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 :versioning_metadata_updates, on: :update
31 validate :past_versions_cannot_be_updated, on: :update
32 validate :file_count_and_size_cannot_be_changed
33 before_save :set_file_names
34 before_save :set_file_count_and_total_size
35 around_update :manage_versioning
37 api_accessible :user, extend: :common do |t|
41 t.add :portable_data_hash
42 t.add :signed_manifest_text, as: :manifest_text
43 t.add :manifest_text, as: :unsigned_manifest_text
44 t.add :replication_desired
45 t.add :replication_confirmed
46 t.add :replication_confirmed_at
47 t.add :storage_classes_desired
48 t.add :storage_classes_confirmed
49 t.add :storage_classes_confirmed_at
54 t.add :current_version_uuid
55 t.add :preserve_version
57 t.add :file_size_total
61 @signatures_checked = false
62 @computed_pdh_for_manifest_text = false
65 def self.attributes_required_columns
67 # If we don't list manifest_text explicitly, the
68 # params[:select] code gets confused by the way we
69 # expose signed_manifest_text as manifest_text in the
70 # API response, and never let clients select the
71 # manifest_text column.
73 # We need trash_at and is_trashed to determine the
74 # correct timestamp in signed_manifest_text.
75 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
76 'unsigned_manifest_text' => ['manifest_text'],
80 def self.ignored_select_attributes
81 super + ["updated_at", "file_names"]
84 def self.limit_index_columns_read
88 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
90 return false if self.manifest_text.nil?
92 return true if current_user.andand.is_admin
94 # Provided the manifest_text hasn't changed materially since an
95 # earlier validation, it's safe to pass this validation on
96 # subsequent passes without checking any signatures. This is
97 # important because the signatures have probably been stripped off
98 # by the time we get to a second validation pass!
99 if @signatures_checked && @signatures_checked == computed_pdh
103 if self.manifest_text_changed?
104 # Check permissions on the collection manifest.
105 # If any signature cannot be verified, raise PermissionDeniedError
106 # which will return 403 Permission denied to the client.
107 api_token = Thread.current[:token]
109 api_token: api_token,
110 now: @validation_timestamp.to_i,
112 self.manifest_text.each_line do |entry|
113 entry.split.each do |tok|
114 if tok == '.' or tok.starts_with? './'
116 elsif tok =~ FILE_TOKEN
117 # This is a filename token, not a blob locator. Note that we
118 # keep checking tokens after this, even though manifest
119 # format dictates that all subsequent tokens will also be
120 # filenames. Safety first!
121 elsif Blob.verify_signature tok, signing_opts
123 elsif Keep::Locator.parse(tok).andand.signature
124 # Signature provided, but verify_signature did not like it.
125 logger.warn "Invalid signature on locator #{tok}"
126 raise ArvadosModel::PermissionDeniedError
127 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
128 # No signature provided, but we are running in insecure mode.
129 logger.debug "Missing signature on locator #{tok} ignored"
130 elsif Blob.new(tok).empty?
131 # No signature provided -- but no data to protect, either.
133 logger.warn "Missing signature on locator #{tok}"
134 raise ArvadosModel::PermissionDeniedError
139 @signatures_checked = computed_pdh
142 def strip_signatures_and_update_replication_confirmed
143 if self.manifest_text_changed?
145 if not self.replication_confirmed.nil?
146 self.class.each_manifest_locator(manifest_text_was) do |match|
147 in_old_manifest[match[1]] = true
151 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
152 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
153 # If the new manifest_text contains locators whose hashes
154 # weren't in the old manifest_text, storage replication is no
156 self.replication_confirmed_at = nil
157 self.replication_confirmed = nil
160 # Return the locator with all permission signatures removed,
161 # but otherwise intact.
162 match[0].gsub(/\+A[^+]*/, '')
165 if @computed_pdh_for_manifest_text == manifest_text
166 # If the cached PDH was valid before stripping, it is still
167 # valid after stripping.
