From 3d3501347eac5ca2db4bfd37d0af471984c56b48 Mon Sep 17 00:00:00 2001
From: Aaron Klingaman <alk@cs.princeton.edu>
Date: Wed, 1 Jun 2005 19:54:57 +0000
Subject: [PATCH] better handle the kexec failure cases by notifying the user
 of the exact problem, and canceling further boot processes

---
 source/BootManager.py         | 21 +++++++++++----------
 source/steps/ChainBootNode.py | 24 +++++++++++++++++++++---
 2 files changed, 32 insertions(+), 13 deletions(-)

diff --git a/source/BootManager.py b/source/BootManager.py
index 478c714..ea8015a 100755
--- a/source/BootManager.py
+++ b/source/BootManager.py
@@ -120,16 +120,17 @@ class log:
         """
         upload the contents of the log to the server
         """
-    
-        self.LogEntry( "Uploading logs to %s" % UPLOAD_LOG_URL )
-        
-        self.OutputFile.close()
-        self.OutputFile= None
-                
-        curl_cmd= "%s -s --connect-timeout 60 --max-time 600 " \
-                  "--form log=@%s --form ifconfig=\</tmp/ifconfig %s" % \
-                  (CURL_PATH, self.OutputFilePath, UPLOAD_LOG_URL)
-        os.system( curl_cmd )
+
+        if self.OutputFile is not None:
+            self.LogEntry( "Uploading logs to %s" % UPLOAD_LOG_URL )
+            
+            self.OutputFile.close()
+            self.OutputFile= None
+            
+            curl_cmd= "%s -s --connect-timeout 60 --max-time 600 " \
+                      "--form log=@%s --form ifconfig=\</tmp/ifconfig %s" % \
+                      (CURL_PATH, self.OutputFilePath, UPLOAD_LOG_URL)
+            os.system( curl_cmd )
         
     
 
diff --git a/source/steps/ChainBootNode.py b/source/steps/ChainBootNode.py
index 9cc16b8..181b934 100644
--- a/source/steps/ChainBootNode.py
+++ b/source/steps/ChainBootNode.py
@@ -109,6 +109,11 @@ def Run( vars, log ):
     # further use of log after Upload will only output to screen
     log.Upload()
 
+    # regardless of whether kexec works or not, we need to stop trying to
+    # run anything
+    cancel_boot_flag= "/tmp/CANCEL_BOOT"
+    utils.sysexec( "touch %s" % cancel_boot_flag, log )
+
     # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything to
     # get kexec to work correctly
     
@@ -131,8 +136,21 @@ def Run( vars, log ):
                 utils.sysexec_noerr( "modprobe -r %s" % module, log )
     except IOError:
         log.write( "Couldn't load /tmp/loadedmodules to unload, continuing.\n" )
-    
-    utils.sysexec( "kexec --force --initrd=/tmp/initrd " \
-                   "--append=ramdisk_size=8192 /tmp/kernel" )
+
+    try:
+        utils.sysexec( "kexec --force --initrd=/tmp/initrd " \
+                       "--append=ramdisk_size=8192 /tmp/kernel" )
+    except BootManagerException, e:
+        # if kexec fails, we've shut the machine down to a point where nothing
+        # can run usefully anymore (network down, all modules unloaded, file
+        # systems unmounted. write out the error, and cancel the boot process
+
+        log.write( "\n\n" )
+        log.write( "-------------------------------------------------------\n" )
+        log.write( "kexec failed with the following error. Please report\n" )
+        log.write( "this problem to support@planet-lab.org.\n\n" )
+        log.write( str(e) + "\n\n" )
+        log.write( "The boot process has been canceled.\n" )
+        log.write( "-------------------------------------------------------\n\n" )
 
     return
-- 
2.43.0