2 Commits
v0.1.5 ... sym

Author SHA1 Message Date
7bc711c296 Update pythonbpf/functions_pass.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
2025-10-02 05:01:32 +05:30
80c3519b95 Fix local_sym_tab usage in binary_ops 2025-10-02 04:58:39 +05:30
95 changed files with 204895 additions and 5095 deletions

View File

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

3
.gitignore vendored
View File

@ -7,6 +7,3 @@ __pycache__/
*.ll
*.o
.ipynb_checkpoints/
vmlinux.py
~*
vmlinux.h

View File

@ -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"
@ -21,7 +21,7 @@ ci:
repos:
# Standard hooks
- repo: https://github.com/pre-commit/pre-commit-hooks
rev: v6.0.0
rev: v4.6.0
hooks:
- id: check-added-large-files
- id: check-case-conflict
@ -36,16 +36,16 @@ repos:
- id: trailing-whitespace
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: "v0.13.2"
rev: "v0.4.2"
hooks:
- 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
rev: "v1.18.2"
rev: "v1.10.0"
hooks:
- id: mypy
exclude: ^(tests)|^(examples)

View File

@ -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)

13
TODO.md Normal file
View File

@ -0,0 +1,13 @@
## Short term
- Implement enough functionality to port the BCC tutorial examples in PythonBPF
- Static Typing
- 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?

View File

@ -12,7 +12,7 @@
"from pythonbpf import bpf, map, section, bpfglobal, BPF\n",
"from pythonbpf.helper import pid\n",
"from pythonbpf.maps import HashMap\n",
"from pylibbpf import BpfMap\n",
"from pylibbpf import *\n",
"from ctypes import c_void_p, c_int64, c_uint64, c_int32\n",
"import matplotlib.pyplot as plt"
]
@ -308,7 +308,6 @@
"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",
@ -330,7 +329,6 @@
"def LICENSE() -> str:\n",
" return \"GPL\"\n",
"\n",
"\n",
"b = BPF()"
]
},
@ -359,6 +357,7 @@
}
],
"source": [
"\n",
"b.load_and_attach()\n",
"hist = BpfMap(b, hist)\n",
"print(\"Recording\")\n",

View File

@ -1,29 +0,0 @@
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.

View File

@ -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{dataobj.pid}, comm {strobj}")
print(f"clone called at {dataobj.ts} by pid" f"{dataobj.pid}, comm {strobj}")
events.output(dataobj)
return c_int32(0)

View File

@ -21,17 +21,17 @@ def last() -> HashMap:
@section("tracepoint/syscalls/sys_enter_sync")
def do_trace(ctx: c_void_p) -> c_int64:
key = 0
tsp = last.lookup(key)
tsp = last().lookup(key)
if tsp:
kt = ktime()
delta = kt - tsp
if delta < 1000000000:
time_ms = delta // 1000000
print(f"sync called within last second, last {time_ms} ms ago")
last.delete(key)
last().delete(key)
else:
kt = ktime()
last.update(key, kt)
last().update(key, kt)
return c_int64(0)

203381
examples/vmlinux.py Normal file

File diff suppressed because it is too large Load Diff

View File

@ -1,8 +1,8 @@
from pythonbpf import bpf, map, section, bpfglobal, compile, compile_to_ir
from pythonbpf import bpf, map, section, bpfglobal, compile
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
@ -41,5 +41,4 @@ def LICENSE() -> str:
return "GPL"
compile_to_ir("xdp_pass.py", "xdp_pass.ll")
compile()

View File

@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
[project]
name = "pythonbpf"
version = "0.1.5"
version = "0.1.3"
description = "Reduced Python frontend for eBPF"
authors = [
{ name = "r41k0u", email="pragyanshchaturvedi18@gmail.com" },

View File

@ -1,178 +0,0 @@
import ast
import logging
from llvmlite import ir
from dataclasses import dataclass
from typing import Any
from pythonbpf.helper import HelperHandlerRegistry
from pythonbpf.type_deducer import ctypes_to_ir
logger = logging.getLogger(__name__)
@dataclass
class LocalSymbol:
var: ir.AllocaInstr
ir_type: ir.Type
metadata: Any = None
def __iter__(self):
yield self.var
yield self.ir_type
yield self.metadata
def handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab):
"""Handle memory allocation for assignment statements."""
# Validate assignment
if len(stmt.targets) != 1:
logger.warning("Multi-target assignment not supported, skipping allocation")
return
target = stmt.targets[0]
# Skip non-name targets (e.g., struct field assignments)
if isinstance(target, ast.Attribute):
logger.debug(f"Struct field assignment to {target.attr}, no allocation needed")
return
if not isinstance(target, ast.Name):
logger.warning(f"Unsupported assignment target type: {type(target).__name__}")
return
var_name = target.id
rval = stmt.value
# Skip if already allocated
if var_name in local_sym_tab:
logger.debug(f"Variable {var_name} already allocated, skipping")
return
# Determine type and allocate based on rval
if isinstance(rval, ast.Call):
_allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab)
elif isinstance(rval, ast.Constant):
_allocate_for_constant(builder, var_name, rval, local_sym_tab)
elif isinstance(rval, ast.BinOp):
_allocate_for_binop(builder, var_name, local_sym_tab)
else:
logger.warning(
f"Unsupported assignment value type for {var_name}: {type(rval).__name__}"
)
def _allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab):
"""Allocate memory for variable assigned from a call."""
if isinstance(rval.func, ast.Name):
call_type = rval.func.id
# C type constructors
if call_type in ("c_int32", "c_int64", "c_uint32", "c_uint64"):
ir_type = ctypes_to_ir(call_type)
var = builder.alloca(ir_type, name=var_name)
var.align = ir_type.width // 8
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} as {call_type}")
# Helper functions
elif HelperHandlerRegistry.has_handler(call_type):
ir_type = ir.IntType(64) # Assume i64 return type
var = builder.alloca(ir_type, name=var_name)
var.align = 8
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} for helper {call_type}")
# Deref function
elif call_type == "deref":
ir_type = ir.IntType(64) # Assume i64 return type
var = builder.alloca(ir_type, name=var_name)
var.align = 8
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} for deref")
# Struct constructors
elif call_type in structs_sym_tab:
struct_info = structs_sym_tab[call_type]
var = builder.alloca(struct_info.ir_type, name=var_name)
local_sym_tab[var_name] = LocalSymbol(var, struct_info.ir_type, call_type)
logger.info(f"Pre-allocated {var_name} for struct {call_type}")
else:
logger.warning(f"Unknown call type for allocation: {call_type}")
elif isinstance(rval.func, ast.Attribute):
# Map method calls - need double allocation for ptr handling
_allocate_for_map_method(builder, var_name, local_sym_tab)
else:
logger.warning(f"Unsupported call function type for {var_name}")
def _allocate_for_map_method(builder, var_name, local_sym_tab):
"""Allocate memory for variable assigned from map method (double alloc)."""
# Main variable (pointer to pointer)
ir_type = ir.PointerType(ir.IntType(64))
var = builder.alloca(ir_type, name=var_name)
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
# Temporary variable for computed values
tmp_ir_type = ir.IntType(64)
var_tmp = builder.alloca(tmp_ir_type, name=f"{var_name}_tmp")
local_sym_tab[f"{var_name}_tmp"] = LocalSymbol(var_tmp, tmp_ir_type)
logger.info(f"Pre-allocated {var_name} and {var_name}_tmp for map method")
def _allocate_for_constant(builder, var_name, rval, local_sym_tab):
"""Allocate memory for variable assigned from a constant."""
if isinstance(rval.value, bool):
ir_type = ir.IntType(1)
var = builder.alloca(ir_type, name=var_name)
var.align = 1
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} as bool")
elif isinstance(rval.value, int):
ir_type = ir.IntType(64)
var = builder.alloca(ir_type, name=var_name)
var.align = 8
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} as i64")
elif isinstance(rval.value, str):
ir_type = ir.PointerType(ir.IntType(8))
var = builder.alloca(ir_type, name=var_name)
var.align = 8
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} as string")
else:
logger.warning(
f"Unsupported constant type for {var_name}: {type(rval.value).__name__}"
)
def _allocate_for_binop(builder, var_name, local_sym_tab):
"""Allocate memory for variable assigned from a binary operation."""
ir_type = ir.IntType(64) # Assume i64 result
var = builder.alloca(ir_type, name=var_name)
var.align = 8
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
logger.info(f"Pre-allocated {var_name} for binop result")
def allocate_temp_pool(builder, max_temps, local_sym_tab):
"""Allocate the temporary scratch space pool for helper arguments."""
if max_temps == 0:
return
logger.info(f"Allocating temp pool of {max_temps} variables")
for i in range(max_temps):
temp_name = f"__helper_temp_{i}"
temp_var = builder.alloca(ir.IntType(64), name=temp_name)
temp_var.align = 8
local_sym_tab[temp_name] = LocalSymbol(temp_var, ir.IntType(64))

View File

@ -1,108 +0,0 @@
import ast
import logging
from llvmlite import ir
from pythonbpf.expr import eval_expr
logger = logging.getLogger(__name__)
def handle_struct_field_assignment(
func, module, builder, target, rval, local_sym_tab, map_sym_tab, structs_sym_tab
):
"""Handle struct field assignment (obj.field = value)."""
var_name = target.value.id
field_name = target.attr
if var_name not in local_sym_tab:
logger.error(f"Variable '{var_name}' not found in symbol table")
return
struct_type = local_sym_tab[var_name].metadata
struct_info = structs_sym_tab[struct_type]
if field_name not in struct_info.fields:
logger.error(f"Field '{field_name}' not found in struct '{struct_type}'")
return
# Get field pointer and evaluate value
field_ptr = struct_info.gep(builder, local_sym_tab[var_name].var, field_name)
val = eval_expr(
func, module, builder, rval, local_sym_tab, map_sym_tab, structs_sym_tab
)
if val is None:
logger.error(f"Failed to evaluate value for {var_name}.{field_name}")
return
# TODO: Handle string assignment to char array (not a priority)
field_type = struct_info.field_type(field_name)
if isinstance(field_type, ir.ArrayType) and val[1] == ir.PointerType(ir.IntType(8)):
logger.warning(
f"String to char array assignment not implemented for {var_name}.{field_name}"
)
return
# Store the value
builder.store(val[0], field_ptr)
logger.info(f"Assigned to struct field {var_name}.{field_name}")
def handle_variable_assignment(
func, module, builder, var_name, rval, local_sym_tab, map_sym_tab, structs_sym_tab
):
"""Handle single named variable assignment."""
if var_name not in local_sym_tab:
logger.error(f"Variable {var_name} not declared.")
return False
var_ptr = local_sym_tab[var_name].var
var_type = local_sym_tab[var_name].ir_type
# NOTE: Special case for struct initialization
if isinstance(rval, ast.Call) and isinstance(rval.func, ast.Name):
struct_name = rval.func.id
if struct_name in structs_sym_tab and len(rval.args) == 0:
struct_info = structs_sym_tab[struct_name]
ir_struct = struct_info.ir_type
builder.store(ir.Constant(ir_struct, None), var_ptr)
logger.info(f"Initialized struct {struct_name} for variable {var_name}")
return True
val_result = eval_expr(
func, module, builder, rval, local_sym_tab, map_sym_tab, structs_sym_tab
)
if val_result is None:
logger.error(f"Failed to evaluate value for {var_name}")
return False
val, val_type = val_result
logger.info(f"Evaluated value for {var_name}: {val} of type {val_type}, {var_type}")
if val_type != var_type:
if isinstance(val_type, ir.IntType) and isinstance(var_type, ir.IntType):
# Allow implicit int widening
if val_type.width < var_type.width:
val = builder.sext(val, var_type)
logger.info(f"Implicitly widened int for variable {var_name}")
elif val_type.width > var_type.width:
val = builder.trunc(val, var_type)
logger.info(f"Implicitly truncated int for variable {var_name}")
elif isinstance(val_type, ir.IntType) and isinstance(var_type, ir.PointerType):
# NOTE: This is assignment to a PTR_TO_MAP_VALUE_OR_NULL
logger.info(
f"Creating temporary variable for pointer assignment to {var_name}"
)
var_ptr_tmp = local_sym_tab[f"{var_name}_tmp"].var
builder.store(val, var_ptr_tmp)
val = var_ptr_tmp
else:
logger.error(
f"Type mismatch for variable {var_name}: {val_type} vs {var_type}"
)
return False
builder.store(val, var_ptr)
logger.info(f"Assigned value to variable {var_name}")
return True

71
pythonbpf/binary_ops.py Normal file
View File

@ -0,0 +1,71 @@
import ast
from llvmlite import ir
def recursive_dereferencer(var, builder):
"""dereference until primitive type comes out"""
if var.type == ir.PointerType(ir.PointerType(ir.IntType(64))):
a = builder.load(var)
return recursive_dereferencer(a, builder)
elif var.type == ir.PointerType(ir.IntType(64)):
a = builder.load(var)
return recursive_dereferencer(a, builder)
elif var.type == ir.IntType(64):
return var
else:
raise TypeError(f"Unsupported type for dereferencing: {var.type}")
def handle_binary_op(rval, module, builder, var_name, local_sym_tab, map_sym_tab, func):
print(module)
left = rval.left
right = rval.right
op = rval.op
# Handle left operand
if isinstance(left, ast.Name):
if left.id in local_sym_tab:
left = recursive_dereferencer(local_sym_tab[left.id].var, builder)
else:
raise SyntaxError(f"Undefined variable: {left.id}")
elif isinstance(left, ast.Constant):
left = ir.Constant(ir.IntType(64), left.value)
else:
raise SyntaxError("Unsupported left operand type")
if isinstance(right, ast.Name):
if right.id in local_sym_tab:
right = recursive_dereferencer(local_sym_tab[right.id].var, builder)
else:
raise SyntaxError(f"Undefined variable: {right.id}")
elif isinstance(right, ast.Constant):
right = ir.Constant(ir.IntType(64), right.value)
else:
raise SyntaxError("Unsupported right operand type")
print(f"left is {left}, right is {right}, op is {op}")
if isinstance(op, ast.Add):
builder.store(builder.add(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.Sub):
builder.store(builder.sub(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.Mult):
builder.store(builder.mul(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.Div):
builder.store(builder.sdiv(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.Mod):
builder.store(builder.srem(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.LShift):
builder.store(builder.shl(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.RShift):
builder.store(builder.lshr(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.BitOr):
builder.store(builder.or_(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.BitXor):
builder.store(builder.xor(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.BitAnd):
builder.store(builder.and_(left, right), local_sym_tab[var_name].var)
elif isinstance(op, ast.FloorDiv):
builder.store(builder.udiv(left, right), local_sym_tab[var_name].var)
else:
raise SyntaxError("Unsupported binary operation")

View File

@ -1,38 +1,19 @@
import ast
from llvmlite import ir
from .license_pass import license_processing
from .functions import func_proc
from .functions_pass import func_proc
from .maps import maps_proc
from .structs import structs_proc
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
from .globals_pass import globals_processing
from .debuginfo import DW_LANG_C11, DwarfBehaviorEnum
import os
import subprocess
import inspect
from pathlib import Path
from pylibbpf import BpfProgram
import tempfile
from logging import Logger
import logging
import re
logger: Logger = logging.getLogger(__name__)
VERSION = "v0.1.5"
def finalize_module(original_str):
"""After all IR generation is complete, we monkey patch btf_ama attribute"""
# Create a string with applied transformation of btf_ama attribute addition to BTF struct field accesses.
pattern = r'(@"llvm\.[^"]+:[^"]*" = external global i64, !llvm\.preserve\.access\.index ![0-9]+)'
replacement = r'\1 "btf_ama"'
return re.sub(pattern, replacement, original_str)
VERSION = "v0.1.3"
def find_bpf_chunks(tree):
@ -49,28 +30,21 @@ def find_bpf_chunks(tree):
def processor(source_code, filename, module):
tree = ast.parse(source_code, filename)
logger.debug(ast.dump(tree, indent=4))
print(ast.dump(tree, indent=4))
bpf_chunks = find_bpf_chunks(tree)
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)
print(f"Found BPF function/struct: {func_node.name}")
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)
globals_list_creation(tree, module)
license_processing(tree, module)
globals_processing(tree, module)
def compile_to_ir(filename: str, output: str, loglevel=logging.INFO):
logging.basicConfig(
level=loglevel, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s"
)
def compile_to_ir(filename: str, output: str):
with open(filename) as f:
source = f.read()
@ -79,17 +53,33 @@ def compile_to_ir(filename: str, output: str, loglevel=logging.INFO):
module.triple = "bpf"
if not hasattr(module, "_debug_compile_unit"):
debug_generator = DebugInfoGenerator(module)
debug_generator.generate_file_metadata(filename, os.path.dirname(filename))
debug_generator.generate_debug_cu(
DW_LANG_C11,
f"PythonBPF {VERSION}",
True, # TODO: This is probably not true
# TODO: add a global field here that keeps track of all the globals. Works without it, but I think it might
# be required for kprobes.
True,
module._file_metadata = module.add_debug_info(
"DIFile",
{ # type: ignore
"filename": filename,
"directory": os.path.dirname(filename),
},
)
module._debug_compile_unit = module.add_debug_info(
"DICompileUnit",
{ # type: ignore
"language": DW_LANG_C11,
"file": module._file_metadata, # type: ignore
"producer": f"PythonBPF {VERSION}",
"isOptimized": True, # TODO: This is probably not true
# TODO: add a global field here that keeps track of all the globals. Works without it, but I think it might
# be required for kprobes.
"runtimeVersion": 0,
"emissionKind": 1,
"splitDebugInlining": False,
"nameTableKind": 0,
},
is_distinct=True,
)
module.add_named_metadata("llvm.dbg.cu", module._debug_compile_unit) # type: ignore
processor(source, filename, module)
wchar_size = module.add_metadata(
@ -131,45 +121,16 @@ def compile_to_ir(filename: str, output: str, loglevel=logging.INFO):
module.add_named_metadata("llvm.ident", [f"PythonBPF {VERSION}"])
module_string = finalize_module(str(module))
logger.info(f"IR written to {output}")
print(f"IR written to {output}")
with open(output, "w") as f:
f.write(f'source_filename = "{filename}"\n')
f.write(module_string)
f.write(str(module))
f.write("\n")
return output
def _run_llc(ll_file, obj_file):
"""Compile LLVM IR to BPF object file using llc."""
logger.info(f"Compiling IR to object: {ll_file} -> {obj_file}")
result = subprocess.run(
[
"llc",
"-march=bpf",
"-filetype=obj",
"-O2",
str(ll_file),
"-o",
str(obj_file),
],
check=True,
capture_output=True,
text=True,
)
if result.returncode == 0:
logger.info(f"Object file written to {obj_file}")
return True
else:
logger.error(f"llc compilation failed: {result.stderr}")
return False
def compile(loglevel=logging.INFO) -> bool:
def compile() -> bool:
# Look one level up the stack to the caller of this function
caller_frame = inspect.stack()[1]
caller_file = Path(caller_frame.filename).resolve()
@ -178,17 +139,29 @@ def compile(loglevel=logging.INFO) -> bool:
o_file = caller_file.with_suffix(".o")
success = True
success = (
compile_to_ir(str(caller_file), str(ll_file), loglevel=loglevel) and success
success = compile_to_ir(str(caller_file), str(ll_file)) and success
success = bool(
subprocess.run(
[
"llc",
"-march=bpf",
"-filetype=obj",
"-O2",
str(ll_file),
"-o",
str(o_file),
],
check=True,
)
and success
)
success = _run_llc(ll_file, o_file) and success
logger.info(f"Object written to {o_file}")
print(f"Object written to {o_file}")
return success
def BPF(loglevel=logging.INFO) -> BpfProgram:
def BPF() -> BpfProgram:
caller_frame = inspect.stack()[1]
src = inspect.getsource(caller_frame.frame)
with tempfile.NamedTemporaryFile(
@ -201,7 +174,18 @@ def BPF(loglevel=logging.INFO) -> BpfProgram:
f.write(src)
f.flush()
source = f.name
compile_to_ir(source, str(inter.name), loglevel=loglevel)
_run_llc(str(inter.name), str(obj_file.name))
compile_to_ir(source, str(inter.name))
subprocess.run(
[
"llc",
"-march=bpf",
"-filetype=obj",
"-O2",
str(inter.name),
"-o",
str(obj_file.name),
],
check=True,
)
return BpfProgram(str(obj_file.name))

View File

@ -12,34 +12,6 @@ class DebugInfoGenerator:
self.module = module
self._type_cache = {} # Cache for common debug types
def generate_file_metadata(self, filename, dirname):
self.module._file_metadata = self.module.add_debug_info(
"DIFile",
{ # type: ignore
"filename": filename,
"directory": dirname,
},
)
def generate_debug_cu(
self, language, producer: str, is_optimized: bool, is_distinct: bool
):
self.module._debug_compile_unit = self.module.add_debug_info(
"DICompileUnit",
{ # type: ignore
"language": language,
"file": self.module._file_metadata, # type: ignore
"producer": producer,
"isOptimized": is_optimized,
"runtimeVersion": 0,
"emissionKind": 1,
"splitDebugInlining": False,
"nameTableKind": 0,
},
is_distinct=is_distinct,
)
self.module.add_named_metadata("llvm.dbg.cu", self.module._debug_compile_unit) # type: ignore
def get_basic_type(self, name: str, size: int, encoding: int) -> Any:
"""Get or create a basic type with caching"""
key = (name, size, encoding)

View File

@ -1,14 +0,0 @@
from .expr_pass import eval_expr, handle_expr, get_operand_value
from .type_normalization import convert_to_bool, get_base_type_and_depth
from .ir_ops import deref_to_depth
from .call_registry import CallHandlerRegistry
__all__ = [
"eval_expr",
"handle_expr",
"convert_to_bool",
"get_base_type_and_depth",
"deref_to_depth",
"get_operand_value",
"CallHandlerRegistry",
]

View File

@ -1,20 +0,0 @@
class CallHandlerRegistry:
"""Registry for handling different types of calls (helpers, etc.)"""
_handler = None
@classmethod
def set_handler(cls, handler):
"""Set the handler for unknown calls"""
cls._handler = handler
@classmethod
def handle_call(
cls, call, module, builder, func, local_sym_tab, map_sym_tab, structs_sym_tab
):
"""Handle a call using the registered handler"""
if cls._handler is None:
return None
return cls._handler(
call, module, builder, func, local_sym_tab, map_sym_tab, structs_sym_tab
)

View File

@ -1,566 +0,0 @@
import ast
from llvmlite import ir
from logging import Logger
import logging
from typing import Dict
from pythonbpf.type_deducer import ctypes_to_ir, is_ctypes
from .call_registry import CallHandlerRegistry
from .type_normalization import (
convert_to_bool,
handle_comparator,
get_base_type_and_depth,
deref_to_depth,
)
logger: Logger = logging.getLogger(__name__)
# ============================================================================
# Leaf Handlers (No Recursive eval_expr calls)
# ============================================================================
def _handle_name_expr(expr: ast.Name, local_sym_tab: Dict, builder: ir.IRBuilder):
"""Handle ast.Name expressions."""
if expr.id in local_sym_tab:
var = local_sym_tab[expr.id].var
val = builder.load(var)
return val, local_sym_tab[expr.id].ir_type
else:
logger.info(f"Undefined variable {expr.id}")
return None
def _handle_constant_expr(module, builder, expr: ast.Constant):
"""Handle ast.Constant expressions."""
if isinstance(expr.value, int) or isinstance(expr.value, bool):
return ir.Constant(ir.IntType(64), int(expr.value)), ir.IntType(64)
elif isinstance(expr.value, str):
str_name = f".str.{id(expr)}"
str_bytes = expr.value.encode("utf-8") + b"\x00"
str_type = ir.ArrayType(ir.IntType(8), len(str_bytes))
str_constant = ir.Constant(str_type, bytearray(str_bytes))
# Create global variable
global_str = ir.GlobalVariable(module, str_type, name=str_name)
global_str.linkage = "internal"
global_str.global_constant = True
global_str.initializer = str_constant
str_ptr = builder.bitcast(global_str, ir.PointerType(ir.IntType(8)))
return str_ptr, ir.PointerType(ir.IntType(8))
else:
logger.error(f"Unsupported constant type {ast.dump(expr)}")
return None
def _handle_attribute_expr(
expr: ast.Attribute,
local_sym_tab: Dict,
structs_sym_tab: Dict,
builder: ir.IRBuilder,
):
"""Handle ast.Attribute expressions for struct field access."""
if isinstance(expr.value, ast.Name):
var_name = expr.value.id
attr_name = expr.attr
if var_name in local_sym_tab:
var_ptr, var_type, var_metadata = local_sym_tab[var_name]
logger.info(f"Loading attribute {attr_name} from variable {var_name}")
logger.info(f"Variable type: {var_type}, Variable ptr: {var_ptr}")
metadata = structs_sym_tab[var_metadata]
if attr_name in metadata.fields:
gep = metadata.gep(builder, var_ptr, attr_name)
val = builder.load(gep)
field_type = metadata.field_type(attr_name)
return val, field_type
return None
def _handle_deref_call(expr: ast.Call, local_sym_tab: Dict, builder: ir.IRBuilder):
"""Handle deref function calls."""
logger.info(f"Handling deref {ast.dump(expr)}")
if len(expr.args) != 1:
logger.info("deref takes exactly one argument")
return None
arg = expr.args[0]
if (
isinstance(arg, ast.Call)
and isinstance(arg.func, ast.Name)
and arg.func.id == "deref"
):
logger.info("Multiple deref not supported")
return None
if isinstance(arg, ast.Name):
if arg.id in local_sym_tab:
arg_ptr = local_sym_tab[arg.id].var
else:
logger.info(f"Undefined variable {arg.id}")
return None
else:
logger.info("Unsupported argument type for deref")
return None
if arg_ptr is None:
logger.info("Failed to evaluate deref argument")
return None
# Load the value from pointer
val = builder.load(arg_ptr)
return val, local_sym_tab[arg.id].ir_type
# ============================================================================
# Binary Operations
# ============================================================================
def get_operand_value(
func, module, operand, builder, local_sym_tab, map_sym_tab, structs_sym_tab=None
):
"""Extract the value from an operand, handling variables and constants."""
logger.info(f"Getting operand value for: {ast.dump(operand)}")
if isinstance(operand, ast.Name):
if operand.id in local_sym_tab:
var = local_sym_tab[operand.id].var
var_type = var.type
base_type, depth = get_base_type_and_depth(var_type)
logger.info(f"var is {var}, base_type is {base_type}, depth is {depth}")
val = deref_to_depth(func, builder, var, depth)
return val
raise ValueError(f"Undefined variable: {operand.id}")
elif isinstance(operand, ast.Constant):
if isinstance(operand.value, int):
cst = ir.Constant(ir.IntType(64), int(operand.value))
return cst
raise TypeError(f"Unsupported constant type: {type(operand.value)}")
elif isinstance(operand, ast.BinOp):
res = _handle_binary_op_impl(
func, module, operand, builder, local_sym_tab, map_sym_tab, structs_sym_tab
)
return res
else:
res = eval_expr(
func, module, builder, operand, local_sym_tab, map_sym_tab, structs_sym_tab
)
if res is None:
raise ValueError(f"Failed to evaluate call expression: {operand}")
val, _ = res
logger.info(f"Evaluated expr to {val} of type {val.type}")
base_type, depth = get_base_type_and_depth(val.type)
if depth > 0:
val = deref_to_depth(func, builder, val, depth)
return val
raise TypeError(f"Unsupported operand type: {type(operand)}")
def _handle_binary_op_impl(
func, module, rval, builder, local_sym_tab, map_sym_tab, structs_sym_tab=None
):
op = rval.op
left = get_operand_value(
func, module, rval.left, builder, local_sym_tab, map_sym_tab, structs_sym_tab
)
right = get_operand_value(
func, module, rval.right, builder, local_sym_tab, map_sym_tab, structs_sym_tab
)
logger.info(f"left is {left}, right is {right}, op is {op}")
# NOTE: Before doing the operation, if the operands are integers
# we always extend them to i64. The assignment to LHS will take
# care of truncation if needed.
if isinstance(left.type, ir.IntType) and left.type.width < 64:
left = builder.sext(left, ir.IntType(64))
if isinstance(right.type, ir.IntType) and right.type.width < 64:
right = builder.sext(right, ir.IntType(64))
# Map AST operation nodes to LLVM IR builder methods
op_map = {
ast.Add: builder.add,
ast.Sub: builder.sub,
ast.Mult: builder.mul,
ast.Div: builder.sdiv,
ast.Mod: builder.srem,
ast.LShift: builder.shl,
ast.RShift: builder.lshr,
ast.BitOr: builder.or_,
ast.BitXor: builder.xor,
ast.BitAnd: builder.and_,
ast.FloorDiv: builder.udiv,
}
if type(op) in op_map:
result = op_map[type(op)](left, right)
return result
else:
raise SyntaxError("Unsupported binary operation")
def _handle_binary_op(
func,
module,
rval,
builder,
var_name,
local_sym_tab,
map_sym_tab,
structs_sym_tab=None,
):
result = _handle_binary_op_impl(
func, module, rval, builder, local_sym_tab, map_sym_tab, structs_sym_tab
)
if var_name and var_name in local_sym_tab:
logger.info(
f"Storing result {result} into variable {local_sym_tab[var_name].var}"
)
builder.store(result, local_sym_tab[var_name].var)
return result, result.type
# ============================================================================
# Comparison and Unary Operations
# ============================================================================
def _handle_ctypes_call(
func,
module,
builder,
expr,
local_sym_tab,
map_sym_tab,
structs_sym_tab=None,
):
"""Handle ctypes type constructor calls."""
if len(expr.args) != 1:
logger.info("ctypes constructor takes exactly one argument")
return None
arg = expr.args[0]
val = eval_expr(
func,
module,
builder,
arg,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
if val is None:
logger.info("Failed to evaluate argument to ctypes constructor")
return None
call_type = expr.func.id
expected_type = ctypes_to_ir(call_type)
if val[1] != expected_type:
# NOTE: We are only considering casting to and from int types for now
if isinstance(val[1], ir.IntType) and isinstance(expected_type, ir.IntType):
if val[1].width < expected_type.width:
val = (builder.sext(val[0], expected_type), expected_type)
else:
val = (builder.trunc(val[0], expected_type), expected_type)
else:
raise ValueError(f"Type mismatch: expected {expected_type}, got {val[1]}")
return val
def _handle_compare(
func, module, builder, cond, local_sym_tab, map_sym_tab, structs_sym_tab=None
):
"""Handle ast.Compare expressions."""
if len(cond.ops) != 1 or len(cond.comparators) != 1:
logger.error("Only single comparisons are supported")
return None
lhs = eval_expr(
func,
module,
builder,
cond.left,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
rhs = eval_expr(
func,
module,
builder,
cond.comparators[0],
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
if lhs is None or rhs is None:
logger.error("Failed to evaluate comparison operands")
return None
lhs, _ = lhs
rhs, _ = rhs
return handle_comparator(func, builder, cond.ops[0], lhs, rhs)
def _handle_unary_op(
func,
module,
builder,
expr: ast.UnaryOp,
local_sym_tab,
map_sym_tab,
structs_sym_tab=None,
):
"""Handle ast.UnaryOp expressions."""
if not isinstance(expr.op, ast.Not) and not isinstance(expr.op, ast.USub):
logger.error("Only 'not' and '-' unary operators are supported")
return None
operand = get_operand_value(
func, module, expr.operand, builder, local_sym_tab, map_sym_tab, structs_sym_tab
)
if operand is None:
logger.error("Failed to evaluate operand for unary operation")
return None
if isinstance(expr.op, ast.Not):
true_const = ir.Constant(ir.IntType(1), 1)
result = builder.xor(convert_to_bool(builder, operand), true_const)
return result, ir.IntType(1)
elif isinstance(expr.op, ast.USub):
# Multiply by -1
neg_one = ir.Constant(ir.IntType(64), -1)
result = builder.mul(operand, neg_one)
return result, ir.IntType(64)
# ============================================================================
# Boolean Operations
# ============================================================================
def _handle_and_op(func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab):
"""Handle `and` boolean operations."""
logger.debug(f"Handling 'and' operator with {len(expr.values)} operands")
merge_block = func.append_basic_block(name="and.merge")
false_block = func.append_basic_block(name="and.false")
incoming_values = []
for i, value in enumerate(expr.values):
is_last = i == len(expr.values) - 1
# Evaluate current operand
operand_result = eval_expr(
func, None, builder, value, local_sym_tab, map_sym_tab, structs_sym_tab
)
if operand_result is None:
logger.error(f"Failed to evaluate operand {i} in 'and' expression")
return None
operand_val, operand_type = operand_result
# Convert to boolean if needed
operand_bool = convert_to_bool(builder, operand_val)
current_block = builder.block
if is_last:
# Last operand: result is this value
builder.branch(merge_block)
incoming_values.append((operand_bool, current_block))
else:
# Not last: check if true, continue or short-circuit
next_check = func.append_basic_block(name=f"and.check_{i + 1}")
builder.cbranch(operand_bool, next_check, false_block)
builder.position_at_end(next_check)
# False block: short-circuit with false
builder.position_at_end(false_block)
builder.branch(merge_block)
false_value = ir.Constant(ir.IntType(1), 0)
incoming_values.append((false_value, false_block))
# Merge block: phi node
builder.position_at_end(merge_block)
phi = builder.phi(ir.IntType(1), name="and.result")
for val, block in incoming_values:
phi.add_incoming(val, block)
logger.debug(f"Generated 'and' with {len(incoming_values)} incoming values")
return phi, ir.IntType(1)
def _handle_or_op(func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab):
"""Handle `or` boolean operations."""
logger.debug(f"Handling 'or' operator with {len(expr.values)} operands")
merge_block = func.append_basic_block(name="or.merge")
true_block = func.append_basic_block(name="or.true")
incoming_values = []
for i, value in enumerate(expr.values):
is_last = i == len(expr.values) - 1
# Evaluate current operand
operand_result = eval_expr(
func, None, builder, value, local_sym_tab, map_sym_tab, structs_sym_tab
)
if operand_result is None:
logger.error(f"Failed to evaluate operand {i} in 'or' expression")
return None
operand_val, operand_type = operand_result
# Convert to boolean if needed
operand_bool = convert_to_bool(builder, operand_val)
current_block = builder.block
if is_last:
# Last operand: result is this value
builder.branch(merge_block)
incoming_values.append((operand_bool, current_block))
else:
# Not last: check if false, continue or short-circuit
next_check = func.append_basic_block(name=f"or.check_{i + 1}")
builder.cbranch(operand_bool, true_block, next_check)
builder.position_at_end(next_check)
# True block: short-circuit with true
builder.position_at_end(true_block)
builder.branch(merge_block)
true_value = ir.Constant(ir.IntType(1), 1)
incoming_values.append((true_value, true_block))
# Merge block: phi node
builder.position_at_end(merge_block)
phi = builder.phi(ir.IntType(1), name="or.result")
for val, block in incoming_values:
phi.add_incoming(val, block)
logger.debug(f"Generated 'or' with {len(incoming_values)} incoming values")
return phi, ir.IntType(1)
def _handle_boolean_op(
func,
module,
builder,
expr: ast.BoolOp,
local_sym_tab,
map_sym_tab,
structs_sym_tab=None,
):
"""Handle `and` and `or` boolean operations."""
if isinstance(expr.op, ast.And):
return _handle_and_op(
func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
)
elif isinstance(expr.op, ast.Or):
return _handle_or_op(
func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
)
else:
logger.error(f"Unsupported boolean operator: {type(expr.op).__name__}")
return None
# ============================================================================
# Expression Dispatcher
# ============================================================================
def eval_expr(
func,
module,
builder,
expr,
local_sym_tab,
map_sym_tab,
structs_sym_tab=None,
):
logger.info(f"Evaluating expression: {ast.dump(expr)}")
if isinstance(expr, ast.Name):
return _handle_name_expr(expr, local_sym_tab, builder)
elif isinstance(expr, ast.Constant):
return _handle_constant_expr(module, builder, expr)
elif isinstance(expr, ast.Call):
if isinstance(expr.func, ast.Name) and expr.func.id == "deref":
return _handle_deref_call(expr, local_sym_tab, builder)
if isinstance(expr.func, ast.Name) and is_ctypes(expr.func.id):
return _handle_ctypes_call(
func,
module,
builder,
expr,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
result = CallHandlerRegistry.handle_call(
expr, module, builder, func, local_sym_tab, map_sym_tab, structs_sym_tab
)
if result is not None:
return result
logger.warning(f"Unknown call: {ast.dump(expr)}")
return None
elif isinstance(expr, ast.Attribute):
return _handle_attribute_expr(expr, local_sym_tab, structs_sym_tab, builder)
elif isinstance(expr, ast.BinOp):
return _handle_binary_op(
func,
module,
expr,
builder,
None,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
elif isinstance(expr, ast.Compare):
return _handle_compare(
func, module, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
)
elif isinstance(expr, ast.UnaryOp):
return _handle_unary_op(
func, module, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
)
elif isinstance(expr, ast.BoolOp):
return _handle_boolean_op(
func, module, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
)
logger.info("Unsupported expression evaluation")
return None
def handle_expr(
func,
module,
builder,
expr,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
):
"""Handle expression statements in the function body."""
logger.info(f"Handling expression: {ast.dump(expr)}")
call = expr.value
if isinstance(call, ast.Call):
eval_expr(
func,
module,
builder,
call,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
else:
logger.info("Unsupported expression type")

