Issue #522 has been updated by Nico Huber.
I agree with Werner that these regions should never overflow in the first place.
Looking into our smihandlers, I believe the problem actually starts in smm_points_to_smram(). Which doesn't validate the inputs. Maybe we should introduce a function to create a region from `offset` and `size`? and use that instead of manually filling the struct. That being said, I wonder why `struct region` isn't opaque in the first place. Maybe we should change that in the long run.
---------------------------------------- Bug #522: `region_overlap()` function might not work as expected due to an integer overflow in `region_end()` function. https://ticket.coreboot.org/issues/522#change-1728
* Author: Vadim Zaliva * Status: New * Priority: Normal * Category: coreboot common code * Target version: none * Start date: 2023-12-27 * Affected versions: master ---------------------------------------- `region_overlap()` function checks whether or not two memory regions overlap. Memory regions are represented as a region struct that contains the region's offset and size. This function then relies on `region_end()` function to compute the end of the region. `region_end()` function is susceptible to an integer overflow, which might result in the incorrect behaviour of `region_overlap()` function.
An example of inputs that lead to wrong behaviour: ``` struct region r1 = {SIZE_MAX - 10, 20}; struct region r2 = {SIZE_MAX - 20, 15}; ``` It returns 0, but since the regions actually overlap, it should return 1.
`region_overlap()` function is used in `smm_region_overlaps_handler()` function, which is itself used in SMI handlers to validate address values that come from an untrusted environment. This is necessary to prevent security vulnerabilities such as described in [BARing the System by Yuriy Bulygin, Oleksandr Bazhaniuk et al.](https://www.c7zero.info/stuff/REConBrussels2017_BARing_the_system.pdf)
We do not have an example of an exploit based on this incorrect behaviour and are aware of the existence of one. However, theoretically, this could lead to security vulnerabilities.
This bug was found during an ongoing [Coreboot Formal Verification Project](https://zaliva.org/UCSC-Twisted-Presentation-20231211.pdf), which aims to prove some important security properties of the coreboot’s SMI handler for the Gemini Lake/Octopus platform using Coq proof assistant and Verified Software Toolchain framework.
---Files-------------------------------- diff.txt (930 Bytes)