Given the following snippet from my schema:
create table users (
id serial primary key,
name text not null
);
create table user_groups (
id serial primary key,
name text not null
);
create table user_user_group (
user_id integer not null references users(id),
user_group_id integer not null references user_groups(id)
);
grant all on users to staff;
grant all on user_groups to staff;
grant all on user_user_group to staff;
create function can_access_user_group(id integer) returns boolean as $$
select exists(
select 1
from user_user_group
where user_group_id = id
and user_id = current_user_id()
);
$$ language sql stable security invoker;
create function can_access_user(id integer) returns boolean as $$
select exists(
select 1
from user_user_group
where user_id = id
and can_access_user_group(user_group_id)
);
$$ language sql stable security invoker;
alter table users enable row level security;
create policy staff_users_policy
on users
to staff
using (
can_access_user(id)
);
Please assume the staff
role, and current_user_id()
function are tested and working correctly. I'm hoping to allow the "staff" role to create users in user groups they can access via the user_user_group
table. The following statement fails the staff_users_policy
:
begin;
set local role staff;
with new_user as (
insert into users (
name
) values (
'Some name'
)
returning id
)
insert into user_user_group (
user_id,
user_group_id
)
select
new_user.id,
1 as user_group_id
from new_user;
commit;
I can add a staff_insert_users_policy
like this:
create policy staff_insert_users_policy
on users
for insert
to staff
with check (
true
);
Which allows me to insert the user but fails on returning id
, and I need the new user id in order to add the row to the user_user_group
table.
I understand why it fails, but conceptually how can I avoid this problem? I could create a "definer" function, or a new role with it's own policy just for this but I'm hoping there's a more straightforward approach.