recursive parsing fix without ctypes in recursed type

Signed-off-by: varun-r-mallya <varunrmallya@gmail.com>

.github/workflows/format.yml

@@ -12,8 +12,8 @@ jobs:
     name: Format
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v4
-    - uses: actions/setup-python@v5
+    - uses: actions/checkout@v5
+    - uses: actions/setup-python@v6
       with:
         python-version: "3.x"
     - uses: pre-commit/action@v3.0.1

.gitignore

@@ -7,3 +7,5 @@ __pycache__/
 *.ll
 *.o
 .ipynb_checkpoints/
+vmlinux.py
+~*

.pre-commit-config.yaml

@@ -12,7 +12,7 @@
 #
 # See https://github.com/pre-commit/pre-commit
 
-exclude: 'vmlinux.*\.py$'
+exclude: 'vmlinux.py'
 
 ci:
   autoupdate_commit_msg: "chore: update pre-commit hooks"
@@ -41,7 +41,7 @@ repos:
     - id: ruff
       args: ["--fix", "--show-fixes"]
     - id: ruff-format
-      exclude: ^(docs)|^(tests)|^(examples)
+#      exclude: ^(docs)|^(tests)|^(examples)
 
 # Checking static types
 - repo: https://github.com/pre-commit/mirrors-mypy

README.md

@@ -83,14 +83,14 @@ def hist() -> HashMap:
 def hello(ctx: c_void_p) -> c_int64:
     process_id = pid()
     one = 1
-    prev = hist().lookup(process_id)
+    prev = hist.lookup(process_id)
     if prev:
         previous_value = prev + 1
         print(f"count: {previous_value} with {process_id}")
-        hist().update(process_id, previous_value)
+        hist.update(process_id, previous_value)
         return c_int64(0)
     else:
-        hist().update(process_id, one)
+        hist.update(process_id, one)
     return c_int64(0)
 
 

TODO.md

@@ -1,12 +0,0 @@
-## Short term
-
-- Implement enough functionality to port the BCC tutorial examples in PythonBPF
-- Add all maps
-- XDP support in pylibbpf
-- ringbuf support
-- recursive expression resolution
-
-## Long term
-
-- Refactor the codebase to be better than a hackathon project
-- Port to C++ and use actual LLVM?

examples/clone-matplotlib.ipynb

@@ -308,6 +308,7 @@
     "def hist() -> HashMap:\n",
     "    return HashMap(key=c_int32, value=c_uint64, max_entries=4096)\n",
     "\n",
+    "\n",
     "@bpf\n",
     "@section(\"tracepoint/syscalls/sys_enter_clone\")\n",
     "def hello(ctx: c_void_p) -> c_int64:\n",
@@ -329,6 +330,7 @@
     "def LICENSE() -> str:\n",
     "    return \"GPL\"\n",
     "\n",
+    "\n",
     "b = BPF()"
    ]
   },
@@ -357,7 +359,6 @@
     }
    ],
    "source": [
-    "\n",
     "b.load_and_attach()\n",
     "hist = BpfMap(b, hist)\n",
     "print(\"Recording\")\n",

examples/kprobes.py

@@ -0,0 +1,29 @@
+from pythonbpf import bpf, section, bpfglobal, BPF
+from ctypes import c_void_p, c_int64
+
+
+@bpf
+@section("kretprobe/do_unlinkat")
+def hello_world(ctx: c_void_p) -> c_int64:
+    print("Hello, World!")
+    return c_int64(0)
+
+
+@bpf
+@section("kprobe/do_unlinkat")
+def hello_world2(ctx: c_void_p) -> c_int64:
+    print("Hello, World!")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+b = BPF()
+b.load_and_attach()
+while True:
+    print("running")
+# Now cat /sys/kernel/debug/tracing/trace_pipe to see results of unlink kprobe.

examples/struct_and_perf.py

@@ -27,7 +27,7 @@ def hello(ctx: c_void_p) -> c_int32:
     dataobj.pid = pid()
     dataobj.ts = ktime()
     # dataobj.comm = strobj
-    print(f"clone called at {dataobj.ts} by pid" f"{dataobj.pid}, comm {strobj}")
+    print(f"clone called at {dataobj.ts} by pid{dataobj.pid}, comm {strobj}")
     events.output(dataobj)
     return c_int32(0)
 

examples/xdp_pass.py

@@ -1,8 +1,8 @@
-from pythonbpf import bpf, map, section, bpfglobal, compile
+from pythonbpf import bpf, map, section, bpfglobal, compile, compile_to_ir
 from pythonbpf.helper import XDP_PASS
 from pythonbpf.maps import HashMap
+from ctypes import c_int64, c_void_p
 
-from ctypes import c_void_p, c_int64
 
 # Instructions to how to run this program
 # 1. Install PythonBPF: pip install pythonbpf
@@ -41,4 +41,5 @@ def LICENSE() -> str:
     return "GPL"
 
 
+compile_to_ir("xdp_pass.py", "xdp_pass.ll")
 compile()

pythonbpf/codegen.py

@@ -4,7 +4,12 @@ from .license_pass import license_processing
 from .functions import func_proc
 from .maps import maps_proc
 from .structs import structs_proc
-from .globals_pass import globals_processing
+from .vmlinux_parser import vmlinux_proc
+from .globals_pass import (
+    globals_list_creation,
+    globals_processing,
+    populate_global_symbol_table,
+)
 from .debuginfo import DW_LANG_C11, DwarfBehaviorEnum, DebugInfoGenerator
 import os
 import subprocess
@@ -40,12 +45,16 @@ def processor(source_code, filename, module):
     for func_node in bpf_chunks:
         logger.info(f"Found BPF function/struct: {func_node.name}")
 
+    vmlinux_proc(tree, module)
+    populate_global_symbol_table(tree, module)
+    license_processing(tree, module)
+    globals_processing(tree, module)
+
     structs_sym_tab = structs_proc(tree, module, bpf_chunks)
     map_sym_tab = maps_proc(tree, module, bpf_chunks)
     func_proc(tree, module, bpf_chunks, map_sym_tab, structs_sym_tab)
 
-    license_processing(tree, module)
-    globals_processing(tree, module)
+    globals_list_creation(tree, module)
 
 
 def compile_to_ir(filename: str, output: str, loglevel=logging.INFO):

pythonbpf/globals_pass.py

@@ -1,8 +1,121 @@
 from llvmlite import ir
 import ast
 
+from logging import Logger
+import logging
+from .type_deducer import ctypes_to_ir
 
