Hai Chen

from collections import deque
from enum import Enum

from binaryninja.enums import (
    BranchType,
    InstructionTextTokenType,
)

# Support both package and standalone imports
try:
    from ..utils.hashes import calculate_md5, calculate_sha256
except ImportError:
    # Fallback to absolute imports (for multiprocessing spawned processes)
    from redb.extractors.decompiler.bninja.utils.hashes import calculate_md5, calculate_sha256


class CFGAnalysis:
    def __init__(self, function):
        self.function = function

    def determine_block_type(self, block) -> str:
        """Determine the type of a basic block."""
        # Check if it's a thunk function (usually just a jump or call)
        if len(block.disassembly_text) <= 2 and any(
            "jmp" in line.tokens[0].text.lower() for line in block.disassembly_text
        ):
            return "THUNK"

        # Check if it contains only data (no valid instructions)
        if all(not line.tokens for line in block.disassembly_text):
            return "DATA"

        # Default to code
        return "CODE"

    def extract_cyclomatic_complexity(self):
        """
        Cyclomatic complexity (McCabe’s metric) measures the number of linearly independent paths
        through a function’s control flow graph (CFG).
        The standard formula is:

            M = E - N + 2

        where:
            - E = number of edges in the CFG
            - N = number of nodes (basic blocks)
            - 2 accounts for the entry and exit nodes of a single connected graph
        """
        if self.function is None:
            return 0

        # number of basic blocks
        num_blocks = len(self.function.basic_blocks)
        # number of edges in the graph
        num_edges = sum(
            len(basic_block.outgoing_edges)
            for basic_block in self.function.basic_blocks
        )
        return num_edges - num_blocks + 2

    def extract_function_cfg(self):
        """Extract information about a function CFG and return it as a dictionary."""

        function = self.function
        function_data = {
            "function_address": self.function.start,
            "blocks": [],
            "measures": {
                "cyclomatic_complexity": self.extract_cyclomatic_complexity(),
            },
        }

        if self.function is None:
            return function_data

        # Get the map of the depth associated to every block
        depths = self.get_map_depth()

        # Get the map of the positions associated to every block
        id_maps = self.get_block_id_map()

        # Extract block data with graph structure information
        for block in function.basic_blocks:
            # dominators per every block translated
            dominators = sorted(self.extract_dominators(block, id_maps))

            # post dominators
            post_dominators = sorted(self.extract_post_dominators(block, id_maps))

            # Build block instructions string
            block_instructions = "\n".join(str(line) for line in block.disassembly_text)

            # Determine block type
            block_type = self.determine_block_type(block)

            # Extract successors directly from basic block
            successor_blocks = [edge.target.start for edge in block.outgoing_edges]
            # We ensure a canonical order and we sort the edges
            successor_blocks.sort()

            # Extract predecessors directly from basic block
            predecessor_blocks = [edge.source.start for edge in block.incoming_edges]
            # We ensure a canonical order and we sort the edges
            predecessor_blocks.sort()

            # Determine branch type from outgoing edges
            branch_type = self.determine_branch_type(block)

            instructions_count = len(block.disassembly_text)

            # Create block record
            block_json = {
                "function_address": self.function.start,
                "block_start_address": block.start,
                "block_end_address": block.end,
                "block_size": block.end - block.start,
                "instructions_count": instructions_count,
                "block_instructions_hash": calculate_sha256(block_instructions),
                "predecessor_blocks": predecessor_blocks,
                "successor_blocks": successor_blocks,
                "depth": depths[block.start],
                "position": id_maps[block.start],
                "branch_type": branch_type,
                "block_type": block_type,
                "flags": self.extract_block_flags(block),
                "dominators": dominators,
                "post_dominators": post_dominators,
            }
            function_data["blocks"].append(block_json)

        return function_data

    def extract_dominators(self, bb, id_maps):
        """Extract the dominators normalized"""
        dom_idx = [id_maps[d.start] for d in bb.dominators]
        return dom_idx

    def extract_post_dominators(self, bb, id_maps):
        """Extract the post-dominators normalized"""
        post_dom_idx = [id_maps[d.start] for d in bb.post_dominators]
        return post_dom_idx

    def determine_branch_type(self, block):
        """
        Determine the type of branch at the end of a basic block.
        This combines edge type information with instruction analysis.
        """
        # If no outgoing edges, it might be a return or terminal block
        if not block.outgoing_edges:
            # Check if the last instruction is a return
            for line in reversed(list(block.disassembly_text)):
                if line.tokens and any(
                    token.text.lower() in ["ret", "retn"] for token in line.tokens
                ):
                    return "RETURN"
            return "UNKNOWN"

