3 class Collection < ArvadosModel
7 include CommonApiTemplate
9 serialize :properties, Hash
11 before_validation :check_encoding
12 before_validation :check_manifest_validity
13 before_validation :check_signatures
14 before_validation :strip_signatures_and_update_replication_confirmed
15 validate :ensure_pdh_matches_manifest_text
16 before_save :set_file_names
18 # Query only undeleted collections by default.
19 default_scope where("expires_at IS NULL or expires_at > CURRENT_TIMESTAMP")
21 api_accessible :user, extend: :common do |t|
25 t.add :portable_data_hash
26 t.add :signed_manifest_text, as: :manifest_text
27 t.add :replication_desired
28 t.add :replication_confirmed
29 t.add :replication_confirmed_at
32 def self.attributes_required_columns
34 # If we don't list manifest_text explicitly, the
35 # params[:select] code gets confused by the way we
36 # expose signed_manifest_text as manifest_text in the
37 # API response, and never let clients select the
38 # manifest_text column.
39 'manifest_text' => ['manifest_text'],
43 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
45 return false if self.manifest_text.nil?
47 return true if current_user.andand.is_admin
49 # Provided the manifest_text hasn't changed materially since an
50 # earlier validation, it's safe to pass this validation on
51 # subsequent passes without checking any signatures. This is
52 # important because the signatures have probably been stripped off
53 # by the time we get to a second validation pass!
54 return true if @signatures_checked and @signatures_checked == computed_pdh
56 if self.manifest_text_changed?
57 # Check permissions on the collection manifest.
58 # If any signature cannot be verified, raise PermissionDeniedError
59 # which will return 403 Permission denied to the client.
60 api_token = current_api_client_authorization.andand.api_token
63 now: db_current_time.to_i,
65 self.manifest_text.each_line do |entry|
66 entry.split.each do |tok|
67 if tok == '.' or tok.starts_with? './'
69 elsif tok =~ FILE_TOKEN
70 # This is a filename token, not a blob locator. Note that we
71 # keep checking tokens after this, even though manifest
72 # format dictates that all subsequent tokens will also be
73 # filenames. Safety first!
74 elsif Blob.verify_signature tok, signing_opts
76 elsif Keep::Locator.parse(tok).andand.signature
77 # Signature provided, but verify_signature did not like it.
78 logger.warn "Invalid signature on locator #{tok}"
79 raise ArvadosModel::PermissionDeniedError
80 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
81 # No signature provided, but we are running in insecure mode.
82 logger.debug "Missing signature on locator #{tok} ignored"
83 elsif Blob.new(tok).empty?
84 # No signature provided -- but no data to protect, either.
86 logger.warn "Missing signature on locator #{tok}"
87 raise ArvadosModel::PermissionDeniedError
92 @signatures_checked = computed_pdh
95 def strip_signatures_and_update_replication_confirmed
96 if self.manifest_text_changed?
98 if not self.replication_confirmed.nil?
99 self.class.each_manifest_locator(manifest_text_was) do |match|
100 in_old_manifest[match[1]] = true
104 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
105 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
106 # If the new manifest_text contains locators whose hashes
107 # weren't in the old manifest_text, storage replication is no
109 self.replication_confirmed_at = nil
110 self.replication_confirmed = nil
113 # Return the locator with all permission signatures removed,
114 # but otherwise intact.
115 match[0].gsub(/\+A[^+]*/, '')
118 if @computed_pdh_for_manifest_text == manifest_text
119 # If the cached PDH was valid before stripping, it is still
120 # valid after stripping.
121 @computed_pdh_for_manifest_text = stripped_manifest.dup
124 self[:manifest_text] = stripped_manifest
129 def ensure_pdh_matches_manifest_text
130 if not manifest_text_changed? and not portable_data_hash_changed?
132 elsif portable_data_hash.nil? or not portable_data_hash_changed?
133 self.portable_data_hash = computed_pdh
134 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
135 errors.add(:portable_data_hash, "is not a valid locator")
137 elsif portable_data_hash[0..31] != computed_pdh[0..31]
138 errors.add(:portable_data_hash,
139 "does not match computed hash #{computed_pdh}")
142 # Ignore the client-provided size part: always store
143 # computed_pdh in the database.
144 self.portable_data_hash = computed_pdh
149 if self.manifest_text_changed?