-def emit_globals(module: ir.Module, names: list[str]):
+logger: Logger = logging.getLogger(__name__)
+
+# TODO: this is going to be a huge fuck of a headache in the future.
+global_sym_tab = []
+
+
+def populate_global_symbol_table(tree, module: ir.Module):
+    for node in tree.body:
+        if isinstance(node, ast.FunctionDef):
+            for dec in node.decorator_list:
+                if (
+                    isinstance(dec, ast.Call)
+                    and isinstance(dec.func, ast.Name)
+                    and dec.func.id == "section"
+                    and len(dec.args) == 1
+                    and isinstance(dec.args[0], ast.Constant)
+                    and isinstance(dec.args[0].value, str)
+                ):
+                    global_sym_tab.append(node)
+                elif isinstance(dec, ast.Name) and dec.id == "bpfglobal":
+                    global_sym_tab.append(node)
+
+                elif isinstance(dec, ast.Name) and dec.id == "map":
+                    global_sym_tab.append(node)
+    return False
+
+
+def emit_global(module: ir.Module, node, name):
+    logger.info(f"global identifier {name} processing")
+    # deduce LLVM type from the annotated return
+    if not isinstance(node.returns, ast.Name):
+        raise ValueError(f"Unsupported return annotation {ast.dump(node.returns)}")
+    ty = ctypes_to_ir(node.returns.id)
+
+    # extract the return expression
+    # TODO: turn this return extractor into a generic function I can use everywhere.
+    ret_stmt = node.body[0]
+    if not isinstance(ret_stmt, ast.Return) or ret_stmt.value is None:
+        raise ValueError(f"Global '{name}' has no valid return")
+
+    init_val = ret_stmt.value
+
+    # simple constant like "return 0"
+    if isinstance(init_val, ast.Constant):
+        llvm_init = ir.Constant(ty, init_val.value)
+
+    # variable reference like "return SOME_CONST"
+    elif isinstance(init_val, ast.Name):
+        # need symbol resolution here, stub as 0 for now
+        raise ValueError(f"Name reference {init_val.id} not yet supported")
+
+    # constructor call like "return c_int64(0)" or dataclass(...)
+    elif isinstance(init_val, ast.Call):
+        if len(init_val.args) >= 1 and isinstance(init_val.args[0], ast.Constant):
+            llvm_init = ir.Constant(ty, init_val.args[0].value)
+        else:
+            logger.info("Defaulting to zero as no constant argument found")
+            llvm_init = ir.Constant(ty, 0)
+    else:
+        raise ValueError(f"Unsupported return expr {ast.dump(init_val)}")
+
+    gvar = ir.GlobalVariable(module, ty, name=name)
+    gvar.initializer = llvm_init
+    gvar.align = 8
+    gvar.linkage = "dso_local"
+    gvar.global_constant = False
+    return gvar
+
+
+def globals_processing(tree, module):
+    """Process stuff decorated with @bpf and @bpfglobal except license and return the section name"""
+    globals_sym_tab = []
+
+    for node in tree.body:
+        # Skip non-assignment and non-function nodes
+        if not (isinstance(node, ast.FunctionDef)):
+            continue
+
+        # Get the name based on node type
+        if isinstance(node, ast.FunctionDef):
+            name = node.name
+        else:
+            continue
+
+        # Check for duplicate names
+        if name in globals_sym_tab:
+            raise SyntaxError(f"ERROR: Global name '{name}' previously defined")
+        else:
+            globals_sym_tab.append(name)
+
+        if isinstance(node, ast.FunctionDef) and node.name != "LICENSE":
+            decorators = [
+                dec.id for dec in node.decorator_list if isinstance(dec, ast.Name)
+            ]
+            if "bpf" in decorators and "bpfglobal" in decorators:
+                if (
+                    len(node.body) == 1
+                    and isinstance(node.body[0], ast.Return)
+                    and node.body[0].value is not None
+                    and isinstance(
+                        node.body[0].value, (ast.Constant, ast.Name, ast.Call)
+                    )
+                ):
+                    emit_global(module, node, name)
+                else:
+                    raise SyntaxError(f"ERROR: Invalid syntax for {name} global")
+
+    return None
+
+
+def emit_llvm_compiler_used(module: ir.Module, names: list[str]):
     """
     Emit the @llvm.compiler.used global given a list of function/global names.
     """
@@ -24,7 +137,7 @@ def emit_globals(module: ir.Module, names: list[str]):
     gv.section = "llvm.metadata"
 
 
-def globals_processing(tree, module: ir.Module):
+def globals_list_creation(tree, module: ir.Module):
     collected = ["LICENSE"]
 
     for node in tree.body:
@@ -40,10 +153,11 @@ def globals_processing(tree, module: ir.Module):
                 ):
                     collected.append(node.name)
 
-                elif isinstance(dec, ast.Name) and dec.id == "bpfglobal":
-                    collected.append(node.name)
+                # NOTE: all globals other than
+                # elif isinstance(dec, ast.Name) and dec.id == "bpfglobal":
+                #     collected.append(node.name)
 
                 elif isinstance(dec, ast.Name) and dec.id == "map":
                     collected.append(node.name)
 
-    emit_globals(module, collected)
+    emit_llvm_compiler_used(module, collected)

pythonbpf/helper/helpers.py

@@ -15,5 +15,8 @@ def deref(ptr):
     return result if result is not None else 0
 
 
+XDP_ABORTED = ctypes.c_int64(0)
 XDP_DROP = ctypes.c_int64(1)
 XDP_PASS = ctypes.c_int64(2)
+XDP_TX = ctypes.c_int64(3)
+XDP_REDIRECT = ctypes.c_int64(4)

pythonbpf/vmlinux_parser/__init__.py

@@ -0,0 +1,3 @@
+from .import_detector import vmlinux_proc
+
+__all__ = ["vmlinux_proc"]

