Blind Index

Securely search encrypted database fields

Works with Lockbox (full example) and attr_encrypted (full example)

Learn more about securing sensitive data in Rails

We use this approach by Scott Arciszewski. To summarize, we compute a keyed hash of the sensitive data and store it in a column. To query, we apply the keyed hash function to the value we’re searching and then perform a database search. This results in performant queries for exact matches. Efficient LIKE queries are not possible, but you can index expressions.

An important consideration in searchable encryption is leakage, which is information an attacker can gain. Blind indexing leaks that rows have the same value. If you use this for a field like last name, an attacker can use frequency analysis to predict the values. In an active attack where an attacker can control the input values, they can learn which other values in the database match.

Here’s a great article on leakage in searchable encryption. Blind indexing has the same leakage as deterministic encryption.

Add this line to your application’s Gemfile:

gem "blind_index"

Your model should already be set up with Lockbox or attr_encrypted. The examples are for a User model with has_encrypted :email or attr_encrypted :email. See the full examples for Lockbox and attr_encrypted if needed.

Also, if you use attr_encrypted, generate a key.

Create a migration to add a column for the blind index

add_column :users, :email_bidx, :string
add_index :users, :email_bidx # unique: true if needed

Add to your model

class User < ApplicationRecord
  blind_index :email
end

For more sensitive fields, use

class User < ApplicationRecord
  blind_index :email, slow: true
end

Backfill existing records

BlindIndex.backfill(User)

And query away

User.where(email: "test@example.org")

You can apply expressions to attributes before indexing and searching. This gives you the the ability to perform case-insensitive searches and more.

class User < ApplicationRecord
  blind_index :email, expression: ->(v) { v.downcase }
end

You can use blind indexes for uniqueness validations.

class User < ApplicationRecord
  validates :email, uniqueness: true
end

We recommend adding a unique index to the blind index column through a database migration.

add_index :users, :email_bidx, unique: true

For allow_blank: true, use:

class User < ApplicationRecord
  blind_index :email, expression: ->(v) { v.presence }
  validates :email, uniqueness: {allow_blank: true}
end

For case_sensitive: false, use:

class User < ApplicationRecord
  blind_index :email, expression: ->(v) { v.downcase }
  validates :email, uniqueness: true # for best performance, leave out {case_sensitive: false}
end

You may want multiple blind indexes for an attribute. To do this, add another column:

add_column :users, :email_ci_bidx, :string
add_index :users, :email_ci_bidx

Update your model

class User < ApplicationRecord
  blind_index :email
  blind_index :email_ci, attribute: :email, expression: ->(v) { v.downcase }
end

Backfill existing records

BlindIndex.backfill(User, columns: [:email_ci_bidx])

And query away

User.where(email_ci: "test@example.org")

If you don’t need to store the original value (for instance, when just checking duplicates), use a virtual attribute:

class User < ApplicationRecord
  attribute :email, :string
  blind_index :email
end

You can also use virtual attributes to index data from multiple columns:

class User < ApplicationRecord
  attribute :initials, :string
  blind_index :initials

  before_validation :set_initials, if: -> { changes.key?(:first_name) || changes.key?(:last_name) }

  def set_initials
    self.initials = "#{first_name[0]}#{last_name[0]}"
  end
end

If you’re encrypting a column and adding a blind index at the same time, use the migrating option.

class User < ApplicationRecord
  blind_index :email, migrating: true
end

This allows you to backfill records while still querying the unencrypted field.

BlindIndex.backfill(User)

Once that completes, you can remove the migrating option.

To rotate keys without downtime, add a new column:

add_column :users, :email_bidx_v2, :string
add_index :users, :email_bidx_v2

And add to your model

class User < ApplicationRecord
  blind_index :email, rotate: {version: 2, master_key: ENV["BLIND_INDEX_MASTER_KEY_V2"]}
end

This will keep the new column synced going forward. Next, backfill the data:

BlindIndex.backfill(User, columns: [:email_bidx_v2])

Then update your model

class User < ApplicationRecord
  blind_index :email, version: 2, master_key: ENV["BLIND_INDEX_MASTER_KEY_V2"]
end

Finally, drop the old column.

