X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/5f3df47f1369ca700f2405fca4231055989765c5..8c30c649c667f7ccd205cdfcf1864563ac503585:/services/api/app/models/collection.rb?ds=sidebyside diff --git a/services/api/app/models/collection.rb b/services/api/app/models/collection.rb index 91d1d1e108..ddb85862a2 100644 --- a/services/api/app/models/collection.rb +++ b/services/api/app/models/collection.rb @@ -1,35 +1,56 @@ +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: AGPL-3.0 + require 'arvados/keep' +require 'sweep_trashed_objects' +require 'trashable' class Collection < ArvadosModel + extend CurrentApiClient extend DbCurrentTime include HasUuid include KindAndEtag include CommonApiTemplate + include Trashable serialize :properties, Hash + serialize :storage_classes_desired, Array + serialize :storage_classes_confirmed, Array + before_validation :default_empty_manifest + before_validation :default_storage_classes, on: :create before_validation :check_encoding + before_validation :check_manifest_validity before_validation :check_signatures before_validation :strip_signatures_and_update_replication_confirmed - before_validation :set_portable_data_hash validate :ensure_pdh_matches_manifest_text + validate :ensure_storage_classes_desired_is_not_empty + validate :ensure_storage_classes_contain_non_empty_strings before_save :set_file_names - # Query only undeleted collections by default. - default_scope where("expires_at IS NULL or expires_at > CURRENT_TIMESTAMP") - api_accessible :user, extend: :common do |t| t.add :name t.add :description t.add :properties t.add :portable_data_hash t.add :signed_manifest_text, as: :manifest_text + t.add :manifest_text, as: :unsigned_manifest_text t.add :replication_desired t.add :replication_confirmed t.add :replication_confirmed_at + t.add :storage_classes_desired + t.add :storage_classes_confirmed + t.add :storage_classes_confirmed_at + t.add :delete_at + t.add :trash_at + t.add :is_trashed end - LOCATOR_REGEXP = /^([[:xdigit:]]{32})(\+([[:digit:]]+))?(\+([[:upper:]][[:alnum:]+@_-]*))?$/ + after_initialize do + @signatures_checked = false + @computed_pdh_for_manifest_text = false + end def self.attributes_required_columns super.merge( @@ -38,10 +59,22 @@ class Collection < ArvadosModel # expose signed_manifest_text as manifest_text in the # API response, and never let clients select the # manifest_text column. - 'manifest_text' => ['manifest_text'], + # + # We need trash_at and is_trashed to determine the + # correct timestamp in signed_manifest_text. + 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'], + 'unsigned_manifest_text' => ['manifest_text'], ) end + def self.ignored_select_attributes + super + ["updated_at", "file_names"] + end + + def self.limit_index_columns_read + ["manifest_text"] + end + FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/ def check_signatures return false if self.manifest_text.nil? @@ -53,20 +86,24 @@ class Collection < ArvadosModel # subsequent passes without checking any signatures. This is # important because the signatures have probably been stripped off # by the time we get to a second validation pass! - return true if @signatures_checked and @signatures_checked == computed_pdh + if @signatures_checked && @signatures_checked == computed_pdh + return true + end if self.manifest_text_changed? # Check permissions on the collection manifest. # If any signature cannot be verified, raise PermissionDeniedError # which will return 403 Permission denied to the client. - api_token = current_api_client_authorization.andand.api_token + api_token = Thread.current[:token] signing_opts = { api_token: api_token, - now: db_current_time.to_i, + now: @validation_timestamp.to_i, } - self.manifest_text.lines.each do |entry| - entry.split[1..-1].each do |tok| - if tok =~ FILE_TOKEN + self.manifest_text.each_line do |entry| + entry.split.each do |tok| + if tok == '.' or tok.starts_with? './' + # Stream name token. + elsif tok =~ FILE_TOKEN # This is a filename token, not a blob locator. Note that we # keep checking tokens after this, even though manifest # format dictates that all subsequent tokens will also be @@ -95,66 +132,53 @@ class Collection < ArvadosModel def strip_signatures_and_update_replication_confirmed if self.manifest_text_changed? in_old_manifest = {} - self.class.munge_manifest_locators!(manifest_text_was) do |match| - in_old_manifest[match[1]] = true + if not self.replication_confirmed.nil? + self.class.each_manifest_locator(manifest_text_was) do |match| + in_old_manifest[match[1]] = true + end end - cleared_replication_confirmed = false - - # Remove any permission signatures from the manifest. - self[:manifest_text] = self.class.munge_manifest_locators!