pythonbpf/vmlinux_parser/class_handler.py

@@ -0,0 +1,120 @@
+import logging
+from functools import lru_cache
+import importlib
+from .dependency_handler import DependencyHandler
+from .dependency_node import DependencyNode
+import ctypes
+from typing import Optional, Any
+
+logger = logging.getLogger(__name__)
+
+
+@lru_cache(maxsize=1)
+def get_module_symbols(module_name: str):
+    imported_module = importlib.import_module(module_name)
+    return [name for name in dir(imported_module)], imported_module
+
+
+def process_vmlinux_class(node, llvm_module, handler: DependencyHandler):
+    symbols_in_module, imported_module = get_module_symbols("vmlinux")
+    if node.name in symbols_in_module:
+        vmlinux_type = getattr(imported_module, node.name)
+        process_vmlinux_post_ast(vmlinux_type, llvm_module, handler)
+    else:
+        raise ImportError(f"{node.name} not in vmlinux")
+
+
+def process_vmlinux_post_ast(
+    elem_type_class, llvm_handler, handler: DependencyHandler, processing_stack=None
+):
+    # Initialize processing stack on first call
+    if processing_stack is None:
+        processing_stack = set()
+    symbols_in_module, imported_module = get_module_symbols("vmlinux")
+
+    current_symbol_name = elem_type_class.__name__
+    field_table = {}
+    is_complex_type = False
+    containing_type: Optional[Any] = None
+    ctype_complex_type: Optional[Any] = None
+    type_length: Optional[int] = None
+    module_name = getattr(elem_type_class, "__module__", None)
+
+    if hasattr(elem_type_class, "_length_") and is_complex_type:
+        type_length = elem_type_class._length_
+
+    if current_symbol_name in processing_stack:
+        logger.debug(
+            f"Circular dependency detected for {current_symbol_name}, skipping"
+        )
+        return True
+
+    # Check if already processed
+    if handler.has_node(current_symbol_name):
+        existing_node = handler.get_node(current_symbol_name)
+        # If the node exists and is ready, we're done
+        if existing_node and existing_node.is_ready:
+            logger.info(f"Node {current_symbol_name} already processed and ready")
+            return True
+
+    processing_stack.add(current_symbol_name)
+
+    if module_name == "vmlinux":
+        if hasattr(elem_type_class, "_type_"):
+            is_complex_type = True
+            containing_type = elem_type_class._type_
+            if containing_type.__module__ == "vmlinux":
+                print("Very weird type ig for containing type", containing_type)
+            elif containing_type.__module__ == ctypes.__name__:
+                if isinstance(elem_type_class, type):
+                    if issubclass(elem_type_class, ctypes.Array):
+                        ctype_complex_type = ctypes.Array
+                    elif issubclass(elem_type_class, ctypes._Pointer):
+                        ctype_complex_type = ctypes._Pointer
+                else:
+                    raise TypeError("Unsupported ctypes subclass")
+                # handle ctype complex type
+
+            else:
+                raise ImportError(f"Unsupported module of {containing_type}")
+        else:
+            new_dep_node = DependencyNode(name=current_symbol_name)
+            handler.add_node(new_dep_node)
+            class_obj = getattr(imported_module, current_symbol_name)
+            # Inspect the class fields
+            if hasattr(class_obj, "_fields_"):
+                for field_name, field_type in class_obj._fields_:
+                    field_table[field_name] = field_type
+            elif hasattr(class_obj, "__annotations__"):
+                for field_name, field_type in class_obj.__annotations__.items():
+                    field_table[field_name] = field_type
+            else:
+                raise TypeError("Could not get required class and definition")
+
+            logger.info(f"Extracted fields for {current_symbol_name}: {field_table}")
+
+            for elem_name, elem_type in field_table.items():
+                local_module_name = getattr(elem_type, "__module__", None)
+                if local_module_name == ctypes.__name__:
+                    new_dep_node.add_field(elem_name, elem_type, ready=True)
+                    logger.info(f"Field {elem_name} is direct ctypes type: {elem_type}")
+                elif local_module_name == "vmlinux":
+                    new_dep_node.add_field(elem_name, elem_type, ready=False)
+                    logger.debug(
+                        f"Processing vmlinux field: {elem_name}, type: {elem_type}"
+                    )
+                    if process_vmlinux_post_ast(
+                        elem_type, llvm_handler, handler, processing_stack
+                    ):
+                        new_dep_node.set_field_ready(elem_name, True)
+                else:
+                    raise ValueError(
+                        f"{elem_name} with type {elem_type} from module {module_name} not supported in recursive resolver"
+                    )
+            print("")
+
+    else:
+        raise ImportError("UNSUPPORTED Module")
+
+    print(current_symbol_name, "DONE")
+    print(f"handler readiness {handler.is_ready}")

pythonbpf/vmlinux_parser/dependency_handler.py

@@ -0,0 +1,149 @@
+from typing import Optional, Dict, List, Iterator
+from .dependency_node import DependencyNode
+
+
+class DependencyHandler:
+    """
+    Manages a collection of DependencyNode objects with no duplicates.
+
+    Ensures that no two nodes with the same name can be added and provides
+    methods to check readiness and retrieve specific nodes.
+
+    Example usage:
+        # Create a handler
+        handler = DependencyHandler()
+
+        # Create some dependency nodes
+        node1 = DependencyNode(name="node1")
+        node1.add_field("field1", str)
+        node1.set_field_value("field1", "value1")
+
+        node2 = DependencyNode(name="node2")
+        node2.add_field("field1", int)
+
+        # Add nodes to the handler
+        handler.add_node(node1)
+        handler.add_node(node2)
+
+        # Check if a specific node exists
+        print(handler.has_node("node1"))  # True
+
+        # Get a reference to a node and modify it
+        node = handler.get_node("node2")
+        node.set_field_value("field1", 42)
+
+        # Check if all nodes are ready
+        print(handler.is_ready)  # False (node2 is ready, but node1 isn't)
+    """
+
+    def __init__(self):
+        # Using a dictionary with node names as keys ensures name uniqueness
+        # and provides efficient lookups
+        self._nodes: Dict[str, DependencyNode] = {}
+
+    def add_node(self, node: DependencyNode) -> bool:
+        """
+        Add a dependency node to the handler.
+
+        Args:
+            node: The DependencyNode to add
+
+        Returns:
+            bool: True if the node was added, False if a node with the same name already exists
+
+        Raises:
+            TypeError: If the provided object is not a DependencyNode
+        """
+        if not isinstance(node, DependencyNode):
+            raise TypeError(f"Expected DependencyNode, got {type(node).__name__}")
+
+        # Check if a node with this name already exists
+        if node.name in self._nodes:
+            return False
+
+        self._nodes[node.name] = node
+        return True
+
+    @property
+    def is_ready(self) -> bool:
+        """
+        Check if all nodes are ready.
+
+        Returns:
+            bool: True if all nodes are ready (or if there are no nodes), False otherwise
+        """
+        if not self._nodes:
+            return True
+
+        return all(node.is_ready for node in self._nodes.values())
+
+    def has_node(self, name: str) -> bool:
+        """
+        Check if a node with the given name exists.
+
+        Args:
+            name: The name to check
+
+        Returns:
+            bool: True if a node with the given name exists, False otherwise
+        """
+        return name in self._nodes
+
+    def get_node(self, name: str) -> Optional[DependencyNode]:
+        """
+        Get a node by name for manipulation.
+
+        Args:
+            name: The name of the node to retrieve
+
+        Returns:
+            Optional[DependencyNode]: The node with the given name, or None if not found
+        """
+        return self._nodes.get(name)
+
+    def remove_node(self, node_or_name) -> bool:
+        """
+        Remove a node by name or reference.
+
+        Args:
+            node_or_name: The node to remove or its name
+
+        Returns:
+            bool: True if the node was removed, False if not found
+        """
+        if isinstance(node_or_name, DependencyNode):
+            name = node_or_name.name
+        else:
+            name = node_or_name
+
+        if name in self._nodes:
+            del self._nodes[name]
+            return True
+        return False
+
+    def get_all_nodes(self) -> List[DependencyNode]:
+        """
+        Get all nodes stored in the handler.
+
+        Returns:
+            List[DependencyNode]: List of all nodes
+        """
+        return list(self._nodes.values())
+
+    def __iter__(self) -> Iterator[DependencyNode]:
+        """
+        Iterate over all nodes.
+
+        Returns:
+            Iterator[DependencyNode]: Iterator over all nodes
+        """
+        return iter(self._nodes.values())
+
+    def __len__(self) -> int:
+        """
+        Get the number of nodes in the handler.
+
+        Returns:
+            int: The number of nodes
+        """
+        return len(self._nodes)

