Add sandboxee embedding

- Implement `--sapi_embed_name` and `--sapi_embed_dir` flags
- Do not emit full AST-serialization for C++ classes

sandboxed_api/tools/clang_generator/emitter.cc

@@ -17,9 +17,11 @@
 #include "absl/random/random.h"
 #include "absl/strings/ascii.h"
 #include "absl/strings/escaping.h"
+#include "absl/strings/match.h"
 #include "absl/strings/str_cat.h"
 #include "absl/strings/str_join.h"
 #include "absl/strings/str_replace.h"
+#include "absl/strings/strip.h"
 #include "clang/AST/DeclCXX.h"
 #include "sandboxed_api/tools/clang_generator/diagnostics.h"
 #include "sandboxed_api/util/status_macros.h"
@@ -57,21 +59,13 @@ std::string GetIncludeGuard(absl::string_view filename) {
   return guard;
 }
 
-// Returns the components of a declaration's qualified name, excluding the
-// declaration itself.
-sapi::StatusOr<std::vector<std::string>> GetQualifiedNamePath(
-    const clang::TypeDecl* decl) {
+// Returns the namespace components of a declaration's qualified name.
+std::vector<std::string> GetNamespacePath(const clang::TypeDecl* decl) {
   std::vector<std::string> comps;
   for (const auto* ctx = decl->getDeclContext(); ctx; ctx = ctx->getParent()) {
-    if (llvm::isa<clang::TranslationUnitDecl>(ctx)) {
-      continue;
+    if (const auto* nd = llvm::dyn_cast<clang::NamespaceDecl>(ctx)) {
+      comps.push_back(nd->getNameAsString());
     }
-    const auto* nd = llvm::dyn_cast<clang::NamespaceDecl>(ctx);
-    if (!nd) {
-      return MakeStatusWithDiagnostic(decl->getBeginLoc(),
-                                      absl::StrCat("not in a namespace decl"));
-    }
-    comps.push_back(nd->getNameAsString());
   }
   std::reverse(comps.begin(), comps.end());
   return comps;
@@ -79,6 +73,15 @@ sapi::StatusOr<std::vector<std::string>> GetQualifiedNamePath(
 
 // Serializes the given Clang AST declaration back into compilable source code.
 std::string PrintAstDecl(const clang::Decl* decl) {
+  // TODO(cblichmann): Make types nicer
+  //   - Rewrite typedef to using
+  //   - Rewrite function pointers using std::add_pointer_t<>;
+
+  if (const auto* record = llvm::dyn_cast<clang::CXXRecordDecl>(decl)) {
+    // For C++ classes/structs, only emit a forward declaration.
+    return absl::StrCat(record->isClass() ? "class " : "struct ",
+                        std::string(record->getName()));
+  }
   std::string pretty;
   llvm::raw_string_ostream os(pretty);
   decl->print(os);
@@ -120,9 +123,11 @@ sapi::StatusOr<std::string> EmitFunction(const clang::FunctionDecl* decl) {
   const clang::QualType return_type = decl->getDeclaredReturnType();
   const bool returns_void = return_type->isVoidType();
 
+  const clang::ASTContext& context = decl->getASTContext();
+
   // "Status<OptionalReturn> FunctionName("
-  absl::StrAppend(&out, MapQualTypeReturn(return_type), " ", function_name,
-                  "(");
+  absl::StrAppend(&out, MapQualTypeReturn(context, return_type), " ",
+                  function_name, "(");
 
   struct ParameterInfo {
     clang::QualType qual;
@@ -134,20 +139,22 @@ sapi::StatusOr<std::string> EmitFunction(const clang::FunctionDecl* decl) {
   for (int i = 0; i < decl->getNumParams(); ++i) {
     const clang::ParmVarDecl* param = decl->getParamDecl(i);
 
