Merge branch 'master' into request-struct

pythonbpf/vmlinux_parser/class_handler.py

@@ -16,6 +16,33 @@ def get_module_symbols(module_name: str):
     return [name for name in dir(imported_module)], imported_module
 
 
+def unwrap_pointer_type(type_obj: Any) -> Any:
+    """
+    Recursively unwrap all pointer layers to get the base type.
+
+    This handles multiply nested pointers like LP_LP_struct_attribute_group
+    and returns the base type (struct_attribute_group).
+
+    Stops unwrapping when reaching a non-pointer type (one without _type_ attribute).
+
+    Args:
+        type_obj: The type object to unwrap
+
+    Returns:
+        The base type after unwrapping all pointer layers
+    """
+    current_type = type_obj
+    # Keep unwrapping while it's a pointer/array type (has _type_)
+    # But stop if _type_ is just a string or basic type marker
+    while hasattr(current_type, "_type_"):
+        next_type = current_type._type_
+        # Stop if _type_ is a string (like 'c' for c_char)
+        if isinstance(next_type, str):
+            break
+        current_type = next_type
+    return current_type
+
+
 def process_vmlinux_class(
     node,
     llvm_module,
@@ -158,28 +185,19 @@ def process_vmlinux_post_ast(
                         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 ImportError(
-                                        "Non Array and Pointer type ctype imports not supported in current version"
-                                    )
-                            else:
-                                raise TypeError("Unsupported ctypes subclass")
-                        else:
-                            raise ImportError(
-                                f"Unsupported module of {containing_type}"
+                        # Unwrap all pointer layers to get the base type for dependency tracking
+                        base_type = unwrap_pointer_type(elem_type)
+                        base_type_module = getattr(base_type, "__module__", None)
+
+                        if base_type_module == "vmlinux":
+                            base_type_name = (
+                                base_type.__name__
+                                if hasattr(base_type, "__name__")
+                                else str(base_type)
                             )
+                            # ONLY add vmlinux types as dependencies
+                            new_dep_node.add_dependent(base_type_name)
+
                             logger.debug(
                                 f"{containing_type} containing type of parent {elem_name} with {elem_type} and ctype {ctype_complex_type} and length {type_length}"
                             )
@@ -191,11 +209,20 @@ def process_vmlinux_post_ast(
                                 elem_name, ctype_complex_type
                             )
                             new_dep_node.set_field_type(elem_name, elem_type)
-                        if containing_type.__module__ == "vmlinux":
+
+                            # Check the containing_type module to decide whether to recurse
+                            containing_type_module = getattr(
+                                containing_type, "__module__", None
+                            )
+                            if containing_type_module == "vmlinux":
+                                # Also unwrap containing_type to get base type name
+                                base_containing_type = unwrap_pointer_type(
+                                    containing_type
+                                )
                                 containing_type_name = (
-                                containing_type.__name__
-                                if hasattr(containing_type, "__name__")
-                                else str(containing_type)
+                                    base_containing_type.__name__
+                                    if hasattr(base_containing_type, "__name__")
+                                    else str(base_containing_type)
                                 )
 
