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 :manifest_text, as: :unsigned_manifest_text
30 t.add :replication_desired
31 t.add :replication_confirmed
32 t.add :replication_confirmed_at
36 def self.attributes_required_columns
38 # If we don't list manifest_text explicitly, the
39 # params[:select] code gets confused by the way we
40 # expose signed_manifest_text as manifest_text in the
41 # API response, and never let clients select the
42 # manifest_text column.
44 # We need expires_at to determine the correct
45 # timestamp in signed_manifest_text.
46 'manifest_text' => ['manifest_text', 'expires_at'],
47 'unsigned_manifest_text' => ['manifest_text'],
51 def self.ignored_select_attributes
52 super + ["updated_at", "file_names"]
55 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
57 return false if self.manifest_text.nil?
59 return true if current_user.andand.is_admin
61 # Provided the manifest_text hasn't changed materially since an
62 # earlier validation, it's safe to pass this validation on
63 # subsequent passes without checking any signatures. This is
64 # important because the signatures have probably been stripped off
65 # by the time we get to a second validation pass!
66 return true if @signatures_checked and @signatures_checked == computed_pdh
68 if self.manifest_text_changed?
69 # Check permissions on the collection manifest.
70 # If any signature cannot be verified, raise PermissionDeniedError
71 # which will return 403 Permission denied to the client.
72 api_token = current_api_client_authorization.andand.api_token
75 now: db_current_time.to_i,
77 self.manifest_text.each_line do |entry|
78 entry.split.each do |tok|
79 if tok == '.' or tok.starts_with? './'
81 elsif tok =~ FILE_TOKEN
82 # This is a filename token, not a blob locator. Note that we
83 # keep checking tokens after this, even though manifest
84 # format dictates that all subsequent tokens will also be
85 # filenames. Safety first!
86 elsif Blob.verify_signature tok, signing_opts
88 elsif Keep::Locator.parse(tok).andand.signature
89 # Signature provided, but verify_signature did not like it.
90 logger.warn "Invalid signature on locator #{tok}"
91 raise ArvadosModel::PermissionDeniedError
92 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
93 # No signature provided, but we are running in insecure mode.
94 logger.debug "Missing signature on locator #{tok} ignored"
95 elsif Blob.new(tok).empty?
96 # No signature provided -- but no data to protect, either.
98 logger.warn "Missing signature on locator #{tok}"
99 raise ArvadosModel::PermissionDeniedError
104 @signatures_checked = computed_pdh
107 def strip_signatures_and_update_replication_confirmed
108 if self.manifest_text_changed?
110 if not self.replication_confirmed.nil?
111 self.class.each_manifest_locator(manifest_text_was) do |match|
112 in_old_manifest[match[1]] = true
116 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
117 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
118 # If the new manifest_text contains locators whose hashes
119 # weren't in the old manifest_text, storage replication is no
121 self.replication_confirmed_at = nil
122 self.replication_confirmed = nil
125 # Return the locator with all permission signatures removed,
126 # but otherwise intact.
127 match[0].gsub(/\+A[^+]*/, '')
130 if @computed_pdh_for_manifest_text == manifest_text
131 # If the cached PDH was valid before stripping, it is still
132 # valid after stripping.
133 @computed_pdh_for_manifest_text = stripped_manifest.dup
136 self[:manifest_text] = stripped_manifest
141 def ensure_pdh_matches_manifest_text
142 if not manifest_text_changed? and not portable_data_hash_changed?
144 elsif portable_data_hash.nil? or not portable_data_hash_changed?
145 self.portable_data_hash = computed_pdh
146 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
147 errors.add(:portable_data_hash, "is not a valid locator")
149 elsif portable_data_hash[0..31] != computed_pdh[0..31]
150 errors.add(:portable_data_hash,
151 "does not match computed hash #{computed_pdh}")
154 # Ignore the client-provided size part: always store
155 # computed_pdh in the database.
156 self.portable_data_hash = computed_pdh
161 if self.manifest_text_changed?
