
     return 0;
 }                              /*CheckSignal */
 
+static void *
+ShutdownWatchdogLWP(void *unused)
+{
+    sleep(panic_timeout);
+    ViceLog(0, ("ShutdownWatchdogLWP: Failed to shutdown and panic "
+                "within %d seconds; forcing panic\n", panic_timeout));
+    assert(0);
+    return NULL;
+}
+
 void
 ShutDownAndCore(int dopanic)
 {
     time_t now = time(0);
     char tbuffer[32];
 
+    if (dopanic) {
+#ifdef AFS_PTHREAD_ENV
+       pthread_t watchdogPid;
+       pthread_attr_t tattr;
+       assert(pthread_attr_init(&tattr) == 0);
+       assert(pthread_create(&watchdogPid, &tattr, ShutdownWatchdogLWP, NULL) == 0);
+#else
+       PROCESS watchdogPid;
+       assert(LWP_CreateProcess
+              (ShutdownWatchdogLWP, stack * 1024, LWP_MAX_PRIORITY - 2,
+               NULL, "ShutdownWatchdog", &watchdogPid) == LWP_SUCCESS);
+#endif
+    }
+
     /* do not allows new reqests to be served from now on, all new requests
      * are returned with an error code of RX_RESTARTING ( transient failure ) */
     rx_SetRxTranquil();                /* dhruba */
@@ -814,22 +848,27 @@ ShutDownAndCore(int dopanic)
         * demand attach fs
         * save fileserver state to disk */
 
-       /* make sure background threads have finished all of their asynchronous 
-        * work on host and callback structures */
-       FS_STATE_RDLOCK;
-       while (!fs_state.FiveMinuteLWP_tranquil ||
-              !fs_state.HostCheckLWP_tranquil ||
-              !fs_state.FsyncCheckLWP_tranquil) {
-           FS_LOCK;
-           FS_STATE_UNLOCK;
-           ViceLog(0, ("waiting for background host/callback threads to quiesce before saving fileserver state...\n"));
-           assert(pthread_cond_wait(&fs_state.worker_done_cv, &fileproc_glock_mutex) == 0);
-           FS_UNLOCK;
+       if (dopanic) {
+           ViceLog(0, ("Not saving fileserver state; abnormal shutdown\n"));
+
+       } else {
+           /* make sure background threads have finished all of their asynchronous
+            * work on host and callback structures */
            FS_STATE_RDLOCK;
-       }
+           while (!fs_state.FiveMinuteLWP_tranquil ||
+                  !fs_state.HostCheckLWP_tranquil ||
+                  !fs_state.FsyncCheckLWP_tranquil) {
+               FS_LOCK;
+               FS_STATE_UNLOCK;
+               ViceLog(0, ("waiting for background host/callback threads to quiesce before saving fileserver state...\n"));
+               assert(pthread_cond_wait(&fs_state.worker_done_cv, &fileproc_glock_mutex) == 0);
+               FS_UNLOCK;
+               FS_STATE_RDLOCK;
+           }
 
-       /* ok. it should now be fairly safe. let's do the state dump */
-       fs_stateSave();
+           /* ok. it should now be fairly safe. let's do the state dump */
+           fs_stateSave();
+       }
     }
 #endif /* AFS_DEMAND_ATTACH_FS */
 
@@ -1909,6 +1948,7 @@ main(int argc, char *argv[])
     int curLimit;
     time_t t;
     afs_uint32 rx_bindhost;
+    VolumePackageOptions opts;
 
 #ifdef AFS_AIX32_ENV
     struct sigaction nsa;
@@ -2215,7 +2255,12 @@ main(int argc, char *argv[])
      * will be available "real soon now".  Worry about whether we can satisfy the 
      * calls in the volume package itself.
      */
-    if (VInitVolumePackage(fileServer, large, nSmallVns, 0, volcache)) {
+    VOptDefaults(fileServer, &opts);
+    opts.nLargeVnodes = large;
+    opts.nSmallVnodes = nSmallVns;
+    opts.volcache = volcache;
+
+    if (VInitVolumePackage2(fileServer, &opts)) {
        ViceLog(0,
                ("Shutting down: errors encountered initializing volume package\n"));
        VShutdown();