View File

@ -1,50 +0,0 @@
import logging
from llvmlite import ir
logger = logging.getLogger(__name__)
def deref_to_depth(func, builder, val, target_depth):
"""Dereference a pointer to a certain depth."""
cur_val = val
cur_type = val.type
for depth in range(target_depth):
if not isinstance(val.type, ir.PointerType):
logger.error("Cannot dereference further, non-pointer type")
return None
# dereference with null check
pointee_type = cur_type.pointee
null_check_block = builder.block
not_null_block = func.append_basic_block(name=f"deref_not_null_{depth}")
merge_block = func.append_basic_block(name=f"deref_merge_{depth}")
null_ptr = ir.Constant(cur_type, None)
is_not_null = builder.icmp_signed("!=", cur_val, null_ptr)
logger.debug(f"Inserted null check for pointer at depth {depth}")
builder.cbranch(is_not_null, not_null_block, merge_block)
builder.position_at_end(not_null_block)
dereferenced_val = builder.load(cur_val)
logger.debug(f"Dereferenced to depth {depth - 1}, type: {pointee_type}")
builder.branch(merge_block)
builder.position_at_end(merge_block)
phi = builder.phi(pointee_type, name=f"deref_result_{depth}")
zero_value = (
ir.Constant(pointee_type, 0)
if isinstance(pointee_type, ir.IntType)
else ir.Constant(pointee_type, None)
)
phi.add_incoming(zero_value, null_check_block)
phi.add_incoming(dereferenced_val, not_null_block)
# Continue with phi result
cur_val = phi
cur_type = pointee_type
return cur_val

View File

@ -1,83 +0,0 @@
import logging
import ast
from llvmlite import ir
from .ir_ops import deref_to_depth
logger = logging.getLogger(__name__)
COMPARISON_OPS = {
ast.Eq: "==",
ast.NotEq: "!=",
ast.Lt: "<",
ast.LtE: "<=",
ast.Gt: ">",
ast.GtE: ">=",
ast.Is: "==",
ast.IsNot: "!=",
}
def get_base_type_and_depth(ir_type):
"""Get the base type for pointer types."""
cur_type = ir_type
depth = 0
while isinstance(cur_type, ir.PointerType):
depth += 1
cur_type = cur_type.pointee
return cur_type, depth
def _normalize_types(func, builder, lhs, rhs):
"""Normalize types for comparison."""
logger.info(f"Normalizing types: {lhs.type} vs {rhs.type}")
if isinstance(lhs.type, ir.IntType) and isinstance(rhs.type, ir.IntType):
if lhs.type.width < rhs.type.width:
lhs = builder.sext(lhs, rhs.type)
else:
rhs = builder.sext(rhs, lhs.type)
return lhs, rhs
elif not isinstance(lhs.type, ir.PointerType) and not isinstance(
rhs.type, ir.PointerType
):
logger.error(f"Type mismatch: {lhs.type} vs {rhs.type}")
return None, None
else:
lhs_base, lhs_depth = get_base_type_and_depth(lhs.type)
rhs_base, rhs_depth = get_base_type_and_depth(rhs.type)
if lhs_base == rhs_base:
if lhs_depth < rhs_depth:
rhs = deref_to_depth(func, builder, rhs, rhs_depth - lhs_depth)
elif rhs_depth < lhs_depth:
lhs = deref_to_depth(func, builder, lhs, lhs_depth - rhs_depth)
return _normalize_types(func, builder, lhs, rhs)
def convert_to_bool(builder, val):
"""Convert a value to boolean."""
if val.type == ir.IntType(1):
return val
if isinstance(val.type, ir.PointerType):
zero = ir.Constant(val.type, None)
else:
zero = ir.Constant(val.type, 0)
return builder.icmp_signed("!=", val, zero)
def handle_comparator(func, builder, op, lhs, rhs):
"""Handle comparison operations."""
if lhs.type != rhs.type:
lhs, rhs = _normalize_types(func, builder, lhs, rhs)
if lhs is None or rhs is None:
return None
if type(op) not in COMPARISON_OPS:
logger.error(f"Unsupported comparison operator: {type(op)}")
return None
predicate = COMPARISON_OPS[type(op)]
result = builder.icmp_signed(predicate, lhs, rhs)
logger.debug(f"Comparison result: {result}")
return result, ir.IntType(1)

145
pythonbpf/expr_pass.py Normal file
View File

@ -0,0 +1,145 @@
import ast
from llvmlite import ir
def eval_expr(
func,
module,
builder,
expr,
local_sym_tab,
map_sym_tab,
structs_sym_tab=None,
):
print(f"Evaluating expression: {ast.dump(expr)}")
if isinstance(expr, ast.Name):
if expr.id in local_sym_tab:
var = local_sym_tab[expr.id].var
val = builder.load(var)
return val, local_sym_tab[expr.id].ir_type # return value and type
else:
print(f"Undefined variable {expr.id}")
return None
elif isinstance(expr, ast.Constant):
if isinstance(expr.value, int):
return ir.Constant(ir.IntType(64), expr.value), ir.IntType(64)
elif isinstance(expr.value, bool):
return ir.Constant(ir.IntType(1), int(expr.value)), ir.IntType(1)
else:
print("Unsupported constant type")
return None
elif isinstance(expr, ast.Call):
# delayed import to avoid circular dependency
from pythonbpf.helper import HelperHandlerRegistry, handle_helper_call
if isinstance(expr.func, ast.Name):
# check deref
if expr.func.id == "deref":
print(f"Handling deref {ast.dump(expr)}")
if len(expr.args) != 1:
print("deref takes exactly one argument")
return None
arg = expr.args[0]
if (
isinstance(arg, ast.Call)
and isinstance(arg.func, ast.Name)
and arg.func.id == "deref"
):
print("Multiple deref not supported")
return None
if isinstance(arg, ast.Name):
if arg.id in local_sym_tab:
arg = local_sym_tab[arg.id].var
else:
print(f"Undefined variable {arg.id}")
return None
if arg is None:
print("Failed to evaluate deref argument")
return None
# Since we are handling only name case, directly take type from sym tab
val = builder.load(arg)
return val, local_sym_tab[expr.args[0].id].ir_type
# check for helpers
if HelperHandlerRegistry.has_handler(expr.func.id):
return handle_helper_call(
expr,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
elif isinstance(expr.func, ast.Attribute):
print(f"Handling method call: {ast.dump(expr.func)}")
if isinstance(expr.func.value, ast.Call) and isinstance(
expr.func.value.func, ast.Name
):
method_name = expr.func.attr
if HelperHandlerRegistry.has_handler(method_name):
return handle_helper_call(
expr,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
elif isinstance(expr.func.value, ast.Name):
obj_name = expr.func.value.id
method_name = expr.func.attr
if obj_name in map_sym_tab:
if HelperHandlerRegistry.has_handler(method_name):
return handle_helper_call(
expr,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
elif isinstance(expr, ast.Attribute):
if isinstance(expr.value, ast.Name):
var_name = expr.value.id
attr_name = expr.attr
if var_name in local_sym_tab:
var_ptr, var_type, var_metadata = local_sym_tab[var_name]
print(f"Loading attribute {attr_name} from variable {var_name}")
print(f"Variable type: {var_type}, Variable ptr: {var_ptr}")
metadata = structs_sym_tab[var_metadata]
if attr_name in metadata.fields:
gep = metadata.gep(builder, var_ptr, attr_name)
val = builder.load(gep)
field_type = metadata.field_type(attr_name)
return val, field_type
print("Unsupported expression evaluation")
return None
def handle_expr(
func,
module,
builder,
expr,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
):
"""Handle expression statements in the function body."""
print(f"Handling expression: {ast.dump(expr)}")
call = expr.value
if isinstance(call, ast.Call):
eval_expr(
func,
module,
builder,
call,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
else:
print("Unsupported expression type")

View File

@ -1,3 +0,0 @@
from .functions_pass import func_proc
__all__ = ["func_proc"]

View File

@ -1,88 +0,0 @@
import ast
def get_probe_string(func_node):
"""Extract the probe string from the decorator of the function node"""
# TODO: right now we have the whole string in the section decorator
# But later we can implement typed tuples for tracepoints and kprobes
# For helper functions, we return "helper"
for decorator in func_node.decorator_list:
if isinstance(decorator, ast.Name) and decorator.id == "bpfglobal":
return None
if isinstance(decorator, ast.Call) and isinstance(decorator.func, ast.Name):
if decorator.func.id == "section" and len(decorator.args) == 1:
arg = decorator.args[0]
if isinstance(arg, ast.Constant) and isinstance(arg.value, str):
return arg.value
return "helper"
def is_global_function(func_node):
"""Check if the function is a global"""
for decorator in func_node.decorator_list:
if isinstance(decorator, ast.Name) and decorator.id in (
"map",
"bpfglobal",
"struct",
):
return True
return False
def infer_return_type(func_node: ast.FunctionDef):
if not isinstance(func_node, (ast.FunctionDef, ast.AsyncFunctionDef)):
raise TypeError("Expected ast.FunctionDef")
if func_node.returns is not None:
try:
return ast.unparse(func_node.returns)
except Exception:
node = func_node.returns
if isinstance(node, ast.Name):
return node.id
if isinstance(node, ast.Attribute):
return getattr(node, "attr", type(node).__name__)
try:
return str(node)
except Exception:
return type(node).__name__
found_type = None
def _expr_type(e):
if e is None:
return "None"
if isinstance(e, ast.Constant):
return type(e.value).__name__
if isinstance(e, ast.Name):
return e.id
if isinstance(e, ast.Call):
f = e.func
if isinstance(f, ast.Name):
return f.id
if isinstance(f, ast.Attribute):
try:
return ast.unparse(f)
except Exception:
return getattr(f, "attr", type(f).__name__)
try:
return ast.unparse(f)
except Exception:
return type(f).__name__
if isinstance(e, ast.Attribute):
try:
return ast.unparse(e)
except Exception:
return getattr(e, "attr", type(e).__name__)
try:
return ast.unparse(e)
except Exception:
return type(e).__name__
for walked_node in ast.walk(func_node):
if isinstance(walked_node, ast.Return):
t = _expr_type(walked_node.value)
if found_type is None:
found_type = t
elif found_type != t:
raise ValueError(f"Conflicting return types: {found_type} vs {t}")
return found_type or "None"

View File

@ -1,456 +0,0 @@
from llvmlite import ir
import ast
import logging
from pythonbpf.helper import (
HelperHandlerRegistry,
reset_scratch_pool,
)
from pythonbpf.type_deducer import ctypes_to_ir
from pythonbpf.expr import eval_expr, handle_expr, convert_to_bool
from pythonbpf.assign_pass import (
handle_variable_assignment,
handle_struct_field_assignment,
)
from pythonbpf.allocation_pass import handle_assign_allocation, allocate_temp_pool
from .return_utils import handle_none_return, handle_xdp_return, is_xdp_name
from .function_metadata import get_probe_string, is_global_function, infer_return_type
logger = logging.getLogger(__name__)
# ============================================================================
# SECTION 1: Memory Allocation
# ============================================================================
def count_temps_in_call(call_node, local_sym_tab):
"""Count the number of temporary variables needed for a function call."""
count = 0
is_helper = False
# NOTE: We exclude print calls for now
if isinstance(call_node.func, ast.Name):
if (
HelperHandlerRegistry.has_handler(call_node.func.id)
and call_node.func.id != "print"
):
is_helper = True
elif isinstance(call_node.func, ast.Attribute):
if HelperHandlerRegistry.has_handler(call_node.func.attr):
is_helper = True
if not is_helper:
return 0
for arg in call_node.args:
# NOTE: Count all non-name arguments
# For struct fields, if it is being passed as an argument,
# The struct object should already exist in the local_sym_tab
if not isinstance(arg, ast.Name) and not (
isinstance(arg, ast.Attribute) and arg.value.id in local_sym_tab
):
count += 1
return count
def handle_if_allocation(
module, builder, stmt, func, ret_type, map_sym_tab, local_sym_tab, structs_sym_tab
):
"""Recursively handle allocations in if/else branches."""
if stmt.body:
allocate_mem(
module,
builder,
stmt.body,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
if stmt.orelse:
allocate_mem(
module,
builder,
stmt.orelse,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
def allocate_mem(
module, builder, body, func, ret_type, map_sym_tab, local_sym_tab, structs_sym_tab
):
max_temps_needed = 0
def update_max_temps_for_stmt(stmt):
nonlocal max_temps_needed
temps_needed = 0
if isinstance(stmt, ast.If):
for s in stmt.body:
update_max_temps_for_stmt(s)
for s in stmt.orelse:
update_max_temps_for_stmt(s)
return
for node in ast.walk(stmt):
if isinstance(node, ast.Call):
temps_needed += count_temps_in_call(node, local_sym_tab)
max_temps_needed = max(max_temps_needed, temps_needed)
for stmt in body:
update_max_temps_for_stmt(stmt)
# Handle allocations
if isinstance(stmt, ast.If):
handle_if_allocation(
module,
builder,
stmt,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
elif isinstance(stmt, ast.Assign):
handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab)
allocate_temp_pool(builder, max_temps_needed, local_sym_tab)
return local_sym_tab
# ============================================================================
# SECTION 2: Statement Handlers
# ============================================================================
def handle_assign(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab
):
"""Handle assignment statements in the function body."""
# TODO: Support this later
# GH #37
if len(stmt.targets) != 1:
logger.error("Multi-target assignment is not supported for now")
return
target = stmt.targets[0]
rval = stmt.value
if isinstance(target, ast.Name):
# NOTE: Simple variable assignment case: x = 5
var_name = target.id
result = handle_variable_assignment(
func,
module,
builder,
var_name,
rval,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
if not result:
logger.error(f"Failed to handle assignment to {var_name}")
return
if isinstance(target, ast.Attribute):
# NOTE: Struct field assignment case: pkt.field = value
handle_struct_field_assignment(
func,
module,
builder,
target,
rval,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
return
# Unsupported target type
logger.error(f"Unsupported assignment target: {ast.dump(target)}")
def handle_cond(
func, module, builder, cond, local_sym_tab, map_sym_tab, structs_sym_tab=None
):
val = eval_expr(
func, module, builder, cond, local_sym_tab, map_sym_tab, structs_sym_tab
)[0]
return convert_to_bool(builder, val)
def handle_if(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab=None
):
"""Handle if statements in the function body."""
logger.info("Handling if statement")
# start = builder.block.parent
then_block = func.append_basic_block(name="if.then")
merge_block = func.append_basic_block(name="if.end")
if stmt.orelse:
else_block = func.append_basic_block(name="if.else")
else:
else_block = None
cond = handle_cond(
func, module, builder, stmt.test, local_sym_tab, map_sym_tab, structs_sym_tab
)
if else_block:
builder.cbranch(cond, then_block, else_block)
else:
builder.cbranch(cond, then_block, merge_block)
builder.position_at_end(then_block)
for s in stmt.body:
process_stmt(
func, module, builder, s, local_sym_tab, map_sym_tab, structs_sym_tab, False
)
if not builder.block.is_terminated:
builder.branch(merge_block)
if else_block:
builder.position_at_end(else_block)
for s in stmt.orelse:
process_stmt(
func,
module,
builder,
s,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
False,
)
if not builder.block.is_terminated:
builder.branch(merge_block)
builder.position_at_end(merge_block)
def handle_return(builder, stmt, local_sym_tab, ret_type):
logger.info(f"Handling return statement: {ast.dump(stmt)}")
if stmt.value is None:
return handle_none_return(builder)
elif isinstance(stmt.value, ast.Name) and is_xdp_name(stmt.value.id):
return handle_xdp_return(stmt, builder, ret_type)
else:
val = eval_expr(
func=None,
module=None,
builder=builder,
expr=stmt.value,
local_sym_tab=local_sym_tab,
map_sym_tab={},
structs_sym_tab={},
)
logger.info(f"Evaluated return expression to {val}")
builder.ret(val[0])
return True
def process_stmt(
func,
module,
builder,
stmt,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
did_return,
ret_type=ir.IntType(64),
):
logger.info(f"Processing statement: {ast.dump(stmt)}")
reset_scratch_pool()
if isinstance(stmt, ast.Expr):
handle_expr(
func,
module,
builder,
stmt,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
elif isinstance(stmt, ast.Assign):
handle_assign(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab
)
elif isinstance(stmt, ast.AugAssign):
raise SyntaxError("Augmented assignment not supported")
elif isinstance(stmt, ast.If):
handle_if(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab
)
elif isinstance(stmt, ast.Return):
did_return = handle_return(
builder,
stmt,
local_sym_tab,
ret_type,
)
return did_return
# ============================================================================
# SECTION 3: Function Body Processing
# ============================================================================
def process_func_body(
module, builder, func_node, func, ret_type, map_sym_tab, structs_sym_tab
):
"""Process the body of a bpf function"""
# TODO: A lot. We just have print -> bpf_trace_printk for now
did_return = False
local_sym_tab = {}
# pre-allocate dynamic variables
local_sym_tab = allocate_mem(
module,
builder,
func_node.body,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
logger.info(f"Local symbol table: {local_sym_tab.keys()}")
for stmt in func_node.body:
did_return = process_stmt(
func,
module,
builder,
stmt,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
did_return,
ret_type,
)
if not did_return:
builder.ret(ir.Constant(ir.IntType(64), 0))
def process_bpf_chunk(func_node, module, return_type, map_sym_tab, structs_sym_tab):
"""Process a single BPF chunk (function) and emit corresponding LLVM IR."""
func_name = func_node.name
ret_type = return_type
# TODO: parse parameters
param_types = []
if func_node.args.args:
# Assume first arg to be ctx
param_types.append(ir.PointerType())
func_ty = ir.FunctionType(ret_type, param_types)
func = ir.Function(module, func_ty, func_name)
func.linkage = "dso_local"
func.attributes.add("nounwind")
func.attributes.add("noinline")
func.attributes.add("optnone")
if func_node.args.args:
# Only look at the first argument for now
param = func.args[0]
param.add_attribute("nocapture")
probe_string = get_probe_string(func_node)
if probe_string is not None:
func.section = probe_string
block = func.append_basic_block(name="entry")
builder = ir.IRBuilder(block)
process_func_body(
module, builder, func_node, func, ret_type, map_sym_tab, structs_sym_tab
)
return func
# ============================================================================
# SECTION 4: Top-Level Function Processor
# ============================================================================
def func_proc(tree, module, chunks, map_sym_tab, structs_sym_tab):
for func_node in chunks:
if is_global_function(func_node):
continue
func_type = get_probe_string(func_node)
logger.info(f"Found probe_string of {func_node.name}: {func_type}")
process_bpf_chunk(
func_node,
module,
ctypes_to_ir(infer_return_type(func_node)),
map_sym_tab,
structs_sym_tab,
)
# TODO: WIP, for string assignment to fixed-size arrays
def assign_string_to_array(builder, target_array_ptr, source_string_ptr, array_length):
"""
Copy a string (i8*) to a fixed-size array ([N x i8]*)
"""
# Create a loop to copy characters one by one
# entry_block = builder.block
copy_block = builder.append_basic_block("copy_char")
end_block = builder.append_basic_block("copy_end")
# Create loop counter
i = builder.alloca(ir.IntType(32))
builder.store(ir.Constant(ir.IntType(32), 0), i)
# Start the loop
builder.branch(copy_block)
# Copy loop
builder.position_at_end(copy_block)
idx = builder.load(i)
in_bounds = builder.icmp_unsigned(
"<", idx, ir.Constant(ir.IntType(32), array_length)
)
builder.cbranch(in_bounds, copy_block, end_block)
with builder.if_then(in_bounds):
# Load character from source
src_ptr = builder.gep(source_string_ptr, [idx])
char = builder.load(src_ptr)
# Store character in target
dst_ptr = builder.gep(target_array_ptr, [ir.Constant(ir.IntType(32), 0), idx])
builder.store(char, dst_ptr)
# Increment counter
next_idx = builder.add(idx, ir.Constant(ir.IntType(32), 1))
builder.store(next_idx, i)
builder.position_at_end(end_block)
# Ensure null termination
last_idx = ir.Constant(ir.IntType(32), array_length - 1)
null_ptr = builder.gep(target_array_ptr, [ir.Constant(ir.IntType(32), 0), last_idx])
builder.store(ir.Constant(ir.IntType(8), 0), null_ptr)

