1 files changed, 41 insertions, 0 deletions
diff --git a/todo/henri.txt b/todo/henri.txt
new file mode 100644
index 0000000..cca8814
--- /dev/null
+++ b/todo/henri.txt
@@ -0,0 +1,41 @@
+o Proper SSL support in proxy mode.
+  - A naive implementation relying on the current code would probably look
+    horrible (at least my own attempts did). I believe that nsock should
+    internally be able to SSLify a plain TCP connection. It doesn't have to be
+    exported but it should be implemented just like the other operations. Then
+    it would be trivial (and clean) for the library to SSLify the channel
+    established by the proxy hooks.
+  - When redesigning nsock SSL code, keep in mind the ability to establish a SSL
+    session and still expose the raw TCP. That can be convenient when auditing
+    the SSL/TLS layer.
+o Don't drop pending writes when deleting the corresponding IOD. For nsock to
+  behave a bit like standard BSD sockets we should flush writes on close. (OTOH
+  anything which isn't ack'ed has no meaning, caller can still cancel it
+  typically...)
+o Give IODs their own methods to streamline the code and get rid of all
+  the special cases in nsock_core.c. This would also make it easier to
+  hook operations (typically: override the default iod_connect() method
+  to establish a proxy chain).
+o Fix the read API (!)
+o Profile the pcap code. It needs cleanup (for sure) and optimizations (maybe).
+o Proxy authentication
+o Handle socks4a
+  - This requires to figure out how to trigger proxy code without
+    resolving target hostname first. The problem is that the proxy code
+    is supposed to be a transparent hook of connect()... Extending the
+    exported API will probably be needed :(
+  - Async hostname resolution available from within nsock would let us
+    try clever tricks... I'm not sure whether nsock should provide it
+    or if it should simply provide an API to plug an external system.
+o Socks5 support
+o Some code is copied from ncat. I should move it to nbase.
+o Replace event lists by more efficient data structures. Consider using
+  a radix tree to map event IDs to pointers. Another solution would
+  be to put them all into a single RB-tree (TODO: validate BSD_HACK_MODE
+  & stuff). Encoding the event type in the ID's MSB would let us do inorder
+  traversal with connect events first, then read, then write...
+  {NOTE: It'd be cool for the beauty of it, but my tests reveal that as of Oct.
+  2013 there's no big bottleneck there.}
+o Rework the filespace code to avoid unneeded data copy. Scatter/gather
+  I/O might be useful there. Same task can also be expressed as: "profile and
+  optimize the usual nmap nsock I/O patterns."