Developing role-based security question

[craig]

I am working on my first .NET development project that involves custom
role-based security per the project requirements. This lead to a general
design issue this week that really caused us some concern. I have described
the situation below because we are very curious to see what other, more
experienced, developers might suggest. The specific classes and fields are
used just to illustrate the concepts.

Our application uses role-based authorization security. Thus, we allow the
administrator to define roles to which the can define various permissions.
Permissions are defined relative to the various entities in the application,
such as Users, Customers, Services, etc. One such role might be defined as
follows:

Role: Administrator
Permissions: Add User, Delete User, View User, Update User

Thus a user that is added to the Administrator role is granted permission to
add, delete, view and update User entities. Once such a user is
authenticated, they will be authorized to perform those actions.

Now, consider that the User entity is abstracted by a User class, and
suppose that class has a Username field (among others). If an authenticated
user indicated that they would like to edit the Username field of a system
User entity, the system first checks to see what role the authenticated user
is in. If they are in an administrator role, they have the desired "Update
User" privilege described above, so the system instantiates a User object
and presents it to the user for editing. If, however, the user does not have
this permission, their request is denied and the system does not instantiate
an object.

So far, so good.

Now, suppose that another requirement is that ANY user that is logged into
the system must be granted permission to edit their own Username, regardless
of what role they are in. In otherwords, the currently authenticated user
should always be granted the ability edit their own Username.

Now, here is where we ran into a problem. If the currently authenticated
user would like to update their Username field, but they are not in an
Administrator role, the system will not allow the User object to be
instantiated, so they will not be able to edit their Username as desired.
The object that provides the functionality needed is not accessible. As we
began to think about this situation, it was not clear how best to solve the
problem and it lead to quite a bit of discussion. Naturally, we wondered if
this is a common problem for which there is a consensus solution in the
developer community, or whether there was just something inherently wrong
with our role-based security model, our object model or our logic.

Here are some of the solutions we considered. I was hoping to hear what
others think.

1. Create a hidden "System" role which has all possible permissions. Add a
hidden "System" user to this role. Anytime such a situation occurs, log out
the current user, log in the System user, perform the desired actions, then
log the current user back in.

2. Create a different class, with different security logic, that can be used
to expose the proper Username field to the currently authenticated user for
editing. This would mean that a single field in the DB would be editable
through two different classes.

3. Add additionaly security logic to the User class such that it functions
as follows: If the authenticated user requesting a User object has Update
User permission, instantiate the requesed User object. If the authenticated
user
does not have Update User permission, allow them the ability to instantiate
only the User entity that corresponds to themselves, and expose only the
Username field for editing (make all other fields read-only).

Any thoughts on the problem in general or the possible solutions? As we
thought about this situation and its possible solutions, it raised a few
other questions:

1. Is it considered good design for two different classes to both provide
read/write access to the same field in the database?

2. Does the use of a "SuperUser" account as described constitute a breech of
security?

3. Is it possible for a situation to arise in which the system might not be
able to discern the security context of a request to instantiate an object?
If so, would it be acceptable to use a parameter to communicate that
context?

Thanks!

[Angelos Karantzalis]

Perhaps a solution - although a slightly more complex one - would be to
explictly set permissions on the user object, rather than the role. What i
mean is, in Windows, any user can belong to groups and get some permission
based on this group membership. However, you can also explicitly set a
permission on the user to say, read a file.

I'd think that in the plumbing, upon instantiation of the User (or perhaps
Identity ) object, it is assigned permissions directly. So, I pseudo-code,
on the constructor of your data class there should be something like:

public myClass() {
new DataAccessPermission(DataAccessPermission.Read)Dem and();
// etc etc.
}

.... that permission demanding code should also be present in the properties
of your class.

What needs to be done for this though, is create your Permission sub-classes
( if an existing Permission class doesn't cover your requirements ), amd
possible implement some "plumbing" inside them to access the permission
you've set on the current user object. It's a bit tricky coding, but once
you've got it, you've a perfect case of re-use for almost any project
afterwards.

I think it's worth taking the time, actually I was planning to create that
sort of thing for the needs of my company as well in the near future,
because we're gonna re-use it in almost any project that requires security.

( This is actually a Java-driven approach, but I think the 2 frameworks are
more-or-less equivalent in functionality, so it shouldn't be very difficult
to implement without any major difficulties )

Hope this helped,

Angel
O:]
"craig" <[e-mail]> wrote in message
news:eD**************@TK2MSFTNGP14.phx.gbl...

I am working on my first .NET development project that involves custom
role-based security per the project requirements. This lead to a general
design issue this week that really caused us some concern. I have described the situation below because we are very curious to see what other, more
experienced, developers might suggest. The specific classes and fields are
used just to illustrate the concepts.

