i#3161: Add pure-static-client-DR-app support (#7177)

Adds support for statically linking a client and DR into a pure-static
application.

The code support for this involves expecting to not find symbols via
dlsym and to instead directly invoke dr_client_main(). A weak copy of
dr_client_main() is added to avoid problems with static DR used in
standalone mode.

Since even static client libraries today pull in the shared DR library,
getting CMake to build a pure static binary with a client occupies the
bulk of the changes here:

+ A new cmake routine configure_DynamoRIO_static_client() is added which
links with dynamorio_static instead of dynamorio.

+ Static libz libraries are identified in an admittedly hacky manner.

+ A new client library variant "_drstatic" is added (alongside today's
base shared library and "_static" variant) for all the ext/ libraries.
This is identical to "_static" except it uses
configure_DynamoRIO_static_client(). A drmemtrace_drstatic variant is
also added, which also links with static libz.

+ The drmemtrace weak dr_client_main no longer works with the new
dynamorio_static weak symbol so it is instead made strong and removed
completely when DRMEMTRACE_NO_MAIN is defined; a drmemtrace_nomain
variant is created with this set.

A new test tool.drcacheoff.purestatic is added which verifies that a
static client and DR linked into a pure-static binary all work together.
It is difficult to get it to link on Ubuntu20, when cross-compiling, and
in other setups due to the need for a static libpthread and other
requirements, so the test is limited to Ubuntu22 64-bit where it is
added to the test list there and confirmed to be built and run:
```
 8/11 Test #414: code_api|tool.drcacheoff.purestatic ...   Passed    2.04 sec
```

Fixes #3161

Author: derekbruening

CMakeLists.txt

@@ -1,5 +1,5 @@
 # **********************************************************
-# Copyright (c) 2010-2024 Google, Inc.    All rights reserved.
+# Copyright (c) 2010-2025 Google, Inc.    All rights reserved.
 # Copyright (c) 2009-2010 VMware, Inc.    All rights reserved.
 # Copyright (c) 2018 Arm Limited          All rights reserved.
 # **********************************************************
@@ -1868,6 +1868,13 @@ if (WIN32 AND NOT DISABLE_ZLIB)
   set(ZLIB_FOUND OFF)
 else ()
   find_package(ZLIB)
+  # FindZLIB doesn't give us any library split so we manually
+  # convert to static.
+  set(ZLIB_STATIC_LIBRARIES "")
+  foreach (lib ${ZLIB_LIBRARIES})
+    string(REGEX REPLACE "\\.so$" ".a" newlib "${lib}")
+    list(APPEND ZLIB_STATIC_LIBRARIES ${newlib})
+  endforeach ()
 endif ()
 # On Ubuntu 14.10, 32-bit builds fail to link with -lsnappy, just ignore.
 if (UNIX AND X64)

api/docs/release.dox

@@ -1,5 +1,5 @@
 /* ******************************************************************************
- * Copyright (c) 2010-2024 Google, Inc.  All rights reserved.
+ * Copyright (c) 2010-2025 Google, Inc.  All rights reserved.
  * Copyright (c) 2011 Massachusetts Institute of Technology  All rights reserved.
  * Copyright (c) 2008-2010 VMware, Inc.  All rights reserved.
  * ******************************************************************************/
@@ -129,15 +129,18 @@ changes:
  - No compatibility changes yet.
 
 Further non-compatibility-affecting changes include:
- - No changes yet.
+ - Added support for statically linking clients and the DynamoRIO library into
+   a pure-static application via a new configure_DynamoRIO_static_client() CMake
+   function and new "_drstatic" static library CMake targets for the provided
+   extension libraries.
 
 **************************************************
 <hr>
 
 The changes between version 11.2.0 and 11.1.0 include the following compatibility
 changes:
- - Added #dynamorio::drmemtrace::TRACE_ENTRY_VERSION_RETIRED_INSTRUCTIONS_ONLY to increase the
-   trace version for drmemtraces with uncompleted instructions removed.
+ - Added #dynamorio::drmemtrace::TRACE_ENTRY_VERSION_RETIRED_INSTRUCTIONS_ONLY to
+   increase the trace version for drmemtraces with uncompleted instructions removed.
 
