more ELF OS detection techniques

CHANGELOG.md

@@ -95,6 +95,7 @@ Also a big thanks to the other contributors: @aaronatp, @Aayush-Goel-04, @bkojus
 - binja: use `binaryninja.load` to open files @xusheng6
 - binja: bump binja version to 3.5 #1789 @xusheng6
 - elf: better detect ELF OS via GCC .ident directives #1928 @williballenthin
+- elf: better detect ELF OS via Android dependencies #1947 @williballenthin
 - fix setuptools package discovery #1886 @gmacon @mr-tz
 
 ### capa explorer IDA Pro plugin

capa/features/extractors/elf.py

@@ -866,6 +866,8 @@ def guess_os_from_ident_directive(elf: ELF) -> Optional[OS]:
             return OS.LINUX
         elif "Red Hat" in comment:
             return OS.LINUX
+        elif "Android" in comment:
+            return OS.ANDROID
 
     return None
 
@@ -921,6 +923,8 @@ def guess_os_from_needed_dependencies(elf: ELF) -> Optional[OS]:
             return OS.HURD
         if needed.startswith("libandroid.so"):
             return OS.ANDROID
+        if needed.startswith("liblog.so"):
+            return OS.ANDROID
 
     return None
 
@@ -1023,10 +1027,6 @@ def detect_elf_os(f) -> str:
     if osabi_guess:
         ret = osabi_guess
 
-    elif ident_guess:
-        # we don't trust this too much due to non-cross-compilation assumptions
-        ret = ident_guess
-
     elif ph_notes_guess:
         ret = ph_notes_guess
 
@@ -1045,6 +1045,11 @@ def detect_elf_os(f) -> str:
     elif symtab_guess:
         ret = symtab_guess
 
+    elif ident_guess:
+        # at the bottom because we don't trust this too much
+        # due to potential for bugs with cross-compilation.
+        ret = ident_guess
+
     return ret.value if ret is not None else "unknown"