pythonbpf/vmlinux_parser/dependency_node.py

@@ -0,0 +1,191 @@
+from dataclasses import dataclass, field
+from typing import Dict, Any, Optional
+
+
+# TODO: FIX THE FUCKING TYPE NAME CONVENTION.
+@dataclass
+class Field:
+    """Represents a field in a dependency node with its type and readiness state."""
+
+    name: str
+    type: type
+    containing_type: Optional[Any]
+    type_size: Optional[int]
+    value: Any = None
+    ready: bool = False
+
+    def set_ready(self, is_ready: bool = True) -> None:
+        """Set the readiness state of this field."""
+        self.ready = is_ready
+
+    def set_value(self, value: Any, mark_ready: bool = True) -> None:
+        """Set the value of this field and optionally mark it as ready."""
+        self.value = value
+        if mark_ready:
+            self.ready = True
+
+    def set_type(self, given_type, mark_ready: bool = True) -> None:
+        """Set value of the type field and mark as ready"""
+        self.type = given_type
+        if mark_ready:
+            self.ready = True
+
+    def set_containing_type(
+        self, containing_type: Optional[Any], mark_ready: bool = True
+    ) -> None:
+        """Set the containing_type of this field and optionally mark it as ready."""
+        self.containing_type = containing_type
+        if mark_ready:
+            self.ready = True
+
+    def set_type_size(self, type_size: Any, mark_ready: bool = True) -> None:
+        """Set the type_size of this field and optionally mark it as ready."""
+        self.type_size = type_size
+        if mark_ready:
+            self.ready = True
+
+
+@dataclass
+class DependencyNode:
+    """
+    A node with typed fields and readiness tracking.
+
+    Example usage:
+        # Create a dependency node for a Person
+        somestruct = DependencyNode(name="struct_1")
+
+        # Add fields with their types
+        somestruct.add_field("field_1", str)
+        somestruct.add_field("field_2", int)
+        somestruct.add_field("field_3", str)
+
+        # Check if the node is ready (should be False initially)
+        print(f"Is node ready? {somestruct.is_ready}")  # False
+
+        # Set some field values
+        somestruct.set_field_value("field_1", "someproperty")
+        somestruct.set_field_value("field_2", 30)
+
+        # Check if the node is ready (still False because email is not ready)
+        print(f"Is node ready? {somestruct.is_ready}")  # False
+
+        # Set the last field and make the node ready
+        somestruct.set_field_value("field_3", "anotherproperty")
+
+        # Now the node should be ready
+        print(f"Is node ready? {somestruct.is_ready}")  # True
+
+        # You can also mark a field as not ready
+        somestruct.set_field_ready("field_3", False)
+
+        # Now the node is not ready again
+        print(f"Is node ready? {somestruct.is_ready}")  # False
+
+        # Get all field values
+        print(somestruct.get_field_values())  # {'field_1': 'someproperty', 'field_2': 30, 'field_3': 'anotherproperty'}
+
+        # Get only ready fields
+        ready_fields = somestruct.get_ready_fields()
+        print(f"Ready fields: {[field.name for field in ready_fields.values()]}")  # ['field_1', 'field_2']
+    """
+
+    name: str
+    fields: Dict[str, Field] = field(default_factory=dict)
+    _ready_cache: Optional[bool] = field(default=None, repr=False)
+
+    def add_field(
+        self,
+        name: str,
+        field_type: type,
+        initial_value: Any = None,
+        containing_type: Optional[Any] = None,
+        type_size: Optional[int] = None,
+        ready: bool = False,
+    ) -> None:
+        """Add a field to the node with an optional initial value and readiness state."""
+        self.fields[name] = Field(
+            name=name,
+            type=field_type,
+            value=initial_value,
+            ready=ready,
+            containing_type=containing_type,
+            type_size=type_size,
+        )
+        # Invalidate readiness cache
+        self._ready_cache = None
+
+    def get_field(self, name: str) -> Field:
+        """Get a field by name."""
+        return self.fields[name]
+
+    def set_field_value(self, name: str, value: Any, mark_ready: bool = True) -> None:
+        """Set a field's value and optionally mark it as ready."""
+        if name not in self.fields:
+            raise KeyError(f"Field '{name}' does not exist in node '{self.name}'")
+
+        self.fields[name].set_value(value, mark_ready)
+        # Invalidate readiness cache
+        self._ready_cache = None
+
+    def set_field_type(self, name: str, type: Any, mark_ready: bool = True) -> None:
+        """Set a field's type and optionally mark it as ready."""
+        if name not in self.fields:
+            raise KeyError(f"Field '{name}' does not exist in node '{self.name}'")
+
+        self.fields[name].set_type(type, mark_ready)
+        # Invalidate readiness cache
+        self._ready_cache = None
+
+    def set_field_containing_type(
+        self, name: str, containing_type: Any, mark_ready: bool = True
+    ) -> None:
+        """Set a field's containing_type and optionally mark it as ready."""
+        if name not in self.fields:
+            raise KeyError(f"Field '{name}' does not exist in node '{self.name}'")
+
+        self.fields[name].set_containing_type(containing_type, mark_ready)
+        # Invalidate readiness cache
+        self._ready_cache = None
+
+    def set_field_type_size(
+        self, name: str, type_size: Any, mark_ready: bool = True
+    ) -> None:
+        """Set a field's type_size and optionally mark it as ready."""
+        if name not in self.fields:
+            raise KeyError(f"Field '{name}' does not exist in node '{self.name}'")
+
+        self.fields[name].set_type_size(type_size, mark_ready)
+        # Invalidate readiness cache
+        self._ready_cache = None
+
+    def set_field_ready(self, name: str, is_ready: bool = True) -> None:
+        """Mark a field as ready or not ready."""
+        if name not in self.fields:
+            raise KeyError(f"Field '{name}' does not exist in node '{self.name}'")
+
+        self.fields[name].set_ready(is_ready)
+        # Invalidate readiness cache
+        self._ready_cache = None
+
+    @property
+    def is_ready(self) -> bool:
+        """Check if the node is ready (all fields are ready)."""
+        # Use cached value if available
+        if self._ready_cache is not None:
+            return self._ready_cache
+
+        # Calculate readiness only when needed
+        if not self.fields:
+            self._ready_cache = False
+            return False
+
+        self._ready_cache = all(elem.ready for elem in self.fields.values())
+        return self._ready_cache
+
+    def get_field_values(self) -> Dict[str, Any]:
+        """Get a dictionary of field names to their values."""
+        return {name: elem.value for name, elem in self.fields.items()}
+
+    def get_ready_fields(self) -> Dict[str, Field]:
+        """Get all fields that are marked as ready."""
+        return {name: elem for name, elem in self.fields.items() if elem.ready}

