angr 9.2.103__py3-none-macosx_11_0_arm64.whl

angr/analyses/vfg.py

@@ -0,0 +1,1945 @@
+from typing import TYPE_CHECKING, Any, DefaultDict, Optional
+from collections.abc import Callable, Generator
+import logging
+from collections import defaultdict
+
+import archinfo
+from archinfo.arch_arm import is_arm_arch
+import claripy
+from claripy.vsa.strided_interval import StridedInterval
+import networkx
+
+from angr.utils.graph import GraphUtils
+from angr.analyses import ForwardAnalysis
+from .cfg.cfg_job_base import BlockID, FunctionKey, CFGJobBase
+from .. import sim_options
+from ..engines.procedure import ProcedureEngine
+from ..engines import SimSuccessors
+from ..errors import (
+    AngrDelayJobNotice,
+    AngrSkipJobNotice,
+    AngrVFGError,
+    AngrError,
+    AngrVFGRestartAnalysisNotice,
+    AngrJobMergingFailureNotice,
+    SimValueError,
+    SimIRSBError,
+    SimError,
+)
+from ..procedures import SIM_PROCEDURES
+from ..state_plugins.callstack import CallStack
+from ..analyses import AnalysesHub
+from ..sim_state import SimState
+from . import Analysis, CFGEmulated
+
+if TYPE_CHECKING:
+    from ..knowledge_base import KnowledgeBase
+
+l = logging.getLogger(name=__name__)
+
+
+class VFGJob(CFGJobBase):
+    """
+    A job descriptor that contains local variables used during VFG analysis.
+    """
+
+    def __init__(self, *args, **kwargs) -> None:
+        super().__init__(*args, **kwargs)
+
+        self.call_stack_suffix: list = []
+        self.vfg_node: VFGNode | None = None
+        self.is_call_jump = None
+        self.call_target = None
+        self.dbg_exit_status = {}
+        self.is_return_jump = None
+
+        self.sim_successors: SimSuccessors | None = None
+
+        # if this job has a call successor, do we plan to skip the call successor or not
+        self.call_skipped = False
+        # if the call is skipped, calling stack of the skipped function is saved in `call_context_key`
+        self.call_function_key: FunctionKey | None = None
+
+        self.call_task: CallAnalysis | None = None
+
+    @property
+    def block_id(self) -> BlockID | None:
+        return self._block_id
+
+    def callstack_repr(self, kb: "KnowledgeBase"):
+        s = []
+        for i in range(0, len(self.call_stack_suffix), 2):
+            call_site, func_addr = (
+                self.call_stack_suffix[i],
+                self.call_stack_suffix[i + 1],
+            )  # pylint:disable=unsubscriptable-object
+            if func_addr is None:
+                continue
+
+            call_site_str = "%#x" % call_site if call_site is not None else "None"
+
+            if func_addr in kb.functions:
+                s.append(f"{kb.functions[func_addr].name}[{call_site_str}]")
+            else:
+                s.append(f"{func_addr:#x}[{call_site_str}]")
+
+        return "//".join(s)
+
+
+class PendingJob:
+    """
+    Describes a pending job during VFG analysis.
+    """
+
+    __slots__ = (
+        "block_id",
+        "state",
+        "call_stack",
+        "src_block_id",
+        "src_stmt_idx",
+        "src_ins_addr",
+    )
+
+    def __init__(
+        self,
+        block_id: BlockID,
+        state: SimState,
+        call_stack: CallStack,
+        src_block_id: BlockID,
+        src_stmt_idx: int,
+        src_ins_addr: int,
+    ) -> None:
+        self.block_id = block_id
+        self.state = state
+        self.call_stack = call_stack
+        self.src_block_id = src_block_id
+        self.src_stmt_idx = src_stmt_idx
+        self.src_ins_addr = src_ins_addr
+
+
+class AnalysisTask:
+    """
+    An analysis task describes a task that should be done before popping this task out of the task stack and discard it.
+    """
+
+    def __init__(self) -> None:
+        pass
+
+    @property
+    def done(self):
+        raise NotImplementedError()
+
+
+class FunctionAnalysis(AnalysisTask):
+    """
+    Analyze a function, generate fix-point states from all endpoints of that function, and then merge them to one state.
+    """
+
+    def __init__(self, function_address: int, return_address: int | None) -> None:
+        super().__init__()
+
+        self.function_address = function_address
+        self.return_address = return_address
+
+        self.call_analysis: AnalysisTask | None = None
+
+        # tracks all jobs that are live currently
+        self.jobs = []
+
+    def __repr__(self):
+        s = "<Function @ %#08x with %d jobs>" % (self.function_address, len(self.jobs))
+        return s
+
+    #
+    # Properties
+    #
+
+    @property
+    def done(self) -> bool:
+        return not self.jobs
+
+
+class CallAnalysis(AnalysisTask):
+    """
+    Analyze a call by analyze all functions this call might be calling, collect all final states generated by analyzing
+    those functions, and merge them into one state.
+    """
+
+    def __init__(
+        self,
+        address: int,
+        return_address: None,
+        function_analysis_tasks: list[Any] | None = None,
+        mergeable_plugins: tuple[str, str] | None = None,
+    ) -> None:
+        super().__init__()
+
+        self.address = address
+        self.return_address = return_address
+        self.function_analysis_tasks = function_analysis_tasks if function_analysis_tasks is not None else []
+        self._mergeable_plugins = mergeable_plugins
+
+        self.skipped = False
+        self._final_jobs = []
+
+    def __repr__(self):
+        s = "<Call @ %#08x with %d function tasks>" % (self.address, len(self.function_analysis_tasks))
+        return s
+
+    #
+    # Properties
+    #
+
+    @property
+    def done(self) -> bool:
+        for task in self.function_analysis_tasks:
+            if not task.done:
+                return False
+        return True
+
+    #
+    # Public methods
+    #
+
+    def register_function_analysis(self, task: FunctionAnalysis) -> None:
+        assert isinstance(task, FunctionAnalysis)
+
+        self.function_analysis_tasks.append(task)
+        task.call_analysis = self
+
+    def add_final_job(self, job: VFGJob) -> None:
+        self._final_jobs.append(job)
+
+    def merge_jobs(self) -> VFGJob:
+        assert self._final_jobs
+
+        job = self._final_jobs[0]
+
+        for other in self._final_jobs[1:]:
+            job.state = job.state.merge(other.state, plugin_whitelist=self._mergeable_plugins)[0]
+
+        return job
+
+
+class VFGNode:
+    """
+    A descriptor of nodes in a Value-Flow Graph
+    """
+
+    def __init__(self, addr: int, key: BlockID, state: Optional["SimState"] = None) -> None:
+        """
+        Constructor.
+
+        :param int addr:
+        :param BlockID key:
+        :param SimState state:
+        """
+        self.key = key
+        self.addr = addr
+        self.state: SimState | None = None
+        self.widened_state: SimState | None = None
+        self.narrowing_times: int = 0
+        self.all_states: list[SimState] = []
+        self.events: list = []
+        self.input_variables: list = []
+        self.actions: list = []
+        self.final_states: list[SimState] = []
+
+        if state:
+            self.all_states.append(state)
+            self.state = state
+
+    def __hash__(self) -> int:
+        return hash(self.key)
+
+    def __eq__(self, o):
+        return (
+            type(self) == type(o)
+            and self.key == o.key  # pylint:disable=unidiomatic-typecheck
+            and self.addr == o.addr
+            and self.state == o.state
+            and self.actions == o.actions
+            and self.events == o.events
+            and self.narrowing_times == o.narrowing_times
+            and self.all_states == o.all_states
+            and self.widened_state == o.widened_state
+            and self.input_variables == o.input_variables
+        )
+
+    def __repr__(self):
+        s = f"VFGNode[{self.addr:#x}] <{repr(self.key)}>"
+        return s
+
+    def append_state(self, s, is_widened_state=False):
+        """
+        Appended a new state to this VFGNode.
+        :param s: The new state to append
+        :param is_widened_state: Whether it is a widened state or not.
+        """
+
+        if not is_widened_state:
+            self.all_states.append(s)
+            self.state = s
+
+        else:
+            self.widened_state = s
+
+
+class VFG(ForwardAnalysis[SimState, VFGNode, VFGJob, BlockID], Analysis):  # pylint:disable=abstract-method
+    """
+    This class represents a control-flow graph with static analysis result.
+
+    Perform abstract interpretation analysis starting from the given function address. The output is an invariant at
+    the beginning (or the end) of each basic block.
+
+    Steps:
+
+        - Generate a CFG first if CFG is not provided.
+        - Identify all merge points (denote the set of merge points as Pw) in the CFG.