Our application uses role-based authorization security. Thus, we allow the
administrator to define roles to which the can define various permissions.
Permissions are defined relative to the various entities in the application, such as Users, Customers, Services, etc. One such role might be defined as
follows:

Role: Administrator
Permissions: Add User, Delete User, View User, Update User

Thus a user that is added to the Administrator role is granted permission to add, delete, view and update User entities. Once such a user is
authenticated, they will be authorized to perform those actions.

Now, consider that the User entity is abstracted by a User class, and
suppose that class has a Username field (among others). If an authenticated user indicated that they would like to edit the Username field of a system
User entity, the system first checks to see what role the authenticated user is in. If they are in an administrator role, they have the desired "Update
User" privilege described above, so the system instantiates a User object
and presents it to the user for editing. If, however, the user does not have this permission, their request is denied and the system does not instantiate an object.

So far, so good.

Now, suppose that another requirement is that ANY user that is logged into
the system must be granted permission to edit their own Username, regardless of what role they are in. In otherwords, the currently authenticated user
should always be granted the ability edit their own Username.

Now, here is where we ran into a problem. If the currently authenticated
user would like to update their Username field, but they are not in an
Administrator role, the system will not allow the User object to be
instantiated, so they will not be able to edit their Username as desired.
The object that provides the functionality needed is not accessible. As we
began to think about this situation, it was not clear how best to solve the problem and it lead to quite a bit of discussion. Naturally, we wondered if this is a common problem for which there is a consensus solution in the
developer community, or whether there was just something inherently wrong
with our role-based security model, our object model or our logic.

Here are some of the solutions we considered. I was hoping to hear what
others think.

1. Create a hidden "System" role which has all possible permissions. Add a
hidden "System" user to this role. Anytime such a situation occurs, log out the current user, log in the System user, perform the desired actions, then log the current user back in.

2. Create a different class, with different security logic, that can be used to expose the proper Username field to the currently authenticated user for editing. This would mean that a single field in the DB would be editable
through two different classes.

3. Add additionaly security logic to the User class such that it functions
as follows: If the authenticated user requesting a User object has Update
User permission, instantiate the requesed User object. If the authenticated user
does not have Update User permission, allow them the ability to instantiate only the User entity that corresponds to themselves, and expose only the
Username field for editing (make all other fields read-only).

Any thoughts on the problem in general or the possible solutions? As we
thought about this situation and its possible solutions, it raised a few
other questions:

1. Is it considered good design for two different classes to both provide
read/write access to the same field in the database?

2. Does the use of a "SuperUser" account as described constitute a breech of security?

3. Is it possible for a situation to arise in which the system might not be able to discern the security context of a request to instantiate an object? If so, would it be acceptable to use a parameter to communicate that
context?

Thanks!

[Angelos Karantzalis]

Actually, I've just found something that might make your life much easier,
and you can combine it with the previous post:

String id1 = "Bob";
String role1 = null;
PrincipalPermission PrincipalPerm1 = new PrincipalPermission(id1, role1);

String id2 = null;
String role2 = "Supervisor";
PrincipalPermission PrincipalPerm2 = new PrincipalPermission(id2, role2);

(PrincipalPerm1.Union(PrincipalPerm2)).Demand();

The following code will demand that the current IPrincipal of the system is
either user "Bob", or that the current IPrincipal is a "Supervisor" ... in
that manner, you can specify that the user accessing your data class is
either an Admin, or the user with the username specified by the class
instance itself - the only disadvantage is that upon construction of the
data object, you don't really know the username contained inside the
instance you're loading without doing a "select" from the DB ... but, almost
nothing is perfect, innit ? You get some small performance overhead, but
great ease-of-use codewise

O:]


"Angelos Karantzalis" <ak**********@agiltech.gr> wrote in message
news:#K**************@TK2MSFTNGP14.phx.gbl...

Perhaps a solution - although a slightly more complex one - would be to
explictly set permissions on the user object, rather than the role. What i
mean is, in Windows, any user can belong to groups and get some permission
based on this group membership. However, you can also explicitly set a
permission on the user to say, read a file.

I'd think that in the plumbing, upon instantiation of the User (or perhaps
Identity ) object, it is assigned permissions directly. So, I pseudo-code,
on the constructor of your data class there should be something like:

public myClass() {
new DataAccessPermission(DataAccessPermission.Read)Dem and();
// etc etc.
}

... that permission demanding code should also be present in the properties of your class.

What needs to be done for this though, is create your Permission sub-class es ( if an existing Permission class doesn't cover your requirements ), amd
possible implement some "plumbing" inside them to access the permission
you've set on the current user object. It's a bit tricky coding, but once
you've got it, you've a perfect case of re-use for almost any project
afterwards.