                                 # Check for self-reference or already processed
@@ -212,21 +239,61 @@ def process_vmlinux_post_ast(
                                     )
                                     new_dep_node.set_field_ready(elem_name, True)
                                 else:
-                                # Process recursively - THIS WAS MISSING
+                                    # Process recursively - use base containing type, not the pointer wrapper
                                     new_dep_node.add_dependent(containing_type_name)
                                     process_vmlinux_post_ast(
-                                    containing_type,
+                                        base_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}")
+                            elif (
+                                containing_type_module == ctypes.__name__
+                                or containing_type_module is None
+                            ):
+                                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"
+                                    f"Module not supported in recursive resolution: {containing_type_module}"
+                                )
+                        elif (
+                            base_type_module == ctypes.__name__
+                            or base_type_module is None
+                        ):
+                            # Handle ctypes or types with no module (like some internal ctypes types)
+                            # DO NOT add ctypes as dependencies - just set field metadata and mark ready
+                            logger.debug(
+                                f"Base type {base_type} is ctypes - NOT adding as dependency, just processing field"
+                            )
+                            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 ImportError(
+                                        "Non Array and Pointer type ctype imports not supported in current version"
+                                    )
+                            else:
+                                raise TypeError("Unsupported ctypes subclass")
+
+                            # Set field metadata but DO NOT add dependency or recurse
+                            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)
+                            new_dep_node.set_field_ready(elem_name, True)
+                        else:
+                            raise ImportError(
+                                f"Unsupported module of {base_type}: {base_type_module}"
                             )
                     else:
                         new_dep_node.add_dependent(
@@ -245,9 +312,12 @@ def process_vmlinux_post_ast(
                     raise ValueError(
                         f"{elem_name} with type {elem_type} from module {module_name} not supported in recursive resolver"
                     )
-
+    elif module_name == ctypes.__name__ or module_name is None:
+        # Handle ctypes types - these don't need processing, just return
+        logger.debug(f"Skipping ctypes type {current_symbol_name}")
+        return True
     else:
-        raise ImportError("UNSUPPORTED Module")
+        raise ImportError(f"UNSUPPORTED Module {module_name}")
 
     logger.info(
         f"{current_symbol_name} processed and handler readiness {handler.is_ready}"

pythonbpf/vmlinux_parser/import_detector.py

@@ -11,7 +11,9 @@ from .class_handler import process_vmlinux_class
 logger = logging.getLogger(__name__)
 
 
-def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
+def detect_import_statement(
+    tree: ast.AST,
+) -> list[tuple[str, ast.ImportFrom, str, str]]:
     """
     Parse AST and detect import statements from vmlinux.
 
@@ -25,7 +27,7 @@ def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
         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
+        SyntaxError: If import * is used
     """
     vmlinux_imports = []
 
@@ -40,28 +42,19 @@ def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
                         "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
+                # Support multiple imports: from vmlinux import A, B, C
                 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}")
+                    # Use alias if provided, otherwise use the original name
+                    as_name = alias.asname if alias.asname else alias.name
+                    vmlinux_imports.append(("vmlinux", node, import_name, as_name))
+                    logger.info(f"Found vmlinux import: {import_name} as {as_name}")
 
         # Handle "import vmlinux" statements (not typical but should be rejected)
         elif isinstance(node, ast.Import):
@@ -103,39 +96,36 @@ def vmlinux_proc(tree: ast.AST, module):
     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
+    for import_mod, import_node, imported_name, as_name in import_statements:
         found = False
         for mod_node in mod_ast.body:
-                if (
-                    isinstance(mod_node, ast.ClassDef)
-                    and mod_node.name == imported_name
-                ):
+            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)
+                        process_vmlinux_assign(mod_node, module, assignments, as_name)
                         found = True
                         break
             if found:
                 break
         if not found:
-                logger.info(
-                    f"{imported_name} not found as ClassDef or Assign in vmlinux"
-                )
+            logger.info(f"{imported_name} not found as ClassDef or Assign in vmlinux")
 
     IRGenerator(module, handler, assignments)
     return assignments
 
 
-def process_vmlinux_assign(node, module, assignments: dict[str, AssignmentInfo]):
+def process_vmlinux_assign(
+    node, module, assignments: dict[str, AssignmentInfo], target_name=None
+):
     """Process assignments from vmlinux module."""
     # Only handle single-target assignments
     if len(node.targets) == 1 and isinstance(node.targets[0], ast.Name):
+        # Use provided target_name (for aliased imports) or fall back to original name
+        if target_name is None:
             target_name = node.targets[0].id
 