+        - Cut those loop back edges (can be derived from Pw) so that we gain an acyclic CFG.
+        - Identify all variables that are 1) from memory loading 2) from initial values, or 3) phi functions. Denote
+            the set of those variables as S_{var}.
+        - Start real AI analysis and try to compute a fix point of each merge point. Perform widening/narrowing only on
+            variables \\in S_{var}.
+    """
+
+    # TODO: right now the graph traversal method is not optimal. A new solution is needed to minimize the iteration we
+    # TODO: access each node in the graph
+
+    def __init__(
+        self,
+        cfg: CFGEmulated | None = None,
+        context_sensitivity_level: int = 2,
+        start: int | None = None,
+        function_start: int | None = None,
+        interfunction_level: int = 0,
+        initial_state: Optional["SimState"] = None,
+        avoid_runs: list[int] | None = None,
+        remove_options: set[str] | None = None,
+        timeout: int | None = None,
+        max_iterations_before_widening: int = 8,
+        max_iterations: int = 40,
+        widening_interval: int = 3,
+        final_state_callback: Callable[["SimState", CallStack], Any] | None = None,
+        status_callback: Callable[["VFG"], Any] | None = None,
+        record_function_final_states: bool = False,
+    ) -> None:
+        """
+        :param cfg: The control-flow graph to base this analysis on. If none is provided, we will
+                    construct a CFGEmulated.
+        :param context_sensitivity_level: The level of context-sensitivity of this VFG.
+                                        It ranges from 0 to infinity. Default 2.
+        :param function_start: The address of the function to analyze.
+        :param interfunction_level: The level of interfunction-ness to be
+        :param initial_state: A state to use as the initial one
+        :param avoid_runs: A list of runs to avoid
+        :param remove_options: State options to remove from the initial state. It only works when `initial_state` is
+                                None
+        :param int timeout:
+        :param final_state_callback: callback function when countering final state
+        :param status_callback: callback function used in _analysis_core_baremetal
+        """
+
+        ForwardAnalysis.__init__(
+            self, order_jobs=True, allow_merging=True, allow_widening=True, status_callback=status_callback
+        )  # type:ignore
+
+        # Related CFG.
+        # We can still perform analysis if you don't specify a CFG. But providing a CFG may give you better result.
+        self._cfg = cfg
+
+        # Where to start the analysis
+        self._start: int = start if start is not None else self.project.entry
+        self._function_start: int = function_start if function_start is not None else self._start
+
+        # Other parameters
+        self._avoid_runs: list[int] = [] if avoid_runs is None else avoid_runs
+        self._context_sensitivity_level = context_sensitivity_level
+        self._interfunction_level = interfunction_level
+        self._state_options_to_remove = set() if remove_options is None else remove_options
+        self._timeout = timeout
+        self._start_at_function = self._start == self._function_start
+
+        self._initial_state = initial_state
+
+        self._max_iterations_before_widening = max_iterations_before_widening
+        self._max_iterations = max_iterations
+        self._widening_interval = widening_interval
+
+        self._final_state_callback = final_state_callback
+
+        self._record_function_final_states = record_function_final_states
+
+        self._nodes: dict[BlockID, VFGNode] = {}  # all the vfg nodes, keyed on block IDs
+        self._normal_states: dict[BlockID, SimState] = (
+            {}
+        )  # Last available state for each program point without widening
+        self._widened_states: dict[BlockID, SimState] = {}  # States on which widening has occurred
+
+        # Initial states of each function, which is context sensitive
+        # It maps function key to its states
+        self._function_initial_states: DefaultDict[int, dict[int, SimState]] = defaultdict(dict)
+        # Final states of each function, right after `ret` is called. Also context sensitive.
+        # even if a function may have multiple return sites, as long as they all return to the same place, there is
+        # only one final state of that function.
+        self._function_final_states: DefaultDict[int, dict[int, SimState]] = defaultdict(dict)
+
+        # All final states are put in this list
+        self.final_states: list[SimState] = []
+
+        self._state_initialization_map: DefaultDict[int, list[tuple[int, int]]] = defaultdict(list)
+
+        self._exit_targets: DefaultDict[tuple[int | None, ...], list[tuple[BlockID, str]]] = defaultdict(
+            list
+        )  # A dict to log edges and the jumpkind between each basic block
+        # A dict to record all blocks that returns to a specific address
+        self._return_target_sources: DefaultDict[int, list[int]] = defaultdict(list)
+
+        self._pending_returns: dict[BlockID, PendingJob] = {}
+
+        self._thumb_addrs: set[int] = set()  # set of all addresses that are code in thumb mode
+
+        self._final_address: int | None = (
+            None  # Address of the very last instruction. The analysis is terminated there.
+        )
+
+        self._function_merge_points: dict[int, list[int]] = {}
+        self._function_widening_points: dict[int, list[int]] = {}
+        self._function_node_addrs: dict[int, list[int]] = {}  # sorted in reverse post-order
+
+        self._mergeable_plugins = ("memory", "registers")
+
+        self._task_stack: list[FunctionAnalysis] = []
+
+        self._tracing_times: DefaultDict[BlockID, int] = defaultdict(int)
+
+        # counters for debugging
+        self._execution_counter: DefaultDict[BlockID, int] = defaultdict(int)
+
+        # Start analysis
+        self._analyze()
+
+    #
+    # Internal properties
+    #
+
+    @property
+    def _current_function_address(self):
+        return self._task_stack[-1].function_address
+
+    @property
+    def _top_task(self) -> FunctionAnalysis | None:
+        """
+        Get the first task in the stack.
+
+        :return: The top task in the stack, or None if the stack is empty.
+        :rtype: FunctionAnalysis
+        """
+        return self._task_stack[-1] if len(self._task_stack) != 0 else None
+
+    @property
+    def _top_function_analysis_task(self):
+        """
+        Get the first FunctionAnalysis task in the stack.
+
+        :return: The top function analysis task in the stack, or None if there isn't any.
+        :rtype: FunctionAnalysis
+        """
+
+        for r in reversed(self._task_stack):
+            if isinstance(r, FunctionAnalysis):
+                return r
+        return None
+
+    @property
+    def function_initial_states(self):
+        return self._function_initial_states
+
+    @property
+    def function_final_states(self):
+        return self._function_final_states
+
+    #
+    # Public methods
+    #
+
+    def get_any_node(self, addr: int) -> VFGNode | None:
+        """
+        Get any VFG node corresponding to the basic block at @addr.
+        Note that depending on the context sensitivity level, there might be
+        multiple nodes corresponding to different contexts. This function will
+        return the first one it encounters, which might not be what you want.
+        """
+        for n in self.graph.nodes():
+            if n.addr == addr:
+                return n
+        return None
+
+    def get_all_nodes(self, addr) -> Generator[VFGNode, None, None]:
+        for n in self.graph.nodes():
+            if n.addr == addr:
+                yield n
+
+    def irsb_from_node(self, node):
+        return self.project.factory.successors(node.state, addr=node.addr)
+
+    #
+    # Operations
+    #
+
+    def copy(self):
+        new_vfg = VFG(self.project)  # type:ignore
+        new_vfg._cfg = self._cfg
+        new_vfg._graph = networkx.DiGraph(self.graph)
+        new_vfg._nodes = self._nodes.copy()
+        new_vfg._exit_targets = defaultdict(list, self._exit_targets)
+        return new_vfg
+
+    # Pickling helpers
+    def __setstate__(self, s):
+        self.__dict__.update(s)
+
+    def __getstate__(self):
+        return dict(self.__dict__)
+
+    #
+    # Main analysis routines, mostly overriding methods of ForwardAnalysis
+    #
+
+    def _pre_analysis(self) -> None:
+        """
+        Executed before analysis starts. Necessary initializations are performed here.
+
+        :return: None
+        """
+
+        l.debug("Starting from %#x", self._start)
+
+        # initialize the task stack
+        self._task_stack = []
+
+        # initialize the execution counter dict
+        self._execution_counter = defaultdict(int)
+
+        # Generate a CFG if no CFG is provided
+        if not self._cfg:
+            l.debug("Generating a CFG, since none was given...")
+            # TODO: can we use a fast CFG instead? note that fast CFG does not care of context sensitivity at all, but
+            # TODO: for state merging, we also don't really care about context sensitivity.
+            self._cfg: CFGEmulated = self.project.analyses[CFGEmulated].prep()(
+                context_sensitivity_level=self._context_sensitivity_level, starts=(self._start,)
+            )
+
+        if not self._cfg.normalized:
+            l.warning(
+                "The given CFG is not normalized, which might impact the performance/accuracy of the VFG analysis."
