1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: AGPL-3.0
6 require 'sweep_trashed_objects'
9 class Collection < ArvadosModel
10 extend CurrentApiClient
14 include CommonApiTemplate
17 serialize :properties, Hash
18 serialize :storage_classes_desired, Array
19 serialize :storage_classes_confirmed, Array
21 before_validation :default_empty_manifest
22 before_validation :default_storage_classes, on: :create
23 before_validation :check_encoding
24 before_validation :check_manifest_validity
25 before_validation :check_signatures
26 before_validation :strip_signatures_and_update_replication_confirmed
27 validate :ensure_pdh_matches_manifest_text
28 validate :ensure_storage_classes_desired_is_not_empty
29 validate :ensure_storage_classes_contain_non_empty_strings
30 validate :past_versions_cannot_be_updated, on: :update
31 before_save :set_file_names
32 around_update :prepare_for_versioning
33 after_update :save_old_version, if: Proc.new { |c| c.should_preserve_version? }
34 after_update :sync_past_versions, if: Proc.new { |c| c.syncable_updates?(self.changes.keys) }
36 api_accessible :user, extend: :common do |t|
40 t.add :portable_data_hash
41 t.add :signed_manifest_text, as: :manifest_text
42 t.add :manifest_text, as: :unsigned_manifest_text
43 t.add :replication_desired
44 t.add :replication_confirmed
45 t.add :replication_confirmed_at
46 t.add :storage_classes_desired
47 t.add :storage_classes_confirmed
48 t.add :storage_classes_confirmed_at
53 t.add :current_version_uuid
54 t.add :preserve_version
58 @signatures_checked = false
59 @computed_pdh_for_manifest_text = false
62 def self.attributes_required_columns
64 # If we don't list manifest_text explicitly, the
65 # params[:select] code gets confused by the way we
66 # expose signed_manifest_text as manifest_text in the
67 # API response, and never let clients select the
68 # manifest_text column.
70 # We need trash_at and is_trashed to determine the
71 # correct timestamp in signed_manifest_text.
72 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
73 'unsigned_manifest_text' => ['manifest_text'],
77 def self.ignored_select_attributes
78 super + ["updated_at", "file_names"]
81 def self.limit_index_columns_read
85 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
87 return false if self.manifest_text.nil?
89 return true if current_user.andand.is_admin
91 # Provided the manifest_text hasn't changed materially since an
92 # earlier validation, it's safe to pass this validation on
93 # subsequent passes without checking any signatures. This is
94 # important because the signatures have probably been stripped off
95 # by the time we get to a second validation pass!
96 if @signatures_checked && @signatures_checked == computed_pdh
100 if self.manifest_text_changed?
101 # Check permissions on the collection manifest.
102 # If any signature cannot be verified, raise PermissionDeniedError
103 # which will return 403 Permission denied to the client.
104 api_token = Thread.current[:token]
106 api_token: api_token,
107 now: @validation_timestamp.to_i,
109 self.manifest_text.each_line do |entry|
110 entry.split.each do |tok|
111 if tok == '.' or tok.starts_with? './'
113 elsif tok =~ FILE_TOKEN
114 # This is a filename token, not a blob locator. Note that we
115 # keep checking tokens after this, even though manifest
116 # format dictates that all subsequent tokens will also be
117 # filenames. Safety first!
118 elsif Blob.verify_signature tok, signing_opts
120 elsif Keep::Locator.parse(tok).andand.signature
121 # Signature provided, but verify_signature did not like it.
122 logger.warn "Invalid signature on locator #{tok}"
123 raise ArvadosModel::PermissionDeniedError
124 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
125 # No signature provided, but we are running in insecure mode.
126 logger.debug "Missing signature on locator #{tok} ignored"
127 elsif Blob.new(tok).empty?
128 # No signature provided -- but no data to protect, either.
130 logger.warn "Missing signature on locator #{tok}"
131 raise ArvadosModel::PermissionDeniedError
136 @signatures_checked = computed_pdh
139 def strip_signatures_and_update_replication_confirmed
140 if self.manifest_text_changed?
