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.
43 # We need expires_at to determine the correct
44 # timestamp in signed_manifest_text.
45 'manifest_text' => ['manifest_text', 'expires_at'],
49 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
51 return false if self.manifest_text.nil?
53 return true if current_user.andand.is_admin
55 # Provided the manifest_text hasn't changed materially since an
56 # earlier validation, it's safe to pass this validation on
57 # subsequent passes without checking any signatures. This is
58 # important because the signatures have probably been stripped off
59 # by the time we get to a second validation pass!
60 return true if @signatures_checked and @signatures_checked == computed_pdh
62 if self.manifest_text_changed?
63 # Check permissions on the collection manifest.
64 # If any signature cannot be verified, raise PermissionDeniedError
65 # which will return 403 Permission denied to the client.
66 api_token = current_api_client_authorization.andand.api_token
69 now: db_current_time.to_i,
71 self.manifest_text.each_line do |entry|
72 entry.split.each do |tok|
73 if tok == '.' or tok.starts_with? './'
75 elsif tok =~ FILE_TOKEN
76 # This is a filename token, not a blob locator. Note that we
77 # keep checking tokens after this, even though manifest
78 # format dictates that all subsequent tokens will also be
79 # filenames. Safety first!
80 elsif Blob.verify_signature tok, signing_opts
82 elsif Keep::Locator.parse(tok).andand.signature
83 # Signature provided, but verify_signature did not like it.
84 logger.warn "Invalid signature on locator #{tok}"
85 raise ArvadosModel::PermissionDeniedError
86 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
87 # No signature provided, but we are running in insecure mode.
88 logger.debug "Missing signature on locator #{tok} ignored"
89 elsif Blob.new(tok).empty?
90 # No signature provided -- but no data to protect, either.
92 logger.warn "Missing signature on locator #{tok}"
93 raise ArvadosModel::PermissionDeniedError
98 @signatures_checked = computed_pdh
101 def strip_signatures_and_update_replication_confirmed
102 if self.manifest_text_changed?
104 if not self.replication_confirmed.nil?
105 self.class.each_manifest_locator(manifest_text_was) do |match|
106 in_old_manifest[match[1]] = true
110 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
111 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
112 # If the new manifest_text contains locators whose hashes
113 # weren't in the old manifest_text, storage replication is no
115 self.replication_confirmed_at = nil
116 self.replication_confirmed = nil
119 # Return the locator with all permission signatures removed,
120 # but otherwise intact.
121 match[0].gsub(/\+A[^+]*/, '')
124 if @computed_pdh_for_manifest_text == manifest_text
125 # If the cached PDH was valid before stripping, it is still
126 # valid after stripping.
127 @computed_pdh_for_manifest_text = stripped_manifest.dup
130 self[:manifest_text] = stripped_manifest
135 def ensure_pdh_matches_manifest_text
136 if not manifest_text_changed? and not portable_data_hash_changed?
138 elsif portable_data_hash.nil? or not portable_data_hash_changed?
139 self.portable_data_hash = computed_pdh
140 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
141 errors.add(:portable_data_hash, "is not a valid locator")
143 elsif portable_data_hash[0..31] != computed_pdh[0..31]
144 errors.add(:portable_data_hash,
145 "does not match computed hash #{computed_pdh}")
148 # Ignore the client-provided size part: always store
149 # computed_pdh in the database.
150 self.portable_data_hash = computed_pdh
155 if self.manifest_text_changed?
156 self.file_names = manifest_files
163 if self.manifest_text
164 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
165 names << name.first.gsub('\040',' ') + "\n"
166 break if names.length > 2**12
170 if self.manifest_text and names.length < 2**12
171 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
172 names << stream_name.first.gsub('\040',' ') + "\n"
173 break if names.length > 2**12
180 def default_empty_manifest
181 self.manifest_text ||= ''
185 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
189 # If Ruby thinks the encoding is something else, like 7-bit
190 # ASCII, but its stored bytes are equal to the (valid) UTF-8
191 # encoding of the same string, we declare it to be a UTF-8
194 utf8.force_encoding Encoding::UTF_8
195 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
201 errors.add :manifest_text, "must use UTF-8 encoding"
206 def check_manifest_validity
208 Keep::Manifest.validate! manifest_text
210 rescue ArgumentError => e
211 errors.add :manifest_text, e.message
216 def signed_manifest_text
217 if has_attribute? :manifest_text
218 token = current_api_client_authorization.andand.api_token
219 exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
220 expires_at].compact.map(&:to_i).min
221 @signed_manifest_text = self.class.sign_manifest manifest_text, token, exp
225 def self.sign_manifest manifest, token, exp=nil
227 exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
233 m = munge_manifest_locators(manifest) do |match|
234 Blob.sign_locator(match[0], signing_opts)
239 def self.munge_manifest_locators manifest
240 # Given a manifest text and a block, yield the regexp MatchData
241 # for each locator. Return a new manifest in which each locator
242 # has been replaced by the block's return value.