+            )
+
+        # Prepare the state
+        initial_state = self._prepare_initial_state(self._start, self._initial_state)
+        initial_state.ip = self._start
+
+        if self.project.arch.name.startswith("MIPS"):
+            initial_state.regs.t9 = self._start
+
+        # clear function merge points cache
+        self._function_merge_points = {}
+
+        # Create the initial state
+        state = initial_state.copy()
+
+        if self._start_at_function:
+            # set the return address to an address so we can catch it and terminate the VSA analysis
+            # TODO: Properly pick an address that will not conflict with any existing code and data in the program
+            self._final_address = 0x4FFF0000
+            self._set_return_address(state, self._final_address)
+
+        call_stack = None
+        if not self._start_at_function:
+            # we should build a custom call stack
+            call_stack = CallStack()
+            call_stack = call_stack.call(None, self._function_start, retn_target=self._final_address)
+
+        job = VFGJob(
+            state.addr,
+            state,
+            self._context_sensitivity_level,
+            jumpkind="Ijk_Boring",
+            final_return_address=self._final_address,
+            call_stack=call_stack,
+        )
+        block_id = BlockID.new(state.addr, job.get_call_stack_suffix(), job.jumpkind)
+        job._block_id = block_id
+
+        self._insert_job(job)
+
+        # create the task
+        function_analysis_task = FunctionAnalysis(self._function_start, self._final_address)
+        function_analysis_task.jobs.append(job)
+        self._task_stack.append(function_analysis_task)
+
+    def _job_sorting_key(self, job):
+        """
+        Get the sorting key of a VFGJob instance.
+
+        :param VFGJob job: the VFGJob object.
+        :return: An integer that determines the order of this job in the queue.
+        :rtype: int
+        """
+
+        MAX_BLOCKS_PER_FUNCTION = 1000000
+
+        task_functions = list(
+            reversed(list(task.function_address for task in self._task_stack if isinstance(task, FunctionAnalysis)))
+        )
+        try:
+            function_pos = task_functions.index(job.func_addr)
+        except ValueError:
+            # not in the list
+            # it might be because we followed the wrong path, or there is a bug in the traversal algorithm
+            # anyways, do it first
+            l.warning("Function address %#x is not found in task stack.", job.func_addr)
+            return 0
+
+        try:
+            block_in_function_pos = self._ordered_node_addrs(job.func_addr).index(job.addr)
+        except ValueError:
+            # block not found. what?
+            block_in_function_pos = min(job.addr - job.func_addr, MAX_BLOCKS_PER_FUNCTION - 1)
+
+        return block_in_function_pos + MAX_BLOCKS_PER_FUNCTION * function_pos
+
+        # return self._cfg.get_topological_order(self._cfg.get_node(job.block_id))
+
+    def _job_key(self, job: VFGJob) -> BlockID | None:
+        """
+        Return the block ID of the job. Two or more jobs owning the same block ID will be merged together.
+
+        :param VFGJob job: The VFGJob instance.
+        :return:           The block ID of the job
+        :rtype:            BlockID
+        """
+
+        return job.block_id
+
+    def _pre_job_handling(self, job: VFGJob) -> None:
+        """
+        Some code executed before actually processing the job.
+
+        :param VFGJob job: the VFGJob object.
+        :return: None
+        """
+
+        # did we reach the final address?
+        if self._final_address is not None and job.addr == self._final_address:
+            # our analysis should be termianted here
+            l.debug("%s is viewed as a final state. Skip.", job)
+            raise AngrSkipJobNotice()
+
+        l.debug("Handling VFGJob %s", job)
+
+        if not self._top_task:
+            l.debug("No more tasks available. Skip the job.")
+            raise AngrSkipJobNotice()
+
+        assert isinstance(self._top_task, FunctionAnalysis)
+
+        if job not in self._top_task.jobs:
+            # it seems that all jobs of the top task has been done. unwind the task stack
+            # make sure this job is at least recorded somewhere
+            unwind_count = None
+            for i, task in enumerate(reversed(self._task_stack)):
+                if isinstance(task, FunctionAnalysis):
+                    if job in task.jobs:
+                        # nice
+                        unwind_count = i
+
+            if unwind_count is None:
+                l.debug("%s is not recorded. Skip the job.", job)
+                raise AngrSkipJobNotice()
+            else:
+                # unwind the stack till the target, unless we see any pending jobs for each new top task
+                for i in range(unwind_count):
+                    if isinstance(self._top_task, FunctionAnalysis):
+                        # are there any pending job belonging to the current function that we should handle first?
+                        pending_job_key = self._get_pending_job(self._top_task.function_address)
+                        if pending_job_key is not None:
+                            # ah there is
+                            # analyze it first
+                            self._trace_pending_job(pending_job_key)
+                            l.debug(
+                                "A pending job is found for function %#x. Delay %s.",
+                                self._top_task.function_address,
+                                job,
+                            )
+                            raise AngrDelayJobNotice()
+
+                    task = self._task_stack.pop()
+
+                    if not task.done:
+                        l.warning("Removing an unfinished task %s. Might be a bug.", task)
+
+                assert job in self._top_task.jobs
+
+        # check if this is considered to be a final state
+        if self._final_state_callback is not None and self._final_state_callback(job.state, job.call_stack):
+            l.debug("%s.state is considered as a final state. Skip the job.", job)
+            self.final_states.append(job.state)
+            raise AngrSkipJobNotice()
+
+        # increment the execution counter
+        self._execution_counter[job.addr] += 1
+
+        self._top_task.jobs.remove(job)
+
+        # set up some essential variables and parameters
+        job.call_stack_suffix = job.get_call_stack_suffix()  # type:ignore
+        job.jumpkind = "Ijk_Boring" if job.state.history.jumpkind is None else job.state.history.jumpkind
+
+        src_block_id = job.src_block_id
+        src_exit_stmt_idx = job.src_exit_stmt_idx
+
+        addr = job.state.solver.eval(job.state.regs.ip)
+        input_state = job.state
+        block_id = BlockID.new(addr, job.call_stack_suffix, job.jumpkind)
+
+        if self._tracing_times[block_id] > self._max_iterations:
+            l.debug("%s has been traced too many times. Skip", job)
+            raise AngrSkipJobNotice()
+
+        self._tracing_times[block_id] += 1
+
+        if block_id not in self._nodes:
+            vfg_node = VFGNode(addr, block_id, state=input_state)
+            self._nodes[block_id] = vfg_node
+
+        else:
+            vfg_node = self._nodes[block_id]
+
+        job.vfg_node = vfg_node
+        # log the current state
+        vfg_node.state = input_state
+
+        # Execute this basic block with input state, and get a new SimSuccessors instance
+        # unused result var is `error_occured`
+        job.sim_successors, _, restart_analysis = self._get_simsuccessors(input_state, addr)
+
+        if restart_analysis:
+            # We should restart the analysis because of something must be changed in the very initial state
+            raise AngrVFGRestartAnalysisNotice()
+
+        if job.sim_successors is None:
+            # Ouch, we cannot get the SimSuccessors for some reason
+            # Skip this guy
+            l.debug("Cannot create SimSuccessors for %s. Skip.", job)
+            raise AngrSkipJobNotice()
+
+        self._graph_add_edge(src_block_id, block_id, jumpkind=job.jumpkind, src_exit_stmt_idx=src_exit_stmt_idx)
+
+    def _get_successors(self, job: VFGJob) -> list[SimState]:
+        # Extract initial values
+        state = job.state
+        addr = job.addr
+
+        # Obtain successors
+        if addr not in self._avoid_runs:
+            assert job.sim_successors is not None
+            all_successors: list["SimState"] = (
+                job.sim_successors.flat_successors + job.sim_successors.unconstrained_successors
+            )
+        else:
+            all_successors = []
+
+        assert job.vfg_node is not None
+        # save those states
+        if job.vfg_node.final_states is None:
+            job.vfg_node.final_states = []
+        job.vfg_node.final_states.extend(all_successors)
+
+        # Update thumb_addrs
+        assert job.sim_successors
+        if job.sim_successors.sort == "IRSB" and state.thumb:
+            self._thumb_addrs.update(job.sim_successors.artifacts["insn_addrs"])
+
+        if not all_successors:
+            if job.sim_successors.sort == "SimProcedure" and isinstance(
+                job.sim_successors.artifacts["procedure"], SIM_PROCEDURES["stubs"]["PathTerminator"]
+            ):
+                # If there is no valid exit in this branch and it's not
+                # intentional (e.g. caused by a SimProcedure that does not
+                # do_return) , we should make it return to its callsite.
+                # However, we don't want to use its state as it might be
+                # corrupted. Just create a link in the exit_targets map.