168 @computed_pdh_for_manifest_text = stripped_manifest.dup
171 self[:manifest_text] = stripped_manifest
176 def ensure_pdh_matches_manifest_text
177 if not manifest_text_changed? and not portable_data_hash_changed?
179 elsif portable_data_hash.nil? or not portable_data_hash_changed?
180 self.portable_data_hash = computed_pdh
181 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
182 errors.add(:portable_data_hash, "is not a valid locator")
184 elsif portable_data_hash[0..31] != computed_pdh[0..31]
185 errors.add(:portable_data_hash,
186 "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
189 # Ignore the client-provided size part: always store
190 # computed_pdh in the database.
191 self.portable_data_hash = computed_pdh
196 if self.manifest_text_changed?
197 self.file_names = manifest_files
202 def set_file_count_and_total_size
203 if self.manifest_text_changed? && self.valid?
204 m = Keep::Manifest.new(self.manifest_text)
205 self.file_size_total = m.files_size
206 self.file_count = m.files_count
212 return '' if !self.manifest_text
216 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
219 names << name.first.gsub('\040',' ') + "\n"
221 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
222 next if done[stream_name]
223 done[stream_name] = true
224 names << stream_name.first.gsub('\040',' ') + "\n"
229 def default_empty_manifest
230 self.manifest_text ||= ''
233 def skip_uuid_existence_check
234 # Avoid checking the existence of current_version_uuid, as it's
235 # assigned on creation of a new 'current version' collection, so
236 # the collection's UUID only lives on memory when the validation check
238 ['current_version_uuid']
241 def manage_versioning
242 should_preserve_version = should_preserve_version? # Time sensitive, cache value
243 return(yield) unless (should_preserve_version || syncable_updates.any?)
245 # Put aside the changes because with_lock forces a record reload
246 changes = self.changes
249 # Copy the original state to save it as old version
250 if should_preserve_version
252 snapshot.uuid = nil # Reset UUID so it's created as a new record
253 snapshot.created_at = self.created_at
256 # Restore requested changes on the current version
257 changes.keys.each do |attr|
258 if attr == 'preserve_version' && changes[attr].last == false
259 next # Ignore false assignment, once true it'll be true until next version
261 self.attributes = {attr => changes[attr].last}
263 # Also update the current version reference
264 self.attributes = {'current_version_uuid' => changes[attr].last}
268 if should_preserve_version
270 self.preserve_version = false
275 sync_past_versions if syncable_updates.any?
277 snapshot.attributes = self.syncable_updates
278 snapshot.manifest_text = snapshot.signed_manifest_text
286 (syncable_attrs & self.changes.keys).each do |attr|
288 # Point old versions to current version's new UUID
289 updates['current_version_uuid'] = self.changes[attr].last
291 updates[attr] = self.changes[attr].last
297 def sync_past_versions
298 updates = self.syncable_updates
299 Collection.where('current_version_uuid = ? AND uuid != ?', self.uuid_was, self.uuid_was).each do |c|
300 c.attributes = updates
301 # Use a different validation context to skip the 'old_versions_cannot_be_updated'
302 # validator, as on this case it is legal to update some fields.
303 leave_modified_by_user_alone do
304 leave_modified_at_alone do
305 c.save(context: :update_old_versions)
311 def versionable_updates?(attrs)
312 (['manifest_text', 'description', 'properties', 'name'] & attrs).any?
316 ['uuid', 'owner_uuid', 'delete_at', 'trash_at', 'is_trashed', 'replication_desired', 'storage_classes_desired']
319 def should_preserve_version?