150 self.file_names = manifest_files
157 if self.manifest_text
158 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
159 names << name.first.gsub('\040',' ') + "\n"
160 break if names.length > 2**12
164 if self.manifest_text and names.length < 2**12
165 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
166 names << stream_name.first.gsub('\040',' ') + "\n"
167 break if names.length > 2**12
175 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
179 # If Ruby thinks the encoding is something else, like 7-bit
180 # ASCII, but its stored bytes are equal to the (valid) UTF-8
181 # encoding of the same string, we declare it to be a UTF-8
184 utf8.force_encoding Encoding::UTF_8
185 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
191 errors.add :manifest_text, "must use UTF-8 encoding"
196 def check_manifest_validity
198 Keep::Manifest.validate! manifest_text if manifest_text
205 def signed_manifest_text
206 if has_attribute? :manifest_text
207 token = current_api_client_authorization.andand.api_token
208 @signed_manifest_text = self.class.sign_manifest manifest_text, token
212 def self.sign_manifest manifest, token
215 expire: db_current_time.to_i + Rails.configuration.blob_signature_ttl,
217 m = munge_manifest_locators(manifest) do |match|
218 Blob.sign_locator(match[0], signing_opts)
223 def self.munge_manifest_locators manifest
224 # Given a manifest text and a block, yield the regexp MatchData
225 # for each locator. Return a new manifest in which each locator
226 # has been replaced by the block's return value.
227 return nil if !manifest
228 return '' if manifest == ''
231 manifest.each_line do |line|
234 line.split(' ').each do |word|
237 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
238 new_words << yield(match)
243 new_lines << new_words.join(' ')
245 new_lines.join("\n") + "\n"
248 def self.each_manifest_locator manifest
249 # Given a manifest text and a block, yield the regexp match object
251 manifest.each_line do |line|
252 # line will have a trailing newline, but the last token is never
253 # a locator, so it's harmless here.
254 line.split(' ').each do |word|
255 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
262 def self.normalize_uuid uuid
265 uuid.split('+').each do |token|
266 if token.match /^[0-9a-f]{32,}$/
267 raise "uuid #{uuid} has multiple hash parts" if hash_part
269 elsif token.match /^\d+$/
270 raise "uuid #{uuid} has multiple size parts" if size_part
274 raise "uuid #{uuid} has no hash part" if !hash_part
275 [hash_part, size_part].compact.join '+'
278 # Return array of Collection objects
279 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
280 readers ||= [Thread.current[:user]]
282 readable_by(*readers).
283 readable_by(*readers, table_name: "collections").
284 joins("JOIN collections ON links.head_uuid = collections.uuid").
285 order("links.created_at DESC")
287 # If the search term is a Collection locator that contains one file
288 # that looks like a Docker image, return it.
289 if loc = Keep::Locator.parse(search_term)
291 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
293 # Check if the Collection contains exactly one file whose name
294 # looks like a saved Docker image.
295 manifest = Keep::Manifest.new(coll_match.manifest_text)
296 if manifest.exact_file_count?(1) and
297 (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/)
303 if search_tag.nil? and (n = search_term.index(":"))
304 search_tag = search_term[n+1..-1]
305 search_term = search_term[0..n-1]
308 # Find Collections with matching Docker image repository+tag pairs.
309 matches = base_search.
310 where(link_class: "docker_image_repo+tag",
311 name: "#{search_term}:#{search_tag || 'latest'}")
313 # If that didn't work, find Collections with matching Docker image hashes.
315 matches = base_search.
316 where("link_class = ? and links.name LIKE ?",
317 "docker_image_hash", "#{search_term}%")
320 # Generate an order key for each result. We want to order the results
321 # so that anything with an image timestamp is considered more recent than
322 # anything without; then we use the link's created_at as a tiebreaker.
324 matches.all.map do |link|
325 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
326 -link.created_at.to_i]
328 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
331 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
332 find_all_for_docker_image(search_term, search_tag, readers).first
335 def self.searchable_columns operator
336 super - ["manifest_text"]
339 def self.full_text_searchable_columns
340 super - ["manifest_text"]
344 def portable_manifest_text
345 self.class.munge_manifest_locators(manifest_text) do |match|
355 portable_manifest = portable_manifest_text
356 (Digest::MD5.hexdigest(portable_manifest) +
358 portable_manifest.bytesize.to_s)
362 if @computed_pdh_for_manifest_text == manifest_text
365 @computed_pdh = compute_pdh
366 @computed_pdh_for_manifest_text = manifest_text.dup
370 def ensure_permission_to_save
371 if (not current_user.andand.is_admin and
372 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
373 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
374 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")