 Further non-compatibility-affecting changes include:
  - Added instr_get_operation_size() and instr_set_operation_size() APIs for

clients/drcachesim/CMakeLists.txt

@@ -1,5 +1,5 @@
 # **********************************************************
-# Copyright (c) 2015-2024 Google, Inc.    All rights reserved.
+# Copyright (c) 2015-2025 Google, Inc.    All rights reserved.
 # **********************************************************
 
 # Redistribution and use in source and binary forms, with or without
@@ -98,6 +98,7 @@ if (ZLIB_FOUND)
       "until then, disabling zip output and fast seeking")
     set(zip_reader "")
     set(zlib_libs ${ZLIB_LIBRARIES})
+    set(zlib_static_libs ${ZLIB_STATIC_LIBRARIES})
     set(ZIP_FOUND OFF)
   else ()
     file(GLOB minizip_srcs "${minizip_dir}/*.c")
@@ -117,6 +118,7 @@ if (ZLIB_FOUND)
     install_exported_target(minizip ${INSTALL_CLIENTS_LIB})
     set(zip_reader reader/zipfile_file_reader.cpp)
     set(zlib_libs ${ZLIB_LIBRARIES} minizip)
+    set(zlib_static_libs ${ZLIB_STATIC_LIBRARIES} minizip)
   endif ()
 else ()
   # XXX: We could ship with a zlib for Windows, or build the third_party/zlib.
@@ -473,9 +475,13 @@ if (NOT AARCH64 AND NOT APPLE)
 endif ()
 target_link_libraries(scheduler_launcher ${zlib_libs})
 
-macro(add_drmemtrace name type)
+macro(add_drmemtrace name type static_DR no_main)
   if (${type} STREQUAL "STATIC")
-    set(ext_sfx "_static")
+    if (${static_DR})
+      set(ext_sfx "_drstatic")
+    else ()
+      set(ext_sfx "_static")
+    endif ()
   else ()
     set(ext_sfx "")
   endif ()
@@ -519,7 +525,14 @@ macro(add_drmemtrace name type)
   endif()
 
   add_library(${name} ${type} ${drmemtrace_srcs})
-  configure_DynamoRIO_client(${name})
+  if (${no_main})
+    append_property_list(TARGET ${name} COMPILE_DEFINITIONS "DRMEMTRACE_NO_MAIN")
+  endif ()
+  if (${static_DR})
+    configure_DynamoRIO_static_client(${name})
+  else ()
+    configure_DynamoRIO_client(${name})
+  endif()
   use_DynamoRIO_extension(${name} drmgr${ext_sfx})
   use_DynamoRIO_extension(${name} drsyms${ext_sfx})
   use_DynamoRIO_extension(${name} drwrap${ext_sfx})
@@ -536,7 +549,11 @@ macro(add_drmemtrace name type)
   if (libsnappy)
     target_link_libraries(${name} snappy)
   endif ()
-  target_link_libraries(${name} ${zlib_libs})
+  if (${static_DR})
+    target_link_libraries(${name} ${zlib_static_libs})
+  else ()
+    target_link_libraries(${name} ${zlib_libs})
+  endif ()
   if (liblz4)
     target_link_libraries(${name} lz4)
   endif ()
@@ -547,8 +564,13 @@ macro(add_drmemtrace name type)
   install_target(${name} ${INSTALL_CLIENTS_LIB})
 endmacro()
 