View File

@ -1,44 +0,0 @@
import logging
import ast
from llvmlite import ir
logger: logging.Logger = logging.getLogger(__name__)
XDP_ACTIONS = {
"XDP_ABORTED": 0,
"XDP_DROP": 1,
"XDP_PASS": 2,
"XDP_TX": 3,
"XDP_REDIRECT": 4,
}
def handle_none_return(builder) -> bool:
"""Handle return or return None -> returns 0."""
builder.ret(ir.Constant(ir.IntType(64), 0))
logger.debug("Generated default return: 0")
return True
def is_xdp_name(name: str) -> bool:
"""Check if a name is an XDP action"""
return name in XDP_ACTIONS
def handle_xdp_return(stmt: ast.Return, builder, ret_type) -> bool:
"""Handle XDP returns"""
if not isinstance(stmt.value, ast.Name):
return False
action_name = stmt.value.id
if action_name not in XDP_ACTIONS:
raise ValueError(
f"Unknown XDP action: {action_name}. Available: {XDP_ACTIONS.keys()}"
)
value = XDP_ACTIONS[action_name]
builder.ret(ir.Constant(ret_type, value))
logger.debug(f"Generated XDP action return: {action_name} = {value}")
return True

726
pythonbpf/functions_pass.py Normal file
View File

@ -0,0 +1,726 @@
from llvmlite import ir
import ast
import logging
from typing import Any
from dataclasses import dataclass
from .helper import HelperHandlerRegistry, handle_helper_call
from .type_deducer import ctypes_to_ir
from .binary_ops import handle_binary_op
from .expr_pass import eval_expr, handle_expr
logger = logging.getLogger(__name__)
@dataclass
class LocalSymbol:
var: ir.AllocaInstr
ir_type: ir.Type
metadata: Any = None
def __iter__(self):
yield self.var
yield self.ir_type
yield self.metadata
def get_probe_string(func_node):
"""Extract the probe string from the decorator of the function node."""
# TODO: right now we have the whole string in the section decorator
# But later we can implement typed tuples for tracepoints and kprobes
# For helper functions, we return "helper"
for decorator in func_node.decorator_list:
if isinstance(decorator, ast.Name) and decorator.id == "bpfglobal":
return None
if isinstance(decorator, ast.Call) and isinstance(decorator.func, ast.Name):
if decorator.func.id == "section" and len(decorator.args) == 1:
arg = decorator.args[0]
if isinstance(arg, ast.Constant) and isinstance(arg.value, str):
return arg.value
return "helper"
def handle_assign(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab
):
"""Handle assignment statements in the function body."""
if len(stmt.targets) != 1:
print("Unsupported multiassignment")
return
num_types = ("c_int32", "c_int64", "c_uint32", "c_uint64")
target = stmt.targets[0]
print(f"Handling assignment to {ast.dump(target)}")
if not isinstance(target, ast.Name) and not isinstance(target, ast.Attribute):
print("Unsupported assignment target")
return
var_name = target.id if isinstance(target, ast.Name) else target.value.id
rval = stmt.value
if isinstance(target, ast.Attribute):
# struct field assignment
field_name = target.attr
if var_name in local_sym_tab:
struct_type = local_sym_tab[var_name].metadata
struct_info = structs_sym_tab[struct_type]
if field_name in struct_info.fields:
field_ptr = struct_info.gep(
builder, local_sym_tab[var_name].var, field_name
)
val = eval_expr(
func,
module,
builder,
rval,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
if isinstance(struct_info.field_type(field_name), ir.ArrayType) and val[
1
] == ir.PointerType(ir.IntType(8)):
# TODO: Figure it out, not a priority rn
# Special case for string assignment to char array
# str_len = struct_info["field_types"][field_idx].count
# assign_string_to_array(builder, field_ptr, val[0], str_len)
# print(f"Assigned to struct field {var_name}.{field_name}")
pass
if val is None:
print("Failed to evaluate struct field assignment")
return
print(field_ptr)
builder.store(val[0], field_ptr)
print(f"Assigned to struct field {var_name}.{field_name}")
return
elif isinstance(rval, ast.Constant):
if isinstance(rval.value, bool):
if rval.value:
builder.store(
ir.Constant(ir.IntType(1), 1), local_sym_tab[var_name].var
)
else:
builder.store(
ir.Constant(ir.IntType(1), 0), local_sym_tab[var_name].var
)
print(f"Assigned constant {rval.value} to {var_name}")
elif isinstance(rval.value, int):
# Assume c_int64 for now
# var = builder.alloca(ir.IntType(64), name=var_name)
# var.align = 8
builder.store(
ir.Constant(ir.IntType(64), rval.value), local_sym_tab[var_name].var
)
print(f"Assigned constant {rval.value} to {var_name}")
elif isinstance(rval.value, str):
str_val = rval.value.encode("utf-8") + b"\x00"
str_const = ir.Constant(
ir.ArrayType(ir.IntType(8), len(str_val)), bytearray(str_val)
)
global_str = ir.GlobalVariable(
module, str_const.type, name=f"{var_name}_str"
)
global_str.linkage = "internal"
global_str.global_constant = True
global_str.initializer = str_const
str_ptr = builder.bitcast(global_str, ir.PointerType(ir.IntType(8)))
builder.store(str_ptr, local_sym_tab[var_name].var)
print(f"Assigned string constant '{rval.value}' to {var_name}")
else:
print("Unsupported constant type")
elif isinstance(rval, ast.Call):
if isinstance(rval.func, ast.Name):
call_type = rval.func.id
print(f"Assignment call type: {call_type}")
if (
call_type in num_types
and len(rval.args) == 1
and isinstance(rval.args[0], ast.Constant)
and isinstance(rval.args[0].value, int)
):
ir_type = ctypes_to_ir(call_type)
# var = builder.alloca(ir_type, name=var_name)
# var.align = ir_type.width // 8
builder.store(
ir.Constant(ir_type, rval.args[0].value),
local_sym_tab[var_name].var,
)
print(
f"Assigned {call_type} constant "
f"{rval.args[0].value} to {var_name}"
)
elif HelperHandlerRegistry.has_handler(call_type):
# var = builder.alloca(ir.IntType(64), name=var_name)
# var.align = 8
val = handle_helper_call(
rval,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
builder.store(val[0], local_sym_tab[var_name].var)
print(f"Assigned constant {rval.func.id} to {var_name}")
elif call_type == "deref" and len(rval.args) == 1:
print(f"Handling deref assignment {ast.dump(rval)}")
val = eval_expr(
func,
module,
builder,
rval,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
if val is None:
print("Failed to evaluate deref argument")
return
print(f"Dereferenced value: {val}, storing in {var_name}")
builder.store(val[0], local_sym_tab[var_name].var)
print(f"Dereferenced and assigned to {var_name}")
elif call_type in structs_sym_tab and len(rval.args) == 0:
struct_info = structs_sym_tab[call_type]
ir_type = struct_info.ir_type
# var = builder.alloca(ir_type, name=var_name)
# Null init
builder.store(ir.Constant(ir_type, None), local_sym_tab[var_name].var)
print(f"Assigned struct {call_type} to {var_name}")
else:
print(f"Unsupported assignment call type: {call_type}")
elif isinstance(rval.func, ast.Attribute):
print(f"Assignment call attribute: {ast.dump(rval.func)}")
if isinstance(rval.func.value, ast.Name):
# TODO: probably a struct access
print(f"TODO STRUCT ACCESS {ast.dump(rval)}")
elif isinstance(rval.func.value, ast.Call) and isinstance(
rval.func.value.func, ast.Name
):
map_name = rval.func.value.func.id
method_name = rval.func.attr
if map_name in map_sym_tab:
if HelperHandlerRegistry.has_handler(method_name):
val = handle_helper_call(
rval,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
# var = builder.alloca(ir.IntType(64), name=var_name)
# var.align = 8
builder.store(val[0], local_sym_tab[var_name].var)
else:
print("Unsupported assignment call structure")
else:
print("Unsupported assignment call function type")
elif isinstance(rval, ast.BinOp):
handle_binary_op(
rval, module, builder, var_name, local_sym_tab, map_sym_tab, func
)
else:
print("Unsupported assignment value type")
def handle_cond(func, module, builder, cond, local_sym_tab, map_sym_tab):
if isinstance(cond, ast.Constant):
if isinstance(cond.value, bool):
return ir.Constant(ir.IntType(1), int(cond.value))
elif isinstance(cond.value, int):
return ir.Constant(ir.IntType(1), int(bool(cond.value)))
else:
print("Unsupported constant type in condition")
return None
elif isinstance(cond, ast.Name):
if cond.id in local_sym_tab:
var = local_sym_tab[cond.id].var
val = builder.load(var)
if val.type != ir.IntType(1):
# Convert nonzero values to true, zero to false
if isinstance(val.type, ir.PointerType):
# For pointer types, compare with null pointer
zero = ir.Constant(val.type, None)
else:
# For integer types, compare with zero
zero = ir.Constant(val.type, 0)
val = builder.icmp_signed("!=", val, zero)
return val
else:
print(f"Undefined variable {cond.id} in condition")
return None
elif isinstance(cond, ast.Compare):
lhs = eval_expr(func, module, builder, cond.left, local_sym_tab, map_sym_tab)[0]
if len(cond.ops) != 1 or len(cond.comparators) != 1:
print("Unsupported complex comparison")
return None
rhs = eval_expr(
func, module, builder, cond.comparators[0], local_sym_tab, map_sym_tab
)[0]
op = cond.ops[0]
if lhs.type != rhs.type:
if isinstance(lhs.type, ir.IntType) and isinstance(rhs.type, ir.IntType):
# Extend the smaller type to the larger type
if lhs.type.width < rhs.type.width:
lhs = builder.sext(lhs, rhs.type)
elif lhs.type.width > rhs.type.width:
rhs = builder.sext(rhs, lhs.type)
else:
print("Type mismatch in comparison")
return None
if isinstance(op, ast.Eq):
return builder.icmp_signed("==", lhs, rhs)
elif isinstance(op, ast.NotEq):
return builder.icmp_signed("!=", lhs, rhs)
elif isinstance(op, ast.Lt):
return builder.icmp_signed("<", lhs, rhs)
elif isinstance(op, ast.LtE):
return builder.icmp_signed("<=", lhs, rhs)
elif isinstance(op, ast.Gt):
return builder.icmp_signed(">", lhs, rhs)
elif isinstance(op, ast.GtE):
return builder.icmp_signed(">=", lhs, rhs)
else:
print("Unsupported comparison operator")
return None
else:
print("Unsupported condition expression")
return None
def handle_if(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab=None
):
"""Handle if statements in the function body."""
print("Handling if statement")
# start = builder.block.parent
then_block = func.append_basic_block(name="if.then")
merge_block = func.append_basic_block(name="if.end")
if stmt.orelse:
else_block = func.append_basic_block(name="if.else")
else:
else_block = None
cond = handle_cond(func, module, builder, stmt.test, local_sym_tab, map_sym_tab)
if else_block:
builder.cbranch(cond, then_block, else_block)
else:
builder.cbranch(cond, then_block, merge_block)
builder.position_at_end(then_block)
for s in stmt.body:
process_stmt(
func, module, builder, s, local_sym_tab, map_sym_tab, structs_sym_tab, False
)
if not builder.block.is_terminated:
builder.branch(merge_block)
if else_block:
builder.position_at_end(else_block)
for s in stmt.orelse:
process_stmt(
func,
module,
builder,
s,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
False,
)
if not builder.block.is_terminated:
builder.branch(merge_block)
builder.position_at_end(merge_block)
def process_stmt(
func,
module,
builder,
stmt,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
did_return,
ret_type=ir.IntType(64),
):
print(f"Processing statement: {ast.dump(stmt)}")
if isinstance(stmt, ast.Expr):
handle_expr(
func,
module,
builder,
stmt,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
elif isinstance(stmt, ast.Assign):
handle_assign(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab
)
elif isinstance(stmt, ast.AugAssign):
raise SyntaxError("Augmented assignment not supported")
elif isinstance(stmt, ast.If):
handle_if(
func, module, builder, stmt, map_sym_tab, local_sym_tab, structs_sym_tab
)
elif isinstance(stmt, ast.Return):
if stmt.value is None:
builder.ret(ir.Constant(ir.IntType(32), 0))
did_return = True
elif (
isinstance(stmt.value, ast.Call)
and isinstance(stmt.value.func, ast.Name)
and len(stmt.value.args) == 1
and isinstance(stmt.value.args[0], ast.Constant)
and isinstance(stmt.value.args[0].value, int)
):
call_type = stmt.value.func.id
if ctypes_to_ir(call_type) != ret_type:
raise ValueError(
"Return type mismatch: expected"
f"{ctypes_to_ir(call_type)}, got {call_type}"
)
else:
builder.ret(ir.Constant(ret_type, stmt.value.args[0].value))
did_return = True
elif isinstance(stmt.value, ast.Name):
if stmt.value.id == "XDP_PASS":
builder.ret(ir.Constant(ret_type, 2))
did_return = True
elif stmt.value.id == "XDP_DROP":
builder.ret(ir.Constant(ret_type, 1))
did_return = True
else:
raise ValueError("Failed to evaluate return expression")
else:
raise ValueError("Unsupported return value")
return did_return
def allocate_mem(
module, builder, body, func, ret_type, map_sym_tab, local_sym_tab, structs_sym_tab
):
for stmt in body:
has_metadata = False
if isinstance(stmt, ast.If):
if stmt.body:
local_sym_tab = allocate_mem(
module,
builder,
stmt.body,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
if stmt.orelse:
local_sym_tab = allocate_mem(
module,
builder,
stmt.orelse,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
elif isinstance(stmt, ast.Assign):
if len(stmt.targets) != 1:
print("Unsupported multiassignment")
continue
target = stmt.targets[0]
if not isinstance(target, ast.Name):
print("Unsupported assignment target")
continue
var_name = target.id
rval = stmt.value
if isinstance(rval, ast.Call):
if isinstance(rval.func, ast.Name):
call_type = rval.func.id
if call_type in ("c_int32", "c_int64", "c_uint32", "c_uint64"):
ir_type = ctypes_to_ir(call_type)
var = builder.alloca(ir_type, name=var_name)
var.align = ir_type.width // 8
print(f"Pre-allocated variable {var_name} of type {call_type}")
elif HelperHandlerRegistry.has_handler(call_type):
# Assume return type is int64 for now
ir_type = ir.IntType(64)
var = builder.alloca(ir_type, name=var_name)
var.align = ir_type.width // 8
print(f"Pre-allocated variable {var_name} for helper")
elif call_type == "deref" and len(rval.args) == 1:
# Assume return type is int64 for now
ir_type = ir.IntType(64)
var = builder.alloca(ir_type, name=var_name)
var.align = ir_type.width // 8
print(f"Pre-allocated variable {var_name} for deref")
elif call_type in structs_sym_tab:
struct_info = structs_sym_tab[call_type]
ir_type = struct_info.ir_type
var = builder.alloca(ir_type, name=var_name)
has_metadata = True
print(
f"Pre-allocated variable {var_name} "
f"for struct {call_type}"
)
elif isinstance(rval.func, ast.Attribute):
ir_type = ir.PointerType(ir.IntType(64))
var = builder.alloca(ir_type, name=var_name)
# var.align = ir_type.width // 8
print(f"Pre-allocated variable {var_name} for map")
else:
print("Unsupported assignment call function type")
continue
elif isinstance(rval, ast.Constant):
if isinstance(rval.value, bool):
ir_type = ir.IntType(1)
var = builder.alloca(ir_type, name=var_name)
var.align = 1
print(f"Pre-allocated variable {var_name} of type c_bool")
elif isinstance(rval.value, int):
# Assume c_int64 for now
ir_type = ir.IntType(64)
var = builder.alloca(ir_type, name=var_name)
var.align = ir_type.width // 8
print(f"Pre-allocated variable {var_name} of type c_int64")
elif isinstance(rval.value, str):
ir_type = ir.PointerType(ir.IntType(8))
var = builder.alloca(ir_type, name=var_name)
var.align = 8
print(f"Pre-allocated variable {var_name} of type string")
else:
print("Unsupported constant type")
continue
elif isinstance(rval, ast.BinOp):
# Assume c_int64 for now
ir_type = ir.IntType(64)
var = builder.alloca(ir_type, name=var_name)
var.align = ir_type.width // 8
print(f"Pre-allocated variable {var_name} of type c_int64")
else:
print("Unsupported assignment value type")
continue
if has_metadata:
local_sym_tab[var_name] = LocalSymbol(var, ir_type, call_type)
else:
local_sym_tab[var_name] = LocalSymbol(var, ir_type)
return local_sym_tab
def process_func_body(
module, builder, func_node, func, ret_type, map_sym_tab, structs_sym_tab
):
"""Process the body of a bpf function"""
# TODO: A lot. We just have print -> bpf_trace_printk for now
did_return = False
local_sym_tab = {}
# pre-allocate dynamic variables
local_sym_tab = allocate_mem(
module,
builder,
func_node.body,
func,
ret_type,
map_sym_tab,
local_sym_tab,
structs_sym_tab,
)
print(f"Local symbol table: {local_sym_tab.keys()}")
for stmt in func_node.body:
did_return = process_stmt(
func,
module,
builder,
stmt,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
did_return,
ret_type,
)
if not did_return:
builder.ret(ir.Constant(ir.IntType(32), 0))
def process_bpf_chunk(func_node, module, return_type, map_sym_tab, structs_sym_tab):
"""Process a single BPF chunk (function) and emit corresponding LLVM IR."""
func_name = func_node.name
ret_type = return_type
# TODO: parse parameters
param_types = []
if func_node.args.args:
# Assume first arg to be ctx
param_types.append(ir.PointerType())
func_ty = ir.FunctionType(ret_type, param_types)
func = ir.Function(module, func_ty, func_name)
func.linkage = "dso_local"
func.attributes.add("nounwind")
func.attributes.add("noinline")
func.attributes.add("optnone")
if func_node.args.args:
# Only look at the first argument for now
param = func.args[0]
param.add_attribute("nocapture")
probe_string = get_probe_string(func_node)
if probe_string is not None:
func.section = probe_string
block = func.append_basic_block(name="entry")
builder = ir.IRBuilder(block)
process_func_body(
module, builder, func_node, func, ret_type, map_sym_tab, structs_sym_tab
)
return func
def func_proc(tree, module, chunks, map_sym_tab, structs_sym_tab):
for func_node in chunks:
is_global = False
for decorator in func_node.decorator_list:
if isinstance(decorator, ast.Name) and decorator.id in (
"map",
"bpfglobal",
"struct",
):
is_global = True
break
if is_global:
continue
func_type = get_probe_string(func_node)
print(f"Found probe_string of {func_node.name}: {func_type}")
process_bpf_chunk(
func_node,
module,
ctypes_to_ir(infer_return_type(func_node)),
map_sym_tab,
structs_sym_tab,
)
def infer_return_type(func_node: ast.FunctionDef):
if not isinstance(func_node, (ast.FunctionDef, ast.AsyncFunctionDef)):
raise TypeError("Expected ast.FunctionDef")
if func_node.returns is not None:
try:
return ast.unparse(func_node.returns)
except Exception:
node = func_node.returns
if isinstance(node, ast.Name):
return node.id
if isinstance(node, ast.Attribute):
return getattr(node, "attr", type(node).__name__)
try:
return str(node)
except Exception:
return type(node).__name__
found_type = None
def _expr_type(e):
if e is None:
return "None"
if isinstance(e, ast.Constant):
return type(e.value).__name__
if isinstance(e, ast.Name):
return e.id
if isinstance(e, ast.Call):
f = e.func
if isinstance(f, ast.Name):
return f.id
if isinstance(f, ast.Attribute):
try:
return ast.unparse(f)
except Exception:
return getattr(f, "attr", type(f).__name__)
try:
return ast.unparse(f)
except Exception:
return type(f).__name__
if isinstance(e, ast.Attribute):
try:
return ast.unparse(e)
except Exception:
return getattr(e, "attr", type(e).__name__)
try:
return ast.unparse(e)
except Exception:
return type(e).__name__
for walked_node in ast.walk(func_node):
if isinstance(walked_node, ast.Return):
t = _expr_type(walked_node.value)
if found_type is None:
found_type = t
elif found_type != t:
raise ValueError(f"Conflicting return types: {found_type} vs {t}")
return found_type or "None"
# For string assignment to fixed-size arrays
def assign_string_to_array(builder, target_array_ptr, source_string_ptr, array_length):
"""
Copy a string (i8*) to a fixed-size array ([N x i8]*)
"""
# Create a loop to copy characters one by one
# entry_block = builder.block
copy_block = builder.append_basic_block("copy_char")
end_block = builder.append_basic_block("copy_end")
# Create loop counter
i = builder.alloca(ir.IntType(32))
builder.store(ir.Constant(ir.IntType(32), 0), i)
# Start the loop
builder.branch(copy_block)
# Copy loop
builder.position_at_end(copy_block)
idx = builder.load(i)
in_bounds = builder.icmp_unsigned(
"<", idx, ir.Constant(ir.IntType(32), array_length)
)
builder.cbranch(in_bounds, copy_block, end_block)
with builder.if_then(in_bounds):
# Load character from source
src_ptr = builder.gep(source_string_ptr, [idx])
char = builder.load(src_ptr)
# Store character in target
dst_ptr = builder.gep(target_array_ptr, [ir.Constant(ir.IntType(32), 0), idx])
builder.store(char, dst_ptr)
# Increment counter
next_idx = builder.add(idx, ir.Constant(ir.IntType(32), 1))
builder.store(next_idx, i)
builder.position_at_end(end_block)
# Ensure null termination
last_idx = ir.Constant(ir.IntType(32), array_length - 1)
null_ptr = builder.gep(target_array_ptr, [ir.Constant(ir.IntType(32), 0), last_idx])
builder.store(ir.Constant(ir.IntType(8), 0), null_ptr)

