Issue SimonKagstrom#92: system-mode: Implement system mode recording …

…via dyninst Instruments binaries using dyninst.
clayne · Apr 18, 2017 · e43715b · e43715b
1 parent ac63ebf
commit e43715b
Show file tree

Hide file tree

Showing 16 changed files with 1,525 additions and 14 deletions.
diff --git a/INSTALL.md b/INSTALL.md
@@ -8,14 +8,17 @@ and (optional) binutils and libiberty to build kcov. Note that elfutils is
 found in multiple variants, and at least in RH/Centos/Fedora you'll need
 elfutils-devel and *not* elfutils-libelf-devel.
 
+On Linux, if [dyninst](http://www.dyninst.org) is present, kcov will also be
+able to do full-system instrumentation.
+
 Ubuntu
 ------
 Install binutils-dev libcurl4-openssl-dev zlib1g-dev libdw-dev libiberty-dev
 
 
 Fedora / Centos / RHEL
 ----------------------
-Install elfutils-libelf-devel libcurl-devel binutils-devel elfutils-devel
+Install elfutils-libelf-devel libcurl-devel binutils-devel elfutils-devel dyninst-devel
 
 Mac OS X
 --------

diff --git a/README.md b/README.md
@@ -84,6 +84,11 @@ after_success:
 
 the -s /path/to/outdir part can be skipped if kcov produces output in the current directory.
 
+Full-system instrumentation
+---------------------------
+Using [dyninst](http://www.dyninst.org), Kcov can instrument all binaries with very low overhead for embedded systems.
+Refer to the [full system instrumentation](doc/full-system-instrumentation.md) documentation for details.
+
 More information
 ----------------
 kcov is written by Simon Kagstrom <[email protected]> and more

diff --git a/cmake/FindDyninst.cmake b/cmake/FindDyninst.cmake
@@ -0,0 +1,58 @@
+# - Try to find dyninst
+# Once done this will define
+#
+#  DYNINST_FOUND - system has dyninst
+#  DYNINST_INCLUDE_DIRS - the dyninst include directory
+#  DYNINST_LIBRARIES - Link these to use dyninst
+#  DYNINST_DEFINITIONS - Compiler switches required for using dyninst
+#
+
+if (DYNINST_LIBRARIES AND DYNINST_INCLUDE_DIRS)
+  set (Dyninst_FIND_QUIETLY TRUE)
+endif (DYNINST_LIBRARIES AND DYNINST_INCLUDE_DIRS)
+
+find_path (DYNINST_INCLUDE_DIR
+    NAMES
+      BPatch.h
+    PATHS
+      /usr/include/dyninst
+      /usr/local/include/dyninst
+      /opt/local/include/dyninst
+      /sw/include/dyninst
+      /sw/include/dyninst
+      /usr/local/include
+      ENV CPATH) # PATH and INCLUDE will also work
+if (DYNINST_INCLUDE_DIR)
+    set (DYNINST_INCLUDE_DIRS  ${DYNINST_INCLUDE_DIR})
+endif (DYNINST_INCLUDE_DIR)
+
+find_library (DYNINST_LIBRARIES
+    NAMES
+      dyninstAPI
+    PATHS
+      /usr/lib
+      /usr/lib64
+      /usr/local/lib
+      /opt/local/lib
+      /usr/local/lib64
+      /opt/local/lib64
+      /sw/lib
+      /usr/lib/dyninst
+      /usr/lib64/dyninst
+      /usr/local/lib/dyninst
+      /opt/local/lib/dyninst
+      /usr/local/lib64/dyninst
+      /opt/local/lib64/dyninst
+      ENV LIBRARY_PATH   # PATH and LIB will also work
+      ENV LD_LIBRARY_PATH)
+
+include (FindPackageHandleStandardArgs)
+
+
+# handle the QUIETLY and REQUIRED arguments and set DYNINST_FOUND to TRUE
+# if all listed variables are TRUE
+FIND_PACKAGE_HANDLE_STANDARD_ARGS(DYNINST DEFAULT_MSG
+    DYNINST_LIBRARIES
+    DYNINST_INCLUDE_DIR)
+
+mark_as_advanced(DYNINST_INCLUDE_DIR DYNINST_LIBRARIES)
diff --git a/doc/full-system-instrumentation.md b/doc/full-system-instrumentation.md
@@ -0,0 +1,22 @@
+## *Full system instrumentation with kcov*
+Using [dyninst](http://www.dyninst.org), Kcov can instrument all binaries with very low overhead for embedded systems.
+
+Instrumenting binaries for a target system
+------------------------------------------
+If your binaries (with debug symbols) are in e.g., sysroot, you can instrument binaries using
+
+```
+kcov-system --system-record /tmp/out-sysroot sysroot
+```
+
+After this finishes, build your filesystem (squashfs etc) using the instrumented binaries
+and install and run on your target.
+
+
+Creating a report from the collected data
+-----------------------------------------
+Copy `/tmp/kcov-data` from your target system and run
+
+```
+kcov-system --system-report /tmp/kcov /tmp/kcov-data
+```
diff --git a/doc/kcov.1 b/doc/kcov.1
@@ -103,6 +103,12 @@ executed as /bin/bash). Does not work well on some systems, so the default is no
 .TP
 \fB\-\-replace\-src\-path\fP=\fIP1\fP:\fIP2\fP
 Replace source file path P1 with P2, if found.
+.TP
+\fB\-\-system\-record
+Perform full-system instrumentation on a sysroot, outputting patched binaries which collect coverage data
+.TP
+\fB\-\-system\-report
+Produce coverage output for a full-system coverage run.
 .RE
 .SH EXAMPLES
 .PP

