[Mai-Ooizumi]

Summary

Fixes a bug in PolyHook2's Linux PLH::MemAccessor::mem_protect implementation where memory ranges that start at a non-page-aligned address and cross a page boundary fail to update protection flags on the trailing page.

Root Cause

Previously, mem_protect rounded dest down to the nearest page boundary (aligned_dest), but rounded size up independently (MEMORY_ROUND_UP(size, pageSize)):

uint64_t aligned_dest = MEMORY_ROUND(dest, PLH::getPageSize());
uint64_t aligned_size = MEMORY_ROUND_UP(size, PLH::getPageSize());

When dest has a large intra-page offset (e.g., pageSize - 16) and size is small (e.g., 32 bytes), [dest, dest + size) spans two pages, whereas MEMORY_ROUND_UP(size, pageSize) only equals 1 * pageSize. As a result, mprotect only updates the first page, leaving the second page of a trampoline or patched region without the requested protection (e.g., non-executable, causing a SIGSEGV when executed).

Changes

• patches/polyhook.patch (sources/MemAccessor.cpp): Include the intra-page offset (dest - aligned_dest) when rounding up the protection length, while preserving aligned_size = 0 when size == 0 (used for querying current page protection):

  uint64_t aligned_size = size == 0 ? 0 : MEMORY_ROUND_UP(size + (dest - aligned_dest), PLH::getPageSize());

• patches/polyhook.patch (UnitTests/linux/TestMemProtector.cpp): Add unit tests covering ranges that cross page boundaries, end exactly at a page boundary, start on a page boundary, and have zero length at an unaligned address, as well as verifying restoration of original protections.
• CMakeLists.txt: Wire up CTest (polyhook_mem_protect_tests) on Linux when BUILD_TESTING is enabled.

[qubka]

Thanks @Mai-Ooizumi - clean fix, well explained.