(self[:manifest_text]) do |match| - if not cleared_replication_confirmed and not in_old_manifest[match[1]] + stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match| + if not self.replication_confirmed.nil? and not in_old_manifest[match[1]] + # If the new manifest_text contains locators whose hashes + # weren't in the old manifest_text, storage replication is no + # longer confirmed. self.replication_confirmed_at = nil self.replication_confirmed = nil - cleared_replication_confirmed = true end - self.class.locator_without_signature(match) - end - end - @computed_pdh_for_manifest_text = self[:manifest_text] if @computed_pdh_for_manifest_text - true - end - def self.locator_without_signature match - without_signature = match[1] - without_signature += match[2] if match[2] - if match[4] - hints = match[4].split('+').reject { |hint| hint.start_with?("A") } - without_signature += hints.join('+') - end - without_signature - end + # Return the locator with all permission signatures removed, + # but otherwise intact. + match[0].gsub(/\+A[^+]*/, '') + end - def set_portable_data_hash - if (portable_data_hash.nil? or - portable_data_hash == "" or - (manifest_text_changed? and !portable_data_hash_changed?)) - self.portable_data_hash = computed_pdh - elsif portable_data_hash_changed? - begin - loc = Keep::Locator.parse!(self.portable_data_hash) - loc.strip_hints! - if loc.size - self.portable_data_hash = loc.to_s - else - self.portable_data_hash = "#{loc.hash}+#{portable_manifest_text.bytesize}" - end - rescue ArgumentError => e - errors.add(:portable_data_hash, "#{e}") - return false + if @computed_pdh_for_manifest_text == manifest_text + # If the cached PDH was valid before stripping, it is still + # valid after stripping. + @computed_pdh_for_manifest_text = stripped_manifest.dup end + + self[:manifest_text] = stripped_manifest end true end def ensure_pdh_matches_manifest_text - return true unless manifest_text_changed? or portable_data_hash_changed? - # No need verify it if :set_portable_data_hash just computed it! - expect_pdh = computed_pdh - if expect_pdh != portable_data_hash + if not manifest_text_changed? and not portable_data_hash_changed? + true + elsif portable_data_hash.nil? or not portable_data_hash_changed? + self.portable_data_hash = computed_pdh + elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP + errors.add(:portable_data_hash, "is not a valid locator") + false + elsif portable_data_hash[0..31] != computed_pdh[0..31] errors.add(:portable_data_hash, - "does not match computed hash #{expect_pdh}") - return false + "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'") + false + else + # Ignore the client-provided size part: always store + # computed_pdh in the database. + self.portable_data_hash = computed_pdh end end @@ -166,22 +190,20 @@ class Collection < ArvadosModel end def manifest_files + return '' if !self.manifest_text + names = '' - if self.manifest_text - self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name| - names << name.first.gsub('\040',' ') + "\n" - break if names.length > 2**12 - end + self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name| + names << name.first.gsub('\040',' ') + "\n" end - - if self.manifest_text and names.length < 2**12 - self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name| - names << stream_name.first.gsub('\040',' ') + "\n" - break if names.length > 2**12 - end + self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name| + names << stream_name.first.gsub('\040',' ') + "\n" end + names + end - names[0,2**12] + def default_empty_manifest + self.manifest_text ||= '' end def check_encoding @@ -196,7 +218,7 @@ class Collection < ArvadosModel utf8 = manifest_text utf8.force_encoding Encoding::UTF_8 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8) - manifest_text = utf8 + self.manifest_text = utf8 return true end rescue @@ -206,60 +228,90 @@ class Collection < ArvadosModel end end + def check_manifest_validity + begin + Keep::Manifest.validate! manifest_text + true + rescue ArgumentError => e + errors.add :manifest_text, e.message + false + end + end + def signed_manifest_text - if has_attribute? :manifest_text - token = current_api_client_authorization.andand.api_token - @signed_manifest_text = self.class.sign_manifest manifest_text, token + if !has_attribute? :manifest_text + return nil + elsif is_trashed + return manifest_text + else + token = Thread.current[:token] + exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl, + trash_at].compact.map(&:to_i).min + self.class.sign_manifest manifest_text, token, exp end end - def self.sign_manifest manifest, token + def self.sign_manifest manifest, token, exp=nil + if exp.nil? + exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl + end signing_opts = { api_token: token, - expire: db_current_time.to_i + Rails.configuration.blob_signature_ttl, + expire: exp, } - m = manifest.dup - m = munge_manifest_locators!