3 class Collection < ArvadosModel
7 include CommonApiTemplate
9 serialize :properties, Hash
11 before_validation :default_empty_manifest
12 before_validation :check_encoding
13 before_validation :check_manifest_validity
14 before_validation :check_signatures
15 before_validation :strip_signatures_and_update_replication_confirmed
16 validate :ensure_pdh_matches_manifest_text
17 before_save :set_file_names
18 before_save :expires_at_not_in_past
20 # Query only undeleted collections by default.
21 default_scope where("expires_at IS NULL or expires_at > statement_timestamp()")
23 api_accessible :user, extend: :common do |t|
27 t.add :portable_data_hash
28 t.add :signed_manifest_text, as: :manifest_text
29 t.add :replication_desired
30 t.add :replication_confirmed
31 t.add :replication_confirmed_at
35 def self.attributes_required_columns
37 # If we don't list manifest_text explicitly, the
38 # params[:select] code gets confused by the way we
39 # expose signed_manifest_text as manifest_text in the
40 # API response, and never let clients select the
41 # manifest_text column.
42 'manifest_text' => ['manifest_text'],
46 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
48 return false if self.manifest_text.nil?
50 return true if current_user.andand.is_admin
52 # Provided the manifest_text hasn't changed materially since an
53 # earlier validation, it's safe to pass this validation on
54 # subsequent passes without checking any signatures. This is
55 # important because the signatures have probably been stripped off
56 # by the time we get to a second validation pass!
57 return true if @signatures_checked and @signatures_checked == computed_pdh
59 if self.manifest_text_changed?
60 # Check permissions on the collection manifest.
61 # If any signature cannot be verified, raise PermissionDeniedError
62 # which will return 403 Permission denied to the client.
63 api_token = current_api_client_authorization.andand.api_token
66 now: db_current_time.to_i,
68 self.manifest_text.each_line do |entry|
69 entry.split.each do |tok|
70 if tok == '.' or tok.starts_with? './'
72 elsif tok =~ FILE_TOKEN
73 # This is a filename token, not a blob locator. Note that we
74 # keep checking tokens after this, even though manifest
75 # format dictates that all subsequent tokens will also be
76 # filenames. Safety first!
77 elsif Blob.verify_signature tok, signing_opts
79 elsif Keep::Locator.parse(tok).andand.signature
80 # Signature provided, but verify_signature did not like it.
81 logger.warn "Invalid signature on locator #{tok}"
82 raise ArvadosModel::PermissionDeniedError
83 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
84 # No signature provided, but we are running in insecure mode.
85 logger.debug "Missing signature on locator #{tok} ignored"
86 elsif Blob.new(tok).empty?
87 # No signature provided -- but no data to protect, either.
89 logger.warn "Missing signature on locator #{tok}"
90 raise ArvadosModel::PermissionDeniedError
95 @signatures_checked = computed_pdh
98 def strip_signatures_and_update_replication_confirmed
99 if self.manifest_text_changed?
101 if not self.replication_confirmed.nil?
102 self.class.each_manifest_locator(manifest_text_was) do |match|
103 in_old_manifest[match[1]] = true
107 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
108 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
109 # If the new manifest_text contains locators whose hashes
110 # weren't in the old manifest_text, storage replication is no
112 self.replication_confirmed_at = nil
113 self.replication_confirmed = nil
116 # Return the locator with all permission signatures removed,
117 # but otherwise intact.
118 match[0].gsub(/\+A[^+]*/, '')
121 if @computed_pdh_for_manifest_text == manifest_text
122 # If the cached PDH was valid before stripping, it is still
123 # valid after stripping.
124 @computed_pdh_for_manifest_text = stripped_manifest.dup
127 self[:manifest_text] = stripped_manifest
132 def ensure_pdh_matches_manifest_text
133 if not manifest_text_changed? and not portable_data_hash_changed?
135 elsif portable_data_hash.nil? or not portable_data_hash_changed?
136 self.portable_data_hash = computed_pdh
137 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
138 errors.add(:portable_data_hash, "is not a valid locator")
140 elsif portable_data_hash[0..31] != computed_pdh[0..31]
141 errors.add(:portable_data_hash,
142 "does not match computed hash #{computed_pdh}")
145 # Ignore the client-provided size part: always store
146 # computed_pdh in the database.
147 self.portable_data_hash = computed_pdh
152 if self.manifest_text_changed?
