[Ocfs2-commits] zab commits r1748 - trunk/cluster

[svn-commits at oss.oracle.com]

Author: zab
Date: 2005-01-07 17:39:44 -0600 (Fri, 07 Jan 2005)
New Revision: 1748

Modified:
   trunk/cluster/tcp.c
Log:
o add a little todo list from an audit that we can clean
  up in the next few days


Modified: trunk/cluster/tcp.c
===================================================================
--- trunk/cluster/tcp.c	2005-01-07 06:00:41 UTC (rev 1747)
+++ trunk/cluster/tcp.c	2005-01-07 23:39:44 UTC (rev 1748)
@@ -23,6 +23,34 @@
  * Boston, MA 021110-1307, USA.
  *
  * Authors: Kurt Hackel
+ *
+ * XXX we should resolve these before release
+ * 	- no recv_list, use data_ready
+ * 	- kill all 'lock; list_for_each() { unlock; lock; }' patterns
+ * 	- make sure max lens are enforced.. alloc max, don't send > max
+ * 	- disable preemt before calling rx handler when debugging
+ * 	- pack/swab packets
+ * 	- more conservative locking patterns to avoid double unlocks
+ * 	- find explicit stack call to drain rx queue
+ * 	- simplify rx thread exit path (completion, etc)
+ * 	- goto out style exiting
+ * 	- push net_handler_lock into lookup_handler, add add call
+ * 	- implement net_remove_handlers
+ * 	- check stack to be sure that blocking rx waits for all data
+ * 	- if spin sending be sure to exclude concurrent racing senders
+ * 	- refcounting around sock against tx/teardown/etc
+ * 	- get sin/iov/msg off the stack, per sock structures
+ * 	- share code between net_send_message{,arr}() 
+ * 	- clean up net_msg_num
+ * 	- net_send_tcp_msg() callers should pass an iov
+ * 	- just rely on rx calls to discover busted sockets?
+ * 	- make sk_state another define
+ * 	- add trivial version trading message at the start of a conn
+ *
+ * 	- document caller block, serialized blocking rx handling
+ * 	- move gsd into its own file
+ * 	- move to userspace connection management?
+ * 	
  */
 #include "warning_hack.h"