3 class Collection < ArvadosModel
7 include CommonApiTemplate
9 serialize :properties, Hash
11 before_validation :check_encoding
12 before_validation :check_signatures
13 before_validation :strip_signatures_and_update_replication_confirmed
14 validate :ensure_pdh_matches_manifest_text
15 before_save :set_file_names
17 # Query only undeleted collections by default.
18 default_scope where("expires_at IS NULL or expires_at > CURRENT_TIMESTAMP")
20 api_accessible :user, extend: :common do |t|
24 t.add :portable_data_hash
25 t.add :signed_manifest_text, as: :manifest_text
26 t.add :replication_desired
27 t.add :replication_confirmed
28 t.add :replication_confirmed_at
31 def self.attributes_required_columns
33 # If we don't list manifest_text explicitly, the
34 # params[:select] code gets confused by the way we
35 # expose signed_manifest_text as manifest_text in the
36 # API response, and never let clients select the
37 # manifest_text column.
38 'manifest_text' => ['manifest_text'],
42 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
44 return false if self.manifest_text.nil?
46 return true if current_user.andand.is_admin
48 # Provided the manifest_text hasn't changed materially since an
49 # earlier validation, it's safe to pass this validation on
50 # subsequent passes without checking any signatures. This is
51 # important because the signatures have probably been stripped off
52 # by the time we get to a second validation pass!
53 return true if @signatures_checked and @signatures_checked == computed_pdh
55 if self.manifest_text_changed?
56 # Check permissions on the collection manifest.
57 # If any signature cannot be verified, raise PermissionDeniedError
58 # which will return 403 Permission denied to the client.
59 api_token = current_api_client_authorization.andand.api_token
62 now: db_current_time.to_i,
64 self.manifest_text.each_line do |entry|
65 entry.split.each do |tok|
66 if tok == '.' or tok.starts_with? './'
68 elsif tok =~ FILE_TOKEN
69 # This is a filename token, not a blob locator. Note that we
70 # keep checking tokens after this, even though manifest
71 # format dictates that all subsequent tokens will also be
72 # filenames. Safety first!
73 elsif Blob.verify_signature tok, signing_opts
75 elsif Keep::Locator.parse(tok).andand.signature
76 # Signature provided, but verify_signature did not like it.
77 logger.warn "Invalid signature on locator #{tok}"
78 raise ArvadosModel::PermissionDeniedError
79 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
80 # No signature provided, but we are running in insecure mode.
81 logger.debug "Missing signature on locator #{tok} ignored"
82 elsif Blob.new(tok).empty?
83 # No signature provided -- but no data to protect, either.
85 logger.warn "Missing signature on locator #{tok}"
86 raise ArvadosModel::PermissionDeniedError
91 @signatures_checked = computed_pdh
94 def strip_signatures_and_update_replication_confirmed
95 if self.manifest_text_changed?
97 if not self.replication_confirmed.nil?
98 self.class.each_manifest_locator(manifest_text_was) do |match|
99 in_old_manifest[match[1]] = true
103 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
104 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
105 # If the new manifest_text contains locators whose hashes
106 # weren't in the old manifest_text, storage replication is no
108 self.replication_confirmed_at = nil
109 self.replication_confirmed = nil
112 # Return the locator with all permission signatures removed,
113 # but otherwise intact.
114 match[0].gsub(/\+A[^+]*/, '')
117 if @computed_pdh_for_manifest_text == manifest_text
118 # If the cached PDH was valid before stripping, it is still
119 # valid after stripping.
120 @computed_pdh_for_manifest_text = stripped_manifest.dup
123 self[:manifest_text] = stripped_manifest
128 def ensure_pdh_matches_manifest_text
129 if not manifest_text_changed? and not portable_data_hash_changed?
131 elsif portable_data_hash.nil? or not portable_data_hash_changed?
132 self.portable_data_hash = computed_pdh
133 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
134 errors.add(:portable_data_hash, "is not a valid locator")
136 elsif portable_data_hash[0..31] != computed_pdh[0..31]
137 errors.add(:portable_data_hash,
138 "does not match computed hash #{computed_pdh}")
141 # Ignore the client-provided size part: always store
142 # computed_pdh in the database.
143 self.portable_data_hash = computed_pdh
148 if self.manifest_text_changed?