-add_drmemtrace(drmemtrace SHARED)
-add_drmemtrace(drmemtrace_static STATIC)
+add_drmemtrace(drmemtrace SHARED OFF OFF)
+# This static library links with the DR shared library and has dr_client_main:
+add_drmemtrace(drmemtrace_static STATIC OFF OFF)
+# This static library links with the DR shared library and has no dr_client_main:
+add_drmemtrace(drmemtrace_nomain STATIC OFF ON)
+# This static library links with the DR static library and has dr_client_main:
+add_drmemtrace(drmemtrace_drstatic STATIC ON OFF)
 append_property_list(TARGET drmemtrace_static COMPILE_DEFINITIONS "DRMEMTRACE_STATIC")
 # We export drmemtrace.h to the same place as the analysis tool headers
 # for simplicity, rather than sticking it into ext/include or sthg.
@@ -1083,7 +1105,7 @@ if (BUILD_TESTS)
     if (LINUX) # Uses mremap.
       add_executable(tool.drcacheoff.burst_maps tests/burst_maps.cpp)
       configure_DynamoRIO_static(tool.drcacheoff.burst_maps)
-      use_DynamoRIO_static_client(tool.drcacheoff.burst_maps drmemtrace_static)
+      use_DynamoRIO_static_client(tool.drcacheoff.burst_maps drmemtrace_nomain)
       target_link_libraries(tool.drcacheoff.burst_maps test_helpers)
       add_win32_flags(tool.drcacheoff.burst_maps)
     endif ()
@@ -1130,10 +1152,25 @@ if (BUILD_TESTS)
       append_property_list(TARGET tool.drcacheoff.burst_client
         COMPILE_DEFINITIONS "TEST_APP_DR_CLIENT_MAIN")
       configure_DynamoRIO_static(tool.drcacheoff.burst_client)
-      use_DynamoRIO_static_client(tool.drcacheoff.burst_client drmemtrace_static)
+      use_DynamoRIO_static_client(tool.drcacheoff.burst_client drmemtrace_nomain)
       target_link_libraries(tool.drcacheoff.burst_client test_helpers)
       # A nop, keep it for the future Windows support.
       add_win32_flags(tool.drcacheoff.burst_client)
+
+      # Test a pure-static-link app.
+      # We ignore "warning: Using 'dlopen' in statically linked applications requires
+      # at runtime the shared libraries from the glibc version used for linking".
+      # XXX: We can only successfully build this if static libpthread is available,
+      # which is only recent glibc 2.34+ such as Ubuntu22+. We also limit to 64-bit
+      # due to pain building 32-bit on a 64-bit machine.
+      if (X64 AND "${libpthread}" MATCHES ".a$")
+        add_executable(tool.drcacheoff.purestatic tests/burst_static.cpp)
+        append_property_list(TARGET tool.drcacheoff.purestatic
+          LINK_FLAGS "-static -Wl,--no-export-dynamic")
+        configure_DynamoRIO_static(tool.drcacheoff.purestatic)
+        use_DynamoRIO_static_client(tool.drcacheoff.purestatic drmemtrace_drstatic)
+        target_link_libraries(tool.drcacheoff.purestatic test_helpers)
+      endif ()
     endif ()
 
     if (NOT RISCV64) # TODO i#3544: Port tests to RISC-V 64

clients/drcachesim/tests/offline-purestatic.templatex

@@ -0,0 +1,46 @@
+pre-DR init
+pre-DR start
+pre-DR detach
+DynamoRIO statistics:
+ *Peak threads under DynamoRIO control : *1
+.*
+all done
+pre-DR init
+pre-DR start
+pre-DR detach
+DynamoRIO statistics:
+ *Peak threads under DynamoRIO control : *1
+.*
+all done
+pre-DR init
+pre-DR start
+pre-DR detach
+DynamoRIO statistics:
+ *Peak threads under DynamoRIO control : *1
+.*
+all done
+Cache simulation results:
+Core #0 \(traced CPU\(s\): #0\)
+  L1I0 .* stats:
+    Hits:                         *[0-9,\.]*
+    Misses:                       *[0-9,\.]*
+    Compulsory misses:            *[0-9,\.]*
+    Invalidations:                *0
+.*    Miss rate:                        [0-3][,\.]..%
+  L1D0 .* stats:
+    Hits:                         *[0-9,\.]*
+    Misses:                       *[0-9,\.]*
+    Compulsory misses:            *[0-9,\.]*
+    Invalidations:                *0
+.*   Miss rate:                        [0-3][,\.]..%
+Core #1 \(traced CPU\(s\): \)
+Core #2 \(traced CPU\(s\): \)
+Core #3 \(traced CPU\(s\): \)
+LL .* stats:
+    Hits:                         *[0-9,\.]*
+    Misses:                       *[0-9,\.]*
+    Compulsory misses:            *[0-9,\.]*
+    Invalidations:                *0
+.*   Local miss rate:        *[0-9,.]*%
+    Child hits:                   *[0-9,\.]*
+    Total miss rate:                  [0-1][,\.]..%

clients/drcachesim/tracer/tracer.cpp

@@ -1,5 +1,5 @@
 /* ******************************************************************************
- * Copyright (c) 2011-2024 Google, Inc.  All rights reserved.
+ * Copyright (c) 2011-2025 Google, Inc.  All rights reserved.
  * Copyright (c) 2010 Massachusetts Institute of Technology  All rights reserved.
  * ******************************************************************************/
 
