Merge pull request #2511 from xusheng6/fix_llil_access

binja: retrieve the LLIL instruction itself without requesting the entire IL function

CHANGELOG.md

@@ -31,6 +31,7 @@
 - ghidra: fix saving of base address @mr-tz
 - binja: support loading raw x86/x86_64 shellcode #2489 @xusheng6
 - binja: fix crash when the IL of certain functions are not available. #2249 @xusheng6
+- binja: major performance improvement on the binja extractor. #1414 @xusheng6
 
 ### capa Explorer Web
 

capa/features/extractors/binja/function.py

@@ -7,12 +7,13 @@
 # See the License for the specific language governing permissions and limitations under the License.
 from typing import Iterator
 
-from binaryninja import Function, BinaryView, SymbolType, ILException, RegisterValueType, LowLevelILOperation
+from binaryninja import Function, BinaryView, SymbolType, LowLevelILOperation
 
 from capa.features.file import FunctionName
 from capa.features.common import Feature, Characteristic
 from capa.features.address import Address, AbsoluteVirtualAddress
 from capa.features.extractors import loops
+from capa.features.extractors.binja.helpers import get_llil_instr_at_addr
 from capa.features.extractors.base_extractor import FunctionHandle
 
 
@@ -24,14 +25,7 @@ def extract_function_calls_to(fh: FunctionHandle):
         # Everything that is a code reference to the current function is considered a caller, which actually includes
         # many other references that are NOT a caller. For example, an instruction `push function_start` will also be
         # considered a caller to the function
-        llil = None
-        try:
-            # Temporary fix for https://github.com/Vector35/binaryninja-api/issues/6020. Since `.llil` can throw an
-            # exception rather than returning None
-            llil = caller.llil
-        except ILException:
-            continue
-
+        llil = get_llil_instr_at_addr(func.view, caller.address)
         if (llil is None) or llil.operation not in [
             LowLevelILOperation.LLIL_CALL,
             LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
@@ -40,14 +34,13 @@ def extract_function_calls_to(fh: FunctionHandle):
         ]:
             continue
 
-        if llil.dest.value.type not in [
-            RegisterValueType.ImportedAddressValue,
-            RegisterValueType.ConstantValue,
-            RegisterValueType.ConstantPointerValue,
+        if llil.dest.operation not in [
+            LowLevelILOperation.LLIL_CONST,
+            LowLevelILOperation.LLIL_CONST_PTR,
         ]:
             continue
 
-        address = llil.dest.value.value
+        address = llil.dest.constant
         if address != func.start:
             continue
 

capa/features/extractors/binja/helpers.py

@@ -6,10 +6,10 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 import re
-from typing import Callable
+from typing import Callable, Optional
 from dataclasses import dataclass
 
-from binaryninja import BinaryView, LowLevelILInstruction
+from binaryninja import BinaryView, LowLevelILFunction, LowLevelILInstruction
 from binaryninja.architecture import InstructionTextToken
 
 
@@ -67,3 +67,13 @@ def read_c_string(bv: BinaryView, offset: int, max_len: int) -> str:
         s.append(chr(c))
 
     return "".join(s)
+
+
+def get_llil_instr_at_addr(bv: BinaryView, addr: int) -> Optional[LowLevelILInstruction]:
+    arch = bv.arch
+    buffer = bv.read(addr, arch.max_instr_length)
+    llil = LowLevelILFunction(arch=arch)
+    llil.current_address = addr
+    if arch.get_instruction_low_level_il(buffer, addr, llil) == 0:
+        return None
+    return llil[0]

capa/features/extractors/binja/insn.py

@@ -13,7 +13,6 @@
     BinaryView,
     ILRegister,
     SymbolType,
-    ILException,
     BinaryReader,
     RegisterValueType,
     LowLevelILOperation,
@@ -24,7 +23,7 @@
 from capa.features.insn import API, MAX_STRUCTURE_SIZE, Number, Offset, Mnemonic, OperandNumber, OperandOffset
 from capa.features.common import MAX_BYTES_FEATURE_SIZE, Bytes, String, Feature, Characteristic
 from capa.features.address import Address, AbsoluteVirtualAddress
-from capa.features.extractors.binja.helpers import DisassemblyInstruction, visit_llil_exprs
+from capa.features.extractors.binja.helpers import DisassemblyInstruction, visit_llil_exprs, get_llil_instr_at_addr
 from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
 
 # security cookie checks may perform non-zeroing XORs, these are expected within a certain
@@ -37,40 +36,23 @@
 # 2. The function must only make one call/jump to another address
 # If the function being checked is a stub function, returns the target address. Otherwise, return None.
 def is_stub_function(bv: BinaryView, addr: int) -> Optional[int]:
-    funcs = bv.get_functions_at(addr)
-    for func in funcs:
-        if len(func.basic_blocks) != 1:
-            continue
-
-        call_count = 0
-        call_target = None
-        try:
-            llil = func.llil
-        except ILException:
-            return None
+    llil = get_llil_instr_at_addr(bv, addr)
+    if llil is None or llil.operation not in [
+        LowLevelILOperation.LLIL_CALL,
+        LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
+        LowLevelILOperation.LLIL_JUMP,
+        LowLevelILOperation.LLIL_TAILCALL,
+    ]:
+        return None
 
-        if llil is None:
-            continue
+    if llil.dest.value.type not in [
+        RegisterValueType.ImportedAddressValue,
+        RegisterValueType.ConstantValue,
+        RegisterValueType.ConstantPointerValue,
+    ]:
+        return None
 
-        for il in llil.instructions:
-            if il.operation in [
-                LowLevelILOperation.LLIL_CALL,
-                LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
-                LowLevelILOperation.LLIL_JUMP,
-                LowLevelILOperation.LLIL_TAILCALL,
-            ]:
-                call_count += 1
-                if il.dest.value.type in [
-                    RegisterValueType.ImportedAddressValue,
-                    RegisterValueType.ConstantValue,
-                    RegisterValueType.ConstantPointerValue,
-                ]:
-                    call_target = il.dest.value.value
-
-        if call_count == 1 and call_target is not None:
-            return call_target
-
-    return None
+    return llil.dest.value.value
 
 
 def extract_insn_api_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[tuple[Feature, Address]]: