Setup a sanitized build for CSP

Signed-off-by: Adam Glustein <[email protected]>

Author: AdamGlustein

CMakeLists.txt

@@ -67,6 +67,8 @@ option(CSP_MANYLINUX "Build for python's manylinux setup" OFF)
 option(CSP_USE_VCPKG "Build with vcpkg dependencies" ON)
 option(CSP_USE_CCACHE "Build with ccache caching" OFF)
 option(CSP_USE_LD_CLASSIC_MAC "On macOS, link with ld_classic" OFF)
+option(CSP_ENABLE_ASAN "Build with address sanitizer" OFF)
+option(CSP_ENABLE_UBSAN "Build with undefined behavior sanitizer" OFF)
 
 # Extension options
 option(CSP_BUILD_KAFKA_ADAPTER "Build kafka adapter" ON)
@@ -213,6 +215,18 @@ else()
     endif()
 endif()
 
+if(CSP_ENABLE_ASAN)
+    message(STATUS "Enabling Address Sanitizer")
+    add_compile_options(-fsanitize=address -fno-omit-frame-pointer)
+    add_link_options(-fsanitize=address -fno-omit-frame-pointer)
+endif()
+
+if(CSP_ENABLE_UBSAN)
+    message(STATUS "Enabling Undefined Behavior Sanitizer")
+    add_compile_options(-fsanitize=undefined -fno-omit-frame-pointer)
+    add_link_options(-fsanitize=undefined -fno-omit-frame-pointer)
+endif()
+
 
 ###################################################################################################################################################
 # Messages #

Makefile

@@ -20,12 +20,18 @@ develop: requirements  ## install dependencies and build library
 build:  ## build the library
 	python setup.py build build_ext --inplace
 
+build-sanitizer:  ## build the library with ASAN and UBSAN enabled
+	CSP_ENABLE_ASAN=1 CSP_ENABLE_UBSAN=1 python setup.py build build_ext --inplace
+
 build-debug:  ## build the library ( DEBUG ) - May need a make clean when switching from regular build to build-debug and vice versa
 	SKBUILD_CONFIGURE_OPTIONS="" DEBUG=1 python setup.py build build_ext --inplace
 
 build-conda:  ## build the library in Conda
 	python setup.py build build_ext --csp-no-vcpkg --inplace
 
+build-conda-sanitizer:  ## build the library in Conda with ASAN and UBSAN enabled
+	CSP_ENABLE_ASAN=1 CSP_ENABLE_UBSAN=1 python setup.py build build_ext --csp-no-vcpkg --inplace
+
 install:  ## install library
 	python -m pip install .
 
@@ -84,6 +90,10 @@ TEST_ARGS :=
 test-py: ## Clean and Make unit tests
 	python -m pytest -v csp/tests --junitxml=junit.xml $(TEST_ARGS)
 
+test-py-sanitizer: ## Clean and Make unit tests with sanitizers enabled
+	ASAN_OPTIONS=detect_leaks=0,detect_stack_use_after_return=true,use_odr_indicator=1,strict_init_order=true,strict_string_checks=true LD_PRELOAD=$$(gcc -print-file-name=libasan.so) \
+	python -m pytest -v csp/tests --junitxml=junit.xml $(TEST_ARGS)
+
 test-cpp: ## Make C++ unit tests
 ifneq ($(OS),Windows_NT)
 	for f in ./csp/tests/bin/*; do $$f; done || (echo "TEST FAILED" && exit 1)

cpp/cmake/modules/Findcsp_autogen.cmake

@@ -26,8 +26,20 @@ function(csp_autogen MODULE_NAME DEST_FILENAME HEADER_NAME_OUTVAR SOURCE_NAME_OU
         set(CSP_AUTOGEN_PYTHONPATH ${PROJECT_BINARY_DIR}/lib:${CMAKE_SOURCE_DIR}:$$PYTHONPATH)
     endif()
 
+    if(CSP_ENABLE_ASAN)
+        execute_process(
+            COMMAND gcc -print-file-name=libasan.so
+            OUTPUT_VARIABLE ASAN_LIB_PATH
+            OUTPUT_STRIP_TRAILING_WHITESPACE
+        )
+        # Turn off leak checks as we are using PyMalloc when we run autogen
+        set(ASAN_PRELOAD_CMD "ASAN_OPTIONS=detect_leaks=0" "LD_PRELOAD=${ASAN_LIB_PATH}")
+    else()
+        set(ASAN_PRELOAD_CMD "")
+    endif()
+
     add_custom_command(OUTPUT "${CSP_AUTOGEN_CPP_OUT}" "${CSP_AUTOGEN_H_OUT}"
-        COMMAND ${CMAKE_COMMAND} -E env "PYTHONPATH=${CSP_AUTOGEN_PYTHONPATH}" ${Python_EXECUTABLE} ${CSP_AUTOGEN_MODULE_PATH} -m ${MODULE_NAME} -d ${CSP_AUTOGEN_DESTINATION_FOLDER} -o ${DEST_FILENAME} ${CSP_AUTOGEN_EXTRA_ARGS}
+        COMMAND ${CMAKE_COMMAND} -E env "PYTHONPATH=${CSP_AUTOGEN_PYTHONPATH}" ${ASAN_PRELOAD_CMD} ${Python_EXECUTABLE} ${CSP_AUTOGEN_MODULE_PATH} -m ${MODULE_NAME} -d ${CSP_AUTOGEN_DESTINATION_FOLDER} -o ${DEST_FILENAME} ${CSP_AUTOGEN_EXTRA_ARGS}
         COMMENT "generating csp c++ types from module ${MODULE_NAME}"
         DEPENDS mkdir_autogen_${MODULE_TARGETNAME} 
                     ${CSP_AUTOGEN_MODULE_PATH}