         # Handle constant value assignments

pythonbpf/vmlinux_parser/ir_gen/debug_info_gen.py

@@ -21,7 +21,7 @@ def debug_info_generation(
         generated_debug_info: List of tuples (struct, debug_info) to track generated debug info
 
     Returns:
-        The generated global variable debug info
+        The generated global variable debug info, or None for unsupported types
     """
     # Set up debug info generator
     generator = DebugInfoGenerator(llvm_module)
@@ -31,23 +31,39 @@ def debug_info_generation(
         if existing_struct.name == struct.name:
             return debug_info
 
+    # Check if this is a union (not supported yet)
+    if not struct.name.startswith("struct_"):
+        logger.warning(f"Skipping debug info generation for union: {struct.name}")
+        # Create a minimal forward declaration for unions
+        union_type = generator.create_struct_type(
+            [], struct.__sizeof__() * 8, is_distinct=True
+        )
+        return union_type
+
     # Process all fields and create members for the struct
     members = []
     for field_name, field in struct.fields.items():
+        try:
             # Get appropriate debug type for this field
             field_type = _get_field_debug_type(
                 field_name, field, generator, struct, generated_debug_info
             )
+
+            # Ensure field_type is a tuple
+            if not isinstance(field_type, tuple) or len(field_type) != 2:
+                logger.error(f"Invalid field_type for {field_name}: {field_type}")
+                continue
+
             # Create struct member with proper offset
             member = generator.create_struct_member_vmlinux(
                 field_name, field_type, field.offset * 8
             )
             members.append(member)
+        except Exception as e:
+            logger.error(f"Failed to process field {field_name} in {struct.name}: {e}")
+            continue
 
-    if struct.name.startswith("struct_"):
     struct_name = struct.name.removeprefix("struct_")
-    else:
-        raise ValueError("Unions are not supported in the current version")
     # Create struct type with all members
     struct_type = generator.create_struct_type_with_name(
         struct_name, members, struct.__sizeof__() * 8, is_distinct=True
@@ -74,11 +90,17 @@ def _get_field_debug_type(
         generated_debug_info: List of already generated debug info
 
     Returns:
-        The debug info type for this field
+        A tuple of (debug_type, size_in_bits)
     """
-    # Handle complex types (arrays, pointers)
+    # Handle complex types (arrays, pointers, function pointers)
     if field.ctype_complex_type is not None:
-        if issubclass(field.ctype_complex_type, ctypes.Array):
+        # Handle function pointer types (CFUNCTYPE)
+        if callable(field.ctype_complex_type):
+            # Function pointers are represented as void pointers
+            logger.info(f"Field {field_name} is a function pointer, using void pointer")
+            void_ptr = generator.create_pointer_type(None, 64)
+            return void_ptr, 64
+        elif issubclass(field.ctype_complex_type, ctypes.Array):
             # Handle array types
             element_type, base_type_size = _get_basic_debug_type(
                 field.containing_type, generator
@@ -100,11 +122,13 @@ def _get_field_debug_type(
         for existing_struct, debug_info in generated_debug_info:
             if existing_struct.name == struct_name:
                 # Use existing debug info
-                return debug_info, existing_struct.__sizeof__()
+                return debug_info, existing_struct.__sizeof__() * 8
 
         # If not found, create a forward declaration
         # This will be completed when the actual struct is processed
-        logger.warning("Forward declaration in struct created")
+        logger.warning(
+            f"Forward declaration created for {struct_name} in {parent_struct.name}"
+        )
         forward_type = generator.create_struct_type([], 0, is_distinct=True)
         return forward_type, 0
 

pythonbpf/vmlinux_parser/ir_gen/ir_generation.py

@@ -11,6 +11,10 @@ logger = logging.getLogger(__name__)
 
 
 class IRGenerator:
+    # This field keeps track of the non_struct names to avoid duplicate name errors.
+    type_number = 0
+    unprocessed_store: list[str] = []
+
     # get the assignments dict and add this stuff to it.
     def __init__(self, llvm_module, handler: DependencyHandler, assignments):
         self.llvm_module = llvm_module
@@ -129,7 +133,19 @@ class IRGenerator:
 