View File

@ -1,121 +1,8 @@
from llvmlite import ir
import ast
from logging import Logger
import logging
from .type_deducer import ctypes_to_ir
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]):
def emit_globals(module: ir.Module, names: list[str]):
"""
Emit the @llvm.compiler.used global given a list of function/global names.
"""
@ -137,7 +24,7 @@ def emit_llvm_compiler_used(module: ir.Module, names: list[str]):
gv.section = "llvm.metadata"
def globals_list_creation(tree, module: ir.Module):
def globals_processing(tree, module: ir.Module):
collected = ["LICENSE"]
for node in tree.body:
@ -153,11 +40,10 @@ def globals_list_creation(tree, module: ir.Module):
):
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 == "bpfglobal":
collected.append(node.name)
elif isinstance(dec, ast.Name) and dec.id == "map":
collected.append(node.name)
emit_llvm_compiler_used(module, collected)
emit_globals(module, collected)

View File

@ -1,63 +1,9 @@
from .helper_registry import HelperHandlerRegistry
from .helper_utils import reset_scratch_pool
from .helper_utils import HelperHandlerRegistry
from .bpf_helper_handler import handle_helper_call
from .helpers import ktime, pid, deref, XDP_DROP, XDP_PASS
# Register the helper handler with expr module
def _register_helper_handler():
"""Register helper call handler with the expression evaluator"""
from pythonbpf.expr.expr_pass import CallHandlerRegistry
def helper_call_handler(
call, module, builder, func, local_sym_tab, map_sym_tab, structs_sym_tab
):
"""Check if call is a helper and handle it"""
import ast
# Check for direct helper calls (e.g., ktime(), print())
if isinstance(call.func, ast.Name):
if HelperHandlerRegistry.has_handler(call.func.id):
return handle_helper_call(
call,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
# Check for method calls (e.g., map.lookup())
elif isinstance(call.func, ast.Attribute):
method_name = call.func.attr
# Handle: my_map.lookup(key)
if isinstance(call.func.value, ast.Name):
obj_name = call.func.value.id
if map_sym_tab and obj_name in map_sym_tab:
if HelperHandlerRegistry.has_handler(method_name):
return handle_helper_call(
call,
module,
builder,
func,
local_sym_tab,
map_sym_tab,
structs_sym_tab,
)
return None
CallHandlerRegistry.set_handler(helper_call_handler)
# Register on module import
_register_helper_handler()
__all__ = [
"HelperHandlerRegistry",
"reset_scratch_pool",
"handle_helper_call",
"ktime",
"pid",

View File

@ -1,19 +1,14 @@
import ast
from llvmlite import ir
from enum import Enum
from .helper_registry import HelperHandlerRegistry
from .helper_utils import (
HelperHandlerRegistry,
get_or_create_ptr_from_arg,
get_flags_val,
handle_fstring_print,
simple_string_print,
get_data_ptr_and_size,
)
from .printk_formatter import simple_string_print, handle_fstring_print
from logging import Logger
import logging
logger: Logger = logging.getLogger(__name__)
class BPFHelperID(Enum):
@ -35,7 +30,6 @@ def bpf_ktime_get_ns_emitter(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
"""
Emit LLVM IR for bpf_ktime_get_ns helper function call.
@ -58,26 +52,19 @@ def bpf_map_lookup_elem_emitter(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
"""
Emit LLVM IR for bpf_map_lookup_elem helper function call.
"""
if not call.args or len(call.args) != 1:
raise ValueError(
f"Map lookup expects exactly one argument (key), got {len(call.args)}"
"Map lookup expects exactly one argument (key), got " f"{len(call.args)}"
)
key_ptr = get_or_create_ptr_from_arg(
func, module, call.args[0], builder, local_sym_tab, map_sym_tab, struct_sym_tab
)
key_ptr = get_or_create_ptr_from_arg(call.args[0], builder, local_sym_tab)
map_void_ptr = builder.bitcast(map_ptr, ir.PointerType())
# TODO: I have changed the return type to i64*, as we are
# allocating space for that type in allocate_mem. This is
# temporary, and we will honour other widths later. But this
# allows us to have cool binary ops on the returned value.
fn_type = ir.FunctionType(
ir.PointerType(ir.IntType(64)), # Return type: void*
ir.PointerType(), # Return type: void*
[ir.PointerType(), ir.PointerType()], # Args: (void*, void*)
var_arg=False,
)
@ -100,7 +87,6 @@ def bpf_printk_emitter(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
"""Emit LLVM IR for bpf_printk helper function call."""
if not hasattr(func, "_fmt_counter"):
@ -136,7 +122,7 @@ def bpf_printk_emitter(
fn_ptr = builder.inttoptr(fn_addr, fn_ptr_type)
builder.call(fn_ptr, args, tail=True)
return True
return None
@HelperHandlerRegistry.register("update")
@ -148,7 +134,6 @@ def bpf_map_update_elem_emitter(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
"""
Emit LLVM IR for bpf_map_update_elem helper function call.
@ -156,19 +141,16 @@ def bpf_map_update_elem_emitter(
"""
if not call.args or len(call.args) < 2 or len(call.args) > 3:
raise ValueError(
f"Map update expects 2 or 3 args (key, value, flags), got {len(call.args)}"
"Map update expects 2 or 3 args (key, value, flags), "
f"got {len(call.args)}"
)
key_arg = call.args[0]
value_arg = call.args[1]
flags_arg = call.args[2] if len(call.args) > 2 else None
key_ptr = get_or_create_ptr_from_arg(
func, module, key_arg, builder, local_sym_tab, map_sym_tab, struct_sym_tab
)
value_ptr = get_or_create_ptr_from_arg(
func, module, value_arg, builder, local_sym_tab, map_sym_tab, struct_sym_tab
)
key_ptr = get_or_create_ptr_from_arg(key_arg, builder, local_sym_tab)
value_ptr = get_or_create_ptr_from_arg(value_arg, builder, local_sym_tab)
flags_val = get_flags_val(flags_arg, builder, local_sym_tab)
map_void_ptr = builder.bitcast(map_ptr, ir.PointerType())
@ -203,7 +185,6 @@ def bpf_map_delete_elem_emitter(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
"""
Emit LLVM IR for bpf_map_delete_elem helper function call.
@ -211,11 +192,9 @@ def bpf_map_delete_elem_emitter(
"""
if not call.args or len(call.args) != 1:
raise ValueError(
f"Map delete expects exactly one argument (key), got {len(call.args)}"
"Map delete expects exactly one argument (key), got " f"{len(call.args)}"
)
key_ptr = get_or_create_ptr_from_arg(
func, module, call.args[0], builder, local_sym_tab, map_sym_tab, struct_sym_tab
)
key_ptr = get_or_create_ptr_from_arg(call.args[0], builder, local_sym_tab)
map_void_ptr = builder.bitcast(map_ptr, ir.PointerType())
# Define function type for bpf_map_delete_elem
@ -243,7 +222,6 @@ def bpf_get_current_pid_tgid_emitter(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
"""
Emit LLVM IR for bpf_get_current_pid_tgid helper function call.
@ -270,11 +248,10 @@ def bpf_perf_event_output_handler(
func,
local_sym_tab=None,
struct_sym_tab=None,
map_sym_tab=None,
):
if len(call.args) != 1:
raise ValueError(
f"Perf event output expects exactly one argument, got {len(call.args)}"
"Perf event output expects exactly one argument, " f"got {len(call.args)}"
)
data_arg = call.args[0]
ctx_ptr = func.args[0] # First argument to the function is ctx
@ -335,7 +312,6 @@ def handle_helper_call(
func,
local_sym_tab,
struct_sym_tab,
map_sym_tab,
)
# Handle direct function calls (e.g., print(), ktime())
@ -346,7 +322,7 @@ def handle_helper_call(
elif isinstance(call.func, ast.Attribute):
method_name = call.func.attr
value = call.func.value
logger.info(f"Handling method call: {ast.dump(call.func)}")
print(f"Handling method call: {ast.dump(call.func)}")
# Get map pointer from different styles of map access
if isinstance(value, ast.Call) and isinstance(value.func, ast.Name):
# Func style: my_map().lookup(key)

View File

@ -1,27 +0,0 @@
from typing import Callable
class HelperHandlerRegistry:
"""Registry for BPF helpers"""
_handlers: dict[str, Callable] = {}
@classmethod
def register(cls, helper_name):
"""Decorator to register a handler function for a helper"""
def decorator(func):
cls._handlers[helper_name] = func
return func
return decorator
@classmethod
def get_handler(cls, helper_name):
"""Get the handler function for a helper"""
return cls._handlers.get(helper_name)
@classmethod
def has_handler(cls, helper_name):
"""Check if a handler function is registered for a helper"""
return helper_name in cls._handlers

View File

@ -1,52 +1,37 @@
import ast
import logging
from collections.abc import Callable
from llvmlite import ir
from pythonbpf.expr import (
get_operand_value,
)
from pythonbpf.expr_pass import eval_expr
logger = logging.getLogger(__name__)
class ScratchPoolManager:
"""Manage the temporary helper variables in local_sym_tab"""
class HelperHandlerRegistry:
"""Registry for BPF helpers"""
def __init__(self):
self._counter = 0
_handlers: dict[str, Callable] = {}
@property
def counter(self):
return self._counter
@classmethod
def register(cls, helper_name):
"""Decorator to register a handler function for a helper"""
def reset(self):
self._counter = 0
logger.debug("Scratch pool counter reset to 0")
def decorator(func):
cls._handlers[helper_name] = func
return func
def get_next_temp(self, local_sym_tab):
temp_name = f"__helper_temp_{self._counter}"
self._counter += 1
return decorator
if temp_name not in local_sym_tab:
raise ValueError(
f"Scratch pool exhausted or inadequate: {temp_name}. "
f"Current counter: {self._counter}"
)
@classmethod
def get_handler(cls, helper_name):
"""Get the handler function for a helper"""
return cls._handlers.get(helper_name)
return local_sym_tab[temp_name].var, temp_name
_temp_pool_manager = ScratchPoolManager() # Singleton instance
def reset_scratch_pool():
"""Reset the scratch pool counter"""
_temp_pool_manager.reset()
# ============================================================================
# Argument Preparation
# ============================================================================
@classmethod
def has_handler(cls, helper_name):
"""Check if a handler function is registered for a helper"""
return helper_name in cls._handlers
def get_var_ptr_from_name(var_name, local_sym_tab):
@ -56,41 +41,27 @@ def get_var_ptr_from_name(var_name, local_sym_tab):
raise ValueError(f"Variable '{var_name}' not found in local symbol table")
def create_int_constant_ptr(value, builder, local_sym_tab, int_width=64):
def create_int_constant_ptr(value, builder, int_width=64):
"""Create a pointer to an integer constant."""
# Default to 64-bit integer
ptr, temp_name = _temp_pool_manager.get_next_temp(local_sym_tab)
logger.info(f"Using temp variable '{temp_name}' for int constant {value}")
const_val = ir.Constant(ir.IntType(int_width), value)
builder.store(const_val, ptr)
int_type = ir.IntType(int_width)
ptr = builder.alloca(int_type)
ptr.align = int_type.width // 8
builder.store(ir.Constant(int_type, value), ptr)
return ptr
def get_or_create_ptr_from_arg(
func, module, arg, builder, local_sym_tab, map_sym_tab, struct_sym_tab=None
):
def get_or_create_ptr_from_arg(arg, builder, local_sym_tab):
"""Extract or create pointer from the call arguments."""
if isinstance(arg, ast.Name):
ptr = get_var_ptr_from_name(arg.id, local_sym_tab)
elif isinstance(arg, ast.Constant) and isinstance(arg.value, int):
ptr = create_int_constant_ptr(arg.value, builder, local_sym_tab)
ptr = create_int_constant_ptr(arg.value, builder)
else:
# Evaluate the expression and store the result in a temp variable
val = get_operand_value(
func, module, arg, builder, local_sym_tab, map_sym_tab, struct_sym_tab
raise NotImplementedError(
"Only simple variable names are supported as args in map helpers."
)
if val is None:
raise ValueError("Failed to evaluate expression for helper arg.")
# NOTE: We assume the result is an int64 for now
# if isinstance(arg, ast.Attribute):
# return val
ptr, temp_name = _temp_pool_manager.get_next_temp(local_sym_tab)
logger.info(f"Using temp variable '{temp_name}' for expression result")
builder.store(val, ptr)
return ptr
@ -113,6 +84,204 @@ def get_flags_val(arg, builder, local_sym_tab):
)
def simple_string_print(string_value, module, builder, func):
"""Prepare arguments for bpf_printk from a simple string value"""
fmt_str = string_value + "\n\0"
fmt_ptr = _create_format_string_global(fmt_str, func, module, builder)
args = [fmt_ptr, ir.Constant(ir.IntType(32), len(fmt_str))]
return args
def handle_fstring_print(
joined_str,
module,
builder,
func,
local_sym_tab=None,
struct_sym_tab=None,
):
"""Handle f-string formatting for bpf_printk emitter."""
fmt_parts = []
exprs = []
for value in joined_str.values:
logger.debug(f"Processing f-string value: {ast.dump(value)}")
if isinstance(value, ast.Constant):
_process_constant_in_fstring(value, fmt_parts, exprs)
elif isinstance(value, ast.FormattedValue):
_process_fval(
value,
fmt_parts,
exprs,
local_sym_tab,
struct_sym_tab,
)
else:
raise NotImplementedError(f"Unsupported f-string value type: {type(value)}")
fmt_str = "".join(fmt_parts)
args = simple_string_print(fmt_str, module, builder, func)
# NOTE: Process expressions (limited to 3 due to BPF constraints)
if len(exprs) > 3:
logger.warning("bpf_printk supports up to 3 args, extra args will be ignored.")
for expr in exprs[:3]:
arg_value = _prepare_expr_args(
expr,
func,
module,
builder,
local_sym_tab,
struct_sym_tab,
)
args.append(arg_value)
return args
def _process_constant_in_fstring(cst, fmt_parts, exprs):
"""Process constant values in f-string."""
if isinstance(cst.value, str):
fmt_parts.append(cst.value)
elif isinstance(cst.value, int):
fmt_parts.append("%lld")
exprs.append(ir.Constant(ir.IntType(64), cst.value))
else:
raise NotImplementedError(
f"Unsupported constant type in f-string: {type(cst.value)}"
)
def _process_fval(fval, fmt_parts, exprs, local_sym_tab, struct_sym_tab):
"""Process formatted values in f-string."""
logger.debug(f"Processing formatted value: {ast.dump(fval)}")
if isinstance(fval.value, ast.Name):
_process_name_in_fval(fval.value, fmt_parts, exprs, local_sym_tab)
elif isinstance(fval.value, ast.Attribute):
_process_attr_in_fval(
fval.value,
fmt_parts,
exprs,
local_sym_tab,
struct_sym_tab,
)
else:
raise NotImplementedError(
f"Unsupported formatted value in f-string: {type(fval.value)}"
)
def _process_name_in_fval(name_node, fmt_parts, exprs, local_sym_tab):
"""Process name nodes in formatted values."""
if local_sym_tab and name_node.id in local_sym_tab:
_, var_type, tmp = local_sym_tab[name_node.id]
_populate_fval(var_type, name_node, fmt_parts, exprs)
def _process_attr_in_fval(attr_node, fmt_parts, exprs, local_sym_tab, struct_sym_tab):
"""Process attribute nodes in formatted values."""
if (
isinstance(attr_node.value, ast.Name)
and local_sym_tab
and attr_node.value.id in local_sym_tab
):
var_name = attr_node.value.id
field_name = attr_node.attr
var_type = local_sym_tab[var_name].metadata
if var_type not in struct_sym_tab:
raise ValueError(
f"Struct '{var_type}' for '{var_name}' not in symbol table"
)
struct_info = struct_sym_tab[var_type]
if field_name not in struct_info.fields:
raise ValueError(f"Field '{field_name}' not found in struct '{var_type}'")
field_type = struct_info.field_type(field_name)
_populate_fval(field_type, attr_node, fmt_parts, exprs)
else:
raise NotImplementedError(
"Only simple attribute on local vars is supported in f-strings."
)
def _populate_fval(ftype, node, fmt_parts, exprs):
"""Populate format parts and expressions based on field type."""
if isinstance(ftype, ir.IntType):
# TODO: We print as signed integers only for now
if ftype.width == 64:
fmt_parts.append("%lld")
exprs.append(node)
elif ftype.width == 32:
fmt_parts.append("%d")
exprs.append(node)
else:
raise NotImplementedError(
f"Unsupported integer width in f-string: {ftype.width}"
)
elif ftype == ir.PointerType(ir.IntType(8)):
# NOTE: We assume i8* is a string
fmt_parts.append("%s")
exprs.append(node)
else:
raise NotImplementedError(f"Unsupported field type in f-string: {ftype}")
def _create_format_string_global(fmt_str, func, module, builder):
"""Create a global variable for the format string."""
fmt_name = f"{func.name}____fmt{func._fmt_counter}"
func._fmt_counter += 1
fmt_gvar = ir.GlobalVariable(
module, ir.ArrayType(ir.IntType(8), len(fmt_str)), name=fmt_name
)
fmt_gvar.global_constant = True
fmt_gvar.initializer = ir.Constant(
ir.ArrayType(ir.IntType(8), len(fmt_str)), bytearray(fmt_str.encode("utf8"))
)
fmt_gvar.linkage = "internal"
fmt_gvar.align = 1
return builder.bitcast(fmt_gvar, ir.PointerType())
def _prepare_expr_args(expr, func, module, builder, local_sym_tab, struct_sym_tab):
"""Evaluate and prepare an expression to use as an arg for bpf_printk."""
val, _ = eval_expr(
func,
module,
builder,
expr,
local_sym_tab,
None,
struct_sym_tab,
)
if val:
if isinstance(val.type, ir.PointerType):
val = builder.ptrtoint(val, ir.IntType(64))
elif isinstance(val.type, ir.IntType):
if val.type.width < 64:
val = builder.sext(val, ir.IntType(64))
else:
logger.warning(
"Only int and ptr supported in bpf_printk args. " "Others default to 0."
)
val = ir.Constant(ir.IntType(64), 0)
return val
else:
logger.warning(
"Failed to evaluate expression for bpf_printk argument. "
"It will be converted to 0."
)
return ir.Constant(ir.IntType(64), 0)
def get_data_ptr_and_size(data_arg, local_sym_tab, struct_sym_tab):
"""Extract data pointer and size information for perf event output."""
if isinstance(data_arg, ast.Name):