pythonbpf/vmlinux_parser/import_detector.py

@@ -0,0 +1,135 @@
+import ast
+import logging
+from typing import List, Tuple, Dict
+import importlib
+import inspect
+
+from .dependency_handler import DependencyHandler
+from .ir_generation import IRGenerator
+from .class_handler import process_vmlinux_class
+
+logger = logging.getLogger(__name__)
+
+
+def detect_import_statement(tree: ast.AST) -> List[Tuple[str, ast.ImportFrom]]:
+    """
+    Parse AST and detect import statements from vmlinux.
+
+    Returns a list of tuples (module_name, imported_item) for vmlinux imports.
+    Raises SyntaxError for invalid import patterns.
+
+    Args:
+        tree: The AST to parse
+
+    Returns:
+        List of tuples containing (module_name, imported_item) for each vmlinux import
+
+    Raises:
+        SyntaxError: If multiple imports from vmlinux are attempted or import * is used
+    """
+    vmlinux_imports = []
+
+    for node in ast.walk(tree):
+        # Handle "from vmlinux import ..." statements
+        if isinstance(node, ast.ImportFrom):
+            if node.module == "vmlinux":
+                # Check for wildcard import: from vmlinux import *
+                if any(alias.name == "*" for alias in node.names):
+                    raise SyntaxError(
+                        "Wildcard imports from vmlinux are not supported. "
+                        "Please import specific types explicitly."
+                    )
+
+                # Check for multiple imports: from vmlinux import A, B, C
+                if len(node.names) > 1:
+                    imported_names = [alias.name for alias in node.names]
+                    raise SyntaxError(
+                        f"Multiple imports from vmlinux are not supported. "
+                        f"Found: {', '.join(imported_names)}. "
+                        f"Please use separate import statements for each type."
+                    )
+
+                # Check if no specific import is specified (should not happen with valid Python)
+                if len(node.names) == 0:
+                    raise SyntaxError(
+                        "Import from vmlinux must specify at least one type."
+                    )
+
+                # Valid single import
+                for alias in node.names:
+                    import_name = alias.name
+                    # Use alias if provided, otherwise use the original name (commented)
+                    # as_name = alias.asname if alias.asname else alias.name
+                    vmlinux_imports.append(("vmlinux", node))
+                    logger.info(f"Found vmlinux import: {import_name}")
+
+        # Handle "import vmlinux" statements (not typical but should be rejected)
+        elif isinstance(node, ast.Import):
+            for alias in node.names:
+                if alias.name == "vmlinux" or alias.name.startswith("vmlinux."):
+                    raise SyntaxError(
+                        "Direct import of vmlinux module is not supported. "
+                        "Use 'from vmlinux import <type>' instead."
+                    )
+
+    logger.info(f"Total vmlinux imports detected: {len(vmlinux_imports)}")
+    return vmlinux_imports
+
+
+def vmlinux_proc(tree: ast.AST, module):
+    import_statements = detect_import_statement(tree)
+
+    # initialise dependency handler
+    handler = DependencyHandler()
+    # initialise assignment dictionary of name to type
+    assignments: Dict[str, type] = {}
+
+    if not import_statements:
+        logger.info("No vmlinux imports found")
+        return
+
+    # Import vmlinux module directly
+    try:
+        vmlinux_mod = importlib.import_module("vmlinux")
+    except ImportError:
+        logger.warning("Could not import vmlinux module")
+        return
+
+    source_file = inspect.getsourcefile(vmlinux_mod)
+    if source_file is None:
+        logger.warning("Cannot find source for vmlinux module")
+        return
+
+    with open(source_file, "r") as f:
+        mod_ast = ast.parse(f.read(), filename=source_file)
+
+    for import_mod, import_node in import_statements:
+        for alias in import_node.names:
+            imported_name = alias.name
+            found = False
+            for mod_node in mod_ast.body:
+                if (
+                    isinstance(mod_node, ast.ClassDef)
+                    and mod_node.name == imported_name
+                ):
+                    process_vmlinux_class(mod_node, module, handler)
+                    found = True
+                    break
+                if isinstance(mod_node, ast.Assign):
+                    for target in mod_node.targets:
+                        if isinstance(target, ast.Name) and target.id == imported_name:
+                            process_vmlinux_assign(mod_node, module, assignments)
+                            found = True
+                            break
+                if found:
+                    break
+            if not found:
+                logger.info(
+                    f"{imported_name} not found as ClassDef or Assign in vmlinux"
+                )
+
+    IRGenerator(module, handler)
+
+
+def process_vmlinux_assign(node, module, assignments: Dict[str, type]):
+    raise NotImplementedError("Assignment handling has not been implemented yet")

pythonbpf/vmlinux_parser/ir_generation.py

@@ -0,0 +1,8 @@
+# here, we will iterate through the dependencies and generate IR once dependencies are resolved fully
+from .dependency_handler import DependencyHandler
+
+
+class IRGenerator:
+    def __init__(self, module, handler):
+        self.module = module
+        self.handler: DependencyHandler = handler

tests/c-form/ex2.bpf.c

@@ -1,11 +1,10 @@
-#include <linux/bpf.h>
+#include "vmlinux.h"
 #include <bpf/bpf_helpers.h>
-#define u64 unsigned long long
-#define u32 unsigned int
+#include <bpf/bpf_endian.h>
 
