Watermarking: Difference between revisions

From NESdev Wiki
Jump to navigationJump to search
(→‎Shuffling: EPIC win)
(→‎Compression: ELs to BBS topics prompting this page)
Line 42: Line 42:
If your program includes compressed data, you can change the interpretation of bits in the data format.
If your program includes compressed data, you can change the interpretation of bits in the data format.
For example, in RLE compression of graphics, the sense of bits denoting a run of repeated pixels vs. bits denoting a run of several literal pixels can be inverted.
For example, in RLE compression of graphics, the sense of bits denoting a run of repeated pixels vs. bits denoting a run of several literal pixels can be inverted.
== External links ==
*Based on BBS topics [http://nesdev.parodius.com/bbs/viewtopic.php?t=6046 6046] and [http://nesdev.parodius.com/bbs/viewtopic.php?t=6197 6197]

Revision as of 18:35, 20 January 2011

Watermarking is defined by Wikipedia as "the process of embedding information into a digital signal in a way that is difficult to remove." In some cases, the developer of an unreleased NES program wants to distribute to beta testers but still trace any leaked copies of the program to the tester who broke the non-disclosure agreement. There are several ways to produce binaries that can be traced back to a particular recipient.

Shuffling

One way to make each copy unique is to shuffle, or randomly rearrange, pieces of a program at compile time.

A code preprocessor can randomize the order of statically allocated variables in a program. This causes the addresses embedded in the code to change every time the program is compiled. It has benefits beyond watermarking: as the program is shuffled, the effects of a buffer overflow may become more apparent.

A code preprocessor can shuffle the order of subroutines in a program. Watch out: A common technique on the NES is the "inline tail call", in which a subroutine doesn't return but instead falls off the end into the following subroutine. You'll need to take this into account when adding markup to control the preprocessor.

A code preprocessor can shuffle the order of instructions in a subroutine. In a lot of cases, the order of instructions doesn't matter, such as LDA vs. CLC, or LDX vs. LDY where neither is indexed, or STA (d),Y vs. DEX, or STA of the same value to several different variables. Writes to some PPU registers act similarly: when setting up $2000 and $2001 at the end of vertical blanking, it doesn't matter which is written first. The markup for such "bundles of instructions" resembles the markup for explicitly parallel instruction computing. Adding the required markup also has a benefit beyond watermarking: it forces a code review, which allows a programmer to refamiliarize himself with a code base and possibly discover defects.

A common method to cope with bus conflicts on discrete mappers brings in another trick. For example, a game using UNROM might load from a table and write back to that table to make sure that the written bits match the bits in ROM. (See the banktable example at Programming UNROM.) But if you shuffle the data in banks 0 through 6 and shuffle the bank numbers in the table, you can make 7! = 5040 different binaries from this alone.

Even in a game no bigger than NROM-128, shuffling alone allows for more distinct binaries than the number of atoms in the known universe squared. With the size of NES games and with modern solid archiving tools such as 7-Zip, you can save each binary that you send out to each tester and still not fill a 4 GB USB flash drive. As long as the binary doesn't get leaked to someone with the knowledge to disassemble and reassemble a binary (as in SMBDis), computing the Hamming distance between the leaked copy and your saved copies is likely to result in a high-confidence match to the leaker.

Instruction encoding

A few instructions have multiple encodings. A code preprocessor can introduce any of several NOP instructions at random points in a non-time-critical subroutine.

Graphics changes

Graphics can depend on the build:

  • Choose one of several alternatives for grass and other noisy tiles
  • Tester's name or something derived from tester's name on the title screen. This is easy to remove, but it acts as a deterrent.
  • Tester's name or something derived from tester's name on a sign in a building in the game

Compression

If your program includes compressed data, you can change the interpretation of bits in the data format. For example, in RLE compression of graphics, the sense of bits denoting a run of repeated pixels vs. bits denoting a run of several literal pixels can be inverted.

External links