The master key is used to generate unique keys for each blind index. This technique comes from CipherSweet. The table name and blind index column name are both used in this process.

You can get an individual key with:

BlindIndex.index_key(table: "users", bidx_attribute: "email_bidx")

To rename a table with blind indexes, use:

class User < ApplicationRecord
  blind_index :email, key_table: "original_table"
end

To rename a blind index column, use:

class User < ApplicationRecord
  blind_index :email, key_attribute: "original_column"
end

Argon2id is used for best security. The default cost parameters are 3 iterations and 4 MB of memory. For slow: true, the cost parameters are 4 iterations and 32 MB of memory.

A number of other algorithms are also supported. Unless you have specific reasons to use them, go with Argon2id.

You can use blind indexes in fixtures with:

test_user:
  email_bidx: <%= User.generate_email_bidx("test@example.org").inspect %>

Be sure to include the inspect at the end or it won’t be encoded properly in YAML.

For Mongoid, use:

class User
  field :email_bidx, type: String
  index({email_bidx: 1})
end

This is optional for Lockbox, as its master key is used by default.

Generate a key with:

BlindIndex.generate_key

Store the key with your other secrets. This is typically Rails credentials or an environment variable (dotenv is great for this). Be sure to use different keys in development and production. Keys don’t need to be hex-encoded, but it’s often easier to store them this way.

Set the following environment variable with your key (you can use this one in development)

BLIND_INDEX_MASTER_KEY=ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff

or create config/initializers/blind_index.rb with something like

BlindIndex.master_key = Rails.application.credentials.blind_index_master_key

Unfortunately, blind indexes can’t be used for LIKE, ILIKE, or full-text searching. Instead, records must be loaded, decrypted, and searched in memory.

For LIKE, use:

User.select { |u| u.email.include?("value") }

For ILIKE, use:

User.select { |u| u.email =~ /value/i }

For full-text or fuzzy searching, use a gem like FuzzyMatch:

FuzzyMatch.new(User.all, read: :email).find("value")

If the number of records is large, try to find a way to narrow it down. An expression index is one way to do this, but leaks which records have the same value of the expression, so use it carefully.

Set default options in an initializer with:

BlindIndex.default_options = {algorithm: :pbkdf2_sha256}

By default, blind indexes are encoded in Base64. Set a different encoding with:

class User < ApplicationRecord
  blind_index :email, encode: ->(v) { [v].pack("H*") }
end

By default, blind indexes are 32 bytes. Set a smaller size with:

class User < ApplicationRecord
  blind_index :email, size: 16
end

Set a key directly for an index with:

class User < ApplicationRecord
  blind_index :email, key: ENV["USER_EMAIL_BLIND_INDEX_KEY"]
end

You can generate blind indexes from other languages as well. For Python, you can use argon2-cffi.

from argon2.low_level import Type, hash_secret_raw
from base64 import b64encode

key = '289737bab72fa97b1f4b081cef00d7b7d75034bcf3183c363feaf3e6441777bc'
value = 'test@example.org'

bidx = b64encode(hash_secret_raw(
    secret=value.encode(),
    salt=bytes.fromhex(key),
    time_cost=3,
    memory_cost=2**12,
    parallelism=1,
    hash_len=32,
    type=Type.ID
))

One alternative to blind indexing is to use a deterministic encryption scheme, like AES-SIV. In this approach, the encrypted data will be the same for matches. We recommend blind indexing over deterministic encryption because:

1. You can keep encryption consistent for all fields (both searchable and non-searchable)
2. Blind indexing supports expressions

View the changelog

Everyone is encouraged to help improve this project. Here are a few ways you can help:

• Report bugs
• Fix bugs and submit pull requests
• Write, clarify, or fix documentation
• Suggest or add new features

To get started with development and testing:

git clone https://github.com/ankane/blind_index.git
cd blind_index
bundle install
bundle exec rake test

For security issues, send an email to the address on this page.