 SEC("xdp")
 int hello(struct xdp_md *ctx) {
-    bpf_printk("Hello, World!\n");
+    bpf_printk("Hello, World! %ud \n", ctx->data);
     return XDP_PASS;
 }
 

tests/c-form/globals.bpf.c

@@ -0,0 +1,27 @@
+// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include <linux/types.h>
+
+struct test_struct {
+    __u64 a;
+    __u64 b;
+};
+
+struct test_struct w = {};
+volatile __u64 prev_time = 0;
+
+SEC("tracepoint/syscalls/sys_enter_execve")
+int trace_execve(void *ctx)
+{
+    bpf_printk("previous %ul now %ul", w.b, w.a);
+    __u64 ts = bpf_ktime_get_ns();
+    bpf_printk("prev %ul now %ul", prev_time, ts);
+    w.a = ts;
+    w.b = prev_time;
+    prev_time = ts;
+    return 0;
+}
+
+char LICENSE[] SEC("license") = "GPL";

tests/c-form/kprobe.bpf.c

@@ -0,0 +1,19 @@
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char LICENSE[] SEC("license") = "Dual BSD/GPL";
+
+SEC("kprobe/do_unlinkat")
+int kprobe_execve(struct pt_regs *ctx)
+{
+    bpf_printk("unlinkat created");
+    return 0;
+}
+
+SEC("kretprobe/do_unlinkat")
+int kretprobe_execve(struct pt_regs *ctx)
+{
+    bpf_printk("unlinkat returned\n");
+    return 0;
+}

tests/failing_tests/conditionals/oneline.py

@@ -0,0 +1,18 @@
+from pythonbpf import bpf, section, bpfglobal, compile
+from ctypes import c_void_p, c_int64
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def hello_world(ctx: c_void_p) -> c_int64:
+    print("Hello, World!") if True else print("Goodbye, World!")
+    return
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/failing_tests/globals.py

@@ -0,0 +1,109 @@
+import logging
+
+from pythonbpf import compile, bpf, section, bpfglobal, compile_to_ir
+from ctypes import c_void_p, c_int64, c_int32
+
+
+@bpf
+@bpfglobal
+def somevalue() -> c_int32:
+    return c_int32(42)
+
+
+@bpf
+@bpfglobal
+def somevalue2() -> c_int64:
+    return c_int64(69)
+
+
+@bpf
+@bpfglobal
+def somevalue1() -> c_int32:
+    return c_int32(42)
+
+
+# --- Passing examples ---
+
+
+# Simple constant return
+@bpf
+@bpfglobal
+def g1() -> c_int64:
+    return c_int64(42)
+
+
+# Constructor with one constant argument
+@bpf
+@bpfglobal
+def g2() -> c_int64:
+    return c_int64(69)
+
+
+# --- Failing examples ---
+
+# No return annotation
+# @bpf
+# @bpfglobal
+# def g3():
+#     return 42
+
+# Return annotation is complex
+# @bpf
+# @bpfglobal
+# def g4() -> List[int]:
+#     return []
+
+# # Return is missing
+# @bpf
+# @bpfglobal
+# def g5() -> c_int64:
+#     pass
+
+# # Return is a variable reference
+# #TODO: maybe fix this sometime later. It defaults to 0
+# CONST = 5
+# @bpf
+# @bpfglobal
+# def g6() -> c_int64:
+#     return c_int64(CONST)
+
+
+# Constructor with multiple args
+# TODO: this is not working. should it work ?
+@bpf
+@bpfglobal
+def g7() -> c_int64:
+    return c_int64(1)
+
+
+# Dataclass call
+# TODO: fails with dataclass
+# @dataclass
+# class Point:
+#     x: c_int64
+#     y: c_int64
+
+# @bpf
+# @bpfglobal
+# def g8() -> Point:
+#     return Point(1, 2)
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def sometag(ctx: c_void_p) -> c_int64:
+    print("test")
+    global somevalue
+    somevalue = 2
+    print(f"{somevalue}")
+    return c_int64(1)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("globals.py", "globals.ll", loglevel=logging.INFO)
+compile()

tests/failing_tests/named_arg.py

@@ -11,6 +11,7 @@ from ctypes import c_void_p, c_int64
 # We cannot allocate space for the intermediate type now.
 # We probably need to track the ref/deref chain for each variable.
 
+
 @bpf
 @map
 def count() -> HashMap:

tests/failing_tests/undeclared_values.py

@@ -0,0 +1,23 @@
+import logging
+
+from pythonbpf import compile, bpf, section, bpfglobal, compile_to_ir
+from ctypes import c_void_p, c_int64
+
+
+# This should not pass as somevalue is not declared at all.
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def sometag(ctx: c_void_p) -> c_int64:
+    print("test")
+    print(f"{somevalue}")  # noqa: F821
+    return c_int64(1)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("globals.py", "globals.ll", loglevel=logging.INFO)
+compile()

tests/failing_tests/xdp_pass.py

@@ -0,0 +1,48 @@
+from pythonbpf import bpf, map, section, bpfglobal, compile, compile_to_ir
+from pythonbpf.maps import HashMap
+from pythonbpf.helper import XDP_PASS
+from vmlinux import struct_xdp_md
+from vmlinux import struct_ring_buffer_per_cpu  # noqa: F401
+
+# from vmlinux import struct_xdp_buff  # noqa: F401
+# from vmlinux import struct_xdp_md
+from ctypes import c_int64
+
+# Instructions to how to run this program
+# 1. Install PythonBPF: pip install pythonbpf
+# 2. Run the program: python examples/xdp_pass.py
+# 3. Run the program with sudo: sudo tools/check.sh run examples/xdp_pass.o
+# 4. Attach object file to any network device with something like ./check.sh xdp examples/xdp_pass.o tailscale0
+# 5. send traffic through the device and observe effects
+
+
+@bpf
+@map
+def count() -> HashMap:
+    return HashMap(key=c_int64, value=c_int64, max_entries=1)
+
+
+@bpf
+@section("xdp")
+def hello_world(ctx: struct_xdp_md) -> c_int64:
+    key = 0
+    one = 1
+    prev = count().lookup(key)
+    if prev:
+        prevval = prev + 1
+        print(f"count: {prevval}")
+        count().update(key, prevval)
+        return XDP_PASS
+    else:
+        count().update(key, one)
+
+    return XDP_PASS
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("xdp_pass.py", "xdp_pass.ll")

tests/passing_tests/return/typecast_var.py

@@ -6,8 +6,8 @@ from ctypes import c_void_p, c_int32
 @section("tracepoint/syscalls/sys_enter_execve")
 def hello_world(ctx: c_void_p) -> c_int32:
     print("Hello, World!")
-    a = 1 # int64
-    return c_int32(a) # typecast to int32
+    a = 1  # int64
+    return c_int32(a)  # typecast to int32
 
