On 2020/03/05 12:32, Kyotaro Horiguchi wrote:
At Wed, 4 Mar 2020 21:07:05 +0900, Fujii Masao <masao.fu...@oss.nttdata.com>
wrote in
The function that you are proposing is really enough for this use
case?
What if malicious users directly change the oid of function
to < FirstNormalObjectId? Or you're assuming that malicious users will
never log in as superuser and not be able to change the oid?
That's a good point! I'm surprised that user is allowed to update an
oid of database object. In addition, surprisingly we can update it to
0, which in turn leads the assertion failure:
Since non-superusers are not allowed to do that by default,
that's not so bad? That is, to avoid such unexpected change of oid,
admin just should prevent malicious users from logging in as
superusers
and not give the permission on system catalogs to such users.
I think there is still insider threats. As long as we depend on
superuser privilege to do some DBA work, a malicious DBA might be able
to log in as superuser and modify oid.
Yes. But I'm sure that DBA has already considered the measures
againt such threads. Otherwise malicious users can do anything
more malicious rather than changing oid.
Agreed. So that's not a serious problem in practice but we cannot say
the checking by pg_is_user_object() is totally enough for checking
whether malicious object exists or not. Is that right?
Yes.
My opinion is that, if malious users are not allowed to log in
as superusers and the admin give no permission on the system
schema/catalog to them, checking whether the object is defined
under pg_catalog schema or not is enough for your purpose.
Because they are also not allowed to create the object under
pg_catalog. pg_is_user_object() seems not necessary.
OTOH, if you address the case where malicious users can create
the object under pg_catalog, of course, checking whether
the object is defined under pg_catalog schema or not is enough
for the purpose. But pg_is_user_object() is also not enough
because such users can change oid.
The discussion seems assuming the feature is related to some security
measure. But I think I haven't seen the objective or use case for the
feature. I don't see how we should treat them according the result
from the "user-defined objects detection" feature.
For example, we could decide a function whether to be pushed-out or
not to remote server on postgres_fdw. In this case, we need to ask "is
the behavior of this function known to us?", in short, "is this
function is predefined?". In this use case, we have no concern if DBA
have added some functions as "not user-defined", since it's their own
risk.
I don't come up with another use cases but, anyway, I think we need to
clarify the scope of the feature.
Agreed. Also we would need to consider that the existing approach
(e.g., checking whether the object is defined under pg_catalog or not,
or seeing pg_stat_user_functions, _indexes, and _tables) is enough
for the use cases. If enough, new function might not be necessary.
If not enough, we might also need to reconsider the definitions of
pg_stat_user_xxx after considering the function.
Regards,
--
Fujii Masao
NTT DATA CORPORATION
Advanced Platform Technology Group
Research and Development Headquarters