-    ParameterInfo& pi = params.emplace_back();
-    pi.qual = param->getType();
-    pi.name = GetParamName(param, i);
+    ParameterInfo& param_info = params.emplace_back();
+    param_info.qual = param->getType();
+    param_info.name = GetParamName(param, i);
 
     absl::StrAppend(&out, print_separator);
     print_separator = ", ";
-    absl::StrAppend(&out, MapQualTypeParameter(pi.qual), " ", pi.name);
+    absl::StrAppend(&out, MapQualTypeParameter(context, param_info.qual), " ",
+                    param_info.name);
   }
 
   absl::StrAppend(&out, ") {\n");
-  absl::StrAppend(&out, MapQualType(return_type), " v_ret_;\n");
+  absl::StrAppend(&out, MapQualType(context, return_type), " v_ret_;\n");
   for (const auto& [qual, name] : params) {
     if (!IsPointerOrReference(qual)) {
-      absl::StrAppend(&out, MapQualType(qual), " v_", name, "(", name, ");\n");
+      absl::StrAppend(&out, MapQualType(context, qual), " v_", name, "(", name,
+                      ");\n");
     }
   }
   absl::StrAppend(&out, "\nSAPI_RETURN_IF_ERROR(sandbox_->Call(\"",
@@ -167,44 +174,77 @@ sapi::StatusOr<std::string> EmitHeader(
   std::string out;
   const std::string include_guard = GetIncludeGuard(options.out_file);
   absl::StrAppendFormat(&out, kHeaderProlog, include_guard);
+
+  // When embedding the sandboxee, add embed header include
+  if (!options.embed_name.empty()) {
+    // Not using JoinPath() because even on Windows include paths use plain
+    // slashes.
+    std::string include_file(absl::StripSuffix(
+        absl::StrReplaceAll(options.embed_dir, {{"\\", "/"}}), "/"));
+    if (!include_file.empty()) {
+      absl::StrAppend(&include_file, "/");
+    }
+    absl::StrAppend(&include_file, options.embed_name);
+    absl::StrAppendFormat(&out, kEmbedInclude, include_file);
+  }
+
+  // If specified, wrap the generated API in a namespace
   if (options.has_namespace()) {
     absl::StrAppendFormat(&out, kNamespaceBeginTemplate,
                           options.namespace_name);
   }
-  {
-    std::string out_types = "// Types this API depends on\n";
-    bool added_dependent_types = false;
-    for (const clang::QualType& qual : types) {
-      clang::TypeDecl* decl = nullptr;
-      if (const auto* typedef_type = qual->getAs<clang::TypedefType>()) {
-        decl = typedef_type->getDecl();
-      } else if (const auto* enum_type = qual->getAs<clang::EnumType>()) {
-        decl = enum_type->getDecl();
-      } else {
-        decl = qual->getAsRecordDecl();
-      }
 
-      if (decl) {
-        SAPI_ASSIGN_OR_RETURN(auto ns, GetQualifiedNamePath(decl));
-        if (!ns.empty()) {
-          absl::StrAppend(&out_types, "namespace", (ns[0].empty() ? "" : " "),
-                          absl::StrJoin(ns, "::"), " {\n");
-        }
-        // TODO(cblichmann): Make types nicer
-        //   - Rewrite typedef to using
-        //   - Rewrite function pointers using std::add_pointer_t<>;
-        absl::StrAppend(&out_types, PrintAstDecl(decl), ";");
-        if (!ns.empty()) {
-          absl::StrAppend(&out_types, "\n}");
-        }
-        absl::StrAppend(&out_types, "\n");
-        added_dependent_types = true;
+  // Emit type dependencies
+  // TODO(cblichmann): Coalesce namespaces
+  std::string out_types = "// Types this API depends on\n";
+  bool added_types = false;
+  for (const clang::QualType& qual : types) {
+    clang::TypeDecl* decl = nullptr;
+    if (const auto* typedef_type = qual->getAs<clang::TypedefType>()) {
+      decl = typedef_type->getDecl();
+    } else if (const auto* enum_type = qual->getAs<clang::EnumType>()) {
+      decl = enum_type->getDecl();
+    } else {
+      decl = qual->getAsRecordDecl();
+    }
+    if (!decl) {
+      continue;
+    }
+
+    const std::vector<std::string> ns_path = GetNamespacePath(decl);
+    std::string nested_ns_name;
+    if (!ns_path.empty()) {
+      if (const auto& ns_root = ns_path.front();
+          ns_root == "std" || ns_root == "sapi" || ns_root == "__gnu_cxx") {
+        // Filter out any and all declarations from the C++ standard library,
+        // from SAPI itself and from other well-known namespaces. This avoids
+        // re-declaring things like standard integer types, for example.
+        continue;
       }
+      nested_ns_name = absl::StrCat(ns_path[0].empty() ? "" : " ",
+                                    absl::StrJoin(ns_path, "::"));
+      absl::StrAppend(&out_types, "namespace", nested_ns_name, " {\n");
     }
-    if (added_dependent_types) {
-      absl::StrAppend(&out, out_types);
+    absl::StrAppend(&out_types, PrintAstDecl(decl), ";");
+    if (!ns_path.empty()) {
+      absl::StrAppend(&out_types, "\n}  // namespace", nested_ns_name);
     }
+    absl::StrAppend(&out_types, "\n");
+    added_types = true;
   }
+  if (added_types) {
+    absl::StrAppend(&out, out_types);
+  }
+
+  // Optionally emit a default sandbox that instantiates an embedded sandboxee
+  if (!options.embed_name.empty()) {
+    // TODO(cblichmann): Make the "Sandbox" suffix configurable.
+    absl::StrAppendFormat(
+        &out, kEmbedClassTemplate, absl::StrCat(options.name, "Sandbox"),
+        absl::StrReplaceAll(options.embed_name, {{"-", "_"}}));
+  }
+
+  // Emit the actual Sandboxed API
   // TODO(cblichmann): Make the "Api" suffix configurable or at least optional.
   absl::StrAppendFormat(&out, kClassHeaderTemplate,
                         absl::StrCat(options.name, "Api"));
@@ -215,6 +255,7 @@ sapi::StatusOr<std::string> EmitHeader(
   }
   absl::StrAppend(&out, kClassFooterTemplate);
 