         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(
+            if callable(field.ctype_complex_type):
+                # Function pointer case - generate a simple field accessor
+                field_co_re_name, returned = 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)
+                self.generated_field_names[struct.name][field_name] = globvar
+            elif field.ctype_complex_type is not None and issubclass(
                 field.ctype_complex_type, ctypes.Array
             ):
                 array_size = field.type_size
@@ -137,7 +153,7 @@ class IRGenerator:
                 if containing_type.__module__ == ctypes.__name__:
                     containing_type_size = ctypes.sizeof(containing_type)
                     if array_size == 0:
-                        field_co_re_name = self._struct_name_generator(
+                        field_co_re_name, returned = self._struct_name_generator(
                             struct, field, field_index, True, 0, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
@@ -149,7 +165,7 @@ class IRGenerator:
                         field_index += 1
                         continue
                     for i in range(0, array_size):
-                        field_co_re_name = self._struct_name_generator(
+                        field_co_re_name, returned = self._struct_name_generator(
                             struct, field, field_index, True, i, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
@@ -163,12 +179,28 @@ class IRGenerator:
                 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
+                    # Unwrap all pointer layers to get the base struct type
+                    base_containing_type = containing_type
+                    while hasattr(base_containing_type, "_type_"):
+                        next_type = base_containing_type._type_
+                        # Stop if _type_ is a string (like 'c' for c_char)
+                        # TODO: stacked pointers not handl;ing ctypes check here as well
+                        if isinstance(next_type, str):
+                            break
+                        base_containing_type = next_type
+
+                    # Get the base struct name
+                    base_struct_name = (
+                        base_containing_type.__name__
+                        if hasattr(base_containing_type, "__name__")
+                        else str(base_containing_type)
+                    )
+
+                    # Look up the size using the base struct name
+                    containing_type_size = self.handler[base_struct_name].current_offset
+                    if array_size == 0:
+                        field_co_re_name, returned = self._struct_name_generator(
+                            struct, field, field_index, True, 0, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
                             self.llvm_module, ir.IntType(64), name=field_co_re_name
@@ -178,7 +210,28 @@ class IRGenerator:
                         self.generated_field_names[struct.name][field_name] = globvar
                         field_index += 1
                     else:
-                field_co_re_name = self._struct_name_generator(
+                        for i in range(0, array_size):
+                            field_co_re_name, returned = 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
+                            )
+                            self.generated_field_names[struct.name][field_name] = (
+                                globvar
+                            )
+                        field_index += 1
+            else:
+                field_co_re_name, returned = self._struct_name_generator(
                     struct, field, field_index
                 )
                 field_index += 1
@@ -198,7 +251,7 @@ class IRGenerator:
         is_indexed: bool = False,
         index: int = 0,
         containing_type_size: int = 0,
-    ) -> str:
+    ) -> tuple[str, bool]:
         # TODO: Does not support Unions as well as recursive pointer and array type naming
         if is_indexed:
             name = (
@@ -208,7 +261,7 @@ class IRGenerator:
                 + "$"
                 + f"0:{field_index}:{index}"
             )
-            return name
+            return name, True
         elif struct.name.startswith("struct_"):
             name = (
                 "llvm."
@@ -217,9 +270,18 @@ class IRGenerator:
                 + "$"
                 + f"0:{field_index}"
             )
-            return name
+            return name, True
         else:
-            print(self.handler[struct.name])
-            raise TypeError(
-                "Name generation cannot occur due to type name not starting with struct"
+            logger.warning(
+                "Blindly handling non-struct type to avoid type errors in vmlinux IR generation. Possibly a union."
             )
+            self.type_number += 1
+            unprocessed_type = "unprocessed_type_" + str(self.handler[struct.name].name)
+            if self.unprocessed_store.__contains__(unprocessed_type):
+                return unprocessed_type + "_" + str(self.type_number), False
+            else:
+                self.unprocessed_store.append(unprocessed_type)
+                return unprocessed_type, False
+            # raise TypeError(
+            #     "Name generation cannot occur due to type name not starting with struct"
+            # )

