feat: Add support for ISO 32000-2 AES256 encryption

[Darkheir]

Pull request

ISO 32000-2 is bringing a new password derivation algorithm that must be applied before using the key to decrypt the document.

This algorithm is needed for documents using the Version 5, Revision 6. PDF 2.0 is also deprecating Version 5 Revision 5 because of security issues.

This PR aims to support this new algorithm.

How Has This Been Tested?

Added tests

Checklist

• I have added tests that prove my fix is effective or that my feature
  works
• I have added docstrings to newly created methods and classes
• I have optimized the code at least one time after creating the initial
  version
• I have updated the README.md or I am verified that this
  is not necessary
• I have updated the readthedocs documentation or I
  verified that this is not necessary
• I have added a consice human-readable description of the change to
  CHANGELOG.md

[pietermarsman]

Hi @Darkheir,

Thanks for this work.

If I understand correctly this adds functionality that is added with the PDF 2.0 specification.

To do a sensible review I want to look up the PDF 2.0 specification but it seems that you have to pay for it. Do you know if that is true? And where did you get your copy of the 2.0 specification?

[Darkheir]

Hi @pietermarsman

Sadly I didn't get access to the specification since you have to pay for it as you said.

I found a python project that implemented the feature: https://github.com/MatthiasValvekens/pyHanko/blob/master/pyhanko/pdf_utils/crypt.py

The code is based on this implementation, I only changed it to fit with the pdfminer code base.

I also added an encrypted pdf to test the implemented feature.

[MatthiasValvekens]

Hi, I'm the author of the linked implementation on which this one was based. I happen to own a (legal) copy of ISO 32000-2:2020 that I purchased from ISO through the usual channels. Unfortunately, since ISO licenses are single-user, I'm not at liberty to share the document. I also can't vouch for the correctness of this adaptation.

However, even if you don't have access to the official standard, that doesn't mean that you can't do any interoperability testing at all. There are quite a few implementations of PDF 2.0 encryption out there (both open-source and not) that you could use to experiment. For example, iText has had support for this stuff pretty much since the very first revision of ISO 32000-2 came out: https://github.com/itext/itext7/blob/develop/kernel/src/main/java/com/itextpdf/kernel/crypto/securityhandler/StandardHandlerUsingAes256.java.

---

Also: my implementation is MIT licensed, so you're of course free to modify & integrate it into this project, but would you mind adding a notice to make sure the attribution clause in the license is adhered to? It's no big deal, but I'd still appreciate the courtesy. Thanks ;)

(I also have a copy of the license for PyPDF2 in my repo, since a lot of the low-level PDF code heavily borrows from it.)

[pietermarsman]

I'm going to assume that this works, because the tests work.

[pietermarsman]

So this MR needs 3 more changes I think:

• Use None instead of b''
• Fix CHANGELOG.md conflict
• Include reference to original implementation as @MatthiasValvekens suggested

[MatthiasValvekens]

I also added a couple of comments.

Another general one: correct me if I'm wrong, but I don't think you're validating the integrity of the Perms entry anywhere in the authentication process. Was that intentional? It's part of the key retrieval/password validation procedure for revision 6 security handlers. Right now, it doesn't provide much in the way of integrity protection, but If nothing else, it's a good sanity check.

[pietermarsman]

@Darkheir @MatthiasValvekens Thanks for the work and support on this!

[MartinThoma]

@Darkheir would you be interested in adding modern encryption support to PyPDF2 as well?

https://github.com/py-pdf/PyPDF2

We recently added decryption support. Only the encryption is missing