+  // Close out the header: close namespace (if needed) and end include guard
   if (options.has_namespace()) {
     absl::StrAppendFormat(&out, kNamespaceEndTemplate, options.namespace_name);
   }

sandboxed_api/tools/clang_generator/emitter.h

@@ -43,6 +43,8 @@ sapi::StatusOr<std::string> EmitHeader(
 // Common file prolog with auto-generation notice.
 // Note: The includes will be adjusted by Copybara when converting to/from
 //       internal code. This is intentional.
+// Text template arguments:
+//   1. Header guard
 inline constexpr absl::string_view kHeaderProlog =
     R"(// AUTO-GENERATED by the Sandboxed API generator.
 // Edits will be discarded when regenerating this file.
@@ -65,6 +67,12 @@ inline constexpr absl::string_view kHeaderEpilog =
 R"(
 #endif  // %1$s)";
 
+// Text template arguments:
+//   1. Include for embedded sandboxee objects
+inline constexpr absl::string_view kEmbedInclude = R"(#include "%1$s_embed.h"
+
+)";
+
 // Text template arguments:
 //   1. Namespace name
 inline constexpr absl::string_view kNamespaceBeginTemplate =
@@ -77,6 +85,18 @@ R"(
 }  // namespace %1$s
 )";
 
+// Text template arguments:
+//   1. Class name
+//   2. Embedded object identifier
+inline constexpr absl::string_view kEmbedClassTemplate = R"(
+// Sandbox with embedded sandboxee and default policy
+class %1$s : public ::sapi::Sandbox {
+ public:
+  %1$s() : ::sapi::Sandbox(%2$s_embed_create()) {}
+};
+
+)";
+
 // Text template arguments:
 //   1. Class name
 inline constexpr absl::string_view kClassHeaderTemplate = R"(

sandboxed_api/tools/clang_generator/generator.h

@@ -46,6 +46,8 @@ struct GeneratorOptions {
   std::string name;            // Name of the Sandboxed API
   std::string namespace_name;  // Namespace to wrap the SAPI in
   std::string out_file;        // Output path of the generated header
+  std::string embed_dir;       // Directory with embedded includes
+  std::string embed_name;      // Identifier of the embed object
 };
 
 class GeneratorASTVisitor
@@ -92,8 +94,6 @@ class GeneratorAction : public clang::ASTFrontendAction {
       : options_(options) {}
 
  private:
-  friend class GeneratorFactory;
-
   std::unique_ptr<clang::ASTConsumer> CreateASTConsumer(
       clang::CompilerInstance&, llvm::StringRef in_file) override {
     return absl::make_unique<GeneratorASTConsumer>(std::string(in_file),

sandboxed_api/tools/clang_generator/generator_tool.cc

@@ -83,6 +83,8 @@ GeneratorOptions GeneratorOptionsFromFlags() {
   options.name = *g_sapi_name;
   options.namespace_name = *g_sapi_ns;
   options.out_file = *g_sapi_out;
+  options.embed_dir = *g_sapi_embed_dir;
+  options.embed_name = *g_sapi_embed_name;
   return options;
 }
 

sandboxed_api/tools/clang_generator/types.cc

@@ -78,7 +78,24 @@ void GatherRelatedTypes(const clang::ASTContext& context, clang::QualType qual,
   }
 }
 
