2 require 'sweep_trashed_collections'
4 class Collection < ArvadosModel
5 extend CurrentApiClient
9 include CommonApiTemplate
11 serialize :properties, Hash
13 before_validation :set_validation_timestamp
14 before_validation :default_empty_manifest
15 before_validation :check_encoding
16 before_validation :check_manifest_validity
17 before_validation :check_signatures
18 before_validation :strip_signatures_and_update_replication_confirmed
19 before_validation :ensure_trash_at_not_in_past
20 before_validation :sync_trash_state
21 before_validation :default_trash_interval
22 validate :ensure_pdh_matches_manifest_text
23 validate :validate_trash_and_delete_timing
24 before_save :set_file_names
26 # Query only untrashed collections by default.
27 default_scope { where("is_trashed = false") }
29 api_accessible :user, extend: :common do |t|
33 t.add :portable_data_hash
34 t.add :signed_manifest_text, as: :manifest_text
35 t.add :manifest_text, as: :unsigned_manifest_text
36 t.add :replication_desired
37 t.add :replication_confirmed
38 t.add :replication_confirmed_at
45 @signatures_checked = false
46 @computed_pdh_for_manifest_text = false
49 def self.attributes_required_columns
51 # If we don't list manifest_text explicitly, the
52 # params[:select] code gets confused by the way we
53 # expose signed_manifest_text as manifest_text in the
54 # API response, and never let clients select the
55 # manifest_text column.
57 # We need trash_at and is_trashed to determine the
58 # correct timestamp in signed_manifest_text.
59 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
60 'unsigned_manifest_text' => ['manifest_text'],
64 def self.ignored_select_attributes
65 super + ["updated_at", "file_names"]
68 def self.limit_index_columns_read
72 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
74 return false if self.manifest_text.nil?
76 return true if current_user.andand.is_admin
78 # Provided the manifest_text hasn't changed materially since an
79 # earlier validation, it's safe to pass this validation on
80 # subsequent passes without checking any signatures. This is
81 # important because the signatures have probably been stripped off
82 # by the time we get to a second validation pass!
83 if @signatures_checked && @signatures_checked == computed_pdh
87 if self.manifest_text_changed?
88 # Check permissions on the collection manifest.
89 # If any signature cannot be verified, raise PermissionDeniedError
90 # which will return 403 Permission denied to the client.
91 api_token = current_api_client_authorization.andand.api_token
94 now: @validation_timestamp.to_i,
96 self.manifest_text.each_line do |entry|
97 entry.split.each do |tok|
98 if tok == '.' or tok.starts_with? './'
100 elsif tok =~ FILE_TOKEN
101 # This is a filename token, not a blob locator. Note that we
102 # keep checking tokens after this, even though manifest
103 # format dictates that all subsequent tokens will also be
104 # filenames. Safety first!
105 elsif Blob.verify_signature tok, signing_opts
107 elsif Keep::Locator.parse(tok).andand.signature
108 # Signature provided, but verify_signature did not like it.
109 logger.warn "Invalid signature on locator #{tok}"
110 raise ArvadosModel::PermissionDeniedError
111 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
112 # No signature provided, but we are running in insecure mode.
113 logger.debug "Missing signature on locator #{tok} ignored"
114 elsif Blob.new(tok).empty?
115 # No signature provided -- but no data to protect, either.
117 logger.warn "Missing signature on locator #{tok}"
118 raise ArvadosModel::PermissionDeniedError
123 @signatures_checked = computed_pdh
126 def strip_signatures_and_update_replication_confirmed
127 if self.manifest_text_changed?
129 if not self.replication_confirmed.nil?
130 self.class.each_manifest_locator(manifest_text_was) do |match|
131 in_old_manifest[match[1]] = true
135 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
136 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
137 # If the new manifest_text contains locators whose hashes
138 # weren't in the old manifest_text, storage replication is no
140 self.replication_confirmed_at = nil
141 self.replication_confirmed = nil
144 # Return the locator with all permission signatures removed,
145 # but otherwise intact.
146 match[0].gsub(/\+A[^+]*/, '')
149 if @computed_pdh_for_manifest_text == manifest_text
150 # If the cached PDH was valid before stripping, it is still
151 # valid after stripping.
152 @computed_pdh_for_manifest_text = stripped_manifest.dup
155 self[:manifest_text] = stripped_manifest
160 def ensure_pdh_matches_manifest_text
161 if not manifest_text_changed? and not portable_data_hash_changed?
163 elsif portable_data_hash.nil? or not portable_data_hash_changed?