+                retn_target = job.call_stack.current_return_target
+                if retn_target is not None:
+                    new_call_stack = job.call_stack_copy()
+                    exit_target_tpl = new_call_stack.stack_suffix(self._context_sensitivity_level) + (retn_target,)
+                    self._exit_targets[job.call_stack_suffix + (addr,)].append((exit_target_tpl, "Ijk_Ret"))
+            else:
+                # This is intentional. We shall remove all the pending returns generated before along this path.
+                self._remove_pending_return(job, self._pending_returns)
+
+        # If this is a call exit, we shouldn't put the default exit (which
+        # is artificial) into the CFG. The exits will be Ijk_Call and
+        # Ijk_FakeRet, and Ijk_Call always goes first
+        job.is_call_jump = any([self._is_call_jumpkind(i.history.jumpkind) for i in all_successors])
+        call_targets = []
+        for succ in all_successors:
+            if self._is_call_jumpkind(succ.history.jumpkind):
+                try:
+                    call_targets.append(succ.solver.eval_one(succ._ip))
+                    # finding one is enough!
+                    break
+                except SimValueError:
+                    # this call has multiple possible targets. ignore it
+                    pass
+
+        job.call_target = None if not call_targets else call_targets[0]
+
+        job.is_return_jump = len(all_successors) and all_successors[0].history.jumpkind == "Ijk_Ret"
+
+        if job.is_call_jump:
+            # create the call task
+
+            # TODO: correctly fill the return address
+            call_task = CallAnalysis(job.addr, None, [], mergeable_plugins=self._mergeable_plugins)
+            self._task_stack.append(call_task)
+
+            job.call_task = call_task
+
+        return all_successors
+
+    def _handle_successor(
+        self, job: VFGJob, successor: SimState, all_successors: list[SimState]
+    ) -> list[VFGJob | Any]:  # pylint:disable=arguments-renamed
+        """
+        Process each successor generated by the job, and return a new list of succeeding jobs.
+
+        :param VFGJob job: The VFGJob instance.
+        :param SimState successor: The succeeding state.
+        :param list all_successors:  A list of all successors.
+        :return: A list of newly created jobs from the successor.
+        :rtype: list
+        """
+
+        # Initialize parameters
+        addr = job.addr
+        jumpkind = successor.history.jumpkind
+
+        #
+        # Get instruction pointer
+        #
+
+        if job.is_return_jump:
+            ret_target = job.call_stack.current_return_target
+            if ret_target is None:
+                # We have no where to go according to our call stack. However, the callstack might be corrupted
+                l.debug("According to the call stack, we have nowhere to return to.")
+                return []
+
+            successor.ip = ret_target
+
+        # this try-except block is to handle cases where the instruction pointer is symbolic
+        try:
+            successor_addrs = successor.solver.eval_upto(successor.ip, 2)
+        except SimValueError:
+            # TODO: Should fall back to reading targets from CFG
+            # It cannot be concretized currently. Maybe we could handle
+            # it later, maybe it just cannot be concretized
+            return []
+
+        if len(successor_addrs) > 1:
+            # multiple concrete targets
+            if job.is_return_jump:
+                # It might be caused by state merging
+                # We may retrieve the correct ip from call stack
+                successor.ip = job.call_stack.current_return_target
+
+            else:
+                return self._handle_successor_multitargets(job, successor, all_successors)
+
+        # Now there should be one single target for the successor
+        successor_addr = successor.solver.eval_one(successor.ip)
+
+        # Get the fake ret successor
+        fakeret_successor = None
+        if self._is_call_jumpkind(jumpkind):
+            fakeret_successor = all_successors[-1]
+
+            # If the function we're calling into doesn't return, we should discard it
+            if self._cfg is not None:
+                func = self.kb.functions.function(addr=job.call_target)
+                if func is not None and func.returning is False and len(all_successors) == 2:
+                    del all_successors[-1]
+                    fakeret_successor = None
+
+        if self._is_call_jumpkind(jumpkind):
+            # Create a new call stack for the successor
+            new_call_stack = self._create_callstack(job, successor_addr, jumpkind, fakeret_successor)
+            if new_call_stack is None:
+                l.debug("Cannot create a new callstack for address %#x", successor_addr)
+                job.dbg_exit_status[successor] = ""
+                return []
+            new_call_stack_suffix = new_call_stack.stack_suffix(self._context_sensitivity_level)
+
+            new_function_key = FunctionKey.new(successor_addr, new_call_stack_suffix)
+            # Save the initial state for the function
+            self._save_function_initial_state(new_function_key, successor_addr, successor.copy())
+
+            # bail out if we hit the interfunction_level cap
+            if len(job.call_stack) >= self._interfunction_level:
+                l.debug(
+                    "We are not tracing into a new function %#08x as we hit interfunction_level limit", successor_addr
+                )
+
+                # mark it as skipped
+                job.dbg_exit_status[successor] = "Skipped"
+
+                job.call_skipped = True
+                job.call_function_key = new_function_key
+
+                job.call_task.skipped = True
+
+                return []
+
+        elif jumpkind == "Ijk_Ret":
+            # Pop the current function out from the call stack
+            new_call_stack = self._create_callstack(job, successor_addr, jumpkind, fakeret_successor)
+            if new_call_stack is None:
+                l.debug("Cannot create a new callstack for address %#x", successor_addr)
+                job.dbg_exit_status[successor] = ""
+                return []
+            new_call_stack_suffix = new_call_stack.stack_suffix(self._context_sensitivity_level)
+
+        else:
+            new_call_stack = job.call_stack
+            new_call_stack_suffix = job.call_stack_suffix
+
+        # Generate the new block ID
+        new_block_id = BlockID.new(successor_addr, new_call_stack_suffix, jumpkind)
+
+        #
+        # Generate new VFG jobs
+        #
+
+        if jumpkind == "Ijk_Ret":
+            assert not job.is_call_jump
+
+            # Record this return
+            self._return_target_sources[successor_addr].append(job.call_stack_suffix + (addr,))
+
+            # Check if this return is inside our pending returns list
+            if new_block_id in self._pending_returns:
+                del self._pending_returns[new_block_id]
+
+        # Check if we have reached a fix-point
+        if jumpkind != "Ijk_FakeRet" and new_block_id in self._nodes:
+            last_state = self._nodes[new_block_id].state
+
+            _, _, merged = last_state.merge(successor, plugin_whitelist=self._mergeable_plugins)
+
+            if merged:
+                l.debug("%s didn't reach a fix-point", new_block_id)
+            else:
+                l.debug("%s reaches a fix-point.", new_block_id)
+                job.dbg_exit_status[successor] = "Merged due to reaching a fix-point"
+                return []
+
+        new_jobs = self._create_new_jobs(job, successor, new_block_id, new_call_stack)
+
+        return new_jobs
+
+    def _handle_successor_multitargets(self, job, successor, all_successors):
+        """
+        Generate new jobs for all possible successor targets when there are more than one possible concrete value for
+        successor.ip
+
+        :param VFGJob job: The VFGJob instance.
+        :param SimState successor: The succeeding state.
+        :param list all_successors: All succeeding states from the same VFGJob.
+        :return: A list of new succeeding jobs
+        :rtype: list
+        """
+
+        new_jobs = []
+
+        # Currently we assume a legit jumping target cannot have more than 256 concrete values
+        # TODO: make it a setting on VFG
+        MAX_NUMBER_OF_CONCRETE_VALUES = 256
+
+        all_possible_ips = successor.solver.eval_upto(successor.ip, MAX_NUMBER_OF_CONCRETE_VALUES + 1)
+
+        if len(all_possible_ips) > MAX_NUMBER_OF_CONCRETE_VALUES:
+            l.warning(
+                "IP can be concretized to more than %d values, which means it might be corrupted.",
+                MAX_NUMBER_OF_CONCRETE_VALUES,
+            )
+            return []
+
+        # Call this function to generate a successor for each possible IP
+        for ip in all_possible_ips:
+            concrete_successor = successor.copy()
+            concrete_successor.ip = ip
+
+            concrete_jobs = self._handle_successor(job, concrete_successor, all_successors)
+
+            if job.is_call_jump:  # TODO: take care of syscalls
+                for new_job in concrete_jobs:
+                    # TODO: correctly fill the return address. The return address can be found from the
+                    # TODO: fakeret successor in the `successors` list
+                    function_analysis_task = FunctionAnalysis(new_job.addr, None)
+                    # log the new job
+                    function_analysis_task.jobs.append(new_job)
+                    # put it onto the stack
+                    self._task_stack.append(function_analysis_task)
+                    # log it in the call_task
+                    job.call_task.register_function_analysis(function_analysis_task)
+
+            new_jobs.extend(concrete_jobs)
+
+        return new_jobs
+
+    def _post_job_handling(
+        self, job: VFGJob, new_jobs: list[VFGJob | Any], successors: list[SimState]
+    ) -> None:  # pylint:disable=unused-argument
+        # Debugging output
+        if l.level == logging.DEBUG:
+            self._post_job_handling_debug(job, successors)
+
+        # pop all finished tasks from the task stack
+
+        pending_task_func_addrs = {k.func_addr for k in self._pending_returns}
+        while True:
+            task = self._top_task
+
+            if task is None:
+                # the task stack is empty
+                break
+
+            if isinstance(task, CallAnalysis):
+                # the call never returns
+                if task.skipped:
+                    l.debug("Calls from %s are skipped.", task)
+                else:
+                    l.debug("%s never returns.", task)
+                self._task_stack.pop()
+
+            else:
+                if not task.done or task.function_address in pending_task_func_addrs:
+                    break
+
+                l.debug("%s is finished.", task)
+                self._task_stack.pop()
+
+                # the next guy *might be* a call analysis task
+                task = self._top_task
+                if isinstance(task, CallAnalysis):
+                    if task.done:
+                        # awesome!