 
 @bpf

tools/vmlinux-gen.py

@@ -0,0 +1,379 @@
+#!/usr/bin/env python3
+"""
+BTF to Python ctypes Converter
+Converts Linux kernel BTF (BPF Type Format) to Python ctypes definitions.
+
+This tool automates the process of:
+1. Dumping BTF from vmlinux
+2. Preprocessing enum definitions
+3. Processing struct kioctx to extract anonymous nested structs
+4. Running C preprocessor
+5. Converting to Python ctypes using clang2py
+6. Post-processing the output
+
+Requirements:
+- bpftool
+- clang
+- ctypeslib2 (pip install ctypeslib2)
+"""
+
+import argparse
+import os
+import re
+import subprocess
+import sys
+import tempfile
+
+
+class BTFConverter:
+    def __init__(
+        self,
+        btf_source="/sys/kernel/btf/vmlinux",
+        output_file="vmlinux.py",
+        keep_intermediate=False,
+        verbose=False,
+    ):
+        self.btf_source = btf_source
+        self.output_file = output_file
+        self.keep_intermediate = keep_intermediate
+        self.verbose = verbose
+        self.temp_dir = tempfile.mkdtemp() if not keep_intermediate else "."
+
+    def log(self, message):
+        """Print message if verbose mode is enabled."""
+        if self.verbose:
+            print(f"[*] {message}")
+
+    def run_command(self, cmd, description):
+        """Run a shell command and handle errors."""
+        self.log(f"{description}...")
+        try:
+            result = subprocess.run(
+                cmd, shell=True, check=True, capture_output=True, text=True
+            )
+            if self.verbose and result.stdout:
+                print(result.stdout)
+            return result
+        except subprocess.CalledProcessError as e:
+            print(f"Error during {description}:", file=sys.stderr)
+            print(e.stderr, file=sys.stderr)
+            sys.exit(1)
+
+    def step1_dump_btf(self):
+        """Step 1: Dump BTF from vmlinux."""
+        vmlinux_h = os.path.join(self.temp_dir, "vmlinux.h")
+        cmd = f"bpftool btf dump file {self.btf_source} format c > {vmlinux_h}"
+        self.run_command(cmd, "Dumping BTF from vmlinux")
+        return vmlinux_h
+
+    def step2_preprocess_enums(self, input_file):
+        """Step 1.5: Preprocess enum definitions."""
+        self.log("Preprocessing enum definitions...")
+
+        with open(input_file, "r") as f:
+            original_code = f.read()
+
+        # Extract anonymous enums
+        enums = re.findall(
+            r"(?<!typedef\s)(enum\s*\{[^}]*\})\s*(\w+)\s*(?::\s*\d+)?\s*;",
+            original_code,
+        )
+        enum_defs = [enum_block + ";" for enum_block, _ in enums]
+
+        # Replace anonymous enums with int declarations
+        processed_code = re.sub(
+            r"(?<!typedef\s)enum\s*\{[^}]*\}\s*(\w+)\s*(?::\s*\d+)?\s*;",
+            r"int \1;",
+            original_code,
+        )
+
+        # Prepend enum definitions
+        if enum_defs:
+            enum_text = "\n".join(enum_defs) + "\n\n"
+            processed_code = enum_text + processed_code
+
+        output_file = os.path.join(self.temp_dir, "vmlinux_processed.h")
+        with open(output_file, "w") as f:
+            f.write(processed_code)
+
+        return output_file
+
+    def step2_5_process_kioctx(self, input_file):
+        # TODO: this is a very bad bug and design decision. A single struct has an issue mostly.
+        """Step 2.5: Process struct kioctx to extract nested anonymous structs."""
+        self.log("Processing struct kioctx nested structs...")
+
+        with open(input_file, "r") as f:
+            content = f.read()
+
+        # Pattern to match struct kioctx with its full body (handles multiple nesting levels)
+        kioctx_pattern = (
+            r"struct\s+kioctx\s*\{(?:[^{}]|\{(?:[^{}]|\{[^{}]*\})*\})*\}\s*;"
+        )
+
+        def process_kioctx_replacement(match):
+            full_struct = match.group(0)
+            self.log(f"Found struct kioctx, length: {len(full_struct)} chars")
+
+            # Extract the struct body (everything between outermost { and })
+            body_match = re.search(
+                r"struct\s+kioctx\s*\{(.*)\}\s*;", full_struct, re.DOTALL
+            )
+            if not body_match:
+                return full_struct
+
+            body = body_match.group(1)
+
+            # Find all anonymous structs within the body
+            # Pattern: struct { ... } followed by ; (not a member name)
+            # anon_struct_pattern = r"struct\s*\{[^}]*\}"
+
+            anon_structs = []
+            anon_counter = 4  # Start from 4, counting down to 1
+
+            def replace_anonymous_struct(m):
+                nonlocal anon_counter
+                anon_struct_content = m.group(0)
+
+                # Extract the body of the anonymous struct
+                anon_body_match = re.search(
+                    r"struct\s*\{(.*)\}", anon_struct_content, re.DOTALL
+                )
+                if not anon_body_match:
+                    return anon_struct_content
+
+                anon_body = anon_body_match.group(1)
+
+                # Create the named struct definition
+                anon_name = f"__anon{anon_counter}"
+                member_name = f"a{anon_counter}"
+
+                # Store the struct definition
+                anon_structs.append(f"struct {anon_name} {{{anon_body}}};")
+
+                anon_counter -= 1
+
+                # Return the member declaration
+                return f"struct {anon_name} {member_name}"
+
+            # Process the body, finding and replacing anonymous structs
+            # We need to be careful to only match anonymous structs followed by ;
+            processed_body = body
+
+            # Find all occurrences and process them
+            pattern_with_semicolon = r"struct\s*\{([^}]*)\}\s*;"
+            matches = list(re.finditer(pattern_with_semicolon, body, re.DOTALL))
+
+            if not matches:
+                self.log("No anonymous structs found in kioctx")
+                return full_struct
+
+            self.log(f"Found {len(matches)} anonymous struct(s)")
+
+            # Process in reverse order to maintain string positions
+            for match in reversed(matches):
+                anon_struct_content = match.group(1)
+                start_pos = match.start()
+                end_pos = match.end()
+
+                # Create the named struct definition
+                anon_name = f"__anon{anon_counter}"
+                member_name = f"a{anon_counter}"
+
+                # Store the struct definition
+                anon_structs.insert(0, f"struct {anon_name} {{{anon_struct_content}}};")
+
+                # Replace in the body
+                replacement = f"struct {anon_name} {member_name};"