153 self.file_names = manifest_files
160 if self.manifest_text
161 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
162 names << name.first.gsub('\040',' ') + "\n"
163 break if names.length > 2**12
167 if self.manifest_text and names.length < 2**12
168 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
169 names << stream_name.first.gsub('\040',' ') + "\n"
170 break if names.length > 2**12
177 def default_empty_manifest
178 self.manifest_text ||= ''
182 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
186 # If Ruby thinks the encoding is something else, like 7-bit
187 # ASCII, but its stored bytes are equal to the (valid) UTF-8
188 # encoding of the same string, we declare it to be a UTF-8
191 utf8.force_encoding Encoding::UTF_8
192 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
198 errors.add :manifest_text, "must use UTF-8 encoding"
203 def check_manifest_validity
205 Keep::Manifest.validate! manifest_text
207 rescue ArgumentError => e
208 errors.add :manifest_text, e.message
213 def signed_manifest_text
214 if has_attribute? :manifest_text
215 token = current_api_client_authorization.andand.api_token
216 @signed_manifest_text = self.class.sign_manifest manifest_text, token
220 def self.sign_manifest manifest, token
223 expire: db_current_time.to_i + Rails.configuration.blob_signature_ttl,
225 m = munge_manifest_locators(manifest) do |match|
226 Blob.sign_locator(match[0], signing_opts)
231 def self.munge_manifest_locators manifest
232 # Given a manifest text and a block, yield the regexp MatchData
233 # for each locator. Return a new manifest in which each locator
234 # has been replaced by the block's return value.
235 return nil if !manifest
236 return '' if manifest == ''
239 manifest.each_line do |line|
242 line.split(' ').each do |word|
245 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
246 new_words << yield(match)
251 new_lines << new_words.join(' ')
253 new_lines.join("\n") + "\n"
256 def self.each_manifest_locator manifest
257 # Given a manifest text and a block, yield the regexp match object
259 manifest.each_line do |line|
260 # line will have a trailing newline, but the last token is never
261 # a locator, so it's harmless here.
262 line.split(' ').each do |word|
263 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
270 def self.normalize_uuid uuid
273 uuid.split('+').each do |token|
274 if token.match /^[0-9a-f]{32,}$/
275 raise "uuid #{uuid} has multiple hash parts" if hash_part
277 elsif token.match /^\d+$/
278 raise "uuid #{uuid} has multiple size parts" if size_part
282 raise "uuid #{uuid} has no hash part" if !hash_part
283 [hash_part, size_part].compact.join '+'
286 # Return array of Collection objects
287 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
288 readers ||= [Thread.current[:user]]
290 readable_by(*readers).
291 readable_by(*readers, table_name: "collections").
292 joins("JOIN collections ON links.head_uuid = collections.uuid").
293 order("links.created_at DESC")
295 # If the search term is a Collection locator that contains one file
296 # that looks like a Docker image, return it.
297 if loc = Keep::Locator.parse(search_term)
299 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
301 # Check if the Collection contains exactly one file whose name
302 # looks like a saved Docker image.
303 manifest = Keep::Manifest.new(coll_match.manifest_text)
304 if manifest.exact_file_count?(1) and
305 (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/)
311 if search_tag.nil? and (n = search_term.index(":"))
312 search_tag = search_term[n+1..-1]
313 search_term = search_term[0..n-1]
316 # Find Collections with matching Docker image repository+tag pairs.
317 matches = base_search.
318 where(link_class: "docker_image_repo+tag",
319 name: "#{search_term}:#{search_tag || 'latest'}")
321 # If that didn't work, find Collections with matching Docker image hashes.
323 matches = base_search.
324 where("link_class = ? and links.name LIKE ?",
325 "docker_image_hash", "#{search_term}%")
328 # Generate an order key for each result. We want to order the results
329 # so that anything with an image timestamp is considered more recent than
330 # anything without; then we use the link's created_at as a tiebreaker.
332 matches.all.map do |link|
333 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
334 -link.created_at.to_i]
336 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
339 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
340 find_all_for_docker_image(search_term, search_tag, readers).first
343 def self.searchable_columns operator
344 super - ["manifest_text"]
347 def self.full_text_searchable_columns
348 super - ["manifest_text"]
352 def portable_manifest_text
353 self.class.munge_manifest_locators(manifest_text) do |match|
363 portable_manifest = portable_manifest_text
364 (Digest::MD5.hexdigest(portable_manifest) +
366 portable_manifest.bytesize.to_s)
370 if @computed_pdh_for_manifest_text == manifest_text
373 @computed_pdh = compute_pdh
374 @computed_pdh_for_manifest_text = manifest_text.dup
378 def ensure_permission_to_save
379 if (not current_user.andand.is_admin and
380 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
381 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
382 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
387 # If expires_at is being changed to a time in the past, change it to
388 # now. This allows clients to say "expires {client-current-time}"
389 # without failing due to clock skew, while avoiding odd log entries
390 # like "expiry date changed to {1 year ago}".
391 def expires_at_not_in_past
392 if expires_at_changed? and expires_at
393 self.expires_at = [db_current_time, expires_at].max