Correctly parse ELF for musllinux on Big Endian (pypa#538)

packaging/_musllinux.py

@@ -39,9 +39,11 @@ def _parse_ld_musl_from_elf(f: IO[bytes]) -> Optional[str]:
         # p_fmt: Format for section header.
         # p_idx: Indexes to find p_type, p_offset, and p_filesz.
         e_fmt, p_fmt, p_idx = {
-            1: ("IIIIHHH", "IIIIIIII", (0, 1, 4)),  # 32-bit.
-            2: ("QQQIHHH", "IIQQQQQQ", (0, 2, 5)),  # 64-bit.
-        }[ident[4]]
+            (1, 1): ("<IIIIHHH", "<IIIIIIII", (0, 1, 4)),  # 32-bit LSB.
+            (1, 2): (">IIIIHHH", ">IIIIIIII", (0, 1, 4)),  # 32-bit MSB.
+            (2, 1): ("<QQQIHHH", "<IIQQQQQQ", (0, 2, 5)),  # 64-bit LSB.
+            (2, 2): (">QQQIHHH", ">IIQQQQQQ", (0, 2, 5)),  # 64-bit MSB.
+        }[(ident[4], ident[5])]
     except KeyError:
         return None
     else:

tests/test_musllinux.py

@@ -101,14 +101,15 @@ def test_parse_ld_musl_from_elf_no_interpreter_section():
     with BIN_MUSL_X86_64.open("rb") as f:
         data = f.read()
 
-    # Change all sections to *not* PT_INTERP.
-    unpacked = struct.unpack("16BHHIQQQIHHH", data[:58])
+    # Change all sections to *not* PT_INTERP. We are explicitly using LSB rules
+    # because the binaries are in LSB.
+    unpacked = struct.unpack("<16BHHIQQQIHHH", data[:58])
     *_, e_phoff, _, _, _, e_phentsize, e_phnum = unpacked
     for i in range(e_phnum + 1):
         sb = e_phoff + e_phentsize * i
         se = sb + 56
-        section = struct.unpack("IIQQQQQQ", data[sb:se])
-        data = data[:sb] + struct.pack("IIQQQQQQ", 0, *section[1:]) + data[se:]
+        section = struct.unpack("<IIQQQQQQ", data[sb:se])
+        data = data[:sb] + struct.pack("<IIQQQQQQ", 0, *section[1:]) + data[se:]
 
     assert _parse_ld_musl_from_elf(io.BytesIO(data)) is None