+                processed_body = (
+                    processed_body[:start_pos] + replacement + processed_body[end_pos:]
+                )
+
+                anon_counter -= 1
+
+            # Rebuild the complete definition
+            if anon_structs:
+                # Prepend the anonymous struct definitions
+                anon_definitions = "\n".join(anon_structs) + "\n\n"
+                new_struct = f"struct kioctx {{{processed_body}}};"
+                return anon_definitions + new_struct
+            else:
+                return full_struct
+
+        # Apply the transformation
+        processed_content = re.sub(
+            kioctx_pattern, process_kioctx_replacement, content, flags=re.DOTALL
+        )
+
+        output_file = os.path.join(self.temp_dir, "vmlinux_kioctx_processed.h")
+        with open(output_file, "w") as f:
+            f.write(processed_content)
+
+        self.log(f"Saved kioctx-processed output to {output_file}")
+        return output_file
+
+    def step3_run_preprocessor(self, input_file):
+        """Step 2: Run C preprocessor."""
+        output_file = os.path.join(self.temp_dir, "vmlinux.i")
+        cmd = f"clang -E {input_file} > {output_file}"
+        self.run_command(cmd, "Running C preprocessor")
+        return output_file
+
+    def step4_convert_to_ctypes(self, input_file):
+        """Step 3: Convert to Python ctypes using clang2py."""
+        output_file = os.path.join(self.temp_dir, "vmlinux_raw.py")
+        cmd = (
+            f"clang2py {input_file} -o {output_file} "
+            f'--clang-args="-fno-ms-extensions -I/usr/include -I/usr/include/linux"'
+        )
+        self.run_command(cmd, "Converting to Python ctypes")
+        return output_file
+
+    def step5_postprocess(self, input_file):
+        """Step 4: Post-process the generated Python file."""
+        self.log("Post-processing Python ctypes definitions...")
+
+        with open(input_file, "r") as f:
+            data = f.read()
+
+        # Remove lines like ('_45', ctypes.c_int64, 0)
+        data = re.sub(r"\('_[0-9]+',\s*ctypes\.[a-zA-Z0-9_]+,\s*0\),?\s*\n?", "", data)
+
+        # Replace ('_20', ctypes.c_uint64, 64) → ('_20', ctypes.c_uint64)
+        data = re.sub(
+            r"\('(_[0-9]+)',\s*(ctypes\.[a-zA-Z0-9_]+),\s*[0-9]+\)", r"('\1', \2)", data
+        )
+
+        # Replace ('_20', ctypes.c_char, 8) with ('_20', ctypes.c_uint8, 8)
+        data = re.sub(r"(ctypes\.c_char)(\s*,\s*\d+\))", r"ctypes.c_uint8\2", data)
+
+        # below to replace those c_bool with bitfield greater than 8
+        def repl(m):
+            name, bits = m.groups()
+            return (
+                f"('{name}', ctypes.c_uint32, {bits})" if int(bits) > 8 else m.group(0)
+            )
+
+        data = re.sub(r"\('([^']+)',\s*ctypes\.c_bool,\s*(\d+)\)", repl, data)
+
+        # Remove ctypes. prefix from invalid entries
+        invalid_ctypes = ["bpf_iter_state", "_cache_type", "fs_context_purpose"]
+        for name in invalid_ctypes:
+            data = re.sub(rf"\bctypes\.{name}\b", name, data)
+
+        with open(self.output_file, "w") as f:
+            f.write(data)
+
+        self.log(f"Saved final output to {self.output_file}")
+
+    def cleanup(self):
+        """Remove temporary files if not keeping them."""
+        if not self.keep_intermediate and self.temp_dir != ".":
+            self.log(f"Cleaning up temporary directory: {self.temp_dir}")
+            import shutil
+
+            shutil.rmtree(self.temp_dir, ignore_errors=True)
+
+    def convert(self):
+        """Run the complete conversion pipeline."""
+        try:
+            self.log("Starting BTF to Python ctypes conversion...")
+
+            # Check dependencies
+            self.check_dependencies()
+
+            # Run conversion pipeline
+            vmlinux_h = self.step1_dump_btf()
+            vmlinux_processed_h = self.step2_preprocess_enums(vmlinux_h)
+            vmlinux_kioctx_h = self.step2_5_process_kioctx(vmlinux_processed_h)
+            vmlinux_i = self.step3_run_preprocessor(vmlinux_kioctx_h)
+            vmlinux_raw_py = self.step4_convert_to_ctypes(vmlinux_i)
+            self.step5_postprocess(vmlinux_raw_py)
+
+            print(f"\n✓ Conversion complete! Output saved to: {self.output_file}")
+
+        except Exception as e:
+            print(f"\n✗ Error during conversion: {e}", file=sys.stderr)
+            import traceback
+
+            traceback.print_exc()
+            sys.exit(1)
+        finally:
+            self.cleanup()
+
+    def check_dependencies(self):
+        """Check if required tools are available."""
+        self.log("Checking dependencies...")
+
+        dependencies = {
+            "bpftool": "bpftool --version",
+            "clang": "clang --version",
+            "clang2py": "clang2py --version",
+        }
+
+        missing = []
+        for tool, cmd in dependencies.items():
+            try:
+                subprocess.run(cmd, shell=True, check=True, capture_output=True)
+            except subprocess.CalledProcessError:
+                missing.append(tool)
+
+        if missing:
+            print("Error: Missing required dependencies:", file=sys.stderr)
+            for tool in missing:
+                print(f"  - {tool}", file=sys.stderr)
+            if "clang2py" in missing:
+                print("\nInstall ctypeslib2: pip install ctypeslib2", file=sys.stderr)
+            sys.exit(1)
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Convert Linux kernel BTF to Python ctypes definitions",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+  %(prog)s
+  %(prog)s -o kernel_types.py
+  %(prog)s --btf-source /sys/kernel/btf/custom_module -k -v
+        """,
+    )
+
+    parser.add_argument(
+        "--btf-source",
+        default="/sys/kernel/btf/vmlinux",
+        help="Path to BTF source (default: /sys/kernel/btf/vmlinux)",
+    )
+
+    parser.add_argument(
+        "-o",
+        "--output",
+        default="vmlinux.py",
+        help="Output Python file (default: vmlinux.py)",
+    )
+
+    parser.add_argument(
+        "-k",
+        "--keep-intermediate",
+        action="store_true",
+        help="Keep intermediate files (vmlinux.h, vmlinux_processed.h, etc.)",
+    )
+
+    parser.add_argument(
+        "-v", "--verbose", action="store_true", help="Enable verbose output"
+    )
+
+    args = parser.parse_args()
+
+    converter = BTFConverter(
+        btf_source=args.btf_source,
+        output_file=args.output,
+        keep_intermediate=args.keep_intermediate,
+        verbose=args.verbose,
+    )
+
+    converter.convert()
+
+
+if __name__ == "__main__":
+    main()