tests/c-form/Makefile

@@ -3,21 +3,20 @@ CFLAGS := -emit-llvm -target bpf -c
 
 SRC := $(wildcard *.bpf.c)
 LL := $(SRC:.bpf.c=.bpf.ll)
-LL2 := $(SRC:.bpf.c=.bpf.o2.ll)
 OBJ := $(SRC:.bpf.c=.bpf.o)
-
+LL0 := $(SRC:.bpf.c=.bpf.o0.ll)
 .PHONY: all clean
 
-all: $(LL) $(OBJ) $(LL2)
+all: $(LL) $(OBJ) $(LL0)
 
 %.bpf.o: %.bpf.c
 	$(BPF_CLANG) -O2 -g -target bpf -c $< -o $@
 
 %.bpf.ll: %.bpf.c
-	$(BPF_CLANG) -O0 $(CFLAGS) -g -S $< -o $@
+	$(BPF_CLANG) $(CFLAGS) -O2 -g -S $< -o $@
 
-%.bpf.o2.ll: %.bpf.c
-	$(BPF_CLANG) -O2 $(CFLAGS) -g -S $< -o $@
+%.bpf.o0.ll: %.bpf.c
+	$(BPF_CLANG) $(CFLAGS) -O0 -g -S $< -o $@
 
 clean:
-	rm -f $(LL) $(OBJ) $(LL2)
+	rm -f $(LL) $(OBJ) $(LL0)

tests/c-form/requests.bpf.c

@@ -0,0 +1,15 @@
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char LICENSE[] SEC("license") = "GPL";
+
+SEC("kprobe/blk_mq_start_request")
+int example(struct pt_regs *ctx)
+{
+    struct request *req = (struct request *)(ctx->di);
+    u32 data_len = req->__data_len;
+    bpf_printk("data length %u\n", data_len);
+
+    return 0;
+}

tests/failing_tests/vmlinux/assignment_handling.py

@@ -0,0 +1,22 @@
+from vmlinux import XDP_PASS
+from pythonbpf import bpf, section, bpfglobal, compile_to_ir
+import logging
+from ctypes import c_int64, c_void_p
+
+
+@bpf
+@section("kprobe/blk_mq_start_request")
+def example(ctx: c_void_p) -> c_int64:
+    d = XDP_PASS  # This gives an error, but
+    e = XDP_PASS + 0  # this does not
+    print(f"test1 {e} test2 {d}")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("assignment_handling.py", "assignment_handling.ll", loglevel=logging.INFO)

tests/failing_tests/vmlinux/requests.py

@@ -0,0 +1,22 @@
+from vmlinux import struct_request, struct_pt_regs
+from pythonbpf import bpf, section, bpfglobal, compile_to_ir
+import logging
+from ctypes import c_int64
+
+
+@bpf
+@section("kprobe/blk_mq_start_request")
+def example(ctx: struct_pt_regs) -> c_int64:
+    req = struct_request(ctx.di)
+    c = req.__data_len
+    print(f"data length {c}")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("requests.py", "requests.ll", loglevel=logging.INFO)

tests/passing_tests/vmlinux/requests2.py

@@ -0,0 +1,21 @@
+from vmlinux import struct_pt_regs
+from pythonbpf import bpf, section, bpfglobal, compile_to_ir
+import logging
+from ctypes import c_int64
+
+
+@bpf
+@section("kprobe/blk_mq_start_request")
+def example(ctx: struct_pt_regs) -> c_int64:
+    req = ctx.di
+    print(f"data length {req}")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("requests2.py", "requests2.ll", loglevel=logging.INFO)