(m) do |match| - Blob.sign_locator(locator_without_signature(match), signing_opts) + m = munge_manifest_locators(manifest) do |match| + Blob.sign_locator(match[0], signing_opts) end return m end - def self.munge_manifest_locators! manifest - # Given a manifest text and a block, yield each locator, - # and replace it with whatever the block returns. + def self.munge_manifest_locators manifest + # Given a manifest text and a block, yield the regexp MatchData + # for each locator. Return a new manifest in which each locator + # has been replaced by the block's return value. + return nil if !manifest + return '' if manifest == '' + new_lines = [] - lines = manifest.andand.split("\n") - lines.andand.each do |line| - words = line.split(' ') + manifest.each_line do |line| + line.rstrip! new_words = [] - words.each do |word| - if match = LOCATOR_REGEXP.match(word.strip) + line.split(' ').each do |word| + if new_words.empty? + new_words << word + elsif match = Keep::Locator::LOCATOR_REGEXP.match(word) new_words << yield(match) else - new_words << word.strip + new_words << word end end new_lines << new_words.join(' ') end + new_lines.join("\n") + "\n" + end - if !new_lines.empty? - ends_with_newline = manifest.end_with?("\n") - manifest = new_lines.join("\n") - manifest += "\n" if ends_with_newline + def self.each_manifest_locator manifest + # Given a manifest text and a block, yield the regexp match object + # for each locator. + manifest.each_line do |line| + # line will have a trailing newline, but the last token is never + # a locator, so it's harmless here. + line.split(' ').each do |word| + if match = Keep::Locator::LOCATOR_REGEXP.match(word) + yield(match) + end + end end - - manifest end def self.normalize_uuid uuid hash_part = nil size_part = nil uuid.split('+').each do |token| - if token.match /^[0-9a-f]{32,}$/ + if token.match(/^[0-9a-f]{32,}$/) raise "uuid #{uuid} has multiple hash parts" if hash_part hash_part = token - elsif token.match /^\d+$/ + elsif token.match(/^\d+$/) raise "uuid #{uuid} has multiple size parts" if size_part size_part = token end @@ -268,8 +320,53 @@ class Collection < ArvadosModel [hash_part, size_part].compact.join '+' end - # Return array of Collection objects - def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil) + def self.get_compatible_images(readers, pattern, collections) + if collections.empty? + return [] + end + + migrations = Hash[ + Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?', + collections.map(&:portable_data_hash), + 'docker_image_migration', + system_user_uuid). + order('links.created_at asc'). + map { |l| + [l.tail_uuid, l.head_uuid] + }] + + migrated_collections = Hash[ + Collection.readable_by(*readers). + where('portable_data_hash in (?)', migrations.values). + map { |c| + [c.portable_data_hash, c] + }] + + collections.map { |c| + # Check if the listed image is compatible first, if not, then try the + # migration link. + manifest = Keep::Manifest.new(c.manifest_text) + if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern + c + elsif m = migrated_collections[migrations[c.portable_data_hash]] + manifest = Keep::Manifest.new(m.manifest_text) + if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern + m + end + end + }.compact + end + + # Resolve a Docker repo+tag, hash, or collection PDH to an array of + # Collection objects, sorted by timestamp starting with the most recent + # match. + # + # If filter_compatible_format is true (the default), only return image + # collections which are support by the installation as indicated by + # Rails.configuration.docker_image_formats. Will follow + # 'docker_image_migration' links if search_term resolves to an incompatible + # image, but an equivalent compatible image is available. + def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true) readers ||= [Thread.current[:user]] base_search = Link. readable_by(*readers). @@ -277,20 +374,23 @@ class Collection < ArvadosModel joins("JOIN collections ON links.head_uuid = collections.uuid"). order("links.created_at DESC") + if (Rails.configuration.docker_image_formats.include? 