fixup! [HWASan] Fix symbol indexing

Author: Stefan Bossbaly

compiler-rt/lib/hwasan/scripts/hwasan_symbolize

@@ -37,12 +37,12 @@ if sys.version_info.major < 3:
 Ehdr_size = 64
 e_shnum_offset = 60
 e_shoff_offset = 40
-
+e_shstrndx_offset = 62
 Shdr_size = 64
+sh_name_offset = 0
 sh_type_offset = 4
 sh_offset_offset = 24
 sh_size_offset = 32
-SHT_SYMTAB = 2
 SHT_NOTE = 7
 
 Nhdr_size = 12
@@ -63,18 +63,32 @@ def handle_Nhdr(mv, sh_size):
     offset += Nhdr_size + align_up(n_namesz, 4) + align_up(n_descsz, 4)
   return None
 
+def handle_shstrtab(mv, e_shoff):
+  e_shstrndx, = struct.unpack_from('<H', buffer=mv, offset=e_shstrndx_offset)
+  
+  start_shstrndx = e_shoff + e_shstrndx * Shdr_size
+  shstrndx_sh = mv[start_shstrndx: start_shstrndx + Shdr_size]
+  _, shstrndx_sh_offset, shstrndx_sh_size = handle_Shdr(shstrndx_sh)
+  return mv[shstrndx_sh_offset:shstrndx_sh_offset + shstrndx_sh_size]
+
+def shstrtab_sh_name(mv):
+  name = ""
+  for byte in mv:
+    char = chr(byte)
+    if char == '\x00':
+      break
+    name += char
+  return name
+
 def unpack_sh_type(mv):
   sh_type, = struct.unpack_from('<I', buffer=mv, offset=sh_type_offset)
   return sh_type
 
 def handle_Shdr(mv):
-  sh_type = unpack_sh_type(mv)
-  # Sanity check
-  if sh_type != SHT_NOTE:
-    return None, None
+  name_offset, = struct.unpack_from('<I', buffer=mv, offset=sh_name_offset)
   sh_offset, = struct.unpack_from('<Q', buffer=mv, offset=sh_offset_offset)
   sh_size, = struct.unpack_from('<Q', buffer=mv, offset=sh_size_offset)
-  return sh_offset, sh_size
+  return name_offset, sh_offset, sh_size
 
 def handle_elf(mv):
   # \x02 is ELFCLASS64, \x01 is ELFDATA2LSB. HWASan currently only works on
@@ -86,23 +100,31 @@ def handle_elf(mv):
   bid = None
   e_shnum, = struct.unpack_from('<H', buffer=mv, offset=e_shnum_offset)
   e_shoff, = struct.unpack_from('<Q', buffer=mv, offset=e_shoff_offset)
+
+  # Section where all the section header names are stored
+  shstr = handle_shstrtab(mv, e_shoff)
+
   for i in range(0, e_shnum):
     start = e_shoff + i * Shdr_size
     sh = mv[start: start + Shdr_size]
+    sh_name_offset, sh_offset, sh_size = handle_Shdr(sh)
+    sh_name = shstrtab_sh_name(shstr[sh_name_offset:])
     sh_type = unpack_sh_type(sh)
 
-    if sh_type == SHT_SYMTAB:
+    if sh_name == ".debug_info":
       found_symbols = True
-    elif sh_type == SHT_NOTE:
-      sh_offset, sh_size = handle_Shdr(sh)
+    if sh_type == SHT_NOTE:
       if sh_offset is None:
         continue
       note_hdr = mv[sh_offset: sh_offset + sh_size]
       result = handle_Nhdr(note_hdr, sh_size)
       if result is not None:
         bid = result
 
-  return (found_symbols, bid)
+  if found_symbols:
+    return bid
+  else:
+    return None
 
 def read_elf(filename):
   with open(filename, "r") as fd:
@@ -284,13 +306,13 @@ class Symbolizer:
         for fn in fnames:
           filename = os.path.join(dname, fn)
           try:
-            found_symbols, bid = read_elf(filename)
+            bid = read_elf(filename)
           except FileNotFoundError:
             continue
           except Exception as e:
             print("Failed to parse {}: {}".format(filename, e), file=sys.stderr)
             continue
-          if found_symbols and bid is not None:
+          if bid is not None:
             if bid in self.__index:
               print("Duplicate build ID {} for {} and {}".format(bid, self.__index[bid], filename), file=sys.stderr)
               sys.exit(1)