GitLab

This cuts down on the number of copies and fixes some other bits:

- Add a fuzzer.

- Document a place where the original implementation did not match the
  specification. (The final record's size is a little iffy. We probably
  want a small spec tweak.)

- Use a streaming SHA-256 implementation, rather than making a copy to
  stick the 0 or 1 in the hash.

- If there is no more room in the output (the consumer may be issuing
  smaller reads), stop processing input. There is no need to make a copy
  of the entire input. MerkleIntegeritySourceStream only needs to buffer
  at most one record. (Ideally we wouldn't even do and instead
  coordinate with the base class's read buffer, but that would require
  tweaking the FilteredSourceStream interface. This CL addresses the
  easy stuff.)

- Fix O(N^2) behavior if the caller issues tiny reads in the buffered
  output.

- If the record is entirely in the input buffer (common case), don't
  make a copy to extract it.

- If the output fits entirely in the output buffer (common case), don't
  make a copy to return it.

- Flesh out missing tests, based on code coverage tools and important
  security checks (notably truncation).

  (For others trying to repeat the coverage bits: this file was
  unfortunately placed in //content rather than //net, so I wasn't able
  to get the coverage tools to work without hacking it into
  net_unittests locally. It seems the X server dependency is
  problematic for tools/code_coverage?? Also content_unittests is huge.)

- s/MI-256/MI-SHA256/. There are other 256-bit hashes.

Bug: 814591
Change-Id: If927d3f49085a5bec31939846c9a55f8903da34a
Reviewed-on: https://chromium-review.googlesource.com/981798Reviewed-by: Yutaka Hirano <yhirano@chromium.org>
Reviewed-by: Matt Menke <mmenke@chromium.org>
Reviewed-by: Max Moroz <mmoroz@chromium.org>
Commit-Queue: David Benjamin <davidben@chromium.org>
Cr-Commit-Position: refs/heads/master@{#547874}