+                        # pop it from the task stack
+                        self._task_stack.pop()
+
+                        if task._final_jobs:
+                            # merge all jobs, and create a new job
+                            new_job = task.merge_jobs()
+
+                            # register the job to the top task
+                            self._top_task.jobs.append(new_job)
+
+                            # insert the job
+                            self._insert_job(new_job)
+
+        # if not new_jobs:
+        #    # task stack is empty
+        #    self.final_states.append(job.state)
+
+    def _intra_analysis(self) -> None:
+        pass
+
+    def _merge_jobs(self, *jobs: VFGJob):
+        l.debug("Merging jobs %s", jobs)
+
+        # there should not be more than two jobs being merged at the same time
+        assert len(jobs) == 2
+
+        addr = jobs[0].addr
+
+        if self.project.is_hooked(addr) and self.project.hooked_by(addr).is_continuation:
+            raise AngrJobMergingFailureNotice()
+
+        # update jobs
+        for job in jobs:
+            if job in self._top_function_analysis_task.jobs:
+                self._top_function_analysis_task.jobs.remove(job)
+
+        state_0 = jobs[0].state
+        state_1 = jobs[1].state
+
+        merged_state, _ = self._merge_states(state_0, state_1)
+
+        new_job = VFGJob(
+            jobs[0].addr,
+            merged_state,
+            self._context_sensitivity_level,
+            jumpkind=jobs[0].jumpkind,
+            block_id=jobs[0].block_id,
+            call_stack=jobs[0].call_stack,
+            src_block_id=jobs[0].src_block_id,
+            src_exit_stmt_idx=jobs[0].src_exit_stmt_idx,
+            src_ins_addr=jobs[0].src_ins_addr,
+        )
+
+        self._top_function_analysis_task.jobs.append(new_job)
+
+        return new_job
+
+    def _should_widen_jobs(self, *jobs):
+        """
+
+        :param iterable jobs:
+        :return: True if should widen, False otherwise
+        :rtype: bool
+        """
+
+        job_0: VFGJob
+        _: VFGJob
+        job_0, _ = jobs[-2:]  # pylint:disable=unbalanced-tuple-unpacking
+
+        addr = job_0.addr
+
+        if addr not in self._widening_points(job_0.func_addr):
+            return False
+
+        tracing_times = self._tracing_times[job_0.block_id]
+        if tracing_times > self._max_iterations_before_widening and tracing_times % self._widening_interval == 0:
+            return True
+
+        return False
+
+    def _widen_jobs(self, *jobs: VFGJob):
+        """
+
+        :param iterable jobs:
+        :return:
+        """
+
+        job_0: VFGJob
+        job_1: VFGJob
+        job_0, job_1 = jobs[-2:]  # pylint:disable=unbalanced-tuple-unpacking
+
+        assert self._top_function_analysis_task is not None
+
+        # update jobs
+        for job in jobs:
+            if job in self._top_function_analysis_task.jobs:
+                self._top_function_analysis_task.jobs.remove(job)
+
+        l.debug("Widening %s", job_1)
+
+        new_state, _ = self._widen_states(job_0.state, job_1.state)
+
+        new_job = VFGJob(
+            jobs[0].addr,
+            new_state,
+            self._context_sensitivity_level,
+            jumpkind=jobs[0].jumpkind,
+            block_id=jobs[0].block_id,
+            call_stack=jobs[0].call_stack,
+            src_block_id=jobs[0].src_block_id,
+            src_exit_stmt_idx=jobs[0].src_exit_stmt_idx,
+            src_ins_addr=jobs[0].src_ins_addr,
+        )
+        self._top_function_analysis_task.jobs.append(new_job)
+
+        return new_job
+
+    def _job_queue_empty(self) -> None:
+        if self._pending_returns:
+            # We don't have any paths remaining. Let's pop a previously-missing return to
+            # process
+
+            top_task: FunctionAnalysis = self._top_task
+            func_addr = top_task.function_address
+
+            pending_ret_key = self._get_pending_job(func_addr)
+
+            if pending_ret_key is None:
+                # analysis of the current function is somehow terminated
+                # we have to rewind the stack, and try the function that calls the current function
+                l.debug("No pending return for the current function %#x. Unwind the stack.", func_addr)
+                if not self._top_function_analysis_task.done:
+                    l.warning(
+                        "The top function analysis task is not done yet. This might be a bug. Please report to Fish."
+                    )
+                # stack unwinding
+                while True:
+                    s = self._task_stack.pop()
+                    if isinstance(s, CallAnalysis):
+                        break
+
+                return self._job_queue_empty()
+
+            self._trace_pending_job(pending_ret_key)
+
+            l.debug("Tracing a missing return %s", repr(pending_ret_key))
+        return None
+
+    def _post_analysis(self) -> None:
+        pass
+
+    #
+    # State widening, merging, and narrowing
+    #
+
+    def _merge_states(self, *states: SimState):
+        """
+        Merge two given states, and return a new one.
+
+        :param states:  All states to merge.
+        :returns: The merged state, and whether a merging has occurred
+        """
+
+        # print old_state.dbg_print_stack()
+        # print new_state.dbg_print_stack()
+
+        merged = states[0]
+        merging_occurred = False
+        for state in states[1:]:
+            merged, _, merging_occurred_ = merged.merge(state, plugin_whitelist=self._mergeable_plugins)
+            merging_occurred |= merging_occurred_
+
+        # print "Merged: "
+        # print merged_state.dbg_print_stack()
+
+        return merged, merging_occurred
+
+    @staticmethod
+    def _widen_states(old_state, new_state):
+        """
+        Perform widen operation on the given states, and return a new one.
+
+        :param old_state:
+        :param new_state:
+        :returns: The widened state, and whether widening has occurred
+        """
+
+        # print old_state.dbg_print_stack()
+        # print new_state.dbg_print_stack()
+
+        l.debug("Widening state at IP %s", old_state.ip)
+
+        widened_state, widening_occurred = old_state.widen(new_state)
+
+        # print "Widened: "
+        # print widened_state.dbg_print_stack()
+
+        return widened_state, widening_occurred
+
+    @staticmethod
+    def _narrow_states(
+        node, old_state, new_state, previously_widened_state
+    ):  # pylint:disable=unused-argument,no-self-use
+        """
+        Try to narrow the state!
+
+        :param old_state:
+        :param new_state:
+        :param previously_widened_state:
+        :returns: The narrowed state, and whether a narrowing has occurred
+        """
+
+        l.debug("Narrowing state at IP %s", previously_widened_state.ip)
+
+        s = previously_widened_state.copy()
+
+        narrowing_occurred = False
+
+        # TODO: Finish the narrowing logic
+
+        return s, narrowing_occurred
+
+    #
+    # Helper methods
+    #
+
+    def _prepare_initial_state(self, function_start: int, state: SimState | None) -> SimState:
+        """
+        Get the state to start the analysis for function.
+
+        :param int function_start: Address of the function
+        :param SimState state: The program state to base on.
+        """
+
+        if state is None:
+            state = self.project.factory.blank_state(mode="static", remove_options=self._state_options_to_remove)
+
+        assert state is not None
+        # make room for arguments passed to the function
+        sp = state.regs.sp
+        sp_val = state.solver.eval_one(sp)
+        state.memory.set_stack_address_mapping(sp_val, state.memory.stack_id(function_start) + "_pre", 0)
+        state.registers.store("sp", sp - 0x100)
+
+        # Set the stack address mapping for the initial stack
+        state.memory.set_stack_size(state.arch.stack_size)
+        initial_sp = state.solver.eval(
+            state.regs.sp
+        )  # FIXME: This is bad, as it may lose tracking of multiple sp values
+        initial_sp -= state.arch.bytes
+        state.memory.set_stack_address_mapping(initial_sp, state.memory.stack_id(function_start), function_start)
+
+        return state
+
+    def _set_return_address(self, state: "SimState", ret_addr: int) -> None:
+        """
+        Set the return address of the current state to a specific address. We assume we are at the beginning of a
+        function, or in other words, we are about to execute the very first instruction of the function.