I think it's worth taking the time, actually I was planning to create that
sort of thing for the needs of my company as well in the near future,
because we're gonna re-use it in almost any project that requires security.
( This is actually a Java-driven approach, but I think the 2 frameworks are more-or-less equivalent in functionality, so it shouldn't be very difficult to implement without any major difficulties )

Hope this helped,

Angel
O:]
"craig" <[e-mail]> wrote in message
news:eD**************@TK2MSFTNGP14.phx.gbl...

I am working on my first .NET development project that involves custom
role-based security per the project requirements. This lead to a general
design issue this week that really caused us some concern. I have described

the situation below because we are very curious to see what other, more
experienced, developers might suggest. The specific classes and fields are used just to illustrate the concepts.

Our application uses role-based authorization security. Thus, we allow the administrator to define roles to which the can define various permissions. Permissions are defined relative to the various entities in the

application,

such as Users, Customers, Services, etc. One such role might be defined as follows:

Role: Administrator
Permissions: Add User, Delete User, View User, Update User

Thus a user that is added to the Administrator role is granted permission to

add, delete, view and update User entities. Once such a user is
authenticated, they will be authorized to perform those actions.

Now, consider that the User entity is abstracted by a User class, and
suppose that class has a Username field (among others). If an authenticated

user indicated that they would like to edit the Username field of a

system User entity, the system first checks to see what role the authenticated

user

is in. If they are in an administrator role, they have the desired "Update User" privilege described above, so the system instantiates a User object and presents it to the user for editing. If, however, the user does not

have

this permission, their request is denied and the system does not

instantiate

an object.

So far, so good.

Now, suppose that another requirement is that ANY user that is logged into the system must be granted permission to edit their own Username,

regardless

of what role they are in. In otherwords, the currently authenticated user should always be granted the ability edit their own Username.

Now, here is where we ran into a problem. If the currently authenticated
user would like to update their Username field, but they are not in an
Administrator role, the system will not allow the User object to be
instantiated, so they will not be able to edit their Username as desired. The object that provides the functionality needed is not accessible. As we began to think about this situation, it was not clear how best to solve

the

problem and it lead to quite a bit of discussion. Naturally, we wondered

if

this is a common problem for which there is a consensus solution in the
developer community, or whether there was just something inherently wrong with our role-based security model, our object model or our logic.

Here are some of the solutions we considered. I was hoping to hear what
others think.

1. Create a hidden "System" role which has all possible permissions. Add a hidden "System" user to this role. Anytime such a situation occurs, log

out

the current user, log in the System user, perform the desired actions,

then

log the current user back in.

2. Create a different class, with different security logic, that can be

used

to expose the proper Username field to the currently authenticated user

for

editing. This would mean that a single field in the DB would be editable
through two different classes.

3. Add additionaly security logic to the User class such that it functions as follows: If the authenticated user requesting a User object has Update User permission, instantiate the requesed User object. If the

authenticated

user
does not have Update User permission, allow them the ability to

instantiate

only the User entity that corresponds to themselves, and expose only the
Username field for editing (make all other fields read-only).

Any thoughts on the problem in general or the possible solutions? As we
thought about this situation and its possible solutions, it raised a few
other questions:

1. Is it considered good design for two different classes to both provide read/write access to the same field in the database?

2. Does the use of a "SuperUser" account as described constitute a

breech of

security?

3. Is it possible for a situation to arise in which the system might not

be

able to discern the security context of a request to instantiate an

object?

If so, would it be acceptable to use a parameter to communicate that
context?

Thanks!

[craig]

Thanks Angelos. I really appreciate your taking the time to respond.

I will spend some time to study your posts. Looks like some good input.

"Angelos Karantzalis" <ak**********@agiltech.gr> wrote in message
news:eW**************@TK2MSFTNGP10.phx.gbl...

Actually, I've just found something that might make your life much easier,
and you can combine it with the previous post:

String id1 = "Bob";
String role1 = null;
PrincipalPermission PrincipalPerm1 = new PrincipalPermission(id1, role1);

String id2 = null;
String role2 = "Supervisor";
PrincipalPermission PrincipalPerm2 = new PrincipalPermission(id2, role2);

(PrincipalPerm1.Union(PrincipalPerm2)).Demand();

The following code will demand that the current IPrincipal of the system
is
either user "Bob", or that the current IPrincipal is a "Supervisor" ... in
that manner, you can specify that the user accessing your data class is
either an Admin, or the user with the username specified by the class
instance itself - the only disadvantage is that upon construction of the
data object, you don't really know the username contained inside the
instance you're loading without doing a "select" from the DB ... but,
almost
nothing is perfect, innit ? You get some small performance overhead, but
great ease-of-use codewise

O:]


"Angelos Karantzalis" <ak**********@agiltech.gr> wrote in message
news:#K**************@TK2MSFTNGP14.phx.gbl...