149 self.file_names = manifest_files
156 if self.manifest_text
157 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
158 names << name.first.gsub('\040',' ') + "\n"
159 break if names.length > 2**12
163 if self.manifest_text and names.length < 2**12
164 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
165 names << stream_name.first.gsub('\040',' ') + "\n"
166 break if names.length > 2**12
174 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
178 # If Ruby thinks the encoding is something else, like 7-bit
179 # ASCII, but its stored bytes are equal to the (valid) UTF-8
180 # encoding of the same string, we declare it to be a UTF-8
183 utf8.force_encoding Encoding::UTF_8
184 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
190 errors.add :manifest_text, "must use UTF-8 encoding"
195 def signed_manifest_text
196 if has_attribute? :manifest_text
197 token = current_api_client_authorization.andand.api_token
198 @signed_manifest_text = self.class.sign_manifest manifest_text, token
202 def self.sign_manifest manifest, token
205 expire: db_current_time.to_i + Rails.configuration.blob_signature_ttl,
207 m = munge_manifest_locators(manifest) do |match|
208 Blob.sign_locator(match[0], signing_opts)
213 def self.munge_manifest_locators manifest
214 # Given a manifest text and a block, yield the regexp MatchData
215 # for each locator. Return a new manifest in which each locator
216 # has been replaced by the block's return value.
217 return nil if !manifest
218 return '' if manifest == ''
221 manifest.each_line do |line|
224 line.split(' ').each do |word|
227 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
228 new_words << yield(match)
233 new_lines << new_words.join(' ')
235 new_lines.join("\n") + "\n"
238 def self.each_manifest_locator manifest
239 # Given a manifest text and a block, yield the regexp match object
241 manifest.each_line do |line|
242 # line will have a trailing newline, but the last token is never
243 # a locator, so it's harmless here.
244 line.split(' ').each do |word|
245 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
252 def self.normalize_uuid uuid
255 uuid.split('+').each do |token|
256 if token.match /^[0-9a-f]{32,}$/
257 raise "uuid #{uuid} has multiple hash parts" if hash_part
259 elsif token.match /^\d+$/
260 raise "uuid #{uuid} has multiple size parts" if size_part
264 raise "uuid #{uuid} has no hash part" if !hash_part
265 [hash_part, size_part].compact.join '+'
268 # Return array of Collection objects
269 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
270 readers ||= [Thread.current[:user]]
272 readable_by(*readers).
273 readable_by(*readers, table_name: "collections").
274 joins("JOIN collections ON links.head_uuid = collections.uuid").
275 order("links.created_at DESC")
277 # If the search term is a Collection locator that contains one file
278 # that looks like a Docker image, return it.
279 if loc = Keep::Locator.parse(search_term)
281 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
283 # Check if the Collection contains exactly one file whose name
284 # looks like a saved Docker image.
285 manifest = Keep::Manifest.new(coll_match.manifest_text)
286 if manifest.exact_file_count?(1) and
287 (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/)
293 if search_tag.nil? and (n = search_term.index(":"))
294 search_tag = search_term[n+1..-1]
295 search_term = search_term[0..n-1]
298 # Find Collections with matching Docker image repository+tag pairs.
299 matches = base_search.
300 where(link_class: "docker_image_repo+tag",
301 name: "#{search_term}:#{search_tag || 'latest'}")
303 # If that didn't work, find Collections with matching Docker image hashes.
305 matches = base_search.
306 where("link_class = ? and links.name LIKE ?",
307 "docker_image_hash", "#{search_term}%")
310 # Generate an order key for each result. We want to order the results
311 # so that anything with an image timestamp is considered more recent than
312 # anything without; then we use the link's created_at as a tiebreaker.
314 matches.all.map do |link|
315 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
316 -link.created_at.to_i]
318 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
321 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
322 find_all_for_docker_image(search_term, search_tag, readers).first
325 def self.searchable_columns operator
326 super - ["manifest_text"]
329 def self.full_text_searchable_columns
330 super - ["manifest_text"]
334 def portable_manifest_text
335 self.class.munge_manifest_locators(manifest_text) do |match|
345 portable_manifest = portable_manifest_text
346 (Digest::MD5.hexdigest(portable_manifest) +
348 portable_manifest.bytesize.to_s)
352 if @computed_pdh_for_manifest_text == manifest_text
355 @computed_pdh = compute_pdh
356 @computed_pdh_for_manifest_text = manifest_text.dup
360 def ensure_permission_to_save
361 if (not current_user.andand.is_admin and
362 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
363 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
364 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")