require 'serializers'
require 'request_error'
-class ArvadosModel < ActiveRecord::Base
+class ArvadosModel < ApplicationRecord
self.abstract_class = true
include ArvadosModelUpdates
include DbCurrentTime
extend RecordFilters
+ after_find :schedule_restoring_changes
after_initialize :log_start_state
before_save :ensure_permission_to_save
before_save :ensure_owner_uuid_is_permitted
after_create :log_create
after_update :log_update
after_destroy :log_destroy
- after_find :convert_serialized_symbols_to_strings
before_validation :normalize_collection_uuids
before_validation :set_default_owner
validate :ensure_valid_uuids
class_name: 'Link',
primary_key: :uuid)
+ # If async is true at create or update, permission graph
+ # update is deferred allowing making multiple calls without the performance
+ # penalty.
+ attr_accessor :async_permissions_update
+
+ # Ignore listed attributes on mass assignments
+ def self.protected_attributes
+ []
+ end
+
class PermissionDeniedError < RequestError
def http_status
403
# The following permit! is necessary even with
# "ActionController::Parameters.permit_all_parameters = true",
# because permit_all does not permit nested attributes.
+ raw_params ||= {}
+
if raw_params
+ raw_params = raw_params.to_hash
+ raw_params.delete_if { |k, _| self.protected_attributes.include? k }
serialized_attributes.each do |colname, coder|
param = raw_params[colname.to_sym]
if param.nil?
raise ArgumentError.new("#{colname} parameter cannot have non-string hash keys")
end
end
+ # Check JSONB columns that aren't listed on serialized_attributes
+ columns.select{|c| c.type == :jsonb}.collect{|j| j.name}.each do |colname|
+ if serialized_attributes.include? colname || raw_params[colname.to_sym].nil?
+ next
+ end
+ if has_nonstring_keys?(raw_params[colname.to_sym])
+ raise ArgumentError.new("#{colname} parameter cannot have non-string hash keys")
+ end
+ end
end
ActionController::Parameters.new(raw_params).permit!
end
def reload(*args)
super
log_start_state
+ self
end
def self.create raw_params={}, *args
sql_conds = nil
user_uuids = users_list.map { |u| u.uuid }
+ all_user_uuids = []
+
+ # For details on how the trashed_groups table is constructed, see
+ # see db/migrate/20200501150153_permission_table.rb
exclude_trashed_records = ""
if !include_trash and (sql_table == "groups" or sql_table == "collections") then
- # Only include records that are not explicitly trashed
- exclude_trashed_records = "AND #{sql_table}.is_trashed = false"
+ # Only include records that are not trashed
+ exclude_trashed_records = "AND (#{sql_table}.trash_at is NULL or #{sql_table}.trash_at > statement_timestamp())"
+ end
+
+ trashed_check = ""
+ if !include_trash && sql_table != "api_client_authorizations"
+ trashed_check = "#{sql_table}.owner_uuid NOT IN (SELECT group_uuid FROM #{TRASHED_GROUPS} " +
+ "where trash_at <= statement_timestamp()) #{exclude_trashed_records}"
end
if users_list.select { |u| u.is_admin }.any?
# Admin skips most permission checks, but still want to filter on trashed items.
- if !include_trash
- if sql_table != "api_client_authorizations"
- # Only include records where the owner is not trashed
- sql_conds = "NOT EXISTS(SELECT 1 FROM #{PERMISSION_VIEW} "+
- "WHERE trashed = 1 AND "+
- "(#{sql_table}.owner_uuid = target_uuid)) #{exclude_trashed_records}"
- end
+ if !include_trash && sql_table != "api_client_authorizations"
+ # Only include records where the owner is not trashed
+ sql_conds = trashed_check
end
else
- trashed_check = ""
- if !include_trash then
- trashed_check = "AND trashed = 0"
- end
+ # The core of the permission check is a join against the
+ # materialized_permissions table to determine if the user has at
+ # least read permission to either the object itself or its
+ # direct owner (if traverse_owned is true). See
+ # db/migrate/20200501150153_permission_table.rb for details on
+ # how the permissions are computed.
+
+ # A user can have can_manage access to another user, this grants
+ # full access to all that user's stuff. To implement that we
+ # need to include those other users in the permission query.