142 if not self.replication_confirmed.nil?
143 self.class.each_manifest_locator(manifest_text_was) do |match|
144 in_old_manifest[match[1]] = true
148 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
149 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
150 # If the new manifest_text contains locators whose hashes
151 # weren't in the old manifest_text, storage replication is no
153 self.replication_confirmed_at = nil
154 self.replication_confirmed = nil
157 # Return the locator with all permission signatures removed,
158 # but otherwise intact.
159 match[0].gsub(/\+A[^+]*/, '')
162 if @computed_pdh_for_manifest_text == manifest_text
163 # If the cached PDH was valid before stripping, it is still
164 # valid after stripping.
165 @computed_pdh_for_manifest_text = stripped_manifest.dup
168 self[:manifest_text] = stripped_manifest
173 def ensure_pdh_matches_manifest_text
174 if not manifest_text_changed? and not portable_data_hash_changed?
176 elsif portable_data_hash.nil? or not portable_data_hash_changed?
177 self.portable_data_hash = computed_pdh
178 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
179 errors.add(:portable_data_hash, "is not a valid locator")
181 elsif portable_data_hash[0..31] != computed_pdh[0..31]
182 errors.add(:portable_data_hash,
183 "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
186 # Ignore the client-provided size part: always store
187 # computed_pdh in the database.
188 self.portable_data_hash = computed_pdh
193 if self.manifest_text_changed?
194 self.file_names = manifest_files
200 return '' if !self.manifest_text
204 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
207 names << name.first.gsub('\040',' ') + "\n"
209 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
210 next if done[stream_name]
211 done[stream_name] = true
212 names << stream_name.first.gsub('\040',' ') + "\n"
217 def default_empty_manifest
218 self.manifest_text ||= ''
221 def skip_uuid_existence_check
222 # Avoid checking the existence of current_version_uuid, as it's
223 # assigned on creation of a new 'current version' collection, so
224 # the collection's UUID only lives on memory when the validation check
226 ['current_version_uuid']
229 def prepare_for_versioning
230 # Put aside the changes because with_lock forces a record reload
231 changes = self.changes
234 # Copy the original state to save it as old version
235 if versionable_updates?(changes.keys) && should_preserve_version?
237 @snapshot.uuid = nil # Reset UUID so it's created as a new record
238 @snapshot.created_at = self.created_at
241 # Restore requested changes on the current version
242 changes.keys.each do |attr|
245 elsif attr == 'preserve_version' && changes[attr].last == false
246 next # Ignore false assignment, once true it'll be true until next version
248 self.attributes = {attr => changes[attr].last}
250 # Also update the current version reference
251 self.attributes = {'current_version_uuid' => changes[attr].last}
255 if versionable_updates?(changes.keys) && should_preserve_version?
257 self.preserve_version = false
265 (syncable_attrs & self.changes.keys).each do |attr|
267 # Point old versions to current version's new UUID
268 updates['current_version_uuid'] = self.changes[attr].last
270 updates[attr] = self.changes[attr].last
276 def sync_past_versions
277 updates = self.syncable_updates
278 Collection.where('current_version_uuid = ? AND uuid != ?', self.uuid_was, self.uuid_was).each do |c|
279 c.attributes = updates
280 # Use a different validation context to skip the 'old_versions_cannot_be_updated'
281 # validator, as on this case it is legal to update some fields.
282 leave_modified_by_user_alone do
283 c.save(context: :update_old_versions)
288 def versionable_updates?(attrs)
289 (['manifest_text', 'description', 'properties', 'name'] & attrs).any?
293 ['uuid', 'owner_uuid', 'delete_at', 'trash_at', 'is_trashed', 'replication_desired', 'storage_classes_desired']
296 def syncable_updates?(attrs)
297 (self.syncable_attrs & attrs).any?
301 if @snapshot && should_preserve_version?
302 @snapshot.attributes = self.syncable_updates
307 def should_preserve_version?