164 self.portable_data_hash = computed_pdh
165 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
166 errors.add(:portable_data_hash, "is not a valid locator")
168 elsif portable_data_hash[0..31] != computed_pdh[0..31]
169 errors.add(:portable_data_hash,
170 "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
173 # Ignore the client-provided size part: always store
174 # computed_pdh in the database.
175 self.portable_data_hash = computed_pdh
180 if self.manifest_text_changed?
181 self.file_names = manifest_files
188 if self.manifest_text
189 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
190 names << name.first.gsub('\040',' ') + "\n"
191 break if names.length > 2**12
195 if self.manifest_text and names.length < 2**12
196 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
197 names << stream_name.first.gsub('\040',' ') + "\n"
198 break if names.length > 2**12
205 def default_empty_manifest
206 self.manifest_text ||= ''
210 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
214 # If Ruby thinks the encoding is something else, like 7-bit
215 # ASCII, but its stored bytes are equal to the (valid) UTF-8
216 # encoding of the same string, we declare it to be a UTF-8
219 utf8.force_encoding Encoding::UTF_8
220 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
221 self.manifest_text = utf8
226 errors.add :manifest_text, "must use UTF-8 encoding"
231 def check_manifest_validity
233 Keep::Manifest.validate! manifest_text
235 rescue ArgumentError => e
236 errors.add :manifest_text, e.message
241 def signed_manifest_text
242 if !has_attribute? :manifest_text
247 token = current_api_client_authorization.andand.api_token
248 exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
249 trash_at].compact.map(&:to_i).min
250 self.class.sign_manifest manifest_text, token, exp
254 def self.sign_manifest manifest, token, exp=nil
256 exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
262 m = munge_manifest_locators(manifest) do |match|
263 Blob.sign_locator(match[0], signing_opts)
268 def self.munge_manifest_locators manifest
269 # Given a manifest text and a block, yield the regexp MatchData
270 # for each locator. Return a new manifest in which each locator
271 # has been replaced by the block's return value.
272 return nil if !manifest
273 return '' if manifest == ''
276 manifest.each_line do |line|
279 line.split(' ').each do |word|
282 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
283 new_words << yield(match)
288 new_lines << new_words.join(' ')
290 new_lines.join("\n") + "\n"
293 def self.each_manifest_locator manifest
294 # Given a manifest text and a block, yield the regexp match object
296 manifest.each_line do |line|
297 # line will have a trailing newline, but the last token is never
298 # a locator, so it's harmless here.
299 line.split(' ').each do |word|
300 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
307 def self.normalize_uuid uuid
310 uuid.split('+').each do |token|
311 if token.match(/^[0-9a-f]{32,}$/)
312 raise "uuid #{uuid} has multiple hash parts" if hash_part
314 elsif token.match(/^\d+$/)
315 raise "uuid #{uuid} has multiple size parts" if size_part
319 raise "uuid #{uuid} has no hash part" if !hash_part
320 [hash_part, size_part].compact.join '+'
323 def self.get_compatible_images(readers, pattern, collections)
324 if collections.empty?
329 Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
330 collections.map(&:portable_data_hash),
331 'docker_image_migration',
333 order('links.created_at asc').
335 [l.tail_uuid, l.head_uuid]
338 migrated_collections = Hash[
339 Collection.readable_by(*readers).
340 where('portable_data_hash in (?)', migrations.values).
342 [c.portable_data_hash, c]
345 collections.map { |c|
346 # Check if the listed image is compatible first, if not, then try the
348 manifest = Keep::Manifest.new(c.manifest_text)
349 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
351 elsif m = migrated_collections[migrations[c.portable_data_hash]]
352 manifest = Keep::Manifest.new(m.manifest_text)
353 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
360 # Resolve a Docker repo+tag, hash, or collection PDH to an array of
361 # Collection objects, sorted by timestamp starting with the most recent
364 # If filter_compatible_format is true (the default), only return image
365 # collections which are support by the installation as indicated by
366 # Rails.configuration.docker_image_formats. Will follow
367 # 'docker_image_migration' links if search_term resolves to an incompatible
368 # image, but an equivalent compatible image is available.
369 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
370 readers ||= [Thread.current[:user]]
372 readable_by(*readers).
373 readable_by(*readers, table_name: "collections").
374 joins("JOIN collections ON links.head_uuid = collections.uuid").
375 order("links.created_at DESC")
377 if (Rails.configuration.docker_image_formats.include? 'v1' and
378 Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
379 pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
380 elsif Rails.configuration.docker_image_formats.include? 'v2'
381 pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
382 elsif Rails.configuration.docker_image_formats.include? 'v1'
383 pattern = /^[0-9A-Fa-f]{64}\.tar$/
385 raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
388 # If the search term is a Collection locator that contains one file
389 # that looks like a Docker image, return it.
