X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/6eb3d1fb8fe71623fa63da46c250184cf2e4fbb8..441ef97e93a951b349356df96d8a6ef604c6cab7:/services/api/app/models/arvados_model.rb diff --git a/services/api/app/models/arvados_model.rb b/services/api/app/models/arvados_model.rb index 936f823a5e..d1a0bc5794 100644 --- a/services/api/app/models/arvados_model.rb +++ b/services/api/app/models/arvados_model.rb @@ -1,15 +1,14 @@ require 'has_uuid' +require 'record_filters' +require 'serializers' class ArvadosModel < ActiveRecord::Base self.abstract_class = true include CurrentApiClient # current_user, current_api_client, etc. include DbCurrentTime + extend RecordFilters - attr_protected :created_at - attr_protected :modified_by_user_uuid - attr_protected :modified_by_client_uuid - attr_protected :modified_at after_initialize :log_start_state before_save :ensure_permission_to_save before_save :ensure_owner_uuid_is_permitted @@ -23,13 +22,17 @@ class ArvadosModel < ActiveRecord::Base after_destroy :log_destroy after_find :convert_serialized_symbols_to_strings before_validation :normalize_collection_uuids - validate :ensure_serialized_attribute_type + before_validation :set_default_owner validate :ensure_valid_uuids # Note: This only returns permission links. It does not account for # permissions obtained via user.is_admin or # user.uuid==object.owner_uuid. - has_many :permissions, :foreign_key => :head_uuid, :class_name => 'Link', :primary_key => :uuid, :conditions => "link_class = 'permission'" + has_many(:permissions, + ->{where(link_class: 'permission')}, + foreign_key: :head_uuid, + class_name: 'Link', + primary_key: :uuid) class PermissionDeniedError < StandardError def http_status @@ -39,7 +42,19 @@ class ArvadosModel < ActiveRecord::Base class AlreadyLockedError < StandardError def http_status - 403 + 422 + end + end + + class LockFailedError < StandardError + def http_status + 422 + end + end + + class InvalidStateTransitionError < StandardError + def http_status + 422 end end @@ -49,6 +64,12 @@ class ArvadosModel < ActiveRecord::Base end end + class UnresolvableContainerError < StandardError + def http_status + 422 + end + end + def self.kind_class(kind) kind.match(/^arvados\#(.+)$/)[1].classify.safe_constantize rescue nil end @@ -57,6 +78,46 @@ class ArvadosModel < ActiveRecord::Base "#{current_api_base}/#{self.class.to_s.pluralize.underscore}/#{self.uuid}" end + def self.permit_attribute_params raw_params + # strong_parameters does not provide security: permissions are + # implemented with before_save hooks. + # + # The following permit! is necessary even with + # "ActionController::Parameters.permit_all_parameters = true", + # because permit_all does not permit nested attributes. + if raw_params + serialized_attributes.each do |colname, coder| + param = raw_params[colname.to_sym] + if param.nil? + # ok + elsif !param.is_a?(coder.object_class) + raise ArgumentError.new("#{colname} parameter must be #{coder.object_class}, not #{param.class}") + elsif has_nonstring_keys?(param) + raise ArgumentError.new("#{colname} parameter cannot have non-string hash keys") + end + end + end + ActionController::Parameters.new(raw_params).permit! + end + + def initialize raw_params={}, *args + super(self.class.permit_attribute_params(raw_params), *args) + end + + # Reload "old attributes" for logging, too. + def reload(*args) + super + log_start_state + end + + def self.create raw_params={}, *args + super(permit_attribute_params(raw_params), *args) + end + + def update_attributes raw_params={}, *args + super(self.class.permit_attribute_params(raw_params), *args) + end + def self.selectable_attributes(template=:user) # Return an array of attribute name strings that can be selected # in the given template. @@ -103,10 +164,27 @@ class ArvadosModel < ActiveRecord::Base api_column_map end + def self.ignored_select_attributes + ["href", "kind", "etag"] + end + def self.columns_for_attributes(select_attributes) + if select_attributes.empty? + raise ArgumentError.new("Attribute selection list cannot be empty") + end + api_column_map = attributes_required_columns + invalid_attrs = [] + select_attributes.each do |s| + next if ignored_select_attributes.include? s + if not s.is_a? String or not api_column_map.include? s + invalid_attrs << s + end + end + if not invalid_attrs.empty? + raise ArgumentError.new("Invalid attribute(s): #{invalid_attrs.inspect}") + end # Given an array of attribute names to select, return an array of column # names that must be fetched from the database to satisfy the request. - api_column_map = attributes_required_columns select_attributes.flat_map { |attr| api_column_map[attr] }.uniq end @@ -114,6 +192,18 @@ class ArvadosModel < ActiveRecord::Base ["#{table_name}.modified_at desc", "#{table_name}.uuid"] end + def self.unique_columns + ["id", "uuid"] + end + + def self.limit_index_columns_read + # This method returns a list of column names. + # If an index request reads that column from the database, + # APIs that return lists will only fetch objects until reaching + # max_index_database_read bytes of data from those columns. + [] + end + # If current user can manage the object, return an array of uuids of # users and groups that have permission to write the object. The # first two elements are always [self.owner_uuid, current user's @@ -160,91 +250,140 @@ class ArvadosModel < ActiveRecord::Base # Check if any of the users are admin. If so, we're done. if users_list.select { |u| u.is_admin }.any? - return self + # Return existing relation with no new filters. + return where({}) end - # Collect the uuids for each user and any groups readable by each user. + # Collect the UUIDs of the authorized users. user_uuids = users_list.map { |u| u.uuid } - uuid_list = user_uuids + users_list.flat_map { |u| u.groups_i_can(:read) } - sql_conds = [] - sql_params = [] - sql_table = kwargs.fetch(:table_name, table_name) - or_object_uuid = '' - - # This row is owned by a member of users_list, or owned by a group - # readable by a member of users_list - # or - # This row uuid is the uuid of a member of users_list - # or - # A permission link exists ('write' and 'manage' implicitly include - # 'read') from a member of users_list, or a group readable by users_list, - # to this row, or to the owner of this row (see join() below). - sql_conds += ["#{sql_table}.uuid in (?)"] - sql_params += [user_uuids] - - if uuid_list.any? - sql_conds += ["#{sql_table}.owner_uuid in (?)"] - sql_params += [uuid_list] - sanitized_uuid_list = uuid_list. - collect { |uuid| sanitize(uuid) }.join(', ') - permitted_uuids = "(SELECT head_uuid FROM links WHERE link_class='permission' AND tail_uuid IN (#{sanitized_uuid_list}))" - sql_conds += ["#{sql_table}.uuid IN #{permitted_uuids}"] - end + # Collect the UUIDs of all groups readable by any of the + # authorized users. If one of these (or the UUID of one of the + # authorized users themselves) is an object's owner_uuid, that + # object is readable. + owner_uuids = user_uuids + users_list.flat_map { |u| u.groups_i_can(:read) } + owner_uuids.uniq! - if sql_table == "links" and users_list.any? - # This row is a 'permission' or 'resources' link class - # The uuid for a member of users_list is referenced in either the head - # or tail of the link - sql_conds += ["(#{sql_table}.link_class in (#{sanitize 'permission'}, #{sanitize 'resources'}) AND (#{sql_table}.head_uuid IN (?) OR #{sql_table}.tail_uuid IN (?)))"] - sql_params += [user_uuids, user_uuids] - end + sql_conds = [] + sql_table = kwargs.fetch(:table_name, table_name) - if sql_table == "logs" and users_list.any? - # Link head points to the object described by this row - sql_conds += ["#{sql_table}.object_uuid IN #{permitted_uuids}"] + # Match any object (evidently a group or user) whose UUID is + # listed explicitly in owner_uuids. + sql_conds += ["#{sql_table}.uuid in (:owner_uuids)"] + + # Match any object whose owner is listed explicitly in + # owner_uuids. + sql_conds += ["#{sql_table}.owner_uuid IN (:owner_uuids)"] + + # Match the head of any permission link whose tail is listed + # explicitly in owner_uuids. + sql_conds += ["#{sql_table}.uuid IN (SELECT head_uuid FROM links WHERE link_class='permission' AND tail_uuid IN (:owner_uuids))"] + + if sql_table == "links" + # Match any permission link that gives one of the authorized + # users some permission _or_ gives anyone else permission to + # view one of the authorized users. + sql_conds += ["(#{sql_table}.link_class in (:permission_link_classes) AND "+ + "(#{sql_table}.head_uuid IN (:user_uuids) OR #{sql_table}.tail_uuid IN (:user_uuids)))"] + end + + where(sql_conds.join(' OR '), + owner_uuids: owner_uuids, + user_uuids: user_uuids, + permission_link_classes: ['permission', 'resources']) + end + + def save_with_unique_name! + uuid_was = uuid + name_was = name + max_retries = 2 + transaction do + conn = ActiveRecord::Base.connection + conn.exec_query 'SAVEPOINT save_with_unique_name' + begin + save! + rescue ActiveRecord::RecordNotUnique => rn + raise if max_retries == 0 + max_retries -= 1 + + conn.exec_query 'ROLLBACK TO SAVEPOINT save_with_unique_name' + + # Dig into the error to determine if it is specifically calling out a + # (owner_uuid, name) uniqueness violation. In this specific case, and + # the client requested a unique name with ensure_unique_name==true, + # update the name field and try to save again. Loop as necessary to + # discover a unique name. It is necessary to handle name choosing at + # this level (as opposed to the client) to ensure that record creation + # never fails due to a race condition. + err = rn.original_exception + raise unless err.is_a?(PG::UniqueViolation) + + # Unfortunately ActiveRecord doesn't abstract out any of the + # necessary information to figure out if this the error is actually + # the specific case where we want to apply the ensure_unique_name + # behavior, so the following code is specialized to Postgres. + detail = err.result.error_field(PG::Result::PG_DIAG_MESSAGE_DETAIL) + raise unless /^Key \(owner_uuid, name\)=\([a-z0-9]{5}-[a-z0-9]{5}-[a-z0-9]{15}, .*?\) already exists\./.match detail + + new_name = "#{name_was} (#{db_current_time.utc.iso8601(3)})" + if new_name == name + # If the database is fast enough to do two attempts in the + # same millisecond, we need to wait to ensure we try a + # different timestamp on each attempt. + sleep 0.002 + new_name = "#{name_was} (#{db_current_time.utc.iso8601(3)})" + end - # This object described by this row is owned by this user, or owned by a group readable by this user - sql_conds += ["#{sql_table}.object_owner_uuid in (?)"] - sql_params += [uuid_list] + self[:name] = new_name + self[:uuid] = nil if uuid_was.nil? && !uuid.nil? + conn.exec_query 'SAVEPOINT save_with_unique_name' + retry + ensure + conn.exec_query 'RELEASE SAVEPOINT save_with_unique_name' + end end - - # Link head points to this row, or to the owner of this row (the - # thing to be read) - # - # Link tail originates from this user, or a group that is readable - # by this user (the identity with authorization to read) - # - # Link class is 'permission' ('write' and 'manage' implicitly - # include 'read') - where(sql_conds.join(' OR '), *sql_params) end def logged_attributes - attributes + attributes.except(*Rails.configuration.unlogged_attributes) end def self.full_text_searchable_columns self.columns.select do |col| - if col.type == :string or col.type == :text - true - end + col.type == :string or col.type == :text end.map(&:name) end def self.full_text_tsvector - tsvector_str = "to_tsvector('english', " - first = true - self.full_text_searchable_columns.each do |column| - tsvector_str += " || ' ' || " if not first - tsvector_str += "coalesce(#{column},'')" - first = false + parts = full_text_searchable_columns.collect do |column| + "coalesce(#{column},'')" end - tsvector_str += ")" + "to_tsvector('english', #{parts.join(" || ' ' || ")})" + end + + def self.apply_filters query, filters + ft = record_filters filters, self + if not ft[:cond_out].any? + return query + end + query.where('(' + ft[:cond_out].join(') AND (') + ')', + *ft[:param_out]) end protected + def self.deep_sort_hash(x) + if x.is_a? Hash + x.sort.collect do |k, v| + [k, deep_sort_hash(v)] + end.to_h + elsif x.is_a? Array + x.collect { |v| deep_sort_hash(v) } + else + x + end + end + def ensure_ownership_path_leads_to_user if new_record? or owner_uuid_changed? uuid_in_path = {owner_uuid => true, uuid => true} @@ -277,12 +416,14 @@ class ArvadosModel < ActiveRecord::Base true end - def ensure_owner_uuid_is_permitted - raise PermissionDeniedError if !current_user - - if new_record? and respond_to? :owner_uuid= + def set_default_owner + if new_record? and current_user and respond_to? :owner_uuid= self.owner_uuid ||= current_user.uuid end + end + + def ensure_owner_uuid_is_permitted + raise PermissionDeniedError if !current_user if self.owner_uuid.nil? errors.add :owner_uuid, "cannot be nil" @@ -295,36 +436,31 @@ class ArvadosModel < ActiveRecord::Base raise PermissionDeniedError end - # Verify "write" permission on old owner - # default fail unless one of: - # owner_uuid did not change - # previous owner_uuid is nil - # current user is the old owner - # current user is this object - # current user can_write old owner - unless !owner_uuid_changed? or - owner_uuid_was.nil? or - current_user.uuid == self.owner_uuid_was or - current_user.uuid == self.uuid or - current_user.can? write: self.owner_uuid_was - logger.warn "User #{current_user.uuid} tried to modify #{self.class.to_s} #{uuid} but does not have permission to write old owner_uuid #{owner_uuid_was}" - errors.add :owner_uuid, "cannot be changed without write permission on old owner" - raise PermissionDeniedError - end - - # Verify "write" permission on new owner - # default fail unless one of: - # current_user is this object - # current user can_write new owner, or this object if owner unchanged - if new_record? or owner_uuid_changed? or is_a?(ApiClientAuthorization) - write_target = owner_uuid + if new_record? || owner_uuid_changed? + # Permission on owner_uuid_was is needed to move an existing + # object away from its previous owner (which implies permission + # to modify this object itself, so we don't need to check that + # separately). Permission on the new owner_uuid is also needed. + [['old', owner_uuid_was], + ['new', owner_uuid] + ].each do |which, check_uuid| + if check_uuid.nil? + # old_owner_uuid is nil? New record, no need to check. + elsif !current_user.can?(write: check_uuid) + logger.warn "User #{current_user.uuid} tried to set ownership of #{self.class.to_s} #{self.uuid} but does not have permission to write #{which} owner_uuid #{check_uuid}" + errors.add :owner_uuid, "cannot be set or changed without write permission on #{which} owner" + raise PermissionDeniedError + end + end else - write_target = uuid - end - unless current_user == self or current_user.can? write: write_target - logger.warn "User #{current_user.uuid} tried to modify #{self.class.to_s} #{uuid} but does not have permission to write new owner_uuid #{owner_uuid}" - errors.add :owner_uuid, "cannot be changed without write permission on new owner" - raise PermissionDeniedError + # If the object already existed and we're not changing + # owner_uuid, we only need write permission on the object + # itself. + if !current_user.can?(write: self.uuid) + logger.warn "User #{current_user.uuid} tried to modify #{self.class.to_s} #{self.uuid} without write permission" + errors.add :uuid, "is not writable" + raise PermissionDeniedError + end end true @@ -372,6 +508,7 @@ class ArvadosModel < ActiveRecord::Base def update_modified_by_fields current_time = db_current_time + self.created_at = created_at_was || current_time self.updated_at = current_time self.owner_uuid ||= current_default_owner if self.respond_to? :owner_uuid= self.modified_at = current_time @@ -380,20 +517,34 @@ class ArvadosModel < ActiveRecord::Base true end + def self.has_nonstring_keys? x + if x.is_a? Hash + x.each do |k,v| + return true if !(k.is_a?(String) || k.is_a?(Symbol)) || has_nonstring_keys?(v) + end + elsif x.is_a? Array + x.each do |v| + return true if has_nonstring_keys?(v) + end + end + false + end + def self.has_symbols? x if x.is_a? Hash x.each do |k,v| return true if has_symbols?