+
+        :param SimState state: The program state
+        :param int ret_addr: The return address
+        :return: None
+        """
+
+        # TODO: the following code is totally untested other than X86 and AMD64. Don't freak out if you find bugs :)
+        # TODO: Test it
+
+        ret_bvv = state.solver.BVV(ret_addr, self.project.arch.bits)
+
+        if self.project.arch.name in ("X86", "AMD64"):
+            state.stack_push(ret_bvv)
+        elif is_arm_arch(self.project.arch):
+            state.regs.lr = ret_bvv
+        elif self.project.arch.name in ("MIPS32", "MIPS64"):
+            state.regs.ra = ret_bvv
+        elif self.project.arch.name in ("PPC32", "PPC64"):
+            state.regs.lr = ret_bvv
+        else:
+            l.warning(
+                "Return address cannot be set for architecture %s. Please add corresponding logic to "
+                "VFG._set_return_address().",
+                self.project.arch.name,
+            )
+
+    def _create_graph(self, return_target_sources=None):
+        """
+        Create a DiGraph out of the existing edge map.
+        :param return_target_sources: Used for making up those missing returns
+        :returns: A networkx.DiGraph() object
+        """
+        if return_target_sources is None:
+            # We set it to a defaultdict in order to be consistent with the
+            # actual parameter.
+            return_target_sources = defaultdict(list)
+
+        cfg = networkx.DiGraph()
+        # The corner case: add a node to the graph if there is only one block
+        if len(self._nodes) == 1:
+            cfg.add_node(self._nodes[next(iter(self._nodes.keys()))])
+
+        # Adding edges
+        for tpl, targets in self._exit_targets.items():
+            basic_block = self._nodes[tpl]  # Cannot fail :)
+            for ex, jumpkind in targets:
+                if ex in self._nodes:
+                    target_bbl = self._nodes[ex]
+                    cfg.add_edge(basic_block, target_bbl, jumpkind=jumpkind)
+
+                    # Add edges for possibly missing returns
+                    if basic_block.addr in return_target_sources:
+                        for src_irsb_key in return_target_sources[basic_block.addr]:
+                            cfg.add_edge(self._nodes[src_irsb_key], basic_block, jumpkind="Ijk_Ret")
+                else:
+                    # Debugging output
+                    def addr_formalize(addr):
+                        if addr is None:
+                            return "None"
+                        else:
+                            return "%#08x" % addr
+
+                    s = "(["
+                    for addr in ex[:-1]:
+                        s += addr_formalize(addr) + ", "
+                    s += "] %s)" % addr_formalize(ex[-1])
+                    l.warning("Key %s does not exist.", s)
+
+        return cfg
+
+    #
+    # DiGraph manipulation
+    #
+
+    def _graph_get_node(self, block_id: BlockID, terminator_for_nonexistent_node: bool = False) -> VFGNode:
+        """
+        Get an existing VFGNode instance from the graph.
+
+        :param BlockID block_id:                     The block ID for the node to get.
+        :param bool terminator_for_nonexistent_node: True if a Terminator (which is a SimProcedure stub) should be
+                                                     created when there is no existing node available for the given
+                                                     block ID.
+        :return:                                     A node in the graph, or None.
+        :rtype:                                      VFGNode
+        """
+
+        if block_id not in self._nodes:
+            l.error("Trying to look up a node that we don't have yet. Is this okay????")
+            if not terminator_for_nonexistent_node:
+                return None
+            # Generate a PathTerminator node
+            addr = block_id.addr
+            func_addr = block_id.func_addr
+            if func_addr is None:
+                # We'll have to use the current block address instead
+                # TODO: Is it really OK?
+                func_addr = addr
+
+            input_state = self.project.factory.entry_state()
+            input_state.ip = addr
+            pt = VFGNode(addr, block_id, input_state)
+            self._nodes[block_id] = pt
+
+            if isinstance(self.project.arch, archinfo.ArchARM) and addr % 2 == 1:
+                self._thumb_addrs.add(addr)
+                self._thumb_addrs.add(addr - 1)
+
+            l.debug("Block ID %s does not exist. Create a PathTerminator instead.", repr(block_id))
+
+        return self._nodes[block_id]
+
+    def _graph_add_edge(self, src_block_id: BlockID | None, dst_block_id: BlockID, **kwargs) -> None:
+        """
+        Add an edge onto the graph.
+
+        :param BlockID src_block_id: The block ID for source node.
+        :param BlockID dst_block_id: The block Id for destination node.
+        :param str jumpkind:         The jumpkind of the edge.
+        :param exit_stmt_idx:        ID of the statement in the source IRSB where this edge is created from. 'default'
+                                     refers to the default exit.
+        :return: None
+        """
+
+        dst_node = self._graph_get_node(dst_block_id, terminator_for_nonexistent_node=True)
+
+        if src_block_id is None:
+            self.graph.add_node(dst_node)
+
+        else:
+            src_node = self._graph_get_node(src_block_id, terminator_for_nonexistent_node=True)
+            self.graph.add_edge(src_node, dst_node, **kwargs)
+
+    #
+    # Other methods
+    #
+
+    def _get_simsuccessors(self, state: SimState, addr: int) -> tuple[SimSuccessors, bool, bool]:
+        error_occured = False
+        restart_analysis = False
+
+        jumpkind = "Ijk_Boring"
+        if state.history.jumpkind:
+            jumpkind = state.history.jumpkind
+
+        try:
+            node = self._cfg.model.get_any_node(addr)
+            num_inst = None if node is None else len(node.instruction_addrs)
+            sim_successors = self.project.factory.successors(state, jumpkind=jumpkind, num_inst=num_inst)
+        except SimIRSBError as ex:
+            # It's a tragedy that we came across some instructions that VEX
+            # does not support. I'll create a terminating stub there
+            l.error("SimIRSBError occurred(%s). Creating a PathTerminator.", ex)
+            error_occured = True
+            inst = SIM_PROCEDURES["stubs"]["PathTerminator"](state)
+            sim_successors = ProcedureEngine().process(state, procedure=inst)
+        except claripy.ClaripyError:
+            l.error("ClaripyError: ", exc_info=True)
+            error_occured = True
+            # Generate a PathTerminator to terminate the current path
+            inst = SIM_PROCEDURES["stubs"]["PathTerminator"](state)
+            sim_successors = ProcedureEngine().process(state, procedure=inst)
+        except SimError:
+            l.error("SimError: ", exc_info=True)
+
+            error_occured = True
+            # Generate a PathTerminator to terminate the current path
+            inst = SIM_PROCEDURES["stubs"]["PathTerminator"](state)
+            sim_successors = ProcedureEngine().process(state, procedure=inst)
+        except AngrError as ex:
+            # segment = self.project.loader.main_object.in_which_segment(addr)
+            l.error("AngrError %s when generating SimSuccessors at %#x", ex, addr, exc_info=True)
+            # We might be on a wrong branch, and is likely to encounter the
+            # "No bytes in memory xxx" exception
+            # Just ignore it
+            error_occured = True
+            sim_successors = None
+
+        return sim_successors, error_occured, restart_analysis
+
+    def _create_new_jobs(
+        self, job: VFGJob, successor: "SimState", new_block_id: BlockID, new_call_stack: CallStack
+    ) -> list[VFGJob | Any]:
+        """
+        Create a list of new VFG jobs for the successor state.
+
+        :param VFGJob job:                 The VFGJob instance.
+        :param SimState successor: The succeeding state.
+        :param BlockID new_block_id:       Block ID for the new VFGJob
+        :param new_call_stack:             The new callstack.
+        :return:                           A list of newly created VFG jobs.
+        :rtype:                            list
+        """
+
+        # TODO: basic block stack is probably useless
+
+        jumpkind = successor.history.jumpkind
+        stmt_idx = successor.scratch.stmt_idx
+        ins_addr = successor.scratch.ins_addr
+        # Make a copy of the state in case we use it later
+        successor_state = successor.copy()
+        successor_addr = successor_state.solver.eval(successor_state.ip)
+
+        new_jobs = []
+
+        if jumpkind == "Ijk_FakeRet":
+            assert job.is_call_jump
+
+            # This is the default "fake" return successor generated at each call, if and only if the target function
+            # returns.
+
+            # if the call is skipped (for whatever reason, like we reached the interfunction tracing limit), we use
+            # this FakeRet successor as the final state of the function. Otherwise we save the FakeRet state in case the
+            # callee does not return normally, but don't process them right away.
