The LIO monad is a wrapper around IO that keeps track of a current label and current clearance. Safe code cannot execute arbitrary IO actions from the LIO monad. However, trusted runtime functions can use ioTCB to perform IO actions (which they should only do after appropriately checking labels).

Synonym for monad in which LIO is the base monad.

Internal state of an LIO computation.

Given an LIO computation and some initial state, return an IO action which, when executed, will perform the IFC-safe LIO computation.

Because untrusted code cannot execute IO computations, this function should only be useful within trusted code. No harm is done from exposing the evalLIO symbol to untrusted code. (In general, untrusted code is free to produce IO computations, but it cannot execute them.)

Unlike runLIO, this function throws an exception if the underlying LIO action terminates with an exception.

Execute an LIO action, returning its result and the final label state as a pair. Note that it returns a pair whether or not the LIO action throws an exception. Forcing the result value will re-throw the exception, but the label state will always be valid.

See also evalLIO.

Returns the value of the thread's current label.

Raises the current label to the provided label, which must be between the current label and clearance. See taint.

If the current label is oldLabel and the current clearance is clearance, this function allows code to raise the current label to any value newLabel such that canFlowToP priv oldLabel newLabel && canFlowTo newLabel clearance. (Note the privilege argument affects the label check, not the clearance check; call setClearanceP first to raise the clearance.)

Returns the thread's current clearance.

Lowers the current clearance. The new clerance must be between the current label and previous current clerance. One cannot raise the current label or create object with labels higher than the current clearance.

Raises the current clearance (undoing the effects of setClearance) by exercising privileges. If the current label is l and current clearance is c, then setClearanceP p cnew succeeds only if the new clearance is can flow to the current clearance (modulo privileges), i.e., canFlowToP p cnew c == True. Additionally, the current label must flow to the new clearance, i.e., l `canFlowTo` cnew == True.

Since LIO guards that are used when reading/writing data (e.g., guardAllocP) do not use privileges when comparing labels with the current clearance, code must always raise the current clearance, to read/write data above the current clearance.

Runs an LIO action and re-sets the current clearance to its previous value once the action returns. In particular, if the action lowers the current clearance, the clearance will be restored upon return.

Note that scopeClearance always restores the clearance. If that causes the clearance to drop below the current label, a ClearanceViolation exception is thrown. That exception can only be caught outside a second scopeClearance that restores the clearance to higher than the current label.

Temporarily lowers the clearance for a computation, then restores it. Equivalent to:

withClearance c lio = scopeClearance $ setClearance c >> lio

Note that if the computation inside withClearance acquires any Privs, it may still be able to raise its clearance above the supplied argument using setClearanceP.

A variant of withClearance that takes privileges. Equivalent to:

withClearanceP p c lio = scopeClearance $ setClearanceP p c >> lio

Parent of all label-related exceptions.

Main error type thrown by label failures in the LIO monad.

Error indicating insufficient privileges (independent of the current label). This exception is thrown by delegate, and should also be thrown by gates that receive insufficient privilege descriptions (see LIO.Delegate).

Ensures the label argument is between the current IO label and current IO clearance. Use this function in code that allocates objects--untrusted code shouldn't be able to create an object labeled l unless guardAlloc l does not throw an exception. Similarly use this guard in any code that writes to an object labeled l for which the write has no observable side-effects.

Like guardAlloc, but takes a privilege argument to be more permissive. Note: privileges are only used when checking that the current label can flow to the target label; guardAllocP still always throws an exception when the target label is higher than the current clearance.

Use taint l in trusted code before observing an object labeled l. This will raise the current label to a value l' such that l `canFlowTo` l', or throw a LabelError exception if l' would have to be higher than the current clearance.

Like taint, but use privileges to reduce the amount of taint required. Note that taintP will never lower the current label. It simply uses privileges to avoid raising the label as high as taint would raise it.

Use guardWrite l in any (trusted) code before modifying an object labeled l, for which the modification can be observed, i.e., the write implies a read.

The implementation of guardWrite is straight forward:

guardWrite l = guardAlloc l >> taint l

The guardAlloc ensures that we can write(-only) to the object labeled l, i.e., the current label `canFlowTo` l (and l `canFlowTo` the current clearance). If this check succeeds then we raise the current label with taint to reflect the fact that this is a also a read effect. Note that if the write-only guard succeeds, the taint will always suceed (we're simply raising the current label to a label that is below the clearance).

Like guardWrite, but takes a privilege argument to be more permissive.