308 return false if !Rails.configuration.collection_versioning
310 idle_threshold = Rails.configuration.preserve_version_if_idle
311 if !self.preserve_version_was &&
312 (idle_threshold < 0 ||
313 (idle_threshold > 0 && self.modified_at_was > db_current_time-idle_threshold.seconds))
320 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
324 # If Ruby thinks the encoding is something else, like 7-bit
325 # ASCII, but its stored bytes are equal to the (valid) UTF-8
326 # encoding of the same string, we declare it to be a UTF-8
329 utf8.force_encoding Encoding::UTF_8
330 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
331 self.manifest_text = utf8
336 errors.add :manifest_text, "must use UTF-8 encoding"
341 def check_manifest_validity
343 Keep::Manifest.validate! manifest_text
345 rescue ArgumentError => e
346 errors.add :manifest_text, e.message
351 def signed_manifest_text
352 if !has_attribute? :manifest_text
357 token = Thread.current[:token]
358 exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
359 trash_at].compact.map(&:to_i).min
360 self.class.sign_manifest manifest_text, token, exp
364 def self.sign_manifest manifest, token, exp=nil
366 exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
372 m = munge_manifest_locators(manifest) do |match|
373 Blob.sign_locator(match[0], signing_opts)
378 def self.munge_manifest_locators manifest
379 # Given a manifest text and a block, yield the regexp MatchData
380 # for each locator. Return a new manifest in which each locator
381 # has been replaced by the block's return value.
382 return nil if !manifest
383 return '' if manifest == ''
386 manifest.each_line do |line|
389 line.split(' ').each do |word|
392 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
393 new_words << yield(match)
398 new_lines << new_words.join(' ')
400 new_lines.join("\n") + "\n"
403 def self.each_manifest_locator manifest
404 # Given a manifest text and a block, yield the regexp match object
406 manifest.each_line do |line|
407 # line will have a trailing newline, but the last token is never
408 # a locator, so it's harmless here.
409 line.split(' ').each do |word|
410 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
417 def self.normalize_uuid uuid
420 uuid.split('+').each do |token|
421 if token.match(/^[0-9a-f]{32,}$/)
422 raise "uuid #{uuid} has multiple hash parts" if hash_part
424 elsif token.match(/^\d+$/)
425 raise "uuid #{uuid} has multiple size parts" if size_part
429 raise "uuid #{uuid} has no hash part" if !hash_part
430 [hash_part, size_part].compact.join '+'
433 def self.get_compatible_images(readers, pattern, collections)
434 if collections.empty?
439 Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
440 collections.map(&:portable_data_hash),
441 'docker_image_migration',
443 order('links.created_at asc').
445 [l.tail_uuid, l.head_uuid]
448 migrated_collections = Hash[
449 Collection.readable_by(*readers).
450 where('portable_data_hash in (?)', migrations.values).
452 [c.portable_data_hash, c]
455 collections.map { |c|
456 # Check if the listed image is compatible first, if not, then try the
458 manifest = Keep::Manifest.new(c.manifest_text)
459 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
461 elsif m = migrated_collections[migrations[c.portable_data_hash]]
462 manifest = Keep::Manifest.new(m.manifest_text)
463 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
470 # Resolve a Docker repo+tag, hash, or collection PDH to an array of
471 # Collection objects, sorted by timestamp starting with the most recent
474 # If filter_compatible_format is true (the default), only return image
475 # collections which are support by the installation as indicated by
476 # Rails.configuration.docker_image_formats. Will follow
477 # 'docker_image_migration' links if search_term resolves to an incompatible
478 # image, but an equivalent compatible image is available.
479 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
480 readers ||= [Thread.current[:user]]
482 readable_by(*readers).
483 readable_by(*readers, table_name: "collections").
484 joins("JOIN collections ON links.head_uuid = collections.uuid").