+
+ # This was previously implemented by embedding the subquery
+ # directly into the query, but it was discovered later that this
+ # causes the Postgres query planner to do silly things because
+ # the query heuristics assumed the subquery would have a lot
+ # more rows that it does, and choose a bad merge strategy. By
+ # doing the query here and embedding the result as a constant,
+ # Postgres also knows exactly how many items there are and can
+ # choose the right query strategy.
+ #
+ # (note: you could also do this with a temporary table, but that
+ # would require all every request be wrapped in a transaction,
+ # which is not currently the case).
+
+ all_user_uuids = ActiveRecord::Base.connection.exec_query %{
+#{USER_UUIDS_SUBQUERY_TEMPLATE % {user: "'#{user_uuids.join "', '"}'", perm_level: 1}}
+},
+ 'readable_by.user_uuids'
+
+ user_uuids_subquery = ":user_uuids"
# Note: it is possible to combine the direct_check and
- # owner_check into a single EXISTS() clause, however it turns
+ # owner_check into a single IN (SELECT) clause, however it turns
# out query optimizer doesn't like it and forces a sequential
- # table scan. Constructing the query with separate EXISTS()
+ # table scan. Constructing the query with separate IN (SELECT)
# clauses enables it to use the index.
#
# see issue 13208 for details.
# Match a direct read permission link from the user to the record uuid
- direct_check = "EXISTS(SELECT 1 FROM #{PERMISSION_VIEW} "+
- "WHERE user_uuid IN (:user_uuids) AND perm_level >= 1 #{trashed_check} AND target_uuid = #{sql_table}.uuid)"
-
- # Match a read permission link from the user to the record's owner_uuid
+ direct_check = "#{sql_table}.uuid IN (SELECT target_uuid FROM #{PERMISSION_VIEW} "+
+ "WHERE user_uuid IN (#{user_uuids_subquery}) AND perm_level >= 1)"
+
+ # Match a read permission for the user to the record's
+ # owner_uuid. This is so we can have a permissions table that
+ # mostly consists of users and groups (projects are a type of
+ # group) and not have to compute and list user permission to
+ # every single object in the system.
+ #
+ # Don't do this for API keys (special behavior) or groups
+ # (already covered by direct_check).
+ #
+ # The traverse_owned flag indicates whether the permission to
+ # read an object also implies transitive permission to read
+ # things the object owns. The situation where this is important
+ # are determining if we can read an object owned by another
+ # user. This makes it possible to have permission to read the
+ # user record without granting permission to read things the
+ # other user owns.
owner_check = ""
if sql_table != "api_client_authorizations" and sql_table != "groups" then
- owner_check = "OR EXISTS(SELECT 1 FROM #{PERMISSION_VIEW} "+
- "WHERE user_uuid IN (:user_uuids) AND perm_level >= 1 #{trashed_check} AND target_uuid = #{sql_table}.owner_uuid AND target_owner_uuid IS NOT NULL) "
+ owner_check = "#{sql_table}.owner_uuid IN (SELECT target_uuid FROM #{PERMISSION_VIEW} "+
+ "WHERE user_uuid IN (#{user_uuids_subquery}) AND perm_level >= 1 AND traverse_owned) "
+
+ # We want to do owner_check before direct_check in the OR
+ # clause. The order of the OR clause isn't supposed to
+ # matter, but in practice, it does -- apparently in the
+ # absence of other hints, it uses the ordering from the query.
+ # For certain types of queries (like filtering on owner_uuid),
+ # every item will match the owner_check clause, so then
+ # Postgres will optimize out the direct_check entirely.
+ direct_check = " OR " + direct_check
end
links_cond = ""
# users some permission _or_ gives anyone else permission to
# view one of the authorized users.
links_cond = "OR (#{sql_table}.link_class IN (:permission_link_classes) AND "+
- "(#{sql_table}.head_uuid IN (:user_uuids) OR #{sql_table}.tail_uuid IN (:user_uuids)))"
+ "(#{sql_table}.head_uuid IN (#{user_uuids_subquery}) OR #{sql_table}.tail_uuid IN (#{user_uuids_subquery})))"
end
- sql_conds = "(#{direct_check} #{owner_check} #{links_cond}) #{exclude_trashed_records}"
+ sql_conds = "(#{owner_check} #{direct_check} #{links_cond}) #{trashed_check.empty? ? "" : "AND"} #{trashed_check}"
end
end
self.where(sql_conds,
- user_uuids: user_uuids,
+ user_uuids: all_user_uuids.collect{|c| c["target_uuid"]},
permission_link_classes: ['permission', 'resources'])
end
# 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
+ err = rn.cause
raise unless err.is_a?(PG::UniqueViolation)
# Unfortunately ActiveRecord doesn't abstract out any of the
end
end
+ def user_owner_uuid
+ if self.owner_uuid.nil?
