2 39.5. Rules and Privileges #
4 Due to rewriting of queries by the PostgreSQL rule system, other
5 tables/views than those used in the original query get accessed. When
6 update rules are used, this can include write access to tables.
8 Rewrite rules don't have a separate owner. The owner of a relation
9 (table or view) is automatically the owner of the rewrite rules that
10 are defined for it. The PostgreSQL rule system changes the behavior of
11 the default access control system. With the exception of SELECT rules
12 associated with security invoker views (see CREATE VIEW), all relations
13 that are used due to rules get checked against the privileges of the
14 rule owner, not the user invoking the rule. This means that, except for
15 security invoker views, users only need the required privileges for the
16 tables/views that are explicitly named in their queries.
18 For example: A user has a list of phone numbers where some of them are
19 private, the others are of interest for the assistant of the office.
20 The user can construct the following:
21 CREATE TABLE phone_data (person text, phone text, private boolean);
22 CREATE VIEW phone_number AS
23 SELECT person, CASE WHEN NOT private THEN phone END AS phone
25 GRANT SELECT ON phone_number TO assistant;
27 Nobody except that user (and the database superusers) can access the
28 phone_data table. But because of the GRANT, the assistant can run a
29 SELECT on the phone_number view. The rule system will rewrite the
30 SELECT from phone_number into a SELECT from phone_data. Since the user
31 is the owner of phone_number and therefore the owner of the rule, the
32 read access to phone_data is now checked against the user's privileges
33 and the query is permitted. The check for accessing phone_number is
34 also performed, but this is done against the invoking user, so nobody
35 but the user and the assistant can use it.
37 The privileges are checked rule by rule. So the assistant is for now
38 the only one who can see the public phone numbers. But the assistant
39 can set up another view and grant access to that to the public. Then,
40 anyone can see the phone_number data through the assistant's view. What
41 the assistant cannot do is to create a view that directly accesses
42 phone_data. (Actually the assistant can, but it will not work since
43 every access will be denied during the permission checks.) And as soon
44 as the user notices that the assistant opened their phone_number view,
45 the user can revoke the assistant's access. Immediately, any access to
46 the assistant's view would fail.
48 One might think that this rule-by-rule checking is a security hole, but
49 in fact it isn't. But if it did not work this way, the assistant could
50 set up a table with the same columns as phone_number and copy the data
51 to there once per day. Then it's the assistant's own data and the
52 assistant can grant access to everyone they want. A GRANT command
53 means, “I trust you”. If someone you trust does the thing above, it's
54 time to think it over and then use REVOKE.
56 Note that while views can be used to hide the contents of certain
57 columns using the technique shown above, they cannot be used to
58 reliably conceal the data in unseen rows unless the security_barrier
59 flag has been set. For example, the following view is insecure:
60 CREATE VIEW phone_number AS
61 SELECT person, phone FROM phone_data WHERE phone NOT LIKE '412%';
63 This view might seem secure, since the rule system will rewrite any
64 SELECT from phone_number into a SELECT from phone_data and add the
65 qualification that only entries where phone does not begin with 412 are
66 wanted. But if the user can create their own functions, it is not
67 difficult to convince the planner to execute the user-defined function
68 prior to the NOT LIKE expression. For example:
69 CREATE FUNCTION tricky(text, text) RETURNS bool AS $$
71 RAISE NOTICE '% => %', $1, $2;
74 $$ LANGUAGE plpgsql COST 0.0000000000000000000001;
76 SELECT * FROM phone_number WHERE tricky(person, phone);
78 Every person and phone number in the phone_data table will be printed
79 as a NOTICE, because the planner will choose to execute the inexpensive
80 tricky function before the more expensive NOT LIKE. Even if the user is
81 prevented from defining new functions, built-in functions can be used
82 in similar attacks. (For example, most casting functions include their
83 input values in the error messages they produce.)
85 Similar considerations apply to update rules. In the examples of the
86 previous section, the owner of the tables in the example database could
87 grant the privileges SELECT, INSERT, UPDATE, and DELETE on the shoelace
88 view to someone else, but only SELECT on shoelace_log. The rule action
89 to write log entries will still be executed successfully, and that
90 other user could see the log entries. But they could not create fake
91 entries, nor could they manipulate or remove existing ones. In this
92 case, there is no possibility of subverting the rules by convincing the
93 planner to alter the order of operations, because the only rule which
94 references shoelace_log is an unqualified INSERT. This might not be
95 true in more complex scenarios.
97 When it is necessary for a view to provide row-level security, the
98 security_barrier attribute should be applied to the view. This prevents
99 maliciously-chosen functions and operators from being passed values
100 from rows until after the view has done its work. For example, if the
101 view shown above had been created like this, it would be secure:
102 CREATE VIEW phone_number WITH (security_barrier) AS
103 SELECT person, phone FROM phone_data WHERE phone NOT LIKE '412%';
105 Views created with the security_barrier may perform far worse than
106 views created without this option. In general, there is no way to avoid
107 this: the fastest possible plan must be rejected if it may compromise
108 security. For this reason, this option is not enabled by default.
110 The query planner has more flexibility when dealing with functions that
111 have no side effects. Such functions are referred to as LEAKPROOF, and
112 include many simple, commonly used operators, such as many equality
113 operators. The query planner can safely allow such functions to be
114 evaluated at any point in the query execution process, since invoking
115 them on rows invisible to the user will not leak any information about
116 the unseen rows. Further, functions which do not take arguments or
117 which are not passed any arguments from the security barrier view do
118 not have to be marked as LEAKPROOF to be pushed down, as they never
119 receive data from the view. In contrast, a function that might throw an
120 error depending on the values received as arguments (such as one that
121 throws an error in the event of overflow or division by zero) is not
122 leakproof, and could provide significant information about the unseen
123 rows if applied before the security view's row filters.
125 For example, an index scan cannot be selected for queries on security
126 barrier views (or tables with row-level security policies) if an
127 operator used in the WHERE clause is associated with the operator
128 family of the index, but its underlying function is not marked
129 LEAKPROOF. The psql program's \dAo+ meta-command is useful to list
130 operator families and determine which of their operators are marked as
133 It is important to understand that even a view created with the
134 security_barrier option is intended to be secure only in the limited
135 sense that the contents of the invisible tuples will not be passed to
136 possibly-insecure functions. The user may well have other means of
137 making inferences about the unseen data; for example, they can see the
138 query plan using EXPLAIN, or measure the run time of queries against
139 the view. A malicious attacker might be able to infer something about
140 the amount of unseen data, or even gain some information about the data
141 distribution or most common values (since these things may affect the
142 run time of the plan; or even, since they are also reflected in the
143 optimizer statistics, the choice of plan). If these types of "covert
144 channel" attacks are of concern, it is probably unwise to grant any
145 access to the data at all.