I'm getting somewhat confused with SQL Server security
we have a login and a user: test
we have a table: dbo.tblSessionFilter
User test has no select and no delete permission on this table (I tested this!!)
Then we have a procedure:
create procedure dbo.procFilter_Clear with execute as caller
as
delete from dbo.tblSessionfilter
where spid = ##SPID
User test has execute right on this procedure.
And now, user test can call this procedure and can delete entries from the table; although he has no direct delete access on the table, and the procedure is execute as caller !
How is that possible ?
Is it probably because procedure and table are in the same schema?
See Ownership Chains:
When multiple database objects access each other sequentially, the sequence is known as a chain. Although such chains do not independently exist, when SQL Server traverses the links in a chain, SQL Server evaluates permissions on the constituent objects differently than it would if it were accessing the objects separately.
and,
When an object is accessed through a chain, SQL Server first compares the owner of the object to the owner of the calling object. This is the previous link in the chain. If both objects have the same owner, permissions on the referenced object are not evaluated.
(My emphasis)
Related
I have a stored procedure where I'd like to save the execution of an update statement for further down in the procedure, which I'm doing by putting the sql into a varchar variable.
My problem is that the EXECUTE() seems to use the original caller's permissions and then gives an error, as the caller does not (and should not) have UPDATE permissions for table1.
Why doesn't it use the permissions of the stored procedure's owner?
And more importantly: how can I make it use those permissions?
The stored procedure looks basically like this: (except #tsql_string is set dynamically)
create procedure dbo.woopwoop() as
BEGIN
DECLARE #tsql_string varchar(max)
SET #tsql_string = 'UPDATE table1 set c1 = 1'
EXECUTE(#tsql_string)
END
I'm on SQL Server 2008, and the owner (me) of the stored procedure is a Windows Authentication user with domain\user.
Anyone?
That's what the EXECUTE AS clause exists for - to allow you to control this:
By specifying the context in which the module is executed, you can control which user account the Database Engine uses to validate permissions on objects that are referenced by the module.
CALLER
Specifies the statements inside the module are executed in the context of the caller of the module. The user executing the module must have appropriate permissions not only on the module itself, but also on any database objects that are referenced by the module.
CALLER is the default for all modules except queues
You would be wanting to specify that the stored procedure executes as OWNER (or possibly SELF, read the above page and decide which makes most sense for you) instead.
create procedure dbo.woopwoop()
with execute as owner
as
....
From MSDN (This example grants EXECUTE permission on the stored procedure HumanResources.uspUpdateEmployeeHireInfo to an application role named Recruiting11):
USE AdventureWorks2012;
GRANT EXECUTE ON OBJECT::HumanResources.uspUpdateEmployeeHireInfo
TO Recruiting11;
GO
Source
Of course you should have permission to grant the permission.
I want to know what are the default permissions for sql server stored procedures.
For example, I created a user in database but no role or no permissions granted except execute permission like:
GRANT EXECUTE ON SCHEMA :: [dbo] TO [newUser]
later on me (as a sa login) created a stored procedure that makes just a select from a table like:
CREATE PROCEDURE dbo.selectX AS
BEGIN
SELECT ID, NAME FROM MyTable
END
and my user [newUser] can execute and see the result of select statement like:
EXEC dbo.selectX
up until now, I can assume that stored procedures have default permissions that my user don't have.
I tried other dmls like UPDATE, INSERT, DELETE and user managed to execute procedures successfully.
But! when I created a proceure with ddl as create table .. such as:
CREATE PROCEDURE dbo.crtNT AS
BEGIN
CREATE TABLE del(id int) --for test
END
and when the newUser executes this procedure error occurs like:
CREATE TABLE permission denied in database 'MyDb'
And here is my question: By default, do stored procedures have permission with DML(select, insert, update vs.) but not with DDL(create, drop table vs.)? Because I didn't see any explanation about this situation in Books Online or anywhere else.
Any comments will be welcomed.
You are correct in your assumption that DDL statements are treated differently.
The procedures with DML statements work through the mechanism of Ownership Chaining however as this quote from the EXECUTE AS documentation states
Remember that ownership chaining applies only to DML statements.
I have a SQL Server 2008 database where all access to the underlying tables is done through stored procedures. Some stored procedures simply SELECT records from the tables while others UPDATE, INSERT, and DELETE.
If a stored procedure UPDATES a table does the user executing the stored procedure also need UPDATE permissions to the affected tables or is the fact that they have EXECUTE permissions to the stored procedure enough?
Basically I am wondering if giving the user EXECUTE permissions to the stored procedures is enough or do I need to give them SELECT, UPDATE, DELETE, and INSERT permissions to the tables in order for the stored procedures to work. Thank you.
