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 # Posgresql JSONB columns should NOT be declared as serialized, Rails 5
18 # already know how to properly treat them.
19 attribute :properties, :jsonbHash, default: {}
20 attribute :storage_classes_desired, :jsonbArray, default: ["default"]
21 attribute :storage_classes_confirmed, :jsonbArray, default: []
23 before_validation :default_empty_manifest
24 before_validation :default_storage_classes, on: :create
25 before_validation :check_encoding
26 before_validation :check_manifest_validity
27 before_validation :check_signatures
28 before_validation :strip_signatures_and_update_replication_confirmed
29 validate :ensure_pdh_matches_manifest_text
30 validate :ensure_storage_classes_desired_is_not_empty
31 validate :ensure_storage_classes_contain_non_empty_strings
32 validate :versioning_metadata_updates, on: :update
33 validate :past_versions_cannot_be_updated, on: :update
34 after_validation :set_file_count_and_total_size
35 before_save :set_file_names
36 around_update :manage_versioning, unless: :is_past_version?
38 api_accessible :user, extend: :common do |t|
42 t.add :portable_data_hash
43 t.add :signed_manifest_text, as: :manifest_text
44 t.add :manifest_text, as: :unsigned_manifest_text
45 t.add :replication_desired
46 t.add :replication_confirmed
47 t.add :replication_confirmed_at
48 t.add :storage_classes_desired
49 t.add :storage_classes_confirmed
50 t.add :storage_classes_confirmed_at
55 t.add :current_version_uuid
56 t.add :preserve_version
58 t.add :file_size_total
62 @signatures_checked = false
63 @computed_pdh_for_manifest_text = false
66 def self.attributes_required_columns
68 # If we don't list manifest_text explicitly, the
69 # params[:select] code gets confused by the way we
70 # expose signed_manifest_text as manifest_text in the
71 # API response, and never let clients select the
72 # manifest_text column.
74 # We need trash_at and is_trashed to determine the
75 # correct timestamp in signed_manifest_text.
76 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
77 'unsigned_manifest_text' => ['manifest_text'],
81 def self.ignored_select_attributes
82 super + ["updated_at", "file_names"]
85 def self.limit_index_columns_read
89 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
91 throw(:abort) if self.manifest_text.nil?
93 return true if current_user.andand.is_admin
95 # Provided the manifest_text hasn't changed materially since an
96 # earlier validation, it's safe to pass this validation on
97 # subsequent passes without checking any signatures. This is
98 # important because the signatures have probably been stripped off
99 # by the time we get to a second validation pass!
100 if @signatures_checked && @signatures_checked == computed_pdh
104 if self.manifest_text_changed?
105 # Check permissions on the collection manifest.
106 # If any signature cannot be verified, raise PermissionDeniedError
107 # which will return 403 Permission denied to the client.
108 api_token = Thread.current[:token]
110 api_token: api_token,
111 now: @validation_timestamp.to_i,
113 self.manifest_text.each_line do |entry|
114 entry.split.each do |tok|
115 if tok == '.' or tok.starts_with? './'
117 elsif tok =~ FILE_TOKEN
118 # This is a filename token, not a blob locator. Note that we
119 # keep checking tokens after this, even though manifest
120 # format dictates that all subsequent tokens will also be
121 # filenames. Safety first!
122 elsif Blob.verify_signature tok, signing_opts
124 elsif Keep::Locator.parse(tok).andand.signature
125 # Signature provided, but verify_signature did not like it.
126 logger.warn "Invalid signature on locator #{tok}"
127 raise ArvadosModel::PermissionDeniedError
128 elsif !Rails.configuration.Collections.BlobSigning
129 # No signature provided, but we are running in insecure mode.
130 logger.debug "Missing signature on locator #{tok} ignored"
131 elsif Blob.new(tok).empty?
132 # No signature provided -- but no data to protect, either.
134 logger.warn "Missing signature on locator #{tok}"
135 raise ArvadosModel::PermissionDeniedError
140 @signatures_checked = computed_pdh
143 def strip_signatures_and_update_replication_confirmed
144 if self.manifest_text_changed?
146 if not self.replication_confirmed.nil?
147 self.class.each_manifest_locator(manifest_text_was) do |match|
148 in_old_manifest[match[1]] = true
152 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
153 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
154 # If the new manifest_text contains locators whose hashes
155 # weren't in the old manifest_text, storage replication is no
157 self.replication_confirmed_at = nil
158 self.replication_confirmed = nil
161 # Return the locator with all permission signatures removed,
162 # but otherwise intact.
