From 43b8c2c010986d0109b977083132a147daac1231 Mon Sep 17 00:00:00 2001 From: Pragyansh Chaturvedi Date: Sat, 6 Sep 2025 14:40:51 +0530 Subject: [PATCH] Complete example 1, albeit hardcoded somewhat. Implement body function processing skeleton --- pythonbpf/functions_pass.py | 90 +++++++++++++++++++++++++++++++++++++ 1 file changed, 90 insertions(+) diff --git a/pythonbpf/functions_pass.py b/pythonbpf/functions_pass.py index 1eb35e8..674a5ab 100644 --- a/pythonbpf/functions_pass.py +++ b/pythonbpf/functions_pass.py @@ -58,11 +58,101 @@ def get_probe_string(func_node): return "helper" +def process_func_body(module, builder, func_node, func): + """Process the body of a bpf function""" + # TODO: A lot. We just have print -> bpf_trace_printk for now + did_return = False + + for stmt in func_node.body: + if isinstance(stmt, ast.Expr) and isinstance(stmt.value, ast.Call): + call = stmt.value + if isinstance(call.func, ast.Name) and call.func.id == "print": + # Handle print statement + for arg in call.args: + if isinstance(arg, ast.Constant) and isinstance(arg.value, str): + fmt_str = arg.value + "\n" + # Create a global variable for the format string + fmt_gvar = ir.GlobalVariable( + module, ir.ArrayType(ir.IntType(8), len(fmt_str)), name=f"{func.name}____fmt") + 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 + + # Cast the global variable to i8* + fmt_ptr = builder.bitcast( + fmt_gvar, ir.PointerType()) + + # Call bpf_trace_printk (assumed to be at address 6) + fn_type = ir.FunctionType(ir.IntType( + 64), [ir.PointerType(), ir.IntType(32)], var_arg=True) + fn_ptr_type = ir.PointerType(fn_type) + fn_addr = ir.Constant(ir.IntType(64), 6) + fn_ptr = builder.inttoptr(fn_addr, fn_ptr_type) + + # Call the function + builder.call(fn_ptr, [fmt_ptr, ir.Constant( + ir.IntType(32), len(fmt_str))], tail=True) + 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 stmt.value.func.id == "c_int32" and len(stmt.value.args) == 1 and isinstance(stmt.value.args[0], ast.Constant) and isinstance(stmt.value.args[0].value, int): + builder.ret(ir.Constant(ir.IntType( + 32), stmt.value.args[0].value)) + did_return = True + else: + print("Unsupported return value") + if not did_return: + builder.ret(ir.Constant(ir.IntType(32), 0)) + + +def process_bpf_chunk(func_node, module): + """Process a single BPF chunk (function) and emit corresponding LLVM IR.""" + + func_name = func_node.name + + # TODO: parse return type + ret_type = ir.IntType(32) + + # 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") + + if func_node.args.args: + # Only look at the first argument for now + param = func.args[0] + param.add_attribute("nocapture") + + func.section = get_probe_string(func_node) + + block = func.append_basic_block(name="entry") + builder = ir.IRBuilder(block) + + process_func_body(module, builder, func_node, func) + + print(func) + return func + + def func_proc(tree, module, chunks): for func_node in chunks: func_type = get_probe_string(func_node) print(f"Found probe_string of {func_node.name}: {func_type}") + process_bpf_chunk(func_node, module) + def functions_processing(tree, module): bpf_functions = []