debian-mirror-gitlab/lib/gitlab/background_migration/encrypt_columns.rb

105 lines
3.6 KiB
Ruby
Raw Normal View History

2018-12-05 23:21:45 +05:30
# frozen_string_literal: true
module Gitlab
module BackgroundMigration
# EncryptColumn migrates data from an unencrypted column - `foo`, say - to
# an encrypted column - `encrypted_foo`, say.
#
2019-02-15 15:39:39 +05:30
# To avoid depending on a particular version of the model in app/, add a
# model to `lib/gitlab/background_migration/models/encrypt_columns` and use
# it in the migration that enqueues the jobs, so code can be shared.
#
2018-12-05 23:21:45 +05:30
# For this background migration to work, the table that is migrated _has_ to
# have an `id` column as the primary key. Additionally, the encrypted column
# should be managed by attr_encrypted, and map to an attribute with the same
# name as the unencrypted column (i.e., the unencrypted column should be
2019-02-15 15:39:39 +05:30
# shadowed), unless you want to define specific methods / accessors in the
# temporary model in `/models/encrypt_columns/your_model.rb`.
2018-12-05 23:21:45 +05:30
#
class EncryptColumns
def perform(model, attributes, from, to)
model = model.constantize if model.is_a?(String)
# If sidekiq hasn't undergone a restart, its idea of what columns are
# present may be inaccurate, so ensure this is as fresh as possible
model.reset_column_information
model.define_attribute_methods
attributes = expand_attributes(model, Array(attributes).map(&:to_sym))
model.transaction do
# Use SELECT ... FOR UPDATE to prevent the value being changed while
# we are encrypting it
relation = model.where(id: from..to).lock
relation.each do |instance|
encrypt!(instance, attributes)
end
end
end
2019-02-15 15:39:39 +05:30
def clear_migrated_values?
true
end
2018-12-05 23:21:45 +05:30
private
# Build a hash of { attribute => encrypted column name }
def expand_attributes(klass, attributes)
expanded = attributes.flat_map do |attribute|
attr_config = klass.encrypted_attributes[attribute]
crypt_column_name = attr_config&.fetch(:attribute)
raise "Couldn't determine encrypted column for #{klass}##{attribute}" if
crypt_column_name.nil?
raise "#{klass} source column: #{attribute} is missing" unless
klass.column_names.include?(attribute.to_s)
# Running the migration without the destination column being present
# leads to data loss
raise "#{klass} destination column: #{crypt_column_name} is missing" unless
klass.column_names.include?(crypt_column_name.to_s)
[attribute, crypt_column_name]
end
Hash[*expanded]
end
# Generate ciphertext for each column and update the database
def encrypt!(instance, attributes)
to_clear = attributes
.map { |plain, crypt| apply_attribute!(instance, plain, crypt) }
.compact
.flat_map { |plain| [plain, nil] }
to_clear = Hash[*to_clear]
if instance.changed?
instance.save!
2019-02-15 15:39:39 +05:30
if clear_migrated_values?
instance.update_columns(to_clear)
end
2018-12-05 23:21:45 +05:30
end
end
def apply_attribute!(instance, plain_column, crypt_column)
plaintext = instance[plain_column]
ciphertext = instance[crypt_column]
# No need to do anything if the plaintext is nil, or an encrypted
# value already exists
2019-07-07 11:18:12 +05:30
return unless plaintext.present?
return if ciphertext.present?
2018-12-05 23:21:45 +05:30
# attr_encrypted will calculate and set the expected value for us
instance.public_send("#{plain_column}=", plaintext) # rubocop:disable GitlabSecurity/PublicSend
plain_column
end
end
end
end