This is a general list of things which should/could/may be done.
If any of these features interest you let me know- especially if you're
willing and able to help code it.

Legend:
    - = Not started
    + = Done
    O = Mostly done
    o = Started work
    . = Canceled
    ? = To think about

+ Look at VLAN (802.1q) packets
    - others non-vanilla types?
    + Add tags?  Remove tags?  Change tags?
    - Tag only one side of the connection
    - Cisco's ISL trunking?

- Add support for MPLS

- Add support for GRE
  http://www.linuxguruz.com/iptables/howto/2.4routing-5.html
  Perhaps this should be done via the hardware interface rather then the GRE
  virtual interface since libnet doesn't support the GRE virtual

+ Add support for setting the ethernet protocol field so we can use
    -I, -K to fill out an entire ethernet header w/o using -2

+ Add a secondary interface full layer two rewrite option

+ Fix MAC rewriting to allow sending packets with a MAC of 00:00:00:00:00:00

- Add support for more linktypes (Prism Monitor, 802.11, FDDI, etc)
    - Make it easier for others to add support for others

+ Rip out packet munger from tcpreplay and put it into another tool so
  that tcpreplay can be more optimized
    ? perhaps use libnetdude?
    ? make into a library?
    + definately put it into a seperate binary (tcprewrite)

- Add the ability to modify packet data via regex(es) in tcprewrite
  - Should support pcre
  - Support (foo) and $1, etc so new data can include old
  - Limit matching which packets via BPF filter and tcpprep cache
        (client/server)
  - Step through packets ala tcpreplay and provide option to edit (Y/n)

+ Improve config file format
  + better variable names
  + use "var: value" format
  + have tcpreplay, tcpprep, tcprewrite sections
  + Being solved using GNU AutoOpts

. Add support for dual-nic send on one intf, wait for packet, send next.
  would be really useful for testing the effectiveness of how well an IPS
  detects and blocks attacks. (TP's tomahawk does this even better then
  described here, so why re-invent the wheel?)

- Support fragrouter like features 
    - basic IP fragmenation
    - TCP fudging 
    - then more advanced stuff
    - Can we integrate FR's code?

- Support connection tracking and generating 3way handshake for connections
  missing them.

- Bump Syn/Ack numbers by a random or given value so that running 
  the same pcap will behave as different streams.

- Improve flowreplay so it actually works
  o Use libnids to read the pcaps
  - Allow handoff to a socket after user specified client/server exchanges

- Perhaps integrate stick/snot/fpg logic into flowreplay:
  http://www.geschke-online.de/FLoP/fpg.8.html
  to do full 3way handshakes

- IPv6 support?  People ask for this every few months, but nobody actually
  says they "need" or "really want" it; seems more of "gee, wouldn't it be
  nice".

+ When splitting traffic via tcpprep print out each packet (tcpdump style)
  so end users know where each packet is going

+ Improve autoconf detection of libraries

+ Re-organize source tree

O tcpdump decoder should print packets syncronously w/ the main process

- Rewrite do_sleep() to handle sub sleep times by only nanosleep()'ing
  once for multiple packets when the timestamps are close enough.  We
  also need to time nanosleep, since different architectures have lower
  minimum sleep times (Linux/Alpha is 1ms vs. 10ms for Linux/x86)

+ Tcpreplay should say which interface each packet is going out

+ Better use of GNU Autotools

+ Improve CLI/config file parsing

- Tcprewrite should be able to remove the two byte ethernet FCS (checksums)
  at the end of the frame.

- See about removing libnet_init() from all binaries other then tcprewrite
  so we don't have to run as root:
  . libnet_addr2name4 (ignore, doesn't require libnet_t context)
  + libnet_name2addr4
  - libnet_get_hwaddr
  - libnet_do_checksum

+ Support randomization of IP addresses in ARP packets

- Only tcpreplay should need to run as root.

- Tcpreplay should use raw sockets or BPF directly for writing rather then
      libnet where applicable for higher performance.