fix(plugin): Use UTF-8 paths on Windows

Document that plugin paths should be UTF-8 on Windows, and convert them
to UTF-16 internally.

In sinsp-example, make sure our command line arguments (including the
plugin path) are UTF-8 on Windows.

Signed-off-by: Gerald Combs <gerald@wireshark.org>

Author: geraldcombs

userspace/libsinsp/examples/test.cpp

@@ -35,7 +35,10 @@ limitations under the License.
 #include <thread>
 #include <json/json.h>
 
-#ifndef _WIN32
+#ifdef _WIN32
+#include <Windows.h>
+#include <shellapi.h>
+#else
 extern "C" {
 #include <sys/syscall.h>
 #include <sys/stat.h>
@@ -688,13 +691,46 @@ static void print_perftest_table_data(const uint64_t num_events_diff,
 	          << std::setw(14) << curr_vm_rss_in_mebibytes << '\r' << std::flush;
 }
 
+#ifdef _WIN32
+char* utf16_to_utf8(const LPWSTR utf16_str) {
+	int utf8_str_len = WideCharToMultiByte(CP_UTF8, 0, utf16_str, -1, NULL, 0, NULL, NULL);
+	if(utf8_str_len <= 0) {
+		return NULL;
+	}
+	char* utf8_str = (char*)malloc(utf8_str_len * sizeof(char));
+	if(WideCharToMultiByte(CP_UTF8, 0, utf16_str, -1, utf8_str, utf8_str_len, NULL, NULL) <= 0) {
+		free(utf8_str);
+		return NULL;
+	}
+	return utf8_str;
+}
+#endif
+
 //
 // Sample filters:
 //   "evt.category=process or evt.category=net"
 //   "evt.category=net or (evt.type=execveat or evt.type=execve or evt.type=clone or evt.type=fork
 //   or evt.type=vfork)"
 //
 int main(int argc, char** argv) {
+#ifdef _WIN32
+	// main+argv uses the system encoding, which may not be able to handle all
+	// filesystem paths. Fetch the wide command line arguments and convert them
+	// to UTF-8.
+	int wide_argc;
+	LPWSTR* wide_argv = CommandLineToArgvW(GetCommandLineW(), &wide_argc);
+
+	if(wide_argv && wide_argc == argc) {
+		char** new_argv = (char**)malloc((argc + 1) * sizeof(char*));
+		for(int idx = 0; idx < argc; idx++) {
+			new_argv[idx] = utf16_to_utf8(wide_argv[idx]);
+		}
+		new_argv[argc] = NULL;
+		argv = new_argv;
+	}
+	SetConsoleOutputCP(CP_UTF8);
+#endif
+
 	sinsp inspector;
 
 	filter_list.reset(new sinsp_filter_check_list());

userspace/libsinsp/sinsp.h

@@ -734,6 +734,8 @@ class SINSP_PUBLIC sinsp : public capture_stats_source {
 	// Create and register a plugin from a shared library pointed
 	// to by filepath, and add it to the inspector.
 	// The created sinsp_plugin is returned.
+	// filepath must be encoded as UTF-8 on Windows.
+
 	std::shared_ptr<sinsp_plugin> register_plugin(const std::string& filepath);
 
 	// Create and register a plugin given a custom API vtable.

userspace/plugin/plugin_loader.c

@@ -71,7 +71,24 @@ plugin_handle_t* plugin_load(const char* path, char* err) {
 
 	// open dynamic library
 #ifdef _WIN32
-	ret->handle = LoadLibrary(path);
+	// Using LoadLibraryW ensures that we have a valid path independent
+	// of the system or application code page.
+	int wpath_len = MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS, path, -1, NULL, 0);
+	if(wpath_len <= 0) {
+		free(ret);
+		strlcpy(err, "unable to decode plugin path", PLUGIN_MAX_ERRLEN);
+		return NULL;
+	}
+
+	WCHAR* wpath = malloc(wpath_len * sizeof(WCHAR));
+	if(MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS, path, -1, wpath, wpath_len) <= 0) {
+		free(ret);
+		free(wpath);
+		strlcpy(err, "unable to decode plugin path", PLUGIN_MAX_ERRLEN);
+		return NULL;
+	}
+	ret->handle = LoadLibraryW(wpath);
+	free(wpath);
 	if(ret->handle == NULL) {
 		DWORD flg = FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM |
 		            FORMAT_MESSAGE_IGNORE_INSERTS;

userspace/plugin/plugin_loader.h

@@ -72,6 +72,8 @@ plugin_handle_t* plugin_load_api(const plugin_api* api, char* err);
     \brief Loads a dynamic library from the given path and returns a
     plugin_handle_t* representing the loaded plugin. In case of error,
     returns NULL and fills the err string up to PLUGIN_MAX_ERRLEN chars.
+
+    The path must be encoded as UTF-8 on Windows.
 */
 plugin_handle_t* plugin_load(const char* path, char* err);