bump version to v0.1.5

Adds IR and debug info generation capabilities for vmlinux imported structs

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "pythonbpf"
-version = "0.1.4"
+version = "0.1.5"
 description = "Reduced Python frontend for eBPF"
 authors = [
   { name = "r41k0u", email="pragyanshchaturvedi18@gmail.com" },

pythonbpf/codegen.py

@@ -19,10 +19,20 @@ from pylibbpf import BpfProgram
 import tempfile
 from logging import Logger
 import logging
+import re
 
 logger: Logger = logging.getLogger(__name__)
 
-VERSION = "v0.1.4"
+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)
 
 
 def find_bpf_chunks(tree):
@@ -121,10 +131,12 @@ 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}")
     with open(output, "w") as f:
         f.write(f'source_filename = "{filename}"\n')
-        f.write(str(module))
+        f.write(module_string)
         f.write("\n")
 
     return output

pythonbpf/vmlinux_parser/class_handler.py

@@ -60,6 +60,10 @@ def process_vmlinux_post_ast(
             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
@@ -112,7 +116,11 @@ def process_vmlinux_post_ast(
                             type_length = elem_type._length_
 
                         if containing_type.__module__ == "vmlinux":
-                            pass
+                            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):
@@ -137,10 +145,35 @@ def process_vmlinux_post_ast(
                         )
                         new_dep_node.set_field_type(elem_name, elem_type)
                         if containing_type.__module__ == "vmlinux":
-                            process_vmlinux_post_ast(
-                                containing_type, llvm_handler, handler, processing_stack
+                            containing_type_name = (
+                                containing_type.__name__
+                                if hasattr(containing_type, "__name__")
+                                else str(containing_type)
                             )
-                            new_dep_node.set_field_ready(elem_name, True)
+
+                            # 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)
@@ -149,6 +182,11 @@ def process_vmlinux_post_ast(
                                 "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
                         )

pythonbpf/vmlinux_parser/dependency_handler.py

@@ -147,3 +147,27 @@ class DependencyHandler:
             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

pythonbpf/vmlinux_parser/dependency_node.py

@@ -1,5 +1,6 @@
 from dataclasses import dataclass, field
 from typing import Dict, Any, Optional
+import ctypes
 
 
 # TODO: FIX THE FUCKING TYPE NAME CONVENTION.
@@ -13,6 +14,7 @@ class Field:
     containing_type: Optional[Any]
     type_size: Optional[int]
     bitfield_size: Optional[int]
+    offset: int
     value: Any = None
     ready: bool = False
 
@@ -60,6 +62,10 @@ class Field:
         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:
@@ -106,8 +112,11 @@ class DependencyNode:
     """
 
     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,
@@ -119,8 +128,11 @@ class DependencyNode:
         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,
@@ -130,10 +142,21 @@ class DependencyNode:
             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]
@@ -200,15 +223,112 @@ class DependencyNode:
         # Invalidate readiness cache
         self._ready_cache = None
 
-    def set_field_ready(self, name: str, is_ready: bool = False) -> 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)."""
@@ -235,3 +355,9 @@ class DependencyNode:
     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)

pythonbpf/vmlinux_parser/import_detector.py

@@ -1,11 +1,11 @@
 import ast
 import logging
-from typing import List, Tuple, Dict
+from typing import List, Tuple, Any
 import importlib
 import inspect
 
 from .dependency_handler import DependencyHandler
-from .ir_generation import IRGenerator
+from .ir_gen import IRGenerator
 from .class_handler import process_vmlinux_class
 
 logger = logging.getLogger(__name__)
@@ -82,7 +82,7 @@ def vmlinux_proc(tree: ast.AST, module):
     # initialise dependency handler
     handler = DependencyHandler()
     # initialise assignment dictionary of name to type
-    assignments: Dict[str, type] = {}
+    assignments: dict[str, tuple[type, Any]] = {}
 
     if not import_statements:
         logger.info("No vmlinux imports found")
@@ -129,7 +129,19 @@ def vmlinux_proc(tree: ast.AST, module):
                 )
 
     IRGenerator(module, handler)
+    return assignments
 
 
-def process_vmlinux_assign(node, module, assignments: Dict[str, type]):
-    raise NotImplementedError("Assignment handling has not been implemented yet")
+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")

pythonbpf/vmlinux_parser/ir_gen/__init__.py

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

pythonbpf/vmlinux_parser/ir_gen/debug_info_gen.py

@@ -0,0 +1,15 @@
+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

pythonbpf/vmlinux_parser/ir_gen/ir_generation.py

@@ -0,0 +1,161 @@
+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"
+            )

pythonbpf/vmlinux_parser/ir_generation.py

@@ -1,14 +0,0 @@
-import logging
-from .dependency_handler import DependencyHandler
-
-logger = logging.getLogger(__name__)
-
-
-class IRGenerator:
-    def __init__(self, module, handler: DependencyHandler):
-        self.module = module
-        self.handler: DependencyHandler = handler
-        if not handler.is_ready:
-            raise ImportError(
-                "Semantic analysis of vmlinux imports failed. Cannot generate IR"
-            )

tests/c-form/ex7.bpf.c

@@ -1,23 +1,9 @@
 // SPDX-License-Identifier: GPL-2.0
 
-#include <linux/bpf.h>
+#include "vmlinux.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;

tests/failing_tests/xdp_pass.py

@@ -1,10 +1,17 @@
 from pythonbpf import bpf, map, section, bpfglobal, compile_to_ir
 from pythonbpf.maps import HashMap
 from pythonbpf.helper import XDP_PASS
-from vmlinux import struct_xdp_md
-from vmlinux import struct_xdp_buff  # noqa: F401
-from vmlinux import struct_ring_buffer_per_cpu  # noqa: F401
+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