+
+            # Clear the useless values (like return addresses, parameters) on stack if needed
+            if self._cfg is not None:
+                current_function = self.kb.functions.function(job.call_target)
+                if current_function is not None:
+                    sp_difference = current_function.sp_delta
+                else:
+                    sp_difference = 0
+                arch = successor_state.arch
+                reg_sp_offset = arch.sp_offset
+                reg_sp_expr = (
+                    successor_state.registers.load(reg_sp_offset, size=arch.bytes, endness=arch.register_endness)
+                    + sp_difference
+                )  # type: ignore
+                successor_state.registers.store(arch.sp_offset, reg_sp_expr)
+
+                # Clear the return value with a TOP
+                top_si = successor_state.solver.TSI(arch.bits)
+                successor_state.registers.store(arch.ret_offset, top_si)
+
+            if job.call_skipped:
+                # TODO: Make sure the return values make sense
+                # if self.project.arch.name == "X86":
+                #     successor_state.regs.eax = successor_state.solver.BVS(
+                #         "ret_val", 32, min=0, max=0xFFFFFFFF, stride=1
+                #     )
+
+                new_job = VFGJob(
+                    successor_addr,
+                    successor_state,
+                    self._context_sensitivity_level,
+                    block_id=new_block_id,
+                    jumpkind="Ijk_Ret",
+                    call_stack=new_call_stack,
+                    src_block_id=job.block_id,
+                    src_exit_stmt_idx=stmt_idx,
+                    src_ins_addr=ins_addr,
+                )
+
+                new_jobs.append(new_job)
+                assert isinstance(self._task_stack[-2], FunctionAnalysis)
+                self._task_stack[-2].jobs.append(new_job)
+                job.dbg_exit_status[successor] = "Pending"
+
+            else:
+                self._pending_returns[new_block_id] = PendingJob(
+                    new_block_id, successor_state, new_call_stack, job.block_id, stmt_idx, ins_addr
+                )  # type: ignore
+                job.dbg_exit_status[successor] = "Pending"
+
+        else:
+            if sim_options.ABSTRACT_MEMORY in successor.options:
+                if self._is_call_jumpkind(successor.history.jumpkind):
+                    # If this is a call, we create a new stack address mapping
+                    reg_sp_si = self._create_stack_region(successor_state, successor_addr)
+
+                    # Save the new sp register
+                    new_reg_sp_expr = successor_state.solver.ValueSet(successor_state.arch.bits, "global", 0, reg_sp_si)
+                    successor_state.regs.sp = new_reg_sp_expr
+
+                elif successor.history.jumpkind == "Ijk_Ret":
+                    # Remove the existing stack address mapping
+                    # FIXME: Now we are assuming the sp is restored to its original value
+                    reg_sp_expr = successor_state.regs.sp
+
+                    if isinstance(reg_sp_expr._model_vsa, claripy.vsa.StridedInterval):  # type: ignore
+                        reg_sp_si = reg_sp_expr._model_vsa
+                        # reg_sp_si.min  # reg_sp_val
+                    elif isinstance(reg_sp_expr._model_vsa, claripy.vsa.ValueSet):  # type: ignore
+                        reg_sp_si = next(iter(reg_sp_expr._model_vsa.items()))[1]  # type: ignore
+                        # reg_sp_si.min  # reg_sp_val
+                        # TODO: Finish it!
+
+            new_job = VFGJob(
+                successor_addr,
+                successor_state,
+                self._context_sensitivity_level,
+                block_id=new_block_id,
+                jumpkind=successor_state.history.jumpkind,
+                call_stack=new_call_stack,
+                src_block_id=job.block_id,
+                src_exit_stmt_idx=stmt_idx,
+                src_ins_addr=ins_addr,
+            )
+
+            if successor.history.jumpkind == "Ijk_Ret":
+                # it's returning to the return site
+
+                # save the state as a final state of the function that we are returning from
+                if self._record_function_final_states:
+                    # key of the function that we are returning from
+                    source_function_key = FunctionKey.new(job.func_addr, job.call_stack_suffix)
+                    self._save_function_final_state(source_function_key, job.func_addr, successor_state)
+
+                # TODO: add an assertion that requires the returning target being the same as the return address we
+                # TODO: stored before
+
+                current_task = self._top_task
+                if current_task.call_analysis is not None:
+                    current_task.call_analysis.add_final_job(new_job)
+
+                    job.dbg_exit_status[successor] = "Appended to the call analysis task"
+                else:
+                    job.dbg_exit_status[successor] = "Discarded (no call analysis task)"
+
+            else:
+                if self._is_call_jumpkind(successor.history.jumpkind):
+                    # create a function analysis task
+                    # TODO: the return address
+                    task = FunctionAnalysis(new_job.addr, None)
+                    self._task_stack.append(task)
+                    # link it to the call analysis
+                    job.call_task.register_function_analysis(task)
+
+                else:
+                    task = self._top_task
+
+                # register the job to the function task
+                task.jobs.append(new_job)
+                # insert the new job into the new job array
+                new_jobs.append(new_job)
+
+                job.dbg_exit_status[successor] = "Appended"
+
+        if not job.is_call_jump or jumpkind != "Ijk_FakeRet":
+            new_target = (new_block_id, jumpkind)
+        else:
+            new_target = (new_block_id, "Ijk_FakeRet")  # This is the fake return!
+        self._exit_targets[job.call_stack_suffix + (job.addr,)].append(new_target)
+
+        return new_jobs
+
+    def _remove_pending_return(self, job, pending_returns):
+        """
+        Remove all pending returns that are related to the current job.
+        """
+
+        # Build the tuples that we want to remove from the dict fake_func_retn_exits
+        tpls_to_remove = []
+        call_stack_copy = job.call_stack_copy()
+        while call_stack_copy.current_return_target is not None:
+            ret_target = call_stack_copy.current_return_target
+            # Remove the current call stack frame
+            call_stack_copy = call_stack_copy.ret(ret_target)
+            call_stack_suffix = call_stack_copy.stack_suffix(self._context_sensitivity_level)
+            tpl = call_stack_suffix + (ret_target,)
+            tpls_to_remove.append(tpl)
+
+        # Remove those tuples from the dict
+        for tpl in tpls_to_remove:
+            if tpl in pending_returns:
+                del pending_returns[tpl]
+                l.debug(
+                    "Removed (%s) from FakeExits dict.", ",".join([hex(i) if i is not None else "None" for i in tpl])
+                )
+
+    def _post_job_handling_debug(self, job: VFGJob, successors):
+        """
+        Print out debugging information after handling a VFGJob and generating the succeeding jobs.
+
+        :param VFGJob job: The VFGJob instance.
+        :param list successors: A list of succeeding states.
+        :return: None
+        """
+
+        func = self.project.loader.find_symbol(job.addr)
+        function_name = func.name if func is not None else None
+        module_name = self.project.loader.find_object_containing(job.addr).provides
+
+        l.debug(
+            "VFGJob @ %#08x with callstack [ %s ]",
+            job.addr,
+            job.callstack_repr(self.kb),
+        )
+        l.debug("(Function %s of %s)", function_name, module_name)
+        l.debug("-  is call jump: %s", job.is_call_jump)
+        for suc in successors:
+            if suc not in job.dbg_exit_status:
+                l.warning("- %s is not found. FIND OUT WHY.", suc)
+                continue
+
+            try:
+                l.debug(
+                    "-  successor: %#08x of %s [%s]",
+                    suc.solver.eval_one(suc.ip),
+                    suc.history.jumpkind,
+                    job.dbg_exit_status[suc],
+                )
+            except SimValueError:
+                l.debug("-  target cannot be concretized. %s [%s]", job.dbg_exit_status[suc], suc.history.jumpkind)
+        l.debug("Remaining/pending jobs: %d/%d", len(self._job_info_queue), len(self._pending_returns))
+        l.debug("Remaining jobs: %s", ["%s %d" % (ent.job, id(ent.job)) for ent in self._job_info_queue])
+        l.debug("Task stack: %s", self._task_stack)
+
+    @staticmethod
+    def _is_call_jumpkind(jumpkind: str) -> bool:
+        if jumpkind == "Ijk_Call" or jumpkind.startswith("Ijk_Sys_"):
+            return True
+        return False
+
+    @staticmethod
+    def _is_return_jumpkind(jumpkind):
+        return jumpkind in ("Ijk_Ret", "Ijk_FakeRet")
+
+    @staticmethod
+    def _create_stack_region(successor_state: SimState, successor_ip: int) -> StridedInterval:
+        arch = successor_state.arch
+        reg_sp_offset = arch.sp_offset
+        reg_sp_expr = successor_state.registers.load(reg_sp_offset, size=arch.bytes, endness=arch.register_endness)
+
+        if type(reg_sp_expr._model_vsa) is claripy.BVV:  # pylint:disable=unidiomatic-typecheck
+            reg_sp_val = successor_state.solver.eval(reg_sp_expr)
+            reg_sp_si = successor_state.solver.SI(to_conv=reg_sp_expr)
+            reg_sp_si = reg_sp_si._model_vsa
+        elif type(reg_sp_expr._model_vsa) is int:  # pylint:disable=unidiomatic-typecheck
+            reg_sp_val = reg_sp_expr._model_vsa
+            reg_sp_si = successor_state.solver.SI(bits=successor_state.arch.bits, to_conv=reg_sp_val)
+            reg_sp_si = reg_sp_si._model_vsa
+        elif (
+            type(reg_sp_expr._model_vsa) is claripy.vsa.StridedInterval
+        ):  # pylint:disable=unidiomatic-typecheck # type: ignore
+            reg_sp_si = reg_sp_expr._model_vsa
+            reg_sp_val = reg_sp_si.min
+        else:
+            reg_sp_si = next(iter(reg_sp_expr._model_vsa.items()))[1]
+            reg_sp_val = reg_sp_si.min
+
+        reg_sp_val = reg_sp_val - arch.bytes  # TODO: Is it OK?