320 return false unless (Rails.configuration.collection_versioning && versionable_updates?(self.changes.keys))
322 idle_threshold = Rails.configuration.preserve_version_if_idle
323 if !self.preserve_version_was &&
324 (idle_threshold < 0 ||
325 (idle_threshold > 0 && self.modified_at_was > db_current_time-idle_threshold.seconds))
332 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
336 # If Ruby thinks the encoding is something else, like 7-bit
337 # ASCII, but its stored bytes are equal to the (valid) UTF-8
338 # encoding of the same string, we declare it to be a UTF-8
341 utf8.force_encoding Encoding::UTF_8
342 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
343 self.manifest_text = utf8
348 errors.add :manifest_text, "must use UTF-8 encoding"
353 def check_manifest_validity
355 Keep::Manifest.validate! manifest_text
357 rescue ArgumentError => e
358 errors.add :manifest_text, e.message
363 def signed_manifest_text
364 if !has_attribute? :manifest_text
369 token = Thread.current[:token]
370 exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
371 trash_at].compact.map(&:to_i).min
372 self.class.sign_manifest manifest_text, token, exp
376 def self.sign_manifest manifest, token, exp=nil
378 exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
384 m = munge_manifest_locators(manifest) do |match|
385 Blob.sign_locator(match[0], signing_opts)
390 def self.munge_manifest_locators manifest
391 # Given a manifest text and a block, yield the regexp MatchData
392 # for each locator. Return a new manifest in which each locator
393 # has been replaced by the block's return value.
394 return nil if !manifest
395 return '' if manifest == ''
398 manifest.each_line do |line|
401 line.split(' ').each do |word|
404 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
405 new_words << yield(match)
410 new_lines << new_words.join(' ')
412 new_lines.join("\n") + "\n"
415 def self.each_manifest_locator manifest
416 # Given a manifest text and a block, yield the regexp match object
418 manifest.each_line do |line|
419 # line will have a trailing newline, but the last token is never
420 # a locator, so it's harmless here.
421 line.split(' ').each do |word|
422 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
429 def self.normalize_uuid uuid
432 uuid.split('+').each do |token|
433 if token.match(/^[0-9a-f]{32,}$/)
434 raise "uuid #{uuid} has multiple hash parts" if hash_part
436 elsif token.match(/^\d+$/)
437 raise "uuid #{uuid} has multiple size parts" if size_part
441 raise "uuid #{uuid} has no hash part" if !hash_part
442 [hash_part, size_part].compact.join '+'
445 def self.get_compatible_images(readers, pattern, collections)
446 if collections.empty?
451 Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
452 collections.map(&:portable_data_hash),
453 'docker_image_migration',
455 order('links.created_at asc').
457 [l.tail_uuid, l.head_uuid]
460 migrated_collections = Hash[
461 Collection.readable_by(*readers).
462 where('portable_data_hash in (?)', migrations.values).
464 [c.portable_data_hash, c]
467 collections.map { |c|
468 # Check if the listed image is compatible first, if not, then try the
470 manifest = Keep::Manifest.new(c.manifest_text)
471 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
473 elsif m = migrated_collections[migrations[c.portable_data_hash]]
474 manifest = Keep::Manifest.new(m.manifest_text)
475 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
482 # Resolve a Docker repo+tag, hash, or collection PDH to an array of
483 # Collection objects, sorted by timestamp starting with the most recent
486 # If filter_compatible_format is true (the default), only return image
487 # collections which are support by the installation as indicated by
488 # Rails.configuration.docker_image_formats. Will follow
489 # 'docker_image_migration' links if search_term resolves to an incompatible
490 # image, but an equivalent compatible image is available.
491 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
492 readers ||= [Thread.current[:user]]
494 readable_by(*readers).
495 readable_by(*readers, table_name: "collections").
496 joins("JOIN collections ON links.head_uuid = collections.uuid").
497 order("links.created_at DESC")
499 if (Rails.configuration.docker_image_formats.include? 'v1' and
500 Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
501 pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
502 elsif Rails.configuration.docker_image_formats.include? 'v2'
503 pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
504 elsif Rails.configuration.docker_image_formats.include? 'v1'
505 pattern = /^[0-9A-Fa-f]{64}\.tar$/
507 raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
510 # If the search term is a Collection locator that contains one file
511 # that looks like a Docker image, return it.
512 if loc = Keep::Locator.parse(search_term)
514 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
515 if coll_match.any? or Rails.configuration.remote_hosts.length == 0
516 return get_compatible_images(readers, pattern, coll_match)
518 # Allow bare pdh that doesn't exist in the local database so
519 # that federated container requests which refer to remotely
520 # stored containers will validate.
