debian-mirror-gitlab/lib/gitlab/middleware/multipart.rb

137 lines
4.7 KiB
Ruby
Raw Normal View History

2019-02-15 15:39:39 +05:30
# frozen_string_literal: true
2017-08-17 22:00:37 +05:30
# Gitlab::Middleware::Multipart - a Rack::Multipart replacement
#
# Rack::Multipart leaves behind tempfiles in /tmp and uses valuable Ruby
# process time to copy files around. This alternative solution uses
# gitlab-workhorse to clean up the tempfiles and puts the tempfiles in a
# location where copying should not be needed.
#
# When gitlab-workhorse finds files in a multipart MIME body it sends
# a signed message via a request header. This message lists the names of
# the multipart entries that gitlab-workhorse filtered out of the
# multipart structure and saved to tempfiles. Workhorse adds new entries
# in the multipart structure with paths to the tempfiles.
#
# The job of this Rack middleware is to detect and decode the message
# from workhorse. If present, it walks the Rack 'params' hash for the
# current request, opens the respective tempfiles, and inserts the open
# Ruby File objects in the params hash where Rack::Multipart would have
# put them. The goal is that application code deeper down can keep
# working the way it did with Rack::Multipart without changes.
#
# CAVEAT: the code that modifies the params hash is a bit complex. It is
# conceivable that certain Rack params structures will not be modified
# correctly. We are not aware of such bugs at this time though.
#
module Gitlab
module Middleware
class Multipart
2019-12-04 20:38:33 +05:30
RACK_ENV_KEY = 'HTTP_GITLAB_WORKHORSE_MULTIPART_FIELDS'
2017-08-17 22:00:37 +05:30
class Handler
def initialize(env, message)
2019-12-04 20:38:33 +05:30
@request = Rack::Request.new(env)
2017-08-17 22:00:37 +05:30
@rewritten_fields = message['rewritten_fields']
@open_files = []
end
def with_open_files
@rewritten_fields.each do |field, tmp_path|
parsed_field = Rack::Utils.parse_nested_query(field)
raise "unexpected field: #{field.inspect}" unless parsed_field.count == 1
key, value = parsed_field.first
if value.nil?
2018-11-08 19:23:39 +05:30
value = open_file(@request.params, key)
2017-08-17 22:00:37 +05:30
@open_files << value
else
2018-11-08 19:23:39 +05:30
value = decorate_params_value(value, @request.params[key])
2017-08-17 22:00:37 +05:30
end
2018-03-17 18:26:18 +05:30
2019-12-04 20:38:33 +05:30
update_param(key, value)
2017-08-17 22:00:37 +05:30
end
yield
ensure
@open_files.each(&:close)
end
# This function calls itself recursively
2018-11-08 19:23:39 +05:30
def decorate_params_value(path_hash, value_hash)
2017-08-17 22:00:37 +05:30
unless path_hash.is_a?(Hash) && path_hash.count == 1
raise "invalid path: #{path_hash.inspect}"
end
2018-03-17 18:26:18 +05:30
2017-08-17 22:00:37 +05:30
path_key, path_value = path_hash.first
unless value_hash.is_a?(Hash) && value_hash[path_key]
raise "invalid value hash: #{value_hash.inspect}"
end
case path_value
when nil
2018-11-08 19:23:39 +05:30
value_hash[path_key] = open_file(value_hash.dig(path_key), '')
2017-08-17 22:00:37 +05:30
@open_files << value_hash[path_key]
value_hash
when Hash
2018-11-08 19:23:39 +05:30
decorate_params_value(path_value, value_hash[path_key])
2017-08-17 22:00:37 +05:30
value_hash
else
raise "unexpected path value: #{path_value.inspect}"
end
end
2018-11-08 19:23:39 +05:30
def open_file(params, key)
2018-11-20 20:47:30 +05:30
::UploadedFile.from_params(params, key, allowed_paths)
2017-08-17 22:00:37 +05:30
end
2019-12-04 20:38:33 +05:30
# update_params ensures that both rails controllers and rack middleware can find
# workhorse accelerate files in the request
def update_param(key, value)
# we make sure we have key in POST otherwise update_params will add it in GET
@request.POST[key] ||= value
# this will force Rack::Request to properly update env keys
@request.update_param(key, value)
# ActionDispatch::Request is based on Rack::Request but it caches params
# inside other env keys, here we ensure everything is updated correctly
ActionDispatch::Request.new(@request.env).update_param(key, value)
end
2020-03-28 13:19:24 +05:30
private
def allowed_paths
[
::FileUploader.root,
Gitlab.config.uploads.storage_path,
2020-04-15 14:45:12 +05:30
JobArtifactUploader.workhorse_upload_path,
2020-03-28 13:19:24 +05:30
File.join(Rails.root, 'public/uploads/tmp')
]
end
2017-08-17 22:00:37 +05:30
end
def initialize(app)
@app = app
end
def call(env)
encoded_message = env.delete(RACK_ENV_KEY)
return @app.call(env) if encoded_message.blank?
message = Gitlab::Workhorse.decode_jwt(encoded_message)[0]
Handler.new(env, message).with_open_files do
@app.call(env)
end
2020-04-15 14:45:12 +05:30
rescue UploadedFile::InvalidPathError => e
[400, { 'Content-Type' => 'text/plain' }, e.message]
2017-08-17 22:00:37 +05:30
end
end
end
end
2020-03-28 13:19:24 +05:30
::Gitlab::Middleware::Multipart::Handler.prepend_if_ee('EE::Gitlab::Middleware::Multipart::Handler')