@@ -2538,14 +2538,17 @@ drmemtrace_client_main(client_id_t id, int argc, const char *argv[])
  */
 
 /* To support statically linked multiple clients, we add drmemtrace_client_main
- * as the real client init function and make dr_client_main a weak symbol.
- * We could also use alias to link dr_client_main to drmemtrace_client_main.
- * A simple call won't add too much overhead, and works both in Windows and Linux.
- * To automate the process and minimize the code change, we should investigate the
- * approach that uses command-line link option to alias two symbols.
+ * as the real client init function and let a separate dr_client_main call
+ * drmemtrace_client_main. Since dynamorio_static now provides its own weak
+ * dr_client_main symbol, we can't simply always provide a weak dr_client_main here:
+ * it must either be present and strong or not present, controlled by the
+ * DRMEMTRACE_NO_MAIN define.
  */
-DR_EXPORT WEAK void
+#ifndef DRMEMTRACE_NO_MAIN
+DR_EXPORT
+void
 dr_client_main(client_id_t id, int argc, const char *argv[])
 {
     dynamorio::drmemtrace::drmemtrace_client_main(id, argc, argv);
 }
+#endif

core/lib/instrument.c

@@ -1,5 +1,5 @@
 /* ******************************************************************************
- * Copyright (c) 2010-2024 Google, Inc.  All rights reserved.
+ * Copyright (c) 2010-2025 Google, Inc.  All rights reserved.
  * Copyright (c) 2010-2011 Massachusetts Institute of Technology  All rights reserved.
  * Copyright (c) 2002-2010 VMware, Inc.  All rights reserved.
  * ******************************************************************************/
@@ -371,6 +371,19 @@ DECLARE_CXTSWPROT_VAR(static mutex_t client_aux_lib64_lock,
                       INIT_LOCK_FREE(client_aux_lib64_lock));
 #endif
 
+#if defined(STATIC_LIBRARY) && !defined(WINDOWS)
+// To support static DR used for both standalone and clients we need to provide
+// some copy of client main. There is no WEAK on Windows so we do not support
+// this usage there.
+WEAK void
+dr_client_main(client_id_t id, int argc, const char *argv[])
+{
+    // This will be called when using static DR but no client, so we can't
+    // assert even though this should not be reached when there is a real
+    // client whose non-weak dr_client_main overrides this.
+}
+#endif
+
 /****************************************************************************/
 /* INTERNAL ROUTINES */
 
@@ -577,8 +590,16 @@ add_client_lib(const char *path, const char *id_str, const char *options)
         /* PR 250952: version check */
         int *uses_dr_version =
             (int *)lookup_library_routine(client_lib, USES_DR_VERSION_NAME);
-        if (uses_dr_version == NULL || *uses_dr_version < OLDEST_COMPATIBLE_VERSION ||
-            *uses_dr_version > NEWEST_COMPATIBLE_VERSION) {
+        bool pure_static = false;
+#ifdef STATIC_LIBRARY
+        if (uses_dr_version == NULL) {
+            // We assume we're in a pure-static app where dlsym fails.
+            pure_static = true;
+        }
+#endif
+        if (!pure_static &&
+            (uses_dr_version == NULL || *uses_dr_version < OLDEST_COMPATIBLE_VERSION ||
+             *uses_dr_version > NEWEST_COMPATIBLE_VERSION)) {
             /* not a fatal usage error since we want release build to continue */
             CLIENT_ASSERT(false,
                           "client library is incompatible with this version of DR");
@@ -596,8 +617,9 @@ add_client_lib(const char *path, const char *id_str, const char *options)
             // to the dll bounds functions. xref i#3387.
             client_start = get_dynamorio_dll_start();
             client_end = get_dynamorio_dll_end();
-            ASSERT(client_start <= (app_pc)uses_dr_version &&
-                   (app_pc)uses_dr_version < client_end);
+            ASSERT(pure_static ||
+                   (client_start <= (app_pc)uses_dr_version &&
+                    (app_pc)uses_dr_version < client_end));
 #else
             DEBUG_DECLARE(bool ok =)
             shared_library_bounds(client_lib, (byte *)uses_dr_version, NULL,
@@ -773,6 +795,15 @@ instrument_init(void)
             (*init)(client_libs[i].id, client_libs[i].argc, client_libs[i].argv);
         else if (legacy != NULL)
             (*legacy)(client_libs[i].id);
+#if defined(STATIC_LIBRARY) && !defined(WINDOWS)
+        else {
+            // For pure-static apps we support only INSTRUMENT_INIT_NAME.
+            // There is no WEAK support on Windows so we do not support this there
+            // (plus pure-static is not really practical either).
+            extern void dr_client_main(uint id, int argc, const char *argv[]);
+            dr_client_main(client_libs[i].id, client_libs[i].argc, client_libs[i].argv);
+        }
+#endif
     }
 