163 match[0].gsub(/\+A[^+]*/, '')
166 if @computed_pdh_for_manifest_text == manifest_text
167 # If the cached PDH was valid before stripping, it is still
168 # valid after stripping.
169 @computed_pdh_for_manifest_text = stripped_manifest.dup
172 self[:manifest_text] = stripped_manifest
177 def ensure_pdh_matches_manifest_text
178 if not manifest_text_changed? and not portable_data_hash_changed?
180 elsif portable_data_hash.nil? or not portable_data_hash_changed?
181 self.portable_data_hash = computed_pdh
182 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
183 errors.add(:portable_data_hash, "is not a valid locator")
185 elsif portable_data_hash[0..31] != computed_pdh[0..31]
186 errors.add(:portable_data_hash,
187 "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
190 # Ignore the client-provided size part: always store
191 # computed_pdh in the database.
192 self.portable_data_hash = computed_pdh
197 if self.manifest_text_changed?
198 self.file_names = manifest_files
203 def set_file_count_and_total_size
204 # Only update the file stats if the manifest changed
205 if self.manifest_text_changed?
206 m = Keep::Manifest.new(self.manifest_text)
207 self.file_size_total = m.files_size
208 self.file_count = m.files_count
209 # If the manifest didn't change but the attributes did, ignore the changes
210 elsif self.file_count_changed? || self.file_size_total_changed?
211 self.file_count = self.file_count_was
212 self.file_size_total = self.file_size_total_was
218 return '' if !self.manifest_text
222 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
225 names << name.first.gsub('\040',' ') + "\n"
227 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
228 next if done[stream_name]
229 done[stream_name] = true
230 names << stream_name.first.gsub('\040',' ') + "\n"
235 def default_empty_manifest
236 self.manifest_text ||= ''
239 def skip_uuid_existence_check
240 # Avoid checking the existence of current_version_uuid, as it's
241 # assigned on creation of a new 'current version' collection, so
242 # the collection's UUID only lives on memory when the validation check
244 ['current_version_uuid']
247 def manage_versioning
248 should_preserve_version = should_preserve_version? # Time sensitive, cache value
249 return(yield) unless (should_preserve_version || syncable_updates.any?)
251 # Put aside the changes because with_lock forces a record reload
252 changes = self.changes
255 # Copy the original state to save it as old version
256 if should_preserve_version
258 snapshot.uuid = nil # Reset UUID so it's created as a new record
259 snapshot.created_at = self.created_at
262 # Restore requested changes on the current version
263 changes.keys.each do |attr|
264 if attr == 'preserve_version' && changes[attr].last == false
265 next # Ignore false assignment, once true it'll be true until next version
267 self.attributes = {attr => changes[attr].last}
269 # Also update the current version reference
270 self.attributes = {'current_version_uuid' => changes[attr].last}
274 if should_preserve_version
276 self.preserve_version = false
281 sync_past_versions if syncable_updates.any?
283 snapshot.attributes = self.syncable_updates
284 leave_modified_by_user_alone do
285 act_as_system_user do
295 (syncable_attrs & self.changes.keys).each do |attr|
297 # Point old versions to current version's new UUID
298 updates['current_version_uuid'] = self.changes[attr].last
300 updates[attr] = self.changes[attr].last
306 def sync_past_versions
307 updates = self.syncable_updates
308 Collection.where('current_version_uuid = ? AND uuid != ?', self.uuid_was, self.uuid_was).each do |c|
309 c.attributes = updates
310 # Use a different validation context to skip the 'past_versions_cannot_be_updated'
311 # validator, as on this case it is legal to update some fields.
312 leave_modified_by_user_alone do
313 leave_modified_at_alone do
314 c.save(context: :update_old_versions)
320 def versionable_updates?(attrs)
321 (['manifest_text', 'description', 'properties', 'name'] & attrs).any?
325 ['uuid', 'owner_uuid', 'delete_at', 'trash_at', 'is_trashed', 'replication_desired', 'storage_classes_desired']
329 # Check for the '_was' values just in case the update operation
330 # includes a change on current_version_uuid or uuid.
331 if !new_record? && self.current_version_uuid_was != self.uuid_was
338 def should_preserve_version?