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -5,6 +5,7 @@ set (CMAKE_MODULE_PATH  ${CMAKE_MODULE_PATH}
                         ${CMAKE_CURRENT_SOURCE_DIR}/../../cmake)
 find_package (Threads)
 find_package (Bfd)
+find_package (Dyninst)
 
 include(TargetArch)
 target_architecture(CMAKE_TARGET_ARCHITECTURES)
@@ -51,6 +52,9 @@ set (DISASSEMBLER_SRCS
 	parsers/dummy-disassembler.cc
 )
 
+set (DYNINST_SRCS
+)
+
 set (HAS_LIBBFD "0")
 
 if("${CMAKE_TARGET_ARCHITECTURES}" STREQUAL "i386" OR "${CMAKE_TARGET_ARCHITECTURES}" STREQUAL "x86_64")
@@ -67,6 +71,13 @@ if("${CMAKE_TARGET_ARCHITECTURES}" STREQUAL "i386" OR "${CMAKE_TARGET_ARCHITECTU
 	endif()
 endif()
 
+
+if(DYNINST_FOUND)
+	set (DYNINST_SRCS
+		engines/dyninst-engine.cc
+	)
+endif()
+
 set (coveralls_SRCS writers/coveralls-writer.cc)
 
 if ("${KCOV_STATIC_BUILD}" STREQUAL "1")
@@ -84,6 +95,7 @@ set (ELF_SRCS
 set (MACHO_SRCS
 )
 set (SOLIB_generated )
+set (DYNINST_SOLIB_generated )
 
 # Linux-specific sources
 if (${CMAKE_SYSTEM_NAME} MATCHES Linux)
@@ -131,6 +143,7 @@ set (${KCOV}_SRCS
     configuration.cc
     engine-factory.cc
     engines/bash-engine.cc
+    engines/dyninst-file-format.cc
     engines/gcov-engine.cc
     engines/python-engine.cc
     filter.cc
@@ -141,7 +154,7 @@ set (${KCOV}_SRCS
     ${DISASSEMBLER_SRCS}
     parser-manager.cc
     reporter.cc
-	source-file-cache.cc
+    source-file-cache.cc
     utils.cc
     writers/cobertura-writer.cc
     writers/json-writer.cc
@@ -170,6 +183,51 @@ set (${KCOV}_SRCS
     include/phdr_data.h
     )
 
+# Should be some better way of doing this...
+set (KCOV_DYNINST_SRCS
+    capabilities.cc
+    collector.cc
+    configuration.cc
+    dummy-solib-handler.cc
+    engine-factory.cc
+    engines/bash-engine.cc
+    engines/dyninst-engine.cc
+    engines/dyninst-file-format.cc
+    engines/gcov-engine.cc
+    engines/python-engine.cc
+    filter.cc
+    gcov.cc
+    main.cc
+    merge-file-parser.cc
+    output-handler.cc
+    parser-manager.cc
+    reporter.cc
+    source-file-cache.cc
+    utils.cc
+    writers/cobertura-writer.cc
+    writers/json-writer.cc
+    ${coveralls_SRCS}
+    writers/html-writer.cc
+    writers/sonarqube-xml-writer.cc
+    writers/writer-base.cc
+    include/capabilities.hh
+    include/gcov.hh
+    include/reporter.hh
+    include/collector.hh
+    include/generated-data-base.hh
+    include/solib-handler.hh
+    include/configuration.hh
+    include/lineid.hh
+    include/swap-endian.hh
+    include/engine.hh
+    include/manager.hh
+    include/utils.hh
+    include/file-parser.hh
+    include/output-handler.hh
+    include/writer.hh
+    include/filter.hh
+    include/phdr_data.h
+)
 
 set (KCOV_LIBRARY_PREFIX "/tmp")
 
@@ -182,6 +240,7 @@ include_directories(
 	${LIBDW_INCLUDE_DIRS}
 	${LIBZ_INCLUDE_DIRS}
 	${LIBCURL_INCLUDE_DIRS}
+	${DYNINST_INCLUDE_DIRS}
 	)
 
 link_directories (/home/ska/local/lib)
@@ -197,6 +256,12 @@ add_custom_command(
    DEPENDS ${SOLIB} ${CMAKE_CURRENT_SOURCE_DIR}/bin-to-c-source.py
    )
 