+ return current_user.uuid
+ end
+ owner_class = ArvadosModel.resource_class_for_uuid(self.owner_uuid)
+ if owner_class == User
+ self.owner_uuid
+ else
+ owner_class.find_by_uuid(self.owner_uuid).user_owner_uuid
+ end
+ end
+
def logged_attributes
- attributes.except(*Rails.configuration.unlogged_attributes)
+ attributes.except(*Rails.configuration.AuditLogs.UnloggedAttributes.stringify_keys.keys)
end
def self.full_text_searchable_columns
end.map(&:name)
end
+ def self.full_text_coalesce
+ full_text_searchable_columns.collect do |column|
+ is_jsonb = self.columns.select{|x|x.name == column}[0].type == :jsonb
+ cast = (is_jsonb || serialized_attributes[column]) ? '::text' : ''
+ "coalesce(#{column}#{cast},'')"
+ end
+ end
+
+ def self.full_text_trgm
+ "(#{full_text_coalesce.join(" || ' ' || ")})"
+ end
+
def self.full_text_tsvector
parts = full_text_searchable_columns.collect do |column|
- cast = serialized_attributes[column] ? '::text' : ''
+ is_jsonb = self.columns.select{|x|x.name == column}[0].type == :jsonb
+ cast = (is_jsonb || serialized_attributes[column]) ? '::text' : ''
"coalesce(#{column}#{cast},'')"
end
- "to_tsvector('english', #{parts.join(" || ' ' || ")})"
+ "to_tsvector('english', substr(#{parts.join(" || ' ' || ")}, 0, 8000))"
end
def self.apply_filters query, filters
if not ft[:cond_out].any?
return query
end
+ ft[:joins].each do |t|
+ query = query.joins(t)
+ end
query.where('(' + ft[:cond_out].join(') AND (') + ')',
*ft[:param_out])
end
end
rescue ActiveRecord::RecordNotFound => e
errors.add :owner_uuid, "is not owned by any user: #{e}"
- return false
+ throw(:abort)
end
if uuid_in_path[x]
if x == owner_uuid
else
errors.add :owner_uuid, "has an ownership cycle"
end
- return false
+ throw(:abort)
end
uuid_in_path[x] = true
end
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
+ elsif rsc_class == Group && Group.find_by_uuid(owner_uuid).group_class != "project"
+ errors.add :owner_uuid, "must be a project"
+ raise PermissionDeniedError
end
end
else
# 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"
+ errors.add :uuid, " #{uuid} is not writable by #{current_user.uuid}"
raise PermissionDeniedError
end
end
end
def permission_to_destroy
- permission_to_update
+ if [system_user_uuid, system_group_uuid, anonymous_group_uuid,
+ anonymous_user_uuid, public_project_uuid].include? uuid
+ false
+ else
+ permission_to_update
+ end
end
def maybe_update_modified_by_fields
self.owner_uuid ||= current_default_owner if self.respond_to? :owner_uuid=
if !anonymous_updater
self.modified_by_user_uuid = current_user ? current_user.uuid : nil
+ end
+ if !timeless_updater
self.modified_at = current_time
end
self.modified_by_client_uuid = current_api_client ? current_api_client.uuid : nil
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
- elsif x.is_a? Array
- x.each do |k|
- return true if has_symbols?(k)
- end
- 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
- if x.is_a? Hash
- Hash[x.collect do |k,v|
- [recursive_stringify(k), recursive_stringify(v)]
- end]
- elsif x.is_a? Array
- x.collect do |k|
- recursive_stringify k
- 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 self.where_serialized(colname, value, md5: false)
colsql = colname.to_s
if md5
self.class.serialized_attributes
end
- def convert_serialized_symbols_to_strings
- # ensure_serialized_attribute_type should prevent symbols from
- # getting into the database in the first place. If someone managed
- # to get them into the database (perhaps using an older version)
- # we'll convert symbols to strings when loading from the
- # database. (Otherwise, loading and saving an object with existing
- # symbols in a serialized field will crash.)