(k) or has_symbols?(v) end - false elsif x.is_a? Array x.each do |k| return true if has_symbols?(k) end - false - else - (x.class == Symbol) + elsif x.is_a? Symbol + return true + elsif x.is_a? String + return true if x.start_with?(':') && !x.start_with?('::') end + false end def self.recursive_stringify x @@ -407,27 +558,43 @@ class ArvadosModel < ActiveRecord::Base end elsif x.is_a? Symbol x.to_s + elsif x.is_a? String and x.start_with?(':') and !x.start_with?('::') + x[1..-1] else x end end - def ensure_serialized_attribute_type - # Specifying a type in the "serialize" declaration causes rails to - # raise an exception if a different data type is retrieved from - # the database during load(). The validation preventing such - # crash-inducing records from being inserted in the database in - # the first place seems to have been left as an exercise to the - # developer. - self.class.serialized_attributes.each do |colname, attr| - if attr.object_class - if self.attributes[colname].class != attr.object_class - self.errors.add colname.to_sym, "must be a #{attr.object_class.to_s}, not a #{self.attributes[colname].class.to_s}" - elsif self.class.has_symbols? attributes[colname] - self.errors.add colname.to_sym, "must not contain symbols: #{attributes[colname].inspect}" - end - end + def self.where_serialized(colname, value) + if value.empty? + # rails4 stores as null, rails3 stored as serialized [] or {} + sql = "#{colname.to_s} is null or #{colname.to_s} IN (?)" + sorted = value + else + sql = "#{colname.to_s} IN (?)" + sorted = deep_sort_hash(value) end + where(sql, [sorted.to_yaml, SafeJSON.dump(sorted)]) + end + + Serializer = { + Hash => HashSerializer, + Array => ArraySerializer, + } + + def self.serialize(colname, type) + coder = Serializer[type] + @serialized_attributes ||= {} + @serialized_attributes[colname.to_s] = coder + super(colname, coder) + end + + def self.serialized_attributes + @serialized_attributes ||= {} + end + + def serialized_attributes + self.class.serialized_attributes end def convert_serialized_symbols_to_strings @@ -440,14 +607,14 @@ class ArvadosModel < ActiveRecord::Base self.class.serialized_attributes.each do |colname, attr| if self.class.has_symbols? attributes[colname] attributes[colname] = self.class.recursive_stringify attributes[colname] - self.send(colname + '=', - self.class.recursive_stringify(attributes[colname])) + send(colname + '=', + self.class.recursive_stringify(attributes[colname])) end end end def foreign_key_attributes - attributes.keys.select { |a| a.match /_uuid$/ } + attributes.keys.select { |a| a.match(/_uuid$/) } end def skip_uuid_read_permission_check @@ -462,7 +629,7 @@ class ArvadosModel < ActiveRecord::Base foreign_key_attributes.each do |attr| attr_value = send attr if attr_value.is_a? String and - attr_value.match /^[0-9a-f]{32,}(\+[@\w]+)*$/ + attr_value.match(/^[0-9a-f]{32,}(\+[@\w]+)*$/) begin send "#{attr}=", Collection.normalize_uuid(attr_value) rescue @@ -487,7 +654,7 @@ class ArvadosModel < ActiveRecord::Base end def self.uuid_like_pattern - "_____-#{uuid_prefix}-_______________" + "#{Rails.configuration.uuid_prefix}-#{uuid_prefix}-_______________" end def self.uuid_regex @@ -541,13 +708,12 @@ class ArvadosModel < ActiveRecord::Base unless uuid.is_a? String return nil end - resource_class = nil uuid.match HasUuid::UUID_REGEX do |re| return uuid_prefixes[re[1]] if uuid_prefixes[re[1]] end - if uuid.match /.+@.+/ + if uuid.match(/.+@.+/) return Email end @@ -560,7 +726,7 @@ class ArvadosModel < ActiveRecord::Base if self == ArvadosModel # If called directly as ArvadosModel.find_by_uuid rather than via subclass, # delegate to the appropriate subclass based on the given uuid. - self.resource_class_for_uuid(uuid).find_by_uuid(uuid) + self.resource_class_for_uuid(uuid).unscoped.find_by_uuid(uuid) else super end @@ -593,7 +759,7 @@ class ArvadosModel < ActiveRecord::Base end def log_destroy - log_change('destroy') do |log| + log_change('delete') do |log| log.fill_properties('old', etag(@old_attributes), @old_logged_attributes) log.update_to nil end