'v1' and + Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false + pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/ + elsif Rails.configuration.docker_image_formats.include? 'v2' + pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/ + elsif Rails.configuration.docker_image_formats.include? 'v1' + pattern = /^[0-9A-Fa-f]{64}\.tar$/ + else + raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}" + end + # If the search term is a Collection locator that contains one file # that looks like a Docker image, return it. if loc = Keep::Locator.parse(search_term) loc.strip_hints! - coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first - if coll_match - # Check if the Collection contains exactly one file whose name - # looks like a saved Docker image. - manifest = Keep::Manifest.new(coll_match.manifest_text) - if manifest.exact_file_count?(1) and - (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/) - return [coll_match] - end - end + coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1) + return get_compatible_images(readers, pattern, coll_match) end if search_tag.nil? and (n = search_term.index(":")) @@ -314,11 +414,19 @@ class Collection < ArvadosModel # so that anything with an image timestamp is considered more recent than # anything without; then we use the link's created_at as a tiebreaker. uuid_timestamps = {} - matches.all.map do |link| + matches.each do |link| uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0), -link.created_at.to_i] + end + + sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| + uuid_timestamps[c.uuid] + } + compatible = get_compatible_images(readers, pattern, sorted) + if sorted.length > 0 and compatible.empty? + raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration." end - Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] } + compatible end def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil) @@ -330,20 +438,34 @@ class Collection < ArvadosModel end def self.full_text_searchable_columns - super - ["manifest_text"] + super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed"] + end + + def self.where *args + SweepTrashedObjects.sweep_if_stale + super end protected + + # Although the defaults for these columns is already set up on the schema, + # collection creation from an API client seems to ignore them, making the + # validation on empty desired storage classes return an error. + def default_storage_classes + if self.storage_classes_desired.nil? || self.storage_classes_desired.empty? + self.storage_classes_desired = ["default"] + end + self.storage_classes_confirmed ||= [] + end + def portable_manifest_text - portable_manifest = self[:manifest_text].dup - portable_manifest = self.class.munge_manifest_locators!(portable_manifest) do |match| + self.class.munge_manifest_locators(manifest_text) do |match| if match[2] # size match[1] + match[2] else match[1] end end - portable_manifest end def compute_pdh @@ -363,11 +485,30 @@ class Collection < ArvadosModel end def ensure_permission_to_save - if (not current_user.andand.is_admin and - (replication_confirmed_at_changed? or replication_confirmed_changed?) and - not (replication_confirmed_at.nil? and replication_confirmed.nil?)) - raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil") + if (not current_user.andand.is_admin) + if (replication_confirmed_at_changed? or replication_confirmed_changed?) and + not (replication_confirmed_at.nil? and replication_confirmed.nil?) + raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil") + end + if (storage_classes_confirmed_changed? or storage_classes_confirmed_at_changed?) and + not (storage_classes_confirmed == [] and storage_classes_confirmed_at.nil?) + raise ArvadosModel::PermissionDeniedError.new("storage_classes_confirmed and storage_classes_confirmed_at attributes cannot be changed, except by setting them to [] and nil respectively") + end end super end + + def ensure_storage_classes_desired_is_not_empty + if self.storage_classes_desired.empty? + raise ArvadosModel::InvalidStateTransitionError.new("storage_classes_desired shouldn't be empty") + end + end + + def ensure_storage_classes_contain_non_empty_strings + (self.storage_classes_desired + self.storage_classes_confirmed).each do |c| + if !c.is_a?(String) || c == '' + raise ArvadosModel::InvalidStateTransitionError.new("storage classes should only be non-empty strings") + end + end + end end