View File

@ -15,8 +15,5 @@ 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)

View File

@ -1,240 +0,0 @@
import ast
import logging
from llvmlite import ir
from pythonbpf.expr import eval_expr, get_base_type_and_depth, deref_to_depth
logger = logging.getLogger(__name__)
def simple_string_print(string_value, module, builder, func):
"""Prepare arguments for bpf_printk from a simple string value"""
fmt_str = string_value + "\n\0"
fmt_ptr = _create_format_string_global(fmt_str, func, module, builder)
args = [fmt_ptr, ir.Constant(ir.IntType(32), len(fmt_str))]
return args
def handle_fstring_print(
joined_str,
module,
builder,
func,
local_sym_tab=None,
struct_sym_tab=None,
):
"""Handle f-string formatting for bpf_printk emitter."""
fmt_parts = []
exprs = []
for value in joined_str.values:
logger.debug(f"Processing f-string value: {ast.dump(value)}")
if isinstance(value, ast.Constant):
_process_constant_in_fstring(value, fmt_parts, exprs)
elif isinstance(value, ast.FormattedValue):
_process_fval(
value,
fmt_parts,
exprs,
local_sym_tab,
struct_sym_tab,
)
else:
raise NotImplementedError(f"Unsupported f-string value type: {type(value)}")
fmt_str = "".join(fmt_parts)
args = simple_string_print(fmt_str, module, builder, func)
# NOTE: Process expressions (limited to 3 due to BPF constraints)
if len(exprs) > 3:
logger.warning("bpf_printk supports up to 3 args, extra args will be ignored.")
for expr in exprs[:3]:
arg_value = _prepare_expr_args(
expr,
func,
module,
builder,
local_sym_tab,
struct_sym_tab,
)
args.append(arg_value)
return args
# ============================================================================
# Internal Helpers
# ============================================================================
def _process_constant_in_fstring(cst, fmt_parts, exprs):
"""Process constant values in f-string."""
if isinstance(cst.value, str):
fmt_parts.append(cst.value)
elif isinstance(cst.value, int):
fmt_parts.append("%lld")
exprs.append(ir.Constant(ir.IntType(64), cst.value))
else:
raise NotImplementedError(
f"Unsupported constant type in f-string: {type(cst.value)}"
)
def _process_fval(fval, fmt_parts, exprs, local_sym_tab, struct_sym_tab):
"""Process formatted values in f-string."""
logger.debug(f"Processing formatted value: {ast.dump(fval)}")
if isinstance(fval.value, ast.Name):
_process_name_in_fval(fval.value, fmt_parts, exprs, local_sym_tab)
elif isinstance(fval.value, ast.Attribute):
_process_attr_in_fval(
fval.value,
fmt_parts,
exprs,
local_sym_tab,
struct_sym_tab,
)
else:
raise NotImplementedError(
f"Unsupported formatted value in f-string: {type(fval.value)}"
)
def _process_name_in_fval(name_node, fmt_parts, exprs, local_sym_tab):
"""Process name nodes in formatted values."""
if local_sym_tab and name_node.id in local_sym_tab:
_, var_type, tmp = local_sym_tab[name_node.id]
_populate_fval(var_type, name_node, fmt_parts, exprs)
def _process_attr_in_fval(attr_node, fmt_parts, exprs, local_sym_tab, struct_sym_tab):
"""Process attribute nodes in formatted values."""
if (
isinstance(attr_node.value, ast.Name)
and local_sym_tab
and attr_node.value.id in local_sym_tab
):
var_name = attr_node.value.id
field_name = attr_node.attr
var_type = local_sym_tab[var_name].metadata
if var_type not in struct_sym_tab:
raise ValueError(
f"Struct '{var_type}' for '{var_name}' not in symbol table"
)
struct_info = struct_sym_tab[var_type]
if field_name not in struct_info.fields:
raise ValueError(f"Field '{field_name}' not found in struct '{var_type}'")
field_type = struct_info.field_type(field_name)
_populate_fval(field_type, attr_node, fmt_parts, exprs)
else:
raise NotImplementedError(
"Only simple attribute on local vars is supported in f-strings."
)
def _populate_fval(ftype, node, fmt_parts, exprs):
"""Populate format parts and expressions based on field type."""
if isinstance(ftype, ir.IntType):
# TODO: We print as signed integers only for now
if ftype.width == 64:
fmt_parts.append("%lld")
exprs.append(node)
elif ftype.width == 32:
fmt_parts.append("%d")
exprs.append(node)
else:
raise NotImplementedError(
f"Unsupported integer width in f-string: {ftype.width}"
)
elif isinstance(ftype, ir.PointerType):
target, depth = get_base_type_and_depth(ftype)
if isinstance(target, ir.IntType):
if target.width == 64:
fmt_parts.append("%lld")
exprs.append(node)
elif target.width == 32:
fmt_parts.append("%d")
exprs.append(node)
elif target.width == 8 and depth == 1:
# NOTE: Assume i8* is a string
fmt_parts.append("%s")
exprs.append(node)
else:
raise NotImplementedError(
f"Unsupported pointer target type in f-string: {target}"
)
else:
raise NotImplementedError(
f"Unsupported pointer target type in f-string: {target}"
)
else:
raise NotImplementedError(f"Unsupported field type in f-string: {ftype}")
def _create_format_string_global(fmt_str, func, module, builder):
"""Create a global variable for the format string."""
fmt_name = f"{func.name}____fmt{func._fmt_counter}"
func._fmt_counter += 1
fmt_gvar = ir.GlobalVariable(
module, ir.ArrayType(ir.IntType(8), len(fmt_str)), name=fmt_name
)
fmt_gvar.global_constant = True
fmt_gvar.initializer = ir.Constant(
ir.ArrayType(ir.IntType(8), len(fmt_str)), bytearray(fmt_str.encode("utf8"))
)
fmt_gvar.linkage = "internal"
fmt_gvar.align = 1
return builder.bitcast(fmt_gvar, ir.PointerType())
def _prepare_expr_args(expr, func, module, builder, local_sym_tab, struct_sym_tab):
"""Evaluate and prepare an expression to use as an arg for bpf_printk."""
val, _ = eval_expr(
func,
module,
builder,
expr,
local_sym_tab,
None,
struct_sym_tab,
)
if val:
if isinstance(val.type, ir.PointerType):
target, depth = get_base_type_and_depth(val.type)
if isinstance(target, ir.IntType):
if target.width >= 32:
val = deref_to_depth(func, builder, val, depth)
val = builder.sext(val, ir.IntType(64))
elif target.width == 8 and depth == 1:
# NOTE: i8* is string, no need to deref
pass
else:
logger.warning(
"Only int and ptr supported in bpf_printk args. Others default to 0."
)
val = ir.Constant(ir.IntType(64), 0)
elif isinstance(val.type, ir.IntType):
if val.type.width < 64:
val = builder.sext(val, ir.IntType(64))
else:
logger.warning(
"Only int and ptr supported in bpf_printk args. Others default to 0."
)
val = ir.Constant(ir.IntType(64), 0)
return val
else:
logger.warning(
"Failed to evaluate expression for bpf_printk argument. "
"It will be converted to 0."
)
return ir.Constant(ir.IntType(64), 0)

View File

@ -1,9 +1,5 @@
from llvmlite import ir
import ast
from logging import Logger
import logging
logger: Logger = logging.getLogger(__name__)
def emit_license(module: ir.Module, license_str: str):
@ -45,9 +41,9 @@ def license_processing(tree, module):
emit_license(module, node.body[0].value.value)
return "LICENSE"
else:
logger.info("ERROR: LICENSE() must return a string literal")
print("ERROR: LICENSE() must return a string literal")
return None
else:
logger.info("ERROR: LICENSE already defined")
print("ERROR: LICENSE already defined")
return None
return None

View File

@ -1,93 +0,0 @@
from pythonbpf.debuginfo import DebugInfoGenerator
from .map_types import BPFMapType
def create_map_debug_info(module, map_global, map_name, map_params):
"""Generate debug info metadata for BPF maps HASH and PERF_EVENT_ARRAY"""
generator = DebugInfoGenerator(module)
uint_type = generator.get_uint32_type()
ulong_type = generator.get_uint64_type()
array_type = generator.create_array_type(
uint_type, map_params.get("type", BPFMapType.UNSPEC).value
)
type_ptr = generator.create_pointer_type(array_type, 64)
key_ptr = generator.create_pointer_type(
array_type if "key_size" in map_params else ulong_type, 64
)
value_ptr = generator.create_pointer_type(
array_type if "value_size" in map_params else ulong_type, 64
)
elements_arr = []
# Create struct members
# scope field does not appear for some reason
cnt = 0
for elem in map_params:
if elem == "max_entries":
continue
if elem == "type":
ptr = type_ptr
elif "key" in elem:
ptr = key_ptr
else:
ptr = value_ptr
# TODO: the best way to do this is not 64, but get the size each time. this will not work for structs.
member = generator.create_struct_member(elem, ptr, cnt * 64)
elements_arr.append(member)
cnt += 1
if "max_entries" in map_params:
max_entries_array = generator.create_array_type(
uint_type, map_params["max_entries"]
)
max_entries_ptr = generator.create_pointer_type(max_entries_array, 64)
max_entries_member = generator.create_struct_member(
"max_entries", max_entries_ptr, cnt * 64
)
elements_arr.append(max_entries_member)
# Create the struct type
struct_type = generator.create_struct_type(
elements_arr, 64 * len(elements_arr), is_distinct=True
)
# Create global variable debug info
global_var = generator.create_global_var_debug_info(
map_name, struct_type, is_local=False
)
# Attach debug info to the global variable
map_global.set_metadata("dbg", global_var)
return global_var
def create_ringbuf_debug_info(module, map_global, map_name, map_params):
"""Generate debug information metadata for BPF RINGBUF map"""
generator = DebugInfoGenerator(module)
int_type = generator.get_int32_type()
type_array = generator.create_array_type(
int_type, map_params.get("type", BPFMapType.RINGBUF).value
)
type_ptr = generator.create_pointer_type(type_array, 64)
type_member = generator.create_struct_member("type", type_ptr, 0)
max_entries_array = generator.create_array_type(int_type, map_params["max_entries"])
max_entries_ptr = generator.create_pointer_type(max_entries_array, 64)
max_entries_member = generator.create_struct_member(
"max_entries", max_entries_ptr, 64
)
elements_arr = [type_member, max_entries_member]
struct_type = generator.create_struct_type(elements_arr, 128, is_distinct=True)
global_var = generator.create_global_var_debug_info(
map_name, struct_type, is_local=False
)
map_global.set_metadata("dbg", global_var)
return global_var

View File

@ -1,39 +0,0 @@
from enum import Enum
class BPFMapType(Enum):
UNSPEC = 0
HASH = 1
ARRAY = 2
PROG_ARRAY = 3
PERF_EVENT_ARRAY = 4
PERCPU_HASH = 5
PERCPU_ARRAY = 6
STACK_TRACE = 7
CGROUP_ARRAY = 8
LRU_HASH = 9
LRU_PERCPU_HASH = 10
LPM_TRIE = 11
ARRAY_OF_MAPS = 12
HASH_OF_MAPS = 13
DEVMAP = 14
SOCKMAP = 15
CPUMAP = 16
XSKMAP = 17
SOCKHASH = 18
CGROUP_STORAGE_DEPRECATED = 19
CGROUP_STORAGE = 19
REUSEPORT_SOCKARRAY = 20
PERCPU_CGROUP_STORAGE_DEPRECATED = 21
PERCPU_CGROUP_STORAGE = 21
QUEUE = 22
STACK = 23
SK_STORAGE = 24
DEVMAP_HASH = 25
STRUCT_OPS = 26
RINGBUF = 27
INODE_STORAGE = 28
TASK_STORAGE = 29
BLOOM_FILTER = 30
USER_RINGBUF = 31
CGRP_STORAGE = 32