339 return false unless (Rails.configuration.Collections.CollectionVersioning && versionable_updates?(self.changes.keys))
341 idle_threshold = Rails.configuration.Collections.PreserveVersionIfIdle
342 if !self.preserve_version_was &&
343 (idle_threshold < 0 ||
344 (idle_threshold > 0 && self.modified_at_was > db_current_time-idle_threshold.seconds))
351 if !(manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?)
353 # If Ruby thinks the encoding is something else, like 7-bit
354 # ASCII, but its stored bytes are equal to the (valid) UTF-8
355 # encoding of the same string, we declare it to be a UTF-8
358 utf8.force_encoding Encoding::UTF_8
359 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
360 self.manifest_text = utf8
365 errors.add :manifest_text, "must use UTF-8 encoding"
370 def check_manifest_validity
372 Keep::Manifest.validate! manifest_text
374 rescue ArgumentError => e
375 errors.add :manifest_text, e.message
380 def signed_manifest_text
381 if !has_attribute? :manifest_text
386 token = Thread.current[:token]
387 exp = [db_current_time.to_i + Rails.configuration.Collections.BlobSigningTTL.to_i,
388 trash_at].compact.map(&:to_i).min
389 self.class.sign_manifest manifest_text, token, exp
393 def self.sign_manifest manifest, token, exp=nil
395 exp = db_current_time.to_i + Rails.configuration.Collections.BlobSigningTTL.to_i
401 m = munge_manifest_locators(manifest) do |match|
402 Blob.sign_locator(match[0], signing_opts)
407 def self.munge_manifest_locators manifest
408 # Given a manifest text and a block, yield the regexp MatchData
409 # for each locator. Return a new manifest in which each locator
410 # has been replaced by the block's return value.
411 return nil if !manifest
412 return '' if manifest == ''
415 manifest.each_line do |line|
418 line.split(' ').each do |word|
421 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
422 new_words << yield(match)
427 new_lines << new_words.join(' ')
429 new_lines.join("\n") + "\n"
432 def self.each_manifest_locator manifest
433 # Given a manifest text and a block, yield the regexp match object
435 manifest.each_line do |line|
436 # line will have a trailing newline, but the last token is never
437 # a locator, so it's harmless here.
438 line.split(' ').each do |word|
439 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
446 def self.normalize_uuid uuid
449 uuid.split('+').each do |token|
450 if token.match(/^[0-9a-f]{32,}$/)
451 raise "uuid #{uuid} has multiple hash parts" if hash_part
453 elsif token.match(/^\d+$/)
454 raise "uuid #{uuid} has multiple size parts" if size_part
458 raise "uuid #{uuid} has no hash part" if !hash_part
459 [hash_part, size_part].compact.join '+'
462 def self.get_compatible_images(readers, pattern, collections)
463 if collections.empty?
468 Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
469 collections.map(&:portable_data_hash),
470 'docker_image_migration',
472 order('links.created_at asc').
474 [l.tail_uuid, l.head_uuid]
477 migrated_collections = Hash[
478 Collection.readable_by(*readers).
479 where('portable_data_hash in (?)', migrations.values).
481 [c.portable_data_hash, c]
484 collections.map { |c|
485 # Check if the listed image is compatible first, if not, then try the
487 manifest = Keep::Manifest.new(c.manifest_text)
488 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
490 elsif m = migrated_collections[migrations[c.portable_data_hash]]
491 manifest = Keep::Manifest.new(m.manifest_text)
492 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
499 # Resolve a Docker repo+tag, hash, or collection PDH to an array of
500 # Collection objects, sorted by timestamp starting with the most recent
503 # If filter_compatible_format is true (the default), only return image
504 # collections which are support by the installation as indicated by
505 # Rails.configuration.Containers.SupportedDockerImageFormats. Will follow
506 # 'docker_image_migration' links if search_term resolves to an incompatible
507 # image, but an equivalent compatible image is available.
508 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
509 readers ||= [Thread.current[:user]]
511 readable_by(*readers).
512 readable_by(*readers, table_name: "collections").
513 joins("JOIN collections ON links.head_uuid = collections.uuid").