485 order("links.created_at DESC")
487 if (Rails.configuration.docker_image_formats.include? 'v1' and
488 Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
489 pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
490 elsif Rails.configuration.docker_image_formats.include? 'v2'
491 pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
492 elsif Rails.configuration.docker_image_formats.include? 'v1'
493 pattern = /^[0-9A-Fa-f]{64}\.tar$/
495 raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
498 # If the search term is a Collection locator that contains one file
499 # that looks like a Docker image, return it.
500 if loc = Keep::Locator.parse(search_term)
502 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
503 return get_compatible_images(readers, pattern, coll_match)
506 if search_tag.nil? and (n = search_term.index(":"))
507 search_tag = search_term[n+1..-1]
508 search_term = search_term[0..n-1]
511 # Find Collections with matching Docker image repository+tag pairs.
512 matches = base_search.
513 where(link_class: "docker_image_repo+tag",
514 name: "#{search_term}:#{search_tag || 'latest'}")
516 # If that didn't work, find Collections with matching Docker image hashes.
518 matches = base_search.
519 where("link_class = ? and links.name LIKE ?",
520 "docker_image_hash", "#{search_term}%")
523 # Generate an order key for each result. We want to order the results
524 # so that anything with an image timestamp is considered more recent than
525 # anything without; then we use the link's created_at as a tiebreaker.
527 matches.each do |link|
528 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
529 -link.created_at.to_i]
532 sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
533 uuid_timestamps[c.uuid]
535 compatible = get_compatible_images(readers, pattern, sorted)
536 if sorted.length > 0 and compatible.empty?
537 raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
542 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
543 find_all_for_docker_image(search_term, search_tag, readers).first
546 def self.searchable_columns operator
547 super - ["manifest_text"]
550 def self.full_text_searchable_columns
551 super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed", "current_version_uuid"]
555 SweepTrashedObjects.sweep_if_stale
561 # Although the defaults for these columns is already set up on the schema,
562 # collection creation from an API client seems to ignore them, making the
563 # validation on empty desired storage classes return an error.
564 def default_storage_classes
565 if self.storage_classes_desired.nil? || self.storage_classes_desired.empty?
566 self.storage_classes_desired = ["default"]
568 self.storage_classes_confirmed ||= []
571 def portable_manifest_text
572 self.class.munge_manifest_locators(manifest_text) do |match|
582 portable_manifest = portable_manifest_text
583 (Digest::MD5.hexdigest(portable_manifest) +
585 portable_manifest.bytesize.to_s)
589 if @computed_pdh_for_manifest_text == manifest_text
592 @computed_pdh = compute_pdh
593 @computed_pdh_for_manifest_text = manifest_text.dup
597 def ensure_permission_to_save
598 if (not current_user.andand.is_admin)
599 if (replication_confirmed_at_changed? or replication_confirmed_changed?) and
600 not (replication_confirmed_at.nil? and replication_confirmed.nil?)
601 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
603 if (storage_classes_confirmed_changed? or storage_classes_confirmed_at_changed?) and
604 not (storage_classes_confirmed == [] and storage_classes_confirmed_at.nil?)
605 raise ArvadosModel::PermissionDeniedError.new("storage_classes_confirmed and storage_classes_confirmed_at attributes cannot be changed, except by setting them to [] and nil respectively")
611 def ensure_storage_classes_desired_is_not_empty
612 if self.storage_classes_desired.empty?
613 raise ArvadosModel::InvalidStateTransitionError.new("storage_classes_desired shouldn't be empty")
617 def ensure_storage_classes_contain_non_empty_strings
618 (self.storage_classes_desired + self.storage_classes_confirmed).each do |c|
619 if !c.is_a?(String) || c == ''
620 raise ArvadosModel::InvalidStateTransitionError.new("storage classes should only be non-empty strings")
625 def past_versions_cannot_be_updated
626 # We check for the '_was' values just in case the update operation
627 # includes a change on current_version_uuid or uuid.
628 if current_version_uuid_was != uuid_was
629 errors.add(:base, "past versions cannot be updated")
636 self.current_version_uuid ||= self.uuid