View File

@ -1,11 +1,10 @@
import ast
import logging
from logging import Logger
from llvmlite import ir
from enum import Enum
from .maps_utils import MapProcessorRegistry
from .map_types import BPFMapType
from .map_debug_info import create_map_debug_info, create_ringbuf_debug_info
from ..debuginfo import DebugInfoGenerator
import logging
logger: Logger = logging.getLogger(__name__)
@ -27,6 +26,44 @@ def is_map(func_node):
)
class BPFMapType(Enum):
UNSPEC = 0
HASH = 1
ARRAY = 2
PROG_ARRAY = 3
PERF_EVENT_ARRAY = 4
PERCPU_HASH = 5
PERCPU_ARRAY = 6
STACK_TRACE = 7
CGROUP_ARRAY = 8
LRU_HASH = 9
LRU_PERCPU_HASH = 10
LPM_TRIE = 11
ARRAY_OF_MAPS = 12
HASH_OF_MAPS = 13
DEVMAP = 14
SOCKMAP = 15
CPUMAP = 16
XSKMAP = 17
SOCKHASH = 18
CGROUP_STORAGE_DEPRECATED = 19
CGROUP_STORAGE = 19
REUSEPORT_SOCKARRAY = 20
PERCPU_CGROUP_STORAGE_DEPRECATED = 21
PERCPU_CGROUP_STORAGE = 21
QUEUE = 22
STACK = 23
SK_STORAGE = 24
DEVMAP_HASH = 25
STRUCT_OPS = 26
RINGBUF = 27
INODE_STORAGE = 28
TASK_STORAGE = 29
BLOOM_FILTER = 30
USER_RINGBUF = 31
CGRP_STORAGE = 32
def create_bpf_map(module, map_name, map_params):
"""Create a BPF map in the module with given parameters and debug info"""
@ -47,37 +84,116 @@ def create_bpf_map(module, map_name, map_params):
return map_global
def _parse_map_params(rval, expected_args=None):
"""Parse map parameters from call arguments and keywords."""
def create_map_debug_info(module, map_global, map_name, map_params):
"""Generate debug info metadata for BPF maps HASH and PERF_EVENT_ARRAY"""
generator = DebugInfoGenerator(module)
params = {}
uint_type = generator.get_uint32_type()
ulong_type = generator.get_uint64_type()
array_type = generator.create_array_type(
uint_type, map_params.get("type", BPFMapType.UNSPEC).value
)
type_ptr = generator.create_pointer_type(array_type, 64)
key_ptr = generator.create_pointer_type(
array_type if "key_size" in map_params else ulong_type, 64
)
value_ptr = generator.create_pointer_type(
array_type if "value_size" in map_params else ulong_type, 64
)
# Parse positional arguments
if expected_args:
for i, arg_name in enumerate(expected_args):
if i < len(rval.args):
arg = rval.args[i]
if isinstance(arg, ast.Name):
params[arg_name] = arg.id
elif isinstance(arg, ast.Constant):
params[arg_name] = arg.value
elements_arr = []
# Parse keyword arguments (override positional)
for keyword in rval.keywords:
if isinstance(keyword.value, ast.Name):
params[keyword.arg] = keyword.value.id
elif isinstance(keyword.value, ast.Constant):
params[keyword.arg] = keyword.value.value
# Create struct members
# scope field does not appear for some reason
cnt = 0
for elem in map_params:
if elem == "max_entries":
continue
if elem == "type":
ptr = type_ptr
elif "key" in elem:
ptr = key_ptr
else:
ptr = value_ptr
# TODO: the best way to do this is not 64, but get the size each time. this will not work for structs.
member = generator.create_struct_member(elem, ptr, cnt * 64)
elements_arr.append(member)
cnt += 1
return params
if "max_entries" in map_params:
max_entries_array = generator.create_array_type(
uint_type, map_params["max_entries"]
)
max_entries_ptr = generator.create_pointer_type(max_entries_array, 64)
max_entries_member = generator.create_struct_member(
"max_entries", max_entries_ptr, cnt * 64
)
elements_arr.append(max_entries_member)
# Create the struct type
struct_type = generator.create_struct_type(
elements_arr, 64 * len(elements_arr), is_distinct=True
)
# Create global variable debug info
global_var = generator.create_global_var_debug_info(
map_name, struct_type, is_local=False
)
# Attach debug info to the global variable
map_global.set_metadata("dbg", global_var)
return global_var
def create_ringbuf_debug_info(module, map_global, map_name, map_params):
"""Generate debug information metadata for BPF RINGBUF map"""
generator = DebugInfoGenerator(module)
int_type = generator.get_int32_type()
type_array = generator.create_array_type(
int_type, map_params.get("type", BPFMapType.RINGBUF).value
)
type_ptr = generator.create_pointer_type(type_array, 64)
type_member = generator.create_struct_member("type", type_ptr, 0)
max_entries_array = generator.create_array_type(
int_type, map_params["max_entries"])
max_entries_ptr = generator.create_pointer_type(max_entries_array, 64)
max_entries_member = generator.create_struct_member(
"max_entries", max_entries_ptr, 64
)
elements_arr = [type_member, max_entries_member]
struct_type = generator.create_struct_type(
elements_arr, 128, is_distinct=True)
global_var = generator.create_global_var_debug_info(
map_name, struct_type, is_local=False
)
map_global.set_metadata("dbg", global_var)
return global_var
@MapProcessorRegistry.register("RingBuf")
def process_ringbuf_map(map_name, rval, module):
"""Process a BPF_RINGBUF map declaration"""
logger.info(f"Processing Ringbuf: {map_name}")
map_params = _parse_map_params(rval, expected_args=["max_entries"])
map_params["type"] = BPFMapType.RINGBUF
map_params = {"type": BPFMapType.RINGBUF}
# Parse max_entries if present
if len(rval.args) >= 1 and isinstance(rval.args[0], ast.Constant):
const_val = rval.args[0].value
if isinstance(const_val, int):
map_params["max_entries"] = const_val
for keyword in rval.keywords:
if keyword.arg == "max_entries" and isinstance(keyword.value, ast.Constant):
const_val = keyword.value.value
if isinstance(const_val, int):
map_params["max_entries"] = const_val
logger.info(f"Ringbuf map parameters: {map_params}")
@ -90,8 +206,27 @@ def process_ringbuf_map(map_name, rval, module):
def process_hash_map(map_name, rval, module):
"""Process a BPF_HASH map declaration"""
logger.info(f"Processing HashMap: {map_name}")
map_params = _parse_map_params(rval, expected_args=["key", "value", "max_entries"])
map_params["type"] = BPFMapType.HASH
map_params = {"type": BPFMapType.HASH}
# Assuming order: key_type, value_type, max_entries
if len(rval.args) >= 1 and isinstance(rval.args[0], ast.Name):
map_params["key"] = rval.args[0].id
if len(rval.args) >= 2 and isinstance(rval.args[1], ast.Name):
map_params["value"] = rval.args[1].id
if len(rval.args) >= 3 and isinstance(rval.args[2], ast.Constant):
const_val = rval.args[2].value
if isinstance(const_val, (int, str)): # safe check
map_params["max_entries"] = const_val
for keyword in rval.keywords:
if keyword.arg == "key" and isinstance(keyword.value, ast.Name):
map_params["key"] = keyword.value.id
elif keyword.arg == "value" and isinstance(keyword.value, ast.Name):
map_params["value"] = keyword.value.id
elif keyword.arg == "max_entries" and isinstance(keyword.value, ast.Constant):
const_val = keyword.value.value
if isinstance(const_val, (int, str)):
map_params["max_entries"] = const_val
logger.info(f"Map parameters: {map_params}")
map_global = create_bpf_map(module, map_name, map_params)
@ -104,8 +239,18 @@ def process_hash_map(map_name, rval, module):
def process_perf_event_map(map_name, rval, module):
"""Process a BPF_PERF_EVENT_ARRAY map declaration"""
logger.info(f"Processing PerfEventArray: {map_name}")
map_params = _parse_map_params(rval, expected_args=["key_size", "value_size"])
map_params["type"] = BPFMapType.PERF_EVENT_ARRAY
map_params = {"type": BPFMapType.PERF_EVENT_ARRAY}
if len(rval.args) >= 1 and isinstance(rval.args[0], ast.Name):
map_params["key_size"] = rval.args[0].id
if len(rval.args) >= 2 and isinstance(rval.args[1], ast.Name):
map_params["value_size"] = rval.args[1].id
for keyword in rval.keywords:
if keyword.arg == "key_size" and isinstance(keyword.value, ast.Name):
map_params["key_size"] = keyword.value.id
elif keyword.arg == "value_size" and isinstance(keyword.value, ast.Name):
map_params["value_size"] = keyword.value.id
logger.info(f"Map parameters: {map_params}")
map_global = create_bpf_map(module, map_name, map_params)
@ -135,7 +280,9 @@ def process_bpf_map(func_node, module):
if handler:
return handler(map_name, rval, module)
else:
logger.warning(f"Unknown map type {rval.func.id}, defaulting to HashMap")
logger.warning(
f"Unknown map type " f"{rval.func.id}, defaulting to HashMap"
)
return process_hash_map(map_name, rval, module)
else:
raise ValueError("Function under @map must return a map")

View File

@ -19,7 +19,7 @@ def structs_proc(tree, module, chunks):
structs_sym_tab = {}
for cls_node in chunks:
if is_bpf_struct(cls_node):
logger.info(f"Found BPF struct: {cls_node.name}")
print(f"Found BPF struct: {cls_node.name}")
struct_info = process_bpf_struct(cls_node, module)
structs_sym_tab[cls_node.name] = struct_info
return structs_sym_tab

View File

@ -1,28 +1,24 @@
from llvmlite import ir
# TODO: THIS IS NOT SUPPOSED TO MATCH STRINGS :skull:
mapping = {
"c_int8": ir.IntType(8),
"c_uint8": ir.IntType(8),
"c_int16": ir.IntType(16),
"c_uint16": ir.IntType(16),
"c_int32": ir.IntType(32),
"c_uint32": ir.IntType(32),
"c_int64": ir.IntType(64),
"c_uint64": ir.IntType(64),
"c_float": ir.FloatType(),
"c_double": ir.DoubleType(),
"c_void_p": ir.IntType(64),
# Not so sure about this one
"str": ir.PointerType(ir.IntType(8)),
}
def ctypes_to_ir(ctype: str):
mapping = {
"c_int8": ir.IntType(8),
"c_uint8": ir.IntType(8),
"c_int16": ir.IntType(16),
"c_uint16": ir.IntType(16),
"c_int32": ir.IntType(32),
"c_uint32": ir.IntType(32),
"c_int64": ir.IntType(64),
"c_uint64": ir.IntType(64),
"c_float": ir.FloatType(),
"c_double": ir.DoubleType(),
"c_void_p": ir.IntType(64),
# Not so sure about this one
"str": ir.PointerType(ir.IntType(8)),
}
if ctype in mapping:
return mapping[ctype]
raise NotImplementedError(f"No mapping for {ctype}")
def is_ctypes(ctype: str) -> bool:
return ctype in mapping

View File

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

View File

@ -1,205 +0,0 @@
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, Dict
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__
logger.info(f"Begin {current_symbol_name} Processing")
field_table: Dict[str, list] = {}
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)
# Check if already processed
if handler.has_node(current_symbol_name):
logger.debug(f"Node {current_symbol_name} already processed and ready")
return True
# XXX:Check it's use. It's probably not being used.
if current_symbol_name in processing_stack:
logger.debug(
f"Dependency already in processing stack for {current_symbol_name}, skipping"
)
return True
processing_stack.add(current_symbol_name)
if module_name == "vmlinux":
if hasattr(elem_type_class, "_type_"):
pass
else:
new_dep_node = DependencyNode(name=current_symbol_name)
# elem_type_class is the actual vmlinux struct/class
new_dep_node.set_ctype_struct(elem_type_class)
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_elem in class_obj._fields_:
field_name: str = ""
field_type: Optional[Any] = None
bitfield_size: Optional[int] = None
if len(field_elem) == 2:
field_name, field_type = field_elem
elif len(field_elem) == 3:
field_name, field_type, bitfield_size = field_elem
field_table[field_name] = [field_type, bitfield_size]
elif hasattr(class_obj, "__annotations__"):
for field_elem in class_obj.__annotations__.items():
if len(field_elem) == 2:
field_name, field_type = field_elem
bitfield_size = None
elif len(field_elem) == 3:
field_name, field_type, bitfield_size = field_elem
else:
raise ValueError(
"Number of fields in items() of class object unexpected"
)
field_table[field_name] = [field_type, bitfield_size]
else:
raise TypeError("Could not get required class and definition")
logger.debug(f"Extracted fields for {current_symbol_name}: {field_table}")
for elem in field_table.items():
elem_name, elem_temp_list = elem
[elem_type, elem_bitfield_size] = elem_temp_list
local_module_name = getattr(elem_type, "__module__", None)
new_dep_node.add_field(elem_name, elem_type, ready=False)
if local_module_name == ctypes.__name__:
new_dep_node.set_field_bitfield_size(elem_name, elem_bitfield_size)
new_dep_node.set_field_ready(elem_name, is_ready=True)
logger.debug(
f"Field {elem_name} is direct ctypes type: {elem_type}"
)
elif local_module_name == "vmlinux":
new_dep_node.set_field_bitfield_size(elem_name, elem_bitfield_size)
logger.debug(
f"Processing vmlinux field: {elem_name}, type: {elem_type}"
)
if hasattr(elem_type, "_type_"):
is_complex_type = True
containing_type = elem_type._type_
if hasattr(elem_type, "_length_") and is_complex_type:
type_length = elem_type._length_
if containing_type.__module__ == "vmlinux":
new_dep_node.add_dependent(
elem_type._type_.__name__
if hasattr(elem_type._type_, "__name__")
else str(elem_type._type_)
)
elif containing_type.__module__ == ctypes.__name__:
if isinstance(elem_type, type):
if issubclass(elem_type, ctypes.Array):
ctype_complex_type = ctypes.Array
elif issubclass(elem_type, ctypes._Pointer):
ctype_complex_type = ctypes._Pointer
else:
raise TypeError("Unsupported ctypes subclass")
else:
raise ImportError(
f"Unsupported module of {containing_type}"
)
logger.debug(
f"{containing_type} containing type of parent {elem_name} with {elem_type} and ctype {ctype_complex_type} and length {type_length}"
)
new_dep_node.set_field_containing_type(
elem_name, containing_type
)
new_dep_node.set_field_type_size(elem_name, type_length)
new_dep_node.set_field_ctype_complex_type(
elem_name, ctype_complex_type
)
new_dep_node.set_field_type(elem_name, elem_type)
if containing_type.__module__ == "vmlinux":
containing_type_name = (
containing_type.__name__
if hasattr(containing_type, "__name__")
else str(containing_type)
)
# Check for self-reference or already processed
if containing_type_name == current_symbol_name:
# Self-referential pointer
logger.debug(
f"Self-referential pointer in {current_symbol_name}.{elem_name}"
)
new_dep_node.set_field_ready(elem_name, True)
elif handler.has_node(containing_type_name):
# Already processed
logger.debug(
f"Reusing already processed {containing_type_name}"
)
new_dep_node.set_field_ready(elem_name, True)
else:
# Process recursively - THIS WAS MISSING
new_dep_node.add_dependent(containing_type_name)
process_vmlinux_post_ast(
containing_type,
llvm_handler,
handler,
processing_stack,
)
new_dep_node.set_field_ready(elem_name, True)
elif containing_type.__module__ == ctypes.__name__:
logger.debug(f"Processing ctype internal{containing_type}")
new_dep_node.set_field_ready(elem_name, True)
else:
raise TypeError(
"Module not supported in recursive resolution"
)
else:
new_dep_node.add_dependent(
elem_type.__name__
if hasattr(elem_type, "__name__")
else str(elem_type)
)
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"
)
else:
raise ImportError("UNSUPPORTED Module")
logging.info(
f"{current_symbol_name} processed and handler readiness {handler.is_ready}"
)
return True

View File

@ -1,173 +0,0 @@
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)
def __getitem__(self, name: str) -> DependencyNode:
"""
Get a node by name using dictionary-style access.
Args:
name: The name of the node to retrieve
Returns:
DependencyNode: The node with the given name
Raises:
KeyError: If no node with the given name exists
Example:
node = handler["some-dep_node_name"]
"""
if name not in self._nodes:
raise KeyError(f"No node with name '{name}' found")
return self._nodes[name]
@property
def nodes(self):
return self._nodes

View File

@ -1,363 +0,0 @@
from dataclasses import dataclass, field
from typing import Dict, Any, Optional
import ctypes
# 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
ctype_complex_type: Optional[Any]
containing_type: Optional[Any]
type_size: Optional[int]
bitfield_size: Optional[int]
offset: 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 = False) -> 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 = False) -> 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 = False
) -> 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 = False) -> 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
def set_ctype_complex_type(
self, ctype_complex_type: Any, mark_ready: bool = False
) -> None:
"""Set the ctype_complex_type of this field and optionally mark it as ready."""
self.ctype_complex_type = ctype_complex_type
if mark_ready:
self.ready = True
def set_bitfield_size(self, bitfield_size: Any, mark_ready: bool = False) -> None:
"""Set the bitfield_size of this field and optionally mark it as ready."""
self.bitfield_size = bitfield_size
if mark_ready:
self.ready = True
def set_offset(self, offset: int) -> None:
"""Set the offset of this field"""
self.offset = offset
@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
depends_on: Optional[list[str]] = None
fields: Dict[str, Field] = field(default_factory=dict)
_ready_cache: Optional[bool] = field(default=None, repr=False)
current_offset: int = 0
ctype_struct: Optional[Any] = 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,
ctype_complex_type: Optional[int] = None,
bitfield_size: Optional[int] = None,
ready: bool = False,
offset: int = 0,
) -> None:
"""Add a field to the node with an optional initial value and readiness state."""
if self.depends_on is None:
self.depends_on = []
self.fields[name] = Field(
name=name,
type=field_type,
value=initial_value,
ready=ready,
containing_type=containing_type,
type_size=type_size,
ctype_complex_type=ctype_complex_type,
bitfield_size=bitfield_size,
offset=offset,
)
# Invalidate readiness cache
self._ready_cache = None
def set_ctype_struct(self, ctype_struct: Any) -> None:
"""Set the ctypes structure for automatic offset calculation."""
self.ctype_struct = ctype_struct
def __sizeof__(self):
# If we have a ctype_struct, use its size
if self.ctype_struct is not None:
return ctypes.sizeof(self.ctype_struct)
return self.current_offset
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 = False) -> 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 = False) -> 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 = False
) -> 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 = False
) -> 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_ctype_complex_type(
self, name: str, ctype_complex_type: Any, mark_ready: bool = False
) -> None:
"""Set a field's ctype_complex_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_ctype_complex_type(ctype_complex_type, mark_ready)
# Invalidate readiness cache
self._ready_cache = None
def set_field_bitfield_size(
self, name: str, bitfield_size: Any, mark_ready: bool = False
) -> None:
"""Set a field's bitfield_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_bitfield_size(bitfield_size, mark_ready)
# Invalidate readiness cache
self._ready_cache = None
def set_field_ready(
self,
name: str,
is_ready: bool = False,
size_of_containing_type: Optional[int] = None,
) -> 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)
# Use ctypes built-in offset if available
if self.ctype_struct is not None:
try:
self.fields[name].set_offset(getattr(self.ctype_struct, name).offset)
except AttributeError:
# Fallback to manual calculation if field not found in ctype_struct
self.fields[name].set_offset(self.current_offset)
self.current_offset += self._calculate_size(
name, size_of_containing_type
)
else:
# Manual offset calculation when no ctype_struct is available
self.fields[name].set_offset(self.current_offset)
self.current_offset += self._calculate_size(name, size_of_containing_type)
# Invalidate readiness cache
self._ready_cache = None
def _calculate_size(
self, name: str, size_of_containing_type: Optional[int] = None
) -> int:
processing_field = self.fields[name]
# size_of_field will be in bytes
if processing_field.type.__module__ == ctypes.__name__:
size_of_field = ctypes.sizeof(processing_field.type)
return size_of_field
elif processing_field.type.__module__ == "vmlinux":
if processing_field.ctype_complex_type is not None:
if issubclass(processing_field.ctype_complex_type, ctypes.Array):
if processing_field.containing_type.__module__ == ctypes.__name__:
if (
processing_field.containing_type is not None
and processing_field.type_size is not None
):
size_of_field = (
ctypes.sizeof(processing_field.containing_type)
* processing_field.type_size
)
else:
raise RuntimeError(
f"{processing_field} has no containing_type or type_size"
)
return size_of_field
elif processing_field.containing_type.__module__ == "vmlinux":
if (
size_of_containing_type is not None
and processing_field.type_size is not None
):
size_of_field = (
size_of_containing_type * processing_field.type_size
)
else:
raise RuntimeError(
f"{processing_field} has no containing_type or type_size"
)
return size_of_field
elif issubclass(processing_field.ctype_complex_type, ctypes._Pointer):
return ctypes.sizeof(ctypes.c_void_p)
else:
raise NotImplementedError(
"This subclass of ctype not supported yet"
)
elif processing_field.type_size is not None:
# Handle vmlinux types with type_size but no ctype_complex_type
# This means it's a direct vmlinux struct field (not array/pointer wrapped)
# The type_size should already contain the full size of the struct
# But if there's a containing_type from vmlinux, we need that size
if processing_field.containing_type is not None:
if processing_field.containing_type.__module__ == "vmlinux":
# For vmlinux containing types, we need the pre-calculated size
if size_of_containing_type is not None:
return size_of_containing_type * processing_field.type_size
else:
raise RuntimeError(
f"Field {name}: vmlinux containing_type requires size_of_containing_type"
)
else:
raise ModuleNotFoundError(
f"Containing type module {processing_field.containing_type.__module__} not supported"
)
else:
raise RuntimeError("Wrong type found with no containing type")
else:
# No ctype_complex_type and no type_size, must rely on size_of_containing_type
if size_of_containing_type is None:
raise RuntimeError(
f"Size of containing type {size_of_containing_type} is None"
)
return size_of_containing_type
else:
raise ModuleNotFoundError("Module is not supported for the operation")
raise RuntimeError("control should not reach here")
@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 = True
return True
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}
def get_not_ready_fields(self) -> Dict[str, Field]:
"""Get all fields that are marked as not ready."""
return {name: elem for name, elem in self.fields.items() if not elem.ready}
def add_dependent(self, dep_type):
if dep_type in self.depends_on:
return
else:
self.depends_on.append(dep_type)