- self.class.serialized_attributes.each do |colname, attr|
- if self.class.has_symbols? attributes[colname]
- attributes[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$/) }
end
end
def self.uuid_like_pattern
- "#{Rails.configuration.uuid_prefix}-#{uuid_prefix}-_______________"
+ "#{Rails.configuration.ClusterID}-#{uuid_prefix}-_______________"
end
def self.uuid_regex
%r/[a-z0-9]{5}-#{uuid_prefix}-[a-z0-9]{15}/
end
+ def check_readable_uuid attr, attr_value
+ return if attr_value.nil?
+ if (r = ArvadosModel::resource_class_for_uuid attr_value)
+ unless skip_uuid_read_permission_check.include? attr
+ r = r.readable_by(current_user)
+ end
+ if r.where(uuid: attr_value).count == 0
+ errors.add(attr, "'#{attr_value}' not found")
+ end
+ else
+ # Not a valid uuid or PDH, but that (currently) is not an error.
+ end
+ end
+
def ensure_valid_uuids
specials = [system_user_uuid]
next if skip_uuid_existence_check.include? attr
attr_value = send attr
next if specials.include? attr_value
- if attr_value
- if (r = ArvadosModel::resource_class_for_uuid attr_value)
- unless skip_uuid_read_permission_check.include? attr
- r = r.readable_by(current_user)
- end
- if r.where(uuid: attr_value).count == 0
- errors.add(attr, "'#{attr_value}' not found")
- end
- end
- end
+ check_readable_uuid attr, attr_value
end
end
end
+ def ensure_filesystem_compatible_name
+ if name == "." || name == ".."
+ errors.add(:name, "cannot be '.' or '..'")
+ elsif Rails.configuration.Collections.ForwardSlashNameSubstitution == "" && !name.nil? && name.index('/')
+ errors.add(:name, "cannot contain a '/' character")
+ end
+ end
+
class Email
def self.kind
"email"
nil
end
+ # Fill in implied zero/false values in database records that were
+ # created before #17014 made them explicit, and reset the Rails
+ # "changed" state so the record doesn't appear to have been modified
+ # after loading.
+ #
+ # Invoked by Container and ContainerRequest models as an after_find
+ # hook.
+ def fill_container_defaults_after_find
+ fill_container_defaults
+ set_attribute_was('runtime_constraints', runtime_constraints)
+ set_attribute_was('scheduling_parameters', scheduling_parameters)
+ clear_changes_information
+ end
+
+ # Fill in implied zero/false values. Invoked by ContainerRequest as
+ # a before_validation hook in order to (a) ensure every key has a
+ # value in the updated database record and (b) ensure the attribute
+ # whitelist doesn't reject a change from an explicit zero/false
+ # value in the database to an implicit zero/false value in an update
+ # request.
+ def fill_container_defaults
+ self.runtime_constraints = {
+ 'API' => false,
+ 'keep_cache_ram' => 0,
+ 'ram' => 0,
+ 'vcpus' => 0,
+ }.merge(attributes['runtime_constraints'] || {})
+ self.scheduling_parameters = {
+ 'max_run_time' => 0,
+ 'partitions' => [],
+ 'preemptible' => false,
+ }.merge(attributes['scheduling_parameters'] || {})
+ end
+
# ArvadosModel.find_by_uuid needs extra magic to allow it to return
# an object in any class.
def self.find_by_uuid uuid
end
def is_audit_logging_enabled?
- return !(Rails.configuration.max_audit_log_age.to_i == 0 &&
- Rails.configuration.max_audit_log_delete_batch.to_i > 0)
+ return !(Rails.configuration.AuditLogs.MaxAge.to_i == 0 &&
+ Rails.configuration.AuditLogs.MaxDeleteBatch.to_i > 0)
+ end
+
+ def schedule_restoring_changes
+ # This will be checked at log_start_state, to reset any (virtual) changes
+ # produced by the act of reading a serialized attribute.
+ @fresh_from_database = true
end
def log_start_state
if is_audit_logging_enabled?
@old_attributes = Marshal.load(Marshal.dump(attributes))
@old_logged_attributes = Marshal.load(Marshal.dump(logged_attributes))
+ if @fresh_from_database
+ # This instance was created from reading a database record. Attributes
+ # haven't been changed, but those serialized attributes will be reported
+ # as unpersisted, so we restore them to avoid issues with lock!() and
+ # with_lock().
+ restore_attributes
+ @fresh_from_database = nil
+ end
end
end
projects / arvados.git / blobdiff