521 return [Collection.new(portable_data_hash: loc.to_s)]
525 if search_tag.nil? and (n = search_term.index(":"))
526 search_tag = search_term[n+1..-1]
527 search_term = search_term[0..n-1]
530 # Find Collections with matching Docker image repository+tag pairs.
531 matches = base_search.
532 where(link_class: "docker_image_repo+tag",
533 name: "#{search_term}:#{search_tag || 'latest'}")
535 # If that didn't work, find Collections with matching Docker image hashes.
537 matches = base_search.
538 where("link_class = ? and links.name LIKE ?",
539 "docker_image_hash", "#{search_term}%")
542 # Generate an order key for each result. We want to order the results
543 # so that anything with an image timestamp is considered more recent than
544 # anything without; then we use the link's created_at as a tiebreaker.
546 matches.each do |link|
547 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
548 -link.created_at.to_i]
551 sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
552 uuid_timestamps[c.uuid]
554 compatible = get_compatible_images(readers, pattern, sorted)
555 if sorted.length > 0 and compatible.empty?
556 raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
561 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
562 find_all_for_docker_image(search_term, search_tag, readers).first
565 def self.searchable_columns operator
566 super - ["manifest_text"]
569 def self.full_text_searchable_columns
570 super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed", "current_version_uuid"]
574 SweepTrashedObjects.sweep_if_stale
580 # Although the defaults for these columns is already set up on the schema,
581 # collection creation from an API client seems to ignore them, making the
582 # validation on empty desired storage classes return an error.
583 def default_storage_classes
584 if self.storage_classes_desired.nil? || self.storage_classes_desired.empty?
585 self.storage_classes_desired = ["default"]
587 self.storage_classes_confirmed ||= []
590 def portable_manifest_text
591 self.class.munge_manifest_locators(manifest_text) do |match|
601 portable_manifest = portable_manifest_text
602 (Digest::MD5.hexdigest(portable_manifest) +
604 portable_manifest.bytesize.to_s)
608 if @computed_pdh_for_manifest_text == manifest_text
611 @computed_pdh = compute_pdh
612 @computed_pdh_for_manifest_text = manifest_text.dup
616 def ensure_permission_to_save
617 if (not current_user.andand.is_admin)
618 if (replication_confirmed_at_changed? or replication_confirmed_changed?) and
619 not (replication_confirmed_at.nil? and replication_confirmed.nil?)
620 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
622 if (storage_classes_confirmed_changed? or storage_classes_confirmed_at_changed?) and
623 not (storage_classes_confirmed == [] and storage_classes_confirmed_at.nil?)
624 raise ArvadosModel::PermissionDeniedError.new("storage_classes_confirmed and storage_classes_confirmed_at attributes cannot be changed, except by setting them to [] and nil respectively")
630 def ensure_storage_classes_desired_is_not_empty
631 if self.storage_classes_desired.empty?
632 raise ArvadosModel::InvalidStateTransitionError.new("storage_classes_desired shouldn't be empty")
636 def ensure_storage_classes_contain_non_empty_strings
637 (self.storage_classes_desired + self.storage_classes_confirmed).each do |c|
638 if !c.is_a?(String) || c == ''
639 raise ArvadosModel::InvalidStateTransitionError.new("storage classes should only be non-empty strings")
644 def past_versions_cannot_be_updated
645 # We check for the '_was' values just in case the update operation
646 # includes a change on current_version_uuid or uuid.
647 if current_version_uuid_was != uuid_was
648 errors.add(:base, "past versions cannot be updated")
653 def file_count_and_size_cannot_be_changed
655 if file_count_changed?
656 errors.add(:file_count, "cannot be changed")
659 if file_size_total_changed?
660 errors.add(:file_size_total, "cannot be changed")
666 def versioning_metadata_updates
668 if (current_version_uuid_was == uuid_was) && current_version_uuid_changed?
669 errors.add(:current_version_uuid, "cannot be updated")
673 errors.add(:version, "cannot be updated")
681 self.current_version_uuid ||= self.uuid