[Vm-dev] FFI memory access error and pinned

[Esteban Lorenzano]

Hi,

Some days ago I submitted a small change at ThreadedFFIPlugin>>#ffiAddressOf: rcvr startingAt: byteOffset size: byteSize

to restore a security level: 

		"don't you dare to read from object memory (unless is pinned)!"
		(addr = 0 or: [(interpreterProxy isInMemory: addr) or: [(interpreterProxy isPinned: rcvr) not]])
			ifTrue:[^interpreterProxy primitiveFail].

if you remember, before (at the origin) that method was validating #isInMemory, but I took it out because it was making some test failing and because if found it incorrect. 
Then, after talk with Eliot (and after I took some vacations in the middle), he suggested that internal object should be pinned, so I submitted the change you can see.

Now, the problem is now the test FFIPluginTests>>#testPointers fails because it returns a structure and we try to access its fields. 

Problem is solved if I edit the tests to add a pin:

FFIPluginTests>>#testPointers
	"Test passing and returning of pointers to structs"
	| pt1 pt2 pt3 |
	pt1 := FFITestPoint4 new.
	pt1 x: 1. pt1 y: 2. pt1 z: 3. pt1 w: 4.
	pt2 := FFITestPoint4 new.
	pt2 x: 5. pt2 y: 6. pt2 z: 7. pt2 w: 8.
	pt3 := FFITestLibrary ffiTestPointers: pt1 with: pt2.
	pt3 getHandle pin. “HERE THE CHANGE"
	self assert: pt3 x = 6.
	self assert: pt3 y = 8.
	self assert: pt3 z = 10.
	self assert: pt3 w = 12.

I will submit the change, but now I wonder (again) if forcing a pin is correct. 
Of course, pinning it is safe because it prevents errors if object is moved… but well, I wonder :)

Esteban


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20151208/a2e02395/attachment.htm