     /* We now initialize the 1st thread before coming here, so we can

ext/CMakeLists.txt

@@ -1,5 +1,5 @@
 # **********************************************************
-# Copyright (c) 2010-2022 Google, Inc.    All rights reserved.
+# Copyright (c) 2010-2025 Google, Inc.    All rights reserved.
 # Copyright (c) 2010 VMware, Inc.    All rights reserved.
 # **********************************************************
 
@@ -61,8 +61,12 @@ add_dr_defines()
 
 # This lets us share common code among all extensions.  We use a macro to
 # avoid scope issues and avoid having to call configure_DynamoRIO_global().
-macro(configure_extension target is_static)
-  configure_DynamoRIO_client(${target})
+macro(configure_extension target is_static static_DR)
+  if (${static_DR})
+    configure_DynamoRIO_static_client(${target})
+  else ()
+    configure_DynamoRIO_client(${target})
+  endif ()
 
   # ensure we rebuild if includes change
   add_dependencies(${target} api_headers)

ext/drbbdup/CMakeLists.txt

@@ -1,5 +1,5 @@
 # **********************************************************
-# Copyright (c) 2013-2020 Google, Inc.    All rights reserved.
+# Copyright (c) 2013-2025 Google, Inc.    All rights reserved.
 # **********************************************************
 
 # Redistribution and use in source and binary forms, with or without
@@ -47,15 +47,21 @@ set(srcs
 set(srcs_static ${srcs})
 
 add_library(drbbdup SHARED ${srcs})
-configure_extension(drbbdup OFF)
+configure_extension(drbbdup OFF OFF)
 use_DynamoRIO_extension(drbbdup drcontainers)
 use_DynamoRIO_extension(drbbdup drmgr)
 use_DynamoRIO_extension(drbbdup drreg)
 
 add_library(drbbdup_static STATIC ${srcs_static})
-configure_extension(drbbdup_static ON)
+configure_extension(drbbdup_static ON OFF)
 use_DynamoRIO_extension(drbbdup_static drcontainers)
 use_DynamoRIO_extension(drbbdup_static drmgr_static)
 use_DynamoRIO_extension(drbbdup_static drreg_static)
 
+add_library(drbbdup_drstatic STATIC ${srcs_static})
+configure_extension(drbbdup_drstatic ON ON)
+use_DynamoRIO_extension(drbbdup_drstatic drcontainers_drstatic)
+use_DynamoRIO_extension(drbbdup_drstatic drmgr_drstatic)
+use_DynamoRIO_extension(drbbdup_drstatic drreg_drstatic)
+
 install_ext_header(drbbdup.h)

ext/drcallstack/CMakeLists.txt

@@ -1,5 +1,5 @@
 # **********************************************************
-# Copyright (c) 2021 Google, Inc.    All rights reserved.
+# Copyright (c) 2021-2025 Google, Inc.    All rights reserved.
 # **********************************************************
 
 # Redistribution and use in source and binary forms, with or without
@@ -48,11 +48,15 @@ endif ()
 
 add_library(drcallstack SHARED ${srcs})
 set(PREFERRED_BASE 0x79800000)
-configure_extension(drcallstack OFF)
+configure_extension(drcallstack OFF OFF)
 target_link_libraries(drcallstack unwind)
 
 add_library(drcallstack_static STATIC ${srcs_static})
-configure_extension(drcallstack_static ON)
+configure_extension(drcallstack_static ON OFF)
 target_link_libraries(drcallstack_static unwind)
 
+add_library(drcallstack_drstatic STATIC ${srcs_static})
+configure_extension(drcallstack_drstatic ON ON)
+target_link_libraries(drcallstack_drstatic unwind)
+
 install_ext_header(drcallstack.h)