[EDIT] In most of my stored procedures it does indeed appear that EXECUTE is enough. However, I did find that in stored procedures where "Execute sp_Executesql" was used that EXECUTE was not enough. The tables involved needed to have permissions for the actions being performed within "sp_Executesql".
Permissions on tables are not checked (including DENY) if tables and proc have the same owner. They can be in different schemas too as long as the schemas have the same owner.
See Ownership chaining on MSDN
Edit, from a comment from a deleted answer.
The context is always the current login unless EXECUTE AS as been used: only referenced object DML permissions are not checked. Try OBJECT_ID(referencedtable) in a stored proc where no rights are assigned to referencedtable. It gives NULL. If executed by the owner of the stored proc then it would give a value because owener has rights on referencedtable
Execute permissions on the stored procedure is sufficient.
CREATE TABLE dbo.Temp(n int)
GO
DENY INSERT ON dbo.Temp TO <your role>
GO
CREATE PROCEDURE dbo.SPTemp(#Int int)
AS
INSERT dbo.Temp
SELECT #Int
GO
GRANT EXEC ON dbo.SPTemp TO <your role>
GO
Then the (non-db_owner) user will have the following rights:
EXEC dbo.SPTemp 10
GO
INSERT dbo.Temp --INSERT permission was denied on the object 'Temp'
SELECT 10
However, if there is dynamic SQL inside dbo.SPTemp that attempts to insert into dbo.Temp then that will fail. In this case direct permission on the table will need to be granted.
Maybe you can use
"with execute as owner"
when you create the stored procedure, such as below:
create procedure XXX
with execute as owner
as
begin
...
end
go
Then you only need to grant the user the EXECUTE permission for the stored procedure XXX.
Execute permission on a stored procedure that does an insert, update, or delete is sufficient. You do not need to grant those permissions at the table level. In fact, I would discourage that approach. Using a stored procedure gives you more control over how the change occurs. For instance, you may wish to do some checking prior to allowing the update. Using a stored procedure can also help prevent major accidents--like deleting all the rows in the table because someone forgot the WHERE clause!
THANK YOU SO MUCH! I had a similar problem. This lead me to the answer.
I was attempting to trunctate a table in a stored procedure that called other stored procedures that were nested in IF statements.
My error was
The server principal "domain\my_id" is not able to access the database "2nd_DB" under the current security context.
I had given the calling stored procedure rights to do the truncate (EXECUTE AS SELF), which then caused a problem because SELF didn't have rights to the 2nd DB. Our solution was to move the truncate to another SP, include the EXECUTE AS SELF. We now call the truncate SP, execute our data processing, make logic determination, and call the appropriate 3rd SP.
Here's the situation. I'm writing an automated test that walks the list of dependencies for a proc and determines if an acct has rights for all of the dependent objects.
My code looks like this:
exec sp_depends 'the_proc_name'
-- run this query on the results of sp_depends:
select
case
when exists (
select *
from sys.database_permissions dp
where grantee_principal_id=USER_ID('TheAccount')
and major_id=object_id('dbo.theDependentObject')
and minor_id=0
and state_desc = 'GRANT')
then 'true'
else 'false'
end;
It all seems to be working fine, but there's a hiccup when it encounters a function. I have one case where TheAccount doesn't have rights to a function (the query above returns false). However the proc that calls the function in question runs fine when running under TheAccount. So there's either something wrong with my test code or functions have special permission behavior in SQL-Server that I'm not aware of.
Should I change the code to only search for 'DENY' instead of 'GRANT'? Do functions that are called in procs inherit the permissions of the calling proc except when the execute rights are explicitly denied? Does my code suck?
When you are running static SQL from a stored proc, the stored proc runs with the authority of the id that last created/modified the stored proc; not the id of the person running the stored proc.
For example, this is the same ability that allows you to use a stored proc to run an Insert statement without giving the person running the stored proc Insert Authority on the underlying table.
One note - this does not apply when you are using dynamic SQL (with the exec statement). In that case, the person running the stored proc, must have the necessary authority for that SQL statement.
So I'm not sure if your unit tests will provide you what you are looking for since the rights to dependent objects are taken care of, to some extent, by the way SQL Server handles Stored Proc security.
I'm trying to write a table trigger which queries another table that is outside the schema where the trigger will reside. Is this possible? It seems like I have no problem querying tables in my schema but I get:
Error: ORA-00942: table or view does not exist
when trying trying to query tables outside my schema.
EDIT
My apologies for not providing as much information as possible the first time around. I was under the impression this question was more simple.
I'm trying create a trigger on a table that changes some fields on a newly inserted row based on the existence of some data that may or may not be in a table that is in another schema.
The user account that I'm using to create the trigger does have the permissions to run the queries independently. In fact, I've had my trigger print the query I'm trying to run and was able to run it on it's own successfully.
I should also note that I'm building the query dynamically by using the EXECUTE IMMEDIATE statement. Here's an example:
CREATE OR REPLACE TRIGGER MAIN_SCHEMA.EVENTS
BEFORE INSERT
ON MAIN_SCHEMA.EVENTS REFERENCING OLD AS OLD NEW AS NEW
FOR EACH ROW
DECLARE
rtn_count NUMBER := 0;
table_name VARCHAR2(17) := :NEW.SOME_FIELD;
key_field VARCHAR2(20) := :NEW.ANOTHER_FIELD;
BEGIN
CASE
WHEN (key_field = 'condition_a') THEN
EXECUTE IMMEDIATE 'select count(*) from OTHER_SCHEMA_A.'||table_name||' where KEY_FIELD='''||key_field||'''' INTO rtn_count;
WHEN (key_field = 'condition_b') THEN
EXECUTE IMMEDIATE 'select count(*) from OTHER_SCHEMA_B.'||table_name||' where KEY_FIELD='''||key_field||'''' INTO rtn_count;
WHEN (key_field = 'condition_c') THEN
EXECUTE IMMEDIATE 'select count(*) from OTHER_SCHEMA_C.'||table_name||' where KEY_FIELD='''||key_field||'''' INTO rtn_count;
END CASE;
IF (rtn_count > 0) THEN
-- change some fields that are to be inserted
END IF;
END;
The trigger seams to fail on the EXECUTE IMMEDIATE with the previously mentioned error.
EDIT
I have done some more research and I can offer more clarification.
The user account I'm using to create this trigger is not MAIN_SCHEMA or any one of the OTHER_SCHEMA_Xs. The account I'm using (ME) is given privileges to the involved tables via the schema users themselves. For example (USER_TAB_PRIVS):
GRANTOR GRANTEE TABLE_SCHEMA TABLE_NAME PRIVILEGE GRANTABLE HIERARCHY
MAIN_SCHEMA ME MAIN_SCHEMA EVENTS DELETE NO NO
MAIN_SCHEMA ME MAIN_SCHEMA EVENTS INSERT NO NO
MAIN_SCHEMA ME MAIN_SCHEMA EVENTS SELECT NO NO
MAIN_SCHEMA ME MAIN_SCHEMA EVENTS UPDATE NO NO
OTHER_SCHEMA_X ME OTHER_SCHEMA_X TARGET_TBL SELECT NO NO
And I have the following system privileges (USER_SYS_PRIVS):
USERNAME PRIVILEGE ADMIN_OPTION
ME ALTER ANY TRIGGER NO
ME CREATE ANY TRIGGER NO
ME UNLIMITED TABLESPACE NO
And this is what I found in the Oracle documentation:
To create a trigger in another user's
schema, or to reference a table in
another schema from a trigger in your
schema, you must have the CREATE ANY
TRIGGER system privilege. With this
privilege, the trigger can be created
in any schema and can be associated
with any user's table. In addition,
the user creating the trigger must
also have EXECUTE privilege on the
referenced procedures, functions, or
packages.
Here: Oracle Doc
So it looks to me like this should work, but I'm not sure about the "EXECUTE privilege" it's referring to in the doc.
What you are experiencing is a feature of Oracle's security model. The entire point of using schemas is to control access to the data. The tables in my schema are mine, you cannot even see them until I grant you privileges on them.
The syntax is quite simple: the owner schema issues
grant select, insert on my_table to you
/
Alternatively an account with the GRANT ANY privilege (such as a DBA) can pass privileges on any user's objects.
grant select, insert on apc.my_table to you
/
The grantee can be either a user or a role. However, note that we can only build program units - stored procedures, views, triggers - using privileges which have been granted directly through to our user.
So, if you get the other schema owner to grant you the necessary privileges you will be able to build your trigger.
edit
When referencing an object in another schema we need to qualify the object with the schema name ....
insert into apc.whatever_table values ...
or else we need to create a synonym for it
create synonym whatever for apc.whatever_table;
I feel someone should add the obvious - the other schema's table must be qualified with the schema name or a private/public synonym is needed. I wonder if the original problem was merely a name resolution issue. If not, APC's answer is a good explanation of the Oracle security model.
You should execute this for every table and schema involved:
grant select on OTHER_SCHEMA_%.table_name to MAIN_SCHEMA;