[Ocfs2-commits] mfasheh commits r1407 - trunk

svn-commits at oss.oracle.com svn-commits at oss.oracle.com
Tue Aug 31 12:52:10 CDT 2004


Author: mfasheh
Date: 2004-08-31 12:52:08 -0500 (Tue, 31 Aug 2004)
New Revision: 1407

Modified:
   trunk/TODO
Log:
* update the TODO list.



Modified: trunk/TODO
===================================================================
--- trunk/TODO	2004-08-31 01:50:11 UTC (rev 1406)
+++ trunk/TODO	2004-08-31 17:52:08 UTC (rev 1407)
@@ -1,7 +1,3 @@
-* figure out what to do about old style cdsl. Either we error when we see it
-  (in the module), or we provide an fsck to find and fix them.
-	- Probably should be handle by the conversion.
-
 * Make bitmap reads only read those blocks which we care about
 
 * make slabs for: ocfs_journal_handle, and ocfs_journal_copyout
@@ -24,8 +20,8 @@
 
 * check if we really need the INODE_DELETED flag anymore and kill it if we
   don't.
+       - Yes, we want this as it helps us detect stale inodes.
 
-
 Pre 1.0:
 * redo the delete and rename journal credits now that those paths are
   so different
@@ -46,17 +42,23 @@
 * Find all the ocfs_sleep calls and change the values to something
   much much lower.
 
-Post 1.0:
+* start_trans should be done *after* as many cluster locks as
+  possible. The way we do it harkens back to when everything under the
+  sun was protected by "trans_lock". Nowadays, holding a transaction
+  open before you've even got your cluster locks just has alot of
+  deadlock potential. The next two become easier once this is done:
+
 * in ocfs_extend_file (and possibly other places in the code) we read
   an inode and get/check some values from it just before taking the 
   disk lock. We should really be doing that *after* taking the disk lock...
 
+* space allocation has to be fixed to be able to deal with non contiguous space
+
+Post 1.0:
 * restartable truncate
 
 * Endian fixes
 
-* space allocation has to be fixed to be able to deal with non contiguous space
-
 * need to move the commit_bits stuff outside of the journalling code
 
 * we need to re-evaluate all the release_lock calls and see which ones



More information about the Ocfs2-commits mailing list