- Side load SetupAPI.dll, as this is the DLL that was causing the CfgMgr32.dll local load.
This reverts part of 622e60659c since we no longer have to hook into CfgMgr32.dll directly.
- Also set the redefinition of DECLSPEC_IMPORT, which we need for MinGW32 x86, in the global AM_CFLAGS
of configure.ac, so that we no longer have to worry about forgetting to do it in a source and experience
crashes on 32-bit as a result (See 965759f58a).
- Also delay-load crypt32.dll while we're at it.
- Also add provision for enabling /DEPENDENTLOADFLAG:0x800 on MinGW, by leaving a properly crafted entry
in the .rdata section that can then be used with the loadcfg.py Python script.
- Sadly, per https://github.com/pbatard/rufus/issues/2701#issuecomment-2874788564 and subsequent comment,
having DependentLoadFlags set to LOAD_LIBRARY_SEARCH_SYSTEM32 is still not enough to take care of side
loading issues, as, ever since the introduction of wimlib support, we are seeing CRYPTBASE.DLL being
side-loaded in MinGW, and, even with crypt32.dll being delay-loaded there is literally *nothing* we can
do about it!
- The end result of all the above is that we will have no choice but ditch MinGW for release executables
as it's just impossible to properly take care of side-loading vulnerabilities with MinGW (and Microsoft
are REALLY not helping with this whole mess either, when they don't even use LOAD_LIBRARY_SEARCH_SYSTEM32
for Windows' system DLLs).
- In preparation for this, we add UPX compression to the x86_64 and x86_32 MSVC executables.
- Finally, we also fix one last Coverity warning in xml.c and remove duplicates in .vcxproj for ARM64.
Pete Batard
4adfa4f37e