cpp/csp/core/DynamicBitSet.h

@@ -141,7 +141,8 @@ class DynamicBitSet
         {
             node_type * old = m_nodes;
             m_nodes = new node_type[ newNodes ];
-            memcpy( m_nodes, old, m_numNodes * sizeof( node_type ) );
+            if( likely( m_numNodes > 0 ) )
+                memcpy( m_nodes, old, m_numNodes * sizeof( node_type ) );
             memset( m_nodes + m_numNodes, 0, ( newNodes - m_numNodes ) * sizeof( node_type ) );
 
             m_numNodes = newNodes;

cpp/csp/core/Exception.cpp

@@ -34,7 +34,7 @@ static void printBacktrace( char ** messages, int size, std::ostream & dest )
     {
         char *begin_name = 0, *begin_offset = 0;
         char tmp[1024];
-        strncpy( tmp, messages[i], sizeof(tmp) );
+        strncpy( tmp, messages[i], sizeof(tmp) - 1 );
         tmp[ sizeof( tmp ) - 1 ] = 0;
 
         // find parentheses and +address offset surrounding the mangled name:

cpp/csp/python/cspbaselibimpl.cpp

@@ -273,6 +273,14 @@ DECLARE_CPPNODE( exprtk_impl )
             csp.make_passive( inputs );
     }
 
+    virtual ~exprtk_impl() 
+    {
+        // Need to release the expression before clearing values/symbol table
+        // https://github.com/ArashPartow/exprtk/blob/cc1b800c2bd1ac3ac260478c915d2aec6f4eb41c/readme.txt#L909
+        s_expr.release();
+        s_valuesContainer.clear();
+    }
+
     INVOKE()
     {
         if( use_trigger )

cpp/csp/python/npstatsimpl.cpp

@@ -1303,12 +1303,12 @@ DECLARE_CPPNODE( _np_arg_min_max )
             PyArray_Descr *descr;
             PyArray_DescrConverter( date_type, &descr );
             Py_XDECREF( date_type );
-            DateTime * values = new DateTime[s_elem.size()];
+            
+            PyObject * out = PyArray_NewFromDescr( &PyArray_Type, descr, s_shp.m_dims.size(), &s_shp.m_dims[0], NULL, NULL, 0, NULL );
+            DateTime * values = static_cast<DateTime *>( PyArray_DATA( ( PyArrayObject * )out ) );
             for( size_t i = 0; i < s_elem.size(); ++i )
                 values[i] = s_elem[i].compute_dt();
-
-            PyObject * out = PyArray_NewFromDescr( &PyArray_Type, descr, s_shp.m_dims.size(), &s_shp.m_dims[0], NULL, values, 0, NULL );
-            PyArray_ENABLEFLAGS( ( PyArrayObject * ) out, NPY_ARRAY_OWNDATA );
+            
             RETURN( PyObjectPtr::own( out ) );
         }
     }

cpp/tests/engine/test_tick_buffer.cpp

@@ -98,4 +98,8 @@ TEST( TickBufferTest, test_flatten )
         ASSERT_EQ( values_wrap[ i ], i + 3 );
         ASSERT_EQ( values_nowrap[ i ], i );
     }
+
+    free( values_wrap );
+    free( values_nowrap );
+    free( values_single );
 }

csp/tests/test_engine.py

@@ -14,6 +14,7 @@
 
 import numpy as np
 import psutil
+import pytest
 
 import csp
 from csp import PushMode, ts
@@ -1175,6 +1176,10 @@ def list_comprehension_bug_graph():
         rv = csp.run(list_comprehension_bug_graph, starttime=datetime(2020, 1, 1))["Bucket"]
         self.assertEqual([v[1][0] for v in rv[10:]], list(range(20)))
 
+    @unittest.skipIf(
+        os.environ.get("ASAN_OPTIONS") is not None,
+        reason="Test skipped when AddressSanitizer is enabled, RSS usage is much larger than usual",
+    )
     def test_alarm_leak(self):
         """this was a leak in Scheduler.cpp"""
 

setup.py

@@ -23,6 +23,8 @@
     ("CSP_BUILD_KAFKA_ADAPTER", "1"),
     ("CSP_BUILD_PARQUET_ADAPTER", "1"),
     ("CSP_BUILD_WS_CLIENT_ADAPTER", "1"),
+    ("CSP_ENABLE_ASAN", "0"),
+    ("CSP_ENABLE_UBSAN", "0"),
     # NOTE:
     # - omit vcpkg, need to test for presence
     # - omit ccache, need to test for presence