From 9a8faad50f56d748fcd1498c170797c67bb53217 Mon Sep 17 00:00:00 2001 From: Mark Rowe Date: Wed, 22 Oct 2025 19:52:19 -0700 Subject: [KernelCache] Set segment permissions based on how XNU initially maps them XNU maps kernel cache segments in with different permissions than the load commands indicate. For instance, `__DATA_CONST` is initially mapped as read-write before later being re-mapped as read-only. Treating it as read-only results in analysis falsely assuming that global variables cannot change. To work around this we maintain a mapping from segment name to initial permissions (i.e., most lax permissions) and favor them over permissions derived from the segment load command. Section semantics are also derived from the segment's permissions when the segment is present in the mapping. The mapping is based on the initial permissions established by `arm_vm_prot_init` within the XNU source. --- view/kernelcache/core/KernelCache.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'view/kernelcache/core/KernelCache.cpp') diff --git a/view/kernelcache/core/KernelCache.cpp b/view/kernelcache/core/KernelCache.cpp index 3750e795..ef5fd8a5 100644 --- a/view/kernelcache/core/KernelCache.cpp +++ b/view/kernelcache/core/KernelCache.cpp @@ -77,7 +77,7 @@ bool KernelCache::ProcessEntryImage(Ref bv, const std::string& path, // Associate this region with this image, this makes it easier to identify what image owns this region. sectionRegion.imageStart = image.headerFileAddress; - uint32_t flags = SegmentFlagsFromMachOProtections(segment.initprot, segment.maxprot); + uint32_t flags = SegmentFlagsForSegment(segment); // if we're positive we have an entry point for some reason, force the segment // executable. this helps with kernel images. for (const auto& entryPoint : imageHeader->m_entryPoints) -- cgit v1.3.1