514 order("links.created_at DESC")
516 docker_image_formats = Rails.configuration.Containers.SupportedDockerImageFormats
518 if (docker_image_formats.include? 'v1' and
519 docker_image_formats.include? 'v2') or filter_compatible_format == false
520 pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
521 elsif docker_image_formats.include? 'v2'
522 pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
523 elsif docker_image_formats.include? 'v1'
524 pattern = /^[0-9A-Fa-f]{64}\.tar$/
526 raise "Unrecognized configuration for docker_image_formats #{docker_image_formats}"
529 # If the search term is a Collection locator that contains one file
530 # that looks like a Docker image, return it.
531 if loc = Keep::Locator.parse(search_term)
533 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
534 rc = Rails.configuration.RemoteClusters.select{ |k|
535 k != :"*" && k != Rails.configuration.ClusterID}
536 if coll_match.any? or rc.length == 0
537 return get_compatible_images(readers, pattern, coll_match)
539 # Allow bare pdh that doesn't exist in the local database so
540 # that federated container requests which refer to remotely
541 # stored containers will validate.
542 return [Collection.new(portable_data_hash: loc.to_s)]
546 if search_tag.nil? and (n = search_term.index(":"))
547 search_tag = search_term[n+1..-1]
548 search_term = search_term[0..n-1]
551 # Find Collections with matching Docker image repository+tag pairs.
552 matches = base_search.
553 where(link_class: "docker_image_repo+tag",
554 name: "#{search_term}:#{search_tag || 'latest'}")
556 # If that didn't work, find Collections with matching Docker image hashes.
558 matches = base_search.
559 where("link_class = ? and links.name LIKE ?",
560 "docker_image_hash", "#{search_term}%")
563 # Generate an order key for each result. We want to order the results
564 # so that anything with an image timestamp is considered more recent than
565 # anything without; then we use the link's created_at as a tiebreaker.
567 matches.each do |link|
568 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
569 -link.created_at.to_i]
572 sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
573 uuid_timestamps[c.uuid]
575 compatible = get_compatible_images(readers, pattern, sorted)
576 if sorted.length > 0 and compatible.empty?
577 raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
582 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
583 find_all_for_docker_image(search_term, search_tag, readers).first
586 def self.searchable_columns operator
587 super - ["manifest_text"]
590 def self.full_text_searchable_columns
591 super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed", "current_version_uuid"]
595 SweepTrashedObjects.sweep_if_stale
601 # Although the defaults for these columns is already set up on the schema,
602 # collection creation from an API client seems to ignore them, making the
603 # validation on empty desired storage classes return an error.
604 def default_storage_classes
605 if self.storage_classes_desired.nil? || self.storage_classes_desired.empty?
606 self.storage_classes_desired = ["default"]
608 self.storage_classes_confirmed ||= []
611 def portable_manifest_text
612 self.class.munge_manifest_locators(manifest_text) do |match|
622 portable_manifest = portable_manifest_text
623 (Digest::MD5.hexdigest(portable_manifest) +
625 portable_manifest.bytesize.to_s)
629 if @computed_pdh_for_manifest_text == manifest_text
632 @computed_pdh = compute_pdh
633 @computed_pdh_for_manifest_text = manifest_text.dup
637 def ensure_permission_to_save
638 if (not current_user.andand.is_admin)
639 if (replication_confirmed_at_changed? or replication_confirmed_changed?) and
640 not (replication_confirmed_at.nil? and replication_confirmed.nil?)
641 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
643 if (storage_classes_confirmed_changed? or storage_classes_confirmed_at_changed?) and
644 not (storage_classes_confirmed == [] and storage_classes_confirmed_at.nil?)
645 raise ArvadosModel::PermissionDeniedError.new("storage_classes_confirmed and storage_classes_confirmed_at attributes cannot be changed, except by setting them to [] and nil respectively")
651 def ensure_storage_classes_desired_is_not_empty
652 if self.storage_classes_desired.empty?
653 raise ArvadosModel::InvalidStateTransitionError.new("storage_classes_desired shouldn't be empty")
657 def ensure_storage_classes_contain_non_empty_strings
658 (self.storage_classes_desired + self.storage_classes_confirmed).each do |c|
659 if !c.is_a?(String) || c == ''
660 raise ArvadosModel::InvalidStateTransitionError.new("storage classes should only be non-empty strings")
665 def past_versions_cannot_be_updated
667 errors.add(:base, "past versions cannot be updated")
672 def versioning_metadata_updates
674 if !is_past_version? && current_version_uuid_changed?
675 errors.add(:current_version_uuid, "cannot be updated")
679 errors.add(:version, "cannot be updated")
687 self.current_version_uuid ||= self.uuid