162 self.file_names = manifest_files
169 if self.manifest_text
170 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
171 names << name.first.gsub('\040',' ') + "\n"
172 break if names.length > 2**12
176 if self.manifest_text and names.length < 2**12
177 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
178 names << stream_name.first.gsub('\040',' ') + "\n"
179 break if names.length > 2**12
186 def default_empty_manifest
187 self.manifest_text ||= ''
191 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
195 # If Ruby thinks the encoding is something else, like 7-bit
196 # ASCII, but its stored bytes are equal to the (valid) UTF-8
197 # encoding of the same string, we declare it to be a UTF-8
200 utf8.force_encoding Encoding::UTF_8
201 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
207 errors.add :manifest_text, "must use UTF-8 encoding"
212 def check_manifest_validity
214 Keep::Manifest.validate! manifest_text
216 rescue ArgumentError => e
217 errors.add :manifest_text, e.message
222 def signed_manifest_text
223 if has_attribute? :manifest_text
224 token = current_api_client_authorization.andand.api_token
225 exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
226 expires_at].compact.map(&:to_i).min
227 @signed_manifest_text = self.class.sign_manifest manifest_text, token, exp
231 def self.sign_manifest manifest, token, exp=nil
233 exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
239 m = munge_manifest_locators(manifest) do |match|
240 Blob.sign_locator(match[0], signing_opts)
245 def self.munge_manifest_locators manifest
246 # Given a manifest text and a block, yield the regexp MatchData
247 # for each locator. Return a new manifest in which each locator
248 # has been replaced by the block's return value.
249 return nil if !manifest
250 return '' if manifest == ''
253 manifest.each_line do |line|
256 line.split(' ').each do |word|
259 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
260 new_words << yield(match)
265 new_lines << new_words.join(' ')
267 new_lines.join("\n") + "\n"
270 def self.each_manifest_locator manifest
271 # Given a manifest text and a block, yield the regexp match object
273 manifest.each_line do |line|
274 # line will have a trailing newline, but the last token is never
275 # a locator, so it's harmless here.
276 line.split(' ').each do |word|
277 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
284 def self.normalize_uuid uuid
287 uuid.split('+').each do |token|
288 if token.match /^[0-9a-f]{32,}$/
289 raise "uuid #{uuid} has multiple hash parts" if hash_part
291 elsif token.match /^\d+$/
292 raise "uuid #{uuid} has multiple size parts" if size_part
296 raise "uuid #{uuid} has no hash part" if !hash_part
297 [hash_part, size_part].compact.join '+'
300 # Return array of Collection objects
301 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
302 readers ||= [Thread.current[:user]]
304 readable_by(*readers).
305 readable_by(*readers, table_name: "collections").
306 joins("JOIN collections ON links.head_uuid = collections.uuid").
307 order("links.created_at DESC")
309 # If the search term is a Collection locator that contains one file
310 # that looks like a Docker image, return it.
311 if loc = Keep::Locator.parse(search_term)
313 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
315 # Check if the Collection contains exactly one file whose name
316 # looks like a saved Docker image.
317 manifest = Keep::Manifest.new(coll_match.manifest_text)
318 if manifest.exact_file_count?(1) and
319 (manifest.files[0][1] =~ /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/)
325 if search_tag.nil? and (n = search_term.index(":"))
326 search_tag = search_term[n+1..-1]
327 search_term = search_term[0..n-1]
330 # Find Collections with matching Docker image repository+tag pairs.
331 matches = base_search.
332 where(link_class: "docker_image_repo+tag",
333 name: "#{search_term}:#{search_tag || 'latest'}")
335 # If that didn't work, find Collections with matching Docker image hashes.
337 matches = base_search.
338 where("link_class = ? and links.name LIKE ?",
339 "docker_image_hash", "#{search_term}%")
342 # Generate an order key for each result. We want to order the results
343 # so that anything with an image timestamp is considered more recent than
344 # anything without; then we use the link's created_at as a tiebreaker.
346 matches.all.map do |link|
347 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
348 -link.created_at.to_i]
350 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
353 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
354 find_all_for_docker_image(search_term, search_tag, readers).first
357 def self.searchable_columns operator
358 super - ["manifest_text"]
361 def self.full_text_searchable_columns
362 super - ["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