+        new_stack_region_id = successor_state.memory.stack_id(successor_ip)
+        successor_state.memory.set_stack_address_mapping(reg_sp_val, new_stack_region_id, successor_ip)
+
+        return reg_sp_si
+
+    def _create_callstack(
+        self, job: VFGJob, successor_ip: int, jumpkind: str, fakeret_successor: SimState | None
+    ) -> CallStack:
+        addr = job.addr
+
+        if self._is_call_jumpkind(jumpkind):
+            new_call_stack = job.call_stack_copy()
+            # Notice that in ARM, there are some freaking instructions
+            # like
+            # BLEQ <address>
+            # It should give us three exits: Ijk_Call, Ijk_Boring, and
+            # Ijk_Ret. The last exit is simulated.
+            # Notice: We assume the last exit is the simulated one
+            if fakeret_successor is None:
+                retn_target_addr = None
+            else:
+                retn_target_addr = fakeret_successor.solver.eval_one(fakeret_successor.ip)
+
+            # Create call stack
+            new_call_stack = new_call_stack.call(addr, successor_ip, retn_target=retn_target_addr)
+
+        elif jumpkind == "Ijk_Ret":
+            new_call_stack = job.call_stack_copy()
+            new_call_stack = new_call_stack.ret(successor_ip)
+
+        else:
+            # Normal control flow transition
+            new_call_stack = job.call_stack
+
+        return new_call_stack
+
+    def _save_function_initial_state(self, function_key: FunctionKey, function_address: int, state: SimState) -> None:
+        """
+        Save the initial state of a function, and merge it with existing ones if there are any.
+
+        :param FunctionKey function_key: The key to this function.
+        :param int function_address: Address of the function.
+        :param SimState state: Initial state of the function.
+        :return: None
+        """
+
+        l.debug("Saving the initial state for function %#08x with function key %s", function_address, function_key)
+        if function_key in self._function_initial_states[function_address]:
+            existing_state = self._function_initial_states[function_address][function_key]
+            merged_state, _, _ = existing_state.merge(state)
+            self._function_initial_states[function_address][function_key] = merged_state
+
+        else:
+            self._function_initial_states[function_address][function_key] = state
+
+    def _save_function_final_state(self, function_key, function_address, state):
+        """
+        Save the final state of a function, and merge it with existing ones if there are any.
+
+        :param FunctionKey function_key: The key to this function.
+        :param int function_address: Address of the function.
+        :param SimState state: Initial state of the function.
+        :return: None
+        """
+
+        l.debug("Saving the final state for function %#08x with function key %s", function_address, function_key)
+
+        if function_key in self._function_final_states[function_address]:
+            existing_state = self._function_final_states[function_address][function_key]
+            merged_state = existing_state.merge(state, plugin_whitelist=self._mergeable_plugins)[0]
+            self._function_final_states[function_address][function_key] = merged_state
+
+        else:
+            self._function_final_states[function_address][function_key] = state
+
+    def _trace_pending_job(self, job_key):
+        pending_job: PendingJob = self._pending_returns.pop(job_key)
+        addr = job_key.addr
+
+        # Unlike CFG, we will still trace those blocks that have been traced before. In other words, we don't
+        # remove fake returns even if they have been traced - otherwise we cannot come to a fix-point.
+
+        block_id = BlockID.new(addr, pending_job.call_stack.stack_suffix(self._context_sensitivity_level), "Ijk_Ret")
+        job = VFGJob(
+            addr,
+            pending_job.state,
+            self._context_sensitivity_level,
+            block_id=block_id,
+            jumpkind=pending_job.state.history.jumpkind,
+            call_stack=pending_job.call_stack,
+            src_block_id=pending_job.src_block_id,
+            src_exit_stmt_idx=pending_job.src_stmt_idx,
+            src_ins_addr=pending_job.src_ins_addr,
+        )
+        self._insert_job(job)
+        self._top_task.jobs.append(job)
+
+    def _get_pending_job(self, func_addr):
+        pending_ret_key = None
+        k: BlockID
+        for k in self._pending_returns:
+            if k.func_addr == func_addr:
+                pending_ret_key = k
+                break
+
+        return pending_ret_key
+
+    def _get_block_addr(self, b):  # pylint:disable=R0201
+        if isinstance(b, SimSuccessors):
+            return b.addr
+        else:
+            raise TypeError("Unsupported block type %s" % type(b))
+
+    def _get_nx_paths(self, begin, end):
+        """
+        Get the possible (networkx) simple paths between two nodes or addresses
+        corresponding to nodes.
+        Input: addresses or node instances
+        Return: a list of lists of nodes representing paths.
+        """
+        if type(begin) is int and type(end) is int:  # pylint:disable=unidiomatic-typecheck
+            n_begin = self.get_any_node(begin)
+            n_end = self.get_any_node(end)
+
+        elif isinstance(begin, VFGNode) and isinstance(end, VFGNode):  # pylint:disable=unidiomatic-typecheck
+            n_begin = begin
+            n_end = end
+        else:
+            raise AngrVFGError("from and to should be of the same type")
+
+        return networkx.all_simple_paths(self.graph, n_begin, n_end)  # type: ignore
+
+    def _merge_points(self, function_address):
+        """
+        Return the ordered merge points for a specific function.
+
+        :param int function_address: Address of the querying function.
+        :return: A list of sorted merge points (addresses).
+        :rtype: list
+        """
+
+        # we are entering a new function. now it's time to figure out how to optimally traverse the control flow
+        # graph by generating the sorted merge points
+        try:
+            new_function = self.kb.functions[function_address]
+        except KeyError:
+            # the function does not exist
+            return []
+
+        if function_address not in self._function_merge_points:
+            ordered_merge_points = GraphUtils.find_merge_points(
+                function_address, new_function.endpoints, new_function.graph
+            )
+            self._function_merge_points[function_address] = ordered_merge_points
+
+        return self._function_merge_points[function_address]
+
+    def _widening_points(self, function_address):
+        """
+        Return the ordered widening points for a specific function.
+
+        :param int function_address: Address of the querying function.
+        :return: A list of sorted merge points (addresses).
+        :rtype: list
+        """
+
+        # we are entering a new function. now it's time to figure out how to optimally traverse the control flow
+        # graph by generating the sorted merge points
+        try:
+            new_function = self.kb.functions[function_address]
+        except KeyError:
+            # the function does not exist
+            return []
+
+        if function_address not in self._function_widening_points:
+            if not new_function.normalized:
+                new_function.normalize()
+            widening_points = GraphUtils.find_widening_points(
+                function_address, new_function.endpoints, new_function.graph
+            )
+            self._function_widening_points[function_address] = widening_points
+
+        return self._function_widening_points[function_address]
+
+    def _ordered_node_addrs(self, function_address):
+        """
+        For a given function, return all nodes in an optimal traversal order. If the function does not exist, return an
+        empty list.
+
+        :param int function_address: Address of the function.
+        :return: A ordered list of the nodes.
+        :rtype: list
+        """
+
+        try:
+            function = self.kb.functions[function_address]
+        except KeyError:
+            # the function does not exist
+            return []
+
+        if function_address not in self._function_node_addrs:
+            sorted_nodes = GraphUtils.quasi_topological_sort_nodes(function.graph)
+            self._function_node_addrs[function_address] = [n.addr for n in sorted_nodes]
+
+        return self._function_node_addrs[function_address]
+
+
+AnalysesHub.register_default("VFG", VFG)