243 return nil if !manifest
244 return '' if manifest == ''
247 manifest.each_line do |line|
250 line.split(' ').each do |word|
253 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
254 new_words << yield(match)
259 new_lines << new_words.join(' ')
261 new_lines.join("\n") + "\n"
264 def self.each_manifest_locator manifest
265 # Given a manifest text and a block, yield the regexp match object
267 manifest.each_line do |line|
268 # line will have a trailing newline, but the last token is never
269 # a locator, so it's harmless here.
270 line.split(' ').each do |word|
271 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
278 def self.normalize_uuid uuid
281 uuid.split('+').each do |token|
282 if token.match /^[0-9a-f]{32,}$/
283 raise "uuid #{uuid} has multiple hash parts" if hash_part
285 elsif token.match /^\d+$/
286 raise "uuid #{uuid} has multiple size parts" if size_part
290 raise "uuid #{uuid} has no hash part" if !hash_part
291 [hash_part, size_part].compact.join '+'
294 # Return array of Collection objects
295 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
296 readers ||= [Thread.current[:user]]
298 readable_by(*readers).
299 readable_by(*readers, table_name: "collections").
300 joins("JOIN collections ON links.head_uuid = collections.uuid").
301 order("links.created_at DESC")
303 # If the search term is a Collection locator that contains one file
304 # that looks like a Docker image, return it.
305 if loc = Keep::Locator.parse(search_term)
307 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
309 # Check if the Collection contains exactly one file whose name
310 # looks like a saved Docker image.
311 manifest = Keep::Manifest.new(coll_match.manifest_text)
312 if manifest.exact_file_count?(1) and
313 (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/)
319 if search_tag.nil? and (n = search_term.index(":"))
320 search_tag = search_term[n+1..-1]
321 search_term = search_term[0..n-1]
324 # Find Collections with matching Docker image repository+tag pairs.
325 matches = base_search.
326 where(link_class: "docker_image_repo+tag",
327 name: "#{search_term}:#{search_tag || 'latest'}")
329 # If that didn't work, find Collections with matching Docker image hashes.
331 matches = base_search.
332 where("link_class = ? and links.name LIKE ?",
333 "docker_image_hash", "#{search_term}%")
336 # Generate an order key for each result. We want to order the results
337 # so that anything with an image timestamp is considered more recent than
338 # anything without; then we use the link's created_at as a tiebreaker.
340 matches.all.map do |link|
341 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
342 -link.created_at.to_i]
344 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
347 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
348 find_all_for_docker_image(search_term, search_tag, readers).first
351 def self.searchable_columns operator
352 super - ["manifest_text"]
355 def self.full_text_searchable_columns
356 super - ["manifest_text"]
359 def logged_attributes
360 attrs = attributes.dup
361 attrs.delete('manifest_text')
366 def portable_manifest_text
367 self.class.munge_manifest_locators(manifest_text) do |match|
377 portable_manifest = portable_manifest_text
378 (Digest::MD5.hexdigest(portable_manifest) +
380 portable_manifest.bytesize.to_s)
384 if @computed_pdh_for_manifest_text == manifest_text
387 @computed_pdh = compute_pdh
388 @computed_pdh_for_manifest_text = manifest_text.dup
392 def ensure_permission_to_save
393 if (not current_user.andand.is_admin and
394 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
395 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
396 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
401 # If expires_at is being changed to a time in the past, change it to
402 # now. This allows clients to say "expires {client-current-time}"
403 # without failing due to clock skew, while avoiding odd log entries
404 # like "expiry date changed to {1 year ago}".
405 def expires_at_not_in_past
406 if expires_at_changed? and expires_at
407 self.expires_at = [db_current_time, expires_at].max