+add_custom_command(
+   OUTPUT dyninst-binary-library.cc
+   COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/bin-to-c-source.py libkcov-binary-dyninst.so __dyninst_binary_library > dyninst-binary-library.cc
+   DEPENDS kcov-binary-dyninst ${CMAKE_CURRENT_SOURCE_DIR}/bin-to-c-source.py
+   )
+
 add_custom_command(
    OUTPUT bash-redirector-library.cc
    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/bin-to-c-source.py $<TARGET_FILE:bash_execve_redirector> bash_redirector_library > bash-redirector-library.cc
@@ -278,5 +343,25 @@ target_link_libraries(${KCOV}
 	${LLDB_LIBRARY}
 )
 
+if(DYNINST_FOUND)
+	add_library (kcov-binary-dyninst SHARED engines/dyninst-binary-lib.cc engines/dyninst-file-format.cc utils.cc)
+	set_target_properties(kcov-binary-dyninst PROPERTIES SUFFIX ".so")
+
+    add_executable (kcov-system ${KCOV_DYNINST_SRCS} ${SOLIB_generated} bash-redirector-library.cc dyninst-binary-library.cc python-helper.cc bash-helper.cc html-data-files.cc version.c)
+
+    target_link_libraries(kcov-system
+        ${DYNINST_LIBRARIES}
+        ${LIBELF_LIBRARIES}
+        stdc++
+        ${LIBCURL_LIBRARIES}
+        m
+        ${DISASSEMBLER_LIBRARIES}
+        ${LIBZ_LIBRARIES}
+        ${CMAKE_THREAD_LIBS_INIT}
+        dl
+        ${LLDB_LIBRARY}
+)
+endif()
+
 
 install (TARGETS ${PROJECT_NAME} ${INSTALL_TARGETS_PATH})
diff --git a/src/configuration.cc b/src/configuration.cc
@@ -94,7 +94,11 @@ class Configuration : public IConfiguration
 				"                                                       # including /src/frodo\n"
 				"  kcov --collect-only /tmp/kcov ./frodo  # Collect coverage, don't report\n"
 				"  kcov --report-only /tmp/kcov ./frodo   # Report coverage collected above\n"
-				"  kcov --merge /tmp/out /tmp/dir1 /tmp/dir2    # Merge the dir1/dir2 reports\n"
+				"  kcov --merge /tmp/out /tmp/dir1 /tmp/dir2     # Merge the dir1/dir2 reports\n"
+				"  kcov --system-record /tmp/out-dir sysroot     # Perform full-system in-\n"
+				"                                                  strumentation for sysroot\n"
+				"  kcov --system-report  /tmp/data-dir           # Report all data from a full-\n"
+				"                                                  system run.\n"
 				"",
 				keyAsInt("low-limit"), keyAsInt("high-limit"),
 				uncommonOptions().c_str());
@@ -138,6 +142,8 @@ class Configuration : public IConfiguration
 				{"python-parser", required_argument, 0, 'P'},
 				{"bash-parser", required_argument, 0, 'B'},
 				{"bash-method", required_argument, 0, '4'},
+				{"system-record", no_argument, 0, '8'},
+				{"system-report", no_argument, 0, '9'},
 				{"verify", no_argument, 0, 'V'},
 				{"version", no_argument, 0, 'v'},
 				{"uncommon-options", no_argument, 0, 'U'},
@@ -371,6 +377,12 @@ class Configuration : public IConfiguration
 			case 'm':
 				setKey("running-mode", IConfiguration::MODE_MERGE_ONLY);
 				break;
+			case '8': // Full system record
+				setKey("running-mode", IConfiguration::MODE_SYSTEM_RECORD);
+				break;
+			case '9': // Full system report
+				setKey("running-mode", IConfiguration::MODE_SYSTEM_REPORT);
+				break;
 			case 'l': {
 				StrVecMap_t vec = getCommaSeparatedList(std::string(optarg));
 
@@ -537,6 +549,8 @@ class Configuration : public IConfiguration
 		setKey("merged-name", "[merged]");
 		setKey("css-file", "");
 		setKey("lldb-use-raw-breakpoint-writes", 0);
+		setKey("system-mode-write-file", "");
+		setKey("system-mode-read-results-file", 0);
 	}
 
 
@@ -585,6 +599,10 @@ class Configuration : public IConfiguration
 			setKey(key, stoul(std::string(value)));
 		else if (key == "high-limit")
 			setKey(key, stoul(std::string(value)));
+		else if (key == "system-mode-write-file")
+			setKey(key, std::string(value));
+		else if (key == "system-mode-read-results-file")
+			setKey(key, stoul(std::string(value)));
 		else if (key == "bash-use-basic-parser")
 			setKey(key, stoul(std::string(value)));
 		else if (key == "lldb-use-raw-breakpoint-writes")
@@ -622,6 +640,8 @@ class Configuration : public IConfiguration
 				"\n"
 				" --gcov                  use gcov parser instead of DWARF debugging info\n"
 				" --clang                 use Clang Sanitizer-coverage parser\n"
+				" --system-record         perform full-system instrumentation\n"
+				" --system-report         report full-system coverage data\n"
 				" --skip-solibs           don't parse shared libraries (default: parse solibs)\n"
 				" --exit-first-process    exit when the first process exits, i.e., honor the\n"
 				"                         behavior of daemons (default: wait until last)\n"