Add some documentation

doc/implementation_notes.md

@@ -0,0 +1,33 @@
+Implementation notes
+====================
+
+Symlink attacks
+---------------
+
+MAT2 output predictable filenames (like yourfile.jpg.cleaned).
+This may lead to symlink attack. Please check if you OS prevent
+against them
+
+Archives handling
+-----------------
+
+MAT2 doesn't support archives yet, because we haven't found an usable way to ask the user
+what to do when a non-supported files are encountered.
+
+PDF handling
+------------
+
+MAT was doing some kind of rendering for PDF files, on a cairo surface, then
+printed it to a file. This kept the text selectable, but unfortunately, it
+didn't remove any *deep metadata*, like the ones in embedded pictures. This was
+on of the reason MAT was abandoned: the absence of satisfying solution to
+handle PDF. But apparently, people are ok with [pdf redact
+tools](https://github.com/firstlookmedia/pdf-redact-tools), that simply
+transform the PDF into images. So this is what's MAT2 is doing too.
+
+Images handling
+---------------
+
+When possible, images are handled like PDF: rendered on a surface, then saved
+to the filesystem. This ensures that every metadata is removed.
+

doc/threat_model.md

@@ -0,0 +1,85 @@
+Threat Model
+============
+The Metadata Anonymisation Toolkit 2 adversary has a number
+of goals, capabilities, and counter-attack types that can be
+used to guide us towards a set of requirements for the MAT2.
+
+This is an overhaul of MAT's (the first iteration of the software) one.
+
+Warnings
+--------
+
+Mat only removes standard metadata from your files, it does _not_:
+
+  - anonymise their content
+  - handle watermarking
+  - handle steganography
+  - handle any non-standard metadata field/system
+
+If you really want to be anonymous format that does not contain any
+metadata, or better : use plain-text. And as usual, think before clicking.
+
+
+Adversary
+------------
+
+* Goals:
+
+    - Identifying the source of the document, since a document
+    always has one. Who/where/when/how was a picture
+    taken, where was the document leaked from and by
+    whom, ...
+
+    - Identify the author; in some cases documents may be
+    anonymously authored or created. In these cases,
+    identifying the author is the goal.
+
+    - Identify the equipment/software used. If the attacker fails
+    to directly identify the author and/or source, his next
+    goal is to determine the source of the equipment used
+    to produce, copy, and transmit the document. This can
+    include the model of camera used to take a photo, or
+    which software was used to produce an office document.
+
+
+* Adversary Capabilities - Positioning
+    - The adversary created the document specifically for this
+    user. This is the strongest position for the adversary to
+    have. In this case, the adversary is capable of inserting
+    arbitrary, custom watermarks specifically for tracking
+    the user. In general, MAT cannot defend against this
+    adversary, but we list it for completeness.
+
+    - The adversary created the document for a group of users.
+    In this case, the adversary knows that they attempted to
+    limit distribution to a specific group of users. They may
+    or may not have watermarked the document for these
+    users, but they certainly know the format used.
+
+    - The adversary did not create the document, the weakest
+    position for the adversary to have. The file format is (most of the time)
+     standard, nothing custom is added: MAT
+    should be able to remove all meta-information from the
+    file.
+
+Requirements
+---------------
+
+* Processing
+    - The MAT2 *should* avoid interactions with information.
+    Its goal is to remove metadata, and the user is solely
+    responsible for the information of the file.
+
+    - The MAT2 *must* warn when encountering an unknown
+    format. For example, in a zipfile, if MAT encounters an
+    unknown format, it should warn the user, and ask if the
+    file should be added to the anonymised archive that is
+    produced.
+
+    - The MAT2 *must* not add metadata, since its purpose is to
+    anonymise files: every added items of metadata decreases
+    anonymity.
+
+    - The MAT2 *should* handle unknown/hidden metadata fields,
+    like proprietary extensions of open formats.
+