-std::string MapQualType(clang::QualType qual) {
+namespace {
+
+// Removes "const" from a qualified type if it denotes a pointer or reference
+// type.
+clang::QualType MaybeRemoveConst(const clang::ASTContext& context,
+                                 clang::QualType qual) {
+  if (IsPointerOrReference(qual)) {
+    clang::QualType pointee_qual = qual->getPointeeType();
+    pointee_qual.removeLocalConst();
+    qual = context.getPointerType(pointee_qual);
+  }
+  return qual;
+}
+
+}  // namespace
+
+std::string MapQualType(const clang::ASTContext& context,
+                        clang::QualType qual) {
   if (const auto* builtin = qual->getAs<clang::BuiltinType>()) {
     switch (builtin->getKind()) {
       case clang::BuiltinType::Void:
@@ -156,12 +173,18 @@ std::string MapQualType(clang::QualType qual) {
   } else if (const auto* enum_type = qual->getAs<clang::EnumType>()) {
     return absl::StrCat("::sapi::v::IntBase<",
                         enum_type->getDecl()->getQualifiedNameAsString(), ">");
+  } else if (IsPointerOrReference(qual)) {
+    // Remove "const" qualifier from a pointer or reference type's pointee, as
+    // e.g. const pointers do not work well with SAPI.
+    return absl::StrCat("::sapi::v::Reg<",
+                        MaybeRemoveConst(context, qual).getAsString(), ">");
   }
-  // Best-effort mapping to "int"
-  return "::sapi::v::Int";
+  // Best-effort mapping to "int", leave a comment.
+  return absl::StrCat("::sapi::v::Int /* aka '", qual.getAsString(), "' */");
 }
 
-std::string MapQualTypeParameter(clang::QualType qual) {
+std::string MapQualTypeParameter(const clang::ASTContext& /*context*/,
+                                 clang::QualType qual) {
   // TODO(cblichmann): Define additional mappings, as appropriate
   //   _Bool              -> bool
   //   unsigned long long -> uint64_t (where applicable)
@@ -169,10 +192,14 @@ std::string MapQualTypeParameter(clang::QualType qual) {
   return IsPointerOrReference(qual) ? "::sapi::v::Ptr*" : qual.getAsString();
 }
 
-std::string MapQualTypeReturn(clang::QualType qual) {
-  return qual->isVoidType() ? "absl::Status"
-                            : absl::StrCat("::sapi::StatusOr<",
-                                           MapQualTypeParameter(qual), ">");
+std::string MapQualTypeReturn(const clang::ASTContext& context,
+                              clang::QualType qual) {
+  if (qual->isVoidType()) {
+    return "absl::Status";
+  }
+  // Remove const qualifier like in MapQualType().
+  return absl::StrCat("::sapi::StatusOr<",
+                      MaybeRemoveConst(context, qual).getAsString(), ">");
 }
 
 }  // namespace sapi

sandboxed_api/tools/clang_generator/types.h

@@ -60,17 +60,19 @@ void GatherRelatedTypes(const clang::ASTContext& context, clang::QualType qual,
 // Maps a qualified type to a fully qualified SAPI-compatible type name. This
 // is used for the generated code that invokes the actual function call RPC.
 // If no mapping can be found, "int" is assumed.
-std::string MapQualType(clang::QualType qual);
+std::string MapQualType(const clang::ASTContext& context, clang::QualType qual);
 
 // Maps a qualified type used as a function parameter to a type name compatible
 // with the generated Sandboxed API.
-std::string MapQualTypeParameter(clang::QualType qual);
+std::string MapQualTypeParameter(const clang::ASTContext& context,
+                                 clang::QualType qual);
 
 // Maps a qualified type used as a function return type to a type name
-// compatible with the generated Sandboxed API. Uses MapQualTypeParameter() and 
+// compatible with the generated Sandboxed API. Uses MapQualTypeParameter() and
 // wraps the type in a ::sapi::StatusOr<> if qual is non-void. Otherwise returns
 // absl::Status.
-std::string MapQualTypeReturn(clang::QualType qual);
+std::string MapQualTypeReturn(const clang::ASTContext& context,
+                              clang::QualType qual);
 
 }  // namespace sapi