390 if loc = Keep::Locator.parse(search_term)
392 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
393 return get_compatible_images(readers, pattern, coll_match)
396 if search_tag.nil? and (n = search_term.index(":"))
397 search_tag = search_term[n+1..-1]
398 search_term = search_term[0..n-1]
401 # Find Collections with matching Docker image repository+tag pairs.
402 matches = base_search.
403 where(link_class: "docker_image_repo+tag",
404 name: "#{search_term}:#{search_tag || 'latest'}")
406 # If that didn't work, find Collections with matching Docker image hashes.
408 matches = base_search.
409 where("link_class = ? and links.name LIKE ?",
410 "docker_image_hash", "#{search_term}%")
413 # Generate an order key for each result. We want to order the results
414 # so that anything with an image timestamp is considered more recent than
415 # anything without; then we use the link's created_at as a tiebreaker.
417 matches.each do |link|
418 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
419 -link.created_at.to_i]
422 sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
423 uuid_timestamps[c.uuid]
425 compatible = get_compatible_images(readers, pattern, sorted)
426 if sorted.length > 0 and compatible.empty?
427 raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
432 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
433 find_all_for_docker_image(search_term, search_tag, readers).first
436 def self.searchable_columns operator
437 super - ["manifest_text"]
440 def self.full_text_searchable_columns
441 super - ["manifest_text"]
445 SweepTrashedCollections.sweep_if_stale
450 def portable_manifest_text
451 self.class.munge_manifest_locators(manifest_text) do |match|
461 portable_manifest = portable_manifest_text
462 (Digest::MD5.hexdigest(portable_manifest) +
464 portable_manifest.bytesize.to_s)
468 if @computed_pdh_for_manifest_text == manifest_text
471 @computed_pdh = compute_pdh
472 @computed_pdh_for_manifest_text = manifest_text.dup
476 def ensure_permission_to_save
477 if (not current_user.andand.is_admin and
478 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
479 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
480 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
485 # Use a single timestamp for all validations, even though each
486 # validation runs at a different time.
487 def set_validation_timestamp
488 @validation_timestamp = db_current_time
491 # If trash_at is being changed to a time in the past, change it to
492 # now. This allows clients to say "expires {client-current-time}"
493 # without failing due to clock skew, while avoiding odd log entries
494 # like "expiry date changed to {1 year ago}".
495 def ensure_trash_at_not_in_past
496 if trash_at_changed? && trash_at
497 self.trash_at = [@validation_timestamp, trash_at].max
501 # Caller can move into/out of trash by setting/clearing is_trashed
502 # -- however, if the caller also changes trash_at, then any changes
503 # to is_trashed are ignored.
505 if is_trashed_changed? && !trash_at_changed?
507 self.trash_at = @validation_timestamp
513 self.is_trashed = trash_at && trash_at <= @validation_timestamp || false
517 def default_trash_interval
518 if trash_at_changed? && !delete_at_changed?
519 # If trash_at is updated without touching delete_at,
520 # automatically update delete_at to a sensible value.
524 self.delete_at = trash_at + Rails.configuration.default_trash_lifetime.seconds
526 elsif !trash_at || !delete_at || trash_at > delete_at
527 # Not trash, or bogus arguments? Just validate in
528 # validate_trash_and_delete_timing.
529 elsif delete_at_changed? && delete_at >= trash_at
530 # Fix delete_at if needed, so it's not earlier than the expiry
531 # time on any permission tokens that might have been given out.
533 # In any case there are no signatures expiring after now+TTL.
534 # Also, if the existing trash_at time has already passed, we
535 # know we haven't given out any signatures since then.
537 @validation_timestamp,
539 ].compact.min + Rails.configuration.blob_signature_ttl.seconds
541 # The previous value of delete_at is also an upper bound on the
542 # longest-lived permission token. For example, if TTL=14,
543 # trash_at_was=now-7, delete_at_was=now+7, then it is safe to
544 # set trash_at=now+6, delete_at=now+8.
545 earliest_delete = [earliest_delete, delete_at_was].compact.min
547 # If delete_at is too soon, use the earliest possible time.
548 if delete_at < earliest_delete
549 self.delete_at = earliest_delete
554 def validate_trash_and_delete_timing
555 if trash_at.nil? != delete_at.nil?
556 errors.add :delete_at, "must be set if trash_at is set, and must be nil otherwise"
557 elsif delete_at && delete_at < trash_at
558 errors.add :delete_at, "must not be earlier than trash_at"