View File

@ -1,147 +0,0 @@
import ast
import logging
from typing import List, Tuple, Any
import importlib
import inspect
from .dependency_handler import DependencyHandler
from .ir_gen 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, tuple[type, Any]] = {}
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)
return assignments
def process_vmlinux_assign(node, module, assignments: dict[str, tuple[type, Any]]):
# Check if this is a simple assignment with a constant value
if len(node.targets) == 1 and isinstance(node.targets[0], ast.Name):
target_name = node.targets[0].id
if isinstance(node.value, ast.Constant):
assignments[target_name] = (type(node.value.value), node.value.value)
logger.info(
f"Added assignment: {target_name} = {node.value.value!r} of type {type(node.value.value)}"
)
else:
raise ValueError(f"Unsupported assignment type for {target_name}")
else:
raise ValueError("Not a simple assignment")

View File

@ -1,3 +0,0 @@
from .ir_generation import IRGenerator
__all__ = ["IRGenerator"]

View File

@ -1,15 +0,0 @@
from pythonbpf.debuginfo import DebugInfoGenerator
def debug_info_generation(struct, llvm_module):
generator = DebugInfoGenerator(llvm_module)
# this is sample debug info generation
# i64type = generator.get_uint64_type()
struct_type = generator.create_struct_type([], 64 * 4, is_distinct=True)
global_var = generator.create_global_var_debug_info(
struct.name, struct_type, is_local=False
)
return global_var

View File

@ -1,161 +0,0 @@
import ctypes
import logging
from ..dependency_handler import DependencyHandler
from .debug_info_gen import debug_info_generation
from ..dependency_node import DependencyNode
import llvmlite.ir as ir
logger = logging.getLogger(__name__)
class IRGenerator:
# get the assignments dict and add this stuff to it.
def __init__(self, llvm_module, handler: DependencyHandler, assignment=None):
self.llvm_module = llvm_module
self.handler: DependencyHandler = handler
self.generated: list[str] = []
if not handler.is_ready:
raise ImportError(
"Semantic analysis of vmlinux imports failed. Cannot generate IR"
)
for struct in handler:
self.struct_processor(struct)
def struct_processor(self, struct, processing_stack=None):
# Initialize processing stack on first call
if processing_stack is None:
processing_stack = set()
# If already generated, skip
if struct.name in self.generated:
return
# Detect circular dependency
if struct.name in processing_stack:
logger.info(
f"Circular dependency detected for {struct.name}, skipping recursive processing"
)
# For circular dependencies, we can either:
# 1. Use forward declarations (opaque pointers)
# 2. Mark as incomplete and process later
# 3. Generate a placeholder type
# Here we'll just skip and let it be processed in its own call
return
logger.info(f"IR generating for {struct.name}")
# Add to processing stack before processing dependencies
processing_stack.add(struct.name)
try:
# Process all dependencies first
if struct.depends_on is None:
pass
else:
for dependency in struct.depends_on:
if dependency not in self.generated:
# Check if dependency exists in handler
if dependency in self.handler.nodes:
dep_node_from_dependency = self.handler[dependency]
# Pass the processing_stack down to track circular refs
self.struct_processor(
dep_node_from_dependency, processing_stack
)
else:
raise RuntimeError(
f"Warning: Dependency {dependency} not found in handler"
)
# Actual processor logic here after dependencies are resolved
self.gen_ir(struct)
self.generated.append(struct.name)
finally:
# Remove from processing stack after we're done
processing_stack.discard(struct.name)
def gen_ir(self, struct):
# TODO: we add the btf_ama attribute by monkey patching in the end of compilation, but once llvmlite
# accepts our issue, we will resort to normal accessed attribute based attribute addition
# currently we generate all possible field accesses for CO-RE and put into the assignment table
debug_info = debug_info_generation(struct, self.llvm_module)
field_index = 0
for field_name, field in struct.fields.items():
# does not take arrays and similar types into consideration yet.
if field.ctype_complex_type is not None and issubclass(
field.ctype_complex_type, ctypes.Array
):
array_size = field.type_size
containing_type = field.containing_type
if containing_type.__module__ == ctypes.__name__:
containing_type_size = ctypes.sizeof(containing_type)
for i in range(0, array_size):
field_co_re_name = self._struct_name_generator(
struct, field, field_index, True, i, containing_type_size
)
globvar = ir.GlobalVariable(
self.llvm_module, ir.IntType(64), name=field_co_re_name
)
globvar.linkage = "external"
globvar.set_metadata("llvm.preserve.access.index", debug_info)
field_index += 1
elif field.type_size is not None:
array_size = field.type_size
containing_type = field.containing_type
if containing_type.__module__ == "vmlinux":
containing_type_size = self.handler[
containing_type.__name__
].current_offset
for i in range(0, array_size):
field_co_re_name = self._struct_name_generator(
struct, field, field_index, True, i, containing_type_size
)
globvar = ir.GlobalVariable(
self.llvm_module, ir.IntType(64), name=field_co_re_name
)
globvar.linkage = "external"
globvar.set_metadata("llvm.preserve.access.index", debug_info)
field_index += 1
else:
field_co_re_name = self._struct_name_generator(
struct, field, field_index
)
field_index += 1
globvar = ir.GlobalVariable(
self.llvm_module, ir.IntType(64), name=field_co_re_name
)
globvar.linkage = "external"
globvar.set_metadata("llvm.preserve.access.index", debug_info)
def _struct_name_generator(
self,
struct: DependencyNode,
field,
field_index: int,
is_indexed: bool = False,
index: int = 0,
containing_type_size: int = 0,
) -> str:
if is_indexed:
name = (
"llvm."
+ struct.name.removeprefix("struct_")
+ f":0:{field.offset + index * containing_type_size}"
+ "$"
+ f"0:{field_index}:{index}"
)
return name
elif struct.name.startswith("struct_"):
name = (
"llvm."
+ struct.name.removeprefix("struct_")
+ f":0:{field.offset}"
+ "$"
+ f"0:{field_index}"
)
return name
else:
print(self.handler[struct.name])
raise TypeError(
"Name generation cannot occur due to type name not starting with struct"
)

View File

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

View File

@ -1,9 +1,23 @@
// SPDX-License-Identifier: GPL-2.0
#include "vmlinux.h"
#include <linux/bpf.h>
#include <bpf/bpf_helpers.h>
#include <bpf/bpf_tracing.h>
struct trace_entry {
short unsigned int type;
unsigned char flags;
unsigned char preempt_count;
int pid;
};
struct trace_event_raw_sys_enter {
struct trace_entry ent;
long int id;
long unsigned int args[6];
char __data[0];
};
struct event {
__u32 pid;
__u32 uid;

View File

@ -1,27 +0,0 @@
// 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";

View File

@ -1,19 +0,0 @@
#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;
}

View File

@ -1,40 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
# NOTE: This example tries to reinterpret the variable `x` to a different type.
# We do not allow this for now, as stack allocations are typed and have to be
# done in the first basic block. Allowing re-interpretation would require
# re-allocation of stack space (possibly in a new basic block), which is not
# supported in eBPF yet.
# We can allow bitcasts in cases where the width of the types is the same in
# the future. But for now, we do not allow any re-interpretation of variables.
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
last.update(0, 1)
x = last.lookup(0)
x = 20
if x == 2:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -3,9 +3,9 @@ from ctypes import c_void_p, c_int64
@bpf
@section("tracepoint/syscalls/sys_enter_sync")
@section("sometag1")
def sometag(ctx: c_void_p) -> c_int64:
a = 1 + 2 + 1 + 12 + 13
a = 1 + 2 + 1
print(f"{a}")
return c_int64(0)

View File

@ -1,5 +1,3 @@
import logging
from pythonbpf import compile, bpf, section, bpfglobal
from ctypes import c_void_p, c_int64
@ -7,7 +5,8 @@ from ctypes import c_void_p, c_int64
@bpf
@section("sometag1")
def sometag(ctx: c_void_p) -> c_int64:
a = 1 - 1
b = 1 + 2
a = 1 + b
return c_int64(a)
@ -17,4 +16,4 @@ def LICENSE() -> str:
return "GPL"
compile(loglevel=logging.INFO)
compile()

View File

@ -1,34 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
# NOTE: Decided against fixing this
# as a workaround is assigning the result of lookup to a variable
# and then using that variable in the if statement.
# Might fix in future.
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
last.update(0, 1)
if last.lookup(0) > 0:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,18 +0,0 @@
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()

View File

@ -1,34 +0,0 @@
from pythonbpf import bpf, struct, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
# NOTE: Decided against fixing this
# as one workaround is to just check any field of the struct
# in the if statement. Ugly but works.
# Might fix in future.
@bpf
@struct
class data_t:
pid: c_uint64
ts: c_uint64
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
dat = data_t()
if dat:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -4,18 +4,6 @@ from pythonbpf.maps import HashMap
from ctypes import c_void_p, c_int64
# NOTE: I have decided to not fix this example for now.
# The issue is in line 31, where we are passing an expression.
# The update helper expects a pointer type. But the problem is
# that we must allocate the space for said pointer in the first
# basic block. As that usage is in a different basic block, we
# are unable to cast the expression to a pointer type. (as we never
# allocated space for it).
# Shall we change our space allocation logic? That allows users to
# spam the same helper with the same args, and still run out of
# stack space. So we consider this usage invalid for now.
# Might fix it later.
@bpf
@map
@ -26,12 +14,12 @@ def count() -> HashMap:
@bpf
@section("xdp")
def hello_world(ctx: c_void_p) -> c_int64:
prev = count.lookup(0)
prev = count().lookup(0)
if prev:
count.update(0, prev + 1)
count().update(0, prev + 1)
return XDP_PASS
else:
count.update(0, 1)
count().update(0, 1)
return XDP_PASS

View File

@ -1,109 +0,0 @@
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()

View File

@ -1,41 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from pythonbpf.helper import XDP_PASS
from pythonbpf.maps import HashMap
from ctypes import c_void_p, c_int64
# NOTE: This example exposes the problems with our typing system.
# We can't do steps on line 25 and 27.
# prev is of type i64**. For prev + 1, we deref it down to i64
# To assign it back to prev, we need to go back to i64**.
# 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:
return HashMap(key=c_int64, value=c_int64, max_entries=1)
@bpf
@section("xdp")
def hello_world(ctx: c_void_p) -> c_int64:
prev = count.lookup(0)
if prev:
prev = prev + 1
count.update(0, prev)
return XDP_PASS
else:
count.update(0, 1)
return XDP_PASS
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,23 +0,0 @@
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()

View File

@ -1,54 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile_to_ir
from pythonbpf.maps import HashMap
from pythonbpf.helper import XDP_PASS
from vmlinux import TASK_COMM_LEN # noqa: F401
from vmlinux import struct_qspinlock # noqa: F401
# from vmlinux import struct_trace_event_raw_sys_enter # noqa: F401
# from vmlinux import struct_posix_cputimers # noqa: F401
from vmlinux import struct_xdp_md
# from vmlinux import struct_trace_event_raw_sys_enter # noqa: F401
# from vmlinux import struct_ring_buffer_per_cpu # noqa: F401
# from vmlinux import struct_request # noqa: F401
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")

View File

@ -1,74 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile, struct
from ctypes import c_void_p, c_int64, c_int32, c_uint64
from pythonbpf.maps import HashMap
from pythonbpf.helper import ktime
# NOTE: This is a comprehensive test combining struct, helper, and map features
# Please note that at line 50, though we have used an absurd expression to test
# the compiler, it is recommended to use named variables to reduce the amount of
# scratch space that needs to be allocated.
@bpf
@struct
class data_t:
pid: c_uint64
ts: c_uint64
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
dat = data_t()
dat.pid = 123
dat.pid = dat.pid + 1
print(f"pid is {dat.pid}")
tu = 9
last.update(0, tu)
last.update(1, -last.lookup(0))
x = last.lookup(0)
print(f"Map value at index 0: {x}")
x = x + c_int32(1)
print(f"x after adding 32-bit 1 is {x}")
x = ktime() - 121
print(f"ktime - 121 is {x}")
x = last.lookup(0)
x = x + 1
print(f"x is {x}")
if x == 10:
jat = data_t()
jat.ts = 456
print(f"Hello, World!, ts is {jat.ts}")
a = last.lookup(0)
print(f"a is {a}")
last.update(9, 9)
last.update(
0,
last.lookup(last.lookup(0))
+ last.lookup(last.lookup(0))
+ last.lookup(last.lookup(0)),
)
z = last.lookup(0)
print(f"new map val at index 0 is {z}")
else:
a = last.lookup(0)
print("Goodbye, World!")
c = last.lookup(1 - 1)
print(f"c is {c}")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,27 +0,0 @@
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:
x = 1
print(f"Initial x: {x}")
a = 20
x = a
print(f"Updated x with a: {x}")
x = (x + x) * 3
if x == 2:
print("Hello, World!")
else:
print(f"Goodbye, World! {x}")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,34 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
# NOTE: An example of i64** assignment with binops on the RHS
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
last.update(0, 1)
x = last.lookup(0)
print(f"{x}")
x = x + 1
if x == 2:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,40 +0,0 @@
from pythonbpf import bpf, section, bpfglobal, compile, struct
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.helper import ktime
@bpf
@struct
class data_t:
pid: c_uint64
ts: c_uint64
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
dat = data_t()
dat.pid = 123
dat.pid = dat.pid + 1
print(f"pid is {dat.pid}")
x = ktime() - 121
print(f"ktime is {x}")
x = 1
x = x + 1
print(f"x is {x}")
if x == 2:
jat = data_t()
jat.ts = 456
print(f"Hello, World!, ts is {jat.ts}")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,20 +0,0 @@
from pythonbpf import compile, bpf, section, bpfglobal
from ctypes import c_void_p, c_int64
@bpf
@section("tracepoint/syscalls/sys_enter_sync")
def sometag(ctx: c_void_p) -> c_int64:
b = 1 + 2
a = 1 + b
print(f"{a}")
return c_int64(0)
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,32 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
last.update(0, 1)
last.update(1, 2)
x = last.lookup(0)
y = last.lookup(1)
if x and y:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,21 +0,0 @@
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:
if True:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,21 +0,0 @@
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:
if (0 + 1) * 0:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,21 +0,0 @@
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:
if 0:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,30 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
# last.update(0, 1)
tsp = last.lookup(0)
if tsp:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,30 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
last.update(0, 1)
tsp = last.lookup(0)
if tsp > 0:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,30 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
# last.update(0, 1)
tsp = last.lookup(0)
if not tsp:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,32 +0,0 @@
from pythonbpf import bpf, map, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
from pythonbpf.maps import HashMap
@bpf
@map
def last() -> HashMap:
return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
last.update(0, 1)
# last.update(1, 2)
x = last.lookup(0)
y = last.lookup(1)
if x or y:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,29 +0,0 @@
from pythonbpf import bpf, struct, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_uint64
@bpf
@struct
class data_t:
pid: c_uint64
ts: c_uint64
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
dat = data_t()
if dat.ts:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,23 +0,0 @@
from pythonbpf import bpf, section, bpfglobal, compile
from ctypes import c_void_p, c_int64, c_int32
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
x = 0
y = c_int32(0)
if x == y:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,22 +0,0 @@
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:
x = 0
if x:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,22 +0,0 @@
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:
x = 0
if x * 1:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,22 +0,0 @@
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:
x = 2
if x > 3:
print("Hello, World!")
else:
print("Goodbye, World!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,7 +1,7 @@
from pythonbpf import bpf, map, struct, section, bpfglobal, compile, compile_to_ir, BPF
from pythonbpf.helper import ktime, pid
from pythonbpf.maps import PerfEventArray
import logging
from ctypes import c_void_p, c_int32, c_uint64
@ -42,8 +42,8 @@ def LICENSE() -> str:
return "GPL"
compile()
compile_to_ir("perf_buffer_map.py", "perf_buffer_map.ll")
compile(loglevel=logging.INFO)
b = BPF()
b.load_and_attach()

View File

@ -1,18 +0,0 @@
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!")
return 1 + 1 - 2
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,19 +0,0 @@
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!")
a = 2
return a - 2
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,18 +0,0 @@
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!")
return True
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,18 +0,0 @@
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!")
return 1
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,18 +0,0 @@
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!")
return
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,20 +0,0 @@
from pythonbpf import bpf, section, bpfglobal, compile
from ctypes import c_void_p, c_int32
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int32:
print("Hello, World!")
a = 1 # int64
x = 1 # int64
return c_int32(a - x) # typecast to int32
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,18 +0,0 @@
from pythonbpf import bpf, section, bpfglobal, compile
from ctypes import c_void_p, c_int32
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int32:
print("Hello, World!")
return c_int32(1)
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,19 +0,0 @@
from pythonbpf import bpf, section, bpfglobal, compile
from ctypes import c_void_p, c_int32
@bpf
@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
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,19 +0,0 @@
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!")
a = 1
return a
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,19 +0,0 @@
from pythonbpf import bpf, section, bpfglobal, compile
from ctypes import c_void_p, c_int64
from pythonbpf.helper import XDP_PASS
@bpf
@section("tracepoint/syscalls/sys_enter_execve")
def hello_world(ctx: c_void_p) -> c_int64:
print("Hello, World!")
return XDP_PASS
@bpf
@bpfglobal
def LICENSE() -> str:
return "GPL"
compile()

View File

@ -1,379 +0,0 @@
#!/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()