        # Collect branch types from all outgoing edges
        branch_types = []
        for edge in block.outgoing_edges:
            edge_type = edge.type
            # Map edge type to our branch type enum
            if isinstance(edge_type, str):
                if edge_type == "IndirectCall":
                    branch_types.append("CALL")
                else:
                    branch_types.append("UNKNOWN")
            else:
                # Use our mapping for integer/enum values
                type_mapping = {
                    BranchType.UnconditionalBranch: "DIRECT",
                    BranchType.FalseBranch: "CONDITIONAL",
                    BranchType.TrueBranch: "CONDITIONAL",
                    BranchType.CallDestination: "CALL",
                    BranchType.FunctionReturn: "RETURN",
                    BranchType.SystemCall: "CALL",
                    BranchType.IndirectBranch: "INDIRECT",
                    BranchType.ExceptionBranch: "UNKNOWN",
                    BranchType.UnresolvedBranch: "UNKNOWN",
                    BranchType.UserDefinedBranch: "UNKNOWN",
                }
                branch_types.append(type_mapping.get(edge_type, "UNKNOWN"))

        # Determine overall branch type (prioritize CALL > RETURN > CONDITIONAL > DIRECT)
        if "CALL" in branch_types:
            return "CALL"
        elif "RETURN" in branch_types:
            return "RETURN"
        elif "CONDITIONAL" in branch_types:
            return "CONDITIONAL"
        elif "DIRECT" in branch_types:
            return "DIRECT"
        elif len(block.outgoing_edges) == 1:
            return "FALLTHROUGH"

        # If edge analysis was inconclusive, fall back to instruction analysis
        last_instr = None
        for line in reversed(list(block.disassembly_text)):
            if line.tokens:
                last_instr = line
                break

        if last_instr:
            mnemonic = None
            for token in last_instr.tokens:
                if token.type == InstructionTextTokenType.InstructionToken:
                    mnemonic = token.text.lower()
                    break

            if mnemonic:
                if mnemonic == "call":
                    return "CALL"
                elif mnemonic == "jmp":
                    return "DIRECT"
                elif mnemonic.startswith("j") and mnemonic != "jmp":
                    return "CONDITIONAL"
                elif mnemonic in ["ret", "retn"]:
                    return "RETURN"

        return "UNKNOWN"

    def get_map_depth(self):
        """
        Run a BFS on the basic blocks of the function to assign a depth to every block
        """

        depths = {}
        entry = self.function.get_basic_block_at(self.function.start)

        ### Simple BFS
        q = deque()
        q.append(entry)
        depths[entry.start] = 0

        while q:
            b = q.popleft()
            b_depth = depths[b.start]
            for edge in b.outgoing_edges:
                tgt = edge.target

                if tgt is None:
                    continue

                if tgt.start not in depths:
                    depths[tgt.start] = b_depth + 1
                    q.append(tgt)

        return depths

    def get_block_id_map(self):
        """
        Assign a unique, sequential ID to each basic block of the function using a BFS starting from the entry block.
        """

        id_map = {}
        entry = self.function.get_basic_block_at(self.function.start)

        q = deque()
        q.append(entry)

        current_id = 0
        id_map[entry.start] = current_id

        while q:
            b = q.popleft()
            for edge in b.outgoing_edges:
                tgt = edge.target

                if tgt is None:
                    continue

                if tgt.start not in id_map:
                    current_id += 1
                    id_map[tgt.start] = current_id
                    q.append(tgt)

        return id_map

    def extract_block_flags(self, block):
        """
        Get the flags for every basic block. Currently, we implemented these heuristics:
            - if a basic block is the entry node for a function
            - if a basic block is the exit block for a function
            - if a basic block is part of a natural loop
        """
        flags = []

        if block.start == self.function.start:
            flags.append(BlockFlags.EntryBlock.value)

        if any(edge.type == BranchType.FunctionReturn for edge in block.outgoing_edges):
            flags.append(BlockFlags.ExitBlock.value)

        # if this block is in its dominance frontier, then it's part of a natural loop
        if block in block.dominance_frontier:
            flags.append(BlockFlags.LoopBlock.value)

        return flags


class BlockFlags(Enum):
    # generally, the basic block identifying the entry point of the function
    EntryBlock = "EntryBlock"
    # any basic blocks that makes the control flow exiting from the current function
    ExitBlock = "ExitBlock"
    # any block is in a natural loop if it is in its own dominance frontier
    LoopBlock = "LoopBlock"


class BlockType(Enum):
    THUNK = "THUNK"
    DATA = "DATA"
    PADDING = "PADDING"
    CODE = "CODE"