Perhaps a solution - although a slightly more complex one - would be to
explictly set permissions on the user object, rather than the role. What
i
mean is, in Windows, any user can belong to groups and get some
permission
based on this group membership. However, you can also explicitly set a
permission on the user to say, read a file.

I'd think that in the plumbing, upon instantiation of the User (or
perhaps
Identity ) object, it is assigned permissions directly. So, I
pseudo-code,
on the constructor of your data class there should be something like:

public myClass() {
new DataAccessPermission(DataAccessPermission.Read)Dem and();
// etc etc.
}

... that permission demanding code should also be present in the

properties

of your class.

What needs to be done for this though, is create your Permission
sub-class

es

( if an existing Permission class doesn't cover your requirements ), amd
possible implement some "plumbing" inside them to access the permission
you've set on the current user object. It's a bit tricky coding, but once
you've got it, you've a perfect case of re-use for almost any project
afterwards.

I think it's worth taking the time, actually I was planning to create
that
sort of thing for the needs of my company as well in the near future,
because we're gonna re-use it in almost any project that requires

security.

( This is actually a Java-driven approach, but I think the 2 frameworks

are

more-or-less equivalent in functionality, so it shouldn't be very

difficult

to implement without any major difficulties )

Hope this helped,

Angel
O:]
"craig" <[e-mail]> wrote in message
news:eD**************@TK2MSFTNGP14.phx.gbl...

> I am working on my first .NET development project that involves custom
> role-based security per the project requirements. This lead to a
> general
> design issue this week that really caused us some concern. I have

described

> the situation below because we are very curious to see what other, more
> experienced, developers might suggest. The specific classes and fields are > used just to illustrate the concepts.
>
> Our application uses role-based authorization security. Thus, we allow the > administrator to define roles to which the can define various permissions. > Permissions are defined relative to the various entities in the

application,

> such as Users, Customers, Services, etc. One such role might be defined as > follows:
>
> Role: Administrator
> Permissions: Add User, Delete User, View User, Update User
>
> Thus a user that is added to the Administrator role is granted permission

to

> add, delete, view and update User entities. Once such a user is
> authenticated, they will be authorized to perform those actions.
>
> Now, consider that the User entity is abstracted by a User class, and
> suppose that class has a Username field (among others). If an

authenticated

> user indicated that they would like to edit the Username field of a

system > User entity, the system first checks to see what role the authenticated

user

> is in. If they are in an administrator role, they have the desired "Update > User" privilege described above, so the system instantiates a User object > and presents it to the user for editing. If, however, the user does not

have

> this permission, their request is denied and the system does not

instantiate

> an object.
>
> So far, so good.
>
> Now, suppose that another requirement is that ANY user that is logged into > the system must be granted permission to edit their own Username,

regardless

> of what role they are in. In otherwords, the currently authenticated user > should always be granted the ability edit their own Username.
>
> Now, here is where we ran into a problem. If the currently
> authenticated
> user would like to update their Username field, but they are not in an
> Administrator role, the system will not allow the User object to be
> instantiated, so they will not be able to edit their Username as desired. > The object that provides the functionality needed is not accessible. As we > began to think about this situation, it was not clear how best to solve

the

> problem and it lead to quite a bit of discussion. Naturally, we
> wondered

if

> this is a common problem for which there is a consensus solution in the
> developer community, or whether there was just something inherently wrong > with our role-based security model, our object model or our logic.
>
> Here are some of the solutions we considered. I was hoping to hear what
> others think.
>
> 1. Create a hidden "System" role which has all possible permissions.
> Add a > hidden "System" user to this role. Anytime such a situation occurs, log

out

> the current user, log in the System user, perform the desired actions,

then

> log the current user back in.
>
> 2. Create a different class, with different security logic, that can be

used

> to expose the proper Username field to the currently authenticated user

for

> editing. This would mean that a single field in the DB would be
> editable
> through two different classes.
>
> 3. Add additionaly security logic to the User class such that it functions > as follows: If the authenticated user requesting a User object has Update > User permission, instantiate the requesed User object. If the

authenticated

> user
> does not have Update User permission, allow them the ability to

instantiate

> only the User entity that corresponds to themselves, and expose only
> the
> Username field for editing (make all other fields read-only).
>
> Any thoughts on the problem in general or the possible solutions? As we
> thought about this situation and its possible solutions, it raised a
> few
> other questions:
>
> 1. Is it considered good design for two different classes to both provide > read/write access to the same field in the database?
>
> 2. Does the use of a "SuperUser" account as described constitute a

breech

of

> security?
>
> 3. Is it possible for a situation to arise in which the system might
> not

be

> able to discern the security context of a request to instantiate an

object?

> If so, would it be acceptable to use a parameter to communicate that
> context?
>
> Thanks!
>
>